Pileus Git - ~andy/linux/blob - 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_add_vlan - add a new vlan id on bond
 287  * @bond: bond that got the notification
 288  * @vlan_id: the vlan id to add
 289  *
 290  * Returns -ENOMEM if allocation failed.
 291  */
 292 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 293 {
 294         struct vlan_entry *vlan;
 295
 296         pr_debug("bond: %s, vlan id %d\n",
 297                  (bond ? bond->dev->name : "None"), vlan_id);
 298
 299         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 300         if (!vlan)
 301                 return -ENOMEM;
 302
 303         INIT_LIST_HEAD(&vlan->vlan_list);
 304         vlan->vlan_id = vlan_id;
 305
 306         write_lock_bh(&bond->lock);
 307
 308         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 309
 310         write_unlock_bh(&bond->lock);
 311
 312         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 313
 314         return 0;
 315 }
 316
 317 /**
 318  * bond_del_vlan - delete a vlan id from bond
 319  * @bond: bond that got the notification
 320  * @vlan_id: the vlan id to delete
 321  *
 322  * returns -ENODEV if @vlan_id was not found in @bond.
 323  */
 324 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 325 {
 326         struct vlan_entry *vlan;
 327         int res = -ENODEV;
 328
 329         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 330
 331         block_netpoll_tx();
 332         write_lock_bh(&bond->lock);
 333
 334         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 335                 if (vlan->vlan_id == vlan_id) {
 336                         list_del(&vlan->vlan_list);
 337
 338                         if (bond_is_lb(bond))
 339                                 bond_alb_clear_vlan(bond, vlan_id);
 340
 341                         pr_debug("removed VLAN ID %d from bond %s\n",
 342                                  vlan_id, bond->dev->name);
 343
 344                         kfree(vlan);
 345
 346                         res = 0;
 347                         goto out;
 348                 }
 349         }
 350
 351         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 352                  vlan_id, bond->dev->name);
 353
 354 out:
 355         write_unlock_bh(&bond->lock);
 356         unblock_netpoll_tx();
 357         return res;
 358 }
 359
 360 /**
 361  * bond_next_vlan - safely skip to the next item in the vlans list.
 362  * @bond: the bond we're working on
 363  * @curr: item we're advancing from
 364  *
 365  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 366  * or @curr->next otherwise (even if it is @curr itself again).
 367  *
 368  * Caller must hold bond->lock
 369  */
 370 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 371 {
 372         struct vlan_entry *next, *last;
 373
 374         if (list_empty(&bond->vlan_list))
 375                 return NULL;
 376
 377         if (!curr) {
 378                 next = list_entry(bond->vlan_list.next,
 379                                   struct vlan_entry, vlan_list);
 380         } else {
 381                 last = list_entry(bond->vlan_list.prev,
 382                                   struct vlan_entry, vlan_list);
 383                 if (last == curr) {
 384                         next = list_entry(bond->vlan_list.next,
 385                                           struct vlan_entry, vlan_list);
 386                 } else {
 387                         next = list_entry(curr->vlan_list.next,
 388                                           struct vlan_entry, vlan_list);
 389                 }
 390         }
 391
 392         return next;
 393 }
 394
 395 /**
 396  * bond_dev_queue_xmit - Prepare skb for xmit.
 397  *
 398  * @bond: bond device that got this skb for tx.
 399  * @skb: hw accel VLAN tagged skb to transmit
 400  * @slave_dev: slave that is supposed to xmit this skbuff
 401  */
 402 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 403                         struct net_device *slave_dev)
 404 {
 405         skb->dev = slave_dev;
 406
 407         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 408                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 409         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 410
 411         if (unlikely(netpoll_tx_running(bond->dev)))
 412                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 413         else
 414                 dev_queue_xmit(skb);
 415
 416         return 0;
 417 }
 418
 419 /*
 420  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 421  * We don't protect the slave list iteration with a lock because:
 422  * a. This operation is performed in IOCTL context,
 423  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 424  * c. Holding a lock with BH disabled while directly calling a base driver
 425  *    entry point is generally a BAD idea.
 426  *
 427  * The design of synchronization/protection for this operation in the 8021q
 428  * module is good for one or more VLAN devices over a single physical device
 429  * and cannot be extended for a teaming solution like bonding, so there is a
 430  * potential race condition here where a net device from the vlan group might
 431  * be referenced (either by a base driver or the 8021q code) while it is being
 432  * removed from the system. However, it turns out we're not making matters
 433  * worse, and if it works for regular VLAN usage it will work here too.
 434 */
 435
 436 /**
 437  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 438  * @bond_dev: bonding net device that got called
 439  * @vid: vlan id being added
 440  */
 441 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 442                                 __be16 proto, u16 vid)
 443 {
 444         struct bonding *bond = netdev_priv(bond_dev);
 445         struct slave *slave;
 446         int res;
 447
 448         bond_for_each_slave(bond, slave) {
 449                 res = vlan_vid_add(slave->dev, proto, vid);
 450                 if (res)
 451                         goto unwind;
 452         }
 453
 454         res = bond_add_vlan(bond, vid);
 455         if (res) {
 456                 pr_err("%s: Error: Failed to add vlan id %d\n",
 457                        bond_dev->name, vid);
 458                 return res;
 459         }
 460
 461         return 0;
 462
 463 unwind:
 464         /* unwind from head to the slave that failed */
 465         bond_for_each_slave_continue_reverse(bond, slave)
 466                 vlan_vid_del(slave->dev, proto, vid);
 467
 468         return res;
 469 }
 470
 471 /**
 472  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 473  * @bond_dev: bonding net device that got called
 474  * @vid: vlan id being removed
 475  */
 476 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 477                                  __be16 proto, u16 vid)
 478 {
 479         struct bonding *bond = netdev_priv(bond_dev);
 480         struct slave *slave;
 481         int res;
 482
 483         bond_for_each_slave(bond, slave)
 484                 vlan_vid_del(slave->dev, proto, vid);
 485
 486         res = bond_del_vlan(bond, vid);
 487         if (res) {
 488                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 489                        bond_dev->name, vid);
 490                 return res;
 491         }
 492
 493         return 0;
 494 }
 495
 496 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 497 {
 498         struct vlan_entry *vlan;
 499         int res;
 500
 501         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 502                 res = vlan_vid_add(slave_dev, htons(ETH_P_8021Q),
 503                                    vlan->vlan_id);
 504                 if (res)
 505                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 506                                    bond->dev->name, vlan->vlan_id,
 507                                    slave_dev->name);
 508         }
 509 }
 510
 511 static void bond_del_vlans_from_slave(struct bonding *bond,
 512                                       struct net_device *slave_dev)
 513 {
 514         struct vlan_entry *vlan;
 515
 516         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 517                 if (!vlan->vlan_id)
 518                         continue;
 519                 vlan_vid_del(slave_dev, htons(ETH_P_8021Q), vlan->vlan_id);
 520         }
 521 }
 522
 523 /*------------------------------- Link status -------------------------------*/
 524
 525 /*
 526  * Set the carrier state for the master according to the state of its
 527  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 528  * do special 802.3ad magic.
 529  *
 530  * Returns zero if carrier state does not change, nonzero if it does.
 531  */
 532 static int bond_set_carrier(struct bonding *bond)
 533 {
 534         struct slave *slave;
 535
 536         if (list_empty(&bond->slave_list))
 537                 goto down;
 538
 539         if (bond->params.mode == BOND_MODE_8023AD)
 540                 return bond_3ad_set_carrier(bond);
 541
 542         bond_for_each_slave(bond, slave) {
 543                 if (slave->link == BOND_LINK_UP) {
 544                         if (!netif_carrier_ok(bond->dev)) {
 545                                 netif_carrier_on(bond->dev);
 546                                 return 1;
 547                         }
 548                         return 0;
 549                 }
 550         }
 551
 552 down:
 553         if (netif_carrier_ok(bond->dev)) {
 554                 netif_carrier_off(bond->dev);
 555                 return 1;
 556         }
 557         return 0;
 558 }
 559
 560 /*
 561  * Get link speed and duplex from the slave's base driver
 562  * using ethtool. If for some reason the call fails or the
 563  * values are invalid, set speed and duplex to -1,
 564  * and return.
 565  */
 566 static void bond_update_speed_duplex(struct slave *slave)
 567 {
 568         struct net_device *slave_dev = slave->dev;
 569         struct ethtool_cmd ecmd;
 570         u32 slave_speed;
 571         int res;
 572
 573         slave->speed = SPEED_UNKNOWN;
 574         slave->duplex = DUPLEX_UNKNOWN;
 575
 576         res = __ethtool_get_settings(slave_dev, &ecmd);
 577         if (res < 0)
 578                 return;
 579
 580         slave_speed = ethtool_cmd_speed(&ecmd);
 581         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 582                 return;
 583
 584         switch (ecmd.duplex) {
 585         case DUPLEX_FULL:
 586         case DUPLEX_HALF:
 587                 break;
 588         default:
 589                 return;
 590         }
 591
 592         slave->speed = slave_speed;
 593         slave->duplex = ecmd.duplex;
 594
 595         return;
 596 }
 597
 598 /*
 599  * if <dev> supports MII link status reporting, check its link status.
 600  *
 601  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 602  * depending upon the setting of the use_carrier parameter.
 603  *
 604  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 605  * can't tell and just pretend it is), or 0, meaning that the link is
 606  * down.
 607  *
 608  * If reporting is non-zero, instead of faking link up, return -1 if
 609  * both ETHTOOL and MII ioctls fail (meaning the device does not
 610  * support them).  If use_carrier is set, return whatever it says.
 611  * It'd be nice if there was a good way to tell if a driver supports
 612  * netif_carrier, but there really isn't.
 613  */
 614 static int bond_check_dev_link(struct bonding *bond,
 615                                struct net_device *slave_dev, int reporting)
 616 {
 617         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 618         int (*ioctl)(struct net_device *, struct ifreq *, int);
 619         struct ifreq ifr;
 620         struct mii_ioctl_data *mii;
 621
 622         if (!reporting && !netif_running(slave_dev))
 623                 return 0;
 624
 625         if (bond->params.use_carrier)
 626                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 627
 628         /* Try to get link status using Ethtool first. */
 629         if (slave_dev->ethtool_ops->get_link)
 630                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 631                         BMSR_LSTATUS : 0;
 632
 633         /* Ethtool can't be used, fallback to MII ioctls. */
 634         ioctl = slave_ops->ndo_do_ioctl;
 635         if (ioctl) {
 636                 /* TODO: set pointer to correct ioctl on a per team member */
 637                 /*       bases to make this more efficient. that is, once  */
 638                 /*       we determine the correct ioctl, we will always    */
 639                 /*       call it and not the others for that team          */
 640                 /*       member.                                           */
 641
 642                 /*
 643                  * We cannot assume that SIOCGMIIPHY will also read a
 644                  * register; not all network drivers (e.g., e100)
 645                  * support that.
 646                  */
 647
 648                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 649                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 650                 mii = if_mii(&ifr);
 651                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 652                         mii->reg_num = MII_BMSR;
 653                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 654                                 return mii->val_out & BMSR_LSTATUS;
 655                 }
 656         }
 657
 658         /*
 659          * If reporting, report that either there's no dev->do_ioctl,
 660          * or both SIOCGMIIREG and get_link failed (meaning that we
 661          * cannot report link status).  If not reporting, pretend
 662          * we're ok.
 663          */
 664         return reporting ? -1 : BMSR_LSTATUS;
 665 }
 666
 667 /*----------------------------- Multicast list ------------------------------*/
 668
 669 /*
 670  * Push the promiscuity flag down to appropriate slaves
 671  */
 672 static int bond_set_promiscuity(struct bonding *bond, int inc)
 673 {
 674         int err = 0;
 675         if (USES_PRIMARY(bond->params.mode)) {
 676                 /* write lock already acquired */
 677                 if (bond->curr_active_slave) {
 678                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 679                                                   inc);
 680                 }
 681         } else {
 682                 struct slave *slave;
 683
 684                 bond_for_each_slave(bond, slave) {
 685                         err = dev_set_promiscuity(slave->dev, inc);
 686                         if (err)
 687                                 return err;
 688                 }
 689         }
 690         return err;
 691 }
 692
 693 /*
 694  * Push the allmulti flag down to all slaves
 695  */
 696 static int bond_set_allmulti(struct bonding *bond, int inc)
 697 {
 698         int err = 0;
 699         if (USES_PRIMARY(bond->params.mode)) {
 700                 /* write lock already acquired */
 701                 if (bond->curr_active_slave) {
 702                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 703                                                inc);
 704                 }
 705         } else {
 706                 struct slave *slave;
 707
 708                 bond_for_each_slave(bond, slave) {
 709                         err = dev_set_allmulti(slave->dev, inc);
 710                         if (err)
 711                                 return err;
 712                 }
 713         }
 714         return err;
 715 }
 716
 717 /*
 718  * Retrieve the list of registered multicast addresses for the bonding
 719  * device and retransmit an IGMP JOIN request to the current active
 720  * slave.
 721  */
 722 static void bond_resend_igmp_join_requests(struct bonding *bond)
 723 {
 724         if (!rtnl_trylock()) {
 725                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 726                 return;
 727         }
 728         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 729         rtnl_unlock();
 730
 731         /* We use curr_slave_lock to protect against concurrent access to
 732          * igmp_retrans from multiple running instances of this function and
 733          * bond_change_active_slave
 734          */
 735         write_lock_bh(&bond->curr_slave_lock);
 736         if (bond->igmp_retrans > 1) {
 737                 bond->igmp_retrans--;
 738                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 739         }
 740         write_unlock_bh(&bond->curr_slave_lock);
 741 }
 742
 743 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 744 {
 745         struct bonding *bond = container_of(work, struct bonding,
 746                                             mcast_work.work);
 747
 748         bond_resend_igmp_join_requests(bond);
 749 }
 750
 751 /* Flush bond's hardware addresses from slave
 752  */
 753 static void bond_hw_addr_flush(struct net_device *bond_dev,
 754                                struct net_device *slave_dev)
 755 {
 756         struct bonding *bond = netdev_priv(bond_dev);
 757
 758         dev_uc_unsync(slave_dev, bond_dev);
 759         dev_mc_unsync(slave_dev, bond_dev);
 760
 761         if (bond->params.mode == BOND_MODE_8023AD) {
 762                 /* del lacpdu mc addr from mc list */
 763                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 764
 765                 dev_mc_del(slave_dev, lacpdu_multicast);
 766         }
 767 }
 768
 769 /*--------------------------- Active slave change ---------------------------*/
 770
 771 /* Update the hardware address list and promisc/allmulti for the new and
 772  * old active slaves (if any).  Modes that are !USES_PRIMARY keep all
 773  * slaves up date at all times; only the USES_PRIMARY modes need to call
 774  * this function to swap these settings during a failover.
 775  */
 776 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 777                               struct slave *old_active)
 778 {
 779         if (old_active) {
 780                 if (bond->dev->flags & IFF_PROMISC)
 781                         dev_set_promiscuity(old_active->dev, -1);
 782
 783                 if (bond->dev->flags & IFF_ALLMULTI)
 784                         dev_set_allmulti(old_active->dev, -1);
 785
 786                 bond_hw_addr_flush(bond->dev, old_active->dev);
 787         }
 788
 789         if (new_active) {
 790                 /* FIXME: Signal errors upstream. */
 791                 if (bond->dev->flags & IFF_PROMISC)
 792                         dev_set_promiscuity(new_active->dev, 1);
 793
 794                 if (bond->dev->flags & IFF_ALLMULTI)
 795                         dev_set_allmulti(new_active->dev, 1);
 796
 797                 netif_addr_lock_bh(bond->dev);
 798                 dev_uc_sync(new_active->dev, bond->dev);
 799                 dev_mc_sync(new_active->dev, bond->dev);
 800                 netif_addr_unlock_bh(bond->dev);
 801         }
 802 }
 803
 804 /**
 805  * bond_set_dev_addr - clone slave's address to bond
 806  * @bond_dev: bond net device
 807  * @slave_dev: slave net device
 808  *
 809  * Should be called with RTNL held.
 810  */
 811 static void bond_set_dev_addr(struct net_device *bond_dev,
 812                               struct net_device *slave_dev)
 813 {
 814         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 815                  bond_dev, slave_dev, slave_dev->addr_len);
 816         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 817         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 818         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 819 }
 820
 821 /*
 822  * bond_do_fail_over_mac
 823  *
 824  * Perform special MAC address swapping for fail_over_mac settings
 825  *
 826  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 827  */
 828 static void bond_do_fail_over_mac(struct bonding *bond,
 829                                   struct slave *new_active,
 830                                   struct slave *old_active)
 831         __releases(&bond->curr_slave_lock)
 832         __releases(&bond->lock)
 833         __acquires(&bond->lock)
 834         __acquires(&bond->curr_slave_lock)
 835 {
 836         u8 tmp_mac[ETH_ALEN];
 837         struct sockaddr saddr;
 838         int rv;
 839
 840         switch (bond->params.fail_over_mac) {
 841         case BOND_FOM_ACTIVE:
 842                 if (new_active) {
 843                         write_unlock_bh(&bond->curr_slave_lock);
 844                         read_unlock(&bond->lock);
 845                         bond_set_dev_addr(bond->dev, new_active->dev);
 846                         read_lock(&bond->lock);
 847                         write_lock_bh(&bond->curr_slave_lock);
 848                 }
 849                 break;
 850         case BOND_FOM_FOLLOW:
 851                 /*
 852                  * if new_active && old_active, swap them
 853                  * if just old_active, do nothing (going to no active slave)
 854                  * if just new_active, set new_active to bond's MAC
 855                  */
 856                 if (!new_active)
 857                         return;
 858
 859                 write_unlock_bh(&bond->curr_slave_lock);
 860                 read_unlock(&bond->lock);
 861
 862                 if (old_active) {
 863                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 864                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 865                                ETH_ALEN);
 866                         saddr.sa_family = new_active->dev->type;
 867                 } else {
 868                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 869                         saddr.sa_family = bond->dev->type;
 870                 }
 871
 872                 rv = dev_set_mac_address(new_active->dev, &saddr);
 873                 if (rv) {
 874                         pr_err("%s: Error %d setting MAC of slave %s\n",
 875                                bond->dev->name, -rv, new_active->dev->name);
 876                         goto out;
 877                 }
 878
 879                 if (!old_active)
 880                         goto out;
 881
 882                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 883                 saddr.sa_family = old_active->dev->type;
 884
 885                 rv = dev_set_mac_address(old_active->dev, &saddr);
 886                 if (rv)
 887                         pr_err("%s: Error %d setting MAC of slave %s\n",
 888                                bond->dev->name, -rv, new_active->dev->name);
 889 out:
 890                 read_lock(&bond->lock);
 891                 write_lock_bh(&bond->curr_slave_lock);
 892                 break;
 893         default:
 894                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 895                        bond->dev->name, bond->params.fail_over_mac);
 896                 break;
 897         }
 898
 899 }
 900
 901 static bool bond_should_change_active(struct bonding *bond)
 902 {
 903         struct slave *prim = bond->primary_slave;
 904         struct slave *curr = bond->curr_active_slave;
 905
 906         if (!prim || !curr || curr->link != BOND_LINK_UP)
 907                 return true;
 908         if (bond->force_primary) {
 909                 bond->force_primary = false;
 910                 return true;
 911         }
 912         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 913             (prim->speed < curr->speed ||
 914              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 915                 return false;
 916         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 917                 return false;
 918         return true;
 919 }
 920
 921 /**
 922  * find_best_interface - select the best available slave to be the active one
 923  * @bond: our bonding struct
 924  *
 925  * Warning: Caller must hold curr_slave_lock for writing.
 926  */
 927 static struct slave *bond_find_best_slave(struct bonding *bond)
 928 {
 929         struct slave *new_active, *old_active;
 930         struct slave *bestslave = NULL;
 931         int mintime = bond->params.updelay;
 932         int i;
 933
 934         new_active = bond->curr_active_slave;
 935
 936         if (!new_active) { /* there were no active slaves left */
 937                 new_active = bond_first_slave(bond);
 938                 if (!new_active)
 939                         return NULL; /* still no slave, return NULL */
 940         }
 941
 942         if ((bond->primary_slave) &&
 943             bond->primary_slave->link == BOND_LINK_UP &&
 944             bond_should_change_active(bond)) {
 945                 new_active = bond->primary_slave;
 946         }
 947
 948         /* remember where to stop iterating over the slaves */
 949         old_active = new_active;
 950
 951         bond_for_each_slave_from(bond, new_active, i, old_active) {
 952                 if (new_active->link == BOND_LINK_UP) {
 953                         return new_active;
 954                 } else if (new_active->link == BOND_LINK_BACK &&
 955                            IS_UP(new_active->dev)) {
 956                         /* link up, but waiting for stabilization */
 957                         if (new_active->delay < mintime) {
 958                                 mintime = new_active->delay;
 959                                 bestslave = new_active;
 960                         }
 961                 }
 962         }
 963
 964         return bestslave;
 965 }
 966
 967 static bool bond_should_notify_peers(struct bonding *bond)
 968 {
 969         struct slave *slave = bond->curr_active_slave;
 970
 971         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
 972                  bond->dev->name, slave ? slave->dev->name : "NULL");
 973
 974         if (!slave || !bond->send_peer_notif ||
 975             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 976                 return false;
 977
 978         bond->send_peer_notif--;
 979         return true;
 980 }
 981
 982 /**
 983  * change_active_interface - change the active slave into the specified one
 984  * @bond: our bonding struct
 985  * @new: the new slave to make the active one
 986  *
 987  * Set the new slave to the bond's settings and unset them on the old
 988  * curr_active_slave.
 989  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 990  *
 991  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 992  * because it is apparently the best available slave we have, even though its
 993  * updelay hasn't timed out yet.
 994  *
 995  * If new_active is not NULL, caller must hold bond->lock for read and
 996  * curr_slave_lock for write_bh.
 997  */
 998 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 999 {
1000         struct slave *old_active = bond->curr_active_slave;
1001
1002         if (old_active == new_active)
1003                 return;
1004
1005         if (new_active) {
1006                 new_active->jiffies = jiffies;
1007
1008                 if (new_active->link == BOND_LINK_BACK) {
1009                         if (USES_PRIMARY(bond->params.mode)) {
1010                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1011                                         bond->dev->name, new_active->dev->name,
1012                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1013                         }
1014
1015                         new_active->delay = 0;
1016                         new_active->link = BOND_LINK_UP;
1017
1018                         if (bond->params.mode == BOND_MODE_8023AD)
1019                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1020
1021                         if (bond_is_lb(bond))
1022                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1023                 } else {
1024                         if (USES_PRIMARY(bond->params.mode)) {
1025                                 pr_info("%s: making interface %s the new active one.\n",
1026                                         bond->dev->name, new_active->dev->name);
1027                         }
1028                 }
1029         }
1030
1031         if (USES_PRIMARY(bond->params.mode))
1032                 bond_hw_addr_swap(bond, new_active, old_active);
1033
1034         if (bond_is_lb(bond)) {
1035                 bond_alb_handle_active_change(bond, new_active);
1036                 if (old_active)
1037                         bond_set_slave_inactive_flags(old_active);
1038                 if (new_active)
1039                         bond_set_slave_active_flags(new_active);
1040         } else {
1041                 rcu_assign_pointer(bond->curr_active_slave, new_active);
1042         }
1043
1044         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1045                 if (old_active)
1046                         bond_set_slave_inactive_flags(old_active);
1047
1048                 if (new_active) {
1049                         bool should_notify_peers = false;
1050
1051                         bond_set_slave_active_flags(new_active);
1052
1053                         if (bond->params.fail_over_mac)
1054                                 bond_do_fail_over_mac(bond, new_active,
1055                                                       old_active);
1056
1057                         if (netif_running(bond->dev)) {
1058                                 bond->send_peer_notif =
1059                                         bond->params.num_peer_notif;
1060                                 should_notify_peers =
1061                                         bond_should_notify_peers(bond);
1062                         }
1063
1064                         write_unlock_bh(&bond->curr_slave_lock);
1065                         read_unlock(&bond->lock);
1066
1067                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1068                         if (should_notify_peers)
1069                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1070                                                          bond->dev);
1071
1072                         read_lock(&bond->lock);
1073                         write_lock_bh(&bond->curr_slave_lock);
1074                 }
1075         }
1076
1077         /* resend IGMP joins since active slave has changed or
1078          * all were sent on curr_active_slave.
1079          * resend only if bond is brought up with the affected
1080          * bonding modes and the retransmission is enabled */
1081         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1082             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1083              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1084                 bond->igmp_retrans = bond->params.resend_igmp;
1085                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1086         }
1087 }
1088
1089 /**
1090  * bond_select_active_slave - select a new active slave, if needed
1091  * @bond: our bonding struct
1092  *
1093  * This functions should be called when one of the following occurs:
1094  * - The old curr_active_slave has been released or lost its link.
1095  * - The primary_slave has got its link back.
1096  * - A slave has got its link back and there's no old curr_active_slave.
1097  *
1098  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1099  */
1100 void bond_select_active_slave(struct bonding *bond)
1101 {
1102         struct slave *best_slave;
1103         int rv;
1104
1105         best_slave = bond_find_best_slave(bond);
1106         if (best_slave != bond->curr_active_slave) {
1107                 bond_change_active_slave(bond, best_slave);
1108                 rv = bond_set_carrier(bond);
1109                 if (!rv)
1110                         return;
1111
1112                 if (netif_carrier_ok(bond->dev)) {
1113                         pr_info("%s: first active interface up!\n",
1114                                 bond->dev->name);
1115                 } else {
1116                         pr_info("%s: now running without any active interface !\n",
1117                                 bond->dev->name);
1118                 }
1119         }
1120 }
1121
1122 /*--------------------------- slave list handling ---------------------------*/
1123
1124 /*
1125  * This function attaches the slave to the end of list.
1126  *
1127  * bond->lock held for writing by caller.
1128  */
1129 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1130 {
1131         list_add_tail_rcu(&new_slave->list, &bond->slave_list);
1132         bond->slave_cnt++;
1133 }
1134
1135 /*
1136  * This function detaches the slave from the list.
1137  * WARNING: no check is made to verify if the slave effectively
1138  * belongs to <bond>.
1139  * Nothing is freed on return, structures are just unchained.
1140  * If any slave pointer in bond was pointing to <slave>,
1141  * it should be changed by the calling function.
1142  *
1143  * bond->lock held for writing by caller.
1144  */
1145 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1146 {
1147         list_del_rcu(&slave->list);
1148         bond->slave_cnt--;
1149 }
1150
1151 #ifdef CONFIG_NET_POLL_CONTROLLER
1152 static inline int slave_enable_netpoll(struct slave *slave)
1153 {
1154         struct netpoll *np;
1155         int err = 0;
1156
1157         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1158         err = -ENOMEM;
1159         if (!np)
1160                 goto out;
1161
1162         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1163         if (err) {
1164                 kfree(np);
1165                 goto out;
1166         }
1167         slave->np = np;
1168 out:
1169         return err;
1170 }
1171 static inline void slave_disable_netpoll(struct slave *slave)
1172 {
1173         struct netpoll *np = slave->np;
1174
1175         if (!np)
1176                 return;
1177
1178         slave->np = NULL;
1179         __netpoll_free_async(np);
1180 }
1181 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1182 {
1183         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1184                 return false;
1185         if (!slave_dev->netdev_ops->ndo_poll_controller)
1186                 return false;
1187         return true;
1188 }
1189
1190 static void bond_poll_controller(struct net_device *bond_dev)
1191 {
1192 }
1193
1194 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1195 {
1196         struct bonding *bond = netdev_priv(bond_dev);
1197         struct slave *slave;
1198
1199         bond_for_each_slave(bond, slave)
1200                 if (IS_UP(slave->dev))
1201                         slave_disable_netpoll(slave);
1202 }
1203
1204 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1205 {
1206         struct bonding *bond = netdev_priv(dev);
1207         struct slave *slave;
1208         int err = 0;
1209
1210         bond_for_each_slave(bond, slave) {
1211                 err = slave_enable_netpoll(slave);
1212                 if (err) {
1213                         bond_netpoll_cleanup(dev);
1214                         break;
1215                 }
1216         }
1217         return err;
1218 }
1219 #else
1220 static inline int slave_enable_netpoll(struct slave *slave)
1221 {
1222         return 0;
1223 }
1224 static inline void slave_disable_netpoll(struct slave *slave)
1225 {
1226 }
1227 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1228 {
1229 }
1230 #endif
1231
1232 /*---------------------------------- IOCTL ----------------------------------*/
1233
1234 static netdev_features_t bond_fix_features(struct net_device *dev,
1235         netdev_features_t features)
1236 {
1237         struct slave *slave;
1238         struct bonding *bond = netdev_priv(dev);
1239         netdev_features_t mask;
1240
1241         read_lock(&bond->lock);
1242
1243         if (list_empty(&bond->slave_list)) {
1244                 /* Disable adding VLANs to empty bond. But why? --mq */
1245                 features |= NETIF_F_VLAN_CHALLENGED;
1246                 goto out;
1247         }
1248
1249         mask = features;
1250         features &= ~NETIF_F_ONE_FOR_ALL;
1251         features |= NETIF_F_ALL_FOR_ALL;
1252
1253         bond_for_each_slave(bond, slave) {
1254                 features = netdev_increment_features(features,
1255                                                      slave->dev->features,
1256                                                      mask);
1257         }
1258         features = netdev_add_tso_features(features, mask);
1259
1260 out:
1261         read_unlock(&bond->lock);
1262         return features;
1263 }
1264
1265 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1266                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1267                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1268
1269 static void bond_compute_features(struct bonding *bond)
1270 {
1271         struct slave *slave;
1272         struct net_device *bond_dev = bond->dev;
1273         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1274         unsigned short max_hard_header_len = ETH_HLEN;
1275         unsigned int gso_max_size = GSO_MAX_SIZE;
1276         u16 gso_max_segs = GSO_MAX_SEGS;
1277         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1278
1279         read_lock(&bond->lock);
1280
1281         if (list_empty(&bond->slave_list))
1282                 goto done;
1283
1284         bond_for_each_slave(bond, slave) {
1285                 vlan_features = netdev_increment_features(vlan_features,
1286                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1287
1288                 dst_release_flag &= slave->dev->priv_flags;
1289                 if (slave->dev->hard_header_len > max_hard_header_len)
1290                         max_hard_header_len = slave->dev->hard_header_len;
1291
1292                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1293                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1294         }
1295
1296 done:
1297         bond_dev->vlan_features = vlan_features;
1298         bond_dev->hard_header_len = max_hard_header_len;
1299         bond_dev->gso_max_segs = gso_max_segs;
1300         netif_set_gso_max_size(bond_dev, gso_max_size);
1301
1302         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1303         bond_dev->priv_flags = flags | dst_release_flag;
1304
1305         read_unlock(&bond->lock);
1306
1307         netdev_change_features(bond_dev);
1308 }
1309
1310 static void bond_setup_by_slave(struct net_device *bond_dev,
1311                                 struct net_device *slave_dev)
1312 {
1313         bond_dev->header_ops        = slave_dev->header_ops;
1314
1315         bond_dev->type              = slave_dev->type;
1316         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1317         bond_dev->addr_len          = slave_dev->addr_len;
1318
1319         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1320                 slave_dev->addr_len);
1321 }
1322
1323 /* On bonding slaves other than the currently active slave, suppress
1324  * duplicates except for alb non-mcast/bcast.
1325  */
1326 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1327                                             struct slave *slave,
1328                                             struct bonding *bond)
1329 {
1330         if (bond_is_slave_inactive(slave)) {
1331                 if (bond->params.mode == BOND_MODE_ALB &&
1332                     skb->pkt_type != PACKET_BROADCAST &&
1333                     skb->pkt_type != PACKET_MULTICAST)
1334                         return false;
1335                 return true;
1336         }
1337         return false;
1338 }
1339
1340 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1341 {
1342         struct sk_buff *skb = *pskb;
1343         struct slave *slave;
1344         struct bonding *bond;
1345         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1346                           struct slave *);
1347         int ret = RX_HANDLER_ANOTHER;
1348
1349         skb = skb_share_check(skb, GFP_ATOMIC);
1350         if (unlikely(!skb))
1351                 return RX_HANDLER_CONSUMED;
1352
1353         *pskb = skb;
1354
1355         slave = bond_slave_get_rcu(skb->dev);
1356         bond = slave->bond;
1357
1358         if (bond->params.arp_interval)
1359                 slave->dev->last_rx = jiffies;
1360
1361         recv_probe = ACCESS_ONCE(bond->recv_probe);
1362         if (recv_probe) {
1363                 ret = recv_probe(skb, bond, slave);
1364                 if (ret == RX_HANDLER_CONSUMED) {
1365                         consume_skb(skb);
1366                         return ret;
1367                 }
1368         }
1369
1370         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1371                 return RX_HANDLER_EXACT;
1372         }
1373
1374         skb->dev = bond->dev;
1375
1376         if (bond->params.mode == BOND_MODE_ALB &&
1377             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1378             skb->pkt_type == PACKET_HOST) {
1379
1380                 if (unlikely(skb_cow_head(skb,
1381                                           skb->data - skb_mac_header(skb)))) {
1382                         kfree_skb(skb);
1383                         return RX_HANDLER_CONSUMED;
1384                 }
1385                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1386         }
1387
1388         return ret;
1389 }
1390
1391 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1392                                       struct net_device *slave_dev)
1393 {
1394         int err;
1395
1396         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1397         if (err)
1398                 return err;
1399         slave_dev->flags |= IFF_SLAVE;
1400         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1401         return 0;
1402 }
1403
1404 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1405                                   struct net_device *slave_dev)
1406 {
1407         netdev_upper_dev_unlink(slave_dev, bond_dev);
1408         slave_dev->flags &= ~IFF_SLAVE;
1409         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1410 }
1411
1412 /* enslave device <slave> to bond device <master> */
1413 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1414 {
1415         struct bonding *bond = netdev_priv(bond_dev);
1416         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1417         struct slave *new_slave = NULL;
1418         struct sockaddr addr;
1419         int link_reporting;
1420         int res = 0, i;
1421
1422         if (!bond->params.use_carrier &&
1423             slave_dev->ethtool_ops->get_link == NULL &&
1424             slave_ops->ndo_do_ioctl == NULL) {
1425                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1426                            bond_dev->name, slave_dev->name);
1427         }
1428
1429         /* already enslaved */
1430         if (slave_dev->flags & IFF_SLAVE) {
1431                 pr_debug("Error, Device was already enslaved\n");
1432                 return -EBUSY;
1433         }
1434
1435         /* vlan challenged mutual exclusion */
1436         /* no need to lock since we're protected by rtnl_lock */
1437         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1438                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1439                 if (vlan_uses_dev(bond_dev)) {
1440                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1441                                bond_dev->name, slave_dev->name, bond_dev->name);
1442                         return -EPERM;
1443                 } else {
1444                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1445                                    bond_dev->name, slave_dev->name,
1446                                    slave_dev->name, bond_dev->name);
1447                 }
1448         } else {
1449                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1450         }
1451
1452         /*
1453          * Old ifenslave binaries are no longer supported.  These can
1454          * be identified with moderate accuracy by the state of the slave:
1455          * the current ifenslave will set the interface down prior to
1456          * enslaving it; the old ifenslave will not.
1457          */
1458         if ((slave_dev->flags & IFF_UP)) {
1459                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1460                        slave_dev->name);
1461                 res = -EPERM;
1462                 goto err_undo_flags;
1463         }
1464
1465         /* set bonding device ether type by slave - bonding netdevices are
1466          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1467          * there is a need to override some of the type dependent attribs/funcs.
1468          *
1469          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1470          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1471          */
1472         if (list_empty(&bond->slave_list)) {
1473                 if (bond_dev->type != slave_dev->type) {
1474                         pr_debug("%s: change device type from %d to %d\n",
1475                                  bond_dev->name,
1476                                  bond_dev->type, slave_dev->type);
1477
1478                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1479                                                        bond_dev);
1480                         res = notifier_to_errno(res);
1481                         if (res) {
1482                                 pr_err("%s: refused to change device type\n",
1483                                        bond_dev->name);
1484                                 res = -EBUSY;
1485                                 goto err_undo_flags;
1486                         }
1487
1488                         /* Flush unicast and multicast addresses */
1489                         dev_uc_flush(bond_dev);
1490                         dev_mc_flush(bond_dev);
1491
1492                         if (slave_dev->type != ARPHRD_ETHER)
1493                                 bond_setup_by_slave(bond_dev, slave_dev);
1494                         else {
1495                                 ether_setup(bond_dev);
1496                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1497                         }
1498
1499                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1500                                                  bond_dev);
1501                 }
1502         } else if (bond_dev->type != slave_dev->type) {
1503                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1504                        slave_dev->name,
1505                        slave_dev->type, bond_dev->type);
1506                 res = -EINVAL;
1507                 goto err_undo_flags;
1508         }
1509
1510         if (slave_ops->ndo_set_mac_address == NULL) {
1511                 if (list_empty(&bond->slave_list)) {
1512                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1513                                    bond_dev->name);
1514                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1515                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1516                         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",
1517                                bond_dev->name);
1518                         res = -EOPNOTSUPP;
1519                         goto err_undo_flags;
1520                 }
1521         }
1522
1523         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1524
1525         /* If this is the first slave, then we need to set the master's hardware
1526          * address to be the same as the slave's. */
1527         if (list_empty(&bond->slave_list) &&
1528             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1529                 bond_set_dev_addr(bond->dev, slave_dev);
1530
1531         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1532         if (!new_slave) {
1533                 res = -ENOMEM;
1534                 goto err_undo_flags;
1535         }
1536         INIT_LIST_HEAD(&new_slave->list);
1537         /*
1538          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1539          * is set via sysfs or module option if desired.
1540          */
1541         new_slave->queue_id = 0;
1542
1543         /* Save slave's original mtu and then set it to match the bond */
1544         new_slave->original_mtu = slave_dev->mtu;
1545         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1546         if (res) {
1547                 pr_debug("Error %d calling dev_set_mtu\n", res);
1548                 goto err_free;
1549         }
1550
1551         /*
1552          * Save slave's original ("permanent") mac address for modes
1553          * that need it, and for restoring it upon release, and then
1554          * set it to the master's address
1555          */
1556         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1557
1558         if (!bond->params.fail_over_mac) {
1559                 /*
1560                  * Set slave to master's mac address.  The application already
1561                  * set the master's mac address to that of the first slave
1562                  */
1563                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1564                 addr.sa_family = slave_dev->type;
1565                 res = dev_set_mac_address(slave_dev, &addr);
1566                 if (res) {
1567                         pr_debug("Error %d calling set_mac_address\n", res);
1568                         goto err_restore_mtu;
1569                 }
1570         }
1571
1572         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1573         if (res) {
1574                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1575                 goto err_restore_mac;
1576         }
1577
1578         /* open the slave since the application closed it */
1579         res = dev_open(slave_dev);
1580         if (res) {
1581                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1582                 goto err_unset_master;
1583         }
1584
1585         new_slave->bond = bond;
1586         new_slave->dev = slave_dev;
1587         slave_dev->priv_flags |= IFF_BONDING;
1588
1589         if (bond_is_lb(bond)) {
1590                 /* bond_alb_init_slave() must be called before all other stages since
1591                  * it might fail and we do not want to have to undo everything
1592                  */
1593                 res = bond_alb_init_slave(bond, new_slave);
1594                 if (res)
1595                         goto err_close;
1596         }
1597
1598         /* If the mode USES_PRIMARY, then the following is handled by
1599          * bond_change_active_slave().
1600          */
1601         if (!USES_PRIMARY(bond->params.mode)) {
1602                 /* set promiscuity level to new slave */
1603                 if (bond_dev->flags & IFF_PROMISC) {
1604                         res = dev_set_promiscuity(slave_dev, 1);
1605                         if (res)
1606                                 goto err_close;
1607                 }
1608
1609                 /* set allmulti level to new slave */
1610                 if (bond_dev->flags & IFF_ALLMULTI) {
1611                         res = dev_set_allmulti(slave_dev, 1);
1612                         if (res)
1613                                 goto err_close;
1614                 }
1615
1616                 netif_addr_lock_bh(bond_dev);
1617
1618                 dev_mc_sync_multiple(slave_dev, bond_dev);
1619                 dev_uc_sync_multiple(slave_dev, bond_dev);
1620
1621                 netif_addr_unlock_bh(bond_dev);
1622         }
1623
1624         if (bond->params.mode == BOND_MODE_8023AD) {
1625                 /* add lacpdu mc addr to mc list */
1626                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1627
1628                 dev_mc_add(slave_dev, lacpdu_multicast);
1629         }
1630
1631         bond_add_vlans_on_slave(bond, slave_dev);
1632
1633         write_lock_bh(&bond->lock);
1634
1635         bond_attach_slave(bond, new_slave);
1636
1637         new_slave->delay = 0;
1638         new_slave->link_failure_count = 0;
1639
1640         write_unlock_bh(&bond->lock);
1641
1642         bond_compute_features(bond);
1643
1644         bond_update_speed_duplex(new_slave);
1645
1646         read_lock(&bond->lock);
1647
1648         new_slave->last_arp_rx = jiffies -
1649                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1650         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1651                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1652
1653         if (bond->params.miimon && !bond->params.use_carrier) {
1654                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1655
1656                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1657                         /*
1658                          * miimon is set but a bonded network driver
1659                          * does not support ETHTOOL/MII and
1660                          * arp_interval is not set.  Note: if
1661                          * use_carrier is enabled, we will never go
1662                          * here (because netif_carrier is always
1663                          * supported); thus, we don't need to change
1664                          * the messages for netif_carrier.
1665                          */
1666                         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",
1667                                bond_dev->name, slave_dev->name);
1668                 } else if (link_reporting == -1) {
1669                         /* unable get link status using mii/ethtool */
1670                         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",
1671                                    bond_dev->name, slave_dev->name);
1672                 }
1673         }
1674
1675         /* check for initial state */
1676         if (bond->params.miimon) {
1677                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1678                         if (bond->params.updelay) {
1679                                 new_slave->link = BOND_LINK_BACK;
1680                                 new_slave->delay = bond->params.updelay;
1681                         } else {
1682                                 new_slave->link = BOND_LINK_UP;
1683                         }
1684                 } else {
1685                         new_slave->link = BOND_LINK_DOWN;
1686                 }
1687         } else if (bond->params.arp_interval) {
1688                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1689                         BOND_LINK_UP : BOND_LINK_DOWN);
1690         } else {
1691                 new_slave->link = BOND_LINK_UP;
1692         }
1693
1694         if (new_slave->link != BOND_LINK_DOWN)
1695                 new_slave->jiffies = jiffies;
1696         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1697                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1698                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1699
1700         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1701                 /* if there is a primary slave, remember it */
1702                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1703                         bond->primary_slave = new_slave;
1704                         bond->force_primary = true;
1705                 }
1706         }
1707
1708         write_lock_bh(&bond->curr_slave_lock);
1709
1710         switch (bond->params.mode) {
1711         case BOND_MODE_ACTIVEBACKUP:
1712                 bond_set_slave_inactive_flags(new_slave);
1713                 bond_select_active_slave(bond);
1714                 break;
1715         case BOND_MODE_8023AD:
1716                 /* in 802.3ad mode, the internal mechanism
1717                  * will activate the slaves in the selected
1718                  * aggregator
1719                  */
1720                 bond_set_slave_inactive_flags(new_slave);
1721                 /* if this is the first slave */
1722                 if (bond_first_slave(bond) == new_slave) {
1723                         SLAVE_AD_INFO(new_slave).id = 1;
1724                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1725                          * can be called only after the mac address of the bond is set
1726                          */
1727                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1728                 } else {
1729                         struct slave *prev_slave;
1730
1731                         prev_slave = bond_prev_slave(bond, new_slave);
1732                         SLAVE_AD_INFO(new_slave).id =
1733                                 SLAVE_AD_INFO(prev_slave).id + 1;
1734                 }
1735
1736                 bond_3ad_bind_slave(new_slave);
1737                 break;
1738         case BOND_MODE_TLB:
1739         case BOND_MODE_ALB:
1740                 bond_set_active_slave(new_slave);
1741                 bond_set_slave_inactive_flags(new_slave);
1742                 bond_select_active_slave(bond);
1743                 break;
1744         default:
1745                 pr_debug("This slave is always active in trunk mode\n");
1746
1747                 /* always active in trunk mode */
1748                 bond_set_active_slave(new_slave);
1749
1750                 /* In trunking mode there is little meaning to curr_active_slave
1751                  * anyway (it holds no special properties of the bond device),
1752                  * so we can change it without calling change_active_interface()
1753                  */
1754                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1755                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1756
1757                 break;
1758         } /* switch(bond_mode) */
1759
1760         write_unlock_bh(&bond->curr_slave_lock);
1761
1762         bond_set_carrier(bond);
1763
1764 #ifdef CONFIG_NET_POLL_CONTROLLER
1765         slave_dev->npinfo = bond->dev->npinfo;
1766         if (slave_dev->npinfo) {
1767                 if (slave_enable_netpoll(new_slave)) {
1768                         read_unlock(&bond->lock);
1769                         pr_info("Error, %s: master_dev is using netpoll, "
1770                                  "but new slave device does not support netpoll.\n",
1771                                  bond_dev->name);
1772                         res = -EBUSY;
1773                         goto err_detach;
1774                 }
1775         }
1776 #endif
1777
1778         read_unlock(&bond->lock);
1779
1780         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1781         if (res)
1782                 goto err_detach;
1783
1784         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1785                                          new_slave);
1786         if (res) {
1787                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1788                 goto err_dest_symlinks;
1789         }
1790
1791         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1792                 bond_dev->name, slave_dev->name,
1793                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1794                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1795
1796         /* enslave is successful */
1797         return 0;
1798
1799 /* Undo stages on error */
1800 err_dest_symlinks:
1801         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1802
1803 err_detach:
1804         if (!USES_PRIMARY(bond->params.mode))
1805                 bond_hw_addr_flush(bond_dev, slave_dev);
1806
1807         bond_del_vlans_from_slave(bond, slave_dev);
1808         write_lock_bh(&bond->lock);
1809         bond_detach_slave(bond, new_slave);
1810         if (bond->primary_slave == new_slave)
1811                 bond->primary_slave = NULL;
1812         if (bond->curr_active_slave == new_slave) {
1813                 bond_change_active_slave(bond, NULL);
1814                 write_unlock_bh(&bond->lock);
1815                 read_lock(&bond->lock);
1816                 write_lock_bh(&bond->curr_slave_lock);
1817                 bond_select_active_slave(bond);
1818                 write_unlock_bh(&bond->curr_slave_lock);
1819                 read_unlock(&bond->lock);
1820         } else {
1821                 write_unlock_bh(&bond->lock);
1822         }
1823         slave_disable_netpoll(new_slave);
1824
1825 err_close:
1826         slave_dev->priv_flags &= ~IFF_BONDING;
1827         dev_close(slave_dev);
1828
1829 err_unset_master:
1830         bond_upper_dev_unlink(bond_dev, slave_dev);
1831
1832 err_restore_mac:
1833         if (!bond->params.fail_over_mac) {
1834                 /* XXX TODO - fom follow mode needs to change master's
1835                  * MAC if this slave's MAC is in use by the bond, or at
1836                  * least print a warning.
1837                  */
1838                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1839                 addr.sa_family = slave_dev->type;
1840                 dev_set_mac_address(slave_dev, &addr);
1841         }
1842
1843 err_restore_mtu:
1844         dev_set_mtu(slave_dev, new_slave->original_mtu);
1845
1846 err_free:
1847         kfree(new_slave);
1848
1849 err_undo_flags:
1850         bond_compute_features(bond);
1851         /* Enslave of first slave has failed and we need to fix master's mac */
1852         if (list_empty(&bond->slave_list) &&
1853             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1854                 eth_hw_addr_random(bond_dev);
1855
1856         return res;
1857 }
1858
1859 /*
1860  * Try to release the slave device <slave> from the bond device <master>
1861  * It is legal to access curr_active_slave without a lock because all the function
1862  * is write-locked. If "all" is true it means that the function is being called
1863  * while destroying a bond interface and all slaves are being released.
1864  *
1865  * The rules for slave state should be:
1866  *   for Active/Backup:
1867  *     Active stays on all backups go down
1868  *   for Bonded connections:
1869  *     The first up interface should be left on and all others downed.
1870  */
1871 static int __bond_release_one(struct net_device *bond_dev,
1872                               struct net_device *slave_dev,
1873                               bool all)
1874 {
1875         struct bonding *bond = netdev_priv(bond_dev);
1876         struct slave *slave, *oldcurrent;
1877         struct sockaddr addr;
1878         netdev_features_t old_features = bond_dev->features;
1879
1880         /* slave is not a slave or master is not master of this slave */
1881         if (!(slave_dev->flags & IFF_SLAVE) ||
1882             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1883                 pr_err("%s: Error: cannot release %s.\n",
1884                        bond_dev->name, slave_dev->name);
1885                 return -EINVAL;
1886         }
1887
1888         block_netpoll_tx();
1889         write_lock_bh(&bond->lock);
1890
1891         slave = bond_get_slave_by_dev(bond, slave_dev);
1892         if (!slave) {
1893                 /* not a slave of this bond */
1894                 pr_info("%s: %s not enslaved\n",
1895                         bond_dev->name, slave_dev->name);
1896                 write_unlock_bh(&bond->lock);
1897                 unblock_netpoll_tx();
1898                 return -EINVAL;
1899         }
1900
1901         write_unlock_bh(&bond->lock);
1902         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1903          * for this slave anymore.
1904          */
1905         netdev_rx_handler_unregister(slave_dev);
1906         write_lock_bh(&bond->lock);
1907
1908         /* Inform AD package of unbinding of slave. */
1909         if (bond->params.mode == BOND_MODE_8023AD) {
1910                 /* must be called before the slave is
1911                  * detached from the list
1912                  */
1913                 bond_3ad_unbind_slave(slave);
1914         }
1915
1916         pr_info("%s: releasing %s interface %s\n",
1917                 bond_dev->name,
1918                 bond_is_active_slave(slave) ? "active" : "backup",
1919                 slave_dev->name);
1920
1921         oldcurrent = bond->curr_active_slave;
1922
1923         bond->current_arp_slave = NULL;
1924
1925         /* release the slave from its bond */
1926         bond_detach_slave(bond, slave);
1927
1928         if (!all && !bond->params.fail_over_mac) {
1929                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1930                     !list_empty(&bond->slave_list))
1931                         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",
1932                                    bond_dev->name, slave_dev->name,
1933                                    slave->perm_hwaddr,
1934                                    bond_dev->name, slave_dev->name);
1935         }
1936
1937         if (bond->primary_slave == slave)
1938                 bond->primary_slave = NULL;
1939
1940         if (oldcurrent == slave)
1941                 bond_change_active_slave(bond, NULL);
1942
1943         if (bond_is_lb(bond)) {
1944                 /* Must be called only after the slave has been
1945                  * detached from the list and the curr_active_slave
1946                  * has been cleared (if our_slave == old_current),
1947                  * but before a new active slave is selected.
1948                  */
1949                 write_unlock_bh(&bond->lock);
1950                 bond_alb_deinit_slave(bond, slave);
1951                 write_lock_bh(&bond->lock);
1952         }
1953
1954         if (all) {
1955                 rcu_assign_pointer(bond->curr_active_slave, NULL);
1956         } else if (oldcurrent == slave) {
1957                 /*
1958                  * Note that we hold RTNL over this sequence, so there
1959                  * is no concern that another slave add/remove event
1960                  * will interfere.
1961                  */
1962                 write_unlock_bh(&bond->lock);
1963                 read_lock(&bond->lock);
1964                 write_lock_bh(&bond->curr_slave_lock);
1965
1966                 bond_select_active_slave(bond);
1967
1968                 write_unlock_bh(&bond->curr_slave_lock);
1969                 read_unlock(&bond->lock);
1970                 write_lock_bh(&bond->lock);
1971         }
1972
1973         if (list_empty(&bond->slave_list)) {
1974                 bond_set_carrier(bond);
1975                 eth_hw_addr_random(bond_dev);
1976
1977                 if (bond_vlan_used(bond)) {
1978                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1979                                    bond_dev->name, bond_dev->name);
1980                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1981                                    bond_dev->name);
1982                 }
1983         }
1984
1985         write_unlock_bh(&bond->lock);
1986         unblock_netpoll_tx();
1987         synchronize_rcu();
1988
1989         if (list_empty(&bond->slave_list)) {
1990                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1991                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1992         }
1993
1994         bond_compute_features(bond);
1995         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1996             (old_features & NETIF_F_VLAN_CHALLENGED))
1997                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1998                         bond_dev->name, slave_dev->name, bond_dev->name);
1999
2000         /* must do this from outside any spinlocks */
2001         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2002
2003         bond_del_vlans_from_slave(bond, slave_dev);
2004
2005         /* If the mode USES_PRIMARY, then this cases was handled above by
2006          * bond_change_active_slave(..., NULL)
2007          */
2008         if (!USES_PRIMARY(bond->params.mode)) {
2009                 /* unset promiscuity level from slave */
2010                 if (bond_dev->flags & IFF_PROMISC)
2011                         dev_set_promiscuity(slave_dev, -1);
2012
2013                 /* unset allmulti level from slave */
2014                 if (bond_dev->flags & IFF_ALLMULTI)
2015                         dev_set_allmulti(slave_dev, -1);
2016
2017                 bond_hw_addr_flush(bond_dev, slave_dev);
2018         }
2019
2020         bond_upper_dev_unlink(bond_dev, slave_dev);
2021
2022         slave_disable_netpoll(slave);
2023
2024         /* close slave before restoring its mac address */
2025         dev_close(slave_dev);
2026
2027         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2028                 /* restore original ("permanent") mac address */
2029                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2030                 addr.sa_family = slave_dev->type;
2031                 dev_set_mac_address(slave_dev, &addr);
2032         }
2033
2034         dev_set_mtu(slave_dev, slave->original_mtu);
2035
2036         slave_dev->priv_flags &= ~IFF_BONDING;
2037
2038         kfree(slave);
2039
2040         return 0;  /* deletion OK */
2041 }
2042
2043 /* A wrapper used because of ndo_del_link */
2044 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2045 {
2046         return __bond_release_one(bond_dev, slave_dev, false);
2047 }
2048
2049 /*
2050 * First release a slave and then destroy the bond if no more slaves are left.
2051 * Must be under rtnl_lock when this function is called.
2052 */
2053 static int  bond_release_and_destroy(struct net_device *bond_dev,
2054                                      struct net_device *slave_dev)
2055 {
2056         struct bonding *bond = netdev_priv(bond_dev);
2057         int ret;
2058
2059         ret = bond_release(bond_dev, slave_dev);
2060         if (ret == 0 && list_empty(&bond->slave_list)) {
2061                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2062                 pr_info("%s: destroying bond %s.\n",
2063                         bond_dev->name, bond_dev->name);
2064                 unregister_netdevice(bond_dev);
2065         }
2066         return ret;
2067 }
2068
2069 /*
2070  * This function changes the active slave to slave <slave_dev>.
2071  * It returns -EINVAL in the following cases.
2072  *  - <slave_dev> is not found in the list.
2073  *  - There is not active slave now.
2074  *  - <slave_dev> is already active.
2075  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2076  *  - <slave_dev> is not running.
2077  * In these cases, this function does nothing.
2078  * In the other cases, current_slave pointer is changed and 0 is returned.
2079  */
2080 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2081 {
2082         struct bonding *bond = netdev_priv(bond_dev);
2083         struct slave *old_active = NULL;
2084         struct slave *new_active = NULL;
2085         int res = 0;
2086
2087         if (!USES_PRIMARY(bond->params.mode))
2088                 return -EINVAL;
2089
2090         /* Verify that bond_dev is indeed the master of slave_dev */
2091         if (!(slave_dev->flags & IFF_SLAVE) ||
2092             !netdev_has_upper_dev(slave_dev, bond_dev))
2093                 return -EINVAL;
2094
2095         read_lock(&bond->lock);
2096
2097         old_active = bond->curr_active_slave;
2098         new_active = bond_get_slave_by_dev(bond, slave_dev);
2099         /*
2100          * Changing to the current active: do nothing; return success.
2101          */
2102         if (new_active && new_active == old_active) {
2103                 read_unlock(&bond->lock);
2104                 return 0;
2105         }
2106
2107         if (new_active &&
2108             old_active &&
2109             new_active->link == BOND_LINK_UP &&
2110             IS_UP(new_active->dev)) {
2111                 block_netpoll_tx();
2112                 write_lock_bh(&bond->curr_slave_lock);
2113                 bond_change_active_slave(bond, new_active);
2114                 write_unlock_bh(&bond->curr_slave_lock);
2115                 unblock_netpoll_tx();
2116         } else
2117                 res = -EINVAL;
2118
2119         read_unlock(&bond->lock);
2120
2121         return res;
2122 }
2123
2124 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2125 {
2126         struct bonding *bond = netdev_priv(bond_dev);
2127
2128         info->bond_mode = bond->params.mode;
2129         info->miimon = bond->params.miimon;
2130
2131         read_lock(&bond->lock);
2132         info->num_slaves = bond->slave_cnt;
2133         read_unlock(&bond->lock);
2134
2135         return 0;
2136 }
2137
2138 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2139 {
2140         struct bonding *bond = netdev_priv(bond_dev);
2141         int i = 0, res = -ENODEV;
2142         struct slave *slave;
2143
2144         read_lock(&bond->lock);
2145         bond_for_each_slave(bond, slave) {
2146                 if (i++ == (int)info->slave_id) {
2147                         res = 0;
2148                         strcpy(info->slave_name, slave->dev->name);
2149                         info->link = slave->link;
2150                         info->state = bond_slave_state(slave);
2151                         info->link_failure_count = slave->link_failure_count;
2152                         break;
2153                 }
2154         }
2155         read_unlock(&bond->lock);
2156
2157         return res;
2158 }
2159
2160 /*-------------------------------- Monitoring -------------------------------*/
2161
2162
2163 static int bond_miimon_inspect(struct bonding *bond)
2164 {
2165         int link_state, commit = 0;
2166         struct slave *slave;
2167         bool ignore_updelay;
2168
2169         ignore_updelay = !bond->curr_active_slave ? true : false;
2170
2171         bond_for_each_slave(bond, slave) {
2172                 slave->new_link = BOND_LINK_NOCHANGE;
2173
2174                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2175
2176                 switch (slave->link) {
2177                 case BOND_LINK_UP:
2178                         if (link_state)
2179                                 continue;
2180
2181                         slave->link = BOND_LINK_FAIL;
2182                         slave->delay = bond->params.downdelay;
2183                         if (slave->delay) {
2184                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2185                                         bond->dev->name,
2186                                         (bond->params.mode ==
2187                                          BOND_MODE_ACTIVEBACKUP) ?
2188                                         (bond_is_active_slave(slave) ?
2189                                          "active " : "backup ") : "",
2190                                         slave->dev->name,
2191                                         bond->params.downdelay * bond->params.miimon);
2192                         }
2193                         /*FALLTHRU*/
2194                 case BOND_LINK_FAIL:
2195                         if (link_state) {
2196                                 /*
2197                                  * recovered before downdelay expired
2198                                  */
2199                                 slave->link = BOND_LINK_UP;
2200                                 slave->jiffies = jiffies;
2201                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2202                                         bond->dev->name,
2203                                         (bond->params.downdelay - slave->delay) *
2204                                         bond->params.miimon,
2205                                         slave->dev->name);
2206                                 continue;
2207                         }
2208
2209                         if (slave->delay <= 0) {
2210                                 slave->new_link = BOND_LINK_DOWN;
2211                                 commit++;
2212                                 continue;
2213                         }
2214
2215                         slave->delay--;
2216                         break;
2217
2218                 case BOND_LINK_DOWN:
2219                         if (!link_state)
2220                                 continue;
2221
2222                         slave->link = BOND_LINK_BACK;
2223                         slave->delay = bond->params.updelay;
2224
2225                         if (slave->delay) {
2226                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2227                                         bond->dev->name, slave->dev->name,
2228                                         ignore_updelay ? 0 :
2229                                         bond->params.updelay *
2230                                         bond->params.miimon);
2231                         }
2232                         /*FALLTHRU*/
2233                 case BOND_LINK_BACK:
2234                         if (!link_state) {
2235                                 slave->link = BOND_LINK_DOWN;
2236                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2237                                         bond->dev->name,
2238                                         (bond->params.updelay - slave->delay) *
2239                                         bond->params.miimon,
2240                                         slave->dev->name);
2241
2242                                 continue;
2243                         }
2244
2245                         if (ignore_updelay)
2246                                 slave->delay = 0;
2247
2248                         if (slave->delay <= 0) {
2249                                 slave->new_link = BOND_LINK_UP;
2250                                 commit++;
2251                                 ignore_updelay = false;
2252                                 continue;
2253                         }
2254
2255                         slave->delay--;
2256                         break;
2257                 }
2258         }
2259
2260         return commit;
2261 }
2262
2263 static void bond_miimon_commit(struct bonding *bond)
2264 {
2265         struct slave *slave;
2266
2267         bond_for_each_slave(bond, slave) {
2268                 switch (slave->new_link) {
2269                 case BOND_LINK_NOCHANGE:
2270                         continue;
2271
2272                 case BOND_LINK_UP:
2273                         slave->link = BOND_LINK_UP;
2274                         slave->jiffies = jiffies;
2275
2276                         if (bond->params.mode == BOND_MODE_8023AD) {
2277                                 /* prevent it from being the active one */
2278                                 bond_set_backup_slave(slave);
2279                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2280                                 /* make it immediately active */
2281                                 bond_set_active_slave(slave);
2282                         } else if (slave != bond->primary_slave) {
2283                                 /* prevent it from being the active one */
2284                                 bond_set_backup_slave(slave);
2285                         }
2286
2287                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2288                                 bond->dev->name, slave->dev->name,
2289                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2290                                 slave->duplex ? "full" : "half");
2291
2292                         /* notify ad that the link status has changed */
2293                         if (bond->params.mode == BOND_MODE_8023AD)
2294                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2295
2296                         if (bond_is_lb(bond))
2297                                 bond_alb_handle_link_change(bond, slave,
2298                                                             BOND_LINK_UP);
2299
2300                         if (!bond->curr_active_slave ||
2301                             (slave == bond->primary_slave))
2302                                 goto do_failover;
2303
2304                         continue;
2305
2306                 case BOND_LINK_DOWN:
2307                         if (slave->link_failure_count < UINT_MAX)
2308                                 slave->link_failure_count++;
2309
2310                         slave->link = BOND_LINK_DOWN;
2311
2312                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2313                             bond->params.mode == BOND_MODE_8023AD)
2314                                 bond_set_slave_inactive_flags(slave);
2315
2316                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2317                                 bond->dev->name, slave->dev->name);
2318
2319                         if (bond->params.mode == BOND_MODE_8023AD)
2320                                 bond_3ad_handle_link_change(slave,
2321                                                             BOND_LINK_DOWN);
2322
2323                         if (bond_is_lb(bond))
2324                                 bond_alb_handle_link_change(bond, slave,
2325                                                             BOND_LINK_DOWN);
2326
2327                         if (slave == bond->curr_active_slave)
2328                                 goto do_failover;
2329
2330                         continue;
2331
2332                 default:
2333                         pr_err("%s: invalid new link %d on slave %s\n",
2334                                bond->dev->name, slave->new_link,
2335                                slave->dev->name);
2336                         slave->new_link = BOND_LINK_NOCHANGE;
2337
2338                         continue;
2339                 }
2340
2341 do_failover:
2342                 ASSERT_RTNL();
2343                 block_netpoll_tx();
2344                 write_lock_bh(&bond->curr_slave_lock);
2345                 bond_select_active_slave(bond);
2346                 write_unlock_bh(&bond->curr_slave_lock);
2347                 unblock_netpoll_tx();
2348         }
2349
2350         bond_set_carrier(bond);
2351 }
2352
2353 /*
2354  * bond_mii_monitor
2355  *
2356  * Really a wrapper that splits the mii monitor into two phases: an
2357  * inspection, then (if inspection indicates something needs to be done)
2358  * an acquisition of appropriate locks followed by a commit phase to
2359  * implement whatever link state changes are indicated.
2360  */
2361 void bond_mii_monitor(struct work_struct *work)
2362 {
2363         struct bonding *bond = container_of(work, struct bonding,
2364                                             mii_work.work);
2365         bool should_notify_peers = false;
2366         unsigned long delay;
2367
2368         read_lock(&bond->lock);
2369
2370         delay = msecs_to_jiffies(bond->params.miimon);
2371
2372         if (list_empty(&bond->slave_list))
2373                 goto re_arm;
2374
2375         should_notify_peers = bond_should_notify_peers(bond);
2376
2377         if (bond_miimon_inspect(bond)) {
2378                 read_unlock(&bond->lock);
2379
2380                 /* Race avoidance with bond_close cancel of workqueue */
2381                 if (!rtnl_trylock()) {
2382                         read_lock(&bond->lock);
2383                         delay = 1;
2384                         should_notify_peers = false;
2385                         goto re_arm;
2386                 }
2387
2388                 read_lock(&bond->lock);
2389
2390                 bond_miimon_commit(bond);
2391
2392                 read_unlock(&bond->lock);
2393                 rtnl_unlock();  /* might sleep, hold no other locks */
2394                 read_lock(&bond->lock);
2395         }
2396
2397 re_arm:
2398         if (bond->params.miimon)
2399                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2400
2401         read_unlock(&bond->lock);
2402
2403         if (should_notify_peers) {
2404                 if (!rtnl_trylock()) {
2405                         read_lock(&bond->lock);
2406                         bond->send_peer_notif++;
2407                         read_unlock(&bond->lock);
2408                         return;
2409                 }
2410                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2411                 rtnl_unlock();
2412         }
2413 }
2414
2415 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2416 {
2417         struct vlan_entry *vlan;
2418         struct net_device *vlan_dev;
2419
2420         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2421                 return 1;
2422
2423         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2424                 rcu_read_lock();
2425                 vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2426                                                 vlan->vlan_id);
2427                 rcu_read_unlock();
2428                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2429                         return 1;
2430         }
2431
2432         return 0;
2433 }
2434
2435 /*
2436  * We go to the (large) trouble of VLAN tagging ARP frames because
2437  * switches in VLAN mode (especially if ports are configured as
2438  * "native" to a VLAN) might not pass non-tagged frames.
2439  */
2440 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2441 {
2442         struct sk_buff *skb;
2443
2444         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2445                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2446
2447         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2448                          NULL, slave_dev->dev_addr, NULL);
2449
2450         if (!skb) {
2451                 pr_err("ARP packet allocation failed\n");
2452                 return;
2453         }
2454         if (vlan_id) {
2455                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2456                 if (!skb) {
2457                         pr_err("failed to insert VLAN tag\n");
2458                         return;
2459                 }
2460         }
2461         arp_xmit(skb);
2462 }
2463
2464
2465 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2466 {
2467         int i, vlan_id;
2468         __be32 *targets = bond->params.arp_targets;
2469         struct vlan_entry *vlan;
2470         struct net_device *vlan_dev = NULL;
2471         struct rtable *rt;
2472
2473         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2474                 __be32 addr;
2475                 if (!targets[i])
2476                         break;
2477                 pr_debug("basa: target %pI4\n", &targets[i]);
2478                 if (!bond_vlan_used(bond)) {
2479                         pr_debug("basa: empty vlan: arp_send\n");
2480                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2481                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2482                                       addr, 0);
2483                         continue;
2484                 }
2485
2486                 /*
2487                  * If VLANs are configured, we do a route lookup to
2488                  * determine which VLAN interface would be used, so we
2489                  * can tag the ARP with the proper VLAN tag.
2490                  */
2491                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2492                                      RTO_ONLINK, 0);
2493                 if (IS_ERR(rt)) {
2494                         if (net_ratelimit()) {
2495                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2496                                            bond->dev->name, &targets[i]);
2497                         }
2498                         continue;
2499                 }
2500
2501                 /*
2502                  * This target is not on a VLAN
2503                  */
2504                 if (rt->dst.dev == bond->dev) {
2505                         ip_rt_put(rt);
2506                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2507                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2508                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2509                                       addr, 0);
2510                         continue;
2511                 }
2512
2513                 vlan_id = 0;
2514                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2515                         rcu_read_lock();
2516                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2517                                                         htons(ETH_P_8021Q),
2518                                                         vlan->vlan_id);
2519                         rcu_read_unlock();
2520                         if (vlan_dev == rt->dst.dev) {
2521                                 vlan_id = vlan->vlan_id;
2522                                 pr_debug("basa: vlan match on %s %d\n",
2523                                        vlan_dev->name, vlan_id);
2524                                 break;
2525                         }
2526                 }
2527
2528                 if (vlan_id && vlan_dev) {
2529                         ip_rt_put(rt);
2530                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2531                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2532                                       addr, vlan_id);
2533                         continue;
2534                 }
2535
2536                 if (net_ratelimit()) {
2537                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2538                                    bond->dev->name, &targets[i],
2539                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2540                 }
2541                 ip_rt_put(rt);
2542         }
2543 }
2544
2545 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2546 {
2547         int i;
2548
2549         if (!sip || !bond_has_this_ip(bond, tip)) {
2550                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2551                 return;
2552         }
2553
2554         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2555         if (i == -1) {
2556                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2557                 return;
2558         }
2559         slave->last_arp_rx = jiffies;
2560         slave->target_last_arp_rx[i] = jiffies;
2561 }
2562
2563 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2564                         struct slave *slave)
2565 {
2566         struct arphdr *arp = (struct arphdr *)skb->data;
2567         unsigned char *arp_ptr;
2568         __be32 sip, tip;
2569         int alen;
2570
2571         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2572                 return RX_HANDLER_ANOTHER;
2573
2574         read_lock(&bond->lock);
2575
2576         if (!slave_do_arp_validate(bond, slave))
2577                 goto out_unlock;
2578
2579         alen = arp_hdr_len(bond->dev);
2580
2581         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2582                  bond->dev->name, skb->dev->name);
2583
2584         if (alen > skb_headlen(skb)) {
2585                 arp = kmalloc(alen, GFP_ATOMIC);
2586                 if (!arp)
2587                         goto out_unlock;
2588                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2589                         goto out_unlock;
2590         }
2591
2592         if (arp->ar_hln != bond->dev->addr_len ||
2593             skb->pkt_type == PACKET_OTHERHOST ||
2594             skb->pkt_type == PACKET_LOOPBACK ||
2595             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2596             arp->ar_pro != htons(ETH_P_IP) ||
2597             arp->ar_pln != 4)
2598                 goto out_unlock;
2599
2600         arp_ptr = (unsigned char *)(arp + 1);
2601         arp_ptr += bond->dev->addr_len;
2602         memcpy(&sip, arp_ptr, 4);
2603         arp_ptr += 4 + bond->dev->addr_len;
2604         memcpy(&tip, arp_ptr, 4);
2605
2606         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2607                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2608                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2609                  &sip, &tip);
2610
2611         /*
2612          * Backup slaves won't see the ARP reply, but do come through
2613          * here for each ARP probe (so we swap the sip/tip to validate
2614          * the probe).  In a "redundant switch, common router" type of
2615          * configuration, the ARP probe will (hopefully) travel from
2616          * the active, through one switch, the router, then the other
2617          * switch before reaching the backup.
2618          *
2619          * We 'trust' the arp requests if there is an active slave and
2620          * it received valid arp reply(s) after it became active. This
2621          * is done to avoid endless looping when we can't reach the
2622          * arp_ip_target and fool ourselves with our own arp requests.
2623          */
2624         if (bond_is_active_slave(slave))
2625                 bond_validate_arp(bond, slave, sip, tip);
2626         else if (bond->curr_active_slave &&
2627                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2628                             bond->curr_active_slave->jiffies))
2629                 bond_validate_arp(bond, slave, tip, sip);
2630
2631 out_unlock:
2632         read_unlock(&bond->lock);
2633         if (arp != (struct arphdr *)skb->data)
2634                 kfree(arp);
2635         return RX_HANDLER_ANOTHER;
2636 }
2637
2638 /* function to verify if we're in the arp_interval timeslice, returns true if
2639  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2640  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2641  */
2642 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2643                                   int mod)
2644 {
2645         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2646
2647         return time_in_range(jiffies,
2648                              last_act - delta_in_ticks,
2649                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2650 }
2651
2652 /*
2653  * this function is called regularly to monitor each slave's link
2654  * ensuring that traffic is being sent and received when arp monitoring
2655  * is used in load-balancing mode. if the adapter has been dormant, then an
2656  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2657  * arp monitoring in active backup mode.
2658  */
2659 void bond_loadbalance_arp_mon(struct work_struct *work)
2660 {
2661         struct bonding *bond = container_of(work, struct bonding,
2662                                             arp_work.work);
2663         struct slave *slave, *oldcurrent;
2664         int do_failover = 0;
2665
2666         read_lock(&bond->lock);
2667
2668         if (list_empty(&bond->slave_list))
2669                 goto re_arm;
2670
2671         oldcurrent = bond->curr_active_slave;
2672         /* see if any of the previous devices are up now (i.e. they have
2673          * xmt and rcv traffic). the curr_active_slave does not come into
2674          * the picture unless it is null. also, slave->jiffies is not needed
2675          * here because we send an arp on each slave and give a slave as
2676          * long as it needs to get the tx/rx within the delta.
2677          * TODO: what about up/down delay in arp mode? it wasn't here before
2678          *       so it can wait
2679          */
2680         bond_for_each_slave(bond, slave) {
2681                 unsigned long trans_start = dev_trans_start(slave->dev);
2682
2683                 if (slave->link != BOND_LINK_UP) {
2684                         if (bond_time_in_interval(bond, trans_start, 1) &&
2685                             bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2686
2687                                 slave->link  = BOND_LINK_UP;
2688                                 bond_set_active_slave(slave);
2689
2690                                 /* primary_slave has no meaning in round-robin
2691                                  * mode. the window of a slave being up and
2692                                  * curr_active_slave being null after enslaving
2693                                  * is closed.
2694                                  */
2695                                 if (!oldcurrent) {
2696                                         pr_info("%s: link status definitely up for interface %s, ",
2697                                                 bond->dev->name,
2698                                                 slave->dev->name);
2699                                         do_failover = 1;
2700                                 } else {
2701                                         pr_info("%s: interface %s is now up\n",
2702                                                 bond->dev->name,
2703                                                 slave->dev->name);
2704                                 }
2705                         }
2706                 } else {
2707                         /* slave->link == BOND_LINK_UP */
2708
2709                         /* not all switches will respond to an arp request
2710                          * when the source ip is 0, so don't take the link down
2711                          * if we don't know our ip yet
2712                          */
2713                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2714                             !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2715
2716                                 slave->link  = BOND_LINK_DOWN;
2717                                 bond_set_backup_slave(slave);
2718
2719                                 if (slave->link_failure_count < UINT_MAX)
2720                                         slave->link_failure_count++;
2721
2722                                 pr_info("%s: interface %s is now down.\n",
2723                                         bond->dev->name,
2724                                         slave->dev->name);
2725
2726                                 if (slave == oldcurrent)
2727                                         do_failover = 1;
2728                         }
2729                 }
2730
2731                 /* note: if switch is in round-robin mode, all links
2732                  * must tx arp to ensure all links rx an arp - otherwise
2733                  * links may oscillate or not come up at all; if switch is
2734                  * in something like xor mode, there is nothing we can
2735                  * do - all replies will be rx'ed on same link causing slaves
2736                  * to be unstable during low/no traffic periods
2737                  */
2738                 if (IS_UP(slave->dev))
2739                         bond_arp_send_all(bond, slave);
2740         }
2741
2742         if (do_failover) {
2743                 block_netpoll_tx();
2744                 write_lock_bh(&bond->curr_slave_lock);
2745
2746                 bond_select_active_slave(bond);
2747
2748                 write_unlock_bh(&bond->curr_slave_lock);
2749                 unblock_netpoll_tx();
2750         }
2751
2752 re_arm:
2753         if (bond->params.arp_interval)
2754                 queue_delayed_work(bond->wq, &bond->arp_work,
2755                                    msecs_to_jiffies(bond->params.arp_interval));
2756
2757         read_unlock(&bond->lock);
2758 }
2759
2760 /*
2761  * Called to inspect slaves for active-backup mode ARP monitor link state
2762  * changes.  Sets new_link in slaves to specify what action should take
2763  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2764  * to link states must be committed.
2765  *
2766  * Called with bond->lock held for read.
2767  */
2768 static int bond_ab_arp_inspect(struct bonding *bond)
2769 {
2770         unsigned long trans_start, last_rx;
2771         struct slave *slave;
2772         int commit = 0;
2773
2774         bond_for_each_slave(bond, slave) {
2775                 slave->new_link = BOND_LINK_NOCHANGE;
2776                 last_rx = slave_last_rx(bond, slave);
2777
2778                 if (slave->link != BOND_LINK_UP) {
2779                         if (bond_time_in_interval(bond, last_rx, 1)) {
2780                                 slave->new_link = BOND_LINK_UP;
2781                                 commit++;
2782                         }
2783                         continue;
2784                 }
2785
2786                 /*
2787                  * Give slaves 2*delta after being enslaved or made
2788                  * active.  This avoids bouncing, as the last receive
2789                  * times need a full ARP monitor cycle to be updated.
2790                  */
2791                 if (bond_time_in_interval(bond, slave->jiffies, 2))
2792                         continue;
2793
2794                 /*
2795                  * Backup slave is down if:
2796                  * - No current_arp_slave AND
2797                  * - more than 3*delta since last receive AND
2798                  * - the bond has an IP address
2799                  *
2800                  * Note: a non-null current_arp_slave indicates
2801                  * the curr_active_slave went down and we are
2802                  * searching for a new one; under this condition
2803                  * we only take the curr_active_slave down - this
2804                  * gives each slave a chance to tx/rx traffic
2805                  * before being taken out
2806                  */
2807                 if (!bond_is_active_slave(slave) &&
2808                     !bond->current_arp_slave &&
2809                     !bond_time_in_interval(bond, last_rx, 3)) {
2810                         slave->new_link = BOND_LINK_DOWN;
2811                         commit++;
2812                 }
2813
2814                 /*
2815                  * Active slave is down if:
2816                  * - more than 2*delta since transmitting OR
2817                  * - (more than 2*delta since receive AND
2818                  *    the bond has an IP address)
2819                  */
2820                 trans_start = dev_trans_start(slave->dev);
2821                 if (bond_is_active_slave(slave) &&
2822                     (!bond_time_in_interval(bond, trans_start, 2) ||
2823                      !bond_time_in_interval(bond, last_rx, 2))) {
2824                         slave->new_link = BOND_LINK_DOWN;
2825                         commit++;
2826                 }
2827         }
2828
2829         return commit;
2830 }
2831
2832 /*
2833  * Called to commit link state changes noted by inspection step of
2834  * active-backup mode ARP monitor.
2835  *
2836  * Called with RTNL and bond->lock for read.
2837  */
2838 static void bond_ab_arp_commit(struct bonding *bond)
2839 {
2840         unsigned long trans_start;
2841         struct slave *slave;
2842
2843         bond_for_each_slave(bond, slave) {
2844                 switch (slave->new_link) {
2845                 case BOND_LINK_NOCHANGE:
2846                         continue;
2847
2848                 case BOND_LINK_UP:
2849                         trans_start = dev_trans_start(slave->dev);
2850                         if (bond->curr_active_slave != slave ||
2851                             (!bond->curr_active_slave &&
2852                              bond_time_in_interval(bond, trans_start, 1))) {
2853                                 slave->link = BOND_LINK_UP;
2854                                 if (bond->current_arp_slave) {
2855                                         bond_set_slave_inactive_flags(
2856                                                 bond->current_arp_slave);
2857                                         bond->current_arp_slave = NULL;
2858                                 }
2859
2860                                 pr_info("%s: link status definitely up for interface %s.\n",
2861                                         bond->dev->name, slave->dev->name);
2862
2863                                 if (!bond->curr_active_slave ||
2864                                     (slave == bond->primary_slave))
2865                                         goto do_failover;
2866
2867                         }
2868
2869                         continue;
2870
2871                 case BOND_LINK_DOWN:
2872                         if (slave->link_failure_count < UINT_MAX)
2873                                 slave->link_failure_count++;
2874
2875                         slave->link = BOND_LINK_DOWN;
2876                         bond_set_slave_inactive_flags(slave);
2877
2878                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2879                                 bond->dev->name, slave->dev->name);
2880
2881                         if (slave == bond->curr_active_slave) {
2882                                 bond->current_arp_slave = NULL;
2883                                 goto do_failover;
2884                         }
2885
2886                         continue;
2887
2888                 default:
2889                         pr_err("%s: impossible: new_link %d on slave %s\n",
2890                                bond->dev->name, slave->new_link,
2891                                slave->dev->name);
2892                         continue;
2893                 }
2894
2895 do_failover:
2896                 ASSERT_RTNL();
2897                 block_netpoll_tx();
2898                 write_lock_bh(&bond->curr_slave_lock);
2899                 bond_select_active_slave(bond);
2900                 write_unlock_bh(&bond->curr_slave_lock);
2901                 unblock_netpoll_tx();
2902         }
2903
2904         bond_set_carrier(bond);
2905 }
2906
2907 /*
2908  * Send ARP probes for active-backup mode ARP monitor.
2909  *
2910  * Called with bond->lock held for read.
2911  */
2912 static void bond_ab_arp_probe(struct bonding *bond)
2913 {
2914         struct slave *slave, *next_slave;
2915         int i;
2916
2917         read_lock(&bond->curr_slave_lock);
2918
2919         if (bond->current_arp_slave && bond->curr_active_slave)
2920                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2921                         bond->current_arp_slave->dev->name,
2922                         bond->curr_active_slave->dev->name);
2923
2924         if (bond->curr_active_slave) {
2925                 bond_arp_send_all(bond, bond->curr_active_slave);
2926                 read_unlock(&bond->curr_slave_lock);
2927                 return;
2928         }
2929
2930         read_unlock(&bond->curr_slave_lock);
2931
2932         /* if we don't have a curr_active_slave, search for the next available
2933          * backup slave from the current_arp_slave and make it the candidate
2934          * for becoming the curr_active_slave
2935          */
2936
2937         if (!bond->current_arp_slave) {
2938                 bond->current_arp_slave = bond_first_slave(bond);
2939                 if (!bond->current_arp_slave)
2940                         return;
2941         }
2942
2943         bond_set_slave_inactive_flags(bond->current_arp_slave);
2944
2945         /* search for next candidate */
2946         next_slave = bond_next_slave(bond, bond->current_arp_slave);
2947         bond_for_each_slave_from(bond, slave, i, next_slave) {
2948                 if (IS_UP(slave->dev)) {
2949                         slave->link = BOND_LINK_BACK;
2950                         bond_set_slave_active_flags(slave);
2951                         bond_arp_send_all(bond, slave);
2952                         slave->jiffies = jiffies;
2953                         bond->current_arp_slave = slave;
2954                         break;
2955                 }
2956
2957                 /* if the link state is up at this point, we
2958                  * mark it down - this can happen if we have
2959                  * simultaneous link failures and
2960                  * reselect_active_interface doesn't make this
2961                  * one the current slave so it is still marked
2962                  * up when it is actually down
2963                  */
2964                 if (slave->link == BOND_LINK_UP) {
2965                         slave->link = BOND_LINK_DOWN;
2966                         if (slave->link_failure_count < UINT_MAX)
2967                                 slave->link_failure_count++;
2968
2969                         bond_set_slave_inactive_flags(slave);
2970
2971                         pr_info("%s: backup interface %s is now down.\n",
2972                                 bond->dev->name, slave->dev->name);
2973                 }
2974         }
2975 }
2976
2977 void bond_activebackup_arp_mon(struct work_struct *work)
2978 {
2979         struct bonding *bond = container_of(work, struct bonding,
2980                                             arp_work.work);
2981         bool should_notify_peers = false;
2982         int delta_in_ticks;
2983
2984         read_lock(&bond->lock);
2985
2986         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2987
2988         if (list_empty(&bond->slave_list))
2989                 goto re_arm;
2990
2991         should_notify_peers = bond_should_notify_peers(bond);
2992
2993         if (bond_ab_arp_inspect(bond)) {
2994                 read_unlock(&bond->lock);
2995
2996                 /* Race avoidance with bond_close flush of workqueue */
2997                 if (!rtnl_trylock()) {
2998                         read_lock(&bond->lock);
2999                         delta_in_ticks = 1;
3000                         should_notify_peers = false;
3001                         goto re_arm;
3002                 }
3003
3004                 read_lock(&bond->lock);
3005
3006                 bond_ab_arp_commit(bond);
3007
3008                 read_unlock(&bond->lock);
3009                 rtnl_unlock();
3010                 read_lock(&bond->lock);
3011         }
3012
3013         bond_ab_arp_probe(bond);
3014
3015 re_arm:
3016         if (bond->params.arp_interval)
3017                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3018
3019         read_unlock(&bond->lock);
3020
3021         if (should_notify_peers) {
3022                 if (!rtnl_trylock()) {
3023                         read_lock(&bond->lock);
3024                         bond->send_peer_notif++;
3025                         read_unlock(&bond->lock);
3026                         return;
3027                 }
3028                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3029                 rtnl_unlock();
3030         }
3031 }
3032
3033 /*-------------------------- netdev event handling --------------------------*/
3034
3035 /*
3036  * Change device name
3037  */
3038 static int bond_event_changename(struct bonding *bond)
3039 {
3040         bond_remove_proc_entry(bond);
3041         bond_create_proc_entry(bond);
3042
3043         bond_debug_reregister(bond);
3044
3045         return NOTIFY_DONE;
3046 }
3047
3048 static int bond_master_netdev_event(unsigned long event,
3049                                     struct net_device *bond_dev)
3050 {
3051         struct bonding *event_bond = netdev_priv(bond_dev);
3052
3053         switch (event) {
3054         case NETDEV_CHANGENAME:
3055                 return bond_event_changename(event_bond);
3056         case NETDEV_UNREGISTER:
3057                 bond_remove_proc_entry(event_bond);
3058                 break;
3059         case NETDEV_REGISTER:
3060                 bond_create_proc_entry(event_bond);
3061                 break;
3062         default:
3063                 break;
3064         }
3065
3066         return NOTIFY_DONE;
3067 }
3068
3069 static int bond_slave_netdev_event(unsigned long event,
3070                                    struct net_device *slave_dev)
3071 {
3072         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3073         struct bonding *bond;
3074         struct net_device *bond_dev;
3075         u32 old_speed;
3076         u8 old_duplex;
3077
3078         /* A netdev event can be generated while enslaving a device
3079          * before netdev_rx_handler_register is called in which case
3080          * slave will be NULL
3081          */
3082         if (!slave)
3083                 return NOTIFY_DONE;
3084         bond_dev = slave->bond->dev;
3085         bond = slave->bond;
3086
3087         switch (event) {
3088         case NETDEV_UNREGISTER:
3089                 if (bond_dev->type != ARPHRD_ETHER)
3090                         bond_release_and_destroy(bond_dev, slave_dev);
3091                 else
3092                         bond_release(bond_dev, slave_dev);
3093                 break;
3094         case NETDEV_UP:
3095         case NETDEV_CHANGE:
3096                 old_speed = slave->speed;
3097                 old_duplex = slave->duplex;
3098
3099                 bond_update_speed_duplex(slave);
3100
3101                 if (bond->params.mode == BOND_MODE_8023AD) {
3102                         if (old_speed != slave->speed)
3103                                 bond_3ad_adapter_speed_changed(slave);
3104                         if (old_duplex != slave->duplex)
3105                                 bond_3ad_adapter_duplex_changed(slave);
3106                 }
3107                 break;
3108         case NETDEV_DOWN:
3109                 /*
3110                  * ... Or is it this?
3111                  */
3112                 break;
3113         case NETDEV_CHANGEMTU:
3114                 /*
3115                  * TODO: Should slaves be allowed to
3116                  * independently alter their MTU?  For
3117                  * an active-backup bond, slaves need
3118                  * not be the same type of device, so
3119                  * MTUs may vary.  For other modes,
3120                  * slaves arguably should have the
3121                  * same MTUs. To do this, we'd need to
3122                  * take over the slave's change_mtu
3123                  * function for the duration of their
3124                  * servitude.
3125                  */
3126                 break;
3127         case NETDEV_CHANGENAME:
3128                 /*
3129                  * TODO: handle changing the primary's name
3130                  */
3131                 break;
3132         case NETDEV_FEAT_CHANGE:
3133                 bond_compute_features(bond);
3134                 break;
3135         case NETDEV_RESEND_IGMP:
3136                 /* Propagate to master device */
3137                 call_netdevice_notifiers(event, slave->bond->dev);
3138                 break;
3139         default:
3140                 break;
3141         }
3142
3143         return NOTIFY_DONE;
3144 }
3145
3146 /*
3147  * bond_netdev_event: handle netdev notifier chain events.
3148  *
3149  * This function receives events for the netdev chain.  The caller (an
3150  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3151  * locks for us to safely manipulate the slave devices (RTNL lock,
3152  * dev_probe_lock).
3153  */
3154 static int bond_netdev_event(struct notifier_block *this,
3155                              unsigned long event, void *ptr)
3156 {
3157         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3158
3159         pr_debug("event_dev: %s, event: %lx\n",
3160                  event_dev ? event_dev->name : "None",
3161                  event);
3162
3163         if (!(event_dev->priv_flags & IFF_BONDING))
3164                 return NOTIFY_DONE;
3165
3166         if (event_dev->flags & IFF_MASTER) {
3167                 pr_debug("IFF_MASTER\n");
3168                 return bond_master_netdev_event(event, event_dev);
3169         }
3170
3171         if (event_dev->flags & IFF_SLAVE) {
3172                 pr_debug("IFF_SLAVE\n");
3173                 return bond_slave_netdev_event(event, event_dev);
3174         }
3175
3176         return NOTIFY_DONE;
3177 }
3178
3179 static struct notifier_block bond_netdev_notifier = {
3180         .notifier_call = bond_netdev_event,
3181 };
3182
3183 /*---------------------------- Hashing Policies -----------------------------*/
3184
3185 /*
3186  * Hash for the output device based upon layer 2 data
3187  */
3188 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3189 {
3190         struct ethhdr *data = (struct ethhdr *)skb->data;
3191
3192         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3193                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3194
3195         return 0;
3196 }
3197
3198 /*
3199  * Hash for the output device based upon layer 2 and layer 3 data. If
3200  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3201  */
3202 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3203 {
3204         const struct ethhdr *data;
3205         const struct iphdr *iph;
3206         const struct ipv6hdr *ipv6h;
3207         u32 v6hash;
3208         const __be32 *s, *d;
3209
3210         if (skb->protocol == htons(ETH_P_IP) &&
3211             pskb_network_may_pull(skb, sizeof(*iph))) {
3212                 iph = ip_hdr(skb);
3213                 data = (struct ethhdr *)skb->data;
3214                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3215                         (data->h_dest[5] ^ data->h_source[5])) % count;
3216         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3217                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3218                 ipv6h = ipv6_hdr(skb);
3219                 data = (struct ethhdr *)skb->data;
3220                 s = &ipv6h->saddr.s6_addr32[0];
3221                 d = &ipv6h->daddr.s6_addr32[0];
3222                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3223                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3224                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3225         }
3226
3227         return bond_xmit_hash_policy_l2(skb, count);
3228 }
3229
3230 /*
3231  * Hash for the output device based upon layer 3 and layer 4 data. If
3232  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3233  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3234  */
3235 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3236 {
3237         u32 layer4_xor = 0;
3238         const struct iphdr *iph;
3239         const struct ipv6hdr *ipv6h;
3240         const __be32 *s, *d;
3241         const __be16 *l4 = NULL;
3242         __be16 _l4[2];
3243         int noff = skb_network_offset(skb);
3244         int poff;
3245
3246         if (skb->protocol == htons(ETH_P_IP) &&
3247             pskb_may_pull(skb, noff + sizeof(*iph))) {
3248                 iph = ip_hdr(skb);
3249                 poff = proto_ports_offset(iph->protocol);
3250
3251                 if (!ip_is_fragment(iph) && poff >= 0) {
3252                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3253                                                 sizeof(_l4), &_l4);
3254                         if (l4)
3255                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3256                 }
3257                 return (layer4_xor ^
3258                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3259         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3260                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3261                 ipv6h = ipv6_hdr(skb);
3262                 poff = proto_ports_offset(ipv6h->nexthdr);
3263                 if (poff >= 0) {
3264                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3265                                                 sizeof(_l4), &_l4);
3266                         if (l4)
3267                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3268                 }
3269                 s = &ipv6h->saddr.s6_addr32[0];
3270                 d = &ipv6h->daddr.s6_addr32[0];
3271                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3272                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3273                                (layer4_xor >> 8);
3274                 return layer4_xor % count;
3275         }
3276
3277         return bond_xmit_hash_policy_l2(skb, count);
3278 }
3279
3280 /*-------------------------- Device entry points ----------------------------*/
3281
3282 static void bond_work_init_all(struct bonding *bond)
3283 {
3284         INIT_DELAYED_WORK(&bond->mcast_work,
3285                           bond_resend_igmp_join_requests_delayed);
3286         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3287         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3288         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3289                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3290         else
3291                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3292         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3293 }
3294
3295 static void bond_work_cancel_all(struct bonding *bond)
3296 {
3297         cancel_delayed_work_sync(&bond->mii_work);
3298         cancel_delayed_work_sync(&bond->arp_work);
3299         cancel_delayed_work_sync(&bond->alb_work);
3300         cancel_delayed_work_sync(&bond->ad_work);
3301         cancel_delayed_work_sync(&bond->mcast_work);
3302 }
3303
3304 static int bond_open(struct net_device *bond_dev)
3305 {
3306         struct bonding *bond = netdev_priv(bond_dev);
3307         struct slave *slave;
3308
3309         /* reset slave->backup and slave->inactive */
3310         read_lock(&bond->lock);
3311         if (!list_empty(&bond->slave_list)) {
3312                 read_lock(&bond->curr_slave_lock);
3313                 bond_for_each_slave(bond, slave) {
3314                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3315                                 && (slave != bond->curr_active_slave)) {
3316                                 bond_set_slave_inactive_flags(slave);
3317                         } else {
3318                                 bond_set_slave_active_flags(slave);
3319                         }
3320                 }
3321                 read_unlock(&bond->curr_slave_lock);
3322         }
3323         read_unlock(&bond->lock);
3324
3325         bond_work_init_all(bond);
3326
3327         if (bond_is_lb(bond)) {
3328                 /* bond_alb_initialize must be called before the timer
3329                  * is started.
3330                  */
3331                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3332                         return -ENOMEM;
3333                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3334         }
3335
3336         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3337                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3338
3339         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3340                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3341                 if (bond->params.arp_validate)
3342                         bond->recv_probe = bond_arp_rcv;
3343         }
3344
3345         if (bond->params.mode == BOND_MODE_8023AD) {
3346                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3347                 /* register to receive LACPDUs */
3348                 bond->recv_probe = bond_3ad_lacpdu_recv;
3349                 bond_3ad_initiate_agg_selection(bond, 1);
3350         }
3351
3352         return 0;
3353 }
3354
3355 static int bond_close(struct net_device *bond_dev)
3356 {
3357         struct bonding *bond = netdev_priv(bond_dev);
3358
3359         write_lock_bh(&bond->lock);
3360         bond->send_peer_notif = 0;
3361         write_unlock_bh(&bond->lock);
3362
3363         bond_work_cancel_all(bond);
3364         if (bond_is_lb(bond)) {
3365                 /* Must be called only after all
3366                  * slaves have been released
3367                  */
3368                 bond_alb_deinitialize(bond);
3369         }
3370         bond->recv_probe = NULL;
3371
3372         return 0;
3373 }
3374
3375 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3376                                                 struct rtnl_link_stats64 *stats)
3377 {
3378         struct bonding *bond = netdev_priv(bond_dev);
3379         struct rtnl_link_stats64 temp;
3380         struct slave *slave;
3381
3382         memset(stats, 0, sizeof(*stats));
3383
3384         read_lock_bh(&bond->lock);
3385         bond_for_each_slave(bond, slave) {
3386                 const struct rtnl_link_stats64 *sstats =
3387                         dev_get_stats(slave->dev, &temp);
3388
3389                 stats->rx_packets += sstats->rx_packets;
3390                 stats->rx_bytes += sstats->rx_bytes;
3391                 stats->rx_errors += sstats->rx_errors;
3392                 stats->rx_dropped += sstats->rx_dropped;
3393
3394                 stats->tx_packets += sstats->tx_packets;
3395                 stats->tx_bytes += sstats->tx_bytes;
3396                 stats->tx_errors += sstats->tx_errors;
3397                 stats->tx_dropped += sstats->tx_dropped;
3398
3399                 stats->multicast += sstats->multicast;
3400                 stats->collisions += sstats->collisions;
3401
3402                 stats->rx_length_errors += sstats->rx_length_errors;
3403                 stats->rx_over_errors += sstats->rx_over_errors;
3404                 stats->rx_crc_errors += sstats->rx_crc_errors;
3405                 stats->rx_frame_errors += sstats->rx_frame_errors;
3406                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3407                 stats->rx_missed_errors += sstats->rx_missed_errors;
3408
3409                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3410                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3411                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3412                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3413                 stats->tx_window_errors += sstats->tx_window_errors;
3414         }
3415         read_unlock_bh(&bond->lock);
3416
3417         return stats;
3418 }
3419
3420 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3421 {
3422         struct net_device *slave_dev = NULL;
3423         struct ifbond k_binfo;
3424         struct ifbond __user *u_binfo = NULL;
3425         struct ifslave k_sinfo;
3426         struct ifslave __user *u_sinfo = NULL;
3427         struct mii_ioctl_data *mii = NULL;
3428         struct net *net;
3429         int res = 0;
3430
3431         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3432
3433         switch (cmd) {
3434         case SIOCGMIIPHY:
3435                 mii = if_mii(ifr);
3436                 if (!mii)
3437                         return -EINVAL;
3438
3439                 mii->phy_id = 0;
3440                 /* Fall Through */
3441         case SIOCGMIIREG:
3442                 /*
3443                  * We do this again just in case we were called by SIOCGMIIREG
3444                  * instead of SIOCGMIIPHY.
3445                  */
3446                 mii = if_mii(ifr);
3447                 if (!mii)
3448                         return -EINVAL;
3449
3450
3451                 if (mii->reg_num == 1) {
3452                         struct bonding *bond = netdev_priv(bond_dev);
3453                         mii->val_out = 0;
3454                         read_lock(&bond->lock);
3455                         read_lock(&bond->curr_slave_lock);
3456                         if (netif_carrier_ok(bond->dev))
3457                                 mii->val_out = BMSR_LSTATUS;
3458
3459                         read_unlock(&bond->curr_slave_lock);
3460                         read_unlock(&bond->lock);
3461                 }
3462
3463                 return 0;
3464         case BOND_INFO_QUERY_OLD:
3465         case SIOCBONDINFOQUERY:
3466                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3467
3468                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3469                         return -EFAULT;
3470
3471                 res = bond_info_query(bond_dev, &k_binfo);
3472                 if (res == 0 &&
3473                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3474                         return -EFAULT;
3475
3476                 return res;
3477         case BOND_SLAVE_INFO_QUERY_OLD:
3478         case SIOCBONDSLAVEINFOQUERY:
3479                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3480
3481                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3482                         return -EFAULT;
3483
3484                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3485                 if (res == 0 &&
3486                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3487                         return -EFAULT;
3488
3489                 return res;
3490         default:
3491                 /* Go on */
3492                 break;
3493         }
3494
3495         net = dev_net(bond_dev);
3496
3497         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3498                 return -EPERM;
3499
3500         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3501
3502         pr_debug("slave_dev=%p:\n", slave_dev);
3503
3504         if (!slave_dev)
3505                 res = -ENODEV;
3506         else {
3507                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3508                 switch (cmd) {
3509                 case BOND_ENSLAVE_OLD:
3510                 case SIOCBONDENSLAVE:
3511                         res = bond_enslave(bond_dev, slave_dev);
3512                         break;
3513                 case BOND_RELEASE_OLD:
3514                 case SIOCBONDRELEASE:
3515                         res = bond_release(bond_dev, slave_dev);
3516                         break;
3517                 case BOND_SETHWADDR_OLD:
3518                 case SIOCBONDSETHWADDR:
3519                         bond_set_dev_addr(bond_dev, slave_dev);
3520                         res = 0;
3521                         break;
3522                 case BOND_CHANGE_ACTIVE_OLD:
3523                 case SIOCBONDCHANGEACTIVE:
3524                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3525                         break;
3526                 default:
3527                         res = -EOPNOTSUPP;
3528                 }
3529
3530                 dev_put(slave_dev);
3531         }
3532
3533         return res;
3534 }
3535
3536 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3537 {
3538         struct bonding *bond = netdev_priv(bond_dev);
3539
3540         if (change & IFF_PROMISC)
3541                 bond_set_promiscuity(bond,
3542                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3543
3544         if (change & IFF_ALLMULTI)
3545                 bond_set_allmulti(bond,
3546                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3547 }
3548
3549 static void bond_set_rx_mode(struct net_device *bond_dev)
3550 {
3551         struct bonding *bond = netdev_priv(bond_dev);
3552         struct slave *slave;
3553
3554         read_lock(&bond->lock);
3555
3556         if (USES_PRIMARY(bond->params.mode)) {
3557                 read_lock(&bond->curr_slave_lock);
3558                 slave = bond->curr_active_slave;
3559                 if (slave) {
3560                         dev_uc_sync(slave->dev, bond_dev);
3561                         dev_mc_sync(slave->dev, bond_dev);
3562                 }
3563                 read_unlock(&bond->curr_slave_lock);
3564         } else {
3565                 bond_for_each_slave(bond, slave) {
3566                         dev_uc_sync_multiple(slave->dev, bond_dev);
3567                         dev_mc_sync_multiple(slave->dev, bond_dev);
3568                 }
3569         }
3570
3571         read_unlock(&bond->lock);
3572 }
3573
3574 static int bond_neigh_init(struct neighbour *n)
3575 {
3576         struct bonding *bond = netdev_priv(n->dev);
3577         const struct net_device_ops *slave_ops;
3578         struct neigh_parms parms;
3579         struct slave *slave;
3580         int ret;
3581
3582         slave = bond_first_slave(bond);
3583         if (!slave)
3584                 return 0;
3585         slave_ops = slave->dev->netdev_ops;
3586         if (!slave_ops->ndo_neigh_setup)
3587                 return 0;
3588
3589         parms.neigh_setup = NULL;
3590         parms.neigh_cleanup = NULL;
3591         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3592         if (ret)
3593                 return ret;
3594
3595         /*
3596          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3597          * after the last slave has been detached.  Assumes that all slaves
3598          * utilize the same neigh_cleanup (true at this writing as only user
3599          * is ipoib).
3600          */
3601         n->parms->neigh_cleanup = parms.neigh_cleanup;
3602
3603         if (!parms.neigh_setup)
3604                 return 0;
3605
3606         return parms.neigh_setup(n);
3607 }
3608
3609 /*
3610  * The bonding ndo_neigh_setup is called at init time beofre any
3611  * slave exists. So we must declare proxy setup function which will
3612  * be used at run time to resolve the actual slave neigh param setup.
3613  *
3614  * It's also called by master devices (such as vlans) to setup their
3615  * underlying devices. In that case - do nothing, we're already set up from
3616  * our init.
3617  */
3618 static int bond_neigh_setup(struct net_device *dev,
3619                             struct neigh_parms *parms)
3620 {
3621         /* modify only our neigh_parms */
3622         if (parms->dev == dev)
3623                 parms->neigh_setup = bond_neigh_init;
3624
3625         return 0;
3626 }
3627
3628 /*
3629  * Change the MTU of all of a master's slaves to match the master
3630  */
3631 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3632 {
3633         struct bonding *bond = netdev_priv(bond_dev);
3634         struct slave *slave;
3635         int res = 0;
3636
3637         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3638                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3639
3640         /* Can't hold bond->lock with bh disabled here since
3641          * some base drivers panic. On the other hand we can't
3642          * hold bond->lock without bh disabled because we'll
3643          * deadlock. The only solution is to rely on the fact
3644          * that we're under rtnl_lock here, and the slaves
3645          * list won't change. This doesn't solve the problem
3646          * of setting the slave's MTU while it is
3647          * transmitting, but the assumption is that the base
3648          * driver can handle that.
3649          *
3650          * TODO: figure out a way to safely iterate the slaves
3651          * list, but without holding a lock around the actual
3652          * call to the base driver.
3653          */
3654
3655         bond_for_each_slave(bond, slave) {
3656                 pr_debug("s %p s->p %p c_m %p\n",
3657                          slave,
3658                          bond_prev_slave(bond, slave),
3659                          slave->dev->netdev_ops->ndo_change_mtu);
3660
3661                 res = dev_set_mtu(slave->dev, new_mtu);
3662
3663                 if (res) {
3664                         /* If we failed to set the slave's mtu to the new value
3665                          * we must abort the operation even in ACTIVE_BACKUP
3666                          * mode, because if we allow the backup slaves to have
3667                          * different mtu values than the active slave we'll
3668                          * need to change their mtu when doing a failover. That
3669                          * means changing their mtu from timer context, which
3670                          * is probably not a good idea.
3671                          */
3672                         pr_debug("err %d %s\n", res, slave->dev->name);
3673                         goto unwind;
3674                 }
3675         }
3676
3677         bond_dev->mtu = new_mtu;
3678
3679         return 0;
3680
3681 unwind:
3682         /* unwind from head to the slave that failed */
3683         bond_for_each_slave_continue_reverse(bond, slave) {
3684                 int tmp_res;
3685
3686                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3687                 if (tmp_res) {
3688                         pr_debug("unwind err %d dev %s\n",
3689                                  tmp_res, slave->dev->name);
3690                 }
3691         }
3692
3693         return res;
3694 }
3695
3696 /*
3697  * Change HW address
3698  *
3699  * Note that many devices must be down to change the HW address, and
3700  * downing the master releases all slaves.  We can make bonds full of
3701  * bonding devices to test this, however.
3702  */
3703 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3704 {
3705         struct bonding *bond = netdev_priv(bond_dev);
3706         struct sockaddr *sa = addr, tmp_sa;
3707         struct slave *slave;
3708         int res = 0;
3709
3710         if (bond->params.mode == BOND_MODE_ALB)
3711                 return bond_alb_set_mac_address(bond_dev, addr);
3712
3713
3714         pr_debug("bond=%p, name=%s\n",
3715                  bond, bond_dev ? bond_dev->name : "None");
3716
3717         /* If fail_over_mac is enabled, do nothing and return success.
3718          * Returning an error causes ifenslave to fail.
3719          */
3720         if (bond->params.fail_over_mac)
3721                 return 0;
3722
3723         if (!is_valid_ether_addr(sa->sa_data))
3724                 return -EADDRNOTAVAIL;
3725
3726         /* Can't hold bond->lock with bh disabled here since
3727          * some base drivers panic. On the other hand we can't
3728          * hold bond->lock without bh disabled because we'll
3729          * deadlock. The only solution is to rely on the fact
3730          * that we're under rtnl_lock here, and the slaves
3731          * list won't change. This doesn't solve the problem
3732          * of setting the slave's hw address while it is
3733          * transmitting, but the assumption is that the base
3734          * driver can handle that.
3735          *
3736          * TODO: figure out a way to safely iterate the slaves
3737          * list, but without holding a lock around the actual
3738          * call to the base driver.
3739          */
3740
3741         bond_for_each_slave(bond, slave) {
3742                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3743                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3744
3745                 if (slave_ops->ndo_set_mac_address == NULL) {
3746                         res = -EOPNOTSUPP;
3747                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3748                         goto unwind;
3749                 }
3750
3751                 res = dev_set_mac_address(slave->dev, addr);
3752                 if (res) {
3753                         /* TODO: consider downing the slave
3754                          * and retry ?
3755                          * User should expect communications
3756                          * breakage anyway until ARP finish
3757                          * updating, so...
3758                          */
3759                         pr_debug("err %d %s\n", res, slave->dev->name);
3760                         goto unwind;
3761                 }
3762         }
3763
3764         /* success */
3765         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3766         return 0;
3767
3768 unwind:
3769         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3770         tmp_sa.sa_family = bond_dev->type;
3771
3772         /* unwind from head to the slave that failed */
3773         bond_for_each_slave_continue_reverse(bond, slave) {
3774                 int tmp_res;
3775
3776                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3777                 if (tmp_res) {
3778                         pr_debug("unwind err %d dev %s\n",
3779                                  tmp_res, slave->dev->name);
3780                 }
3781         }
3782
3783         return res;
3784 }
3785
3786 /**
3787  * bond_xmit_slave_id - transmit skb through slave with slave_id
3788  * @bond: bonding device that is transmitting
3789  * @skb: buffer to transmit
3790  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3791  *
3792  * This function tries to transmit through slave with slave_id but in case
3793  * it fails, it tries to find the first available slave for transmission.
3794  * The skb is consumed in all cases, thus the function is void.
3795  */
3796 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3797 {
3798         struct slave *slave;
3799         int i = slave_id;
3800
3801         /* Here we start from the slave with slave_id */
3802         bond_for_each_slave_rcu(bond, slave) {
3803                 if (--i < 0) {
3804                         if (slave_can_tx(slave)) {
3805                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3806                                 return;
3807                         }
3808                 }
3809         }
3810
3811         /* Here we start from the first slave up to slave_id */
3812         i = slave_id;
3813         bond_for_each_slave_rcu(bond, slave) {
3814                 if (--i < 0)
3815                         break;
3816                 if (slave_can_tx(slave)) {
3817                         bond_dev_queue_xmit(bond, skb, slave->dev);
3818                         return;
3819                 }
3820         }
3821         /* no slave that can tx has been found */
3822         kfree_skb(skb);
3823 }
3824
3825 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3826 {
3827         struct bonding *bond = netdev_priv(bond_dev);
3828         struct iphdr *iph = ip_hdr(skb);
3829         struct slave *slave;
3830
3831         /*
3832          * Start with the curr_active_slave that joined the bond as the
3833          * default for sending IGMP traffic.  For failover purposes one
3834          * needs to maintain some consistency for the interface that will
3835          * send the join/membership reports.  The curr_active_slave found
3836          * will send all of this type of traffic.
3837          */
3838         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3839                 slave = rcu_dereference(bond->curr_active_slave);
3840                 if (slave && slave_can_tx(slave))
3841                         bond_dev_queue_xmit(bond, skb, slave->dev);
3842                 else
3843                         bond_xmit_slave_id(bond, skb, 0);
3844         } else {
3845                 bond_xmit_slave_id(bond, skb,
3846                                    bond->rr_tx_counter++ % bond->slave_cnt);
3847         }
3848
3849         return NETDEV_TX_OK;
3850 }
3851
3852 /*
3853  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3854  * the bond has a usable interface.
3855  */
3856 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3857 {
3858         struct bonding *bond = netdev_priv(bond_dev);
3859         struct slave *slave;
3860
3861         slave = rcu_dereference(bond->curr_active_slave);
3862         if (slave)
3863                 bond_dev_queue_xmit(bond, skb, slave->dev);
3864         else
3865                 kfree_skb(skb);
3866
3867         return NETDEV_TX_OK;
3868 }
3869
3870 /*
3871  * In bond_xmit_xor() , we determine the output device by using a pre-
3872  * determined xmit_hash_policy(), If the selected device is not enabled,
3873  * find the next active slave.
3874  */
3875 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3876 {
3877         struct bonding *bond = netdev_priv(bond_dev);
3878
3879         bond_xmit_slave_id(bond, skb,
3880                            bond->xmit_hash_policy(skb, bond->slave_cnt));
3881
3882         return NETDEV_TX_OK;
3883 }
3884
3885 /* in broadcast mode, we send everything to all usable interfaces. */
3886 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3887 {
3888         struct bonding *bond = netdev_priv(bond_dev);
3889         struct slave *slave = NULL;
3890
3891         bond_for_each_slave_rcu(bond, slave) {
3892                 if (bond_is_last_slave(bond, slave))
3893                         break;
3894                 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3895                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3896
3897                         if (!skb2) {
3898                                 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3899                                        bond_dev->name);
3900                                 continue;
3901                         }
3902                         /* bond_dev_queue_xmit always returns 0 */
3903                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3904                 }
3905         }
3906         if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3907                 bond_dev_queue_xmit(bond, skb, slave->dev);
3908         else
3909                 kfree_skb(skb);
3910
3911         return NETDEV_TX_OK;
3912 }
3913
3914 /*------------------------- Device initialization ---------------------------*/
3915
3916 static void bond_set_xmit_hash_policy(struct bonding *bond)
3917 {
3918         switch (bond->params.xmit_policy) {
3919         case BOND_XMIT_POLICY_LAYER23:
3920                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
3921                 break;
3922         case BOND_XMIT_POLICY_LAYER34:
3923                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
3924                 break;
3925         case BOND_XMIT_POLICY_LAYER2:
3926         default:
3927                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
3928                 break;
3929         }
3930 }
3931
3932 /*
3933  * Lookup the slave that corresponds to a qid
3934  */
3935 static inline int bond_slave_override(struct bonding *bond,
3936                                       struct sk_buff *skb)
3937 {
3938         struct slave *slave = NULL;
3939         struct slave *check_slave;
3940         int res = 1;
3941
3942         if (!skb->queue_mapping)
3943                 return 1;
3944
3945         /* Find out if any slaves have the same mapping as this skb. */
3946         bond_for_each_slave_rcu(bond, check_slave) {
3947                 if (check_slave->queue_id == skb->queue_mapping) {
3948                         slave = check_slave;
3949                         break;
3950                 }
3951         }
3952
3953         /* If the slave isn't UP, use default transmit policy. */
3954         if (slave && slave->queue_id && IS_UP(slave->dev) &&
3955             (slave->link == BOND_LINK_UP)) {
3956                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3957         }
3958
3959         return res;
3960 }
3961
3962
3963 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3964 {
3965         /*
3966          * This helper function exists to help dev_pick_tx get the correct
3967          * destination queue.  Using a helper function skips a call to
3968          * skb_tx_hash and will put the skbs in the queue we expect on their
3969          * way down to the bonding driver.
3970          */
3971         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3972
3973         /*
3974          * Save the original txq to restore before passing to the driver
3975          */
3976         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3977
3978         if (unlikely(txq >= dev->real_num_tx_queues)) {
3979                 do {
3980                         txq -= dev->real_num_tx_queues;
3981                 } while (txq >= dev->real_num_tx_queues);
3982         }
3983         return txq;
3984 }
3985
3986 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3987 {
3988         struct bonding *bond = netdev_priv(dev);
3989
3990         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3991                 if (!bond_slave_override(bond, skb))
3992                         return NETDEV_TX_OK;
3993         }
3994
3995         switch (bond->params.mode) {
3996         case BOND_MODE_ROUNDROBIN:
3997                 return bond_xmit_roundrobin(skb, dev);
3998         case BOND_MODE_ACTIVEBACKUP:
3999                 return bond_xmit_activebackup(skb, dev);
4000         case BOND_MODE_XOR:
4001                 return bond_xmit_xor(skb, dev);
4002         case BOND_MODE_BROADCAST:
4003                 return bond_xmit_broadcast(skb, dev);
4004         case BOND_MODE_8023AD:
4005                 return bond_3ad_xmit_xor(skb, dev);
4006         case BOND_MODE_ALB:
4007         case BOND_MODE_TLB:
4008                 return bond_alb_xmit(skb, dev);
4009         default:
4010                 /* Should never happen, mode already checked */
4011                 pr_err("%s: Error: Unknown bonding mode %d\n",
4012                        dev->name, bond->params.mode);
4013                 WARN_ON_ONCE(1);
4014                 kfree_skb(skb);
4015                 return NETDEV_TX_OK;
4016         }
4017 }
4018
4019 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4020 {
4021         struct bonding *bond = netdev_priv(dev);
4022         netdev_tx_t ret = NETDEV_TX_OK;
4023
4024         /*
4025          * If we risk deadlock from transmitting this in the
4026          * netpoll path, tell netpoll to queue the frame for later tx
4027          */
4028         if (is_netpoll_tx_blocked(dev))
4029                 return NETDEV_TX_BUSY;
4030
4031         rcu_read_lock();
4032         if (!list_empty(&bond->slave_list))
4033                 ret = __bond_start_xmit(skb, dev);
4034         else
4035                 kfree_skb(skb);
4036         rcu_read_unlock();
4037
4038         return ret;
4039 }
4040
4041 /*
4042  * set bond mode specific net device operations
4043  */
4044 void bond_set_mode_ops(struct bonding *bond, int mode)
4045 {
4046         struct net_device *bond_dev = bond->dev;
4047
4048         switch (mode) {
4049         case BOND_MODE_ROUNDROBIN:
4050                 break;
4051         case BOND_MODE_ACTIVEBACKUP:
4052                 break;
4053         case BOND_MODE_XOR:
4054                 bond_set_xmit_hash_policy(bond);
4055                 break;
4056         case BOND_MODE_BROADCAST:
4057                 break;
4058         case BOND_MODE_8023AD:
4059                 bond_set_xmit_hash_policy(bond);
4060                 break;
4061         case BOND_MODE_ALB:
4062                 /* FALLTHRU */
4063         case BOND_MODE_TLB:
4064                 break;
4065         default:
4066                 /* Should never happen, mode already checked */
4067                 pr_err("%s: Error: Unknown bonding mode %d\n",
4068                        bond_dev->name, mode);
4069                 break;
4070         }
4071 }
4072
4073 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4074                                      struct ethtool_cmd *ecmd)
4075 {
4076         struct bonding *bond = netdev_priv(bond_dev);
4077         unsigned long speed = 0;
4078         struct slave *slave;
4079
4080         ecmd->duplex = DUPLEX_UNKNOWN;
4081         ecmd->port = PORT_OTHER;
4082
4083         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4084          * do not need to check mode.  Though link speed might not represent
4085          * the true receive or transmit bandwidth (not all modes are symmetric)
4086          * this is an accurate maximum.
4087          */
4088         read_lock(&bond->lock);
4089         bond_for_each_slave(bond, slave) {
4090                 if (SLAVE_IS_OK(slave)) {
4091                         if (slave->speed != SPEED_UNKNOWN)
4092                                 speed += slave->speed;
4093                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4094                             slave->duplex != DUPLEX_UNKNOWN)
4095                                 ecmd->duplex = slave->duplex;
4096                 }
4097         }
4098         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4099         read_unlock(&bond->lock);
4100
4101         return 0;
4102 }
4103
4104 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4105                                      struct ethtool_drvinfo *drvinfo)
4106 {
4107         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4108         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4109         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4110                  BOND_ABI_VERSION);
4111 }
4112
4113 static const struct ethtool_ops bond_ethtool_ops = {
4114         .get_drvinfo            = bond_ethtool_get_drvinfo,
4115         .get_settings           = bond_ethtool_get_settings,
4116         .get_link               = ethtool_op_get_link,
4117 };
4118
4119 static const struct net_device_ops bond_netdev_ops = {
4120         .ndo_init               = bond_init,
4121         .ndo_uninit             = bond_uninit,
4122         .ndo_open               = bond_open,
4123         .ndo_stop               = bond_close,
4124         .ndo_start_xmit         = bond_start_xmit,
4125         .ndo_select_queue       = bond_select_queue,
4126         .ndo_get_stats64        = bond_get_stats,
4127         .ndo_do_ioctl           = bond_do_ioctl,
4128         .ndo_change_rx_flags    = bond_change_rx_flags,
4129         .ndo_set_rx_mode        = bond_set_rx_mode,
4130         .ndo_change_mtu         = bond_change_mtu,
4131         .ndo_set_mac_address    = bond_set_mac_address,
4132         .ndo_neigh_setup        = bond_neigh_setup,
4133         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4134         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4135 #ifdef CONFIG_NET_POLL_CONTROLLER
4136         .ndo_netpoll_setup      = bond_netpoll_setup,
4137         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4138         .ndo_poll_controller    = bond_poll_controller,
4139 #endif
4140         .ndo_add_slave          = bond_enslave,
4141         .ndo_del_slave          = bond_release,
4142         .ndo_fix_features       = bond_fix_features,
4143 };
4144
4145 static const struct device_type bond_type = {
4146         .name = "bond",
4147 };
4148
4149 static void bond_destructor(struct net_device *bond_dev)
4150 {
4151         struct bonding *bond = netdev_priv(bond_dev);
4152         if (bond->wq)
4153                 destroy_workqueue(bond->wq);
4154         free_netdev(bond_dev);
4155 }
4156
4157 static void bond_setup(struct net_device *bond_dev)
4158 {
4159         struct bonding *bond = netdev_priv(bond_dev);
4160
4161         /* initialize rwlocks */
4162         rwlock_init(&bond->lock);
4163         rwlock_init(&bond->curr_slave_lock);
4164         INIT_LIST_HEAD(&bond->slave_list);
4165         bond->params = bonding_defaults;
4166
4167         /* Initialize pointers */
4168         bond->dev = bond_dev;
4169         INIT_LIST_HEAD(&bond->vlan_list);
4170
4171         /* Initialize the device entry points */
4172         ether_setup(bond_dev);
4173         bond_dev->netdev_ops = &bond_netdev_ops;
4174         bond_dev->ethtool_ops = &bond_ethtool_ops;
4175         bond_set_mode_ops(bond, bond->params.mode);
4176
4177         bond_dev->destructor = bond_destructor;
4178
4179         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4180
4181         /* Initialize the device options */
4182         bond_dev->tx_queue_len = 0;
4183         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4184         bond_dev->priv_flags |= IFF_BONDING;
4185         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4186
4187         /* At first, we block adding VLANs. That's the only way to
4188          * prevent problems that occur when adding VLANs over an
4189          * empty bond. The block will be removed once non-challenged
4190          * slaves are enslaved.
4191          */
4192         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4193
4194         /* don't acquire bond device's netif_tx_lock when
4195          * transmitting */
4196         bond_dev->features |= NETIF_F_LLTX;
4197
4198         /* By default, we declare the bond to be fully
4199          * VLAN hardware accelerated capable. Special
4200          * care is taken in the various xmit functions
4201          * when there are slaves that are not hw accel
4202          * capable
4203          */
4204
4205         bond_dev->hw_features = BOND_VLAN_FEATURES |
4206                                 NETIF_F_HW_VLAN_CTAG_TX |
4207                                 NETIF_F_HW_VLAN_CTAG_RX |
4208                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4209
4210         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4211         bond_dev->features |= bond_dev->hw_features;
4212 }
4213
4214 /*
4215 * Destroy a bonding device.
4216 * Must be under rtnl_lock when this function is called.
4217 */
4218 static void bond_uninit(struct net_device *bond_dev)
4219 {
4220         struct bonding *bond = netdev_priv(bond_dev);
4221         struct slave *slave, *tmp_slave;
4222         struct vlan_entry *vlan, *tmp;
4223
4224         bond_netpoll_cleanup(bond_dev);
4225
4226         /* Release the bonded slaves */
4227         list_for_each_entry_safe(slave, tmp_slave, &bond->slave_list, list)
4228                 __bond_release_one(bond_dev, slave->dev, true);
4229         pr_info("%s: released all slaves\n", bond_dev->name);
4230
4231         list_del(&bond->bond_list);
4232
4233         bond_debug_unregister(bond);
4234
4235         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4236                 list_del(&vlan->vlan_list);
4237                 kfree(vlan);
4238         }
4239 }
4240
4241 /*------------------------- Module initialization ---------------------------*/
4242
4243 /*
4244  * Convert string input module parms.  Accept either the
4245  * number of the mode or its string name.  A bit complicated because
4246  * some mode names are substrings of other names, and calls from sysfs
4247  * may have whitespace in the name (trailing newlines, for example).
4248  */
4249 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4250 {
4251         int modeint = -1, i, rv;
4252         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4253
4254         for (p = (char *)buf; *p; p++)
4255                 if (!(isdigit(*p) || isspace(*p)))
4256                         break;
4257
4258         if (*p)
4259                 rv = sscanf(buf, "%20s", modestr);
4260         else
4261                 rv = sscanf(buf, "%d", &modeint);
4262
4263         if (!rv)
4264                 return -1;
4265
4266         for (i = 0; tbl[i].modename; i++) {
4267                 if (modeint == tbl[i].mode)
4268                         return tbl[i].mode;
4269                 if (strcmp(modestr, tbl[i].modename) == 0)
4270                         return tbl[i].mode;
4271         }
4272
4273         return -1;
4274 }
4275
4276 static int bond_check_params(struct bond_params *params)
4277 {
4278         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4279         int arp_all_targets_value;
4280
4281         /*
4282          * Convert string parameters.
4283          */
4284         if (mode) {
4285                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4286                 if (bond_mode == -1) {
4287                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4288                                mode == NULL ? "NULL" : mode);
4289                         return -EINVAL;
4290                 }
4291         }
4292
4293         if (xmit_hash_policy) {
4294                 if ((bond_mode != BOND_MODE_XOR) &&
4295                     (bond_mode != BOND_MODE_8023AD)) {
4296                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4297                                bond_mode_name(bond_mode));
4298                 } else {
4299                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4300                                                         xmit_hashtype_tbl);
4301                         if (xmit_hashtype == -1) {
4302                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4303                                        xmit_hash_policy == NULL ? "NULL" :
4304                                        xmit_hash_policy);
4305                                 return -EINVAL;
4306                         }
4307                 }
4308         }
4309
4310         if (lacp_rate) {
4311                 if (bond_mode != BOND_MODE_8023AD) {
4312                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4313                                 bond_mode_name(bond_mode));
4314                 } else {
4315                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4316                         if (lacp_fast == -1) {
4317                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4318                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4319                                 return -EINVAL;
4320                         }
4321                 }
4322         }
4323
4324         if (ad_select) {
4325                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4326                 if (params->ad_select == -1) {
4327                         pr_err("Error: Invalid ad_select \"%s\"\n",
4328                                ad_select == NULL ? "NULL" : ad_select);
4329                         return -EINVAL;
4330                 }
4331
4332                 if (bond_mode != BOND_MODE_8023AD) {
4333                         pr_warning("ad_select param only affects 802.3ad mode\n");
4334                 }
4335         } else {
4336                 params->ad_select = BOND_AD_STABLE;
4337         }
4338
4339         if (max_bonds < 0) {
4340                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4341                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4342                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4343         }
4344
4345         if (miimon < 0) {
4346                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4347                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4348                 miimon = BOND_LINK_MON_INTERV;
4349         }
4350
4351         if (updelay < 0) {
4352                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4353                            updelay, INT_MAX);
4354                 updelay = 0;
4355         }
4356
4357         if (downdelay < 0) {
4358                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4359                            downdelay, INT_MAX);
4360                 downdelay = 0;
4361         }
4362
4363         if ((use_carrier != 0) && (use_carrier != 1)) {
4364                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4365                            use_carrier);
4366                 use_carrier = 1;
4367         }
4368
4369         if (num_peer_notif < 0 || num_peer_notif > 255) {
4370                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4371                            num_peer_notif);
4372                 num_peer_notif = 1;
4373         }
4374
4375         /* reset values for 802.3ad */
4376         if (bond_mode == BOND_MODE_8023AD) {
4377                 if (!miimon) {
4378                         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");
4379                         pr_warning("Forcing miimon to 100msec\n");
4380                         miimon = 100;
4381                 }
4382         }
4383
4384         if (tx_queues < 1 || tx_queues > 255) {
4385                 pr_warning("Warning: tx_queues (%d) should be between "
4386                            "1 and 255, resetting to %d\n",
4387                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4388                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4389         }
4390
4391         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4392                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4393                            "not of valid value (0/1), so it was set to "
4394                            "0\n", all_slaves_active);
4395                 all_slaves_active = 0;
4396         }
4397
4398         if (resend_igmp < 0 || resend_igmp > 255) {
4399                 pr_warning("Warning: resend_igmp (%d) should be between "
4400                            "0 and 255, resetting to %d\n",
4401                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4402                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4403         }
4404
4405         /* reset values for TLB/ALB */
4406         if ((bond_mode == BOND_MODE_TLB) ||
4407             (bond_mode == BOND_MODE_ALB)) {
4408                 if (!miimon) {
4409                         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");
4410                         pr_warning("Forcing miimon to 100msec\n");
4411                         miimon = 100;
4412                 }
4413         }
4414
4415         if (bond_mode == BOND_MODE_ALB) {
4416                 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",
4417                           updelay);
4418         }
4419
4420         if (!miimon) {
4421                 if (updelay || downdelay) {
4422                         /* just warn the user the up/down delay will have
4423                          * no effect since miimon is zero...
4424                          */
4425                         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",
4426                                    updelay, downdelay);
4427                 }
4428         } else {
4429                 /* don't allow arp monitoring */
4430                 if (arp_interval) {
4431                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4432                                    miimon, arp_interval);
4433                         arp_interval = 0;
4434                 }
4435
4436                 if ((updelay % miimon) != 0) {
4437                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4438                                    updelay, miimon,
4439                                    (updelay / miimon) * miimon);
4440                 }
4441
4442                 updelay /= miimon;
4443
4444                 if ((downdelay % miimon) != 0) {
4445                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4446                                    downdelay, miimon,
4447                                    (downdelay / miimon) * miimon);
4448                 }
4449
4450                 downdelay /= miimon;
4451         }
4452
4453         if (arp_interval < 0) {
4454                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4455                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4456                 arp_interval = BOND_LINK_ARP_INTERV;
4457         }
4458
4459         for (arp_ip_count = 0, i = 0;
4460              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4461                 /* not complete check, but should be good enough to
4462                    catch mistakes */
4463                 __be32 ip = in_aton(arp_ip_target[i]);
4464                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4465                     ip == htonl(INADDR_BROADCAST)) {
4466                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4467                                    arp_ip_target[i]);
4468                         arp_interval = 0;
4469                 } else {
4470                         if (bond_get_targets_ip(arp_target, ip) == -1)
4471                                 arp_target[arp_ip_count++] = ip;
4472                         else
4473                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4474                                            &ip);
4475                 }
4476         }
4477
4478         if (arp_interval && !arp_ip_count) {
4479                 /* don't allow arping if no arp_ip_target given... */
4480                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4481                            arp_interval);
4482                 arp_interval = 0;
4483         }
4484
4485         if (arp_validate) {
4486                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4487                         pr_err("arp_validate only supported in active-backup mode\n");
4488                         return -EINVAL;
4489                 }
4490                 if (!arp_interval) {
4491                         pr_err("arp_validate requires arp_interval\n");
4492                         return -EINVAL;
4493                 }
4494
4495                 arp_validate_value = bond_parse_parm(arp_validate,
4496                                                      arp_validate_tbl);
4497                 if (arp_validate_value == -1) {
4498                         pr_err("Error: invalid arp_validate \"%s\"\n",
4499                                arp_validate == NULL ? "NULL" : arp_validate);
4500                         return -EINVAL;
4501                 }
4502         } else
4503                 arp_validate_value = 0;
4504
4505         arp_all_targets_value = 0;
4506         if (arp_all_targets) {
4507                 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4508                                                         arp_all_targets_tbl);
4509
4510                 if (arp_all_targets_value == -1) {
4511                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4512                                arp_all_targets);
4513                         arp_all_targets_value = 0;
4514                 }
4515         }
4516
4517         if (miimon) {
4518                 pr_info("MII link monitoring set to %d ms\n", miimon);
4519         } else if (arp_interval) {
4520                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4521                         arp_interval,
4522                         arp_validate_tbl[arp_validate_value].modename,
4523                         arp_ip_count);
4524
4525                 for (i = 0; i < arp_ip_count; i++)
4526                         pr_info(" %s", arp_ip_target[i]);
4527
4528                 pr_info("\n");
4529
4530         } else if (max_bonds) {
4531                 /* miimon and arp_interval not set, we need one so things
4532                  * work as expected, see bonding.txt for details
4533                  */
4534                 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");
4535         }
4536
4537         if (primary && !USES_PRIMARY(bond_mode)) {
4538                 /* currently, using a primary only makes sense
4539                  * in active backup, TLB or ALB modes
4540                  */
4541                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4542                            primary, bond_mode_name(bond_mode));
4543                 primary = NULL;
4544         }
4545
4546         if (primary && primary_reselect) {
4547                 primary_reselect_value = bond_parse_parm(primary_reselect,
4548                                                          pri_reselect_tbl);
4549                 if (primary_reselect_value == -1) {
4550                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4551                                primary_reselect ==
4552                                         NULL ? "NULL" : primary_reselect);
4553                         return -EINVAL;
4554                 }
4555         } else {
4556                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4557         }
4558
4559         if (fail_over_mac) {
4560                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4561                                                       fail_over_mac_tbl);
4562                 if (fail_over_mac_value == -1) {
4563                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4564                                arp_validate == NULL ? "NULL" : arp_validate);
4565                         return -EINVAL;
4566                 }
4567
4568                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4569                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4570         } else {
4571                 fail_over_mac_value = BOND_FOM_NONE;
4572         }
4573
4574         /* fill params struct with the proper values */
4575         params->mode = bond_mode;
4576         params->xmit_policy = xmit_hashtype;
4577         params->miimon = miimon;
4578         params->num_peer_notif = num_peer_notif;
4579         params->arp_interval = arp_interval;
4580         params->arp_validate = arp_validate_value;
4581         params->arp_all_targets = arp_all_targets_value;
4582         params->updelay = updelay;
4583         params->downdelay = downdelay;
4584         params->use_carrier = use_carrier;
4585         params->lacp_fast = lacp_fast;
4586         params->primary[0] = 0;
4587         params->primary_reselect = primary_reselect_value;
4588         params->fail_over_mac = fail_over_mac_value;
4589         params->tx_queues = tx_queues;
4590         params->all_slaves_active = all_slaves_active;
4591         params->resend_igmp = resend_igmp;
4592         params->min_links = min_links;
4593
4594         if (primary) {
4595                 strncpy(params->primary, primary, IFNAMSIZ);
4596                 params->primary[IFNAMSIZ - 1] = 0;
4597         }
4598
4599         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4600
4601         return 0;
4602 }
4603
4604 static struct lock_class_key bonding_netdev_xmit_lock_key;
4605 static struct lock_class_key bonding_netdev_addr_lock_key;
4606 static struct lock_class_key bonding_tx_busylock_key;
4607
4608 static void bond_set_lockdep_class_one(struct net_device *dev,
4609                                        struct netdev_queue *txq,
4610                                        void *_unused)
4611 {
4612         lockdep_set_class(&txq->_xmit_lock,
4613                           &bonding_netdev_xmit_lock_key);
4614 }
4615
4616 static void bond_set_lockdep_class(struct net_device *dev)
4617 {
4618         lockdep_set_class(&dev->addr_list_lock,
4619                           &bonding_netdev_addr_lock_key);
4620         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4621         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4622 }
4623
4624 /*
4625  * Called from registration process
4626  */
4627 static int bond_init(struct net_device *bond_dev)
4628 {
4629         struct bonding *bond = netdev_priv(bond_dev);
4630         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4631         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4632
4633         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4634
4635         /*
4636          * Initialize locks that may be required during
4637          * en/deslave operations.  All of the bond_open work
4638          * (of which this is part) should really be moved to
4639          * a phase prior to dev_open
4640          */
4641         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4642         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4643
4644         bond->wq = create_singlethread_workqueue(bond_dev->name);
4645         if (!bond->wq)
4646                 return -ENOMEM;
4647
4648         bond_set_lockdep_class(bond_dev);
4649
4650         list_add_tail(&bond->bond_list, &bn->dev_list);
4651
4652         bond_prepare_sysfs_group(bond);
4653
4654         bond_debug_register(bond);
4655
4656         /* Ensure valid dev_addr */
4657         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4658             bond_dev->addr_assign_type == NET_ADDR_PERM)
4659                 eth_hw_addr_random(bond_dev);
4660
4661         return 0;
4662 }
4663
4664 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4665 {
4666         if (tb[IFLA_ADDRESS]) {
4667                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4668                         return -EINVAL;
4669                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4670                         return -EADDRNOTAVAIL;
4671         }
4672         return 0;
4673 }
4674
4675 static unsigned int bond_get_num_tx_queues(void)
4676 {
4677         return tx_queues;
4678 }
4679
4680 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4681         .kind                   = "bond",
4682         .priv_size              = sizeof(struct bonding),
4683         .setup                  = bond_setup,
4684         .validate               = bond_validate,
4685         .get_num_tx_queues      = bond_get_num_tx_queues,
4686         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4687                                                              as for TX queues */
4688 };
4689
4690 /* Create a new bond based on the specified name and bonding parameters.
4691  * If name is NULL, obtain a suitable "bond%d" name for us.
4692  * Caller must NOT hold rtnl_lock; we need to release it here before we
4693  * set up our sysfs entries.
4694  */
4695 int bond_create(struct net *net, const char *name)
4696 {
4697         struct net_device *bond_dev;
4698         int res;
4699
4700         rtnl_lock();
4701
4702         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4703                                    name ? name : "bond%d",
4704                                    bond_setup, tx_queues);
4705         if (!bond_dev) {
4706                 pr_err("%s: eek! can't alloc netdev!\n", name);
4707                 rtnl_unlock();
4708                 return -ENOMEM;
4709         }
4710
4711         dev_net_set(bond_dev, net);
4712         bond_dev->rtnl_link_ops = &bond_link_ops;
4713
4714         res = register_netdevice(bond_dev);
4715
4716         netif_carrier_off(bond_dev);
4717
4718         rtnl_unlock();
4719         if (res < 0)
4720                 bond_destructor(bond_dev);
4721         return res;
4722 }
4723
4724 static int __net_init bond_net_init(struct net *net)
4725 {
4726         struct bond_net *bn = net_generic(net, bond_net_id);
4727
4728         bn->net = net;
4729         INIT_LIST_HEAD(&bn->dev_list);
4730
4731         bond_create_proc_dir(bn);
4732         bond_create_sysfs(bn);
4733
4734         return 0;
4735 }
4736
4737 static void __net_exit bond_net_exit(struct net *net)
4738 {
4739         struct bond_net *bn = net_generic(net, bond_net_id);
4740         struct bonding *bond, *tmp_bond;
4741         LIST_HEAD(list);
4742
4743         bond_destroy_sysfs(bn);
4744         bond_destroy_proc_dir(bn);
4745
4746         /* Kill off any bonds created after unregistering bond rtnl ops */
4747         rtnl_lock();
4748         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4749                 unregister_netdevice_queue(bond->dev, &list);
4750         unregister_netdevice_many(&list);
4751         rtnl_unlock();
4752 }
4753
4754 static struct pernet_operations bond_net_ops = {
4755         .init = bond_net_init,
4756         .exit = bond_net_exit,
4757         .id   = &bond_net_id,
4758         .size = sizeof(struct bond_net),
4759 };
4760
4761 static int __init bonding_init(void)
4762 {
4763         int i;
4764         int res;
4765
4766         pr_info("%s", bond_version);
4767
4768         res = bond_check_params(&bonding_defaults);
4769         if (res)
4770                 goto out;
4771
4772         res = register_pernet_subsys(&bond_net_ops);
4773         if (res)
4774                 goto out;
4775
4776         res = rtnl_link_register(&bond_link_ops);
4777         if (res)
4778                 goto err_link;
4779
4780         bond_create_debugfs();
4781
4782         for (i = 0; i < max_bonds; i++) {
4783                 res = bond_create(&init_net, NULL);
4784                 if (res)
4785                         goto err;
4786         }
4787
4788         register_netdevice_notifier(&bond_netdev_notifier);
4789 out:
4790         return res;
4791 err:
4792         rtnl_link_unregister(&bond_link_ops);
4793 err_link:
4794         unregister_pernet_subsys(&bond_net_ops);
4795         goto out;
4796
4797 }
4798
4799 static void __exit bonding_exit(void)
4800 {
4801         unregister_netdevice_notifier(&bond_netdev_notifier);
4802
4803         bond_destroy_debugfs();
4804
4805         rtnl_link_unregister(&bond_link_ops);
4806         unregister_pernet_subsys(&bond_net_ops);
4807
4808 #ifdef CONFIG_NET_POLL_CONTROLLER
4809         /*
4810          * Make sure we don't have an imbalance on our netpoll blocking
4811          */
4812         WARN_ON(atomic_read(&netpoll_block_tx));
4813 #endif
4814 }
4815
4816 module_init(bonding_init);
4817 module_exit(bonding_exit);
4818 MODULE_LICENSE("GPL");
4819 MODULE_VERSION(DRV_VERSION);
4820 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4821 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4822 MODULE_ALIAS_RTNL_LINK("bond");