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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include <linux/rculist.h>
  81 #include "bonding.h"
  82 #include "bond_3ad.h"
  83 #include "bond_alb.h"
  84
  85 /*---------------------------- Module parameters ----------------------------*/
  86
  87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  88 #define BOND_LINK_MON_INTERV    0
  89 #define BOND_LINK_ARP_INTERV    0
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon       = BOND_LINK_MON_INTERV;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval = BOND_LINK_ARP_INTERV;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113
 114 module_param(max_bonds, int, 0);
 115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 116 module_param(tx_queues, int, 0);
 117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 118 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 119 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 120                                "failover event (alias of num_unsol_na)");
 121 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 122 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 123                                "failover event (alias of num_grat_arp)");
 124 module_param(miimon, int, 0);
 125 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 126 module_param(updelay, int, 0);
 127 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 128 module_param(downdelay, int, 0);
 129 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 130                             "in milliseconds");
 131 module_param(use_carrier, int, 0);
 132 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 133                               "0 for off, 1 for on (default)");
 134 module_param(mode, charp, 0);
 135 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 136                        "1 for active-backup, 2 for balance-xor, "
 137                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 138                        "6 for balance-alb");
 139 module_param(primary, charp, 0);
 140 MODULE_PARM_DESC(primary, "Primary network device to use");
 141 module_param(primary_reselect, charp, 0);
 142 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 143                                    "once it comes up; "
 144                                    "0 for always (default), "
 145                                    "1 for only if speed of primary is "
 146                                    "better, "
 147                                    "2 for only on active slave "
 148                                    "failure");
 149 module_param(lacp_rate, charp, 0);
 150 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 151                             "0 for slow, 1 for fast");
 152 module_param(ad_select, charp, 0);
 153 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 154                             "0 for stable (default), 1 for bandwidth, "
 155                             "2 for count");
 156 module_param(min_links, int, 0);
 157 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 158
 159 module_param(xmit_hash_policy, charp, 0);
 160 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 161                                    "0 for layer 2 (default), 1 for layer 3+4, "
 162                                    "2 for layer 2+3");
 163 module_param(arp_interval, int, 0);
 164 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 165 module_param_array(arp_ip_target, charp, NULL, 0);
 166 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 167 module_param(arp_validate, charp, 0);
 168 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 169                                "0 for none (default), 1 for active, "
 170                                "2 for backup, 3 for all");
 171 module_param(arp_all_targets, charp, 0);
 172 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 173 module_param(fail_over_mac, charp, 0);
 174 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 175                                 "the same MAC; 0 for none (default), "
 176                                 "1 for active, 2 for follow");
 177 module_param(all_slaves_active, int, 0);
 178 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 179                                      "by setting active flag for all slaves; "
 180                                      "0 for never (default), 1 for always.");
 181 module_param(resend_igmp, int, 0);
 182 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 183                               "link failure");
 184
 185 /*----------------------------- Global variables ----------------------------*/
 186
 187 #ifdef CONFIG_NET_POLL_CONTROLLER
 188 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 189 #endif
 190
 191 int bond_net_id __read_mostly;
 192
 193 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 194 static int arp_ip_count;
 195 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 196 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 197 static int lacp_fast;
 198
 199 const struct bond_parm_tbl bond_lacp_tbl[] = {
 200 {       "slow",         AD_LACP_SLOW},
 201 {       "fast",         AD_LACP_FAST},
 202 {       NULL,           -1},
 203 };
 204
 205 const struct bond_parm_tbl bond_mode_tbl[] = {
 206 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 207 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 208 {       "balance-xor",          BOND_MODE_XOR},
 209 {       "broadcast",            BOND_MODE_BROADCAST},
 210 {       "802.3ad",              BOND_MODE_8023AD},
 211 {       "balance-tlb",          BOND_MODE_TLB},
 212 {       "balance-alb",          BOND_MODE_ALB},
 213 {       NULL,                   -1},
 214 };
 215
 216 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 217 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 218 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 219 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 220 {       NULL,                   -1},
 221 };
 222
 223 const struct bond_parm_tbl arp_all_targets_tbl[] = {
 224 {       "any",                  BOND_ARP_TARGETS_ANY},
 225 {       "all",                  BOND_ARP_TARGETS_ALL},
 226 {       NULL,                   -1},
 227 };
 228
 229 const struct bond_parm_tbl arp_validate_tbl[] = {
 230 {       "none",                 BOND_ARP_VALIDATE_NONE},
 231 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 232 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 233 {       "all",                  BOND_ARP_VALIDATE_ALL},
 234 {       NULL,                   -1},
 235 };
 236
 237 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 238 {       "none",                 BOND_FOM_NONE},
 239 {       "active",               BOND_FOM_ACTIVE},
 240 {       "follow",               BOND_FOM_FOLLOW},
 241 {       NULL,                   -1},
 242 };
 243
 244 const struct bond_parm_tbl pri_reselect_tbl[] = {
 245 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 246 {       "better",               BOND_PRI_RESELECT_BETTER},
 247 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 248 {       NULL,                   -1},
 249 };
 250
 251 struct bond_parm_tbl ad_select_tbl[] = {
 252 {       "stable",       BOND_AD_STABLE},
 253 {       "bandwidth",    BOND_AD_BANDWIDTH},
 254 {       "count",        BOND_AD_COUNT},
 255 {       NULL,           -1},
 256 };
 257
 258 /*-------------------------- Forward declarations ---------------------------*/
 259
 260 static int bond_init(struct net_device *bond_dev);
 261 static void bond_uninit(struct net_device *bond_dev);
 262
 263 /*---------------------------- General routines -----------------------------*/
 264
 265 const char *bond_mode_name(int mode)
 266 {
 267         static const char *names[] = {
 268                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 269                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 270                 [BOND_MODE_XOR] = "load balancing (xor)",
 271                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 272                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 273                 [BOND_MODE_TLB] = "transmit load balancing",
 274                 [BOND_MODE_ALB] = "adaptive load balancing",
 275         };
 276
 277         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 278                 return "unknown";
 279
 280         return names[mode];
 281 }
 282
 283 /*---------------------------------- VLAN -----------------------------------*/
 284
 285 /**
 286  * bond_dev_queue_xmit - Prepare skb for xmit.
 287  *
 288  * @bond: bond device that got this skb for tx.
 289  * @skb: hw accel VLAN tagged skb to transmit
 290  * @slave_dev: slave that is supposed to xmit this skbuff
 291  */
 292 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 293                         struct net_device *slave_dev)
 294 {
 295         skb->dev = slave_dev;
 296
 297         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 298                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 299         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 300
 301         if (unlikely(netpoll_tx_running(bond->dev)))
 302                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 303         else
 304                 dev_queue_xmit(skb);
 305
 306         return 0;
 307 }
 308
 309 /*
 310  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 311  * We don't protect the slave list iteration with a lock because:
 312  * a. This operation is performed in IOCTL context,
 313  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 314  * c. Holding a lock with BH disabled while directly calling a base driver
 315  *    entry point is generally a BAD idea.
 316  *
 317  * The design of synchronization/protection for this operation in the 8021q
 318  * module is good for one or more VLAN devices over a single physical device
 319  * and cannot be extended for a teaming solution like bonding, so there is a
 320  * potential race condition here where a net device from the vlan group might
 321  * be referenced (either by a base driver or the 8021q code) while it is being
 322  * removed from the system. However, it turns out we're not making matters
 323  * worse, and if it works for regular VLAN usage it will work here too.
 324 */
 325
 326 /**
 327  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 328  * @bond_dev: bonding net device that got called
 329  * @vid: vlan id being added
 330  */
 331 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 332                                 __be16 proto, u16 vid)
 333 {
 334         struct bonding *bond = netdev_priv(bond_dev);
 335         struct slave *slave;
 336         int res;
 337
 338         bond_for_each_slave(bond, slave) {
 339                 res = vlan_vid_add(slave->dev, proto, vid);
 340                 if (res)
 341                         goto unwind;
 342         }
 343
 344         return 0;
 345
 346 unwind:
 347         /* unwind from the slave that failed */
 348         bond_for_each_slave_continue_reverse(bond, slave)
 349                 vlan_vid_del(slave->dev, proto, vid);
 350
 351         return res;
 352 }
 353
 354 /**
 355  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 356  * @bond_dev: bonding net device that got called
 357  * @vid: vlan id being removed
 358  */
 359 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 360                                  __be16 proto, u16 vid)
 361 {
 362         struct bonding *bond = netdev_priv(bond_dev);
 363         struct slave *slave;
 364
 365         bond_for_each_slave(bond, slave)
 366                 vlan_vid_del(slave->dev, proto, vid);
 367
 368         if (bond_is_lb(bond))
 369                 bond_alb_clear_vlan(bond, vid);
 370
 371         return 0;
 372 }
 373
 374 /*------------------------------- Link status -------------------------------*/
 375
 376 /*
 377  * Set the carrier state for the master according to the state of its
 378  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 379  * do special 802.3ad magic.
 380  *
 381  * Returns zero if carrier state does not change, nonzero if it does.
 382  */
 383 static int bond_set_carrier(struct bonding *bond)
 384 {
 385         struct slave *slave;
 386
 387         if (list_empty(&bond->slave_list))
 388                 goto down;
 389
 390         if (bond->params.mode == BOND_MODE_8023AD)
 391                 return bond_3ad_set_carrier(bond);
 392
 393         bond_for_each_slave(bond, slave) {
 394                 if (slave->link == BOND_LINK_UP) {
 395                         if (!netif_carrier_ok(bond->dev)) {
 396                                 netif_carrier_on(bond->dev);
 397                                 return 1;
 398                         }
 399                         return 0;
 400                 }
 401         }
 402
 403 down:
 404         if (netif_carrier_ok(bond->dev)) {
 405                 netif_carrier_off(bond->dev);
 406                 return 1;
 407         }
 408         return 0;
 409 }
 410
 411 /*
 412  * Get link speed and duplex from the slave's base driver
 413  * using ethtool. If for some reason the call fails or the
 414  * values are invalid, set speed and duplex to -1,
 415  * and return.
 416  */
 417 static void bond_update_speed_duplex(struct slave *slave)
 418 {
 419         struct net_device *slave_dev = slave->dev;
 420         struct ethtool_cmd ecmd;
 421         u32 slave_speed;
 422         int res;
 423
 424         slave->speed = SPEED_UNKNOWN;
 425         slave->duplex = DUPLEX_UNKNOWN;
 426
 427         res = __ethtool_get_settings(slave_dev, &ecmd);
 428         if (res < 0)
 429                 return;
 430
 431         slave_speed = ethtool_cmd_speed(&ecmd);
 432         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 433                 return;
 434
 435         switch (ecmd.duplex) {
 436         case DUPLEX_FULL:
 437         case DUPLEX_HALF:
 438                 break;
 439         default:
 440                 return;
 441         }
 442
 443         slave->speed = slave_speed;
 444         slave->duplex = ecmd.duplex;
 445
 446         return;
 447 }
 448
 449 /*
 450  * if <dev> supports MII link status reporting, check its link status.
 451  *
 452  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 453  * depending upon the setting of the use_carrier parameter.
 454  *
 455  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 456  * can't tell and just pretend it is), or 0, meaning that the link is
 457  * down.
 458  *
 459  * If reporting is non-zero, instead of faking link up, return -1 if
 460  * both ETHTOOL and MII ioctls fail (meaning the device does not
 461  * support them).  If use_carrier is set, return whatever it says.
 462  * It'd be nice if there was a good way to tell if a driver supports
 463  * netif_carrier, but there really isn't.
 464  */
 465 static int bond_check_dev_link(struct bonding *bond,
 466                                struct net_device *slave_dev, int reporting)
 467 {
 468         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 469         int (*ioctl)(struct net_device *, struct ifreq *, int);
 470         struct ifreq ifr;
 471         struct mii_ioctl_data *mii;
 472
 473         if (!reporting && !netif_running(slave_dev))
 474                 return 0;
 475
 476         if (bond->params.use_carrier)
 477                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 478
 479         /* Try to get link status using Ethtool first. */
 480         if (slave_dev->ethtool_ops->get_link)
 481                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 482                         BMSR_LSTATUS : 0;
 483
 484         /* Ethtool can't be used, fallback to MII ioctls. */
 485         ioctl = slave_ops->ndo_do_ioctl;
 486         if (ioctl) {
 487                 /* TODO: set pointer to correct ioctl on a per team member */
 488                 /*       bases to make this more efficient. that is, once  */
 489                 /*       we determine the correct ioctl, we will always    */
 490                 /*       call it and not the others for that team          */
 491                 /*       member.                                           */
 492
 493                 /*
 494                  * We cannot assume that SIOCGMIIPHY will also read a
 495                  * register; not all network drivers (e.g., e100)
 496                  * support that.
 497                  */
 498
 499                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 500                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 501                 mii = if_mii(&ifr);
 502                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 503                         mii->reg_num = MII_BMSR;
 504                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 505                                 return mii->val_out & BMSR_LSTATUS;
 506                 }
 507         }
 508
 509         /*
 510          * If reporting, report that either there's no dev->do_ioctl,
 511          * or both SIOCGMIIREG and get_link failed (meaning that we
 512          * cannot report link status).  If not reporting, pretend
 513          * we're ok.
 514          */
 515         return reporting ? -1 : BMSR_LSTATUS;
 516 }
 517
 518 /*----------------------------- Multicast list ------------------------------*/
 519
 520 /*
 521  * Push the promiscuity flag down to appropriate slaves
 522  */
 523 static int bond_set_promiscuity(struct bonding *bond, int inc)
 524 {
 525         int err = 0;
 526         if (USES_PRIMARY(bond->params.mode)) {
 527                 /* write lock already acquired */
 528                 if (bond->curr_active_slave) {
 529                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 530                                                   inc);
 531                 }
 532         } else {
 533                 struct slave *slave;
 534
 535                 bond_for_each_slave(bond, slave) {
 536                         err = dev_set_promiscuity(slave->dev, inc);
 537                         if (err)
 538                                 return err;
 539                 }
 540         }
 541         return err;
 542 }
 543
 544 /*
 545  * Push the allmulti flag down to all slaves
 546  */
 547 static int bond_set_allmulti(struct bonding *bond, int inc)
 548 {
 549         int err = 0;
 550         if (USES_PRIMARY(bond->params.mode)) {
 551                 /* write lock already acquired */
 552                 if (bond->curr_active_slave) {
 553                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 554                                                inc);
 555                 }
 556         } else {
 557                 struct slave *slave;
 558
 559                 bond_for_each_slave(bond, slave) {
 560                         err = dev_set_allmulti(slave->dev, inc);
 561                         if (err)
 562                                 return err;
 563                 }
 564         }
 565         return err;
 566 }
 567
 568 /*
 569  * Retrieve the list of registered multicast addresses for the bonding
 570  * device and retransmit an IGMP JOIN request to the current active
 571  * slave.
 572  */
 573 static void bond_resend_igmp_join_requests(struct bonding *bond)
 574 {
 575         if (!rtnl_trylock()) {
 576                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 577                 return;
 578         }
 579         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 580         rtnl_unlock();
 581
 582         /* We use curr_slave_lock to protect against concurrent access to
 583          * igmp_retrans from multiple running instances of this function and
 584          * bond_change_active_slave
 585          */
 586         write_lock_bh(&bond->curr_slave_lock);
 587         if (bond->igmp_retrans > 1) {
 588                 bond->igmp_retrans--;
 589                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 590         }
 591         write_unlock_bh(&bond->curr_slave_lock);
 592 }
 593
 594 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 595 {
 596         struct bonding *bond = container_of(work, struct bonding,
 597                                             mcast_work.work);
 598
 599         bond_resend_igmp_join_requests(bond);
 600 }
 601
 602 /* Flush bond's hardware addresses from slave
 603  */
 604 static void bond_hw_addr_flush(struct net_device *bond_dev,
 605                                struct net_device *slave_dev)
 606 {
 607         struct bonding *bond = netdev_priv(bond_dev);
 608
 609         dev_uc_unsync(slave_dev, bond_dev);
 610         dev_mc_unsync(slave_dev, bond_dev);
 611
 612         if (bond->params.mode == BOND_MODE_8023AD) {
 613                 /* del lacpdu mc addr from mc list */
 614                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 615
 616                 dev_mc_del(slave_dev, lacpdu_multicast);
 617         }
 618 }
 619
 620 /*--------------------------- Active slave change ---------------------------*/
 621
 622 /* Update the hardware address list and promisc/allmulti for the new and
 623  * old active slaves (if any).  Modes that are !USES_PRIMARY keep all
 624  * slaves up date at all times; only the USES_PRIMARY modes need to call
 625  * this function to swap these settings during a failover.
 626  */
 627 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 628                               struct slave *old_active)
 629 {
 630         ASSERT_RTNL();
 631
 632         if (old_active) {
 633                 if (bond->dev->flags & IFF_PROMISC)
 634                         dev_set_promiscuity(old_active->dev, -1);
 635
 636                 if (bond->dev->flags & IFF_ALLMULTI)
 637                         dev_set_allmulti(old_active->dev, -1);
 638
 639                 bond_hw_addr_flush(bond->dev, old_active->dev);
 640         }
 641
 642         if (new_active) {
 643                 /* FIXME: Signal errors upstream. */
 644                 if (bond->dev->flags & IFF_PROMISC)
 645                         dev_set_promiscuity(new_active->dev, 1);
 646
 647                 if (bond->dev->flags & IFF_ALLMULTI)
 648                         dev_set_allmulti(new_active->dev, 1);
 649
 650                 netif_addr_lock_bh(bond->dev);
 651                 dev_uc_sync(new_active->dev, bond->dev);
 652                 dev_mc_sync(new_active->dev, bond->dev);
 653                 netif_addr_unlock_bh(bond->dev);
 654         }
 655 }
 656
 657 /**
 658  * bond_set_dev_addr - clone slave's address to bond
 659  * @bond_dev: bond net device
 660  * @slave_dev: slave net device
 661  *
 662  * Should be called with RTNL held.
 663  */
 664 static void bond_set_dev_addr(struct net_device *bond_dev,
 665                               struct net_device *slave_dev)
 666 {
 667         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 668                  bond_dev, slave_dev, slave_dev->addr_len);
 669         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 670         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 671         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 672 }
 673
 674 /*
 675  * bond_do_fail_over_mac
 676  *
 677  * Perform special MAC address swapping for fail_over_mac settings
 678  *
 679  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 680  */
 681 static void bond_do_fail_over_mac(struct bonding *bond,
 682                                   struct slave *new_active,
 683                                   struct slave *old_active)
 684         __releases(&bond->curr_slave_lock)
 685         __releases(&bond->lock)
 686         __acquires(&bond->lock)
 687         __acquires(&bond->curr_slave_lock)
 688 {
 689         u8 tmp_mac[ETH_ALEN];
 690         struct sockaddr saddr;
 691         int rv;
 692
 693         switch (bond->params.fail_over_mac) {
 694         case BOND_FOM_ACTIVE:
 695                 if (new_active) {
 696                         write_unlock_bh(&bond->curr_slave_lock);
 697                         read_unlock(&bond->lock);
 698                         bond_set_dev_addr(bond->dev, new_active->dev);
 699                         read_lock(&bond->lock);
 700                         write_lock_bh(&bond->curr_slave_lock);
 701                 }
 702                 break;
 703         case BOND_FOM_FOLLOW:
 704                 /*
 705                  * if new_active && old_active, swap them
 706                  * if just old_active, do nothing (going to no active slave)
 707                  * if just new_active, set new_active to bond's MAC
 708                  */
 709                 if (!new_active)
 710                         return;
 711
 712                 write_unlock_bh(&bond->curr_slave_lock);
 713                 read_unlock(&bond->lock);
 714
 715                 if (old_active) {
 716                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 717                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 718                                ETH_ALEN);
 719                         saddr.sa_family = new_active->dev->type;
 720                 } else {
 721                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 722                         saddr.sa_family = bond->dev->type;
 723                 }
 724
 725                 rv = dev_set_mac_address(new_active->dev, &saddr);
 726                 if (rv) {
 727                         pr_err("%s: Error %d setting MAC of slave %s\n",
 728                                bond->dev->name, -rv, new_active->dev->name);
 729                         goto out;
 730                 }
 731
 732                 if (!old_active)
 733                         goto out;
 734
 735                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 736                 saddr.sa_family = old_active->dev->type;
 737
 738                 rv = dev_set_mac_address(old_active->dev, &saddr);
 739                 if (rv)
 740                         pr_err("%s: Error %d setting MAC of slave %s\n",
 741                                bond->dev->name, -rv, new_active->dev->name);
 742 out:
 743                 read_lock(&bond->lock);
 744                 write_lock_bh(&bond->curr_slave_lock);
 745                 break;
 746         default:
 747                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 748                        bond->dev->name, bond->params.fail_over_mac);
 749                 break;
 750         }
 751
 752 }
 753
 754 static bool bond_should_change_active(struct bonding *bond)
 755 {
 756         struct slave *prim = bond->primary_slave;
 757         struct slave *curr = bond->curr_active_slave;
 758
 759         if (!prim || !curr || curr->link != BOND_LINK_UP)
 760                 return true;
 761         if (bond->force_primary) {
 762                 bond->force_primary = false;
 763                 return true;
 764         }
 765         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 766             (prim->speed < curr->speed ||
 767              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 768                 return false;
 769         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 770                 return false;
 771         return true;
 772 }
 773
 774 /**
 775  * find_best_interface - select the best available slave to be the active one
 776  * @bond: our bonding struct
 777  *
 778  * Warning: Caller must hold curr_slave_lock for writing.
 779  */
 780 static struct slave *bond_find_best_slave(struct bonding *bond)
 781 {
 782         struct slave *new_active, *old_active;
 783         struct slave *bestslave = NULL;
 784         int mintime = bond->params.updelay;
 785         int i;
 786
 787         new_active = bond->curr_active_slave;
 788
 789         if (!new_active) { /* there were no active slaves left */
 790                 new_active = bond_first_slave(bond);
 791                 if (!new_active)
 792                         return NULL; /* still no slave, return NULL */
 793         }
 794
 795         if ((bond->primary_slave) &&
 796             bond->primary_slave->link == BOND_LINK_UP &&
 797             bond_should_change_active(bond)) {
 798                 new_active = bond->primary_slave;
 799         }
 800
 801         /* remember where to stop iterating over the slaves */
 802         old_active = new_active;
 803
 804         bond_for_each_slave_from(bond, new_active, i, old_active) {
 805                 if (new_active->link == BOND_LINK_UP) {
 806                         return new_active;
 807                 } else if (new_active->link == BOND_LINK_BACK &&
 808                            IS_UP(new_active->dev)) {
 809                         /* link up, but waiting for stabilization */
 810                         if (new_active->delay < mintime) {
 811                                 mintime = new_active->delay;
 812                                 bestslave = new_active;
 813                         }
 814                 }
 815         }
 816
 817         return bestslave;
 818 }
 819
 820 static bool bond_should_notify_peers(struct bonding *bond)
 821 {
 822         struct slave *slave = bond->curr_active_slave;
 823
 824         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
 825                  bond->dev->name, slave ? slave->dev->name : "NULL");
 826
 827         if (!slave || !bond->send_peer_notif ||
 828             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 829                 return false;
 830
 831         return true;
 832 }
 833
 834 /**
 835  * change_active_interface - change the active slave into the specified one
 836  * @bond: our bonding struct
 837  * @new: the new slave to make the active one
 838  *
 839  * Set the new slave to the bond's settings and unset them on the old
 840  * curr_active_slave.
 841  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 842  *
 843  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 844  * because it is apparently the best available slave we have, even though its
 845  * updelay hasn't timed out yet.
 846  *
 847  * If new_active is not NULL, caller must hold bond->lock for read and
 848  * curr_slave_lock for write_bh.
 849  */
 850 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 851 {
 852         struct slave *old_active = bond->curr_active_slave;
 853
 854         if (old_active == new_active)
 855                 return;
 856
 857         if (new_active) {
 858                 new_active->jiffies = jiffies;
 859
 860                 if (new_active->link == BOND_LINK_BACK) {
 861                         if (USES_PRIMARY(bond->params.mode)) {
 862                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
 863                                         bond->dev->name, new_active->dev->name,
 864                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
 865                         }
 866
 867                         new_active->delay = 0;
 868                         new_active->link = BOND_LINK_UP;
 869
 870                         if (bond->params.mode == BOND_MODE_8023AD)
 871                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 872
 873                         if (bond_is_lb(bond))
 874                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 875                 } else {
 876                         if (USES_PRIMARY(bond->params.mode)) {
 877                                 pr_info("%s: making interface %s the new active one.\n",
 878                                         bond->dev->name, new_active->dev->name);
 879                         }
 880                 }
 881         }
 882
 883         if (USES_PRIMARY(bond->params.mode))
 884                 bond_hw_addr_swap(bond, new_active, old_active);
 885
 886         if (bond_is_lb(bond)) {
 887                 bond_alb_handle_active_change(bond, new_active);
 888                 if (old_active)
 889                         bond_set_slave_inactive_flags(old_active);
 890                 if (new_active)
 891                         bond_set_slave_active_flags(new_active);
 892         } else {
 893                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 894         }
 895
 896         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
 897                 if (old_active)
 898                         bond_set_slave_inactive_flags(old_active);
 899
 900                 if (new_active) {
 901                         bool should_notify_peers = false;
 902
 903                         bond_set_slave_active_flags(new_active);
 904
 905                         if (bond->params.fail_over_mac)
 906                                 bond_do_fail_over_mac(bond, new_active,
 907                                                       old_active);
 908
 909                         if (netif_running(bond->dev)) {
 910                                 bond->send_peer_notif =
 911                                         bond->params.num_peer_notif;
 912                                 should_notify_peers =
 913                                         bond_should_notify_peers(bond);
 914                         }
 915
 916                         write_unlock_bh(&bond->curr_slave_lock);
 917                         read_unlock(&bond->lock);
 918
 919                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 920                         if (should_notify_peers)
 921                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 922                                                          bond->dev);
 923
 924                         read_lock(&bond->lock);
 925                         write_lock_bh(&bond->curr_slave_lock);
 926                 }
 927         }
 928
 929         /* resend IGMP joins since active slave has changed or
 930          * all were sent on curr_active_slave.
 931          * resend only if bond is brought up with the affected
 932          * bonding modes and the retransmission is enabled */
 933         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 934             ((USES_PRIMARY(bond->params.mode) && new_active) ||
 935              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
 936                 bond->igmp_retrans = bond->params.resend_igmp;
 937                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 938         }
 939 }
 940
 941 /**
 942  * bond_select_active_slave - select a new active slave, if needed
 943  * @bond: our bonding struct
 944  *
 945  * This functions should be called when one of the following occurs:
 946  * - The old curr_active_slave has been released or lost its link.
 947  * - The primary_slave has got its link back.
 948  * - A slave has got its link back and there's no old curr_active_slave.
 949  *
 950  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
 951  */
 952 void bond_select_active_slave(struct bonding *bond)
 953 {
 954         struct slave *best_slave;
 955         int rv;
 956
 957         best_slave = bond_find_best_slave(bond);
 958         if (best_slave != bond->curr_active_slave) {
 959                 bond_change_active_slave(bond, best_slave);
 960                 rv = bond_set_carrier(bond);
 961                 if (!rv)
 962                         return;
 963
 964                 if (netif_carrier_ok(bond->dev)) {
 965                         pr_info("%s: first active interface up!\n",
 966                                 bond->dev->name);
 967                 } else {
 968                         pr_info("%s: now running without any active interface !\n",
 969                                 bond->dev->name);
 970                 }
 971         }
 972 }
 973
 974 /*--------------------------- slave list handling ---------------------------*/
 975
 976 /*
 977  * This function attaches the slave to the end of list.
 978  *
 979  * bond->lock held for writing by caller.
 980  */
 981 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
 982 {
 983         list_add_tail_rcu(&new_slave->list, &bond->slave_list);
 984         bond->slave_cnt++;
 985 }
 986
 987 /*
 988  * This function detaches the slave from the list.
 989  * WARNING: no check is made to verify if the slave effectively
 990  * belongs to <bond>.
 991  * Nothing is freed on return, structures are just unchained.
 992  * If any slave pointer in bond was pointing to <slave>,
 993  * it should be changed by the calling function.
 994  *
 995  * bond->lock held for writing by caller.
 996  */
 997 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
 998 {
 999         list_del_rcu(&slave->list);
1000         bond->slave_cnt--;
1001 }
1002
1003 #ifdef CONFIG_NET_POLL_CONTROLLER
1004 static inline int slave_enable_netpoll(struct slave *slave)
1005 {
1006         struct netpoll *np;
1007         int err = 0;
1008
1009         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1010         err = -ENOMEM;
1011         if (!np)
1012                 goto out;
1013
1014         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1015         if (err) {
1016                 kfree(np);
1017                 goto out;
1018         }
1019         slave->np = np;
1020 out:
1021         return err;
1022 }
1023 static inline void slave_disable_netpoll(struct slave *slave)
1024 {
1025         struct netpoll *np = slave->np;
1026
1027         if (!np)
1028                 return;
1029
1030         slave->np = NULL;
1031         __netpoll_free_async(np);
1032 }
1033 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1034 {
1035         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1036                 return false;
1037         if (!slave_dev->netdev_ops->ndo_poll_controller)
1038                 return false;
1039         return true;
1040 }
1041
1042 static void bond_poll_controller(struct net_device *bond_dev)
1043 {
1044 }
1045
1046 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1047 {
1048         struct bonding *bond = netdev_priv(bond_dev);
1049         struct slave *slave;
1050
1051         bond_for_each_slave(bond, slave)
1052                 if (IS_UP(slave->dev))
1053                         slave_disable_netpoll(slave);
1054 }
1055
1056 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1057 {
1058         struct bonding *bond = netdev_priv(dev);
1059         struct slave *slave;
1060         int err = 0;
1061
1062         bond_for_each_slave(bond, slave) {
1063                 err = slave_enable_netpoll(slave);
1064                 if (err) {
1065                         bond_netpoll_cleanup(dev);
1066                         break;
1067                 }
1068         }
1069         return err;
1070 }
1071 #else
1072 static inline int slave_enable_netpoll(struct slave *slave)
1073 {
1074         return 0;
1075 }
1076 static inline void slave_disable_netpoll(struct slave *slave)
1077 {
1078 }
1079 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1080 {
1081 }
1082 #endif
1083
1084 /*---------------------------------- IOCTL ----------------------------------*/
1085
1086 static netdev_features_t bond_fix_features(struct net_device *dev,
1087                                            netdev_features_t features)
1088 {
1089         struct bonding *bond = netdev_priv(dev);
1090         netdev_features_t mask;
1091         struct slave *slave;
1092
1093         if (list_empty(&bond->slave_list)) {
1094                 /* Disable adding VLANs to empty bond. But why? --mq */
1095                 features |= NETIF_F_VLAN_CHALLENGED;
1096                 return features;
1097         }
1098
1099         mask = features;
1100         features &= ~NETIF_F_ONE_FOR_ALL;
1101         features |= NETIF_F_ALL_FOR_ALL;
1102
1103         bond_for_each_slave(bond, slave) {
1104                 features = netdev_increment_features(features,
1105                                                      slave->dev->features,
1106                                                      mask);
1107         }
1108         features = netdev_add_tso_features(features, mask);
1109
1110         return features;
1111 }
1112
1113 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1114                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1115                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1116
1117 static void bond_compute_features(struct bonding *bond)
1118 {
1119         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1120         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1121         unsigned short max_hard_header_len = ETH_HLEN;
1122         unsigned int gso_max_size = GSO_MAX_SIZE;
1123         struct net_device *bond_dev = bond->dev;
1124         u16 gso_max_segs = GSO_MAX_SEGS;
1125         struct slave *slave;
1126
1127         if (list_empty(&bond->slave_list))
1128                 goto done;
1129
1130         bond_for_each_slave(bond, slave) {
1131                 vlan_features = netdev_increment_features(vlan_features,
1132                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1133
1134                 dst_release_flag &= slave->dev->priv_flags;
1135                 if (slave->dev->hard_header_len > max_hard_header_len)
1136                         max_hard_header_len = slave->dev->hard_header_len;
1137
1138                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1139                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1140         }
1141
1142 done:
1143         bond_dev->vlan_features = vlan_features;
1144         bond_dev->hard_header_len = max_hard_header_len;
1145         bond_dev->gso_max_segs = gso_max_segs;
1146         netif_set_gso_max_size(bond_dev, gso_max_size);
1147
1148         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1149         bond_dev->priv_flags = flags | dst_release_flag;
1150
1151         netdev_change_features(bond_dev);
1152 }
1153
1154 static void bond_setup_by_slave(struct net_device *bond_dev,
1155                                 struct net_device *slave_dev)
1156 {
1157         bond_dev->header_ops        = slave_dev->header_ops;
1158
1159         bond_dev->type              = slave_dev->type;
1160         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1161         bond_dev->addr_len          = slave_dev->addr_len;
1162
1163         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1164                 slave_dev->addr_len);
1165 }
1166
1167 /* On bonding slaves other than the currently active slave, suppress
1168  * duplicates except for alb non-mcast/bcast.
1169  */
1170 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1171                                             struct slave *slave,
1172                                             struct bonding *bond)
1173 {
1174         if (bond_is_slave_inactive(slave)) {
1175                 if (bond->params.mode == BOND_MODE_ALB &&
1176                     skb->pkt_type != PACKET_BROADCAST &&
1177                     skb->pkt_type != PACKET_MULTICAST)
1178                         return false;
1179                 return true;
1180         }
1181         return false;
1182 }
1183
1184 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1185 {
1186         struct sk_buff *skb = *pskb;
1187         struct slave *slave;
1188         struct bonding *bond;
1189         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1190                           struct slave *);
1191         int ret = RX_HANDLER_ANOTHER;
1192
1193         skb = skb_share_check(skb, GFP_ATOMIC);
1194         if (unlikely(!skb))
1195                 return RX_HANDLER_CONSUMED;
1196
1197         *pskb = skb;
1198
1199         slave = bond_slave_get_rcu(skb->dev);
1200         bond = slave->bond;
1201
1202         if (bond->params.arp_interval)
1203                 slave->dev->last_rx = jiffies;
1204
1205         recv_probe = ACCESS_ONCE(bond->recv_probe);
1206         if (recv_probe) {
1207                 ret = recv_probe(skb, bond, slave);
1208                 if (ret == RX_HANDLER_CONSUMED) {
1209                         consume_skb(skb);
1210                         return ret;
1211                 }
1212         }
1213
1214         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1215                 return RX_HANDLER_EXACT;
1216         }
1217
1218         skb->dev = bond->dev;
1219
1220         if (bond->params.mode == BOND_MODE_ALB &&
1221             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1222             skb->pkt_type == PACKET_HOST) {
1223
1224                 if (unlikely(skb_cow_head(skb,
1225                                           skb->data - skb_mac_header(skb)))) {
1226                         kfree_skb(skb);
1227                         return RX_HANDLER_CONSUMED;
1228                 }
1229                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1230         }
1231
1232         return ret;
1233 }
1234
1235 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1236                                       struct net_device *slave_dev)
1237 {
1238         int err;
1239
1240         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1241         if (err)
1242                 return err;
1243         slave_dev->flags |= IFF_SLAVE;
1244         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1245         return 0;
1246 }
1247
1248 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1249                                   struct net_device *slave_dev)
1250 {
1251         netdev_upper_dev_unlink(slave_dev, bond_dev);
1252         slave_dev->flags &= ~IFF_SLAVE;
1253         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1254 }
1255
1256 /* enslave device <slave> to bond device <master> */
1257 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1258 {
1259         struct bonding *bond = netdev_priv(bond_dev);
1260         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1261         struct slave *new_slave = NULL;
1262         struct sockaddr addr;
1263         int link_reporting;
1264         int res = 0, i;
1265
1266         if (!bond->params.use_carrier &&
1267             slave_dev->ethtool_ops->get_link == NULL &&
1268             slave_ops->ndo_do_ioctl == NULL) {
1269                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1270                            bond_dev->name, slave_dev->name);
1271         }
1272
1273         /* already enslaved */
1274         if (slave_dev->flags & IFF_SLAVE) {
1275                 pr_debug("Error, Device was already enslaved\n");
1276                 return -EBUSY;
1277         }
1278
1279         /* vlan challenged mutual exclusion */
1280         /* no need to lock since we're protected by rtnl_lock */
1281         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1282                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1283                 if (vlan_uses_dev(bond_dev)) {
1284                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1285                                bond_dev->name, slave_dev->name, bond_dev->name);
1286                         return -EPERM;
1287                 } else {
1288                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1289                                    bond_dev->name, slave_dev->name,
1290                                    slave_dev->name, bond_dev->name);
1291                 }
1292         } else {
1293                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1294         }
1295
1296         /*
1297          * Old ifenslave binaries are no longer supported.  These can
1298          * be identified with moderate accuracy by the state of the slave:
1299          * the current ifenslave will set the interface down prior to
1300          * enslaving it; the old ifenslave will not.
1301          */
1302         if ((slave_dev->flags & IFF_UP)) {
1303                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1304                        slave_dev->name);
1305                 res = -EPERM;
1306                 goto err_undo_flags;
1307         }
1308
1309         /* set bonding device ether type by slave - bonding netdevices are
1310          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1311          * there is a need to override some of the type dependent attribs/funcs.
1312          *
1313          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1314          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1315          */
1316         if (list_empty(&bond->slave_list)) {
1317                 if (bond_dev->type != slave_dev->type) {
1318                         pr_debug("%s: change device type from %d to %d\n",
1319                                  bond_dev->name,
1320                                  bond_dev->type, slave_dev->type);
1321
1322                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1323                                                        bond_dev);
1324                         res = notifier_to_errno(res);
1325                         if (res) {
1326                                 pr_err("%s: refused to change device type\n",
1327                                        bond_dev->name);
1328                                 res = -EBUSY;
1329                                 goto err_undo_flags;
1330                         }
1331
1332                         /* Flush unicast and multicast addresses */
1333                         dev_uc_flush(bond_dev);
1334                         dev_mc_flush(bond_dev);
1335
1336                         if (slave_dev->type != ARPHRD_ETHER)
1337                                 bond_setup_by_slave(bond_dev, slave_dev);
1338                         else {
1339                                 ether_setup(bond_dev);
1340                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1341                         }
1342
1343                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1344                                                  bond_dev);
1345                 }
1346         } else if (bond_dev->type != slave_dev->type) {
1347                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1348                        slave_dev->name,
1349                        slave_dev->type, bond_dev->type);
1350                 res = -EINVAL;
1351                 goto err_undo_flags;
1352         }
1353
1354         if (slave_ops->ndo_set_mac_address == NULL) {
1355                 if (list_empty(&bond->slave_list)) {
1356                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1357                                    bond_dev->name);
1358                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1359                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1360                         pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1361                                bond_dev->name);
1362                         res = -EOPNOTSUPP;
1363                         goto err_undo_flags;
1364                 }
1365         }
1366
1367         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1368
1369         /* If this is the first slave, then we need to set the master's hardware
1370          * address to be the same as the slave's. */
1371         if (list_empty(&bond->slave_list) &&
1372             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1373                 bond_set_dev_addr(bond->dev, slave_dev);
1374
1375         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1376         if (!new_slave) {
1377                 res = -ENOMEM;
1378                 goto err_undo_flags;
1379         }
1380         INIT_LIST_HEAD(&new_slave->list);
1381         /*
1382          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1383          * is set via sysfs or module option if desired.
1384          */
1385         new_slave->queue_id = 0;
1386
1387         /* Save slave's original mtu and then set it to match the bond */
1388         new_slave->original_mtu = slave_dev->mtu;
1389         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1390         if (res) {
1391                 pr_debug("Error %d calling dev_set_mtu\n", res);
1392                 goto err_free;
1393         }
1394
1395         /*
1396          * Save slave's original ("permanent") mac address for modes
1397          * that need it, and for restoring it upon release, and then
1398          * set it to the master's address
1399          */
1400         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1401
1402         if (!bond->params.fail_over_mac) {
1403                 /*
1404                  * Set slave to master's mac address.  The application already
1405                  * set the master's mac address to that of the first slave
1406                  */
1407                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1408                 addr.sa_family = slave_dev->type;
1409                 res = dev_set_mac_address(slave_dev, &addr);
1410                 if (res) {
1411                         pr_debug("Error %d calling set_mac_address\n", res);
1412                         goto err_restore_mtu;
1413                 }
1414         }
1415
1416         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1417         if (res) {
1418                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1419                 goto err_restore_mac;
1420         }
1421
1422         /* open the slave since the application closed it */
1423         res = dev_open(slave_dev);
1424         if (res) {
1425                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1426                 goto err_unset_master;
1427         }
1428
1429         new_slave->bond = bond;
1430         new_slave->dev = slave_dev;
1431         slave_dev->priv_flags |= IFF_BONDING;
1432
1433         if (bond_is_lb(bond)) {
1434                 /* bond_alb_init_slave() must be called before all other stages since
1435                  * it might fail and we do not want to have to undo everything
1436                  */
1437                 res = bond_alb_init_slave(bond, new_slave);
1438                 if (res)
1439                         goto err_close;
1440         }
1441
1442         /* If the mode USES_PRIMARY, then the following is handled by
1443          * bond_change_active_slave().
1444          */
1445         if (!USES_PRIMARY(bond->params.mode)) {
1446                 /* set promiscuity level to new slave */
1447                 if (bond_dev->flags & IFF_PROMISC) {
1448                         res = dev_set_promiscuity(slave_dev, 1);
1449                         if (res)
1450                                 goto err_close;
1451                 }
1452
1453                 /* set allmulti level to new slave */
1454                 if (bond_dev->flags & IFF_ALLMULTI) {
1455                         res = dev_set_allmulti(slave_dev, 1);
1456                         if (res)
1457                                 goto err_close;
1458                 }
1459
1460                 netif_addr_lock_bh(bond_dev);
1461
1462                 dev_mc_sync_multiple(slave_dev, bond_dev);
1463                 dev_uc_sync_multiple(slave_dev, bond_dev);
1464
1465                 netif_addr_unlock_bh(bond_dev);
1466         }
1467
1468         if (bond->params.mode == BOND_MODE_8023AD) {
1469                 /* add lacpdu mc addr to mc list */
1470                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1471
1472                 dev_mc_add(slave_dev, lacpdu_multicast);
1473         }
1474
1475         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1476         if (res) {
1477                 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1478                        bond_dev->name, slave_dev->name);
1479                 goto err_close;
1480         }
1481
1482         write_lock_bh(&bond->lock);
1483
1484         bond_attach_slave(bond, new_slave);
1485
1486         new_slave->delay = 0;
1487         new_slave->link_failure_count = 0;
1488
1489         write_unlock_bh(&bond->lock);
1490
1491         bond_compute_features(bond);
1492
1493         bond_update_speed_duplex(new_slave);
1494
1495         read_lock(&bond->lock);
1496
1497         new_slave->last_arp_rx = jiffies -
1498                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1499         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1500                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1501
1502         if (bond->params.miimon && !bond->params.use_carrier) {
1503                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1504
1505                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1506                         /*
1507                          * miimon is set but a bonded network driver
1508                          * does not support ETHTOOL/MII and
1509                          * arp_interval is not set.  Note: if
1510                          * use_carrier is enabled, we will never go
1511                          * here (because netif_carrier is always
1512                          * supported); thus, we don't need to change
1513                          * the messages for netif_carrier.
1514                          */
1515                         pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1516                                bond_dev->name, slave_dev->name);
1517                 } else if (link_reporting == -1) {
1518                         /* unable get link status using mii/ethtool */
1519                         pr_warning("%s: Warning: can't get link status from interface %s; 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",
1520                                    bond_dev->name, slave_dev->name);
1521                 }
1522         }
1523
1524         /* check for initial state */
1525         if (bond->params.miimon) {
1526                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1527                         if (bond->params.updelay) {
1528                                 new_slave->link = BOND_LINK_BACK;
1529                                 new_slave->delay = bond->params.updelay;
1530                         } else {
1531                                 new_slave->link = BOND_LINK_UP;
1532                         }
1533                 } else {
1534                         new_slave->link = BOND_LINK_DOWN;
1535                 }
1536         } else if (bond->params.arp_interval) {
1537                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1538                         BOND_LINK_UP : BOND_LINK_DOWN);
1539         } else {
1540                 new_slave->link = BOND_LINK_UP;
1541         }
1542
1543         if (new_slave->link != BOND_LINK_DOWN)
1544                 new_slave->jiffies = jiffies;
1545         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1546                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1547                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1548
1549         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1550                 /* if there is a primary slave, remember it */
1551                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1552                         bond->primary_slave = new_slave;
1553                         bond->force_primary = true;
1554                 }
1555         }
1556
1557         write_lock_bh(&bond->curr_slave_lock);
1558
1559         switch (bond->params.mode) {
1560         case BOND_MODE_ACTIVEBACKUP:
1561                 bond_set_slave_inactive_flags(new_slave);
1562                 bond_select_active_slave(bond);
1563                 break;
1564         case BOND_MODE_8023AD:
1565                 /* in 802.3ad mode, the internal mechanism
1566                  * will activate the slaves in the selected
1567                  * aggregator
1568                  */
1569                 bond_set_slave_inactive_flags(new_slave);
1570                 /* if this is the first slave */
1571                 if (bond_first_slave(bond) == new_slave) {
1572                         SLAVE_AD_INFO(new_slave).id = 1;
1573                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1574                          * can be called only after the mac address of the bond is set
1575                          */
1576                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1577                 } else {
1578                         struct slave *prev_slave;
1579
1580                         prev_slave = bond_prev_slave(bond, new_slave);
1581                         SLAVE_AD_INFO(new_slave).id =
1582                                 SLAVE_AD_INFO(prev_slave).id + 1;
1583                 }
1584
1585                 bond_3ad_bind_slave(new_slave);
1586                 break;
1587         case BOND_MODE_TLB:
1588         case BOND_MODE_ALB:
1589                 bond_set_active_slave(new_slave);
1590                 bond_set_slave_inactive_flags(new_slave);
1591                 bond_select_active_slave(bond);
1592                 break;
1593         default:
1594                 pr_debug("This slave is always active in trunk mode\n");
1595
1596                 /* always active in trunk mode */
1597                 bond_set_active_slave(new_slave);
1598
1599                 /* In trunking mode there is little meaning to curr_active_slave
1600                  * anyway (it holds no special properties of the bond device),
1601                  * so we can change it without calling change_active_interface()
1602                  */
1603                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1604                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1605
1606                 break;
1607         } /* switch(bond_mode) */
1608
1609         write_unlock_bh(&bond->curr_slave_lock);
1610
1611         bond_set_carrier(bond);
1612
1613 #ifdef CONFIG_NET_POLL_CONTROLLER
1614         slave_dev->npinfo = bond->dev->npinfo;
1615         if (slave_dev->npinfo) {
1616                 if (slave_enable_netpoll(new_slave)) {
1617                         read_unlock(&bond->lock);
1618                         pr_info("Error, %s: master_dev is using netpoll, "
1619                                  "but new slave device does not support netpoll.\n",
1620                                  bond_dev->name);
1621                         res = -EBUSY;
1622                         goto err_detach;
1623                 }
1624         }
1625 #endif
1626
1627         read_unlock(&bond->lock);
1628
1629         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1630         if (res)
1631                 goto err_detach;
1632
1633         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1634                                          new_slave);
1635         if (res) {
1636                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1637                 goto err_dest_symlinks;
1638         }
1639
1640         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1641                 bond_dev->name, slave_dev->name,
1642                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1643                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1644
1645         /* enslave is successful */
1646         return 0;
1647
1648 /* Undo stages on error */
1649 err_dest_symlinks:
1650         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1651
1652 err_detach:
1653         if (!USES_PRIMARY(bond->params.mode))
1654                 bond_hw_addr_flush(bond_dev, slave_dev);
1655
1656         vlan_vids_del_by_dev(slave_dev, bond_dev);
1657         write_lock_bh(&bond->lock);
1658         bond_detach_slave(bond, new_slave);
1659         if (bond->primary_slave == new_slave)
1660                 bond->primary_slave = NULL;
1661         if (bond->curr_active_slave == new_slave) {
1662                 bond_change_active_slave(bond, NULL);
1663                 write_unlock_bh(&bond->lock);
1664                 read_lock(&bond->lock);
1665                 write_lock_bh(&bond->curr_slave_lock);
1666                 bond_select_active_slave(bond);
1667                 write_unlock_bh(&bond->curr_slave_lock);
1668                 read_unlock(&bond->lock);
1669         } else {
1670                 write_unlock_bh(&bond->lock);
1671         }
1672         slave_disable_netpoll(new_slave);
1673
1674 err_close:
1675         slave_dev->priv_flags &= ~IFF_BONDING;
1676         dev_close(slave_dev);
1677
1678 err_unset_master:
1679         bond_upper_dev_unlink(bond_dev, slave_dev);
1680
1681 err_restore_mac:
1682         if (!bond->params.fail_over_mac) {
1683                 /* XXX TODO - fom follow mode needs to change master's
1684                  * MAC if this slave's MAC is in use by the bond, or at
1685                  * least print a warning.
1686                  */
1687                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1688                 addr.sa_family = slave_dev->type;
1689                 dev_set_mac_address(slave_dev, &addr);
1690         }
1691
1692 err_restore_mtu:
1693         dev_set_mtu(slave_dev, new_slave->original_mtu);
1694
1695 err_free:
1696         kfree(new_slave);
1697
1698 err_undo_flags:
1699         bond_compute_features(bond);
1700         /* Enslave of first slave has failed and we need to fix master's mac */
1701         if (list_empty(&bond->slave_list) &&
1702             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1703                 eth_hw_addr_random(bond_dev);
1704
1705         return res;
1706 }
1707
1708 /*
1709  * Try to release the slave device <slave> from the bond device <master>
1710  * It is legal to access curr_active_slave without a lock because all the function
1711  * is write-locked. If "all" is true it means that the function is being called
1712  * while destroying a bond interface and all slaves are being released.
1713  *
1714  * The rules for slave state should be:
1715  *   for Active/Backup:
1716  *     Active stays on all backups go down
1717  *   for Bonded connections:
1718  *     The first up interface should be left on and all others downed.
1719  */
1720 static int __bond_release_one(struct net_device *bond_dev,
1721                               struct net_device *slave_dev,
1722                               bool all)
1723 {
1724         struct bonding *bond = netdev_priv(bond_dev);
1725         struct slave *slave, *oldcurrent;
1726         struct sockaddr addr;
1727         int old_flags = bond_dev->flags;
1728         netdev_features_t old_features = bond_dev->features;
1729
1730         /* slave is not a slave or master is not master of this slave */
1731         if (!(slave_dev->flags & IFF_SLAVE) ||
1732             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1733                 pr_err("%s: Error: cannot release %s.\n",
1734                        bond_dev->name, slave_dev->name);
1735                 return -EINVAL;
1736         }
1737
1738         block_netpoll_tx();
1739         write_lock_bh(&bond->lock);
1740
1741         slave = bond_get_slave_by_dev(bond, slave_dev);
1742         if (!slave) {
1743                 /* not a slave of this bond */
1744                 pr_info("%s: %s not enslaved\n",
1745                         bond_dev->name, slave_dev->name);
1746                 write_unlock_bh(&bond->lock);
1747                 unblock_netpoll_tx();
1748                 return -EINVAL;
1749         }
1750
1751         write_unlock_bh(&bond->lock);
1752         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1753          * for this slave anymore.
1754          */
1755         netdev_rx_handler_unregister(slave_dev);
1756         write_lock_bh(&bond->lock);
1757
1758         /* Inform AD package of unbinding of slave. */
1759         if (bond->params.mode == BOND_MODE_8023AD) {
1760                 /* must be called before the slave is
1761                  * detached from the list
1762                  */
1763                 bond_3ad_unbind_slave(slave);
1764         }
1765
1766         pr_info("%s: releasing %s interface %s\n",
1767                 bond_dev->name,
1768                 bond_is_active_slave(slave) ? "active" : "backup",
1769                 slave_dev->name);
1770
1771         oldcurrent = bond->curr_active_slave;
1772
1773         bond->current_arp_slave = NULL;
1774
1775         /* release the slave from its bond */
1776         bond_detach_slave(bond, slave);
1777
1778         if (!all && !bond->params.fail_over_mac) {
1779                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1780                     !list_empty(&bond->slave_list))
1781                         pr_warn("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1782                                    bond_dev->name, slave_dev->name,
1783                                    slave->perm_hwaddr,
1784                                    bond_dev->name, slave_dev->name);
1785         }
1786
1787         if (bond->primary_slave == slave)
1788                 bond->primary_slave = NULL;
1789
1790         if (oldcurrent == slave)
1791                 bond_change_active_slave(bond, NULL);
1792
1793         if (bond_is_lb(bond)) {
1794                 /* Must be called only after the slave has been
1795                  * detached from the list and the curr_active_slave
1796                  * has been cleared (if our_slave == old_current),
1797                  * but before a new active slave is selected.
1798                  */
1799                 write_unlock_bh(&bond->lock);
1800                 bond_alb_deinit_slave(bond, slave);
1801                 write_lock_bh(&bond->lock);
1802         }
1803
1804         if (all) {
1805                 rcu_assign_pointer(bond->curr_active_slave, NULL);
1806         } else if (oldcurrent == slave) {
1807                 /*
1808                  * Note that we hold RTNL over this sequence, so there
1809                  * is no concern that another slave add/remove event
1810                  * will interfere.
1811                  */
1812                 write_unlock_bh(&bond->lock);
1813                 read_lock(&bond->lock);
1814                 write_lock_bh(&bond->curr_slave_lock);
1815
1816                 bond_select_active_slave(bond);
1817
1818                 write_unlock_bh(&bond->curr_slave_lock);
1819                 read_unlock(&bond->lock);
1820                 write_lock_bh(&bond->lock);
1821         }
1822
1823         if (list_empty(&bond->slave_list)) {
1824                 bond_set_carrier(bond);
1825                 eth_hw_addr_random(bond_dev);
1826
1827                 if (vlan_uses_dev(bond_dev)) {
1828                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1829                                    bond_dev->name, bond_dev->name);
1830                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1831                                    bond_dev->name);
1832                 }
1833         }
1834
1835         write_unlock_bh(&bond->lock);
1836         unblock_netpoll_tx();
1837         synchronize_rcu();
1838
1839         if (list_empty(&bond->slave_list)) {
1840                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1841                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1842         }
1843
1844         bond_compute_features(bond);
1845         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1846             (old_features & NETIF_F_VLAN_CHALLENGED))
1847                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1848                         bond_dev->name, slave_dev->name, bond_dev->name);
1849
1850         /* must do this from outside any spinlocks */
1851         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1852
1853         vlan_vids_del_by_dev(slave_dev, bond_dev);
1854
1855         /* If the mode USES_PRIMARY, then this cases was handled above by
1856          * bond_change_active_slave(..., NULL)
1857          */
1858         if (!USES_PRIMARY(bond->params.mode)) {
1859                 /* unset promiscuity level from slave
1860                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1861                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1862                  * value of that flag before that change, as that was the value
1863                  * when this slave was attached, so we cache at the start of the
1864                  * function and use it here. Same goes for ALLMULTI below
1865                  */
1866                 if (old_flags & IFF_PROMISC)
1867                         dev_set_promiscuity(slave_dev, -1);
1868
1869                 /* unset allmulti level from slave */
1870                 if (old_flags & IFF_ALLMULTI)
1871                         dev_set_allmulti(slave_dev, -1);
1872
1873                 bond_hw_addr_flush(bond_dev, slave_dev);
1874         }
1875
1876         bond_upper_dev_unlink(bond_dev, slave_dev);
1877
1878         slave_disable_netpoll(slave);
1879
1880         /* close slave before restoring its mac address */
1881         dev_close(slave_dev);
1882
1883         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1884                 /* restore original ("permanent") mac address */
1885                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1886                 addr.sa_family = slave_dev->type;
1887                 dev_set_mac_address(slave_dev, &addr);
1888         }
1889
1890         dev_set_mtu(slave_dev, slave->original_mtu);
1891
1892         slave_dev->priv_flags &= ~IFF_BONDING;
1893
1894         kfree(slave);
1895
1896         return 0;  /* deletion OK */
1897 }
1898
1899 /* A wrapper used because of ndo_del_link */
1900 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1901 {
1902         return __bond_release_one(bond_dev, slave_dev, false);
1903 }
1904
1905 /*
1906 * First release a slave and then destroy the bond if no more slaves are left.
1907 * Must be under rtnl_lock when this function is called.
1908 */
1909 static int  bond_release_and_destroy(struct net_device *bond_dev,
1910                                      struct net_device *slave_dev)
1911 {
1912         struct bonding *bond = netdev_priv(bond_dev);
1913         int ret;
1914
1915         ret = bond_release(bond_dev, slave_dev);
1916         if (ret == 0 && list_empty(&bond->slave_list)) {
1917                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1918                 pr_info("%s: destroying bond %s.\n",
1919                         bond_dev->name, bond_dev->name);
1920                 unregister_netdevice(bond_dev);
1921         }
1922         return ret;
1923 }
1924
1925 /*
1926  * This function changes the active slave to slave <slave_dev>.
1927  * It returns -EINVAL in the following cases.
1928  *  - <slave_dev> is not found in the list.
1929  *  - There is not active slave now.
1930  *  - <slave_dev> is already active.
1931  *  - The link state of <slave_dev> is not BOND_LINK_UP.
1932  *  - <slave_dev> is not running.
1933  * In these cases, this function does nothing.
1934  * In the other cases, current_slave pointer is changed and 0 is returned.
1935  */
1936 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1937 {
1938         struct bonding *bond = netdev_priv(bond_dev);
1939         struct slave *old_active = NULL;
1940         struct slave *new_active = NULL;
1941         int res = 0;
1942
1943         if (!USES_PRIMARY(bond->params.mode))
1944                 return -EINVAL;
1945
1946         /* Verify that bond_dev is indeed the master of slave_dev */
1947         if (!(slave_dev->flags & IFF_SLAVE) ||
1948             !netdev_has_upper_dev(slave_dev, bond_dev))
1949                 return -EINVAL;
1950
1951         read_lock(&bond->lock);
1952
1953         old_active = bond->curr_active_slave;
1954         new_active = bond_get_slave_by_dev(bond, slave_dev);
1955         /*
1956          * Changing to the current active: do nothing; return success.
1957          */
1958         if (new_active && new_active == old_active) {
1959                 read_unlock(&bond->lock);
1960                 return 0;
1961         }
1962
1963         if (new_active &&
1964             old_active &&
1965             new_active->link == BOND_LINK_UP &&
1966             IS_UP(new_active->dev)) {
1967                 block_netpoll_tx();
1968                 write_lock_bh(&bond->curr_slave_lock);
1969                 bond_change_active_slave(bond, new_active);
1970                 write_unlock_bh(&bond->curr_slave_lock);
1971                 unblock_netpoll_tx();
1972         } else
1973                 res = -EINVAL;
1974
1975         read_unlock(&bond->lock);
1976
1977         return res;
1978 }
1979
1980 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1981 {
1982         struct bonding *bond = netdev_priv(bond_dev);
1983
1984         info->bond_mode = bond->params.mode;
1985         info->miimon = bond->params.miimon;
1986
1987         read_lock(&bond->lock);
1988         info->num_slaves = bond->slave_cnt;
1989         read_unlock(&bond->lock);
1990
1991         return 0;
1992 }
1993
1994 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1995 {
1996         struct bonding *bond = netdev_priv(bond_dev);
1997         int i = 0, res = -ENODEV;
1998         struct slave *slave;
1999
2000         read_lock(&bond->lock);
2001         bond_for_each_slave(bond, slave) {
2002                 if (i++ == (int)info->slave_id) {
2003                         res = 0;
2004                         strcpy(info->slave_name, slave->dev->name);
2005                         info->link = slave->link;
2006                         info->state = bond_slave_state(slave);
2007                         info->link_failure_count = slave->link_failure_count;
2008                         break;
2009                 }
2010         }
2011         read_unlock(&bond->lock);
2012
2013         return res;
2014 }
2015
2016 /*-------------------------------- Monitoring -------------------------------*/
2017
2018
2019 static int bond_miimon_inspect(struct bonding *bond)
2020 {
2021         int link_state, commit = 0;
2022         struct slave *slave;
2023         bool ignore_updelay;
2024
2025         ignore_updelay = !bond->curr_active_slave ? true : false;
2026
2027         bond_for_each_slave(bond, slave) {
2028                 slave->new_link = BOND_LINK_NOCHANGE;
2029
2030                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2031
2032                 switch (slave->link) {
2033                 case BOND_LINK_UP:
2034                         if (link_state)
2035                                 continue;
2036
2037                         slave->link = BOND_LINK_FAIL;
2038                         slave->delay = bond->params.downdelay;
2039                         if (slave->delay) {
2040                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2041                                         bond->dev->name,
2042                                         (bond->params.mode ==
2043                                          BOND_MODE_ACTIVEBACKUP) ?
2044                                         (bond_is_active_slave(slave) ?
2045                                          "active " : "backup ") : "",
2046                                         slave->dev->name,
2047                                         bond->params.downdelay * bond->params.miimon);
2048                         }
2049                         /*FALLTHRU*/
2050                 case BOND_LINK_FAIL:
2051                         if (link_state) {
2052                                 /*
2053                                  * recovered before downdelay expired
2054                                  */
2055                                 slave->link = BOND_LINK_UP;
2056                                 slave->jiffies = jiffies;
2057                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2058                                         bond->dev->name,
2059                                         (bond->params.downdelay - slave->delay) *
2060                                         bond->params.miimon,
2061                                         slave->dev->name);
2062                                 continue;
2063                         }
2064
2065                         if (slave->delay <= 0) {
2066                                 slave->new_link = BOND_LINK_DOWN;
2067                                 commit++;
2068                                 continue;
2069                         }
2070
2071                         slave->delay--;
2072                         break;
2073
2074                 case BOND_LINK_DOWN:
2075                         if (!link_state)
2076                                 continue;
2077
2078                         slave->link = BOND_LINK_BACK;
2079                         slave->delay = bond->params.updelay;
2080
2081                         if (slave->delay) {
2082                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2083                                         bond->dev->name, slave->dev->name,
2084                                         ignore_updelay ? 0 :
2085                                         bond->params.updelay *
2086                                         bond->params.miimon);
2087                         }
2088                         /*FALLTHRU*/
2089                 case BOND_LINK_BACK:
2090                         if (!link_state) {
2091                                 slave->link = BOND_LINK_DOWN;
2092                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2093                                         bond->dev->name,
2094                                         (bond->params.updelay - slave->delay) *
2095                                         bond->params.miimon,
2096                                         slave->dev->name);
2097
2098                                 continue;
2099                         }
2100
2101                         if (ignore_updelay)
2102                                 slave->delay = 0;
2103
2104                         if (slave->delay <= 0) {
2105                                 slave->new_link = BOND_LINK_UP;
2106                                 commit++;
2107                                 ignore_updelay = false;
2108                                 continue;
2109                         }
2110
2111                         slave->delay--;
2112                         break;
2113                 }
2114         }
2115
2116         return commit;
2117 }
2118
2119 static void bond_miimon_commit(struct bonding *bond)
2120 {
2121         struct slave *slave;
2122
2123         bond_for_each_slave(bond, slave) {
2124                 switch (slave->new_link) {
2125                 case BOND_LINK_NOCHANGE:
2126                         continue;
2127
2128                 case BOND_LINK_UP:
2129                         slave->link = BOND_LINK_UP;
2130                         slave->jiffies = jiffies;
2131
2132                         if (bond->params.mode == BOND_MODE_8023AD) {
2133                                 /* prevent it from being the active one */
2134                                 bond_set_backup_slave(slave);
2135                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2136                                 /* make it immediately active */
2137                                 bond_set_active_slave(slave);
2138                         } else if (slave != bond->primary_slave) {
2139                                 /* prevent it from being the active one */
2140                                 bond_set_backup_slave(slave);
2141                         }
2142
2143                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2144                                 bond->dev->name, slave->dev->name,
2145                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2146                                 slave->duplex ? "full" : "half");
2147
2148                         /* notify ad that the link status has changed */
2149                         if (bond->params.mode == BOND_MODE_8023AD)
2150                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2151
2152                         if (bond_is_lb(bond))
2153                                 bond_alb_handle_link_change(bond, slave,
2154                                                             BOND_LINK_UP);
2155
2156                         if (!bond->curr_active_slave ||
2157                             (slave == bond->primary_slave))
2158                                 goto do_failover;
2159
2160                         continue;
2161
2162                 case BOND_LINK_DOWN:
2163                         if (slave->link_failure_count < UINT_MAX)
2164                                 slave->link_failure_count++;
2165
2166                         slave->link = BOND_LINK_DOWN;
2167
2168                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2169                             bond->params.mode == BOND_MODE_8023AD)
2170                                 bond_set_slave_inactive_flags(slave);
2171
2172                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2173                                 bond->dev->name, slave->dev->name);
2174
2175                         if (bond->params.mode == BOND_MODE_8023AD)
2176                                 bond_3ad_handle_link_change(slave,
2177                                                             BOND_LINK_DOWN);
2178
2179                         if (bond_is_lb(bond))
2180                                 bond_alb_handle_link_change(bond, slave,
2181                                                             BOND_LINK_DOWN);
2182
2183                         if (slave == bond->curr_active_slave)
2184                                 goto do_failover;
2185
2186                         continue;
2187
2188                 default:
2189                         pr_err("%s: invalid new link %d on slave %s\n",
2190                                bond->dev->name, slave->new_link,
2191                                slave->dev->name);
2192                         slave->new_link = BOND_LINK_NOCHANGE;
2193
2194                         continue;
2195                 }
2196
2197 do_failover:
2198                 ASSERT_RTNL();
2199                 block_netpoll_tx();
2200                 write_lock_bh(&bond->curr_slave_lock);
2201                 bond_select_active_slave(bond);
2202                 write_unlock_bh(&bond->curr_slave_lock);
2203                 unblock_netpoll_tx();
2204         }
2205
2206         bond_set_carrier(bond);
2207 }
2208
2209 /*
2210  * bond_mii_monitor
2211  *
2212  * Really a wrapper that splits the mii monitor into two phases: an
2213  * inspection, then (if inspection indicates something needs to be done)
2214  * an acquisition of appropriate locks followed by a commit phase to
2215  * implement whatever link state changes are indicated.
2216  */
2217 void bond_mii_monitor(struct work_struct *work)
2218 {
2219         struct bonding *bond = container_of(work, struct bonding,
2220                                             mii_work.work);
2221         bool should_notify_peers = false;
2222         unsigned long delay;
2223
2224         read_lock(&bond->lock);
2225
2226         delay = msecs_to_jiffies(bond->params.miimon);
2227
2228         if (list_empty(&bond->slave_list))
2229                 goto re_arm;
2230
2231         should_notify_peers = bond_should_notify_peers(bond);
2232
2233         if (bond_miimon_inspect(bond)) {
2234                 read_unlock(&bond->lock);
2235
2236                 /* Race avoidance with bond_close cancel of workqueue */
2237                 if (!rtnl_trylock()) {
2238                         read_lock(&bond->lock);
2239                         delay = 1;
2240                         should_notify_peers = false;
2241                         goto re_arm;
2242                 }
2243
2244                 read_lock(&bond->lock);
2245
2246                 bond_miimon_commit(bond);
2247
2248                 read_unlock(&bond->lock);
2249                 rtnl_unlock();  /* might sleep, hold no other locks */
2250                 read_lock(&bond->lock);
2251         }
2252
2253 re_arm:
2254         if (bond->params.miimon)
2255                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2256
2257         read_unlock(&bond->lock);
2258
2259         if (should_notify_peers) {
2260                 if (!rtnl_trylock())
2261                         return;
2262                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2263                 rtnl_unlock();
2264         }
2265 }
2266
2267 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2268 {
2269         struct net_device *upper;
2270         struct list_head *iter;
2271         bool ret = false;
2272
2273         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2274                 return true;
2275
2276         rcu_read_lock();
2277         netdev_for_each_upper_dev_rcu(bond->dev, upper, iter) {
2278                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2279                         ret = true;
2280                         break;
2281                 }
2282         }
2283         rcu_read_unlock();
2284
2285         return ret;
2286 }
2287
2288 /*
2289  * We go to the (large) trouble of VLAN tagging ARP frames because
2290  * switches in VLAN mode (especially if ports are configured as
2291  * "native" to a VLAN) might not pass non-tagged frames.
2292  */
2293 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2294 {
2295         struct sk_buff *skb;
2296
2297         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2298                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2299
2300         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2301                          NULL, slave_dev->dev_addr, NULL);
2302
2303         if (!skb) {
2304                 pr_err("ARP packet allocation failed\n");
2305                 return;
2306         }
2307         if (vlan_id) {
2308                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2309                 if (!skb) {
2310                         pr_err("failed to insert VLAN tag\n");
2311                         return;
2312                 }
2313         }
2314         arp_xmit(skb);
2315 }
2316
2317
2318 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2319 {
2320         struct net_device *upper, *vlan_upper;
2321         struct list_head *iter, *vlan_iter;
2322         struct rtable *rt;
2323         __be32 *targets = bond->params.arp_targets, addr;
2324         int i, vlan_id;
2325
2326         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2327                 pr_debug("basa: target %pI4\n", &targets[i]);
2328
2329                 /* Find out through which dev should the packet go */
2330                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2331                                      RTO_ONLINK, 0);
2332                 if (IS_ERR(rt)) {
2333                         pr_debug("%s: no route to arp_ip_target %pI4\n",
2334                                  bond->dev->name, &targets[i]);
2335                         continue;
2336                 }
2337
2338                 vlan_id = 0;
2339
2340                 /* bond device itself */
2341                 if (rt->dst.dev == bond->dev)
2342                         goto found;
2343
2344                 rcu_read_lock();
2345                 /* first we search only for vlan devices. for every vlan
2346                  * found we verify its upper dev list, searching for the
2347                  * rt->dst.dev. If found we save the tag of the vlan and
2348                  * proceed to send the packet.
2349                  *
2350                  * TODO: QinQ?
2351                  */
2352                 netdev_for_each_upper_dev_rcu(bond->dev, vlan_upper, vlan_iter) {
2353                         if (!is_vlan_dev(vlan_upper))
2354                                 continue;
2355                         netdev_for_each_upper_dev_rcu(vlan_upper, upper, iter) {
2356                                 if (upper == rt->dst.dev) {
2357                                         vlan_id = vlan_dev_vlan_id(vlan_upper);
2358                                         rcu_read_unlock();
2359                                         goto found;
2360                                 }
2361                         }
2362                 }
2363
2364                 /* if the device we're looking for is not on top of any of
2365                  * our upper vlans, then just search for any dev that
2366                  * matches, and in case it's a vlan - save the id
2367                  */
2368                 netdev_for_each_upper_dev_rcu(bond->dev, upper, iter) {
2369                         if (upper == rt->dst.dev) {
2370                                 /* if it's a vlan - get its VID */
2371                                 if (is_vlan_dev(upper))
2372                                         vlan_id = vlan_dev_vlan_id(upper);
2373
2374                                 rcu_read_unlock();
2375                                 goto found;
2376                         }
2377                 }
2378                 rcu_read_unlock();
2379
2380                 /* Not our device - skip */
2381                 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2382                          bond->dev->name, &targets[i],
2383                          rt->dst.dev ? rt->dst.dev->name : "NULL");
2384
2385                 ip_rt_put(rt);
2386                 continue;
2387
2388 found:
2389                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2390                 ip_rt_put(rt);
2391                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2392                               addr, vlan_id);
2393         }
2394 }
2395
2396 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2397 {
2398         int i;
2399
2400         if (!sip || !bond_has_this_ip(bond, tip)) {
2401                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2402                 return;
2403         }
2404
2405         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2406         if (i == -1) {
2407                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2408                 return;
2409         }
2410         slave->last_arp_rx = jiffies;
2411         slave->target_last_arp_rx[i] = jiffies;
2412 }
2413
2414 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2415                  struct slave *slave)
2416 {
2417         struct arphdr *arp = (struct arphdr *)skb->data;
2418         unsigned char *arp_ptr;
2419         __be32 sip, tip;
2420         int alen;
2421
2422         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2423                 return RX_HANDLER_ANOTHER;
2424
2425         read_lock(&bond->lock);
2426
2427         if (!slave_do_arp_validate(bond, slave))
2428                 goto out_unlock;
2429
2430         alen = arp_hdr_len(bond->dev);
2431
2432         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2433                  bond->dev->name, skb->dev->name);
2434
2435         if (alen > skb_headlen(skb)) {
2436                 arp = kmalloc(alen, GFP_ATOMIC);
2437                 if (!arp)
2438                         goto out_unlock;
2439                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2440                         goto out_unlock;
2441         }
2442
2443         if (arp->ar_hln != bond->dev->addr_len ||
2444             skb->pkt_type == PACKET_OTHERHOST ||
2445             skb->pkt_type == PACKET_LOOPBACK ||
2446             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2447             arp->ar_pro != htons(ETH_P_IP) ||
2448             arp->ar_pln != 4)
2449                 goto out_unlock;
2450
2451         arp_ptr = (unsigned char *)(arp + 1);
2452         arp_ptr += bond->dev->addr_len;
2453         memcpy(&sip, arp_ptr, 4);
2454         arp_ptr += 4 + bond->dev->addr_len;
2455         memcpy(&tip, arp_ptr, 4);
2456
2457         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2458                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2459                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2460                  &sip, &tip);
2461
2462         /*
2463          * Backup slaves won't see the ARP reply, but do come through
2464          * here for each ARP probe (so we swap the sip/tip to validate
2465          * the probe).  In a "redundant switch, common router" type of
2466          * configuration, the ARP probe will (hopefully) travel from
2467          * the active, through one switch, the router, then the other
2468          * switch before reaching the backup.
2469          *
2470          * We 'trust' the arp requests if there is an active slave and
2471          * it received valid arp reply(s) after it became active. This
2472          * is done to avoid endless looping when we can't reach the
2473          * arp_ip_target and fool ourselves with our own arp requests.
2474          */
2475         if (bond_is_active_slave(slave))
2476                 bond_validate_arp(bond, slave, sip, tip);
2477         else if (bond->curr_active_slave &&
2478                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2479                             bond->curr_active_slave->jiffies))
2480                 bond_validate_arp(bond, slave, tip, sip);
2481
2482 out_unlock:
2483         read_unlock(&bond->lock);
2484         if (arp != (struct arphdr *)skb->data)
2485                 kfree(arp);
2486         return RX_HANDLER_ANOTHER;
2487 }
2488
2489 /* function to verify if we're in the arp_interval timeslice, returns true if
2490  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2491  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2492  */
2493 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2494                                   int mod)
2495 {
2496         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2497
2498         return time_in_range(jiffies,
2499                              last_act - delta_in_ticks,
2500                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2501 }
2502
2503 /*
2504  * this function is called regularly to monitor each slave's link
2505  * ensuring that traffic is being sent and received when arp monitoring
2506  * is used in load-balancing mode. if the adapter has been dormant, then an
2507  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2508  * arp monitoring in active backup mode.
2509  */
2510 void bond_loadbalance_arp_mon(struct work_struct *work)
2511 {
2512         struct bonding *bond = container_of(work, struct bonding,
2513                                             arp_work.work);
2514         struct slave *slave, *oldcurrent;
2515         int do_failover = 0;
2516
2517         read_lock(&bond->lock);
2518
2519         if (list_empty(&bond->slave_list))
2520                 goto re_arm;
2521
2522         oldcurrent = bond->curr_active_slave;
2523         /* see if any of the previous devices are up now (i.e. they have
2524          * xmt and rcv traffic). the curr_active_slave does not come into
2525          * the picture unless it is null. also, slave->jiffies is not needed
2526          * here because we send an arp on each slave and give a slave as
2527          * long as it needs to get the tx/rx within the delta.
2528          * TODO: what about up/down delay in arp mode? it wasn't here before
2529          *       so it can wait
2530          */
2531         bond_for_each_slave(bond, slave) {
2532                 unsigned long trans_start = dev_trans_start(slave->dev);
2533
2534                 if (slave->link != BOND_LINK_UP) {
2535                         if (bond_time_in_interval(bond, trans_start, 1) &&
2536                             bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2537
2538                                 slave->link  = BOND_LINK_UP;
2539                                 bond_set_active_slave(slave);
2540
2541                                 /* primary_slave has no meaning in round-robin
2542                                  * mode. the window of a slave being up and
2543                                  * curr_active_slave being null after enslaving
2544                                  * is closed.
2545                                  */
2546                                 if (!oldcurrent) {
2547                                         pr_info("%s: link status definitely up for interface %s, ",
2548                                                 bond->dev->name,
2549                                                 slave->dev->name);
2550                                         do_failover = 1;
2551                                 } else {
2552                                         pr_info("%s: interface %s is now up\n",
2553                                                 bond->dev->name,
2554                                                 slave->dev->name);
2555                                 }
2556                         }
2557                 } else {
2558                         /* slave->link == BOND_LINK_UP */
2559
2560                         /* not all switches will respond to an arp request
2561                          * when the source ip is 0, so don't take the link down
2562                          * if we don't know our ip yet
2563                          */
2564                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2565                             !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2566
2567                                 slave->link  = BOND_LINK_DOWN;
2568                                 bond_set_backup_slave(slave);
2569
2570                                 if (slave->link_failure_count < UINT_MAX)
2571                                         slave->link_failure_count++;
2572
2573                                 pr_info("%s: interface %s is now down.\n",
2574                                         bond->dev->name,
2575                                         slave->dev->name);
2576
2577                                 if (slave == oldcurrent)
2578                                         do_failover = 1;
2579                         }
2580                 }
2581
2582                 /* note: if switch is in round-robin mode, all links
2583                  * must tx arp to ensure all links rx an arp - otherwise
2584                  * links may oscillate or not come up at all; if switch is
2585                  * in something like xor mode, there is nothing we can
2586                  * do - all replies will be rx'ed on same link causing slaves
2587                  * to be unstable during low/no traffic periods
2588                  */
2589                 if (IS_UP(slave->dev))
2590                         bond_arp_send_all(bond, slave);
2591         }
2592
2593         if (do_failover) {
2594                 block_netpoll_tx();
2595                 write_lock_bh(&bond->curr_slave_lock);
2596
2597                 bond_select_active_slave(bond);
2598
2599                 write_unlock_bh(&bond->curr_slave_lock);
2600                 unblock_netpoll_tx();
2601         }
2602
2603 re_arm:
2604         if (bond->params.arp_interval)
2605                 queue_delayed_work(bond->wq, &bond->arp_work,
2606                                    msecs_to_jiffies(bond->params.arp_interval));
2607
2608         read_unlock(&bond->lock);
2609 }
2610
2611 /*
2612  * Called to inspect slaves for active-backup mode ARP monitor link state
2613  * changes.  Sets new_link in slaves to specify what action should take
2614  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2615  * to link states must be committed.
2616  *
2617  * Called with bond->lock held for read.
2618  */
2619 static int bond_ab_arp_inspect(struct bonding *bond)
2620 {
2621         unsigned long trans_start, last_rx;
2622         struct slave *slave;
2623         int commit = 0;
2624
2625         bond_for_each_slave(bond, slave) {
2626                 slave->new_link = BOND_LINK_NOCHANGE;
2627                 last_rx = slave_last_rx(bond, slave);
2628
2629                 if (slave->link != BOND_LINK_UP) {
2630                         if (bond_time_in_interval(bond, last_rx, 1)) {
2631                                 slave->new_link = BOND_LINK_UP;
2632                                 commit++;
2633                         }
2634                         continue;
2635                 }
2636
2637                 /*
2638                  * Give slaves 2*delta after being enslaved or made
2639                  * active.  This avoids bouncing, as the last receive
2640                  * times need a full ARP monitor cycle to be updated.
2641                  */
2642                 if (bond_time_in_interval(bond, slave->jiffies, 2))
2643                         continue;
2644
2645                 /*
2646                  * Backup slave is down if:
2647                  * - No current_arp_slave AND
2648                  * - more than 3*delta since last receive AND
2649                  * - the bond has an IP address
2650                  *
2651                  * Note: a non-null current_arp_slave indicates
2652                  * the curr_active_slave went down and we are
2653                  * searching for a new one; under this condition
2654                  * we only take the curr_active_slave down - this
2655                  * gives each slave a chance to tx/rx traffic
2656                  * before being taken out
2657                  */
2658                 if (!bond_is_active_slave(slave) &&
2659                     !bond->current_arp_slave &&
2660                     !bond_time_in_interval(bond, last_rx, 3)) {
2661                         slave->new_link = BOND_LINK_DOWN;
2662                         commit++;
2663                 }
2664
2665                 /*
2666                  * Active slave is down if:
2667                  * - more than 2*delta since transmitting OR
2668                  * - (more than 2*delta since receive AND
2669                  *    the bond has an IP address)
2670                  */
2671                 trans_start = dev_trans_start(slave->dev);
2672                 if (bond_is_active_slave(slave) &&
2673                     (!bond_time_in_interval(bond, trans_start, 2) ||
2674                      !bond_time_in_interval(bond, last_rx, 2))) {
2675                         slave->new_link = BOND_LINK_DOWN;
2676                         commit++;
2677                 }
2678         }
2679
2680         return commit;
2681 }
2682
2683 /*
2684  * Called to commit link state changes noted by inspection step of
2685  * active-backup mode ARP monitor.
2686  *
2687  * Called with RTNL and bond->lock for read.
2688  */
2689 static void bond_ab_arp_commit(struct bonding *bond)
2690 {
2691         unsigned long trans_start;
2692         struct slave *slave;
2693
2694         bond_for_each_slave(bond, slave) {
2695                 switch (slave->new_link) {
2696                 case BOND_LINK_NOCHANGE:
2697                         continue;
2698
2699                 case BOND_LINK_UP:
2700                         trans_start = dev_trans_start(slave->dev);
2701                         if (bond->curr_active_slave != slave ||
2702                             (!bond->curr_active_slave &&
2703                              bond_time_in_interval(bond, trans_start, 1))) {
2704                                 slave->link = BOND_LINK_UP;
2705                                 if (bond->current_arp_slave) {
2706                                         bond_set_slave_inactive_flags(
2707                                                 bond->current_arp_slave);
2708                                         bond->current_arp_slave = NULL;
2709                                 }
2710
2711                                 pr_info("%s: link status definitely up for interface %s.\n",
2712                                         bond->dev->name, slave->dev->name);
2713
2714                                 if (!bond->curr_active_slave ||
2715                                     (slave == bond->primary_slave))
2716                                         goto do_failover;
2717
2718                         }
2719
2720                         continue;
2721
2722                 case BOND_LINK_DOWN:
2723                         if (slave->link_failure_count < UINT_MAX)
2724                                 slave->link_failure_count++;
2725
2726                         slave->link = BOND_LINK_DOWN;
2727                         bond_set_slave_inactive_flags(slave);
2728
2729                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2730                                 bond->dev->name, slave->dev->name);
2731
2732                         if (slave == bond->curr_active_slave) {
2733                                 bond->current_arp_slave = NULL;
2734                                 goto do_failover;
2735                         }
2736
2737                         continue;
2738
2739                 default:
2740                         pr_err("%s: impossible: new_link %d on slave %s\n",
2741                                bond->dev->name, slave->new_link,
2742                                slave->dev->name);
2743                         continue;
2744                 }
2745
2746 do_failover:
2747                 ASSERT_RTNL();
2748                 block_netpoll_tx();
2749                 write_lock_bh(&bond->curr_slave_lock);
2750                 bond_select_active_slave(bond);
2751                 write_unlock_bh(&bond->curr_slave_lock);
2752                 unblock_netpoll_tx();
2753         }
2754
2755         bond_set_carrier(bond);
2756 }
2757
2758 /*
2759  * Send ARP probes for active-backup mode ARP monitor.
2760  *
2761  * Called with bond->lock held for read.
2762  */
2763 static void bond_ab_arp_probe(struct bonding *bond)
2764 {
2765         struct slave *slave, *next_slave;
2766         int i;
2767
2768         read_lock(&bond->curr_slave_lock);
2769
2770         if (bond->current_arp_slave && bond->curr_active_slave)
2771                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2772                         bond->current_arp_slave->dev->name,
2773                         bond->curr_active_slave->dev->name);
2774
2775         if (bond->curr_active_slave) {
2776                 bond_arp_send_all(bond, bond->curr_active_slave);
2777                 read_unlock(&bond->curr_slave_lock);
2778                 return;
2779         }
2780
2781         read_unlock(&bond->curr_slave_lock);
2782
2783         /* if we don't have a curr_active_slave, search for the next available
2784          * backup slave from the current_arp_slave and make it the candidate
2785          * for becoming the curr_active_slave
2786          */
2787
2788         if (!bond->current_arp_slave) {
2789                 bond->current_arp_slave = bond_first_slave(bond);
2790                 if (!bond->current_arp_slave)
2791                         return;
2792         }
2793
2794         bond_set_slave_inactive_flags(bond->current_arp_slave);
2795
2796         /* search for next candidate */
2797         next_slave = bond_next_slave(bond, bond->current_arp_slave);
2798         bond_for_each_slave_from(bond, slave, i, next_slave) {
2799                 if (IS_UP(slave->dev)) {
2800                         slave->link = BOND_LINK_BACK;
2801                         bond_set_slave_active_flags(slave);
2802                         bond_arp_send_all(bond, slave);
2803                         slave->jiffies = jiffies;
2804                         bond->current_arp_slave = slave;
2805                         break;
2806                 }
2807
2808                 /* if the link state is up at this point, we
2809                  * mark it down - this can happen if we have
2810                  * simultaneous link failures and
2811                  * reselect_active_interface doesn't make this
2812                  * one the current slave so it is still marked
2813                  * up when it is actually down
2814                  */
2815                 if (slave->link == BOND_LINK_UP) {
2816                         slave->link = BOND_LINK_DOWN;
2817                         if (slave->link_failure_count < UINT_MAX)
2818                                 slave->link_failure_count++;
2819
2820                         bond_set_slave_inactive_flags(slave);
2821
2822                         pr_info("%s: backup interface %s is now down.\n",
2823                                 bond->dev->name, slave->dev->name);
2824                 }
2825         }
2826 }
2827
2828 void bond_activebackup_arp_mon(struct work_struct *work)
2829 {
2830         struct bonding *bond = container_of(work, struct bonding,
2831                                             arp_work.work);
2832         bool should_notify_peers = false;
2833         int delta_in_ticks;
2834
2835         read_lock(&bond->lock);
2836
2837         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2838
2839         if (list_empty(&bond->slave_list))
2840                 goto re_arm;
2841
2842         should_notify_peers = bond_should_notify_peers(bond);
2843
2844         if (bond_ab_arp_inspect(bond)) {
2845                 read_unlock(&bond->lock);
2846
2847                 /* Race avoidance with bond_close flush of workqueue */
2848                 if (!rtnl_trylock()) {
2849                         read_lock(&bond->lock);
2850                         delta_in_ticks = 1;
2851                         should_notify_peers = false;
2852                         goto re_arm;
2853                 }
2854
2855                 read_lock(&bond->lock);
2856
2857                 bond_ab_arp_commit(bond);
2858
2859                 read_unlock(&bond->lock);
2860                 rtnl_unlock();
2861                 read_lock(&bond->lock);
2862         }
2863
2864         bond_ab_arp_probe(bond);
2865
2866 re_arm:
2867         if (bond->params.arp_interval)
2868                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2869
2870         read_unlock(&bond->lock);
2871
2872         if (should_notify_peers) {
2873                 if (!rtnl_trylock())
2874                         return;
2875                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2876                 rtnl_unlock();
2877         }
2878 }
2879
2880 /*-------------------------- netdev event handling --------------------------*/
2881
2882 /*
2883  * Change device name
2884  */
2885 static int bond_event_changename(struct bonding *bond)
2886 {
2887         bond_remove_proc_entry(bond);
2888         bond_create_proc_entry(bond);
2889
2890         bond_debug_reregister(bond);
2891
2892         return NOTIFY_DONE;
2893 }
2894
2895 static int bond_master_netdev_event(unsigned long event,
2896                                     struct net_device *bond_dev)
2897 {
2898         struct bonding *event_bond = netdev_priv(bond_dev);
2899
2900         switch (event) {
2901         case NETDEV_CHANGENAME:
2902                 return bond_event_changename(event_bond);
2903         case NETDEV_UNREGISTER:
2904                 bond_remove_proc_entry(event_bond);
2905                 break;
2906         case NETDEV_REGISTER:
2907                 bond_create_proc_entry(event_bond);
2908                 break;
2909         case NETDEV_NOTIFY_PEERS:
2910                 if (event_bond->send_peer_notif)
2911                         event_bond->send_peer_notif--;
2912                 break;
2913         default:
2914                 break;
2915         }
2916
2917         return NOTIFY_DONE;
2918 }
2919
2920 static int bond_slave_netdev_event(unsigned long event,
2921                                    struct net_device *slave_dev)
2922 {
2923         struct slave *slave = bond_slave_get_rtnl(slave_dev);
2924         struct bonding *bond;
2925         struct net_device *bond_dev;
2926         u32 old_speed;
2927         u8 old_duplex;
2928
2929         /* A netdev event can be generated while enslaving a device
2930          * before netdev_rx_handler_register is called in which case
2931          * slave will be NULL
2932          */
2933         if (!slave)
2934                 return NOTIFY_DONE;
2935         bond_dev = slave->bond->dev;
2936         bond = slave->bond;
2937
2938         switch (event) {
2939         case NETDEV_UNREGISTER:
2940                 if (bond_dev->type != ARPHRD_ETHER)
2941                         bond_release_and_destroy(bond_dev, slave_dev);
2942                 else
2943                         bond_release(bond_dev, slave_dev);
2944                 break;
2945         case NETDEV_UP:
2946         case NETDEV_CHANGE:
2947                 old_speed = slave->speed;
2948                 old_duplex = slave->duplex;
2949
2950                 bond_update_speed_duplex(slave);
2951
2952                 if (bond->params.mode == BOND_MODE_8023AD) {
2953                         if (old_speed != slave->speed)
2954                                 bond_3ad_adapter_speed_changed(slave);
2955                         if (old_duplex != slave->duplex)
2956                                 bond_3ad_adapter_duplex_changed(slave);
2957                 }
2958                 break;
2959         case NETDEV_DOWN:
2960                 /*
2961                  * ... Or is it this?
2962                  */
2963                 break;
2964         case NETDEV_CHANGEMTU:
2965                 /*
2966                  * TODO: Should slaves be allowed to
2967                  * independently alter their MTU?  For
2968                  * an active-backup bond, slaves need
2969                  * not be the same type of device, so
2970                  * MTUs may vary.  For other modes,
2971                  * slaves arguably should have the
2972                  * same MTUs. To do this, we'd need to
2973                  * take over the slave's change_mtu
2974                  * function for the duration of their
2975                  * servitude.
2976                  */
2977                 break;
2978         case NETDEV_CHANGENAME:
2979                 /*
2980                  * TODO: handle changing the primary's name
2981                  */
2982                 break;
2983         case NETDEV_FEAT_CHANGE:
2984                 bond_compute_features(bond);
2985                 break;
2986         case NETDEV_RESEND_IGMP:
2987                 /* Propagate to master device */
2988                 call_netdevice_notifiers(event, slave->bond->dev);
2989                 break;
2990         default:
2991                 break;
2992         }
2993
2994         return NOTIFY_DONE;
2995 }
2996
2997 /*
2998  * bond_netdev_event: handle netdev notifier chain events.
2999  *
3000  * This function receives events for the netdev chain.  The caller (an
3001  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3002  * locks for us to safely manipulate the slave devices (RTNL lock,
3003  * dev_probe_lock).
3004  */
3005 static int bond_netdev_event(struct notifier_block *this,
3006                              unsigned long event, void *ptr)
3007 {
3008         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3009
3010         pr_debug("event_dev: %s, event: %lx\n",
3011                  event_dev ? event_dev->name : "None",
3012                  event);
3013
3014         if (!(event_dev->priv_flags & IFF_BONDING))
3015                 return NOTIFY_DONE;
3016
3017         if (event_dev->flags & IFF_MASTER) {
3018                 pr_debug("IFF_MASTER\n");
3019                 return bond_master_netdev_event(event, event_dev);
3020         }
3021
3022         if (event_dev->flags & IFF_SLAVE) {
3023                 pr_debug("IFF_SLAVE\n");
3024                 return bond_slave_netdev_event(event, event_dev);
3025         }
3026
3027         return NOTIFY_DONE;
3028 }
3029
3030 static struct notifier_block bond_netdev_notifier = {
3031         .notifier_call = bond_netdev_event,
3032 };
3033
3034 /*---------------------------- Hashing Policies -----------------------------*/
3035
3036 /*
3037  * Hash for the output device based upon layer 2 data
3038  */
3039 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3040 {
3041         struct ethhdr *data = (struct ethhdr *)skb->data;
3042
3043         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3044                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3045
3046         return 0;
3047 }
3048
3049 /*
3050  * Hash for the output device based upon layer 2 and layer 3 data. If
3051  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3052  */
3053 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3054 {
3055         const struct ethhdr *data;
3056         const struct iphdr *iph;
3057         const struct ipv6hdr *ipv6h;
3058         u32 v6hash;
3059         const __be32 *s, *d;
3060
3061         if (skb->protocol == htons(ETH_P_IP) &&
3062             pskb_network_may_pull(skb, sizeof(*iph))) {
3063                 iph = ip_hdr(skb);
3064                 data = (struct ethhdr *)skb->data;
3065                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3066                         (data->h_dest[5] ^ data->h_source[5])) % count;
3067         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3068                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3069                 ipv6h = ipv6_hdr(skb);
3070                 data = (struct ethhdr *)skb->data;
3071                 s = &ipv6h->saddr.s6_addr32[0];
3072                 d = &ipv6h->daddr.s6_addr32[0];
3073                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3074                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3075                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3076         }
3077
3078         return bond_xmit_hash_policy_l2(skb, count);
3079 }
3080
3081 /*
3082  * Hash for the output device based upon layer 3 and layer 4 data. If
3083  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3084  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3085  */
3086 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3087 {
3088         u32 layer4_xor = 0;
3089         const struct iphdr *iph;
3090         const struct ipv6hdr *ipv6h;
3091         const __be32 *s, *d;
3092         const __be16 *l4 = NULL;
3093         __be16 _l4[2];
3094         int noff = skb_network_offset(skb);
3095         int poff;
3096
3097         if (skb->protocol == htons(ETH_P_IP) &&
3098             pskb_may_pull(skb, noff + sizeof(*iph))) {
3099                 iph = ip_hdr(skb);
3100                 poff = proto_ports_offset(iph->protocol);
3101
3102                 if (!ip_is_fragment(iph) && poff >= 0) {
3103                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3104                                                 sizeof(_l4), &_l4);
3105                         if (l4)
3106                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3107                 }
3108                 return (layer4_xor ^
3109                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3110         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3111                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3112                 ipv6h = ipv6_hdr(skb);
3113                 poff = proto_ports_offset(ipv6h->nexthdr);
3114                 if (poff >= 0) {
3115                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3116                                                 sizeof(_l4), &_l4);
3117                         if (l4)
3118                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3119                 }
3120                 s = &ipv6h->saddr.s6_addr32[0];
3121                 d = &ipv6h->daddr.s6_addr32[0];
3122                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3123                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3124                                (layer4_xor >> 8);
3125                 return layer4_xor % count;
3126         }
3127
3128         return bond_xmit_hash_policy_l2(skb, count);
3129 }
3130
3131 /*-------------------------- Device entry points ----------------------------*/
3132
3133 static void bond_work_init_all(struct bonding *bond)
3134 {
3135         INIT_DELAYED_WORK(&bond->mcast_work,
3136                           bond_resend_igmp_join_requests_delayed);
3137         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3138         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3139         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3140                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3141         else
3142                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3143         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3144 }
3145
3146 static void bond_work_cancel_all(struct bonding *bond)
3147 {
3148         cancel_delayed_work_sync(&bond->mii_work);
3149         cancel_delayed_work_sync(&bond->arp_work);
3150         cancel_delayed_work_sync(&bond->alb_work);
3151         cancel_delayed_work_sync(&bond->ad_work);
3152         cancel_delayed_work_sync(&bond->mcast_work);
3153 }
3154
3155 static int bond_open(struct net_device *bond_dev)
3156 {
3157         struct bonding *bond = netdev_priv(bond_dev);
3158         struct slave *slave;
3159
3160         /* reset slave->backup and slave->inactive */
3161         read_lock(&bond->lock);
3162         if (!list_empty(&bond->slave_list)) {
3163                 read_lock(&bond->curr_slave_lock);
3164                 bond_for_each_slave(bond, slave) {
3165                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3166                                 && (slave != bond->curr_active_slave)) {
3167                                 bond_set_slave_inactive_flags(slave);
3168                         } else {
3169                                 bond_set_slave_active_flags(slave);
3170                         }
3171                 }
3172                 read_unlock(&bond->curr_slave_lock);
3173         }
3174         read_unlock(&bond->lock);
3175
3176         bond_work_init_all(bond);
3177
3178         if (bond_is_lb(bond)) {
3179                 /* bond_alb_initialize must be called before the timer
3180                  * is started.
3181                  */
3182                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3183                         return -ENOMEM;
3184                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3185         }
3186
3187         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3188                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3189
3190         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3191                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3192                 if (bond->params.arp_validate)
3193                         bond->recv_probe = bond_arp_rcv;
3194         }
3195
3196         if (bond->params.mode == BOND_MODE_8023AD) {
3197                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3198                 /* register to receive LACPDUs */
3199                 bond->recv_probe = bond_3ad_lacpdu_recv;
3200                 bond_3ad_initiate_agg_selection(bond, 1);
3201         }
3202
3203         return 0;
3204 }
3205
3206 static int bond_close(struct net_device *bond_dev)
3207 {
3208         struct bonding *bond = netdev_priv(bond_dev);
3209
3210         bond_work_cancel_all(bond);
3211         bond->send_peer_notif = 0;
3212         if (bond_is_lb(bond))
3213                 bond_alb_deinitialize(bond);
3214         bond->recv_probe = NULL;
3215
3216         return 0;
3217 }
3218
3219 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3220                                                 struct rtnl_link_stats64 *stats)
3221 {
3222         struct bonding *bond = netdev_priv(bond_dev);
3223         struct rtnl_link_stats64 temp;
3224         struct slave *slave;
3225
3226         memset(stats, 0, sizeof(*stats));
3227
3228         read_lock_bh(&bond->lock);
3229         bond_for_each_slave(bond, slave) {
3230                 const struct rtnl_link_stats64 *sstats =
3231                         dev_get_stats(slave->dev, &temp);
3232
3233                 stats->rx_packets += sstats->rx_packets;
3234                 stats->rx_bytes += sstats->rx_bytes;
3235                 stats->rx_errors += sstats->rx_errors;
3236                 stats->rx_dropped += sstats->rx_dropped;
3237
3238                 stats->tx_packets += sstats->tx_packets;
3239                 stats->tx_bytes += sstats->tx_bytes;
3240                 stats->tx_errors += sstats->tx_errors;
3241                 stats->tx_dropped += sstats->tx_dropped;
3242
3243                 stats->multicast += sstats->multicast;
3244                 stats->collisions += sstats->collisions;
3245
3246                 stats->rx_length_errors += sstats->rx_length_errors;
3247                 stats->rx_over_errors += sstats->rx_over_errors;
3248                 stats->rx_crc_errors += sstats->rx_crc_errors;
3249                 stats->rx_frame_errors += sstats->rx_frame_errors;
3250                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3251                 stats->rx_missed_errors += sstats->rx_missed_errors;
3252
3253                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3254                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3255                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3256                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3257                 stats->tx_window_errors += sstats->tx_window_errors;
3258         }
3259         read_unlock_bh(&bond->lock);
3260
3261         return stats;
3262 }
3263
3264 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3265 {
3266         struct net_device *slave_dev = NULL;
3267         struct ifbond k_binfo;
3268         struct ifbond __user *u_binfo = NULL;
3269         struct ifslave k_sinfo;
3270         struct ifslave __user *u_sinfo = NULL;
3271         struct mii_ioctl_data *mii = NULL;
3272         struct net *net;
3273         int res = 0;
3274
3275         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3276
3277         switch (cmd) {
3278         case SIOCGMIIPHY:
3279                 mii = if_mii(ifr);
3280                 if (!mii)
3281                         return -EINVAL;
3282
3283                 mii->phy_id = 0;
3284                 /* Fall Through */
3285         case SIOCGMIIREG:
3286                 /*
3287                  * We do this again just in case we were called by SIOCGMIIREG
3288                  * instead of SIOCGMIIPHY.
3289                  */
3290                 mii = if_mii(ifr);
3291                 if (!mii)
3292                         return -EINVAL;
3293
3294
3295                 if (mii->reg_num == 1) {
3296                         struct bonding *bond = netdev_priv(bond_dev);
3297                         mii->val_out = 0;
3298                         read_lock(&bond->lock);
3299                         read_lock(&bond->curr_slave_lock);
3300                         if (netif_carrier_ok(bond->dev))
3301                                 mii->val_out = BMSR_LSTATUS;
3302
3303                         read_unlock(&bond->curr_slave_lock);
3304                         read_unlock(&bond->lock);
3305                 }
3306
3307                 return 0;
3308         case BOND_INFO_QUERY_OLD:
3309         case SIOCBONDINFOQUERY:
3310                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3311
3312                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3313                         return -EFAULT;
3314
3315                 res = bond_info_query(bond_dev, &k_binfo);
3316                 if (res == 0 &&
3317                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3318                         return -EFAULT;
3319
3320                 return res;
3321         case BOND_SLAVE_INFO_QUERY_OLD:
3322         case SIOCBONDSLAVEINFOQUERY:
3323                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3324
3325                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3326                         return -EFAULT;
3327
3328                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3329                 if (res == 0 &&
3330                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3331                         return -EFAULT;
3332
3333                 return res;
3334         default:
3335                 /* Go on */
3336                 break;
3337         }
3338
3339         net = dev_net(bond_dev);
3340
3341         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3342                 return -EPERM;
3343
3344         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3345
3346         pr_debug("slave_dev=%p:\n", slave_dev);
3347
3348         if (!slave_dev)
3349                 res = -ENODEV;
3350         else {
3351                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3352                 switch (cmd) {
3353                 case BOND_ENSLAVE_OLD:
3354                 case SIOCBONDENSLAVE:
3355                         res = bond_enslave(bond_dev, slave_dev);
3356                         break;
3357                 case BOND_RELEASE_OLD:
3358                 case SIOCBONDRELEASE:
3359                         res = bond_release(bond_dev, slave_dev);
3360                         break;
3361                 case BOND_SETHWADDR_OLD:
3362                 case SIOCBONDSETHWADDR:
3363                         bond_set_dev_addr(bond_dev, slave_dev);
3364                         res = 0;
3365                         break;
3366                 case BOND_CHANGE_ACTIVE_OLD:
3367                 case SIOCBONDCHANGEACTIVE:
3368                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3369                         break;
3370                 default:
3371                         res = -EOPNOTSUPP;
3372                 }
3373
3374                 dev_put(slave_dev);
3375         }
3376
3377         return res;
3378 }
3379
3380 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3381 {
3382         struct bonding *bond = netdev_priv(bond_dev);
3383
3384         if (change & IFF_PROMISC)
3385                 bond_set_promiscuity(bond,
3386                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3387
3388         if (change & IFF_ALLMULTI)
3389                 bond_set_allmulti(bond,
3390                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3391 }
3392
3393 static void bond_set_rx_mode(struct net_device *bond_dev)
3394 {
3395         struct bonding *bond = netdev_priv(bond_dev);
3396         struct slave *slave;
3397
3398         ASSERT_RTNL();
3399
3400         if (USES_PRIMARY(bond->params.mode)) {
3401                 slave = rtnl_dereference(bond->curr_active_slave);
3402                 if (slave) {
3403                         dev_uc_sync(slave->dev, bond_dev);
3404                         dev_mc_sync(slave->dev, bond_dev);
3405                 }
3406         } else {
3407                 bond_for_each_slave(bond, slave) {
3408                         dev_uc_sync_multiple(slave->dev, bond_dev);
3409                         dev_mc_sync_multiple(slave->dev, bond_dev);
3410                 }
3411         }
3412 }
3413
3414 static int bond_neigh_init(struct neighbour *n)
3415 {
3416         struct bonding *bond = netdev_priv(n->dev);
3417         const struct net_device_ops *slave_ops;
3418         struct neigh_parms parms;
3419         struct slave *slave;
3420         int ret;
3421
3422         slave = bond_first_slave(bond);
3423         if (!slave)
3424                 return 0;
3425         slave_ops = slave->dev->netdev_ops;
3426         if (!slave_ops->ndo_neigh_setup)
3427                 return 0;
3428
3429         parms.neigh_setup = NULL;
3430         parms.neigh_cleanup = NULL;
3431         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3432         if (ret)
3433                 return ret;
3434
3435         /*
3436          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3437          * after the last slave has been detached.  Assumes that all slaves
3438          * utilize the same neigh_cleanup (true at this writing as only user
3439          * is ipoib).
3440          */
3441         n->parms->neigh_cleanup = parms.neigh_cleanup;
3442
3443         if (!parms.neigh_setup)
3444                 return 0;
3445
3446         return parms.neigh_setup(n);
3447 }
3448
3449 /*
3450  * The bonding ndo_neigh_setup is called at init time beofre any
3451  * slave exists. So we must declare proxy setup function which will
3452  * be used at run time to resolve the actual slave neigh param setup.
3453  *
3454  * It's also called by master devices (such as vlans) to setup their
3455  * underlying devices. In that case - do nothing, we're already set up from
3456  * our init.
3457  */
3458 static int bond_neigh_setup(struct net_device *dev,
3459                             struct neigh_parms *parms)
3460 {
3461         /* modify only our neigh_parms */
3462         if (parms->dev == dev)
3463                 parms->neigh_setup = bond_neigh_init;
3464
3465         return 0;
3466 }
3467
3468 /*
3469  * Change the MTU of all of a master's slaves to match the master
3470  */
3471 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3472 {
3473         struct bonding *bond = netdev_priv(bond_dev);
3474         struct slave *slave;
3475         int res = 0;
3476
3477         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3478                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3479
3480         /* Can't hold bond->lock with bh disabled here since
3481          * some base drivers panic. On the other hand we can't
3482          * hold bond->lock without bh disabled because we'll
3483          * deadlock. The only solution is to rely on the fact
3484          * that we're under rtnl_lock here, and the slaves
3485          * list won't change. This doesn't solve the problem
3486          * of setting the slave's MTU while it is
3487          * transmitting, but the assumption is that the base
3488          * driver can handle that.
3489          *
3490          * TODO: figure out a way to safely iterate the slaves
3491          * list, but without holding a lock around the actual
3492          * call to the base driver.
3493          */
3494
3495         bond_for_each_slave(bond, slave) {
3496                 pr_debug("s %p s->p %p c_m %p\n",
3497                          slave,
3498                          bond_prev_slave(bond, slave),
3499                          slave->dev->netdev_ops->ndo_change_mtu);
3500
3501                 res = dev_set_mtu(slave->dev, new_mtu);
3502
3503                 if (res) {
3504                         /* If we failed to set the slave's mtu to the new value
3505                          * we must abort the operation even in ACTIVE_BACKUP
3506                          * mode, because if we allow the backup slaves to have
3507                          * different mtu values than the active slave we'll
3508                          * need to change their mtu when doing a failover. That
3509                          * means changing their mtu from timer context, which
3510                          * is probably not a good idea.
3511                          */
3512                         pr_debug("err %d %s\n", res, slave->dev->name);
3513                         goto unwind;
3514                 }
3515         }
3516
3517         bond_dev->mtu = new_mtu;
3518
3519         return 0;
3520
3521 unwind:
3522         /* unwind from head to the slave that failed */
3523         bond_for_each_slave_continue_reverse(bond, slave) {
3524                 int tmp_res;
3525
3526                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3527                 if (tmp_res) {
3528                         pr_debug("unwind err %d dev %s\n",
3529                                  tmp_res, slave->dev->name);
3530                 }
3531         }
3532
3533         return res;
3534 }
3535
3536 /*
3537  * Change HW address
3538  *
3539  * Note that many devices must be down to change the HW address, and
3540  * downing the master releases all slaves.  We can make bonds full of
3541  * bonding devices to test this, however.
3542  */
3543 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3544 {
3545         struct bonding *bond = netdev_priv(bond_dev);
3546         struct sockaddr *sa = addr, tmp_sa;
3547         struct slave *slave;
3548         int res = 0;
3549
3550         if (bond->params.mode == BOND_MODE_ALB)
3551                 return bond_alb_set_mac_address(bond_dev, addr);
3552
3553
3554         pr_debug("bond=%p, name=%s\n",
3555                  bond, bond_dev ? bond_dev->name : "None");
3556
3557         /* If fail_over_mac is enabled, do nothing and return success.
3558          * Returning an error causes ifenslave to fail.
3559          */
3560         if (bond->params.fail_over_mac)
3561                 return 0;
3562
3563         if (!is_valid_ether_addr(sa->sa_data))
3564                 return -EADDRNOTAVAIL;
3565
3566         /* Can't hold bond->lock with bh disabled here since
3567          * some base drivers panic. On the other hand we can't
3568          * hold bond->lock without bh disabled because we'll
3569          * deadlock. The only solution is to rely on the fact
3570          * that we're under rtnl_lock here, and the slaves
3571          * list won't change. This doesn't solve the problem
3572          * of setting the slave's hw address while it is
3573          * transmitting, but the assumption is that the base
3574          * driver can handle that.
3575          *
3576          * TODO: figure out a way to safely iterate the slaves
3577          * list, but without holding a lock around the actual
3578          * call to the base driver.
3579          */
3580
3581         bond_for_each_slave(bond, slave) {
3582                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3583                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3584
3585                 if (slave_ops->ndo_set_mac_address == NULL) {
3586                         res = -EOPNOTSUPP;
3587                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3588                         goto unwind;
3589                 }
3590
3591                 res = dev_set_mac_address(slave->dev, addr);
3592                 if (res) {
3593                         /* TODO: consider downing the slave
3594                          * and retry ?
3595                          * User should expect communications
3596                          * breakage anyway until ARP finish
3597                          * updating, so...
3598                          */
3599                         pr_debug("err %d %s\n", res, slave->dev->name);
3600                         goto unwind;
3601                 }
3602         }
3603
3604         /* success */
3605         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3606         return 0;
3607
3608 unwind:
3609         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3610         tmp_sa.sa_family = bond_dev->type;
3611
3612         /* unwind from head to the slave that failed */
3613         bond_for_each_slave_continue_reverse(bond, slave) {
3614                 int tmp_res;
3615
3616                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3617                 if (tmp_res) {
3618                         pr_debug("unwind err %d dev %s\n",
3619                                  tmp_res, slave->dev->name);
3620                 }
3621         }
3622
3623         return res;
3624 }
3625
3626 /**
3627  * bond_xmit_slave_id - transmit skb through slave with slave_id
3628  * @bond: bonding device that is transmitting
3629  * @skb: buffer to transmit
3630  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3631  *
3632  * This function tries to transmit through slave with slave_id but in case
3633  * it fails, it tries to find the first available slave for transmission.
3634  * The skb is consumed in all cases, thus the function is void.
3635  */
3636 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3637 {
3638         struct slave *slave;
3639         int i = slave_id;
3640
3641         /* Here we start from the slave with slave_id */
3642         bond_for_each_slave_rcu(bond, slave) {
3643                 if (--i < 0) {
3644                         if (slave_can_tx(slave)) {
3645                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3646                                 return;
3647                         }
3648                 }
3649         }
3650
3651         /* Here we start from the first slave up to slave_id */
3652         i = slave_id;
3653         bond_for_each_slave_rcu(bond, slave) {
3654                 if (--i < 0)
3655                         break;
3656                 if (slave_can_tx(slave)) {
3657                         bond_dev_queue_xmit(bond, skb, slave->dev);
3658                         return;
3659                 }
3660         }
3661         /* no slave that can tx has been found */
3662         kfree_skb(skb);
3663 }
3664
3665 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3666 {
3667         struct bonding *bond = netdev_priv(bond_dev);
3668         struct iphdr *iph = ip_hdr(skb);
3669         struct slave *slave;
3670
3671         /*
3672          * Start with the curr_active_slave that joined the bond as the
3673          * default for sending IGMP traffic.  For failover purposes one
3674          * needs to maintain some consistency for the interface that will
3675          * send the join/membership reports.  The curr_active_slave found
3676          * will send all of this type of traffic.
3677          */
3678         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3679                 slave = rcu_dereference(bond->curr_active_slave);
3680                 if (slave && slave_can_tx(slave))
3681                         bond_dev_queue_xmit(bond, skb, slave->dev);
3682                 else
3683                         bond_xmit_slave_id(bond, skb, 0);
3684         } else {
3685                 bond_xmit_slave_id(bond, skb,
3686                                    bond->rr_tx_counter++ % bond->slave_cnt);
3687         }
3688
3689         return NETDEV_TX_OK;
3690 }
3691
3692 /*
3693  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3694  * the bond has a usable interface.
3695  */
3696 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3697 {
3698         struct bonding *bond = netdev_priv(bond_dev);
3699         struct slave *slave;
3700
3701         slave = rcu_dereference(bond->curr_active_slave);
3702         if (slave)
3703                 bond_dev_queue_xmit(bond, skb, slave->dev);
3704         else
3705                 kfree_skb(skb);
3706
3707         return NETDEV_TX_OK;
3708 }
3709
3710 /*
3711  * In bond_xmit_xor() , we determine the output device by using a pre-
3712  * determined xmit_hash_policy(), If the selected device is not enabled,
3713  * find the next active slave.
3714  */
3715 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3716 {
3717         struct bonding *bond = netdev_priv(bond_dev);
3718
3719         bond_xmit_slave_id(bond, skb,
3720                            bond->xmit_hash_policy(skb, bond->slave_cnt));
3721
3722         return NETDEV_TX_OK;
3723 }
3724
3725 /* in broadcast mode, we send everything to all usable interfaces. */
3726 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3727 {
3728         struct bonding *bond = netdev_priv(bond_dev);
3729         struct slave *slave = NULL;
3730
3731         bond_for_each_slave_rcu(bond, slave) {
3732                 if (bond_is_last_slave(bond, slave))
3733                         break;
3734                 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3735                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3736
3737                         if (!skb2) {
3738                                 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3739                                        bond_dev->name);
3740                                 continue;
3741                         }
3742                         /* bond_dev_queue_xmit always returns 0 */
3743                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3744                 }
3745         }
3746         if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3747                 bond_dev_queue_xmit(bond, skb, slave->dev);
3748         else
3749                 kfree_skb(skb);
3750
3751         return NETDEV_TX_OK;
3752 }
3753
3754 /*------------------------- Device initialization ---------------------------*/
3755
3756 static void bond_set_xmit_hash_policy(struct bonding *bond)
3757 {
3758         switch (bond->params.xmit_policy) {
3759         case BOND_XMIT_POLICY_LAYER23:
3760                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
3761                 break;
3762         case BOND_XMIT_POLICY_LAYER34:
3763                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
3764                 break;
3765         case BOND_XMIT_POLICY_LAYER2:
3766         default:
3767                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
3768                 break;
3769         }
3770 }
3771
3772 /*
3773  * Lookup the slave that corresponds to a qid
3774  */
3775 static inline int bond_slave_override(struct bonding *bond,
3776                                       struct sk_buff *skb)
3777 {
3778         struct slave *slave = NULL;
3779         struct slave *check_slave;
3780         int res = 1;
3781
3782         if (!skb->queue_mapping)
3783                 return 1;
3784
3785         /* Find out if any slaves have the same mapping as this skb. */
3786         bond_for_each_slave_rcu(bond, check_slave) {
3787                 if (check_slave->queue_id == skb->queue_mapping) {
3788                         slave = check_slave;
3789                         break;
3790                 }
3791         }
3792
3793         /* If the slave isn't UP, use default transmit policy. */
3794         if (slave && slave->queue_id && IS_UP(slave->dev) &&
3795             (slave->link == BOND_LINK_UP)) {
3796                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3797         }
3798
3799         return res;
3800 }
3801
3802
3803 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3804 {
3805         /*
3806          * This helper function exists to help dev_pick_tx get the correct
3807          * destination queue.  Using a helper function skips a call to
3808          * skb_tx_hash and will put the skbs in the queue we expect on their
3809          * way down to the bonding driver.
3810          */
3811         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3812
3813         /*
3814          * Save the original txq to restore before passing to the driver
3815          */
3816         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3817
3818         if (unlikely(txq >= dev->real_num_tx_queues)) {
3819                 do {
3820                         txq -= dev->real_num_tx_queues;
3821                 } while (txq >= dev->real_num_tx_queues);
3822         }
3823         return txq;
3824 }
3825
3826 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3827 {
3828         struct bonding *bond = netdev_priv(dev);
3829
3830         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3831                 if (!bond_slave_override(bond, skb))
3832                         return NETDEV_TX_OK;
3833         }
3834
3835         switch (bond->params.mode) {
3836         case BOND_MODE_ROUNDROBIN:
3837                 return bond_xmit_roundrobin(skb, dev);
3838         case BOND_MODE_ACTIVEBACKUP:
3839                 return bond_xmit_activebackup(skb, dev);
3840         case BOND_MODE_XOR:
3841                 return bond_xmit_xor(skb, dev);
3842         case BOND_MODE_BROADCAST:
3843                 return bond_xmit_broadcast(skb, dev);
3844         case BOND_MODE_8023AD:
3845                 return bond_3ad_xmit_xor(skb, dev);
3846         case BOND_MODE_ALB:
3847         case BOND_MODE_TLB:
3848                 return bond_alb_xmit(skb, dev);
3849         default:
3850                 /* Should never happen, mode already checked */
3851                 pr_err("%s: Error: Unknown bonding mode %d\n",
3852                        dev->name, bond->params.mode);
3853                 WARN_ON_ONCE(1);
3854                 kfree_skb(skb);
3855                 return NETDEV_TX_OK;
3856         }
3857 }
3858
3859 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3860 {
3861         struct bonding *bond = netdev_priv(dev);
3862         netdev_tx_t ret = NETDEV_TX_OK;
3863
3864         /*
3865          * If we risk deadlock from transmitting this in the
3866          * netpoll path, tell netpoll to queue the frame for later tx
3867          */
3868         if (is_netpoll_tx_blocked(dev))
3869                 return NETDEV_TX_BUSY;
3870
3871         rcu_read_lock();
3872         if (!list_empty(&bond->slave_list))
3873                 ret = __bond_start_xmit(skb, dev);
3874         else
3875                 kfree_skb(skb);
3876         rcu_read_unlock();
3877
3878         return ret;
3879 }
3880
3881 /*
3882  * set bond mode specific net device operations
3883  */
3884 void bond_set_mode_ops(struct bonding *bond, int mode)
3885 {
3886         struct net_device *bond_dev = bond->dev;
3887
3888         switch (mode) {
3889         case BOND_MODE_ROUNDROBIN:
3890                 break;
3891         case BOND_MODE_ACTIVEBACKUP:
3892                 break;
3893         case BOND_MODE_XOR:
3894                 bond_set_xmit_hash_policy(bond);
3895                 break;
3896         case BOND_MODE_BROADCAST:
3897                 break;
3898         case BOND_MODE_8023AD:
3899                 bond_set_xmit_hash_policy(bond);
3900                 break;
3901         case BOND_MODE_ALB:
3902                 /* FALLTHRU */
3903         case BOND_MODE_TLB:
3904                 break;
3905         default:
3906                 /* Should never happen, mode already checked */
3907                 pr_err("%s: Error: Unknown bonding mode %d\n",
3908                        bond_dev->name, mode);
3909                 break;
3910         }
3911 }
3912
3913 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3914                                      struct ethtool_cmd *ecmd)
3915 {
3916         struct bonding *bond = netdev_priv(bond_dev);
3917         unsigned long speed = 0;
3918         struct slave *slave;
3919
3920         ecmd->duplex = DUPLEX_UNKNOWN;
3921         ecmd->port = PORT_OTHER;
3922
3923         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3924          * do not need to check mode.  Though link speed might not represent
3925          * the true receive or transmit bandwidth (not all modes are symmetric)
3926          * this is an accurate maximum.
3927          */
3928         read_lock(&bond->lock);
3929         bond_for_each_slave(bond, slave) {
3930                 if (SLAVE_IS_OK(slave)) {
3931                         if (slave->speed != SPEED_UNKNOWN)
3932                                 speed += slave->speed;
3933                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3934                             slave->duplex != DUPLEX_UNKNOWN)
3935                                 ecmd->duplex = slave->duplex;
3936                 }
3937         }
3938         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3939         read_unlock(&bond->lock);
3940
3941         return 0;
3942 }
3943
3944 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3945                                      struct ethtool_drvinfo *drvinfo)
3946 {
3947         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3948         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3949         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3950                  BOND_ABI_VERSION);
3951 }
3952
3953 static const struct ethtool_ops bond_ethtool_ops = {
3954         .get_drvinfo            = bond_ethtool_get_drvinfo,
3955         .get_settings           = bond_ethtool_get_settings,
3956         .get_link               = ethtool_op_get_link,
3957 };
3958
3959 static const struct net_device_ops bond_netdev_ops = {
3960         .ndo_init               = bond_init,
3961         .ndo_uninit             = bond_uninit,
3962         .ndo_open               = bond_open,
3963         .ndo_stop               = bond_close,
3964         .ndo_start_xmit         = bond_start_xmit,
3965         .ndo_select_queue       = bond_select_queue,
3966         .ndo_get_stats64        = bond_get_stats,
3967         .ndo_do_ioctl           = bond_do_ioctl,
3968         .ndo_change_rx_flags    = bond_change_rx_flags,
3969         .ndo_set_rx_mode        = bond_set_rx_mode,
3970         .ndo_change_mtu         = bond_change_mtu,
3971         .ndo_set_mac_address    = bond_set_mac_address,
3972         .ndo_neigh_setup        = bond_neigh_setup,
3973         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
3974         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
3975 #ifdef CONFIG_NET_POLL_CONTROLLER
3976         .ndo_netpoll_setup      = bond_netpoll_setup,
3977         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
3978         .ndo_poll_controller    = bond_poll_controller,
3979 #endif
3980         .ndo_add_slave          = bond_enslave,
3981         .ndo_del_slave          = bond_release,
3982         .ndo_fix_features       = bond_fix_features,
3983 };
3984
3985 static const struct device_type bond_type = {
3986         .name = "bond",
3987 };
3988
3989 static void bond_destructor(struct net_device *bond_dev)
3990 {
3991         struct bonding *bond = netdev_priv(bond_dev);
3992         if (bond->wq)
3993                 destroy_workqueue(bond->wq);
3994         free_netdev(bond_dev);
3995 }
3996
3997 static void bond_setup(struct net_device *bond_dev)
3998 {
3999         struct bonding *bond = netdev_priv(bond_dev);
4000
4001         /* initialize rwlocks */
4002         rwlock_init(&bond->lock);
4003         rwlock_init(&bond->curr_slave_lock);
4004         INIT_LIST_HEAD(&bond->slave_list);
4005         bond->params = bonding_defaults;
4006
4007         /* Initialize pointers */
4008         bond->dev = bond_dev;
4009
4010         /* Initialize the device entry points */
4011         ether_setup(bond_dev);
4012         bond_dev->netdev_ops = &bond_netdev_ops;
4013         bond_dev->ethtool_ops = &bond_ethtool_ops;
4014         bond_set_mode_ops(bond, bond->params.mode);
4015
4016         bond_dev->destructor = bond_destructor;
4017
4018         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4019
4020         /* Initialize the device options */
4021         bond_dev->tx_queue_len = 0;
4022         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4023         bond_dev->priv_flags |= IFF_BONDING;
4024         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4025
4026         /* At first, we block adding VLANs. That's the only way to
4027          * prevent problems that occur when adding VLANs over an
4028          * empty bond. The block will be removed once non-challenged
4029          * slaves are enslaved.
4030          */
4031         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4032
4033         /* don't acquire bond device's netif_tx_lock when
4034          * transmitting */
4035         bond_dev->features |= NETIF_F_LLTX;
4036
4037         /* By default, we declare the bond to be fully
4038          * VLAN hardware accelerated capable. Special
4039          * care is taken in the various xmit functions
4040          * when there are slaves that are not hw accel
4041          * capable
4042          */
4043
4044         bond_dev->hw_features = BOND_VLAN_FEATURES |
4045                                 NETIF_F_HW_VLAN_CTAG_TX |
4046                                 NETIF_F_HW_VLAN_CTAG_RX |
4047                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4048
4049         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4050         bond_dev->features |= bond_dev->hw_features;
4051 }
4052
4053 /*
4054 * Destroy a bonding device.
4055 * Must be under rtnl_lock when this function is called.
4056 */
4057 static void bond_uninit(struct net_device *bond_dev)
4058 {
4059         struct bonding *bond = netdev_priv(bond_dev);
4060         struct slave *slave, *tmp_slave;
4061
4062         bond_netpoll_cleanup(bond_dev);
4063
4064         /* Release the bonded slaves */
4065         list_for_each_entry_safe(slave, tmp_slave, &bond->slave_list, list)
4066                 __bond_release_one(bond_dev, slave->dev, true);
4067         pr_info("%s: released all slaves\n", bond_dev->name);
4068
4069         list_del(&bond->bond_list);
4070
4071         bond_debug_unregister(bond);
4072 }
4073
4074 /*------------------------- Module initialization ---------------------------*/
4075
4076 /*
4077  * Convert string input module parms.  Accept either the
4078  * number of the mode or its string name.  A bit complicated because
4079  * some mode names are substrings of other names, and calls from sysfs
4080  * may have whitespace in the name (trailing newlines, for example).
4081  */
4082 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4083 {
4084         int modeint = -1, i, rv;
4085         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4086
4087         for (p = (char *)buf; *p; p++)
4088                 if (!(isdigit(*p) || isspace(*p)))
4089                         break;
4090
4091         if (*p)
4092                 rv = sscanf(buf, "%20s", modestr);
4093         else
4094                 rv = sscanf(buf, "%d", &modeint);
4095
4096         if (!rv)
4097                 return -1;
4098
4099         for (i = 0; tbl[i].modename; i++) {
4100                 if (modeint == tbl[i].mode)
4101                         return tbl[i].mode;
4102                 if (strcmp(modestr, tbl[i].modename) == 0)
4103                         return tbl[i].mode;
4104         }
4105
4106         return -1;
4107 }
4108
4109 static int bond_check_params(struct bond_params *params)
4110 {
4111         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4112         int arp_all_targets_value;
4113
4114         /*
4115          * Convert string parameters.
4116          */
4117         if (mode) {
4118                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4119                 if (bond_mode == -1) {
4120                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4121                                mode == NULL ? "NULL" : mode);
4122                         return -EINVAL;
4123                 }
4124         }
4125
4126         if (xmit_hash_policy) {
4127                 if ((bond_mode != BOND_MODE_XOR) &&
4128                     (bond_mode != BOND_MODE_8023AD)) {
4129                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4130                                bond_mode_name(bond_mode));
4131                 } else {
4132                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4133                                                         xmit_hashtype_tbl);
4134                         if (xmit_hashtype == -1) {
4135                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4136                                        xmit_hash_policy == NULL ? "NULL" :
4137                                        xmit_hash_policy);
4138                                 return -EINVAL;
4139                         }
4140                 }
4141         }
4142
4143         if (lacp_rate) {
4144                 if (bond_mode != BOND_MODE_8023AD) {
4145                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4146                                 bond_mode_name(bond_mode));
4147                 } else {
4148                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4149                         if (lacp_fast == -1) {
4150                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4151                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4152                                 return -EINVAL;
4153                         }
4154                 }
4155         }
4156
4157         if (ad_select) {
4158                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4159                 if (params->ad_select == -1) {
4160                         pr_err("Error: Invalid ad_select \"%s\"\n",
4161                                ad_select == NULL ? "NULL" : ad_select);
4162                         return -EINVAL;
4163                 }
4164
4165                 if (bond_mode != BOND_MODE_8023AD) {
4166                         pr_warning("ad_select param only affects 802.3ad mode\n");
4167                 }
4168         } else {
4169                 params->ad_select = BOND_AD_STABLE;
4170         }
4171
4172         if (max_bonds < 0) {
4173                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4174                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4175                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4176         }
4177
4178         if (miimon < 0) {
4179                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4180                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4181                 miimon = BOND_LINK_MON_INTERV;
4182         }
4183
4184         if (updelay < 0) {
4185                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4186                            updelay, INT_MAX);
4187                 updelay = 0;
4188         }
4189
4190         if (downdelay < 0) {
4191                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4192                            downdelay, INT_MAX);
4193                 downdelay = 0;
4194         }
4195
4196         if ((use_carrier != 0) && (use_carrier != 1)) {
4197                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4198                            use_carrier);
4199                 use_carrier = 1;
4200         }
4201
4202         if (num_peer_notif < 0 || num_peer_notif > 255) {
4203                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4204                            num_peer_notif);
4205                 num_peer_notif = 1;
4206         }
4207
4208         /* reset values for 802.3ad */
4209         if (bond_mode == BOND_MODE_8023AD) {
4210                 if (!miimon) {
4211                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4212                         pr_warning("Forcing miimon to 100msec\n");
4213                         miimon = 100;
4214                 }
4215         }
4216
4217         if (tx_queues < 1 || tx_queues > 255) {
4218                 pr_warning("Warning: tx_queues (%d) should be between "
4219                            "1 and 255, resetting to %d\n",
4220                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4221                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4222         }
4223
4224         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4225                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4226                            "not of valid value (0/1), so it was set to "
4227                            "0\n", all_slaves_active);
4228                 all_slaves_active = 0;
4229         }
4230
4231         if (resend_igmp < 0 || resend_igmp > 255) {
4232                 pr_warning("Warning: resend_igmp (%d) should be between "
4233                            "0 and 255, resetting to %d\n",
4234                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4235                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4236         }
4237
4238         /* reset values for TLB/ALB */
4239         if ((bond_mode == BOND_MODE_TLB) ||
4240             (bond_mode == BOND_MODE_ALB)) {
4241                 if (!miimon) {
4242                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4243                         pr_warning("Forcing miimon to 100msec\n");
4244                         miimon = 100;
4245                 }
4246         }
4247
4248         if (bond_mode == BOND_MODE_ALB) {
4249                 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",
4250                           updelay);
4251         }
4252
4253         if (!miimon) {
4254                 if (updelay || downdelay) {
4255                         /* just warn the user the up/down delay will have
4256                          * no effect since miimon is zero...
4257                          */
4258                         pr_warning("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",
4259                                    updelay, downdelay);
4260                 }
4261         } else {
4262                 /* don't allow arp monitoring */
4263                 if (arp_interval) {
4264                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4265                                    miimon, arp_interval);
4266                         arp_interval = 0;
4267                 }
4268
4269                 if ((updelay % miimon) != 0) {
4270                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4271                                    updelay, miimon,
4272                                    (updelay / miimon) * miimon);
4273                 }
4274
4275                 updelay /= miimon;
4276
4277                 if ((downdelay % miimon) != 0) {
4278                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4279                                    downdelay, miimon,
4280                                    (downdelay / miimon) * miimon);
4281                 }
4282
4283                 downdelay /= miimon;
4284         }
4285
4286         if (arp_interval < 0) {
4287                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4288                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4289                 arp_interval = BOND_LINK_ARP_INTERV;
4290         }
4291
4292         for (arp_ip_count = 0, i = 0;
4293              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4294                 /* not complete check, but should be good enough to
4295                    catch mistakes */
4296                 __be32 ip = in_aton(arp_ip_target[i]);
4297                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4298                     ip == htonl(INADDR_BROADCAST)) {
4299                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4300                                    arp_ip_target[i]);
4301                         arp_interval = 0;
4302                 } else {
4303                         if (bond_get_targets_ip(arp_target, ip) == -1)
4304                                 arp_target[arp_ip_count++] = ip;
4305                         else
4306                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4307                                            &ip);
4308                 }
4309         }
4310
4311         if (arp_interval && !arp_ip_count) {
4312                 /* don't allow arping if no arp_ip_target given... */
4313                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4314                            arp_interval);
4315                 arp_interval = 0;
4316         }
4317
4318         if (arp_validate) {
4319                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4320                         pr_err("arp_validate only supported in active-backup mode\n");
4321                         return -EINVAL;
4322                 }
4323                 if (!arp_interval) {
4324                         pr_err("arp_validate requires arp_interval\n");
4325                         return -EINVAL;
4326                 }
4327
4328                 arp_validate_value = bond_parse_parm(arp_validate,
4329                                                      arp_validate_tbl);
4330                 if (arp_validate_value == -1) {
4331                         pr_err("Error: invalid arp_validate \"%s\"\n",
4332                                arp_validate == NULL ? "NULL" : arp_validate);
4333                         return -EINVAL;
4334                 }
4335         } else
4336                 arp_validate_value = 0;
4337
4338         arp_all_targets_value = 0;
4339         if (arp_all_targets) {
4340                 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4341                                                         arp_all_targets_tbl);
4342
4343                 if (arp_all_targets_value == -1) {
4344                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4345                                arp_all_targets);
4346                         arp_all_targets_value = 0;
4347                 }
4348         }
4349
4350         if (miimon) {
4351                 pr_info("MII link monitoring set to %d ms\n", miimon);
4352         } else if (arp_interval) {
4353                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4354                         arp_interval,
4355                         arp_validate_tbl[arp_validate_value].modename,
4356                         arp_ip_count);
4357
4358                 for (i = 0; i < arp_ip_count; i++)
4359                         pr_info(" %s", arp_ip_target[i]);
4360
4361                 pr_info("\n");
4362
4363         } else if (max_bonds) {
4364                 /* miimon and arp_interval not set, we need one so things
4365                  * work as expected, see bonding.txt for details
4366                  */
4367                 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");
4368         }
4369
4370         if (primary && !USES_PRIMARY(bond_mode)) {
4371                 /* currently, using a primary only makes sense
4372                  * in active backup, TLB or ALB modes
4373                  */
4374                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4375                            primary, bond_mode_name(bond_mode));
4376                 primary = NULL;
4377         }
4378
4379         if (primary && primary_reselect) {
4380                 primary_reselect_value = bond_parse_parm(primary_reselect,
4381                                                          pri_reselect_tbl);
4382                 if (primary_reselect_value == -1) {
4383                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4384                                primary_reselect ==
4385                                         NULL ? "NULL" : primary_reselect);
4386                         return -EINVAL;
4387                 }
4388         } else {
4389                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4390         }
4391
4392         if (fail_over_mac) {
4393                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4394                                                       fail_over_mac_tbl);
4395                 if (fail_over_mac_value == -1) {
4396                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4397                                arp_validate == NULL ? "NULL" : arp_validate);
4398                         return -EINVAL;
4399                 }
4400
4401                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4402                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4403         } else {
4404                 fail_over_mac_value = BOND_FOM_NONE;
4405         }
4406
4407         /* fill params struct with the proper values */
4408         params->mode = bond_mode;
4409         params->xmit_policy = xmit_hashtype;
4410         params->miimon = miimon;
4411         params->num_peer_notif = num_peer_notif;
4412         params->arp_interval = arp_interval;
4413         params->arp_validate = arp_validate_value;
4414         params->arp_all_targets = arp_all_targets_value;
4415         params->updelay = updelay;
4416         params->downdelay = downdelay;
4417         params->use_carrier = use_carrier;
4418         params->lacp_fast = lacp_fast;
4419         params->primary[0] = 0;
4420         params->primary_reselect = primary_reselect_value;
4421         params->fail_over_mac = fail_over_mac_value;
4422         params->tx_queues = tx_queues;
4423         params->all_slaves_active = all_slaves_active;
4424         params->resend_igmp = resend_igmp;
4425         params->min_links = min_links;
4426         params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4427
4428         if (primary) {
4429                 strncpy(params->primary, primary, IFNAMSIZ);
4430                 params->primary[IFNAMSIZ - 1] = 0;
4431         }
4432
4433         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4434
4435         return 0;
4436 }
4437
4438 static struct lock_class_key bonding_netdev_xmit_lock_key;
4439 static struct lock_class_key bonding_netdev_addr_lock_key;
4440 static struct lock_class_key bonding_tx_busylock_key;
4441
4442 static void bond_set_lockdep_class_one(struct net_device *dev,
4443                                        struct netdev_queue *txq,
4444                                        void *_unused)
4445 {
4446         lockdep_set_class(&txq->_xmit_lock,
4447                           &bonding_netdev_xmit_lock_key);
4448 }
4449
4450 static void bond_set_lockdep_class(struct net_device *dev)
4451 {
4452         lockdep_set_class(&dev->addr_list_lock,
4453                           &bonding_netdev_addr_lock_key);
4454         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4455         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4456 }
4457
4458 /*
4459  * Called from registration process
4460  */
4461 static int bond_init(struct net_device *bond_dev)
4462 {
4463         struct bonding *bond = netdev_priv(bond_dev);
4464         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4465         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4466
4467         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4468
4469         /*
4470          * Initialize locks that may be required during
4471          * en/deslave operations.  All of the bond_open work
4472          * (of which this is part) should really be moved to
4473          * a phase prior to dev_open
4474          */
4475         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4476         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4477
4478         bond->wq = create_singlethread_workqueue(bond_dev->name);
4479         if (!bond->wq)
4480                 return -ENOMEM;
4481
4482         bond_set_lockdep_class(bond_dev);
4483
4484         list_add_tail(&bond->bond_list, &bn->dev_list);
4485
4486         bond_prepare_sysfs_group(bond);
4487
4488         bond_debug_register(bond);
4489
4490         /* Ensure valid dev_addr */
4491         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4492             bond_dev->addr_assign_type == NET_ADDR_PERM)
4493                 eth_hw_addr_random(bond_dev);
4494
4495         return 0;
4496 }
4497
4498 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4499 {
4500         if (tb[IFLA_ADDRESS]) {
4501                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4502                         return -EINVAL;
4503                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4504                         return -EADDRNOTAVAIL;
4505         }
4506         return 0;
4507 }
4508
4509 static unsigned int bond_get_num_tx_queues(void)
4510 {
4511         return tx_queues;
4512 }
4513
4514 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4515         .kind                   = "bond",
4516         .priv_size              = sizeof(struct bonding),
4517         .setup                  = bond_setup,
4518         .validate               = bond_validate,
4519         .get_num_tx_queues      = bond_get_num_tx_queues,
4520         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4521                                                              as for TX queues */
4522 };
4523
4524 /* Create a new bond based on the specified name and bonding parameters.
4525  * If name is NULL, obtain a suitable "bond%d" name for us.
4526  * Caller must NOT hold rtnl_lock; we need to release it here before we
4527  * set up our sysfs entries.
4528  */
4529 int bond_create(struct net *net, const char *name)
4530 {
4531         struct net_device *bond_dev;
4532         int res;
4533
4534         rtnl_lock();
4535
4536         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4537                                    name ? name : "bond%d",
4538                                    bond_setup, tx_queues);
4539         if (!bond_dev) {
4540                 pr_err("%s: eek! can't alloc netdev!\n", name);
4541                 rtnl_unlock();
4542                 return -ENOMEM;
4543         }
4544
4545         dev_net_set(bond_dev, net);
4546         bond_dev->rtnl_link_ops = &bond_link_ops;
4547
4548         res = register_netdevice(bond_dev);
4549
4550         netif_carrier_off(bond_dev);
4551
4552         rtnl_unlock();
4553         if (res < 0)
4554                 bond_destructor(bond_dev);
4555         return res;
4556 }
4557
4558 static int __net_init bond_net_init(struct net *net)
4559 {
4560         struct bond_net *bn = net_generic(net, bond_net_id);
4561
4562         bn->net = net;
4563         INIT_LIST_HEAD(&bn->dev_list);
4564
4565         bond_create_proc_dir(bn);
4566         bond_create_sysfs(bn);
4567
4568         return 0;
4569 }
4570
4571 static void __net_exit bond_net_exit(struct net *net)
4572 {
4573         struct bond_net *bn = net_generic(net, bond_net_id);
4574         struct bonding *bond, *tmp_bond;
4575         LIST_HEAD(list);
4576
4577         bond_destroy_sysfs(bn);
4578         bond_destroy_proc_dir(bn);
4579
4580         /* Kill off any bonds created after unregistering bond rtnl ops */
4581         rtnl_lock();
4582         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4583                 unregister_netdevice_queue(bond->dev, &list);
4584         unregister_netdevice_many(&list);
4585         rtnl_unlock();
4586 }
4587
4588 static struct pernet_operations bond_net_ops = {
4589         .init = bond_net_init,
4590         .exit = bond_net_exit,
4591         .id   = &bond_net_id,
4592         .size = sizeof(struct bond_net),
4593 };
4594
4595 static int __init bonding_init(void)
4596 {
4597         int i;
4598         int res;
4599
4600         pr_info("%s", bond_version);
4601
4602         res = bond_check_params(&bonding_defaults);
4603         if (res)
4604                 goto out;
4605
4606         res = register_pernet_subsys(&bond_net_ops);
4607         if (res)
4608                 goto out;
4609
4610         res = rtnl_link_register(&bond_link_ops);
4611         if (res)
4612                 goto err_link;
4613
4614         bond_create_debugfs();
4615
4616         for (i = 0; i < max_bonds; i++) {
4617                 res = bond_create(&init_net, NULL);
4618                 if (res)
4619                         goto err;
4620         }
4621
4622         register_netdevice_notifier(&bond_netdev_notifier);
4623 out:
4624         return res;
4625 err:
4626         rtnl_link_unregister(&bond_link_ops);
4627 err_link:
4628         unregister_pernet_subsys(&bond_net_ops);
4629         goto out;
4630
4631 }
4632
4633 static void __exit bonding_exit(void)
4634 {
4635         unregister_netdevice_notifier(&bond_netdev_notifier);
4636
4637         bond_destroy_debugfs();
4638
4639         rtnl_link_unregister(&bond_link_ops);
4640         unregister_pernet_subsys(&bond_net_ops);
4641
4642 #ifdef CONFIG_NET_POLL_CONTROLLER
4643         /*
4644          * Make sure we don't have an imbalance on our netpoll blocking
4645          */
4646         WARN_ON(atomic_read(&netpoll_block_tx));
4647 #endif
4648 }
4649
4650 module_init(bonding_init);
4651 module_exit(bonding_exit);
4652 MODULE_LICENSE("GPL");
4653 MODULE_VERSION(DRV_VERSION);
4654 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4655 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4656 MODULE_ALIAS_RTNL_LINK("bond");