]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/bonding/bond_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
igb: do not check for vf_data if we didn't enable vfs
[net-next-2.6.git] / drivers / net / bonding / bond_main.c
CommitLineData
1da177e4
LT		 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 *
1da177e4
LT		 32 */
33
1da177e4
LT		 34#include <linux/kernel.h>
35#include <linux/module.h>
1da177e4
LT		 36#include <linux/types.h>
37#include <linux/fcntl.h>
38#include <linux/interrupt.h>
39#include <linux/ptrace.h>
40#include <linux/ioport.h>
41#include <linux/in.h>
169a3e66 42#include <net/ip.h>
1da177e4 43#include <linux/ip.h>
169a3e66
JV		 44#include <linux/tcp.h>
45#include <linux/udp.h>
1da177e4
LT		 46#include <linux/slab.h>
47#include <linux/string.h>
48#include <linux/init.h>
49#include <linux/timer.h>
50#include <linux/socket.h>
51#include <linux/ctype.h>
52#include <linux/inet.h>
53#include <linux/bitops.h>
54#include <asm/system.h>
55#include <asm/io.h>
56#include <asm/dma.h>
57#include <asm/uaccess.h>
58#include <linux/errno.h>
59#include <linux/netdevice.h>
60#include <linux/inetdevice.h>
a816c7c7 61#include <linux/igmp.h>
1da177e4
LT		 62#include <linux/etherdevice.h>
63#include <linux/skbuff.h>
64#include <net/sock.h>
65#include <linux/rtnetlink.h>
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68#include <linux/smp.h>
69#include <linux/if_ether.h>
70#include <net/arp.h>
71#include <linux/mii.h>
72#include <linux/ethtool.h>
73#include <linux/if_vlan.h>
74#include <linux/if_bonding.h>
b63bb739 75#include <linux/jiffies.h>
c3ade5ca 76#include <net/route.h>
457c4cbc 77#include <net/net_namespace.h>
1da177e4
LT		 78#include "bonding.h"
79#include "bond_3ad.h"
80#include "bond_alb.h"
81
82/*---------------------------- Module parameters ----------------------------*/
83
84/* monitor all links that often (in milliseconds). <=0 disables monitoring */
85#define BOND_LINK_MON_INTERV 0
86#define BOND_LINK_ARP_INTERV 0
87
88static int max_bonds = BOND_DEFAULT_MAX_BONDS;
7893b249 89static int num_grat_arp = 1;
305d552a 90static int num_unsol_na = 1;
1da177e4
LT		 91static int miimon = BOND_LINK_MON_INTERV;
92static int updelay = 0;
93static int downdelay = 0;
94static int use_carrier = 1;
95static char *mode = NULL;
96static char *primary = NULL;
97static char *lacp_rate = NULL;
fd989c83 98static char *ad_select = NULL;
169a3e66 99static char *xmit_hash_policy = NULL;
1da177e4
LT		 100static int arp_interval = BOND_LINK_ARP_INTERV;
101static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
f5b2b966 102static char *arp_validate = NULL;
3915c1e8 103static char *fail_over_mac = NULL;
12479f9a 104struct bond_params bonding_defaults;
1da177e4
LT		 105
106module_param(max_bonds, int, 0);
107MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
7893b249
MS		 108module_param(num_grat_arp, int, 0644);
109MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
305d552a
BH		 110module_param(num_unsol_na, int, 0644);
111MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
1da177e4
LT		 112module_param(miimon, int, 0);
113MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
114module_param(updelay, int, 0);
115MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
116module_param(downdelay, int, 0);
2ac47660
MW		 117MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
118 "in milliseconds");
1da177e4 119module_param(use_carrier, int, 0);
2ac47660
MW		 120MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
121 "0 for off, 1 for on (default)");
1da177e4 122module_param(mode, charp, 0);
2ac47660
MW		 123MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
124 "1 for active-backup, 2 for balance-xor, "
125 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
126 "6 for balance-alb");
1da177e4
LT		 127module_param(primary, charp, 0);
128MODULE_PARM_DESC(primary, "Primary network device to use");
129module_param(lacp_rate, charp, 0);
2ac47660
MW		 130MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
131 "(slow/fast)");
fd989c83
JV		 132module_param(ad_select, charp, 0);
133MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
169a3e66 134module_param(xmit_hash_policy, charp, 0);
2ac47660
MW		 135MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
136 ", 1 for layer 3+4");
1da177e4
LT		 137module_param(arp_interval, int, 0);
138MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
139module_param_array(arp_ip_target, charp, NULL, 0);
140MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
f5b2b966
JV		 141module_param(arp_validate, charp, 0);
142MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
3915c1e8
JV		 143module_param(fail_over_mac, charp, 0);
144MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
1da177e4
LT		 145
146/*----------------------------- Global variables ----------------------------*/
147
f71e1309 148static const char * const version =
1da177e4
LT		 149 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
150
12479f9a 151LIST_HEAD(bond_dev_list);
1da177e4
LT		 152
153#ifdef CONFIG_PROC_FS
154static struct proc_dir_entry *bond_proc_dir = NULL;
155#endif
156
d3bb52b0 157static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
1da177e4 158static int arp_ip_count = 0;
1da177e4 159static int bond_mode = BOND_MODE_ROUNDROBIN;
169a3e66 160static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
1da177e4 161static int lacp_fast = 0;
217df670 162
1da177e4 163
e97fd7c6 164const struct bond_parm_tbl bond_lacp_tbl[] = {
1da177e4
LT		 165{ "slow", AD_LACP_SLOW},
166{ "fast", AD_LACP_FAST},
167{ NULL, -1},
168};
169
e97fd7c6 170const struct bond_parm_tbl bond_mode_tbl[] = {
1da177e4
LT		 171{ "balance-rr", BOND_MODE_ROUNDROBIN},
172{ "active-backup", BOND_MODE_ACTIVEBACKUP},
173{ "balance-xor", BOND_MODE_XOR},
174{ "broadcast", BOND_MODE_BROADCAST},
175{ "802.3ad", BOND_MODE_8023AD},
176{ "balance-tlb", BOND_MODE_TLB},
177{ "balance-alb", BOND_MODE_ALB},
178{ NULL, -1},
179};
180
e97fd7c6 181const struct bond_parm_tbl xmit_hashtype_tbl[] = {
169a3e66
JV		 182{ "layer2", BOND_XMIT_POLICY_LAYER2},
183{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
6f6652be 184{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
169a3e66
JV		 185{ NULL, -1},
186};
187
e97fd7c6 188const struct bond_parm_tbl arp_validate_tbl[] = {
f5b2b966
JV		 189{ "none", BOND_ARP_VALIDATE_NONE},
190{ "active", BOND_ARP_VALIDATE_ACTIVE},
191{ "backup", BOND_ARP_VALIDATE_BACKUP},
192{ "all", BOND_ARP_VALIDATE_ALL},
193{ NULL, -1},
194};
195
e97fd7c6 196const struct bond_parm_tbl fail_over_mac_tbl[] = {
3915c1e8
JV		 197{ "none", BOND_FOM_NONE},
198{ "active", BOND_FOM_ACTIVE},
199{ "follow", BOND_FOM_FOLLOW},
200{ NULL, -1},
201};
202
fd989c83
JV		 203struct bond_parm_tbl ad_select_tbl[] = {
204{ "stable", BOND_AD_STABLE},
205{ "bandwidth", BOND_AD_BANDWIDTH},
206{ "count", BOND_AD_COUNT},
207{ NULL, -1},
208};
209
1da177e4
LT		 210/*-------------------------- Forward declarations ---------------------------*/
211
c3ade5ca 212static void bond_send_gratuitous_arp(struct bonding *bond);
c50b85d0 213static void bond_deinit(struct net_device *bond_dev);
1da177e4
LT		 214
215/*---------------------------- General routines -----------------------------*/
216
4ad072c9 217static const char *bond_mode_name(int mode)
1da177e4 218{
77afc92b
HE		 219 static const char *names[] = {
220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
224 [BOND_MODE_8023AD]= "IEEE 802.3ad Dynamic link aggregation",
225 [BOND_MODE_TLB] = "transmit load balancing",
226 [BOND_MODE_ALB] = "adaptive load balancing",
227 };
228
229 if (mode < 0 || mode > BOND_MODE_ALB)
1da177e4 230 return "unknown";
77afc92b
HE		 231
232 return names[mode];
1da177e4
LT		 233}
234
235/*---------------------------------- VLAN -----------------------------------*/
236
237/**
238 * bond_add_vlan - add a new vlan id on bond
239 * @bond: bond that got the notification
240 * @vlan_id: the vlan id to add
241 *
242 * Returns -ENOMEM if allocation failed.
243 */
244static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
245{
246 struct vlan_entry *vlan;
247
5a03cdb7 248 pr_debug("bond: %s, vlan id %d\n",
1da177e4
LT		 249 (bond ? bond->dev->name: "None"), vlan_id);
250
305d552a 251 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
1da177e4
LT		 252 if (!vlan) {
253 return -ENOMEM;
254 }
255
256 INIT_LIST_HEAD(&vlan->vlan_list);
257 vlan->vlan_id = vlan_id;
258
259 write_lock_bh(&bond->lock);
260
261 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
262
263 write_unlock_bh(&bond->lock);
264
5a03cdb7 265 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
1da177e4
LT		 266
267 return 0;
268}
269
270/**
271 * bond_del_vlan - delete a vlan id from bond
272 * @bond: bond that got the notification
273 * @vlan_id: the vlan id to delete
274 *
275 * returns -ENODEV if @vlan_id was not found in @bond.
276 */
277static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
278{
0883beca 279 struct vlan_entry *vlan;
1da177e4
LT		 280 int res = -ENODEV;
281
5a03cdb7 282 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
1da177e4
LT		 283
284 write_lock_bh(&bond->lock);
285
0883beca 286 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
1da177e4
LT		 287 if (vlan->vlan_id == vlan_id) {
288 list_del(&vlan->vlan_list);
289
58402054 290 if (bond_is_lb(bond))
1da177e4 291 bond_alb_clear_vlan(bond, vlan_id);
1da177e4 292
5a03cdb7 293 pr_debug("removed VLAN ID %d from bond %s\n", vlan_id,
1da177e4
LT		 294 bond->dev->name);
295
296 kfree(vlan);
297
298 if (list_empty(&bond->vlan_list) &&
299 (bond->slave_cnt == 0)) {
300 /* Last VLAN removed and no slaves, so
301 * restore block on adding VLANs. This will
302 * be removed once new slaves that are not
303 * VLAN challenged will be added.
304 */
305 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
306 }
307
308 res = 0;
309 goto out;
310 }
311 }
312
5a03cdb7 313 pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id,
1da177e4
LT		 314 bond->dev->name);
315
316out:
317 write_unlock_bh(&bond->lock);
318 return res;
319}
320
321/**
322 * bond_has_challenged_slaves
323 * @bond: the bond we're working on
324 *
325 * Searches the slave list. Returns 1 if a vlan challenged slave
326 * was found, 0 otherwise.
327 *
328 * Assumes bond->lock is held.
329 */
330static int bond_has_challenged_slaves(struct bonding *bond)
331{
332 struct slave *slave;
333 int i;
334
335 bond_for_each_slave(bond, slave, i) {
336 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
5a03cdb7 337 pr_debug("found VLAN challenged slave - %s\n",
1da177e4
LT		 338 slave->dev->name);
339 return 1;
340 }
341 }
342
5a03cdb7 343 pr_debug("no VLAN challenged slaves found\n");
1da177e4
LT		 344 return 0;
345}
346
347/**
348 * bond_next_vlan - safely skip to the next item in the vlans list.
349 * @bond: the bond we're working on
350 * @curr: item we're advancing from
351 *
352 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
353 * or @curr->next otherwise (even if it is @curr itself again).
354 *
355 * Caller must hold bond->lock
356 */
357struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
358{
359 struct vlan_entry *next, *last;
360
361 if (list_empty(&bond->vlan_list)) {
362 return NULL;
363 }
364
365 if (!curr) {
366 next = list_entry(bond->vlan_list.next,
367 struct vlan_entry, vlan_list);
368 } else {
369 last = list_entry(bond->vlan_list.prev,
370 struct vlan_entry, vlan_list);
371 if (last == curr) {
372 next = list_entry(bond->vlan_list.next,
373 struct vlan_entry, vlan_list);
374 } else {
375 next = list_entry(curr->vlan_list.next,
376 struct vlan_entry, vlan_list);
377 }
378 }
379
380 return next;
381}
382
383/**
384 * bond_dev_queue_xmit - Prepare skb for xmit.
385 *
386 * @bond: bond device that got this skb for tx.
387 * @skb: hw accel VLAN tagged skb to transmit
388 * @slave_dev: slave that is supposed to xmit this skbuff
389 *
390 * When the bond gets an skb to transmit that is
391 * already hardware accelerated VLAN tagged, and it
392 * needs to relay this skb to a slave that is not
393 * hw accel capable, the skb needs to be "unaccelerated",
394 * i.e. strip the hwaccel tag and re-insert it as part
395 * of the payload.
396 */
397int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
398{
966bc6f4 399 unsigned short uninitialized_var(vlan_id);
1da177e4
LT		 400
401 if (!list_empty(&bond->vlan_list) &&
402 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
403 vlan_get_tag(skb, &vlan_id) == 0) {
404 skb->dev = slave_dev;
405 skb = vlan_put_tag(skb, vlan_id);
406 if (!skb) {
407 /* vlan_put_tag() frees the skb in case of error,
408 * so return success here so the calling functions
409 * won't attempt to free is again.
410 */
411 return 0;
412 }
413 } else {
414 skb->dev = slave_dev;
415 }
416
417 skb->priority = 1;
418 dev_queue_xmit(skb);
419
420 return 0;
421}
422
423/*
424 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
425 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
426 * lock because:
427 * a. This operation is performed in IOCTL context,
428 * b. The operation is protected by the RTNL semaphore in the 8021q code,
429 * c. Holding a lock with BH disabled while directly calling a base driver
430 * entry point is generally a BAD idea.
431 *
432 * The design of synchronization/protection for this operation in the 8021q
433 * module is good for one or more VLAN devices over a single physical device
434 * and cannot be extended for a teaming solution like bonding, so there is a
435 * potential race condition here where a net device from the vlan group might
436 * be referenced (either by a base driver or the 8021q code) while it is being
437 * removed from the system. However, it turns out we're not making matters
438 * worse, and if it works for regular VLAN usage it will work here too.
439*/
440
441/**
442 * bond_vlan_rx_register - Propagates registration to slaves
443 * @bond_dev: bonding net device that got called
444 * @grp: vlan group being registered
445 */
446static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
447{
454d7c9b 448 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 449 struct slave *slave;
450 int i;
451
452 bond->vlgrp = grp;
453
454 bond_for_each_slave(bond, slave, i) {
455 struct net_device *slave_dev = slave->dev;
eb7cc59a 456 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 457
458 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 459 slave_ops->ndo_vlan_rx_register) {
460 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
1da177e4
LT		 461 }
462 }
463}
464
465/**
466 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
467 * @bond_dev: bonding net device that got called
468 * @vid: vlan id being added
469 */
470static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
471{
454d7c9b 472 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 473 struct slave *slave;
474 int i, res;
475
476 bond_for_each_slave(bond, slave, i) {
477 struct net_device *slave_dev = slave->dev;
eb7cc59a 478 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 479
480 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a
SH		 481 slave_ops->ndo_vlan_rx_add_vid) {
482 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
1da177e4
LT		 483 }
484 }
485
486 res = bond_add_vlan(bond, vid);
487 if (res) {
488 printk(KERN_ERR DRV_NAME
4e0952c7 489 ": %s: Error: Failed to add vlan id %d\n",
1da177e4
LT		 490 bond_dev->name, vid);
491 }
492}
493
494/**
495 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
496 * @bond_dev: bonding net device that got called
497 * @vid: vlan id being removed
498 */
499static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
500{
454d7c9b 501 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 502 struct slave *slave;
503 struct net_device *vlan_dev;
504 int i, res;
505
506 bond_for_each_slave(bond, slave, i) {
507 struct net_device *slave_dev = slave->dev;
eb7cc59a 508 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 509
510 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a 511 slave_ops->ndo_vlan_rx_kill_vid) {
1da177e4
LT		 512 /* Save and then restore vlan_dev in the grp array,
513 * since the slave's driver might clear it.
514 */
5c15bdec 515 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
eb7cc59a 516 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
5c15bdec 517 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
1da177e4
LT		 518 }
519 }
520
521 res = bond_del_vlan(bond, vid);
522 if (res) {
523 printk(KERN_ERR DRV_NAME
4e0952c7 524 ": %s: Error: Failed to remove vlan id %d\n",
1da177e4
LT		 525 bond_dev->name, vid);
526 }
527}
528
529static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
530{
531 struct vlan_entry *vlan;
eb7cc59a 532 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 533
534 write_lock_bh(&bond->lock);
535
eb7cc59a 536 if (list_empty(&bond->vlan_list))
1da177e4 537 goto out;
1da177e4
LT		 538
539 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 540 slave_ops->ndo_vlan_rx_register)
541 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
1da177e4
LT		 542
543 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 544 !(slave_ops->ndo_vlan_rx_add_vid))
1da177e4 545 goto out;
1da177e4 546
eb7cc59a
SH		 547 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
548 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
1da177e4
LT		 549
550out:
551 write_unlock_bh(&bond->lock);
552}
553
554static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
555{
eb7cc59a 556 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 557 struct vlan_entry *vlan;
558 struct net_device *vlan_dev;
559
560 write_lock_bh(&bond->lock);
561
eb7cc59a 562 if (list_empty(&bond->vlan_list))
1da177e4 563 goto out;
1da177e4
LT		 564
565 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 566 !(slave_ops->ndo_vlan_rx_kill_vid))
1da177e4 567 goto unreg;
1da177e4
LT		 568
569 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
570 /* Save and then restore vlan_dev in the grp array,
571 * since the slave's driver might clear it.
572 */
5c15bdec 573 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
eb7cc59a 574 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
5c15bdec 575 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
1da177e4
LT		 576 }
577
578unreg:
579 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 580 slave_ops->ndo_vlan_rx_register)
581 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
1da177e4
LT		 582
583out:
584 write_unlock_bh(&bond->lock);
585}
586
587/*------------------------------- Link status -------------------------------*/
588
ff59c456
JV		 589/*
590 * Set the carrier state for the master according to the state of its
591 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
592 * do special 802.3ad magic.
593 *
594 * Returns zero if carrier state does not change, nonzero if it does.
595 */
596static int bond_set_carrier(struct bonding *bond)
597{
598 struct slave *slave;
599 int i;
600
601 if (bond->slave_cnt == 0)
602 goto down;
603
604 if (bond->params.mode == BOND_MODE_8023AD)
605 return bond_3ad_set_carrier(bond);
606
607 bond_for_each_slave(bond, slave, i) {
608 if (slave->link == BOND_LINK_UP) {
609 if (!netif_carrier_ok(bond->dev)) {
610 netif_carrier_on(bond->dev);
611 return 1;
612 }
613 return 0;
614 }
615 }
616
617down:
618 if (netif_carrier_ok(bond->dev)) {
619 netif_carrier_off(bond->dev);
620 return 1;
621 }
622 return 0;
623}
624
1da177e4
LT		 625/*
626 * Get link speed and duplex from the slave's base driver
627 * using ethtool. If for some reason the call fails or the
628 * values are invalid, fake speed and duplex to 100/Full
629 * and return error.
630 */
631static int bond_update_speed_duplex(struct slave *slave)
632{
633 struct net_device *slave_dev = slave->dev;
1da177e4 634 struct ethtool_cmd etool;
61a44b9c 635 int res;
1da177e4
LT		 636
637 /* Fake speed and duplex */
638 slave->speed = SPEED_100;
639 slave->duplex = DUPLEX_FULL;
640
61a44b9c
MW		 641 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
642 return -1;
1da177e4 643
61a44b9c
MW		 644 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
645 if (res < 0)
1da177e4 646 return -1;
1da177e4 647
1da177e4
LT		 648 switch (etool.speed) {
649 case SPEED_10:
650 case SPEED_100:
651 case SPEED_1000:
94dbffd5 652 case SPEED_10000:
1da177e4
LT		 653 break;
654 default:
655 return -1;
656 }
657
658 switch (etool.duplex) {
659 case DUPLEX_FULL:
660 case DUPLEX_HALF:
661 break;
662 default:
663 return -1;
664 }
665
666 slave->speed = etool.speed;
667 slave->duplex = etool.duplex;
668
669 return 0;
670}
671
672/*
673 * if <dev> supports MII link status reporting, check its link status.
674 *
675 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
676 * depening upon the setting of the use_carrier parameter.
677 *
678 * Return either BMSR_LSTATUS, meaning that the link is up (or we
679 * can't tell and just pretend it is), or 0, meaning that the link is
680 * down.
681 *
682 * If reporting is non-zero, instead of faking link up, return -1 if
683 * both ETHTOOL and MII ioctls fail (meaning the device does not
684 * support them). If use_carrier is set, return whatever it says.
685 * It'd be nice if there was a good way to tell if a driver supports
686 * netif_carrier, but there really isn't.
687 */
688static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
689{
eb7cc59a 690 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 691 static int (* ioctl)(struct net_device *, struct ifreq *, int);
692 struct ifreq ifr;
693 struct mii_ioctl_data *mii;
1da177e4 694
eb7cc59a 695 if (bond->params.use_carrier)
1da177e4 696 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1da177e4 697
eb7cc59a 698 ioctl = slave_ops->ndo_do_ioctl;
1da177e4
LT		 699 if (ioctl) {
700 /* TODO: set pointer to correct ioctl on a per team member */
701 /* bases to make this more efficient. that is, once */
702 /* we determine the correct ioctl, we will always */
703 /* call it and not the others for that team */
704 /* member. */
705
706 /*
707 * We cannot assume that SIOCGMIIPHY will also read a
708 * register; not all network drivers (e.g., e100)
709 * support that.
710 */
711
712 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
713 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
714 mii = if_mii(&ifr);
715 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
716 mii->reg_num = MII_BMSR;
717 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
718 return (mii->val_out & BMSR_LSTATUS);
719 }
720 }
721 }
722
61a44b9c
MW		 723 /*
724 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
725 * attempt to get link status from it if the above MII ioctls fail.
726 */
1da177e4
LT		 727 if (slave_dev->ethtool_ops) {
728 if (slave_dev->ethtool_ops->get_link) {
729 u32 link;
730
731 link = slave_dev->ethtool_ops->get_link(slave_dev);
732
733 return link ? BMSR_LSTATUS : 0;
734 }
735 }
736
1da177e4
LT		 737 /*
738 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 739 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 740 * cannot report link status). If not reporting, pretend
741 * we're ok.
742 */
743 return (reporting ? -1 : BMSR_LSTATUS);
744}
745
746/*----------------------------- Multicast list ------------------------------*/
747
748/*
749 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
750 */
751static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
752{
753 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
754 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
755}
756
757/*
758 * returns dmi entry if found, NULL otherwise
759 */
760static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
761{
762 struct dev_mc_list *idmi;
763
764 for (idmi = mc_list; idmi; idmi = idmi->next) {
765 if (bond_is_dmi_same(dmi, idmi)) {
766 return idmi;
767 }
768 }
769
770 return NULL;
771}
772
773/*
774 * Push the promiscuity flag down to appropriate slaves
775 */
7e1a1ac1 776static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 777{
7e1a1ac1 778 int err = 0;
1da177e4
LT		 779 if (USES_PRIMARY(bond->params.mode)) {
780 /* write lock already acquired */
781 if (bond->curr_active_slave) {
7e1a1ac1
WC		 782 err = dev_set_promiscuity(bond->curr_active_slave->dev,
783 inc);
1da177e4
LT		 784 }
785 } else {
786 struct slave *slave;
787 int i;
788 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 789 err = dev_set_promiscuity(slave->dev, inc);
790 if (err)
791 return err;
1da177e4
LT		 792 }
793 }
7e1a1ac1 794 return err;
1da177e4
LT		 795}
796
797/*
798 * Push the allmulti flag down to all slaves
799 */
7e1a1ac1 800static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 801{
7e1a1ac1 802 int err = 0;
1da177e4
LT		 803 if (USES_PRIMARY(bond->params.mode)) {
804 /* write lock already acquired */
805 if (bond->curr_active_slave) {
7e1a1ac1
WC		 806 err = dev_set_allmulti(bond->curr_active_slave->dev,
807 inc);
1da177e4
LT		 808 }
809 } else {
810 struct slave *slave;
811 int i;
812 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 813 err = dev_set_allmulti(slave->dev, inc);
814 if (err)
815 return err;
1da177e4
LT		 816 }
817 }
7e1a1ac1 818 return err;
1da177e4
LT		 819}
820
821/*
822 * Add a Multicast address to slaves
823 * according to mode
824 */
825static void bond_mc_add(struct bonding *bond, void *addr, int alen)
826{
827 if (USES_PRIMARY(bond->params.mode)) {
828 /* write lock already acquired */
829 if (bond->curr_active_slave) {
830 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
831 }
832 } else {
833 struct slave *slave;
834 int i;
835 bond_for_each_slave(bond, slave, i) {
836 dev_mc_add(slave->dev, addr, alen, 0);
837 }
838 }
839}
840
841/*
842 * Remove a multicast address from slave
843 * according to mode
844 */
845static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
846{
847 if (USES_PRIMARY(bond->params.mode)) {
848 /* write lock already acquired */
849 if (bond->curr_active_slave) {
850 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
851 }
852 } else {
853 struct slave *slave;
854 int i;
855 bond_for_each_slave(bond, slave, i) {
856 dev_mc_delete(slave->dev, addr, alen, 0);
857 }
858 }
859}
860
a816c7c7
JV		 861
862/*
863 * Retrieve the list of registered multicast addresses for the bonding
864 * device and retransmit an IGMP JOIN request to the current active
865 * slave.
866 */
867static void bond_resend_igmp_join_requests(struct bonding *bond)
868{
869 struct in_device *in_dev;
870 struct ip_mc_list *im;
871
872 rcu_read_lock();
873 in_dev = __in_dev_get_rcu(bond->dev);
874 if (in_dev) {
875 for (im = in_dev->mc_list; im; im = im->next) {
876 ip_mc_rejoin_group(im);
877 }
878 }
879
880 rcu_read_unlock();
881}
882
1da177e4
LT		 883/*
884 * Totally destroys the mc_list in bond
885 */
886static void bond_mc_list_destroy(struct bonding *bond)
887{
888 struct dev_mc_list *dmi;
889
890 dmi = bond->mc_list;
891 while (dmi) {
892 bond->mc_list = dmi->next;
893 kfree(dmi);
894 dmi = bond->mc_list;
895 }
a816c7c7 896 bond->mc_list = NULL;
1da177e4
LT		 897}
898
899/*
900 * Copy all the Multicast addresses from src to the bonding device dst
901 */
de54f390 902static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 903 gfp_t gfp_flag)
1da177e4
LT		 904{
905 struct dev_mc_list *dmi, *new_dmi;
906
907 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 908 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT		 909
910 if (!new_dmi) {
911 /* FIXME: Potential memory leak !!! */
912 return -ENOMEM;
913 }
914
915 new_dmi->next = bond->mc_list;
916 bond->mc_list = new_dmi;
917 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
918 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
919 new_dmi->dmi_users = dmi->dmi_users;
920 new_dmi->dmi_gusers = dmi->dmi_gusers;
921 }
922
923 return 0;
924}
925
926/*
927 * flush all members of flush->mc_list from device dev->mc_list
928 */
929static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
930{
454d7c9b 931 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 932 struct dev_mc_list *dmi;
933
934 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
935 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
936 }
937
938 if (bond->params.mode == BOND_MODE_8023AD) {
939 /* del lacpdu mc addr from mc list */
940 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
941
942 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
943 }
944}
945
946/*--------------------------- Active slave change ---------------------------*/
947
948/*
949 * Update the mc list and multicast-related flags for the new and
950 * old active slaves (if any) according to the multicast mode, and
951 * promiscuous flags unconditionally.
952 */
953static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
954{
955 struct dev_mc_list *dmi;
956
957 if (!USES_PRIMARY(bond->params.mode)) {
958 /* nothing to do - mc list is already up-to-date on
959 * all slaves
960 */
961 return;
962 }
963
964 if (old_active) {
965 if (bond->dev->flags & IFF_PROMISC) {
966 dev_set_promiscuity(old_active->dev, -1);
967 }
968
969 if (bond->dev->flags & IFF_ALLMULTI) {
970 dev_set_allmulti(old_active->dev, -1);
971 }
972
973 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
974 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
975 }
976 }
977
978 if (new_active) {
7e1a1ac1 979 /* FIXME: Signal errors upstream. */
1da177e4
LT		 980 if (bond->dev->flags & IFF_PROMISC) {
981 dev_set_promiscuity(new_active->dev, 1);
982 }
983
984 if (bond->dev->flags & IFF_ALLMULTI) {
985 dev_set_allmulti(new_active->dev, 1);
986 }
987
988 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
989 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
990 }
a816c7c7 991 bond_resend_igmp_join_requests(bond);
1da177e4
LT		 992 }
993}
994
3915c1e8
JV		 995/*
996 * bond_do_fail_over_mac
997 *
998 * Perform special MAC address swapping for fail_over_mac settings
999 *
1000 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1001 */
1002static void bond_do_fail_over_mac(struct bonding *bond,
1003 struct slave *new_active,
1004 struct slave *old_active)
1f78d9f9
HE		 1005 __releases(&bond->curr_slave_lock)
1006 __releases(&bond->lock)
1007 __acquires(&bond->lock)
1008 __acquires(&bond->curr_slave_lock)
3915c1e8
JV		 1009{
1010 u8 tmp_mac[ETH_ALEN];
1011 struct sockaddr saddr;
1012 int rv;
1013
1014 switch (bond->params.fail_over_mac) {
1015 case BOND_FOM_ACTIVE:
1016 if (new_active)
1017 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1018 new_active->dev->addr_len);
1019 break;
1020 case BOND_FOM_FOLLOW:
1021 /*
1022 * if new_active && old_active, swap them
1023 * if just old_active, do nothing (going to no active slave)
1024 * if just new_active, set new_active to bond's MAC
1025 */
1026 if (!new_active)
1027 return;
1028
1029 write_unlock_bh(&bond->curr_slave_lock);
1030 read_unlock(&bond->lock);
1031
1032 if (old_active) {
1033 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1034 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1035 ETH_ALEN);
1036 saddr.sa_family = new_active->dev->type;
1037 } else {
1038 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1039 saddr.sa_family = bond->dev->type;
1040 }
1041
1042 rv = dev_set_mac_address(new_active->dev, &saddr);
1043 if (rv) {
1044 printk(KERN_ERR DRV_NAME
1045 ": %s: Error %d setting MAC of slave %s\n",
1046 bond->dev->name, -rv, new_active->dev->name);
1047 goto out;
1048 }
1049
1050 if (!old_active)
1051 goto out;
1052
1053 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1054 saddr.sa_family = old_active->dev->type;
1055
1056 rv = dev_set_mac_address(old_active->dev, &saddr);
1057 if (rv)
1058 printk(KERN_ERR DRV_NAME
1059 ": %s: Error %d setting MAC of slave %s\n",
1060 bond->dev->name, -rv, new_active->dev->name);
1061out:
1062 read_lock(&bond->lock);
1063 write_lock_bh(&bond->curr_slave_lock);
1064 break;
1065 default:
1066 printk(KERN_ERR DRV_NAME
1067 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1068 bond->dev->name, bond->params.fail_over_mac);
1069 break;
1070 }
1071
1072}
1073
1074
1da177e4
LT		 1075/**
1076 * find_best_interface - select the best available slave to be the active one
1077 * @bond: our bonding struct
1078 *
1079 * Warning: Caller must hold curr_slave_lock for writing.
1080 */
1081static struct slave *bond_find_best_slave(struct bonding *bond)
1082{
1083 struct slave *new_active, *old_active;
1084 struct slave *bestslave = NULL;
1085 int mintime = bond->params.updelay;
1086 int i;
1087
1088 new_active = old_active = bond->curr_active_slave;
1089
1090 if (!new_active) { /* there were no active slaves left */
1091 if (bond->slave_cnt > 0) { /* found one slave */
1092 new_active = bond->first_slave;
1093 } else {
1094 return NULL; /* still no slave, return NULL */
1095 }
1096 }
1097
1098 /* first try the primary link; if arping, a link must tx/rx traffic
1099 * before it can be considered the curr_active_slave - also, we would skip
1100 * slaves between the curr_active_slave and primary_slave that may be up
1101 * and able to arp
1102 */
1103 if ((bond->primary_slave) &&
1104 (!bond->params.arp_interval) &&
1105 (IS_UP(bond->primary_slave->dev))) {
1106 new_active = bond->primary_slave;
1107 }
1108
1109 /* remember where to stop iterating over the slaves */
1110 old_active = new_active;
1111
1112 bond_for_each_slave_from(bond, new_active, i, old_active) {
1113 if (IS_UP(new_active->dev)) {
1114 if (new_active->link == BOND_LINK_UP) {
1115 return new_active;
1116 } else if (new_active->link == BOND_LINK_BACK) {
1117 /* link up, but waiting for stabilization */
1118 if (new_active->delay < mintime) {
1119 mintime = new_active->delay;
1120 bestslave = new_active;
1121 }
1122 }
1123 }
1124 }
1125
1126 return bestslave;
1127}
1128
1129/**
1130 * change_active_interface - change the active slave into the specified one
1131 * @bond: our bonding struct
1132 * @new: the new slave to make the active one
1133 *
1134 * Set the new slave to the bond's settings and unset them on the old
1135 * curr_active_slave.
1136 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1137 *
1138 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1139 * because it is apparently the best available slave we have, even though its
1140 * updelay hasn't timed out yet.
1141 *
3915c1e8
JV		 1142 * If new_active is not NULL, caller must hold bond->lock for read and
1143 * curr_slave_lock for write_bh.
1da177e4 1144 */
a77b5325 1145void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1146{
1147 struct slave *old_active = bond->curr_active_slave;
1148
1149 if (old_active == new_active) {
1150 return;
1151 }
1152
1153 if (new_active) {
b2220cad
JV		 1154 new_active->jiffies = jiffies;
1155
1da177e4
LT		 1156 if (new_active->link == BOND_LINK_BACK) {
1157 if (USES_PRIMARY(bond->params.mode)) {
1158 printk(KERN_INFO DRV_NAME
1159 ": %s: making interface %s the new "
1160 "active one %d ms earlier.\n",
1161 bond->dev->name, new_active->dev->name,
1162 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1163 }
1164
1165 new_active->delay = 0;
1166 new_active->link = BOND_LINK_UP;
1da177e4
LT		 1167
1168 if (bond->params.mode == BOND_MODE_8023AD) {
1169 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1170 }
1171
58402054 1172 if (bond_is_lb(bond))
1da177e4 1173 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1da177e4
LT		 1174 } else {
1175 if (USES_PRIMARY(bond->params.mode)) {
1176 printk(KERN_INFO DRV_NAME
1177 ": %s: making interface %s the new "
1178 "active one.\n",
1179 bond->dev->name, new_active->dev->name);
1180 }
1181 }
1182 }
1183
1da177e4
LT		 1184 if (USES_PRIMARY(bond->params.mode)) {
1185 bond_mc_swap(bond, new_active, old_active);
1186 }
1187
58402054 1188 if (bond_is_lb(bond)) {
1da177e4 1189 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1190 if (old_active)
1191 bond_set_slave_inactive_flags(old_active);
1192 if (new_active)
1193 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1194 } else {
1195 bond->curr_active_slave = new_active;
1196 }
c3ade5ca
JV		 1197
1198 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1199 if (old_active) {
1200 bond_set_slave_inactive_flags(old_active);
1201 }
1202
1203 if (new_active) {
1204 bond_set_slave_active_flags(new_active);
2ab82852 1205
709f8a45
OG		 1206 if (bond->params.fail_over_mac)
1207 bond_do_fail_over_mac(bond, new_active,
1208 old_active);
3915c1e8 1209
709f8a45 1210 bond->send_grat_arp = bond->params.num_grat_arp;
b59f9f74 1211 bond_send_gratuitous_arp(bond);
01f3109d 1212
305d552a
BH		 1213 bond->send_unsol_na = bond->params.num_unsol_na;
1214 bond_send_unsolicited_na(bond);
1215
01f3109d
OG		 1216 write_unlock_bh(&bond->curr_slave_lock);
1217 read_unlock(&bond->lock);
1218
1219 netdev_bonding_change(bond->dev);
1220
1221 read_lock(&bond->lock);
1222 write_lock_bh(&bond->curr_slave_lock);
7893b249 1223 }
c3ade5ca 1224 }
1da177e4
LT		 1225}
1226
1227/**
1228 * bond_select_active_slave - select a new active slave, if needed
1229 * @bond: our bonding struct
1230 *
1231 * This functions shoud be called when one of the following occurs:
1232 * - The old curr_active_slave has been released or lost its link.
1233 * - The primary_slave has got its link back.
1234 * - A slave has got its link back and there's no old curr_active_slave.
1235 *
3915c1e8 1236 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1237 */
a77b5325 1238void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1239{
1240 struct slave *best_slave;
ff59c456 1241 int rv;
1da177e4
LT		 1242
1243 best_slave = bond_find_best_slave(bond);
1244 if (best_slave != bond->curr_active_slave) {
1245 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1246 rv = bond_set_carrier(bond);
1247 if (!rv)
1248 return;
1249
1250 if (netif_carrier_ok(bond->dev)) {
1251 printk(KERN_INFO DRV_NAME
1252 ": %s: first active interface up!\n",
1253 bond->dev->name);
1254 } else {
1255 printk(KERN_INFO DRV_NAME ": %s: "
1256 "now running without any active interface !\n",
1257 bond->dev->name);
1258 }
1da177e4
LT		 1259 }
1260}
1261
1262/*--------------------------- slave list handling ---------------------------*/
1263
1264/*
1265 * This function attaches the slave to the end of list.
1266 *
1267 * bond->lock held for writing by caller.
1268 */
1269static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1270{
1271 if (bond->first_slave == NULL) { /* attaching the first slave */
1272 new_slave->next = new_slave;
1273 new_slave->prev = new_slave;
1274 bond->first_slave = new_slave;
1275 } else {
1276 new_slave->next = bond->first_slave;
1277 new_slave->prev = bond->first_slave->prev;
1278 new_slave->next->prev = new_slave;
1279 new_slave->prev->next = new_slave;
1280 }
1281
1282 bond->slave_cnt++;
1283}
1284
1285/*
1286 * This function detaches the slave from the list.
1287 * WARNING: no check is made to verify if the slave effectively
1288 * belongs to <bond>.
1289 * Nothing is freed on return, structures are just unchained.
1290 * If any slave pointer in bond was pointing to <slave>,
1291 * it should be changed by the calling function.
1292 *
1293 * bond->lock held for writing by caller.
1294 */
1295static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1296{
1297 if (slave->next) {
1298 slave->next->prev = slave->prev;
1299 }
1300
1301 if (slave->prev) {
1302 slave->prev->next = slave->next;
1303 }
1304
1305 if (bond->first_slave == slave) { /* slave is the first slave */
1306 if (bond->slave_cnt > 1) { /* there are more slave */
1307 bond->first_slave = slave->next;
1308 } else {
1309 bond->first_slave = NULL; /* slave was the last one */
1310 }
1311 }
1312
1313 slave->next = NULL;
1314 slave->prev = NULL;
1315 bond->slave_cnt--;
1316}
1317
1318/*---------------------------------- IOCTL ----------------------------------*/
1319
4ad072c9
AB		 1320static int bond_sethwaddr(struct net_device *bond_dev,
1321 struct net_device *slave_dev)
1da177e4 1322{
5a03cdb7
HE		 1323 pr_debug("bond_dev=%p\n", bond_dev);
1324 pr_debug("slave_dev=%p\n", slave_dev);
1325 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1da177e4
LT		 1326 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1327 return 0;
1328}
1329
7f353bf2
HX		 1330#define BOND_VLAN_FEATURES \
1331 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1332 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK		 1333
1334/*
8e3babcd 1335 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX		 1336 * feature bits are managed elsewhere, so preserve those feature bits
1337 * on the master device.
8531c5ff
AK		 1338 */
1339static int bond_compute_features(struct bonding *bond)
1340{
8531c5ff
AK		 1341 struct slave *slave;
1342 struct net_device *bond_dev = bond->dev;
7f353bf2 1343 unsigned long features = bond_dev->features;
3158bf7d
MS		 1344 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1345 bond_dev->hard_header_len);
8e3babcd 1346 int i;
8531c5ff 1347
7f353bf2 1348 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
b63365a2
HX		 1349 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1350
1351 if (!bond->first_slave)
1352 goto done;
1353
1354 features &= ~NETIF_F_ONE_FOR_ALL;
7f353bf2 1355
54ef3137 1356 bond_for_each_slave(bond, slave, i) {
b63365a2
HX		 1357 features = netdev_increment_features(features,
1358 slave->dev->features,
1359 NETIF_F_ONE_FOR_ALL);
54ef3137
JV		 1360 if (slave->dev->hard_header_len > max_hard_header_len)
1361 max_hard_header_len = slave->dev->hard_header_len;
1362 }
8531c5ff 1363
b63365a2 1364done:
7f353bf2 1365 features |= (bond_dev->features & BOND_VLAN_FEATURES);
b63365a2 1366 bond_dev->features = netdev_fix_features(features, NULL);
54ef3137 1367 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1368
1369 return 0;
1370}
1371
872254dd
MS		 1372static void bond_setup_by_slave(struct net_device *bond_dev,
1373 struct net_device *slave_dev)
1374{
454d7c9b 1375 struct bonding *bond = netdev_priv(bond_dev);
d90a162a 1376
00829823 1377 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1378
1379 bond_dev->type = slave_dev->type;
1380 bond_dev->hard_header_len = slave_dev->hard_header_len;
1381 bond_dev->addr_len = slave_dev->addr_len;
1382
1383 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1384 slave_dev->addr_len);
d90a162a 1385 bond->setup_by_slave = 1;
872254dd
MS		 1386}
1387
1da177e4 1388/* enslave device <slave> to bond device <master> */
a77b5325 1389int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1390{
454d7c9b 1391 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1392 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 1393 struct slave *new_slave = NULL;
1394 struct dev_mc_list *dmi;
1395 struct sockaddr addr;
1396 int link_reporting;
1397 int old_features = bond_dev->features;
1398 int res = 0;
1399
552709d5 1400 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
eb7cc59a 1401 slave_ops->ndo_do_ioctl == NULL) {
1da177e4 1402 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1403 ": %s: Warning: no link monitoring support for %s\n",
1404 bond_dev->name, slave_dev->name);
1da177e4
LT		 1405 }
1406
1407 /* bond must be initialized by bond_open() before enslaving */
1408 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1409 printk(KERN_WARNING DRV_NAME
1410 " %s: master_dev is not up in bond_enslave\n",
1411 bond_dev->name);
1da177e4
LT		 1412 }
1413
1414 /* already enslaved */
1415 if (slave_dev->flags & IFF_SLAVE) {
5a03cdb7 1416 pr_debug("Error, Device was already enslaved\n");
1da177e4
LT		 1417 return -EBUSY;
1418 }
1419
1420 /* vlan challenged mutual exclusion */
1421 /* no need to lock since we're protected by rtnl_lock */
1422 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
5a03cdb7 1423 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1da177e4
LT		 1424 if (!list_empty(&bond->vlan_list)) {
1425 printk(KERN_ERR DRV_NAME
4e0952c7 1426 ": %s: Error: cannot enslave VLAN "
1da177e4 1427 "challenged slave %s on VLAN enabled "
4e0952c7 1428 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1429 bond_dev->name);
1430 return -EPERM;
1431 } else {
1432 printk(KERN_WARNING DRV_NAME
4e0952c7 1433 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1434 "slave %s. Adding VLANs will be blocked as "
1435 "long as %s is part of bond %s\n",
4e0952c7 1436 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1437 bond_dev->name);
1438 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1439 }
1440 } else {
5a03cdb7 1441 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1da177e4
LT		 1442 if (bond->slave_cnt == 0) {
1443 /* First slave, and it is not VLAN challenged,
1444 * so remove the block of adding VLANs over the bond.
1445 */
1446 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1447 }
1448 }
1449
217df670
JV		 1450 /*
1451 * Old ifenslave binaries are no longer supported. These can
1452 * be identified with moderate accurary by the state of the slave:
1453 * the current ifenslave will set the interface down prior to
1454 * enslaving it; the old ifenslave will not.
1455 */
1456 if ((slave_dev->flags & IFF_UP)) {
1457 printk(KERN_ERR DRV_NAME ": %s is up. "
1458 "This may be due to an out of date ifenslave.\n",
1459 slave_dev->name);
1460 res = -EPERM;
1461 goto err_undo_flags;
1462 }
1da177e4 1463
872254dd
MS		 1464 /* set bonding device ether type by slave - bonding netdevices are
1465 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1466 * there is a need to override some of the type dependent attribs/funcs.
1467 *
1468 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1469 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1470 */
1471 if (bond->slave_cnt == 0) {
1472 if (slave_dev->type != ARPHRD_ETHER)
1473 bond_setup_by_slave(bond_dev, slave_dev);
1474 } else if (bond_dev->type != slave_dev->type) {
1475 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1476 "from other slaves (%d), can not enslave it.\n",
1477 slave_dev->name,
1478 slave_dev->type, bond_dev->type);
1479 res = -EINVAL;
1480 goto err_undo_flags;
1481 }
1482
eb7cc59a 1483 if (slave_ops->ndo_set_mac_address == NULL) {
2ab82852
MS		 1484 if (bond->slave_cnt == 0) {
1485 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1486 ": %s: Warning: The first slave device "
1487 "specified does not support setting the MAC "
3915c1e8 1488 "address. Setting fail_over_mac to active.",
dd957c57 1489 bond_dev->name);
3915c1e8
JV		 1490 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1491 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1492 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1493 ": %s: Error: The slave device specified "
1494 "does not support setting the MAC address, "
3915c1e8 1495 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1496 , bond_dev->name);
1497 res = -EOPNOTSUPP;
1498 goto err_undo_flags;
1499 }
1da177e4
LT		 1500 }
1501
243cb4e5 1502 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1503 if (!new_slave) {
1504 res = -ENOMEM;
1505 goto err_undo_flags;
1506 }
1507
1da177e4
LT		 1508 /* save slave's original flags before calling
1509 * netdev_set_master and dev_open
1510 */
1511 new_slave->original_flags = slave_dev->flags;
1512
217df670
JV		 1513 /*
1514 * Save slave's original ("permanent") mac address for modes
1515 * that need it, and for restoring it upon release, and then
1516 * set it to the master's address
1517 */
1518 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1519
dd957c57 1520 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1521 /*
1522 * Set slave to master's mac address. The application already
1523 * set the master's mac address to that of the first slave
1524 */
1525 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1526 addr.sa_family = slave_dev->type;
1527 res = dev_set_mac_address(slave_dev, &addr);
1528 if (res) {
5a03cdb7 1529 pr_debug("Error %d calling set_mac_address\n", res);
2ab82852
MS		 1530 goto err_free;
1531 }
217df670 1532 }
1da177e4 1533
c2edacf8
JV		 1534 res = netdev_set_master(slave_dev, bond_dev);
1535 if (res) {
5a03cdb7 1536 pr_debug("Error %d calling netdev_set_master\n", res);
569f0c4d 1537 goto err_restore_mac;
c2edacf8 1538 }
217df670
JV		 1539 /* open the slave since the application closed it */
1540 res = dev_open(slave_dev);
1541 if (res) {
5a03cdb7 1542 pr_debug("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1543 goto err_unset_master;
1da177e4
LT		 1544 }
1545
1da177e4 1546 new_slave->dev = slave_dev;
0b680e75 1547 slave_dev->priv_flags |= IFF_BONDING;
1da177e4 1548
58402054 1549 if (bond_is_lb(bond)) {
1da177e4
LT		 1550 /* bond_alb_init_slave() must be called before all other stages since
1551 * it might fail and we do not want to have to undo everything
1552 */
1553 res = bond_alb_init_slave(bond, new_slave);
1554 if (res) {
569f0c4d 1555 goto err_close;
1da177e4
LT		 1556 }
1557 }
1558
1559 /* If the mode USES_PRIMARY, then the new slave gets the
1560 * master's promisc (and mc) settings only if it becomes the
1561 * curr_active_slave, and that is taken care of later when calling
1562 * bond_change_active()
1563 */
1564 if (!USES_PRIMARY(bond->params.mode)) {
1565 /* set promiscuity level to new slave */
1566 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1567 res = dev_set_promiscuity(slave_dev, 1);
1568 if (res)
1569 goto err_close;
1da177e4
LT		 1570 }
1571
1572 /* set allmulti level to new slave */
1573 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1574 res = dev_set_allmulti(slave_dev, 1);
1575 if (res)
1576 goto err_close;
1da177e4
LT		 1577 }
1578
b9e40857 1579 netif_addr_lock_bh(bond_dev);
1da177e4
LT		 1580 /* upload master's mc_list to new slave */
1581 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1582 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1583 }
b9e40857 1584 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1585 }
1586
1587 if (bond->params.mode == BOND_MODE_8023AD) {
1588 /* add lacpdu mc addr to mc list */
1589 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1590
1591 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1592 }
1593
1594 bond_add_vlans_on_slave(bond, slave_dev);
1595
1596 write_lock_bh(&bond->lock);
1597
1598 bond_attach_slave(bond, new_slave);
1599
1600 new_slave->delay = 0;
1601 new_slave->link_failure_count = 0;
1602
8531c5ff
AK		 1603 bond_compute_features(bond);
1604
3915c1e8
JV		 1605 write_unlock_bh(&bond->lock);
1606
1607 read_lock(&bond->lock);
1608
f5b2b966
JV		 1609 new_slave->last_arp_rx = jiffies;
1610
1da177e4
LT		 1611 if (bond->params.miimon && !bond->params.use_carrier) {
1612 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1613
1614 if ((link_reporting == -1) && !bond->params.arp_interval) {
1615 /*
1616 * miimon is set but a bonded network driver
1617 * does not support ETHTOOL/MII and
1618 * arp_interval is not set. Note: if
1619 * use_carrier is enabled, we will never go
1620 * here (because netif_carrier is always
1621 * supported); thus, we don't need to change
1622 * the messages for netif_carrier.
1623 */
1624 printk(KERN_WARNING DRV_NAME
4e0952c7 1625 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1626 "available for interface %s, and "
1627 "arp_interval/arp_ip_target module parameters "
1628 "not specified, thus bonding will not detect "
1629 "link failures! see bonding.txt for details.\n",
4e0952c7 1630 bond_dev->name, slave_dev->name);
1da177e4
LT		 1631 } else if (link_reporting == -1) {
1632 /* unable get link status using mii/ethtool */
1633 printk(KERN_WARNING DRV_NAME
4e0952c7 1634 ": %s: Warning: can't get link status from "
1da177e4
LT		 1635 "interface %s; the network driver associated "
1636 "with this interface does not support MII or "
1637 "ETHTOOL link status reporting, thus miimon "
1638 "has no effect on this interface.\n",
4e0952c7 1639 bond_dev->name, slave_dev->name);
1da177e4
LT		 1640 }
1641 }
1642
1643 /* check for initial state */
1644 if (!bond->params.miimon ||
1645 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1646 if (bond->params.updelay) {
5a03cdb7 1647 pr_debug("Initial state of slave_dev is "
1da177e4
LT		 1648 "BOND_LINK_BACK\n");
1649 new_slave->link = BOND_LINK_BACK;
1650 new_slave->delay = bond->params.updelay;
1651 } else {
5a03cdb7 1652 pr_debug("Initial state of slave_dev is "
1da177e4
LT		 1653 "BOND_LINK_UP\n");
1654 new_slave->link = BOND_LINK_UP;
1655 }
1656 new_slave->jiffies = jiffies;
1657 } else {
5a03cdb7 1658 pr_debug("Initial state of slave_dev is "
1da177e4
LT		 1659 "BOND_LINK_DOWN\n");
1660 new_slave->link = BOND_LINK_DOWN;
1661 }
1662
1663 if (bond_update_speed_duplex(new_slave) &&
1664 (new_slave->link != BOND_LINK_DOWN)) {
1665 printk(KERN_WARNING DRV_NAME
4e0952c7 1666 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1667 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1668 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1669
1670 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1671 printk(KERN_WARNING DRV_NAME
1672 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1673 "support in base driver for proper aggregator "
4e0952c7 1674 "selection.\n", bond_dev->name);
1da177e4
LT		 1675 }
1676 }
1677
1678 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1679 /* if there is a primary slave, remember it */
1680 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1681 bond->primary_slave = new_slave;
1682 }
1683 }
1684
3915c1e8
JV		 1685 write_lock_bh(&bond->curr_slave_lock);
1686
1da177e4
LT		 1687 switch (bond->params.mode) {
1688 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1689 bond_set_slave_inactive_flags(new_slave);
1690 bond_select_active_slave(bond);
1da177e4
LT		 1691 break;
1692 case BOND_MODE_8023AD:
1693 /* in 802.3ad mode, the internal mechanism
1694 * will activate the slaves in the selected
1695 * aggregator
1696 */
1697 bond_set_slave_inactive_flags(new_slave);
1698 /* if this is the first slave */
1699 if (bond->slave_cnt == 1) {
1700 SLAVE_AD_INFO(new_slave).id = 1;
1701 /* Initialize AD with the number of times that the AD timer is called in 1 second
1702 * can be called only after the mac address of the bond is set
1703 */
1704 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1705 bond->params.lacp_fast);
1706 } else {
1707 SLAVE_AD_INFO(new_slave).id =
1708 SLAVE_AD_INFO(new_slave->prev).id + 1;
1709 }
1710
1711 bond_3ad_bind_slave(new_slave);
1712 break;
1713 case BOND_MODE_TLB:
1714 case BOND_MODE_ALB:
1715 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1716 bond_set_slave_inactive_flags(new_slave);
5a29f789 1717 bond_select_active_slave(bond);
1da177e4
LT		 1718 break;
1719 default:
5a03cdb7 1720 pr_debug("This slave is always active in trunk mode\n");
1da177e4
LT		 1721
1722 /* always active in trunk mode */
1723 new_slave->state = BOND_STATE_ACTIVE;
1724
1725 /* In trunking mode there is little meaning to curr_active_slave
1726 * anyway (it holds no special properties of the bond device),
1727 * so we can change it without calling change_active_interface()
1728 */
1729 if (!bond->curr_active_slave) {
1730 bond->curr_active_slave = new_slave;
1731 }
1732 break;
1733 } /* switch(bond_mode) */
1734
3915c1e8
JV		 1735 write_unlock_bh(&bond->curr_slave_lock);
1736
ff59c456
JV		 1737 bond_set_carrier(bond);
1738
3915c1e8 1739 read_unlock(&bond->lock);
1da177e4 1740
b76cdba9
MW		 1741 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1742 if (res)
569f0c4d 1743 goto err_close;
b76cdba9 1744
1da177e4
LT		 1745 printk(KERN_INFO DRV_NAME
1746 ": %s: enslaving %s as a%s interface with a%s link.\n",
1747 bond_dev->name, slave_dev->name,
1748 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1749 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1750
1751 /* enslave is successful */
1752 return 0;
1753
1754/* Undo stages on error */
1da177e4
LT		 1755err_close:
1756 dev_close(slave_dev);
1757
569f0c4d
JV		 1758err_unset_master:
1759 netdev_set_master(slave_dev, NULL);
1760
1da177e4 1761err_restore_mac:
dd957c57 1762 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1763 /* XXX TODO - fom follow mode needs to change master's
1764 * MAC if this slave's MAC is in use by the bond, or at
1765 * least print a warning.
1766 */
2ab82852
MS		 1767 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1768 addr.sa_family = slave_dev->type;
1769 dev_set_mac_address(slave_dev, &addr);
1770 }
1da177e4
LT		 1771
1772err_free:
1773 kfree(new_slave);
1774
1775err_undo_flags:
1776 bond_dev->features = old_features;
8531c5ff 1777
1da177e4
LT		 1778 return res;
1779}
1780
1781/*
1782 * Try to release the slave device <slave> from the bond device <master>
1783 * It is legal to access curr_active_slave without a lock because all the function
1784 * is write-locked.
1785 *
1786 * The rules for slave state should be:
1787 * for Active/Backup:
1788 * Active stays on all backups go down
1789 * for Bonded connections:
1790 * The first up interface should be left on and all others downed.
1791 */
a77b5325 1792int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1793{
454d7c9b 1794 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1795 struct slave *slave, *oldcurrent;
1796 struct sockaddr addr;
1797 int mac_addr_differ;
1798
1799 /* slave is not a slave or master is not master of this slave */
1800 if (!(slave_dev->flags & IFF_SLAVE) ||
1801 (slave_dev->master != bond_dev)) {
1802 printk(KERN_ERR DRV_NAME
4e0952c7 1803 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1804 bond_dev->name, slave_dev->name);
1805 return -EINVAL;
1806 }
1807
1808 write_lock_bh(&bond->lock);
1809
1810 slave = bond_get_slave_by_dev(bond, slave_dev);
1811 if (!slave) {
1812 /* not a slave of this bond */
1813 printk(KERN_INFO DRV_NAME
1814 ": %s: %s not enslaved\n",
1815 bond_dev->name, slave_dev->name);
f5e2a7b2 1816 write_unlock_bh(&bond->lock);
1da177e4
LT		 1817 return -EINVAL;
1818 }
1819
3915c1e8
JV		 1820 if (!bond->params.fail_over_mac) {
1821 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1822 ETH_ALEN);
1823 if (!mac_addr_differ && (bond->slave_cnt > 1))
1824 printk(KERN_WARNING DRV_NAME
1825 ": %s: Warning: the permanent HWaddr of %s - "
e174961c 1826 "%pM - is still in use by %s. "
3915c1e8
JV		 1827 "Set the HWaddr of %s to a different address "
1828 "to avoid conflicts.\n",
1829 bond_dev->name, slave_dev->name,
e174961c 1830 slave->perm_hwaddr,
3915c1e8 1831 bond_dev->name, slave_dev->name);
1da177e4
LT		 1832 }
1833
1834 /* Inform AD package of unbinding of slave. */
1835 if (bond->params.mode == BOND_MODE_8023AD) {
1836 /* must be called before the slave is
1837 * detached from the list
1838 */
1839 bond_3ad_unbind_slave(slave);
1840 }
1841
1842 printk(KERN_INFO DRV_NAME
1843 ": %s: releasing %s interface %s\n",
1844 bond_dev->name,
1845 (slave->state == BOND_STATE_ACTIVE)
1846 ? "active" : "backup",
1847 slave_dev->name);
1848
1849 oldcurrent = bond->curr_active_slave;
1850
1851 bond->current_arp_slave = NULL;
1852
1853 /* release the slave from its bond */
1854 bond_detach_slave(bond, slave);
1855
8531c5ff
AK		 1856 bond_compute_features(bond);
1857
1da177e4
LT		 1858 if (bond->primary_slave == slave) {
1859 bond->primary_slave = NULL;
1860 }
1861
1862 if (oldcurrent == slave) {
1863 bond_change_active_slave(bond, NULL);
1864 }
1865
58402054 1866 if (bond_is_lb(bond)) {
1da177e4
LT		 1867 /* Must be called only after the slave has been
1868 * detached from the list and the curr_active_slave
1869 * has been cleared (if our_slave == old_current),
1870 * but before a new active slave is selected.
1871 */
2543331d 1872 write_unlock_bh(&bond->lock);
1da177e4 1873 bond_alb_deinit_slave(bond, slave);
2543331d 1874 write_lock_bh(&bond->lock);
1da177e4
LT		 1875 }
1876
059fe7a5
JV		 1877 if (oldcurrent == slave) {
1878 /*
1879 * Note that we hold RTNL over this sequence, so there
1880 * is no concern that another slave add/remove event
1881 * will interfere.
1882 */
1883 write_unlock_bh(&bond->lock);
1884 read_lock(&bond->lock);
1885 write_lock_bh(&bond->curr_slave_lock);
1886
1da177e4
LT		 1887 bond_select_active_slave(bond);
1888
059fe7a5
JV		 1889 write_unlock_bh(&bond->curr_slave_lock);
1890 read_unlock(&bond->lock);
1891 write_lock_bh(&bond->lock);
1892 }
1893
1da177e4 1894 if (bond->slave_cnt == 0) {
ff59c456
JV		 1895 bond_set_carrier(bond);
1896
1da177e4
LT		 1897 /* if the last slave was removed, zero the mac address
1898 * of the master so it will be set by the application
1899 * to the mac address of the first slave
1900 */
1901 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1902
1903 if (list_empty(&bond->vlan_list)) {
1904 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1905 } else {
1906 printk(KERN_WARNING DRV_NAME
4e0952c7 1907 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1908 "still has VLANs.\n",
4e0952c7 1909 bond_dev->name, bond_dev->name);
1da177e4 1910 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1911 ": %s: When re-adding slaves, make sure the bond's "
1912 "HW address matches its VLANs'.\n",
1913 bond_dev->name);
1da177e4
LT		 1914 }
1915 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1916 !bond_has_challenged_slaves(bond)) {
1917 printk(KERN_INFO DRV_NAME
4e0952c7 1918 ": %s: last VLAN challenged slave %s "
1da177e4 1919 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1920 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1921 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1922 }
1923
1924 write_unlock_bh(&bond->lock);
1925
b76cdba9
MW		 1926 /* must do this from outside any spinlocks */
1927 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1928
1da177e4
LT		 1929 bond_del_vlans_from_slave(bond, slave_dev);
1930
1931 /* If the mode USES_PRIMARY, then we should only remove its
1932 * promisc and mc settings if it was the curr_active_slave, but that was
1933 * already taken care of above when we detached the slave
1934 */
1935 if (!USES_PRIMARY(bond->params.mode)) {
1936 /* unset promiscuity level from slave */
1937 if (bond_dev->flags & IFF_PROMISC) {
1938 dev_set_promiscuity(slave_dev, -1);
1939 }
1940
1941 /* unset allmulti level from slave */
1942 if (bond_dev->flags & IFF_ALLMULTI) {
1943 dev_set_allmulti(slave_dev, -1);
1944 }
1945
1946 /* flush master's mc_list from slave */
b9e40857 1947 netif_addr_lock_bh(bond_dev);
1da177e4 1948 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 1949 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1950 }
1951
1952 netdev_set_master(slave_dev, NULL);
1953
1954 /* close slave before restoring its mac address */
1955 dev_close(slave_dev);
1956
3915c1e8 1957 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1958 /* restore original ("permanent") mac address */
1959 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1960 addr.sa_family = slave_dev->type;
1961 dev_set_mac_address(slave_dev, &addr);
1962 }
1da177e4 1963
8f903c70 1964 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1965 IFF_SLAVE_INACTIVE | IFF_BONDING |
1966 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1967
1968 kfree(slave);
1969
1970 return 0; /* deletion OK */
1971}
1972
d90a162a
MS		 1973/*
1974* Destroy a bonding device.
1975* Must be under rtnl_lock when this function is called.
1976*/
1977void bond_destroy(struct bonding *bond)
1978{
1979 bond_deinit(bond->dev);
1980 bond_destroy_sysfs_entry(bond);
8cbdeec6 1981 unregister_netdevice(bond->dev);
d90a162a
MS		 1982}
1983
a434e43f
JV		 1984static void bond_destructor(struct net_device *bond_dev)
1985{
244ef9b9 1986 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 1987
1988 if (bond->wq)
1989 destroy_workqueue(bond->wq);
1990
1991 netif_addr_lock_bh(bond_dev);
1992 bond_mc_list_destroy(bond);
1993 netif_addr_unlock_bh(bond_dev);
1994
1995 free_netdev(bond_dev);
1996}
1997
d90a162a
MS		 1998/*
1999* First release a slave and than destroy the bond if no more slaves iare left.
2000* Must be under rtnl_lock when this function is called.
2001*/
2002int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2003{
454d7c9b 2004 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 2005 int ret;
2006
2007 ret = bond_release(bond_dev, slave_dev);
2008 if ((ret == 0) && (bond->slave_cnt == 0)) {
2009 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2010 bond_dev->name, bond_dev->name);
2011 bond_destroy(bond);
2012 }
2013 return ret;
2014}
2015
1da177e4
LT		 2016/*
2017 * This function releases all slaves.
2018 */
2019static int bond_release_all(struct net_device *bond_dev)
2020{
454d7c9b 2021 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2022 struct slave *slave;
2023 struct net_device *slave_dev;
2024 struct sockaddr addr;
2025
2026 write_lock_bh(&bond->lock);
2027
ff59c456
JV		 2028 netif_carrier_off(bond_dev);
2029
1da177e4
LT		 2030 if (bond->slave_cnt == 0) {
2031 goto out;
2032 }
2033
2034 bond->current_arp_slave = NULL;
2035 bond->primary_slave = NULL;
2036 bond_change_active_slave(bond, NULL);
2037
2038 while ((slave = bond->first_slave) != NULL) {
2039 /* Inform AD package of unbinding of slave
2040 * before slave is detached from the list.
2041 */
2042 if (bond->params.mode == BOND_MODE_8023AD) {
2043 bond_3ad_unbind_slave(slave);
2044 }
2045
2046 slave_dev = slave->dev;
2047 bond_detach_slave(bond, slave);
2048
2543331d
JV		 2049 /* now that the slave is detached, unlock and perform
2050 * all the undo steps that should not be called from
2051 * within a lock.
2052 */
2053 write_unlock_bh(&bond->lock);
2054
58402054 2055 if (bond_is_lb(bond)) {
1da177e4
LT		 2056 /* must be called only after the slave
2057 * has been detached from the list
2058 */
2059 bond_alb_deinit_slave(bond, slave);
2060 }
2061
8531c5ff
AK		 2062 bond_compute_features(bond);
2063
b76cdba9 2064 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2065 bond_del_vlans_from_slave(bond, slave_dev);
2066
2067 /* If the mode USES_PRIMARY, then we should only remove its
2068 * promisc and mc settings if it was the curr_active_slave, but that was
2069 * already taken care of above when we detached the slave
2070 */
2071 if (!USES_PRIMARY(bond->params.mode)) {
2072 /* unset promiscuity level from slave */
2073 if (bond_dev->flags & IFF_PROMISC) {
2074 dev_set_promiscuity(slave_dev, -1);
2075 }
2076
2077 /* unset allmulti level from slave */
2078 if (bond_dev->flags & IFF_ALLMULTI) {
2079 dev_set_allmulti(slave_dev, -1);
2080 }
2081
2082 /* flush master's mc_list from slave */
b9e40857 2083 netif_addr_lock_bh(bond_dev);
1da177e4 2084 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2085 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2086 }
2087
2088 netdev_set_master(slave_dev, NULL);
2089
2090 /* close slave before restoring its mac address */
2091 dev_close(slave_dev);
2092
dd957c57 2093 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2094 /* restore original ("permanent") mac address*/
2095 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2096 addr.sa_family = slave_dev->type;
2097 dev_set_mac_address(slave_dev, &addr);
2098 }
1da177e4 2099
8f903c70
JV		 2100 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2101 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2102
2103 kfree(slave);
2104
2105 /* re-acquire the lock before getting the next slave */
2106 write_lock_bh(&bond->lock);
2107 }
2108
2109 /* zero the mac address of the master so it will be
2110 * set by the application to the mac address of the
2111 * first slave
2112 */
2113 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2114
2115 if (list_empty(&bond->vlan_list)) {
2116 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2117 } else {
2118 printk(KERN_WARNING DRV_NAME
4e0952c7 2119 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2120 "still has VLANs.\n",
4e0952c7 2121 bond_dev->name, bond_dev->name);
1da177e4 2122 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2123 ": %s: When re-adding slaves, make sure the bond's "
2124 "HW address matches its VLANs'.\n",
2125 bond_dev->name);
1da177e4
LT		 2126 }
2127
2128 printk(KERN_INFO DRV_NAME
2129 ": %s: released all slaves\n",
2130 bond_dev->name);
2131
2132out:
2133 write_unlock_bh(&bond->lock);
2134
2135 return 0;
2136}
2137
2138/*
2139 * This function changes the active slave to slave <slave_dev>.
2140 * It returns -EINVAL in the following cases.
2141 * - <slave_dev> is not found in the list.
2142 * - There is not active slave now.
2143 * - <slave_dev> is already active.
2144 * - The link state of <slave_dev> is not BOND_LINK_UP.
2145 * - <slave_dev> is not running.
2146 * In these cases, this fuction does nothing.
2147 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2148 */
2149static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2150{
454d7c9b 2151 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2152 struct slave *old_active = NULL;
2153 struct slave *new_active = NULL;
2154 int res = 0;
2155
2156 if (!USES_PRIMARY(bond->params.mode)) {
2157 return -EINVAL;
2158 }
2159
2160 /* Verify that master_dev is indeed the master of slave_dev */
2161 if (!(slave_dev->flags & IFF_SLAVE) ||
2162 (slave_dev->master != bond_dev)) {
2163 return -EINVAL;
2164 }
2165
059fe7a5 2166 read_lock(&bond->lock);
1da177e4 2167
059fe7a5 2168 read_lock(&bond->curr_slave_lock);
1da177e4 2169 old_active = bond->curr_active_slave;
059fe7a5
JV		 2170 read_unlock(&bond->curr_slave_lock);
2171
1da177e4
LT		 2172 new_active = bond_get_slave_by_dev(bond, slave_dev);
2173
2174 /*
2175 * Changing to the current active: do nothing; return success.
2176 */
2177 if (new_active && (new_active == old_active)) {
059fe7a5 2178 read_unlock(&bond->lock);
1da177e4
LT		 2179 return 0;
2180 }
2181
2182 if ((new_active) &&
2183 (old_active) &&
2184 (new_active->link == BOND_LINK_UP) &&
2185 IS_UP(new_active->dev)) {
059fe7a5 2186 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2187 bond_change_active_slave(bond, new_active);
059fe7a5 2188 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2189 } else {
2190 res = -EINVAL;
2191 }
2192
059fe7a5 2193 read_unlock(&bond->lock);
1da177e4
LT		 2194
2195 return res;
2196}
2197
1da177e4
LT		 2198static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2199{
454d7c9b 2200 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2201
2202 info->bond_mode = bond->params.mode;
2203 info->miimon = bond->params.miimon;
2204
6603a6f2 2205 read_lock(&bond->lock);
1da177e4 2206 info->num_slaves = bond->slave_cnt;
6603a6f2 2207 read_unlock(&bond->lock);
1da177e4
LT		 2208
2209 return 0;
2210}
2211
2212static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2213{
454d7c9b 2214 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2215 struct slave *slave;
2216 int i, found = 0;
2217
2218 if (info->slave_id < 0) {
2219 return -ENODEV;
2220 }
2221
6603a6f2 2222 read_lock(&bond->lock);
1da177e4
LT		 2223
2224 bond_for_each_slave(bond, slave, i) {
2225 if (i == (int)info->slave_id) {
2226 found = 1;
2227 break;
2228 }
2229 }
2230
6603a6f2 2231 read_unlock(&bond->lock);
1da177e4
LT		 2232
2233 if (found) {
2234 strcpy(info->slave_name, slave->dev->name);
2235 info->link = slave->link;
2236 info->state = slave->state;
2237 info->link_failure_count = slave->link_failure_count;
2238 } else {
2239 return -ENODEV;
2240 }
2241
2242 return 0;
2243}
2244
2245/*-------------------------------- Monitoring -------------------------------*/
2246
1da177e4 2247
f0c76d61
JV		 2248static int bond_miimon_inspect(struct bonding *bond)
2249{
2250 struct slave *slave;
2251 int i, link_state, commit = 0;
1da177e4
LT		 2252
2253 bond_for_each_slave(bond, slave, i) {
f0c76d61 2254 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2255
f0c76d61 2256 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2257
2258 switch (slave->link) {
f0c76d61
JV		 2259 case BOND_LINK_UP:
2260 if (link_state)
2261 continue;
1da177e4 2262
f0c76d61
JV		 2263 slave->link = BOND_LINK_FAIL;
2264 slave->delay = bond->params.downdelay;
2265 if (slave->delay) {
2266 printk(KERN_INFO DRV_NAME
2267 ": %s: link status down for %s"
2268 "interface %s, disabling it in %d ms.\n",
2269 bond->dev->name,
2270 (bond->params.mode ==
2271 BOND_MODE_ACTIVEBACKUP) ?
2272 ((slave->state == BOND_STATE_ACTIVE) ?
2273 "active " : "backup ") : "",
2274 slave->dev->name,
2275 bond->params.downdelay * bond->params.miimon);
1da177e4 2276 }
f0c76d61
JV		 2277 /*FALLTHRU*/
2278 case BOND_LINK_FAIL:
2279 if (link_state) {
2280 /*
2281 * recovered before downdelay expired
2282 */
2283 slave->link = BOND_LINK_UP;
1da177e4
LT		 2284 slave->jiffies = jiffies;
2285 printk(KERN_INFO DRV_NAME
2286 ": %s: link status up again after %d "
2287 "ms for interface %s.\n",
1b76b316 2288 bond->dev->name,
f0c76d61
JV		 2289 (bond->params.downdelay - slave->delay) *
2290 bond->params.miimon,
2291 slave->dev->name);
2292 continue;
1da177e4 2293 }
f0c76d61
JV		 2294
2295 if (slave->delay <= 0) {
2296 slave->new_link = BOND_LINK_DOWN;
2297 commit++;
2298 continue;
1da177e4 2299 }
1da177e4 2300
f0c76d61
JV		 2301 slave->delay--;
2302 break;
2303
2304 case BOND_LINK_DOWN:
2305 if (!link_state)
2306 continue;
2307
2308 slave->link = BOND_LINK_BACK;
2309 slave->delay = bond->params.updelay;
2310
2311 if (slave->delay) {
2312 printk(KERN_INFO DRV_NAME
2313 ": %s: link status up for "
2314 "interface %s, enabling it in %d ms.\n",
2315 bond->dev->name, slave->dev->name,
2316 bond->params.updelay *
2317 bond->params.miimon);
2318 }
2319 /*FALLTHRU*/
2320 case BOND_LINK_BACK:
2321 if (!link_state) {
2322 slave->link = BOND_LINK_DOWN;
1da177e4
LT		 2323 printk(KERN_INFO DRV_NAME
2324 ": %s: link status down again after %d "
2325 "ms for interface %s.\n",
1b76b316 2326 bond->dev->name,
f0c76d61
JV		 2327 (bond->params.updelay - slave->delay) *
2328 bond->params.miimon,
2329 slave->dev->name);
2330
2331 continue;
2332 }
2333
2334 if (slave->delay <= 0) {
2335 slave->new_link = BOND_LINK_UP;
2336 commit++;
2337 continue;
1da177e4 2338 }
f0c76d61
JV		 2339
2340 slave->delay--;
1da177e4 2341 break;
f0c76d61
JV		 2342 }
2343 }
1da177e4 2344
f0c76d61
JV		 2345 return commit;
2346}
1da177e4 2347
f0c76d61
JV		 2348static void bond_miimon_commit(struct bonding *bond)
2349{
2350 struct slave *slave;
2351 int i;
2352
2353 bond_for_each_slave(bond, slave, i) {
2354 switch (slave->new_link) {
2355 case BOND_LINK_NOCHANGE:
2356 continue;
1da177e4 2357
f0c76d61
JV		 2358 case BOND_LINK_UP:
2359 slave->link = BOND_LINK_UP;
2360 slave->jiffies = jiffies;
2361
2362 if (bond->params.mode == BOND_MODE_8023AD) {
2363 /* prevent it from being the active one */
2364 slave->state = BOND_STATE_BACKUP;
2365 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2366 /* make it immediately active */
2367 slave->state = BOND_STATE_ACTIVE;
2368 } else if (slave != bond->primary_slave) {
2369 /* prevent it from being the active one */
2370 slave->state = BOND_STATE_BACKUP;
1da177e4 2371 }
1da177e4 2372
f0c76d61
JV		 2373 printk(KERN_INFO DRV_NAME
2374 ": %s: link status definitely "
2375 "up for interface %s.\n",
2376 bond->dev->name, slave->dev->name);
1da177e4 2377
f0c76d61
JV		 2378 /* notify ad that the link status has changed */
2379 if (bond->params.mode == BOND_MODE_8023AD)
2380 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2381
58402054 2382 if (bond_is_lb(bond))
f0c76d61
JV		 2383 bond_alb_handle_link_change(bond, slave,
2384 BOND_LINK_UP);
1da177e4 2385
f0c76d61
JV		 2386 if (!bond->curr_active_slave ||
2387 (slave == bond->primary_slave))
2388 goto do_failover;
1da177e4 2389
f0c76d61 2390 continue;
059fe7a5 2391
f0c76d61 2392 case BOND_LINK_DOWN:
fba4acda
JV		 2393 if (slave->link_failure_count < UINT_MAX)
2394 slave->link_failure_count++;
2395
f0c76d61 2396 slave->link = BOND_LINK_DOWN;
1da177e4 2397
f0c76d61
JV		 2398 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2399 bond->params.mode == BOND_MODE_8023AD)
2400 bond_set_slave_inactive_flags(slave);
2401
2402 printk(KERN_INFO DRV_NAME
2403 ": %s: link status definitely down for "
2404 "interface %s, disabling it\n",
2405 bond->dev->name, slave->dev->name);
2406
2407 if (bond->params.mode == BOND_MODE_8023AD)
2408 bond_3ad_handle_link_change(slave,
2409 BOND_LINK_DOWN);
2410
2411 if (bond->params.mode == BOND_MODE_TLB ||
2412 bond->params.mode == BOND_MODE_ALB)
2413 bond_alb_handle_link_change(bond, slave,
2414 BOND_LINK_DOWN);
2415
2416 if (slave == bond->curr_active_slave)
2417 goto do_failover;
2418
2419 continue;
2420
2421 default:
2422 printk(KERN_ERR DRV_NAME
2423 ": %s: invalid new link %d on slave %s\n",
2424 bond->dev->name, slave->new_link,
2425 slave->dev->name);
2426 slave->new_link = BOND_LINK_NOCHANGE;
2427
2428 continue;
2429 }
2430
2431do_failover:
2432 ASSERT_RTNL();
2433 write_lock_bh(&bond->curr_slave_lock);
2434 bond_select_active_slave(bond);
2435 write_unlock_bh(&bond->curr_slave_lock);
2436 }
2437
2438 bond_set_carrier(bond);
1da177e4
LT		 2439}
2440
0b0eef66
JV		 2441/*
2442 * bond_mii_monitor
2443 *
2444 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2445 * inspection, then (if inspection indicates something needs to be done)
2446 * an acquisition of appropriate locks followed by a commit phase to
2447 * implement whatever link state changes are indicated.
0b0eef66
JV		 2448 */
2449void bond_mii_monitor(struct work_struct *work)
2450{
2451 struct bonding *bond = container_of(work, struct bonding,
2452 mii_work.work);
0b0eef66
JV		 2453
2454 read_lock(&bond->lock);
f0c76d61
JV		 2455 if (bond->kill_timers)
2456 goto out;
2457
2458 if (bond->slave_cnt == 0)
2459 goto re_arm;
b59f9f74
JV		 2460
2461 if (bond->send_grat_arp) {
2462 read_lock(&bond->curr_slave_lock);
2463 bond_send_gratuitous_arp(bond);
2464 read_unlock(&bond->curr_slave_lock);
2465 }
2466
305d552a
BH		 2467 if (bond->send_unsol_na) {
2468 read_lock(&bond->curr_slave_lock);
2469 bond_send_unsolicited_na(bond);
2470 read_unlock(&bond->curr_slave_lock);
2471 }
2472
f0c76d61 2473 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2474 read_unlock(&bond->lock);
2475 rtnl_lock();
2476 read_lock(&bond->lock);
f0c76d61
JV		 2477
2478 bond_miimon_commit(bond);
2479
5655662d
JV		 2480 read_unlock(&bond->lock);
2481 rtnl_unlock(); /* might sleep, hold no other locks */
2482 read_lock(&bond->lock);
0b0eef66
JV		 2483 }
2484
f0c76d61
JV		 2485re_arm:
2486 if (bond->params.miimon)
2487 queue_delayed_work(bond->wq, &bond->mii_work,
2488 msecs_to_jiffies(bond->params.miimon));
2489out:
0b0eef66 2490 read_unlock(&bond->lock);
0b0eef66 2491}
c3ade5ca 2492
d3bb52b0 2493static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2494{
2495 struct in_device *idev;
2496 struct in_ifaddr *ifa;
a144ea4b 2497 __be32 addr = 0;
c3ade5ca
JV		 2498
2499 if (!dev)
2500 return 0;
2501
2502 rcu_read_lock();
e5ed6399 2503 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2504 if (!idev)
2505 goto out;
2506
2507 ifa = idev->ifa_list;
2508 if (!ifa)
2509 goto out;
2510
2511 addr = ifa->ifa_local;
2512out:
2513 rcu_read_unlock();
2514 return addr;
2515}
2516
d3bb52b0 2517static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2518{
0883beca 2519 struct vlan_entry *vlan;
f5b2b966
JV		 2520
2521 if (ip == bond->master_ip)
2522 return 1;
2523
0883beca 2524 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2525 if (ip == vlan->vlan_ip)
2526 return 1;
2527 }
2528
2529 return 0;
2530}
2531
c3ade5ca
JV		 2532/*
2533 * We go to the (large) trouble of VLAN tagging ARP frames because
2534 * switches in VLAN mode (especially if ports are configured as
2535 * "native" to a VLAN) might not pass non-tagged frames.
2536 */
d3bb52b0 2537static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2538{
2539 struct sk_buff *skb;
2540
5a03cdb7 2541 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
c3ade5ca
JV		 2542 slave_dev->name, dest_ip, src_ip, vlan_id);
2543
2544 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2545 NULL, slave_dev->dev_addr, NULL);
2546
2547 if (!skb) {
2548 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2549 return;
2550 }
2551 if (vlan_id) {
2552 skb = vlan_put_tag(skb, vlan_id);
2553 if (!skb) {
2554 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2555 return;
2556 }
2557 }
2558 arp_xmit(skb);
2559}
2560
2561
1da177e4
LT		 2562static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2563{
c3ade5ca 2564 int i, vlan_id, rv;
d3bb52b0 2565 __be32 *targets = bond->params.arp_targets;
0883beca 2566 struct vlan_entry *vlan;
c3ade5ca
JV		 2567 struct net_device *vlan_dev;
2568 struct flowi fl;
2569 struct rtable *rt;
1da177e4 2570
6b780567
MW		 2571 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2572 if (!targets[i])
2573 continue;
5a03cdb7 2574 pr_debug("basa: target %x\n", targets[i]);
c3ade5ca 2575 if (list_empty(&bond->vlan_list)) {
5a03cdb7 2576 pr_debug("basa: empty vlan: arp_send\n");
c3ade5ca
JV		 2577 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2578 bond->master_ip, 0);
2579 continue;
2580 }
2581
2582 /*
2583 * If VLANs are configured, we do a route lookup to
2584 * determine which VLAN interface would be used, so we
2585 * can tag the ARP with the proper VLAN tag.
2586 */
2587 memset(&fl, 0, sizeof(fl));
2588 fl.fl4_dst = targets[i];
2589 fl.fl4_tos = RTO_ONLINK;
2590
f206351a 2591 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2592 if (rv) {
2593 if (net_ratelimit()) {
2594 printk(KERN_WARNING DRV_NAME
63779436
HH		 2595 ": %s: no route to arp_ip_target %pI4\n",
2596 bond->dev->name, &fl.fl4_dst);
c3ade5ca
JV		 2597 }
2598 continue;
2599 }
2600
2601 /*
2602 * This target is not on a VLAN
2603 */
2604 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2605 ip_rt_put(rt);
5a03cdb7 2606 pr_debug("basa: rtdev == bond->dev: arp_send\n");
c3ade5ca
JV		 2607 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2608 bond->master_ip, 0);
2609 continue;
2610 }
2611
2612 vlan_id = 0;
0883beca 2613 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2614 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2615 if (vlan_dev == rt->u.dst.dev) {
2616 vlan_id = vlan->vlan_id;
5a03cdb7 2617 pr_debug("basa: vlan match on %s %d\n",
c3ade5ca
JV		 2618 vlan_dev->name, vlan_id);
2619 break;
2620 }
2621 }
2622
2623 if (vlan_id) {
ed4b9f80 2624 ip_rt_put(rt);
c3ade5ca
JV		 2625 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2626 vlan->vlan_ip, vlan_id);
2627 continue;
2628 }
2629
2630 if (net_ratelimit()) {
2631 printk(KERN_WARNING DRV_NAME
63779436
HH		 2632 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2633 bond->dev->name, &fl.fl4_dst,
c3ade5ca
JV		 2634 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2635 }
ed4b9f80 2636 ip_rt_put(rt);
c3ade5ca
JV		 2637 }
2638}
2639
2640/*
2641 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2642 * for each VLAN above us.
b59f9f74
JV		 2643 *
2644 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2645 */
2646static void bond_send_gratuitous_arp(struct bonding *bond)
2647{
2648 struct slave *slave = bond->curr_active_slave;
2649 struct vlan_entry *vlan;
2650 struct net_device *vlan_dev;
2651
5a03cdb7 2652 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
c3ade5ca 2653 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2654
2655 if (!slave || !bond->send_grat_arp ||
2656 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2657 return;
2658
b59f9f74
JV		 2659 bond->send_grat_arp--;
2660
c3ade5ca
JV		 2661 if (bond->master_ip) {
2662 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2663 bond->master_ip, 0);
c3ade5ca
JV		 2664 }
2665
2666 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2667 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2668 if (vlan->vlan_ip) {
2669 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2670 vlan->vlan_ip, vlan->vlan_id);
2671 }
1da177e4
LT		 2672 }
2673}
2674
d3bb52b0 2675static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2676{
2677 int i;
d3bb52b0 2678 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2679
2680 targets = bond->params.arp_targets;
2681 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
5a03cdb7 2682 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
63779436 2683 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
f5b2b966
JV		 2684 if (sip == targets[i]) {
2685 if (bond_has_this_ip(bond, tip))
2686 slave->last_arp_rx = jiffies;
2687 return;
2688 }
2689 }
2690}
2691
2692static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2693{
2694 struct arphdr *arp;
2695 struct slave *slave;
2696 struct bonding *bond;
2697 unsigned char *arp_ptr;
d3bb52b0 2698 __be32 sip, tip;
f5b2b966 2699
c346dca1 2700 if (dev_net(dev) != &init_net)
e730c155
EB		 2701 goto out;
2702
f5b2b966
JV		 2703 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2704 goto out;
2705
454d7c9b 2706 bond = netdev_priv(dev);
f5b2b966
JV		 2707 read_lock(&bond->lock);
2708
5a03cdb7 2709 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
f5b2b966
JV		 2710 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2711 orig_dev ? orig_dev->name : "NULL");
2712
2713 slave = bond_get_slave_by_dev(bond, orig_dev);
2714 if (!slave || !slave_do_arp_validate(bond, slave))
2715 goto out_unlock;
2716
988b7050 2717 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2718 goto out_unlock;
2719
d0a92be0 2720 arp = arp_hdr(skb);
f5b2b966
JV		 2721 if (arp->ar_hln != dev->addr_len ||
2722 skb->pkt_type == PACKET_OTHERHOST ||
2723 skb->pkt_type == PACKET_LOOPBACK ||
2724 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2725 arp->ar_pro != htons(ETH_P_IP) ||
2726 arp->ar_pln != 4)
2727 goto out_unlock;
2728
2729 arp_ptr = (unsigned char *)(arp + 1);
2730 arp_ptr += dev->addr_len;
2731 memcpy(&sip, arp_ptr, 4);
2732 arp_ptr += 4 + dev->addr_len;
2733 memcpy(&tip, arp_ptr, 4);
2734
5a03cdb7 2735 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
63779436
HH		 2736 bond->dev->name, slave->dev->name, slave->state,
2737 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2738 &sip, &tip);
f5b2b966
JV		 2739
2740 /*
2741 * Backup slaves won't see the ARP reply, but do come through
2742 * here for each ARP probe (so we swap the sip/tip to validate
2743 * the probe). In a "redundant switch, common router" type of
2744 * configuration, the ARP probe will (hopefully) travel from
2745 * the active, through one switch, the router, then the other
2746 * switch before reaching the backup.
2747 */
2748 if (slave->state == BOND_STATE_ACTIVE)
2749 bond_validate_arp(bond, slave, sip, tip);
2750 else
2751 bond_validate_arp(bond, slave, tip, sip);
2752
2753out_unlock:
2754 read_unlock(&bond->lock);
2755out:
2756 dev_kfree_skb(skb);
2757 return NET_RX_SUCCESS;
2758}
2759
1da177e4
LT		 2760/*
2761 * this function is called regularly to monitor each slave's link
2762 * ensuring that traffic is being sent and received when arp monitoring
2763 * is used in load-balancing mode. if the adapter has been dormant, then an
2764 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2765 * arp monitoring in active backup mode.
2766 */
1b76b316 2767void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2768{
1b76b316
JV		 2769 struct bonding *bond = container_of(work, struct bonding,
2770 arp_work.work);
1da177e4
LT		 2771 struct slave *slave, *oldcurrent;
2772 int do_failover = 0;
2773 int delta_in_ticks;
2774 int i;
2775
2776 read_lock(&bond->lock);
2777
5ce0da8f 2778 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2779
2780 if (bond->kill_timers) {
2781 goto out;
2782 }
2783
2784 if (bond->slave_cnt == 0) {
2785 goto re_arm;
2786 }
2787
2788 read_lock(&bond->curr_slave_lock);
2789 oldcurrent = bond->curr_active_slave;
2790 read_unlock(&bond->curr_slave_lock);
2791
2792 /* see if any of the previous devices are up now (i.e. they have
2793 * xmt and rcv traffic). the curr_active_slave does not come into
2794 * the picture unless it is null. also, slave->jiffies is not needed
2795 * here because we send an arp on each slave and give a slave as
2796 * long as it needs to get the tx/rx within the delta.
2797 * TODO: what about up/down delay in arp mode? it wasn't here before
2798 * so it can wait
2799 */
2800 bond_for_each_slave(bond, slave, i) {
2801 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2802 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2803 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2804
2805 slave->link = BOND_LINK_UP;
2806 slave->state = BOND_STATE_ACTIVE;
2807
2808 /* primary_slave has no meaning in round-robin
2809 * mode. the window of a slave being up and
2810 * curr_active_slave being null after enslaving
2811 * is closed.
2812 */
2813 if (!oldcurrent) {
2814 printk(KERN_INFO DRV_NAME
2815 ": %s: link status definitely "
2816 "up for interface %s, ",
1b76b316 2817 bond->dev->name,
1da177e4
LT		 2818 slave->dev->name);
2819 do_failover = 1;
2820 } else {
2821 printk(KERN_INFO DRV_NAME
2822 ": %s: interface %s is now up\n",
1b76b316 2823 bond->dev->name,
1da177e4
LT		 2824 slave->dev->name);
2825 }
2826 }
2827 } else {
2828 /* slave->link == BOND_LINK_UP */
2829
2830 /* not all switches will respond to an arp request
2831 * when the source ip is 0, so don't take the link down
2832 * if we don't know our ip yet
2833 */
b63bb739 2834 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2835 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2836
2837 slave->link = BOND_LINK_DOWN;
2838 slave->state = BOND_STATE_BACKUP;
2839
2840 if (slave->link_failure_count < UINT_MAX) {
2841 slave->link_failure_count++;
2842 }
2843
2844 printk(KERN_INFO DRV_NAME
2845 ": %s: interface %s is now down.\n",
1b76b316 2846 bond->dev->name,
1da177e4
LT		 2847 slave->dev->name);
2848
2849 if (slave == oldcurrent) {
2850 do_failover = 1;
2851 }
2852 }
2853 }
2854
2855 /* note: if switch is in round-robin mode, all links
2856 * must tx arp to ensure all links rx an arp - otherwise
2857 * links may oscillate or not come up at all; if switch is
2858 * in something like xor mode, there is nothing we can
2859 * do - all replies will be rx'ed on same link causing slaves
2860 * to be unstable during low/no traffic periods
2861 */
2862 if (IS_UP(slave->dev)) {
2863 bond_arp_send_all(bond, slave);
2864 }
2865 }
2866
2867 if (do_failover) {
059fe7a5 2868 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2869
2870 bond_select_active_slave(bond);
2871
059fe7a5 2872 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2873 }
2874
2875re_arm:
1b76b316
JV		 2876 if (bond->params.arp_interval)
2877 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2878out:
2879 read_unlock(&bond->lock);
2880}
2881
2882/*
b2220cad
JV		 2883 * Called to inspect slaves for active-backup mode ARP monitor link state
2884 * changes. Sets new_link in slaves to specify what action should take
2885 * place for the slave. Returns 0 if no changes are found, >0 if changes
2886 * to link states must be committed.
2887 *
2888 * Called with bond->lock held for read.
1da177e4 2889 */
b2220cad 2890static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2891{
1da177e4 2892 struct slave *slave;
b2220cad 2893 int i, commit = 0;
1da177e4 2894
b2220cad
JV		 2895 bond_for_each_slave(bond, slave, i) {
2896 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2897
b2220cad
JV		 2898 if (slave->link != BOND_LINK_UP) {
2899 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2900 delta_in_ticks)) {
2901 slave->new_link = BOND_LINK_UP;
2902 commit++;
2903 }
1da177e4 2904
b2220cad
JV		 2905 continue;
2906 }
1da177e4 2907
b2220cad
JV		 2908 /*
2909 * Give slaves 2*delta after being enslaved or made
2910 * active. This avoids bouncing, as the last receive
2911 * times need a full ARP monitor cycle to be updated.
2912 */
2913 if (!time_after_eq(jiffies, slave->jiffies +
2914 2 * delta_in_ticks))
2915 continue;
2916
2917 /*
2918 * Backup slave is down if:
2919 * - No current_arp_slave AND
2920 * - more than 3*delta since last receive AND
2921 * - the bond has an IP address
2922 *
2923 * Note: a non-null current_arp_slave indicates
2924 * the curr_active_slave went down and we are
2925 * searching for a new one; under this condition
2926 * we only take the curr_active_slave down - this
2927 * gives each slave a chance to tx/rx traffic
2928 * before being taken out
2929 */
2930 if (slave->state == BOND_STATE_BACKUP &&
2931 !bond->current_arp_slave &&
2932 time_after(jiffies, slave_last_rx(bond, slave) +
2933 3 * delta_in_ticks)) {
2934 slave->new_link = BOND_LINK_DOWN;
2935 commit++;
2936 }
2937
2938 /*
2939 * Active slave is down if:
2940 * - more than 2*delta since transmitting OR
2941 * - (more than 2*delta since receive AND
2942 * the bond has an IP address)
2943 */
2944 if ((slave->state == BOND_STATE_ACTIVE) &&
2945 (time_after_eq(jiffies, slave->dev->trans_start +
2946 2 * delta_in_ticks) ||
2947 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2948 + 2 * delta_in_ticks)))) {
2949 slave->new_link = BOND_LINK_DOWN;
2950 commit++;
2951 }
1da177e4
LT		 2952 }
2953
b2220cad
JV		 2954 read_lock(&bond->curr_slave_lock);
2955
2956 /*
2957 * Trigger a commit if the primary option setting has changed.
1da177e4 2958 */
b2220cad
JV		 2959 if (bond->primary_slave &&
2960 (bond->primary_slave != bond->curr_active_slave) &&
2961 (bond->primary_slave->link == BOND_LINK_UP))
2962 commit++;
1da177e4 2963
b2220cad 2964 read_unlock(&bond->curr_slave_lock);
1da177e4 2965
b2220cad
JV		 2966 return commit;
2967}
1da177e4 2968
b2220cad
JV		 2969/*
2970 * Called to commit link state changes noted by inspection step of
2971 * active-backup mode ARP monitor.
2972 *
2973 * Called with RTNL and bond->lock for read.
2974 */
2975static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2976{
2977 struct slave *slave;
2978 int i;
1da177e4 2979
b2220cad
JV		 2980 bond_for_each_slave(bond, slave, i) {
2981 switch (slave->new_link) {
2982 case BOND_LINK_NOCHANGE:
2983 continue;
ff59c456 2984
b2220cad
JV		 2985 case BOND_LINK_UP:
2986 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2987
b2220cad
JV		 2988 if (!bond->curr_active_slave &&
2989 time_before_eq(jiffies, slave->dev->trans_start +
2990 delta_in_ticks)) {
2991 slave->link = BOND_LINK_UP;
2992 bond_change_active_slave(bond, slave);
2993 bond->current_arp_slave = NULL;
2994
2995 printk(KERN_INFO DRV_NAME
2996 ": %s: %s is up and now the "
2997 "active interface\n",
2998 bond->dev->name, slave->dev->name);
2999
3000 } else if (bond->curr_active_slave != slave) {
3001 /* this slave has just come up but we
3002 * already have a current slave; this can
3003 * also happen if bond_enslave adds a new
3004 * slave that is up while we are searching
3005 * for a new slave
1da177e4 3006 */
b2220cad
JV		 3007 slave->link = BOND_LINK_UP;
3008 bond_set_slave_inactive_flags(slave);
3009 bond->current_arp_slave = NULL;
3010
3011 printk(KERN_INFO DRV_NAME
3012 ": %s: backup interface %s is now up\n",
3013 bond->dev->name, slave->dev->name);
3014 }
1da177e4 3015
b2220cad 3016 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3017
b2220cad 3018 break;
1da177e4 3019
b2220cad
JV		 3020 case BOND_LINK_DOWN:
3021 if (slave->link_failure_count < UINT_MAX)
3022 slave->link_failure_count++;
3023
3024 slave->link = BOND_LINK_DOWN;
3025
3026 if (slave == bond->curr_active_slave) {
3027 printk(KERN_INFO DRV_NAME
3028 ": %s: link status down for active "
3029 "interface %s, disabling it\n",
3030 bond->dev->name, slave->dev->name);
1da177e4
LT		 3031
3032 bond_set_slave_inactive_flags(slave);
3033
b2220cad
JV		 3034 write_lock_bh(&bond->curr_slave_lock);
3035
3036 bond_select_active_slave(bond);
3037 if (bond->curr_active_slave)
3038 bond->curr_active_slave->jiffies =
3039 jiffies;
3040
3041 write_unlock_bh(&bond->curr_slave_lock);
3042
3043 bond->current_arp_slave = NULL;
3044
3045 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3046 printk(KERN_INFO DRV_NAME
3047 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3048 bond->dev->name, slave->dev->name);
3049
3050 bond_set_slave_inactive_flags(slave);
1da177e4 3051 }
b2220cad
JV		 3052 break;
3053
3054 default:
3055 printk(KERN_ERR DRV_NAME
3056 ": %s: impossible: new_link %d on slave %s\n",
3057 bond->dev->name, slave->new_link,
3058 slave->dev->name);
1da177e4
LT		 3059 }
3060 }
3061
b2220cad
JV		 3062 /*
3063 * No race with changes to primary via sysfs, as we hold rtnl.
3064 */
3065 if (bond->primary_slave &&
3066 (bond->primary_slave != bond->curr_active_slave) &&
3067 (bond->primary_slave->link == BOND_LINK_UP)) {
3068 write_lock_bh(&bond->curr_slave_lock);
3069 bond_change_active_slave(bond, bond->primary_slave);
3070 write_unlock_bh(&bond->curr_slave_lock);
3071 }
1da177e4 3072
b2220cad
JV		 3073 bond_set_carrier(bond);
3074}
1da177e4 3075
b2220cad
JV		 3076/*
3077 * Send ARP probes for active-backup mode ARP monitor.
3078 *
3079 * Called with bond->lock held for read.
3080 */
3081static void bond_ab_arp_probe(struct bonding *bond)
3082{
3083 struct slave *slave;
3084 int i;
1da177e4 3085
b2220cad 3086 read_lock(&bond->curr_slave_lock);
1da177e4 3087
b2220cad
JV		 3088 if (bond->current_arp_slave && bond->curr_active_slave)
3089 printk("PROBE: c_arp %s && cas %s BAD\n",
3090 bond->current_arp_slave->dev->name,
3091 bond->curr_active_slave->dev->name);
1da177e4 3092
b2220cad
JV		 3093 if (bond->curr_active_slave) {
3094 bond_arp_send_all(bond, bond->curr_active_slave);
3095 read_unlock(&bond->curr_slave_lock);
3096 return;
3097 }
1da177e4 3098
b2220cad 3099 read_unlock(&bond->curr_slave_lock);
059fe7a5 3100
b2220cad
JV		 3101 /* if we don't have a curr_active_slave, search for the next available
3102 * backup slave from the current_arp_slave and make it the candidate
3103 * for becoming the curr_active_slave
3104 */
1da177e4 3105
b2220cad
JV		 3106 if (!bond->current_arp_slave) {
3107 bond->current_arp_slave = bond->first_slave;
3108 if (!bond->current_arp_slave)
3109 return;
3110 }
1da177e4 3111
b2220cad 3112 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3113
b2220cad
JV		 3114 /* search for next candidate */
3115 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3116 if (IS_UP(slave->dev)) {
3117 slave->link = BOND_LINK_BACK;
3118 bond_set_slave_active_flags(slave);
3119 bond_arp_send_all(bond, slave);
1da177e4 3120 slave->jiffies = jiffies;
b2220cad
JV		 3121 bond->current_arp_slave = slave;
3122 break;
1da177e4
LT		 3123 }
3124
b2220cad
JV		 3125 /* if the link state is up at this point, we
3126 * mark it down - this can happen if we have
3127 * simultaneous link failures and
3128 * reselect_active_interface doesn't make this
3129 * one the current slave so it is still marked
3130 * up when it is actually down
1da177e4 3131 */
b2220cad
JV		 3132 if (slave->link == BOND_LINK_UP) {
3133 slave->link = BOND_LINK_DOWN;
3134 if (slave->link_failure_count < UINT_MAX)
3135 slave->link_failure_count++;
1da177e4 3136
b2220cad
JV		 3137 bond_set_slave_inactive_flags(slave);
3138
3139 printk(KERN_INFO DRV_NAME
3140 ": %s: backup interface %s is now down.\n",
3141 bond->dev->name, slave->dev->name);
1da177e4 3142 }
b2220cad
JV		 3143 }
3144}
1da177e4 3145
b2220cad
JV		 3146void bond_activebackup_arp_mon(struct work_struct *work)
3147{
3148 struct bonding *bond = container_of(work, struct bonding,
3149 arp_work.work);
3150 int delta_in_ticks;
1da177e4 3151
b2220cad 3152 read_lock(&bond->lock);
1da177e4 3153
b2220cad
JV		 3154 if (bond->kill_timers)
3155 goto out;
1da177e4 3156
b2220cad 3157 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3158
b2220cad
JV		 3159 if (bond->slave_cnt == 0)
3160 goto re_arm;
3161
b59f9f74
JV		 3162 if (bond->send_grat_arp) {
3163 read_lock(&bond->curr_slave_lock);
3164 bond_send_gratuitous_arp(bond);
3165 read_unlock(&bond->curr_slave_lock);
3166 }
3167
305d552a
BH		 3168 if (bond->send_unsol_na) {
3169 read_lock(&bond->curr_slave_lock);
3170 bond_send_unsolicited_na(bond);
3171 read_unlock(&bond->curr_slave_lock);
3172 }
3173
b2220cad
JV		 3174 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3175 read_unlock(&bond->lock);
3176 rtnl_lock();
3177 read_lock(&bond->lock);
3178
3179 bond_ab_arp_commit(bond, delta_in_ticks);
3180
3181 read_unlock(&bond->lock);
3182 rtnl_unlock();
3183 read_lock(&bond->lock);
1da177e4
LT		 3184 }
3185
b2220cad
JV		 3186 bond_ab_arp_probe(bond);
3187
1da177e4
LT		 3188re_arm:
3189 if (bond->params.arp_interval) {
1b76b316 3190 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3191 }
3192out:
3193 read_unlock(&bond->lock);
3194}
3195
3196/*------------------------------ proc/seq_file-------------------------------*/
3197
3198#ifdef CONFIG_PROC_FS
3199
1da177e4 3200static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
1f78d9f9
HE		 3201 __acquires(&dev_base_lock)
3202 __acquires(&bond->lock)
1da177e4
LT		 3203{
3204 struct bonding *bond = seq->private;
3205 loff_t off = 0;
3206 struct slave *slave;
3207 int i;
3208
3209 /* make sure the bond won't be taken away */
3210 read_lock(&dev_base_lock);
6603a6f2 3211 read_lock(&bond->lock);
1da177e4
LT		 3212
3213 if (*pos == 0) {
3214 return SEQ_START_TOKEN;
3215 }
3216
3217 bond_for_each_slave(bond, slave, i) {
3218 if (++off == *pos) {
3219 return slave;
3220 }
3221 }
3222
3223 return NULL;
3224}
3225
3226static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3227{
3228 struct bonding *bond = seq->private;
3229 struct slave *slave = v;
3230
3231 ++*pos;
3232 if (v == SEQ_START_TOKEN) {
3233 return bond->first_slave;
3234 }
3235
3236 slave = slave->next;
3237
3238 return (slave == bond->first_slave) ? NULL : slave;
3239}
3240
3241static void bond_info_seq_stop(struct seq_file *seq, void *v)
1f78d9f9
HE		 3242 __releases(&bond->lock)
3243 __releases(&dev_base_lock)
1da177e4
LT		 3244{
3245 struct bonding *bond = seq->private;
3246
6603a6f2 3247 read_unlock(&bond->lock);
1da177e4
LT		 3248 read_unlock(&dev_base_lock);
3249}
3250
3251static void bond_info_show_master(struct seq_file *seq)
3252{
3253 struct bonding *bond = seq->private;
3254 struct slave *curr;
4756b02f 3255 int i;
1da177e4
LT		 3256
3257 read_lock(&bond->curr_slave_lock);
3258 curr = bond->curr_active_slave;
3259 read_unlock(&bond->curr_slave_lock);
3260
dd957c57 3261 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3262 bond_mode_name(bond->params.mode));
3263
dd957c57
JV		 3264 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3265 bond->params.fail_over_mac)
3915c1e8
JV		 3266 seq_printf(seq, " (fail_over_mac %s)",
3267 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3268
3269 seq_printf(seq, "\n");
3270
c61b75ad
MW		 3271 if (bond->params.mode == BOND_MODE_XOR ||
3272 bond->params.mode == BOND_MODE_8023AD) {
3273 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3274 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3275 bond->params.xmit_policy);
3276 }
3277
1da177e4
LT		 3278 if (USES_PRIMARY(bond->params.mode)) {
3279 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3280 (bond->primary_slave) ?
3281 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3282
3283 seq_printf(seq, "Currently Active Slave: %s\n",
3284 (curr) ? curr->dev->name : "None");
3285 }
3286
ff59c456
JV		 3287 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3288 "up" : "down");
1da177e4
LT		 3289 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3290 seq_printf(seq, "Up Delay (ms): %d\n",
3291 bond->params.updelay * bond->params.miimon);
3292 seq_printf(seq, "Down Delay (ms): %d\n",
3293 bond->params.downdelay * bond->params.miimon);
3294
4756b02f
MW		 3295
3296 /* ARP information */
3297 if(bond->params.arp_interval > 0) {
3298 int printed=0;
3299 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3300 bond->params.arp_interval);
3301
3302 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3303
3304 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3305 if (!bond->params.arp_targets[i])
3306 continue;
3307 if (printed)
3308 seq_printf(seq, ",");
8cf14e38 3309 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
4756b02f
MW		 3310 printed = 1;
3311 }
3312 seq_printf(seq, "\n");
3313 }
3314
1da177e4
LT		 3315 if (bond->params.mode == BOND_MODE_8023AD) {
3316 struct ad_info ad_info;
3317
3318 seq_puts(seq, "\n802.3ad info\n");
3319 seq_printf(seq, "LACP rate: %s\n",
3320 (bond->params.lacp_fast) ? "fast" : "slow");
fd989c83
JV		 3321 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3322 ad_select_tbl[bond->params.ad_select].modename);
1da177e4
LT		 3323
3324 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3325 seq_printf(seq, "bond %s has no active aggregator\n",
3326 bond->dev->name);
3327 } else {
3328 seq_printf(seq, "Active Aggregator Info:\n");
3329
3330 seq_printf(seq, "\tAggregator ID: %d\n",
3331 ad_info.aggregator_id);
3332 seq_printf(seq, "\tNumber of ports: %d\n",
3333 ad_info.ports);
3334 seq_printf(seq, "\tActor Key: %d\n",
3335 ad_info.actor_key);
3336 seq_printf(seq, "\tPartner Key: %d\n",
3337 ad_info.partner_key);
e174961c
JB		 3338 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3339 ad_info.partner_system);
1da177e4
LT		 3340 }
3341 }
3342}
3343
3344static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3345{
3346 struct bonding *bond = seq->private;
3347
3348 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3349 seq_printf(seq, "MII Status: %s\n",
3350 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3351 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3352 slave->link_failure_count);
3353
e174961c 3354 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
1da177e4
LT		 3355
3356 if (bond->params.mode == BOND_MODE_8023AD) {
3357 const struct aggregator *agg
3358 = SLAVE_AD_INFO(slave).port.aggregator;
3359
3360 if (agg) {
3361 seq_printf(seq, "Aggregator ID: %d\n",
3362 agg->aggregator_identifier);
3363 } else {
3364 seq_puts(seq, "Aggregator ID: N/A\n");
3365 }
3366 }
3367}
3368
3369static int bond_info_seq_show(struct seq_file *seq, void *v)
3370{
3371 if (v == SEQ_START_TOKEN) {
3372 seq_printf(seq, "%s\n", version);
3373 bond_info_show_master(seq);
3374 } else {
3375 bond_info_show_slave(seq, v);
3376 }
3377
3378 return 0;
3379}
3380
4101dec9 3381static const struct seq_operations bond_info_seq_ops = {
1da177e4
LT		 3382 .start = bond_info_seq_start,
3383 .next = bond_info_seq_next,
3384 .stop = bond_info_seq_stop,
3385 .show = bond_info_seq_show,
3386};
3387
3388static int bond_info_open(struct inode *inode, struct file *file)
3389{
3390 struct seq_file *seq;
3391 struct proc_dir_entry *proc;
3392 int res;
3393
3394 res = seq_open(file, &bond_info_seq_ops);
3395 if (!res) {
3396 /* recover the pointer buried in proc_dir_entry data */
3397 seq = file->private_data;
3398 proc = PDE(inode);
3399 seq->private = proc->data;
3400 }
3401
3402 return res;
3403}
3404
d54b1fdb 3405static const struct file_operations bond_info_fops = {
1da177e4
LT		 3406 .owner = THIS_MODULE,
3407 .open = bond_info_open,
3408 .read = seq_read,
3409 .llseek = seq_lseek,
3410 .release = seq_release,
3411};
3412
3413static int bond_create_proc_entry(struct bonding *bond)
3414{
3415 struct net_device *bond_dev = bond->dev;
3416
3417 if (bond_proc_dir) {
a95609cb
DL		 3418 bond->proc_entry = proc_create_data(bond_dev->name,
3419 S_IRUGO, bond_proc_dir,
3420 &bond_info_fops, bond);
1da177e4
LT		 3421 if (bond->proc_entry == NULL) {
3422 printk(KERN_WARNING DRV_NAME
3423 ": Warning: Cannot create /proc/net/%s/%s\n",
3424 DRV_NAME, bond_dev->name);
3425 } else {
1da177e4
LT		 3426 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3427 }
3428 }
3429
3430 return 0;
3431}
3432
3433static void bond_remove_proc_entry(struct bonding *bond)
3434{
3435 if (bond_proc_dir && bond->proc_entry) {
3436 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3437 memset(bond->proc_file_name, 0, IFNAMSIZ);
3438 bond->proc_entry = NULL;
3439 }
3440}
3441
3442/* Create the bonding directory under /proc/net, if doesn't exist yet.
3443 * Caller must hold rtnl_lock.
3444 */
3445static void bond_create_proc_dir(void)
3446{
1da177e4 3447 if (!bond_proc_dir) {
457c4cbc 3448 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
99b76233 3449 if (!bond_proc_dir)
1da177e4
LT		 3450 printk(KERN_WARNING DRV_NAME
3451 ": Warning: cannot create /proc/net/%s\n",
3452 DRV_NAME);
1da177e4
LT		 3453 }
3454}
3455
3456/* Destroy the bonding directory under /proc/net, if empty.
3457 * Caller must hold rtnl_lock.
3458 */
3459static void bond_destroy_proc_dir(void)
3460{
99b76233 3461 if (bond_proc_dir) {
457c4cbc 3462 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3463 bond_proc_dir = NULL;
3464 }
3465}
3466#endif /* CONFIG_PROC_FS */
3467
3468/*-------------------------- netdev event handling --------------------------*/
3469
3470/*
3471 * Change device name
3472 */
3473static int bond_event_changename(struct bonding *bond)
3474{
3475#ifdef CONFIG_PROC_FS
3476 bond_remove_proc_entry(bond);
3477 bond_create_proc_entry(bond);
3478#endif
b76cdba9
MW		 3479 down_write(&(bonding_rwsem));
3480 bond_destroy_sysfs_entry(bond);
3481 bond_create_sysfs_entry(bond);
3482 up_write(&(bonding_rwsem));
1da177e4
LT		 3483 return NOTIFY_DONE;
3484}
3485
3486static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3487{
454d7c9b 3488 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 3489
3490 switch (event) {
3491 case NETDEV_CHANGENAME:
3492 return bond_event_changename(event_bond);
3493 case NETDEV_UNREGISTER:
3b96c858 3494 bond_release_all(event_bond->dev);
1da177e4
LT		 3495 break;
3496 default:
3497 break;
3498 }
3499
3500 return NOTIFY_DONE;
3501}
3502
3503static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3504{
3505 struct net_device *bond_dev = slave_dev->master;
454d7c9b 3506 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3507
3508 switch (event) {
3509 case NETDEV_UNREGISTER:
3510 if (bond_dev) {
1284cd3a
JV		 3511 if (bond->setup_by_slave)
3512 bond_release_and_destroy(bond_dev, slave_dev);
3513 else
3514 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3515 }
3516 break;
3517 case NETDEV_CHANGE:
17d04500
JV		 3518 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3519 struct slave *slave;
3520
3521 slave = bond_get_slave_by_dev(bond, slave_dev);
3522 if (slave) {
3523 u16 old_speed = slave->speed;
3524 u16 old_duplex = slave->duplex;
3525
3526 bond_update_speed_duplex(slave);
3527
3528 if (bond_is_lb(bond))
3529 break;
3530
3531 if (old_speed != slave->speed)
3532 bond_3ad_adapter_speed_changed(slave);
3533 if (old_duplex != slave->duplex)
3534 bond_3ad_adapter_duplex_changed(slave);
3535 }
3536 }
3537
1da177e4
LT		 3538 break;
3539 case NETDEV_DOWN:
3540 /*
3541 * ... Or is it this?
3542 */
3543 break;
3544 case NETDEV_CHANGEMTU:
3545 /*
3546 * TODO: Should slaves be allowed to
3547 * independently alter their MTU? For
3548 * an active-backup bond, slaves need
3549 * not be the same type of device, so
3550 * MTUs may vary. For other modes,
3551 * slaves arguably should have the
3552 * same MTUs. To do this, we'd need to
3553 * take over the slave's change_mtu
3554 * function for the duration of their
3555 * servitude.
3556 */
3557 break;
3558 case NETDEV_CHANGENAME:
3559 /*
3560 * TODO: handle changing the primary's name
3561 */
3562 break;
8531c5ff
AK		 3563 case NETDEV_FEAT_CHANGE:
3564 bond_compute_features(bond);
3565 break;
1da177e4
LT		 3566 default:
3567 break;
3568 }
3569
3570 return NOTIFY_DONE;
3571}
3572
3573/*
3574 * bond_netdev_event: handle netdev notifier chain events.
3575 *
3576 * This function receives events for the netdev chain. The caller (an
e041c683 3577 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3578 * locks for us to safely manipulate the slave devices (RTNL lock,
3579 * dev_probe_lock).
3580 */
3581static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3582{
3583 struct net_device *event_dev = (struct net_device *)ptr;
3584
c346dca1 3585 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3586 return NOTIFY_DONE;
3587
5a03cdb7 3588 pr_debug("event_dev: %s, event: %lx\n",
1da177e4
LT		 3589 (event_dev ? event_dev->name : "None"),
3590 event);
3591
0b680e75
JV		 3592 if (!(event_dev->priv_flags & IFF_BONDING))
3593 return NOTIFY_DONE;
3594
1da177e4 3595 if (event_dev->flags & IFF_MASTER) {
5a03cdb7 3596 pr_debug("IFF_MASTER\n");
1da177e4
LT		 3597 return bond_master_netdev_event(event, event_dev);
3598 }
3599
3600 if (event_dev->flags & IFF_SLAVE) {
5a03cdb7 3601 pr_debug("IFF_SLAVE\n");
1da177e4
LT		 3602 return bond_slave_netdev_event(event, event_dev);
3603 }
3604
3605 return NOTIFY_DONE;
3606}
3607
c3ade5ca
JV		 3608/*
3609 * bond_inetaddr_event: handle inetaddr notifier chain events.
3610 *
3611 * We keep track of device IPs primarily to use as source addresses in
3612 * ARP monitor probes (rather than spewing out broadcasts all the time).
3613 *
3614 * We track one IP for the main device (if it has one), plus one per VLAN.
3615 */
3616static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3617{
3618 struct in_ifaddr *ifa = ptr;
3619 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3620 struct bonding *bond;
3621 struct vlan_entry *vlan;
c3ade5ca 3622
c346dca1 3623 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3624 return NOTIFY_DONE;
3625
0883beca 3626 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3627 if (bond->dev == event_dev) {
3628 switch (event) {
3629 case NETDEV_UP:
3630 bond->master_ip = ifa->ifa_local;
3631 return NOTIFY_OK;
3632 case NETDEV_DOWN:
3633 bond->master_ip = bond_glean_dev_ip(bond->dev);
3634 return NOTIFY_OK;
3635 default:
3636 return NOTIFY_DONE;
3637 }
3638 }
3639
0883beca 3640 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3641 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3642 if (vlan_dev == event_dev) {
3643 switch (event) {
3644 case NETDEV_UP:
3645 vlan->vlan_ip = ifa->ifa_local;
3646 return NOTIFY_OK;
3647 case NETDEV_DOWN:
3648 vlan->vlan_ip =
3649 bond_glean_dev_ip(vlan_dev);
3650 return NOTIFY_OK;
3651 default:
3652 return NOTIFY_DONE;
3653 }
3654 }
3655 }
3656 }
3657 return NOTIFY_DONE;
3658}
3659
1da177e4
LT		 3660static struct notifier_block bond_netdev_notifier = {
3661 .notifier_call = bond_netdev_event,
3662};
3663
c3ade5ca
JV		 3664static struct notifier_block bond_inetaddr_notifier = {
3665 .notifier_call = bond_inetaddr_event,
3666};
3667
1da177e4
LT		 3668/*-------------------------- Packet type handling ---------------------------*/
3669
3670/* register to receive lacpdus on a bond */
3671static void bond_register_lacpdu(struct bonding *bond)
3672{
3673 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3674
3675 /* initialize packet type */
3676 pk_type->type = PKT_TYPE_LACPDU;
3677 pk_type->dev = bond->dev;
3678 pk_type->func = bond_3ad_lacpdu_recv;
3679
3680 dev_add_pack(pk_type);
3681}
3682
3683/* unregister to receive lacpdus on a bond */
3684static void bond_unregister_lacpdu(struct bonding *bond)
3685{
3686 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3687}
3688
f5b2b966
JV		 3689void bond_register_arp(struct bonding *bond)
3690{
3691 struct packet_type *pt = &bond->arp_mon_pt;
3692
c4f283b1
JV		 3693 if (pt->type)
3694 return;
3695
f5b2b966 3696 pt->type = htons(ETH_P_ARP);
e245cb71 3697 pt->dev = bond->dev;
f5b2b966
JV		 3698 pt->func = bond_arp_rcv;
3699 dev_add_pack(pt);
3700}
3701
3702void bond_unregister_arp(struct bonding *bond)
3703{
c4f283b1
JV		 3704 struct packet_type *pt = &bond->arp_mon_pt;
3705
3706 dev_remove_pack(pt);
3707 pt->type = 0;
f5b2b966
JV		 3708}
3709
169a3e66
JV		 3710/*---------------------------- Hashing Policies -----------------------------*/
3711
6f6652be
JV		 3712/*
3713 * Hash for the output device based upon layer 2 and layer 3 data. If
3714 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3715 */
3716static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3717 struct net_device *bond_dev, int count)
3718{
3719 struct ethhdr *data = (struct ethhdr *)skb->data;
3720 struct iphdr *iph = ip_hdr(skb);
3721
f14c4e4e 3722 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be
JV		 3723 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3724 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3725 }
3726
3727 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3728}
3729
169a3e66 3730/*
59c51591 3731 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3732 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3733 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3734 */
3735static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3736 struct net_device *bond_dev, int count)
3737{
3738 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3739 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3740 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3741 int layer4_xor = 0;
3742
f14c4e4e
BH		 3743 if (skb->protocol == htons(ETH_P_IP)) {
3744 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3745 (iph->protocol == IPPROTO_TCP ||
3746 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3747 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3748 }
3749 return (layer4_xor ^
3750 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3751
3752 }
3753
3754 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3755}
3756
3757/*
3758 * Hash for the output device based upon layer 2 data
3759 */
3760static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3761 struct net_device *bond_dev, int count)
3762{
3763 struct ethhdr *data = (struct ethhdr *)skb->data;
3764
3765 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3766}
3767
1da177e4
LT		 3768/*-------------------------- Device entry points ----------------------------*/
3769
3770static int bond_open(struct net_device *bond_dev)
3771{
454d7c9b 3772 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3773
3774 bond->kill_timers = 0;
3775
58402054 3776 if (bond_is_lb(bond)) {
1da177e4
LT		 3777 /* bond_alb_initialize must be called before the timer
3778 * is started.
3779 */
3780 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3781 /* something went wrong - fail the open operation */
3782 return -1;
3783 }
3784
1b76b316
JV		 3785 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3786 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3787 }
3788
3789 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3790 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3791 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3792 }
3793
3794 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3795 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3796 INIT_DELAYED_WORK(&bond->arp_work,
3797 bond_activebackup_arp_mon);
3798 else
3799 INIT_DELAYED_WORK(&bond->arp_work,
3800 bond_loadbalance_arp_mon);
3801
3802 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3803 if (bond->params.arp_validate)
3804 bond_register_arp(bond);
1da177e4
LT		 3805 }
3806
3807 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3808 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3809 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3810 /* register to receive LACPDUs */
3811 bond_register_lacpdu(bond);
fd989c83 3812 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3813 }
3814
3815 return 0;
3816}
3817
3818static int bond_close(struct net_device *bond_dev)
3819{
454d7c9b 3820 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3821
3822 if (bond->params.mode == BOND_MODE_8023AD) {
3823 /* Unregister the receive of LACPDUs */
3824 bond_unregister_lacpdu(bond);
3825 }
3826
f5b2b966
JV		 3827 if (bond->params.arp_validate)
3828 bond_unregister_arp(bond);
3829
1da177e4
LT		 3830 write_lock_bh(&bond->lock);
3831
b59f9f74 3832 bond->send_grat_arp = 0;
305d552a 3833 bond->send_unsol_na = 0;
1da177e4
LT		 3834
3835 /* signal timers not to re-arm */
3836 bond->kill_timers = 1;
3837
3838 write_unlock_bh(&bond->lock);
3839
1da177e4 3840 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3841 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3842 }
3843
3844 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3845 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3846 }
3847
3848 switch (bond->params.mode) {
3849 case BOND_MODE_8023AD:
1b76b316 3850 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3851 break;
3852 case BOND_MODE_TLB:
3853 case BOND_MODE_ALB:
1b76b316 3854 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3855 break;
3856 default:
3857 break;
3858 }
3859
1da177e4 3860
58402054 3861 if (bond_is_lb(bond)) {
1da177e4
LT		 3862 /* Must be called only after all
3863 * slaves have been released
3864 */
3865 bond_alb_deinitialize(bond);
3866 }
3867
3868 return 0;
3869}
3870
3871static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3872{
454d7c9b 3873 struct bonding *bond = netdev_priv(bond_dev);
eeda3fd6 3874 struct net_device_stats *stats = &bond->stats;
2439f9eb 3875 struct net_device_stats local_stats;
1da177e4
LT		 3876 struct slave *slave;
3877 int i;
3878
2439f9eb 3879 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3880
3881 read_lock_bh(&bond->lock);
3882
3883 bond_for_each_slave(bond, slave, i) {
eeda3fd6
SH		 3884 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3885
2439f9eb
AG		 3886 local_stats.rx_packets += sstats->rx_packets;
3887 local_stats.rx_bytes += sstats->rx_bytes;
3888 local_stats.rx_errors += sstats->rx_errors;
3889 local_stats.rx_dropped += sstats->rx_dropped;
3890
3891 local_stats.tx_packets += sstats->tx_packets;
3892 local_stats.tx_bytes += sstats->tx_bytes;
3893 local_stats.tx_errors += sstats->tx_errors;
3894 local_stats.tx_dropped += sstats->tx_dropped;
3895
3896 local_stats.multicast += sstats->multicast;
3897 local_stats.collisions += sstats->collisions;
3898
3899 local_stats.rx_length_errors += sstats->rx_length_errors;
3900 local_stats.rx_over_errors += sstats->rx_over_errors;
3901 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3902 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3903 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3904 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3905
3906 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3907 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3908 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3909 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3910 local_stats.tx_window_errors += sstats->tx_window_errors;
3911 }
3912
3913 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3914
3915 read_unlock_bh(&bond->lock);
3916
3917 return stats;
3918}
3919
3920static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3921{
3922 struct net_device *slave_dev = NULL;
3923 struct ifbond k_binfo;
3924 struct ifbond __user *u_binfo = NULL;
3925 struct ifslave k_sinfo;
3926 struct ifslave __user *u_sinfo = NULL;
3927 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3928 int res = 0;
3929
5a03cdb7 3930 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
1da177e4
LT		 3931 bond_dev->name, cmd);
3932
3933 switch (cmd) {
1da177e4
LT		 3934 case SIOCGMIIPHY:
3935 mii = if_mii(ifr);
3936 if (!mii) {
3937 return -EINVAL;
3938 }
3939 mii->phy_id = 0;
3940 /* Fall Through */
3941 case SIOCGMIIREG:
3942 /*
3943 * We do this again just in case we were called by SIOCGMIIREG
3944 * instead of SIOCGMIIPHY.
3945 */
3946 mii = if_mii(ifr);
3947 if (!mii) {
3948 return -EINVAL;
3949 }
3950
3951 if (mii->reg_num == 1) {
454d7c9b 3952 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3953 mii->val_out = 0;
6603a6f2 3954 read_lock(&bond->lock);
1da177e4 3955 read_lock(&bond->curr_slave_lock);
4e140079 3956 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3957 mii->val_out = BMSR_LSTATUS;
3958 }
3959 read_unlock(&bond->curr_slave_lock);
6603a6f2 3960 read_unlock(&bond->lock);
1da177e4
LT		 3961 }
3962
3963 return 0;
3964 case BOND_INFO_QUERY_OLD:
3965 case SIOCBONDINFOQUERY:
3966 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3967
3968 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3969 return -EFAULT;
3970 }
3971
3972 res = bond_info_query(bond_dev, &k_binfo);
3973 if (res == 0) {
3974 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3975 return -EFAULT;
3976 }
3977 }
3978
3979 return res;
3980 case BOND_SLAVE_INFO_QUERY_OLD:
3981 case SIOCBONDSLAVEINFOQUERY:
3982 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3983
3984 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3985 return -EFAULT;
3986 }
3987
3988 res = bond_slave_info_query(bond_dev, &k_sinfo);
3989 if (res == 0) {
3990 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3991 return -EFAULT;
3992 }
3993 }
3994
3995 return res;
3996 default:
3997 /* Go on */
3998 break;
3999 }
4000
4001 if (!capable(CAP_NET_ADMIN)) {
4002 return -EPERM;
4003 }
4004
b76cdba9 4005 down_write(&(bonding_rwsem));
881d966b 4006 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4 4007
5a03cdb7 4008 pr_debug("slave_dev=%p: \n", slave_dev);
1da177e4
LT		 4009
4010 if (!slave_dev) {
4011 res = -ENODEV;
4012 } else {
5a03cdb7 4013 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
1da177e4
LT		 4014 switch (cmd) {
4015 case BOND_ENSLAVE_OLD:
4016 case SIOCBONDENSLAVE:
4017 res = bond_enslave(bond_dev, slave_dev);
4018 break;
4019 case BOND_RELEASE_OLD:
4020 case SIOCBONDRELEASE:
4021 res = bond_release(bond_dev, slave_dev);
4022 break;
4023 case BOND_SETHWADDR_OLD:
4024 case SIOCBONDSETHWADDR:
4025 res = bond_sethwaddr(bond_dev, slave_dev);
4026 break;
4027 case BOND_CHANGE_ACTIVE_OLD:
4028 case SIOCBONDCHANGEACTIVE:
4029 res = bond_ioctl_change_active(bond_dev, slave_dev);
4030 break;
4031 default:
4032 res = -EOPNOTSUPP;
4033 }
4034
4035 dev_put(slave_dev);
4036 }
4037
b76cdba9 4038 up_write(&(bonding_rwsem));
1da177e4
LT		 4039 return res;
4040}
4041
4042static void bond_set_multicast_list(struct net_device *bond_dev)
4043{
454d7c9b 4044 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4045 struct dev_mc_list *dmi;
4046
1da177e4
LT		 4047 /*
4048 * Do promisc before checking multicast_mode
4049 */
4050 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4051 /*
4052 * FIXME: Need to handle the error when one of the multi-slaves
4053 * encounters error.
4054 */
1da177e4
LT		 4055 bond_set_promiscuity(bond, 1);
4056 }
4057
4058 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4059 bond_set_promiscuity(bond, -1);
4060 }
4061
4062 /* set allmulti flag to slaves */
4063 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4064 /*
4065 * FIXME: Need to handle the error when one of the multi-slaves
4066 * encounters error.
4067 */
1da177e4
LT		 4068 bond_set_allmulti(bond, 1);
4069 }
4070
4071 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4072 bond_set_allmulti(bond, -1);
4073 }
4074
80ee5ad2
JV		 4075 read_lock(&bond->lock);
4076
1da177e4
LT		 4077 bond->flags = bond_dev->flags;
4078
4079 /* looking for addresses to add to slaves' mc list */
4080 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4081 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4082 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4083 }
4084 }
4085
4086 /* looking for addresses to delete from slaves' list */
4087 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4088 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4089 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4090 }
4091 }
4092
4093 /* save master's multicast list */
4094 bond_mc_list_destroy(bond);
4095 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4096
80ee5ad2 4097 read_unlock(&bond->lock);
1da177e4
LT		 4098}
4099
00829823
SH		 4100static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4101{
4102 struct bonding *bond = netdev_priv(dev);
4103 struct slave *slave = bond->first_slave;
4104
4105 if (slave) {
4106 const struct net_device_ops *slave_ops
4107 = slave->dev->netdev_ops;
4108 if (slave_ops->ndo_neigh_setup)
72e2240f 4109 return slave_ops->ndo_neigh_setup(slave->dev, parms);
00829823
SH		 4110 }
4111 return 0;
4112}
4113
1da177e4
LT		 4114/*
4115 * Change the MTU of all of a master's slaves to match the master
4116 */
4117static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4118{
454d7c9b 4119 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4120 struct slave *slave, *stop_at;
4121 int res = 0;
4122 int i;
4123
5a03cdb7 4124 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
1da177e4
LT		 4125 (bond_dev ? bond_dev->name : "None"), new_mtu);
4126
4127 /* Can't hold bond->lock with bh disabled here since
4128 * some base drivers panic. On the other hand we can't
4129 * hold bond->lock without bh disabled because we'll
4130 * deadlock. The only solution is to rely on the fact
4131 * that we're under rtnl_lock here, and the slaves
4132 * list won't change. This doesn't solve the problem
4133 * of setting the slave's MTU while it is
4134 * transmitting, but the assumption is that the base
4135 * driver can handle that.
4136 *
4137 * TODO: figure out a way to safely iterate the slaves
4138 * list, but without holding a lock around the actual
4139 * call to the base driver.
4140 */
4141
4142 bond_for_each_slave(bond, slave, i) {
5a03cdb7 4143 pr_debug("s %p s->p %p c_m %p\n", slave,
53a3294e 4144 slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
e944ef79 4145
1da177e4
LT		 4146 res = dev_set_mtu(slave->dev, new_mtu);
4147
4148 if (res) {
4149 /* If we failed to set the slave's mtu to the new value
4150 * we must abort the operation even in ACTIVE_BACKUP
4151 * mode, because if we allow the backup slaves to have
4152 * different mtu values than the active slave we'll
4153 * need to change their mtu when doing a failover. That
4154 * means changing their mtu from timer context, which
4155 * is probably not a good idea.
4156 */
5a03cdb7 4157 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 4158 goto unwind;
4159 }
4160 }
4161
4162 bond_dev->mtu = new_mtu;
4163
4164 return 0;
4165
4166unwind:
4167 /* unwind from head to the slave that failed */
4168 stop_at = slave;
4169 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4170 int tmp_res;
4171
4172 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4173 if (tmp_res) {
5a03cdb7 4174 pr_debug("unwind err %d dev %s\n", tmp_res,
1da177e4
LT		 4175 slave->dev->name);
4176 }
4177 }
4178
4179 return res;
4180}
4181
4182/*
4183 * Change HW address
4184 *
4185 * Note that many devices must be down to change the HW address, and
4186 * downing the master releases all slaves. We can make bonds full of
4187 * bonding devices to test this, however.
4188 */
4189static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4190{
454d7c9b 4191 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4192 struct sockaddr *sa = addr, tmp_sa;
4193 struct slave *slave, *stop_at;
4194 int res = 0;
4195 int i;
4196
eb7cc59a
SH		 4197 if (bond->params.mode == BOND_MODE_ALB)
4198 return bond_alb_set_mac_address(bond_dev, addr);
4199
4200
5a03cdb7 4201 pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
1da177e4 4202
dd957c57 4203 /*
3915c1e8
JV		 4204 * If fail_over_mac is set to active, do nothing and return
4205 * success. Returning an error causes ifenslave to fail.
dd957c57 4206 */
3915c1e8 4207 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4208 return 0;
2ab82852 4209
1da177e4
LT		 4210 if (!is_valid_ether_addr(sa->sa_data)) {
4211 return -EADDRNOTAVAIL;
4212 }
4213
4214 /* Can't hold bond->lock with bh disabled here since
4215 * some base drivers panic. On the other hand we can't
4216 * hold bond->lock without bh disabled because we'll
4217 * deadlock. The only solution is to rely on the fact
4218 * that we're under rtnl_lock here, and the slaves
4219 * list won't change. This doesn't solve the problem
4220 * of setting the slave's hw address while it is
4221 * transmitting, but the assumption is that the base
4222 * driver can handle that.
4223 *
4224 * TODO: figure out a way to safely iterate the slaves
4225 * list, but without holding a lock around the actual
4226 * call to the base driver.
4227 */
4228
4229 bond_for_each_slave(bond, slave, i) {
eb7cc59a 4230 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
5a03cdb7 4231 pr_debug("slave %p %s\n", slave, slave->dev->name);
1da177e4 4232
eb7cc59a 4233 if (slave_ops->ndo_set_mac_address == NULL) {
1da177e4 4234 res = -EOPNOTSUPP;
5a03cdb7 4235 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
1da177e4
LT		 4236 goto unwind;
4237 }
4238
4239 res = dev_set_mac_address(slave->dev, addr);
4240 if (res) {
4241 /* TODO: consider downing the slave
4242 * and retry ?
4243 * User should expect communications
4244 * breakage anyway until ARP finish
4245 * updating, so...
4246 */
5a03cdb7 4247 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 4248 goto unwind;
4249 }
4250 }
4251
4252 /* success */
4253 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4254 return 0;
4255
4256unwind:
4257 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4258 tmp_sa.sa_family = bond_dev->type;
4259
4260 /* unwind from head to the slave that failed */
4261 stop_at = slave;
4262 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4263 int tmp_res;
4264
4265 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4266 if (tmp_res) {
5a03cdb7 4267 pr_debug("unwind err %d dev %s\n", tmp_res,
1da177e4
LT		 4268 slave->dev->name);
4269 }
4270 }
4271
4272 return res;
4273}
4274
4275static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4276{
454d7c9b 4277 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4278 struct slave *slave, *start_at;
cf5f9044 4279 int i, slave_no, res = 1;
1da177e4
LT		 4280
4281 read_lock(&bond->lock);
4282
4283 if (!BOND_IS_OK(bond)) {
4284 goto out;
4285 }
4286
cf5f9044
JV		 4287 /*
4288 * Concurrent TX may collide on rr_tx_counter; we accept that
4289 * as being rare enough not to justify using an atomic op here
4290 */
4291 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4292
cf5f9044
JV		 4293 bond_for_each_slave(bond, slave, i) {
4294 slave_no--;
4295 if (slave_no < 0) {
4296 break;
4297 }
1da177e4
LT		 4298 }
4299
cf5f9044 4300 start_at = slave;
1da177e4
LT		 4301 bond_for_each_slave_from(bond, slave, i, start_at) {
4302 if (IS_UP(slave->dev) &&
4303 (slave->link == BOND_LINK_UP) &&
4304 (slave->state == BOND_STATE_ACTIVE)) {
4305 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4306 break;
4307 }
4308 }
4309
1da177e4
LT		 4310out:
4311 if (res) {
4312 /* no suitable interface, frame not sent */
4313 dev_kfree_skb(skb);
4314 }
4315 read_unlock(&bond->lock);
4316 return 0;
4317}
4318
075897ce 4319
1da177e4
LT		 4320/*
4321 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4322 * the bond has a usable interface.
4323 */
4324static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4325{
454d7c9b 4326 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4327 int res = 1;
4328
1da177e4
LT		 4329 read_lock(&bond->lock);
4330 read_lock(&bond->curr_slave_lock);
4331
4332 if (!BOND_IS_OK(bond)) {
4333 goto out;
4334 }
4335
075897ce
JL		 4336 if (!bond->curr_active_slave)
4337 goto out;
4338
075897ce
JL		 4339 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4340
1da177e4
LT		 4341out:
4342 if (res) {
4343 /* no suitable interface, frame not sent */
4344 dev_kfree_skb(skb);
4345 }
4346 read_unlock(&bond->curr_slave_lock);
4347 read_unlock(&bond->lock);
4348 return 0;
4349}
4350
4351/*
169a3e66
JV		 4352 * In bond_xmit_xor() , we determine the output device by using a pre-
4353 * determined xmit_hash_policy(), If the selected device is not enabled,
4354 * find the next active slave.
1da177e4
LT		 4355 */
4356static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4357{
454d7c9b 4358 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4359 struct slave *slave, *start_at;
4360 int slave_no;
4361 int i;
4362 int res = 1;
4363
4364 read_lock(&bond->lock);
4365
4366 if (!BOND_IS_OK(bond)) {
4367 goto out;
4368 }
4369
169a3e66 4370 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4371
4372 bond_for_each_slave(bond, slave, i) {
4373 slave_no--;
4374 if (slave_no < 0) {
4375 break;
4376 }
4377 }
4378
4379 start_at = slave;
4380
4381 bond_for_each_slave_from(bond, slave, i, start_at) {
4382 if (IS_UP(slave->dev) &&
4383 (slave->link == BOND_LINK_UP) &&
4384 (slave->state == BOND_STATE_ACTIVE)) {
4385 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4386 break;
4387 }
4388 }
4389
4390out:
4391 if (res) {
4392 /* no suitable interface, frame not sent */
4393 dev_kfree_skb(skb);
4394 }
4395 read_unlock(&bond->lock);
4396 return 0;
4397}
4398
4399/*
4400 * in broadcast mode, we send everything to all usable interfaces.
4401 */
4402static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4403{
454d7c9b 4404 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4405 struct slave *slave, *start_at;
4406 struct net_device *tx_dev = NULL;
4407 int i;
4408 int res = 1;
4409
4410 read_lock(&bond->lock);
4411
4412 if (!BOND_IS_OK(bond)) {
4413 goto out;
4414 }
4415
4416 read_lock(&bond->curr_slave_lock);
4417 start_at = bond->curr_active_slave;
4418 read_unlock(&bond->curr_slave_lock);
4419
4420 if (!start_at) {
4421 goto out;
4422 }
4423
4424 bond_for_each_slave_from(bond, slave, i, start_at) {
4425 if (IS_UP(slave->dev) &&
4426 (slave->link == BOND_LINK_UP) &&
4427 (slave->state == BOND_STATE_ACTIVE)) {
4428 if (tx_dev) {
4429 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4430 if (!skb2) {
4431 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4432 ": %s: Error: bond_xmit_broadcast(): "
4433 "skb_clone() failed\n",
4434 bond_dev->name);
1da177e4
LT		 4435 continue;
4436 }
4437
4438 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4439 if (res) {
4440 dev_kfree_skb(skb2);
4441 continue;
4442 }
4443 }
4444 tx_dev = slave->dev;
4445 }
4446 }
4447
4448 if (tx_dev) {
4449 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4450 }
4451
4452out:
4453 if (res) {
4454 /* no suitable interface, frame not sent */
4455 dev_kfree_skb(skb);
4456 }
4457 /* frame sent to all suitable interfaces */
4458 read_unlock(&bond->lock);
4459 return 0;
4460}
4461
4462/*------------------------- Device initialization ---------------------------*/
4463
6f6652be
JV		 4464static void bond_set_xmit_hash_policy(struct bonding *bond)
4465{
4466 switch (bond->params.xmit_policy) {
4467 case BOND_XMIT_POLICY_LAYER23:
4468 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4469 break;
4470 case BOND_XMIT_POLICY_LAYER34:
4471 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4472 break;
4473 case BOND_XMIT_POLICY_LAYER2:
4474 default:
4475 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4476 break;
4477 }
4478}
4479
00829823
SH		 4480static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4481{
4482 const struct bonding *bond = netdev_priv(dev);
4483
4484 switch (bond->params.mode) {
4485 case BOND_MODE_ROUNDROBIN:
4486 return bond_xmit_roundrobin(skb, dev);
4487 case BOND_MODE_ACTIVEBACKUP:
4488 return bond_xmit_activebackup(skb, dev);
4489 case BOND_MODE_XOR:
4490 return bond_xmit_xor(skb, dev);
4491 case BOND_MODE_BROADCAST:
4492 return bond_xmit_broadcast(skb, dev);
4493 case BOND_MODE_8023AD:
4494 return bond_3ad_xmit_xor(skb, dev);
4495 case BOND_MODE_ALB:
4496 case BOND_MODE_TLB:
4497 return bond_alb_xmit(skb, dev);
4498 default:
4499 /* Should never happen, mode already checked */
4500 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4501 dev->name, bond->params.mode);
4502 WARN_ON_ONCE(1);
4503 dev_kfree_skb(skb);
4504 return NETDEV_TX_OK;
4505 }
4506}
4507
4508
1da177e4
LT		 4509/*
4510 * set bond mode specific net device operations
4511 */
a77b5325 4512void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4513{
169a3e66
JV		 4514 struct net_device *bond_dev = bond->dev;
4515
1da177e4
LT		 4516 switch (mode) {
4517 case BOND_MODE_ROUNDROBIN:
1da177e4
LT		 4518 break;
4519 case BOND_MODE_ACTIVEBACKUP:
1da177e4
LT		 4520 break;
4521 case BOND_MODE_XOR:
6f6652be 4522 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4523 break;
4524 case BOND_MODE_BROADCAST:
1da177e4
LT		 4525 break;
4526 case BOND_MODE_8023AD:
8f903c70 4527 bond_set_master_3ad_flags(bond);
6f6652be 4528 bond_set_xmit_hash_policy(bond);
1da177e4 4529 break;
1da177e4 4530 case BOND_MODE_ALB:
8f903c70
JV		 4531 bond_set_master_alb_flags(bond);
4532 /* FALLTHRU */
4533 case BOND_MODE_TLB:
1da177e4
LT		 4534 break;
4535 default:
4536 /* Should never happen, mode already checked */
4537 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4538 ": %s: Error: Unknown bonding mode %d\n",
4539 bond_dev->name,
1da177e4
LT		 4540 mode);
4541 break;
4542 }
4543}
4544
217df670
JV		 4545static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4546 struct ethtool_drvinfo *drvinfo)
4547{
4548 strncpy(drvinfo->driver, DRV_NAME, 32);
4549 strncpy(drvinfo->version, DRV_VERSION, 32);
4550 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4551}
4552
7282d491 4553static const struct ethtool_ops bond_ethtool_ops = {
217df670 4554 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac
SH		 4555 .get_link = ethtool_op_get_link,
4556 .get_tx_csum = ethtool_op_get_tx_csum,
4557 .get_sg = ethtool_op_get_sg,
4558 .get_tso = ethtool_op_get_tso,
4559 .get_ufo = ethtool_op_get_ufo,
4560 .get_flags = ethtool_op_get_flags,
8531c5ff
AK		 4561};
4562
eb7cc59a
SH		 4563static const struct net_device_ops bond_netdev_ops = {
4564 .ndo_open = bond_open,
4565 .ndo_stop = bond_close,
00829823 4566 .ndo_start_xmit = bond_start_xmit,
eb7cc59a
SH		 4567 .ndo_get_stats = bond_get_stats,
4568 .ndo_do_ioctl = bond_do_ioctl,
4569 .ndo_set_multicast_list = bond_set_multicast_list,
4570 .ndo_change_mtu = bond_change_mtu,
eb7cc59a 4571 .ndo_set_mac_address = bond_set_mac_address,
00829823 4572 .ndo_neigh_setup = bond_neigh_setup,
eb7cc59a
SH		 4573 .ndo_vlan_rx_register = bond_vlan_rx_register,
4574 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4575 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4576};
4577
1da177e4
LT		 4578/*
4579 * Does not allocate but creates a /proc entry.
4580 * Allowed to fail.
4581 */
3c535952 4582static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4 4583{
454d7c9b 4584 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4585
5a03cdb7 4586 pr_debug("Begin bond_init for %s\n", bond_dev->name);
1da177e4
LT		 4587
4588 /* initialize rwlocks */
4589 rwlock_init(&bond->lock);
4590 rwlock_init(&bond->curr_slave_lock);
4591
4592 bond->params = *params; /* copy params struct */
4593
1b76b316
JV		 4594 bond->wq = create_singlethread_workqueue(bond_dev->name);
4595 if (!bond->wq)
4596 return -ENOMEM;
4597
1da177e4
LT		 4598 /* Initialize pointers */
4599 bond->first_slave = NULL;
4600 bond->curr_active_slave = NULL;
4601 bond->current_arp_slave = NULL;
4602 bond->primary_slave = NULL;
4603 bond->dev = bond_dev;
1053f62c 4604 bond->send_grat_arp = 0;
305d552a 4605 bond->send_unsol_na = 0;
d90a162a 4606 bond->setup_by_slave = 0;
1da177e4
LT		 4607 INIT_LIST_HEAD(&bond->vlan_list);
4608
4609 /* Initialize the device entry points */
eb7cc59a 4610 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 4611 bond_dev->ethtool_ops = &bond_ethtool_ops;
169a3e66 4612 bond_set_mode_ops(bond, bond->params.mode);
1da177e4 4613
a434e43f 4614 bond_dev->destructor = bond_destructor;
1da177e4
LT		 4615
4616 /* Initialize the device options */
4617 bond_dev->tx_queue_len = 0;
4618 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4619 bond_dev->priv_flags |= IFF_BONDING;
6cf3f41e
JV		 4620 if (bond->params.arp_interval)
4621 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
1da177e4
LT		 4622
4623 /* At first, we block adding VLANs. That's the only way to
4624 * prevent problems that occur when adding VLANs over an
4625 * empty bond. The block will be removed once non-challenged
4626 * slaves are enslaved.
4627 */
4628 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4629
932ff279 4630 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4631 * transmitting */
4632 bond_dev->features |= NETIF_F_LLTX;
4633
4634 /* By default, we declare the bond to be fully
4635 * VLAN hardware accelerated capable. Special
4636 * care is taken in the various xmit functions
4637 * when there are slaves that are not hw accel
4638 * capable
4639 */
1da177e4
LT		 4640 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4641 NETIF_F_HW_VLAN_RX |
4642 NETIF_F_HW_VLAN_FILTER);
4643
4644#ifdef CONFIG_PROC_FS
4645 bond_create_proc_entry(bond);
4646#endif
1da177e4
LT		 4647 list_add_tail(&bond->bond_list, &bond_dev_list);
4648
4649 return 0;
4650}
4651
fdaea7a9
JV		 4652static void bond_work_cancel_all(struct bonding *bond)
4653{
4654 write_lock_bh(&bond->lock);
4655 bond->kill_timers = 1;
4656 write_unlock_bh(&bond->lock);
4657
4658 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4659 cancel_delayed_work(&bond->mii_work);
4660
4661 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4662 cancel_delayed_work(&bond->arp_work);
4663
4664 if (bond->params.mode == BOND_MODE_ALB &&
4665 delayed_work_pending(&bond->alb_work))
4666 cancel_delayed_work(&bond->alb_work);
4667
4668 if (bond->params.mode == BOND_MODE_8023AD &&
4669 delayed_work_pending(&bond->ad_work))
4670 cancel_delayed_work(&bond->ad_work);
4671}
4672
a434e43f
JV		 4673/* De-initialize device specific data.
4674 * Caller must hold rtnl_lock.
4675 */
4676static void bond_deinit(struct net_device *bond_dev)
4677{
454d7c9b 4678 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 4679
4680 list_del(&bond->bond_list);
4681
4682 bond_work_cancel_all(bond);
4683
4684#ifdef CONFIG_PROC_FS
4685 bond_remove_proc_entry(bond);
4686#endif
4687}
4688
1da177e4
LT		 4689/* Unregister and free all bond devices.
4690 * Caller must hold rtnl_lock.
4691 */
4692static void bond_free_all(void)
4693{
4694 struct bonding *bond, *nxt;
4695
4696 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4697 struct net_device *bond_dev = bond->dev;
4698
fdaea7a9 4699 bond_work_cancel_all(bond);
70298705 4700 /* Release the bonded slaves */
4701 bond_release_all(bond_dev);
92b41daa 4702 bond_destroy(bond);
1da177e4
LT		 4703 }
4704
4705#ifdef CONFIG_PROC_FS
4706 bond_destroy_proc_dir();
4707#endif
4708}
4709
4710/*------------------------- Module initialization ---------------------------*/
4711
4712/*
4713 * Convert string input module parms. Accept either the
ece95f7f
JV		 4714 * number of the mode or its string name. A bit complicated because
4715 * some mode names are substrings of other names, and calls from sysfs
4716 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4717 */
325dcf7a 4718int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
1da177e4 4719{
54b87323 4720 int modeint = -1, i, rv;
a42e534f 4721 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4722
a42e534f
JV		 4723 for (p = (char *)buf; *p; p++)
4724 if (!(isdigit(*p) || isspace(*p)))
4725 break;
4726
4727 if (*p)
ece95f7f 4728 rv = sscanf(buf, "%20s", modestr);
a42e534f 4729 else
54b87323 4730 rv = sscanf(buf, "%d", &modeint);
a42e534f
JV		 4731
4732 if (!rv)
4733 return -1;
1da177e4
LT		 4734
4735 for (i = 0; tbl[i].modename; i++) {
54b87323 4736 if (modeint == tbl[i].mode)
ece95f7f
JV		 4737 return tbl[i].mode;
4738 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4739 return tbl[i].mode;
1da177e4
LT		 4740 }
4741
4742 return -1;
4743}
4744
4745static int bond_check_params(struct bond_params *params)
4746{
3915c1e8 4747 int arp_validate_value, fail_over_mac_value;
f5b2b966 4748
1da177e4
LT		 4749 /*
4750 * Convert string parameters.
4751 */
4752 if (mode) {
4753 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4754 if (bond_mode == -1) {
4755 printk(KERN_ERR DRV_NAME
4756 ": Error: Invalid bonding mode \"%s\"\n",
4757 mode == NULL ? "NULL" : mode);
4758 return -EINVAL;
4759 }
4760 }
4761
169a3e66
JV		 4762 if (xmit_hash_policy) {
4763 if ((bond_mode != BOND_MODE_XOR) &&
4764 (bond_mode != BOND_MODE_8023AD)) {
4765 printk(KERN_INFO DRV_NAME
4766 ": xor_mode param is irrelevant in mode %s\n",
4767 bond_mode_name(bond_mode));
4768 } else {
4769 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4770 xmit_hashtype_tbl);
4771 if (xmit_hashtype == -1) {
4772 printk(KERN_ERR DRV_NAME
4773 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4774 xmit_hash_policy == NULL ? "NULL" :
4775 xmit_hash_policy);
4776 return -EINVAL;
4777 }
4778 }
4779 }
4780
1da177e4
LT		 4781 if (lacp_rate) {
4782 if (bond_mode != BOND_MODE_8023AD) {
4783 printk(KERN_INFO DRV_NAME
4784 ": lacp_rate param is irrelevant in mode %s\n",
4785 bond_mode_name(bond_mode));
4786 } else {
4787 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4788 if (lacp_fast == -1) {
4789 printk(KERN_ERR DRV_NAME
4790 ": Error: Invalid lacp rate \"%s\"\n",
4791 lacp_rate == NULL ? "NULL" : lacp_rate);
4792 return -EINVAL;
4793 }
4794 }
4795 }
4796
fd989c83
JV		 4797 if (ad_select) {
4798 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4799 if (params->ad_select == -1) {
4800 printk(KERN_ERR DRV_NAME
4801 ": Error: Invalid ad_select \"%s\"\n",
4802 ad_select == NULL ? "NULL" : ad_select);
4803 return -EINVAL;
4804 }
4805
4806 if (bond_mode != BOND_MODE_8023AD) {
4807 printk(KERN_WARNING DRV_NAME
4808 ": ad_select param only affects 802.3ad mode\n");
4809 }
4810 } else {
4811 params->ad_select = BOND_AD_STABLE;
4812 }
4813
b8a9787e 4814 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4815 printk(KERN_WARNING DRV_NAME
4816 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4817 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4818 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4819 max_bonds = BOND_DEFAULT_MAX_BONDS;
4820 }
4821
4822 if (miimon < 0) {
4823 printk(KERN_WARNING DRV_NAME
4824 ": Warning: miimon module parameter (%d), "
4825 "not in range 0-%d, so it was reset to %d\n",
4826 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4827 miimon = BOND_LINK_MON_INTERV;
4828 }
4829
4830 if (updelay < 0) {
4831 printk(KERN_WARNING DRV_NAME
4832 ": Warning: updelay module parameter (%d), "
4833 "not in range 0-%d, so it was reset to 0\n",
4834 updelay, INT_MAX);
4835 updelay = 0;
4836 }
4837
4838 if (downdelay < 0) {
4839 printk(KERN_WARNING DRV_NAME
4840 ": Warning: downdelay module parameter (%d), "
4841 "not in range 0-%d, so it was reset to 0\n",
4842 downdelay, INT_MAX);
4843 downdelay = 0;
4844 }
4845
4846 if ((use_carrier != 0) && (use_carrier != 1)) {
4847 printk(KERN_WARNING DRV_NAME
4848 ": Warning: use_carrier module parameter (%d), "
4849 "not of valid value (0/1), so it was set to 1\n",
4850 use_carrier);
4851 use_carrier = 1;
4852 }
4853
7893b249
MS		 4854 if (num_grat_arp < 0 || num_grat_arp > 255) {
4855 printk(KERN_WARNING DRV_NAME
4856 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4857 "was reset to 1 \n", num_grat_arp);
4858 num_grat_arp = 1;
4859 }
4860
305d552a
BH		 4861 if (num_unsol_na < 0 || num_unsol_na > 255) {
4862 printk(KERN_WARNING DRV_NAME
4863 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4864 "was reset to 1 \n", num_unsol_na);
4865 num_unsol_na = 1;
4866 }
4867
1da177e4
LT		 4868 /* reset values for 802.3ad */
4869 if (bond_mode == BOND_MODE_8023AD) {
4870 if (!miimon) {
4871 printk(KERN_WARNING DRV_NAME
4872 ": Warning: miimon must be specified, "
4873 "otherwise bonding will not detect link "
4874 "failure, speed and duplex which are "
4875 "essential for 802.3ad operation\n");
4876 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4877 miimon = 100;
4878 }
4879 }
4880
4881 /* reset values for TLB/ALB */
4882 if ((bond_mode == BOND_MODE_TLB) ||
4883 (bond_mode == BOND_MODE_ALB)) {
4884 if (!miimon) {
4885 printk(KERN_WARNING DRV_NAME
4886 ": Warning: miimon must be specified, "
4887 "otherwise bonding will not detect link "
4888 "failure and link speed which are essential "
4889 "for TLB/ALB load balancing\n");
4890 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4891 miimon = 100;
4892 }
4893 }
4894
4895 if (bond_mode == BOND_MODE_ALB) {
4896 printk(KERN_NOTICE DRV_NAME
4897 ": In ALB mode you might experience client "
4898 "disconnections upon reconnection of a link if the "
4899 "bonding module updelay parameter (%d msec) is "
4900 "incompatible with the forwarding delay time of the "
4901 "switch\n",
4902 updelay);
4903 }
4904
4905 if (!miimon) {
4906 if (updelay || downdelay) {
4907 /* just warn the user the up/down delay will have
4908 * no effect since miimon is zero...
4909 */
4910 printk(KERN_WARNING DRV_NAME
4911 ": Warning: miimon module parameter not set "
4912 "and updelay (%d) or downdelay (%d) module "
4913 "parameter is set; updelay and downdelay have "
4914 "no effect unless miimon is set\n",
4915 updelay, downdelay);
4916 }
4917 } else {
4918 /* don't allow arp monitoring */
4919 if (arp_interval) {
4920 printk(KERN_WARNING DRV_NAME
4921 ": Warning: miimon (%d) and arp_interval (%d) "
4922 "can't be used simultaneously, disabling ARP "
4923 "monitoring\n",
4924 miimon, arp_interval);
4925 arp_interval = 0;
4926 }
4927
4928 if ((updelay % miimon) != 0) {
4929 printk(KERN_WARNING DRV_NAME
4930 ": Warning: updelay (%d) is not a multiple "
4931 "of miimon (%d), updelay rounded to %d ms\n",
4932 updelay, miimon, (updelay / miimon) * miimon);
4933 }
4934
4935 updelay /= miimon;
4936
4937 if ((downdelay % miimon) != 0) {
4938 printk(KERN_WARNING DRV_NAME
4939 ": Warning: downdelay (%d) is not a multiple "
4940 "of miimon (%d), downdelay rounded to %d ms\n",
4941 downdelay, miimon,
4942 (downdelay / miimon) * miimon);
4943 }
4944
4945 downdelay /= miimon;
4946 }
4947
4948 if (arp_interval < 0) {
4949 printk(KERN_WARNING DRV_NAME
4950 ": Warning: arp_interval module parameter (%d) "
4951 ", not in range 0-%d, so it was reset to %d\n",
4952 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4953 arp_interval = BOND_LINK_ARP_INTERV;
4954 }
4955
4956 for (arp_ip_count = 0;
4957 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4958 arp_ip_count++) {
4959 /* not complete check, but should be good enough to
4960 catch mistakes */
4961 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4962 printk(KERN_WARNING DRV_NAME
4963 ": Warning: bad arp_ip_target module parameter "
4964 "(%s), ARP monitoring will not be performed\n",
4965 arp_ip_target[arp_ip_count]);
4966 arp_interval = 0;
4967 } else {
d3bb52b0 4968 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4969 arp_target[arp_ip_count] = ip;
4970 }
4971 }
4972
4973 if (arp_interval && !arp_ip_count) {
4974 /* don't allow arping if no arp_ip_target given... */
4975 printk(KERN_WARNING DRV_NAME
4976 ": Warning: arp_interval module parameter (%d) "
4977 "specified without providing an arp_ip_target "
4978 "parameter, arp_interval was reset to 0\n",
4979 arp_interval);
4980 arp_interval = 0;
4981 }
4982
f5b2b966
JV		 4983 if (arp_validate) {
4984 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4985 printk(KERN_ERR DRV_NAME
4986 ": arp_validate only supported in active-backup mode\n");
4987 return -EINVAL;
4988 }
4989 if (!arp_interval) {
4990 printk(KERN_ERR DRV_NAME
4991 ": arp_validate requires arp_interval\n");
4992 return -EINVAL;
4993 }
4994
4995 arp_validate_value = bond_parse_parm(arp_validate,
4996 arp_validate_tbl);
4997 if (arp_validate_value == -1) {
4998 printk(KERN_ERR DRV_NAME
4999 ": Error: invalid arp_validate \"%s\"\n",
5000 arp_validate == NULL ? "NULL" : arp_validate);
5001 return -EINVAL;
5002 }
5003 } else
5004 arp_validate_value = 0;
5005
1da177e4
LT		 5006 if (miimon) {
5007 printk(KERN_INFO DRV_NAME
5008 ": MII link monitoring set to %d ms\n",
5009 miimon);
5010 } else if (arp_interval) {
5011 int i;
5012
5013 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 5014 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5015 arp_interval,
5016 arp_validate_tbl[arp_validate_value].modename,
5017 arp_ip_count);
1da177e4
LT		 5018
5019 for (i = 0; i < arp_ip_count; i++)
5020 printk (" %s", arp_ip_target[i]);
5021
5022 printk("\n");
5023
b8a9787e 5024 } else if (max_bonds) {
1da177e4
LT		 5025 /* miimon and arp_interval not set, we need one so things
5026 * work as expected, see bonding.txt for details
5027 */
5028 printk(KERN_WARNING DRV_NAME
5029 ": Warning: either miimon or arp_interval and "
5030 "arp_ip_target module parameters must be specified, "
5031 "otherwise bonding will not detect link failures! see "
5032 "bonding.txt for details.\n");
5033 }
5034
5035 if (primary && !USES_PRIMARY(bond_mode)) {
5036 /* currently, using a primary only makes sense
5037 * in active backup, TLB or ALB modes
5038 */
5039 printk(KERN_WARNING DRV_NAME
5040 ": Warning: %s primary device specified but has no "
5041 "effect in %s mode\n",
5042 primary, bond_mode_name(bond_mode));
5043 primary = NULL;
5044 }
5045
3915c1e8
JV		 5046 if (fail_over_mac) {
5047 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5048 fail_over_mac_tbl);
5049 if (fail_over_mac_value == -1) {
5050 printk(KERN_ERR DRV_NAME
5051 ": Error: invalid fail_over_mac \"%s\"\n",
5052 arp_validate == NULL ? "NULL" : arp_validate);
5053 return -EINVAL;
5054 }
5055
5056 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5057 printk(KERN_WARNING DRV_NAME
5058 ": Warning: fail_over_mac only affects "
5059 "active-backup mode.\n");
5060 } else {
5061 fail_over_mac_value = BOND_FOM_NONE;
5062 }
dd957c57 5063
1da177e4
LT		 5064 /* fill params struct with the proper values */
5065 params->mode = bond_mode;
169a3e66 5066 params->xmit_policy = xmit_hashtype;
1da177e4 5067 params->miimon = miimon;
7893b249 5068 params->num_grat_arp = num_grat_arp;
305d552a 5069 params->num_unsol_na = num_unsol_na;
1da177e4 5070 params->arp_interval = arp_interval;
f5b2b966 5071 params->arp_validate = arp_validate_value;
1da177e4
LT		 5072 params->updelay = updelay;
5073 params->downdelay = downdelay;
5074 params->use_carrier = use_carrier;
5075 params->lacp_fast = lacp_fast;
5076 params->primary[0] = 0;
3915c1e8 5077 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5078
5079 if (primary) {
5080 strncpy(params->primary, primary, IFNAMSIZ);
5081 params->primary[IFNAMSIZ - 1] = 0;
5082 }
5083
5084 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5085
5086 return 0;
5087}
5088
0daa2303 5089static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 5090static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 5091
e8a0464c
DM		 5092static void bond_set_lockdep_class_one(struct net_device *dev,
5093 struct netdev_queue *txq,
5094 void *_unused)
c773e847
DM		 5095{
5096 lockdep_set_class(&txq->_xmit_lock,
5097 &bonding_netdev_xmit_lock_key);
5098}
5099
5100static void bond_set_lockdep_class(struct net_device *dev)
5101{
cf508b12
DM		 5102 lockdep_set_class(&dev->addr_list_lock,
5103 &bonding_netdev_addr_lock_key);
e8a0464c 5104 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 5105}
5106
dfe60397 5107/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5108 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5109 * Caller must NOT hold rtnl_lock; we need to release it here before we
5110 * set up our sysfs entries.
5111 */
0dd646fe 5112int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5113{
5114 struct net_device *bond_dev;
0883beca 5115 struct bonding *bond;
dfe60397
MW		 5116 int res;
5117
5118 rtnl_lock();
027ea041
JV		 5119 down_write(&bonding_rwsem);
5120
5121 /* Check to see if the bond already exists. */
4fe4763c 5122 if (name) {
0883beca 5123 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5124 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5125 printk(KERN_ERR DRV_NAME
027ea041 5126 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5127 name);
5128 res = -EPERM;
5129 goto out_rtnl;
5130 }
5131 }
027ea041 5132
e4b91c48
JV		 5133 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5134 ether_setup);
dfe60397
MW		 5135 if (!bond_dev) {
5136 printk(KERN_ERR DRV_NAME
5137 ": %s: eek! can't alloc netdev!\n",
5138 name);
5139 res = -ENOMEM;
5140 goto out_rtnl;
5141 }
5142
e4b91c48
JV		 5143 if (!name) {
5144 res = dev_alloc_name(bond_dev, "bond%d");
5145 if (res < 0)
5146 goto out_netdev;
5147 }
5148
dfe60397
MW		 5149 /* bond_init() must be called after dev_alloc_name() (for the
5150 * /proc files), but before register_netdevice(), because we
5151 * need to set function pointers.
5152 */
5153
5154 res = bond_init(bond_dev, params);
5155 if (res < 0) {
5156 goto out_netdev;
5157 }
5158
dfe60397
MW		 5159 res = register_netdevice(bond_dev);
5160 if (res < 0) {
5161 goto out_bond;
5162 }
0daa2303 5163
c773e847 5164 bond_set_lockdep_class(bond_dev);
0daa2303 5165
ff59c456
JV		 5166 netif_carrier_off(bond_dev);
5167
027ea041 5168 up_write(&bonding_rwsem);
dfe60397 5169 rtnl_unlock(); /* allows sysfs registration of net device */
454d7c9b 5170 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
09c89279
JV		 5171 if (res < 0) {
5172 rtnl_lock();
027ea041 5173 down_write(&bonding_rwsem);
822973ba
PE		 5174 bond_deinit(bond_dev);
5175 unregister_netdevice(bond_dev);
5176 goto out_rtnl;
09c89279
JV		 5177 }
5178
5179 return 0;
5180
dfe60397
MW		 5181out_bond:
5182 bond_deinit(bond_dev);
5183out_netdev:
5184 free_netdev(bond_dev);
5185out_rtnl:
027ea041 5186 up_write(&bonding_rwsem);
dfe60397 5187 rtnl_unlock();
dfe60397
MW		 5188 return res;
5189}
5190
1da177e4
LT		 5191static int __init bonding_init(void)
5192{
1da177e4
LT		 5193 int i;
5194 int res;
0883beca 5195 struct bonding *bond;
1da177e4
LT		 5196
5197 printk(KERN_INFO "%s", version);
5198
dfe60397 5199 res = bond_check_params(&bonding_defaults);
1da177e4 5200 if (res) {
dfe60397 5201 goto out;
1da177e4
LT		 5202 }
5203
1da177e4
LT		 5204#ifdef CONFIG_PROC_FS
5205 bond_create_proc_dir();
5206#endif
027ea041
JV		 5207
5208 init_rwsem(&bonding_rwsem);
5209
1da177e4 5210 for (i = 0; i < max_bonds; i++) {
0dd646fe 5211 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5212 if (res)
5213 goto err;
1da177e4
LT		 5214 }
5215
b76cdba9
MW		 5216 res = bond_create_sysfs();
5217 if (res)
5218 goto err;
5219
1da177e4 5220 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5221 register_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5222 bond_register_ipv6_notifier();
1da177e4 5223
dfe60397
MW		 5224 goto out;
5225err:
0883beca 5226 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5227 bond_work_cancel_all(bond);
5228 destroy_workqueue(bond->wq);
5229 }
5230
ae68c398
PE		 5231 bond_destroy_sysfs();
5232
40abc270 5233 rtnl_lock();
1da177e4 5234 bond_free_all();
1da177e4 5235 rtnl_unlock();
dfe60397 5236out:
1da177e4 5237 return res;
dfe60397 5238
1da177e4
LT		 5239}
5240
5241static void __exit bonding_exit(void)
5242{
5243 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5244 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5245 bond_unregister_ipv6_notifier();
1da177e4 5246
ae68c398
PE		 5247 bond_destroy_sysfs();
5248
1da177e4
LT		 5249 rtnl_lock();
5250 bond_free_all();
5251 rtnl_unlock();
5252}
5253
5254module_init(bonding_init);
5255module_exit(bonding_exit);
5256MODULE_LICENSE("GPL");
5257MODULE_VERSION(DRV_VERSION);
5258MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5259MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5260MODULE_SUPPORTED_DEVICE("most ethernet devices");
5261
5262/*
5263 * Local variables:
5264 * c-indent-level: 8
5265 * c-basic-offset: 8
5266 * tab-width: 8
5267 * End:
5268 */
5269
Clone of davem's net-next-2.6 tree