]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/bonding/bond_main.c
bonding: Remove redundant argument from bond_create.
[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
34//#define BONDING_DEBUG 1
35
1da177e4
LT
36#include <linux/kernel.h>
37#include <linux/module.h>
1da177e4
LT
38#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
169a3e66 44#include <net/ip.h>
1da177e4 45#include <linux/ip.h>
169a3e66
JV
46#include <linux/tcp.h>
47#include <linux/udp.h>
1da177e4
LT
48#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
56#include <asm/system.h>
57#include <asm/io.h>
58#include <asm/dma.h>
59#include <asm/uaccess.h>
60#include <linux/errno.h>
61#include <linux/netdevice.h>
62#include <linux/inetdevice.h>
a816c7c7 63#include <linux/igmp.h>
1da177e4
LT
64#include <linux/etherdevice.h>
65#include <linux/skbuff.h>
66#include <net/sock.h>
67#include <linux/rtnetlink.h>
68#include <linux/proc_fs.h>
69#include <linux/seq_file.h>
70#include <linux/smp.h>
71#include <linux/if_ether.h>
72#include <net/arp.h>
73#include <linux/mii.h>
74#include <linux/ethtool.h>
75#include <linux/if_vlan.h>
76#include <linux/if_bonding.h>
b63bb739 77#include <linux/jiffies.h>
c3ade5ca 78#include <net/route.h>
457c4cbc 79#include <net/net_namespace.h>
1da177e4
LT
80#include "bonding.h"
81#include "bond_3ad.h"
82#include "bond_alb.h"
83
84/*---------------------------- Module parameters ----------------------------*/
85
86/* monitor all links that often (in milliseconds). <=0 disables monitoring */
87#define BOND_LINK_MON_INTERV 0
88#define BOND_LINK_ARP_INTERV 0
89
90static int max_bonds = BOND_DEFAULT_MAX_BONDS;
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;
169a3e66 98static char *xmit_hash_policy = NULL;
1da177e4
LT
99static int arp_interval = BOND_LINK_ARP_INTERV;
100static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
f5b2b966 101static char *arp_validate = NULL;
dd957c57 102static int fail_over_mac = 0;
12479f9a 103struct bond_params bonding_defaults;
1da177e4
LT
104
105module_param(max_bonds, int, 0);
106MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
107module_param(miimon, int, 0);
108MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
109module_param(updelay, int, 0);
110MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
111module_param(downdelay, int, 0);
2ac47660
MW
112MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
113 "in milliseconds");
1da177e4 114module_param(use_carrier, int, 0);
2ac47660
MW
115MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
116 "0 for off, 1 for on (default)");
1da177e4 117module_param(mode, charp, 0);
2ac47660
MW
118MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
119 "1 for active-backup, 2 for balance-xor, "
120 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
121 "6 for balance-alb");
1da177e4
LT
122module_param(primary, charp, 0);
123MODULE_PARM_DESC(primary, "Primary network device to use");
124module_param(lacp_rate, charp, 0);
2ac47660
MW
125MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
126 "(slow/fast)");
169a3e66 127module_param(xmit_hash_policy, charp, 0);
2ac47660
MW
128MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
129 ", 1 for layer 3+4");
1da177e4
LT
130module_param(arp_interval, int, 0);
131MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
132module_param_array(arp_ip_target, charp, NULL, 0);
133MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
f5b2b966
JV
134module_param(arp_validate, charp, 0);
135MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
dd957c57
JV
136module_param(fail_over_mac, int, 0);
137MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. 0 of off (default), 1 for on.");
1da177e4
LT
138
139/*----------------------------- Global variables ----------------------------*/
140
f71e1309 141static const char * const version =
1da177e4
LT
142 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
143
12479f9a 144LIST_HEAD(bond_dev_list);
1da177e4
LT
145
146#ifdef CONFIG_PROC_FS
147static struct proc_dir_entry *bond_proc_dir = NULL;
148#endif
149
b76cdba9 150extern struct rw_semaphore bonding_rwsem;
d3bb52b0 151static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
1da177e4 152static int arp_ip_count = 0;
1da177e4 153static int bond_mode = BOND_MODE_ROUNDROBIN;
169a3e66 154static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
1da177e4 155static int lacp_fast = 0;
217df670 156
1da177e4 157
12479f9a 158struct bond_parm_tbl bond_lacp_tbl[] = {
1da177e4
LT
159{ "slow", AD_LACP_SLOW},
160{ "fast", AD_LACP_FAST},
161{ NULL, -1},
162};
163
12479f9a 164struct bond_parm_tbl bond_mode_tbl[] = {
1da177e4
LT
165{ "balance-rr", BOND_MODE_ROUNDROBIN},
166{ "active-backup", BOND_MODE_ACTIVEBACKUP},
167{ "balance-xor", BOND_MODE_XOR},
168{ "broadcast", BOND_MODE_BROADCAST},
169{ "802.3ad", BOND_MODE_8023AD},
170{ "balance-tlb", BOND_MODE_TLB},
171{ "balance-alb", BOND_MODE_ALB},
172{ NULL, -1},
173};
174
12479f9a 175struct bond_parm_tbl xmit_hashtype_tbl[] = {
169a3e66
JV
176{ "layer2", BOND_XMIT_POLICY_LAYER2},
177{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
6f6652be 178{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
169a3e66
JV
179{ NULL, -1},
180};
181
f5b2b966
JV
182struct bond_parm_tbl arp_validate_tbl[] = {
183{ "none", BOND_ARP_VALIDATE_NONE},
184{ "active", BOND_ARP_VALIDATE_ACTIVE},
185{ "backup", BOND_ARP_VALIDATE_BACKUP},
186{ "all", BOND_ARP_VALIDATE_ALL},
187{ NULL, -1},
188};
189
1da177e4
LT
190/*-------------------------- Forward declarations ---------------------------*/
191
c3ade5ca 192static void bond_send_gratuitous_arp(struct bonding *bond);
c50b85d0 193static void bond_deinit(struct net_device *bond_dev);
1da177e4
LT
194
195/*---------------------------- General routines -----------------------------*/
196
4ad072c9 197static const char *bond_mode_name(int mode)
1da177e4
LT
198{
199 switch (mode) {
200 case BOND_MODE_ROUNDROBIN :
201 return "load balancing (round-robin)";
202 case BOND_MODE_ACTIVEBACKUP :
203 return "fault-tolerance (active-backup)";
204 case BOND_MODE_XOR :
205 return "load balancing (xor)";
206 case BOND_MODE_BROADCAST :
207 return "fault-tolerance (broadcast)";
208 case BOND_MODE_8023AD:
209 return "IEEE 802.3ad Dynamic link aggregation";
210 case BOND_MODE_TLB:
211 return "transmit load balancing";
212 case BOND_MODE_ALB:
213 return "adaptive load balancing";
214 default:
215 return "unknown";
216 }
217}
218
219/*---------------------------------- VLAN -----------------------------------*/
220
221/**
222 * bond_add_vlan - add a new vlan id on bond
223 * @bond: bond that got the notification
224 * @vlan_id: the vlan id to add
225 *
226 * Returns -ENOMEM if allocation failed.
227 */
228static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
229{
230 struct vlan_entry *vlan;
231
232 dprintk("bond: %s, vlan id %d\n",
233 (bond ? bond->dev->name: "None"), vlan_id);
234
235 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
236 if (!vlan) {
237 return -ENOMEM;
238 }
239
240 INIT_LIST_HEAD(&vlan->vlan_list);
241 vlan->vlan_id = vlan_id;
c3ade5ca 242 vlan->vlan_ip = 0;
1da177e4
LT
243
244 write_lock_bh(&bond->lock);
245
246 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
247
248 write_unlock_bh(&bond->lock);
249
250 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
251
252 return 0;
253}
254
255/**
256 * bond_del_vlan - delete a vlan id from bond
257 * @bond: bond that got the notification
258 * @vlan_id: the vlan id to delete
259 *
260 * returns -ENODEV if @vlan_id was not found in @bond.
261 */
262static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
263{
264 struct vlan_entry *vlan, *next;
265 int res = -ENODEV;
266
267 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
268
269 write_lock_bh(&bond->lock);
270
271 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
272 if (vlan->vlan_id == vlan_id) {
273 list_del(&vlan->vlan_list);
274
275 if ((bond->params.mode == BOND_MODE_TLB) ||
276 (bond->params.mode == BOND_MODE_ALB)) {
277 bond_alb_clear_vlan(bond, vlan_id);
278 }
279
280 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
281 bond->dev->name);
282
283 kfree(vlan);
284
285 if (list_empty(&bond->vlan_list) &&
286 (bond->slave_cnt == 0)) {
287 /* Last VLAN removed and no slaves, so
288 * restore block on adding VLANs. This will
289 * be removed once new slaves that are not
290 * VLAN challenged will be added.
291 */
292 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
293 }
294
295 res = 0;
296 goto out;
297 }
298 }
299
300 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
301 bond->dev->name);
302
303out:
304 write_unlock_bh(&bond->lock);
305 return res;
306}
307
308/**
309 * bond_has_challenged_slaves
310 * @bond: the bond we're working on
311 *
312 * Searches the slave list. Returns 1 if a vlan challenged slave
313 * was found, 0 otherwise.
314 *
315 * Assumes bond->lock is held.
316 */
317static int bond_has_challenged_slaves(struct bonding *bond)
318{
319 struct slave *slave;
320 int i;
321
322 bond_for_each_slave(bond, slave, i) {
323 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
324 dprintk("found VLAN challenged slave - %s\n",
325 slave->dev->name);
326 return 1;
327 }
328 }
329
330 dprintk("no VLAN challenged slaves found\n");
331 return 0;
332}
333
334/**
335 * bond_next_vlan - safely skip to the next item in the vlans list.
336 * @bond: the bond we're working on
337 * @curr: item we're advancing from
338 *
339 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
340 * or @curr->next otherwise (even if it is @curr itself again).
341 *
342 * Caller must hold bond->lock
343 */
344struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
345{
346 struct vlan_entry *next, *last;
347
348 if (list_empty(&bond->vlan_list)) {
349 return NULL;
350 }
351
352 if (!curr) {
353 next = list_entry(bond->vlan_list.next,
354 struct vlan_entry, vlan_list);
355 } else {
356 last = list_entry(bond->vlan_list.prev,
357 struct vlan_entry, vlan_list);
358 if (last == curr) {
359 next = list_entry(bond->vlan_list.next,
360 struct vlan_entry, vlan_list);
361 } else {
362 next = list_entry(curr->vlan_list.next,
363 struct vlan_entry, vlan_list);
364 }
365 }
366
367 return next;
368}
369
370/**
371 * bond_dev_queue_xmit - Prepare skb for xmit.
372 *
373 * @bond: bond device that got this skb for tx.
374 * @skb: hw accel VLAN tagged skb to transmit
375 * @slave_dev: slave that is supposed to xmit this skbuff
376 *
377 * When the bond gets an skb to transmit that is
378 * already hardware accelerated VLAN tagged, and it
379 * needs to relay this skb to a slave that is not
380 * hw accel capable, the skb needs to be "unaccelerated",
381 * i.e. strip the hwaccel tag and re-insert it as part
382 * of the payload.
383 */
384int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
385{
966bc6f4 386 unsigned short uninitialized_var(vlan_id);
1da177e4
LT
387
388 if (!list_empty(&bond->vlan_list) &&
389 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
390 vlan_get_tag(skb, &vlan_id) == 0) {
391 skb->dev = slave_dev;
392 skb = vlan_put_tag(skb, vlan_id);
393 if (!skb) {
394 /* vlan_put_tag() frees the skb in case of error,
395 * so return success here so the calling functions
396 * won't attempt to free is again.
397 */
398 return 0;
399 }
400 } else {
401 skb->dev = slave_dev;
402 }
403
404 skb->priority = 1;
405 dev_queue_xmit(skb);
406
407 return 0;
408}
409
410/*
411 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
412 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
413 * lock because:
414 * a. This operation is performed in IOCTL context,
415 * b. The operation is protected by the RTNL semaphore in the 8021q code,
416 * c. Holding a lock with BH disabled while directly calling a base driver
417 * entry point is generally a BAD idea.
418 *
419 * The design of synchronization/protection for this operation in the 8021q
420 * module is good for one or more VLAN devices over a single physical device
421 * and cannot be extended for a teaming solution like bonding, so there is a
422 * potential race condition here where a net device from the vlan group might
423 * be referenced (either by a base driver or the 8021q code) while it is being
424 * removed from the system. However, it turns out we're not making matters
425 * worse, and if it works for regular VLAN usage it will work here too.
426*/
427
428/**
429 * bond_vlan_rx_register - Propagates registration to slaves
430 * @bond_dev: bonding net device that got called
431 * @grp: vlan group being registered
432 */
433static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
434{
435 struct bonding *bond = bond_dev->priv;
436 struct slave *slave;
437 int i;
438
439 bond->vlgrp = grp;
440
441 bond_for_each_slave(bond, slave, i) {
442 struct net_device *slave_dev = slave->dev;
443
444 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
445 slave_dev->vlan_rx_register) {
446 slave_dev->vlan_rx_register(slave_dev, grp);
447 }
448 }
449}
450
451/**
452 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
453 * @bond_dev: bonding net device that got called
454 * @vid: vlan id being added
455 */
456static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
457{
458 struct bonding *bond = bond_dev->priv;
459 struct slave *slave;
460 int i, res;
461
462 bond_for_each_slave(bond, slave, i) {
463 struct net_device *slave_dev = slave->dev;
464
465 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
466 slave_dev->vlan_rx_add_vid) {
467 slave_dev->vlan_rx_add_vid(slave_dev, vid);
468 }
469 }
470
471 res = bond_add_vlan(bond, vid);
472 if (res) {
473 printk(KERN_ERR DRV_NAME
4e0952c7 474 ": %s: Error: Failed to add vlan id %d\n",
1da177e4
LT
475 bond_dev->name, vid);
476 }
477}
478
479/**
480 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
481 * @bond_dev: bonding net device that got called
482 * @vid: vlan id being removed
483 */
484static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
485{
486 struct bonding *bond = bond_dev->priv;
487 struct slave *slave;
488 struct net_device *vlan_dev;
489 int i, res;
490
491 bond_for_each_slave(bond, slave, i) {
492 struct net_device *slave_dev = slave->dev;
493
494 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
495 slave_dev->vlan_rx_kill_vid) {
496 /* Save and then restore vlan_dev in the grp array,
497 * since the slave's driver might clear it.
498 */
5c15bdec 499 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
1da177e4 500 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
5c15bdec 501 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
1da177e4
LT
502 }
503 }
504
505 res = bond_del_vlan(bond, vid);
506 if (res) {
507 printk(KERN_ERR DRV_NAME
4e0952c7 508 ": %s: Error: Failed to remove vlan id %d\n",
1da177e4
LT
509 bond_dev->name, vid);
510 }
511}
512
513static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
514{
515 struct vlan_entry *vlan;
516
517 write_lock_bh(&bond->lock);
518
519 if (list_empty(&bond->vlan_list)) {
520 goto out;
521 }
522
523 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
524 slave_dev->vlan_rx_register) {
525 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
526 }
527
528 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
529 !(slave_dev->vlan_rx_add_vid)) {
530 goto out;
531 }
532
533 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
534 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
535 }
536
537out:
538 write_unlock_bh(&bond->lock);
539}
540
541static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
542{
543 struct vlan_entry *vlan;
544 struct net_device *vlan_dev;
545
546 write_lock_bh(&bond->lock);
547
548 if (list_empty(&bond->vlan_list)) {
549 goto out;
550 }
551
552 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
553 !(slave_dev->vlan_rx_kill_vid)) {
554 goto unreg;
555 }
556
557 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
558 /* Save and then restore vlan_dev in the grp array,
559 * since the slave's driver might clear it.
560 */
5c15bdec 561 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
1da177e4 562 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
5c15bdec 563 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
1da177e4
LT
564 }
565
566unreg:
567 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
568 slave_dev->vlan_rx_register) {
569 slave_dev->vlan_rx_register(slave_dev, NULL);
570 }
571
572out:
573 write_unlock_bh(&bond->lock);
574}
575
576/*------------------------------- Link status -------------------------------*/
577
ff59c456
JV
578/*
579 * Set the carrier state for the master according to the state of its
580 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
581 * do special 802.3ad magic.
582 *
583 * Returns zero if carrier state does not change, nonzero if it does.
584 */
585static int bond_set_carrier(struct bonding *bond)
586{
587 struct slave *slave;
588 int i;
589
590 if (bond->slave_cnt == 0)
591 goto down;
592
593 if (bond->params.mode == BOND_MODE_8023AD)
594 return bond_3ad_set_carrier(bond);
595
596 bond_for_each_slave(bond, slave, i) {
597 if (slave->link == BOND_LINK_UP) {
598 if (!netif_carrier_ok(bond->dev)) {
599 netif_carrier_on(bond->dev);
600 return 1;
601 }
602 return 0;
603 }
604 }
605
606down:
607 if (netif_carrier_ok(bond->dev)) {
608 netif_carrier_off(bond->dev);
609 return 1;
610 }
611 return 0;
612}
613
1da177e4
LT
614/*
615 * Get link speed and duplex from the slave's base driver
616 * using ethtool. If for some reason the call fails or the
617 * values are invalid, fake speed and duplex to 100/Full
618 * and return error.
619 */
620static int bond_update_speed_duplex(struct slave *slave)
621{
622 struct net_device *slave_dev = slave->dev;
1da177e4 623 struct ethtool_cmd etool;
61a44b9c 624 int res;
1da177e4
LT
625
626 /* Fake speed and duplex */
627 slave->speed = SPEED_100;
628 slave->duplex = DUPLEX_FULL;
629
61a44b9c
MW
630 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
631 return -1;
1da177e4 632
61a44b9c
MW
633 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
634 if (res < 0)
1da177e4 635 return -1;
1da177e4 636
1da177e4
LT
637 switch (etool.speed) {
638 case SPEED_10:
639 case SPEED_100:
640 case SPEED_1000:
94dbffd5 641 case SPEED_10000:
1da177e4
LT
642 break;
643 default:
644 return -1;
645 }
646
647 switch (etool.duplex) {
648 case DUPLEX_FULL:
649 case DUPLEX_HALF:
650 break;
651 default:
652 return -1;
653 }
654
655 slave->speed = etool.speed;
656 slave->duplex = etool.duplex;
657
658 return 0;
659}
660
661/*
662 * if <dev> supports MII link status reporting, check its link status.
663 *
664 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
665 * depening upon the setting of the use_carrier parameter.
666 *
667 * Return either BMSR_LSTATUS, meaning that the link is up (or we
668 * can't tell and just pretend it is), or 0, meaning that the link is
669 * down.
670 *
671 * If reporting is non-zero, instead of faking link up, return -1 if
672 * both ETHTOOL and MII ioctls fail (meaning the device does not
673 * support them). If use_carrier is set, return whatever it says.
674 * It'd be nice if there was a good way to tell if a driver supports
675 * netif_carrier, but there really isn't.
676 */
677static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
678{
679 static int (* ioctl)(struct net_device *, struct ifreq *, int);
680 struct ifreq ifr;
681 struct mii_ioctl_data *mii;
1da177e4
LT
682
683 if (bond->params.use_carrier) {
684 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
685 }
686
687 ioctl = slave_dev->do_ioctl;
688 if (ioctl) {
689 /* TODO: set pointer to correct ioctl on a per team member */
690 /* bases to make this more efficient. that is, once */
691 /* we determine the correct ioctl, we will always */
692 /* call it and not the others for that team */
693 /* member. */
694
695 /*
696 * We cannot assume that SIOCGMIIPHY will also read a
697 * register; not all network drivers (e.g., e100)
698 * support that.
699 */
700
701 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
702 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
703 mii = if_mii(&ifr);
704 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
705 mii->reg_num = MII_BMSR;
706 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
707 return (mii->val_out & BMSR_LSTATUS);
708 }
709 }
710 }
711
61a44b9c
MW
712 /*
713 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
714 * attempt to get link status from it if the above MII ioctls fail.
715 */
1da177e4
LT
716 if (slave_dev->ethtool_ops) {
717 if (slave_dev->ethtool_ops->get_link) {
718 u32 link;
719
720 link = slave_dev->ethtool_ops->get_link(slave_dev);
721
722 return link ? BMSR_LSTATUS : 0;
723 }
724 }
725
1da177e4
LT
726 /*
727 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 728 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT
729 * cannot report link status). If not reporting, pretend
730 * we're ok.
731 */
732 return (reporting ? -1 : BMSR_LSTATUS);
733}
734
735/*----------------------------- Multicast list ------------------------------*/
736
737/*
738 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
739 */
740static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
741{
742 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
743 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
744}
745
746/*
747 * returns dmi entry if found, NULL otherwise
748 */
749static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
750{
751 struct dev_mc_list *idmi;
752
753 for (idmi = mc_list; idmi; idmi = idmi->next) {
754 if (bond_is_dmi_same(dmi, idmi)) {
755 return idmi;
756 }
757 }
758
759 return NULL;
760}
761
762/*
763 * Push the promiscuity flag down to appropriate slaves
764 */
765static void bond_set_promiscuity(struct bonding *bond, int inc)
766{
767 if (USES_PRIMARY(bond->params.mode)) {
768 /* write lock already acquired */
769 if (bond->curr_active_slave) {
770 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
771 }
772 } else {
773 struct slave *slave;
774 int i;
775 bond_for_each_slave(bond, slave, i) {
776 dev_set_promiscuity(slave->dev, inc);
777 }
778 }
779}
780
781/*
782 * Push the allmulti flag down to all slaves
783 */
784static void bond_set_allmulti(struct bonding *bond, int inc)
785{
786 if (USES_PRIMARY(bond->params.mode)) {
787 /* write lock already acquired */
788 if (bond->curr_active_slave) {
789 dev_set_allmulti(bond->curr_active_slave->dev, inc);
790 }
791 } else {
792 struct slave *slave;
793 int i;
794 bond_for_each_slave(bond, slave, i) {
795 dev_set_allmulti(slave->dev, inc);
796 }
797 }
798}
799
800/*
801 * Add a Multicast address to slaves
802 * according to mode
803 */
804static void bond_mc_add(struct bonding *bond, void *addr, int alen)
805{
806 if (USES_PRIMARY(bond->params.mode)) {
807 /* write lock already acquired */
808 if (bond->curr_active_slave) {
809 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
810 }
811 } else {
812 struct slave *slave;
813 int i;
814 bond_for_each_slave(bond, slave, i) {
815 dev_mc_add(slave->dev, addr, alen, 0);
816 }
817 }
818}
819
820/*
821 * Remove a multicast address from slave
822 * according to mode
823 */
824static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
825{
826 if (USES_PRIMARY(bond->params.mode)) {
827 /* write lock already acquired */
828 if (bond->curr_active_slave) {
829 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
830 }
831 } else {
832 struct slave *slave;
833 int i;
834 bond_for_each_slave(bond, slave, i) {
835 dev_mc_delete(slave->dev, addr, alen, 0);
836 }
837 }
838}
839
a816c7c7
JV
840
841/*
842 * Retrieve the list of registered multicast addresses for the bonding
843 * device and retransmit an IGMP JOIN request to the current active
844 * slave.
845 */
846static void bond_resend_igmp_join_requests(struct bonding *bond)
847{
848 struct in_device *in_dev;
849 struct ip_mc_list *im;
850
851 rcu_read_lock();
852 in_dev = __in_dev_get_rcu(bond->dev);
853 if (in_dev) {
854 for (im = in_dev->mc_list; im; im = im->next) {
855 ip_mc_rejoin_group(im);
856 }
857 }
858
859 rcu_read_unlock();
860}
861
1da177e4
LT
862/*
863 * Totally destroys the mc_list in bond
864 */
865static void bond_mc_list_destroy(struct bonding *bond)
866{
867 struct dev_mc_list *dmi;
868
869 dmi = bond->mc_list;
870 while (dmi) {
871 bond->mc_list = dmi->next;
872 kfree(dmi);
873 dmi = bond->mc_list;
874 }
a816c7c7 875 bond->mc_list = NULL;
1da177e4
LT
876}
877
878/*
879 * Copy all the Multicast addresses from src to the bonding device dst
880 */
de54f390 881static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 882 gfp_t gfp_flag)
1da177e4
LT
883{
884 struct dev_mc_list *dmi, *new_dmi;
885
886 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 887 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT
888
889 if (!new_dmi) {
890 /* FIXME: Potential memory leak !!! */
891 return -ENOMEM;
892 }
893
894 new_dmi->next = bond->mc_list;
895 bond->mc_list = new_dmi;
896 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
897 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
898 new_dmi->dmi_users = dmi->dmi_users;
899 new_dmi->dmi_gusers = dmi->dmi_gusers;
900 }
901
902 return 0;
903}
904
905/*
906 * flush all members of flush->mc_list from device dev->mc_list
907 */
908static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
909{
910 struct bonding *bond = bond_dev->priv;
911 struct dev_mc_list *dmi;
912
913 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
914 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
915 }
916
917 if (bond->params.mode == BOND_MODE_8023AD) {
918 /* del lacpdu mc addr from mc list */
919 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
920
921 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
922 }
923}
924
925/*--------------------------- Active slave change ---------------------------*/
926
927/*
928 * Update the mc list and multicast-related flags for the new and
929 * old active slaves (if any) according to the multicast mode, and
930 * promiscuous flags unconditionally.
931 */
932static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
933{
934 struct dev_mc_list *dmi;
935
936 if (!USES_PRIMARY(bond->params.mode)) {
937 /* nothing to do - mc list is already up-to-date on
938 * all slaves
939 */
940 return;
941 }
942
943 if (old_active) {
944 if (bond->dev->flags & IFF_PROMISC) {
945 dev_set_promiscuity(old_active->dev, -1);
946 }
947
948 if (bond->dev->flags & IFF_ALLMULTI) {
949 dev_set_allmulti(old_active->dev, -1);
950 }
951
952 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
953 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
954 }
955 }
956
957 if (new_active) {
958 if (bond->dev->flags & IFF_PROMISC) {
959 dev_set_promiscuity(new_active->dev, 1);
960 }
961
962 if (bond->dev->flags & IFF_ALLMULTI) {
963 dev_set_allmulti(new_active->dev, 1);
964 }
965
966 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
967 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
968 }
a816c7c7 969 bond_resend_igmp_join_requests(bond);
1da177e4
LT
970 }
971}
972
973/**
974 * find_best_interface - select the best available slave to be the active one
975 * @bond: our bonding struct
976 *
977 * Warning: Caller must hold curr_slave_lock for writing.
978 */
979static struct slave *bond_find_best_slave(struct bonding *bond)
980{
981 struct slave *new_active, *old_active;
982 struct slave *bestslave = NULL;
983 int mintime = bond->params.updelay;
984 int i;
985
986 new_active = old_active = bond->curr_active_slave;
987
988 if (!new_active) { /* there were no active slaves left */
989 if (bond->slave_cnt > 0) { /* found one slave */
990 new_active = bond->first_slave;
991 } else {
992 return NULL; /* still no slave, return NULL */
993 }
994 }
995
996 /* first try the primary link; if arping, a link must tx/rx traffic
997 * before it can be considered the curr_active_slave - also, we would skip
998 * slaves between the curr_active_slave and primary_slave that may be up
999 * and able to arp
1000 */
1001 if ((bond->primary_slave) &&
1002 (!bond->params.arp_interval) &&
1003 (IS_UP(bond->primary_slave->dev))) {
1004 new_active = bond->primary_slave;
1005 }
1006
1007 /* remember where to stop iterating over the slaves */
1008 old_active = new_active;
1009
1010 bond_for_each_slave_from(bond, new_active, i, old_active) {
1011 if (IS_UP(new_active->dev)) {
1012 if (new_active->link == BOND_LINK_UP) {
1013 return new_active;
1014 } else if (new_active->link == BOND_LINK_BACK) {
1015 /* link up, but waiting for stabilization */
1016 if (new_active->delay < mintime) {
1017 mintime = new_active->delay;
1018 bestslave = new_active;
1019 }
1020 }
1021 }
1022 }
1023
1024 return bestslave;
1025}
1026
1027/**
1028 * change_active_interface - change the active slave into the specified one
1029 * @bond: our bonding struct
1030 * @new: the new slave to make the active one
1031 *
1032 * Set the new slave to the bond's settings and unset them on the old
1033 * curr_active_slave.
1034 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1035 *
1036 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1037 * because it is apparently the best available slave we have, even though its
1038 * updelay hasn't timed out yet.
1039 *
1040 * Warning: Caller must hold curr_slave_lock for writing.
1041 */
a77b5325 1042void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT
1043{
1044 struct slave *old_active = bond->curr_active_slave;
1045
1046 if (old_active == new_active) {
1047 return;
1048 }
1049
1050 if (new_active) {
1051 if (new_active->link == BOND_LINK_BACK) {
1052 if (USES_PRIMARY(bond->params.mode)) {
1053 printk(KERN_INFO DRV_NAME
1054 ": %s: making interface %s the new "
1055 "active one %d ms earlier.\n",
1056 bond->dev->name, new_active->dev->name,
1057 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1058 }
1059
1060 new_active->delay = 0;
1061 new_active->link = BOND_LINK_UP;
1062 new_active->jiffies = jiffies;
1063
1064 if (bond->params.mode == BOND_MODE_8023AD) {
1065 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1066 }
1067
1068 if ((bond->params.mode == BOND_MODE_TLB) ||
1069 (bond->params.mode == BOND_MODE_ALB)) {
1070 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1071 }
1072 } else {
1073 if (USES_PRIMARY(bond->params.mode)) {
1074 printk(KERN_INFO DRV_NAME
1075 ": %s: making interface %s the new "
1076 "active one.\n",
1077 bond->dev->name, new_active->dev->name);
1078 }
1079 }
1080 }
1081
1da177e4
LT
1082 if (USES_PRIMARY(bond->params.mode)) {
1083 bond_mc_swap(bond, new_active, old_active);
1084 }
1085
1086 if ((bond->params.mode == BOND_MODE_TLB) ||
1087 (bond->params.mode == BOND_MODE_ALB)) {
1088 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV
1089 if (old_active)
1090 bond_set_slave_inactive_flags(old_active);
1091 if (new_active)
1092 bond_set_slave_active_flags(new_active);
1da177e4
LT
1093 } else {
1094 bond->curr_active_slave = new_active;
1095 }
c3ade5ca
JV
1096
1097 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1098 if (old_active) {
1099 bond_set_slave_inactive_flags(old_active);
1100 }
1101
1102 if (new_active) {
1103 bond_set_slave_active_flags(new_active);
1104 }
2ab82852
MS
1105
1106 /* when bonding does not set the slave MAC address, the bond MAC
1107 * address is the one of the active slave.
1108 */
dd957c57 1109 if (new_active && bond->params.fail_over_mac)
2ab82852
MS
1110 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1111 new_active->dev->addr_len);
1053f62c
MS
1112 if (bond->curr_active_slave &&
1113 test_bit(__LINK_STATE_LINKWATCH_PENDING,
1114 &bond->curr_active_slave->dev->state)) {
1115 dprintk("delaying gratuitous arp on %s\n",
1116 bond->curr_active_slave->dev->name);
1117 bond->send_grat_arp = 1;
1118 } else
1119 bond_send_gratuitous_arp(bond);
c3ade5ca 1120 }
1da177e4
LT
1121}
1122
1123/**
1124 * bond_select_active_slave - select a new active slave, if needed
1125 * @bond: our bonding struct
1126 *
1127 * This functions shoud be called when one of the following occurs:
1128 * - The old curr_active_slave has been released or lost its link.
1129 * - The primary_slave has got its link back.
1130 * - A slave has got its link back and there's no old curr_active_slave.
1131 *
1132 * Warning: Caller must hold curr_slave_lock for writing.
1133 */
a77b5325 1134void bond_select_active_slave(struct bonding *bond)
1da177e4
LT
1135{
1136 struct slave *best_slave;
ff59c456 1137 int rv;
1da177e4
LT
1138
1139 best_slave = bond_find_best_slave(bond);
1140 if (best_slave != bond->curr_active_slave) {
1141 bond_change_active_slave(bond, best_slave);
ff59c456
JV
1142 rv = bond_set_carrier(bond);
1143 if (!rv)
1144 return;
1145
1146 if (netif_carrier_ok(bond->dev)) {
1147 printk(KERN_INFO DRV_NAME
1148 ": %s: first active interface up!\n",
1149 bond->dev->name);
1150 } else {
1151 printk(KERN_INFO DRV_NAME ": %s: "
1152 "now running without any active interface !\n",
1153 bond->dev->name);
1154 }
1da177e4
LT
1155 }
1156}
1157
1158/*--------------------------- slave list handling ---------------------------*/
1159
1160/*
1161 * This function attaches the slave to the end of list.
1162 *
1163 * bond->lock held for writing by caller.
1164 */
1165static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1166{
1167 if (bond->first_slave == NULL) { /* attaching the first slave */
1168 new_slave->next = new_slave;
1169 new_slave->prev = new_slave;
1170 bond->first_slave = new_slave;
1171 } else {
1172 new_slave->next = bond->first_slave;
1173 new_slave->prev = bond->first_slave->prev;
1174 new_slave->next->prev = new_slave;
1175 new_slave->prev->next = new_slave;
1176 }
1177
1178 bond->slave_cnt++;
1179}
1180
1181/*
1182 * This function detaches the slave from the list.
1183 * WARNING: no check is made to verify if the slave effectively
1184 * belongs to <bond>.
1185 * Nothing is freed on return, structures are just unchained.
1186 * If any slave pointer in bond was pointing to <slave>,
1187 * it should be changed by the calling function.
1188 *
1189 * bond->lock held for writing by caller.
1190 */
1191static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1192{
1193 if (slave->next) {
1194 slave->next->prev = slave->prev;
1195 }
1196
1197 if (slave->prev) {
1198 slave->prev->next = slave->next;
1199 }
1200
1201 if (bond->first_slave == slave) { /* slave is the first slave */
1202 if (bond->slave_cnt > 1) { /* there are more slave */
1203 bond->first_slave = slave->next;
1204 } else {
1205 bond->first_slave = NULL; /* slave was the last one */
1206 }
1207 }
1208
1209 slave->next = NULL;
1210 slave->prev = NULL;
1211 bond->slave_cnt--;
1212}
1213
1214/*---------------------------------- IOCTL ----------------------------------*/
1215
4ad072c9
AB
1216static int bond_sethwaddr(struct net_device *bond_dev,
1217 struct net_device *slave_dev)
1da177e4
LT
1218{
1219 dprintk("bond_dev=%p\n", bond_dev);
1220 dprintk("slave_dev=%p\n", slave_dev);
1221 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1222 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1223 return 0;
1224}
1225
7f353bf2
HX
1226#define BOND_VLAN_FEATURES \
1227 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1228 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK
1229
1230/*
8e3babcd 1231 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX
1232 * feature bits are managed elsewhere, so preserve those feature bits
1233 * on the master device.
8531c5ff
AK
1234 */
1235static int bond_compute_features(struct bonding *bond)
1236{
8531c5ff
AK
1237 struct slave *slave;
1238 struct net_device *bond_dev = bond->dev;
7f353bf2 1239 unsigned long features = bond_dev->features;
3158bf7d
MS
1240 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1241 bond_dev->hard_header_len);
8e3babcd 1242 int i;
8531c5ff 1243
7f353bf2
HX
1244 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1245 features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1246 NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1247
54ef3137 1248 bond_for_each_slave(bond, slave, i) {
7f353bf2
HX
1249 features = netdev_compute_features(features,
1250 slave->dev->features);
54ef3137
JV
1251 if (slave->dev->hard_header_len > max_hard_header_len)
1252 max_hard_header_len = slave->dev->hard_header_len;
1253 }
8531c5ff 1254
7f353bf2 1255 features |= (bond_dev->features & BOND_VLAN_FEATURES);
8531c5ff 1256 bond_dev->features = features;
54ef3137 1257 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK
1258
1259 return 0;
1260}
1261
872254dd
MS
1262
1263static void bond_setup_by_slave(struct net_device *bond_dev,
1264 struct net_device *slave_dev)
1265{
d90a162a
MS
1266 struct bonding *bond = bond_dev->priv;
1267
872254dd 1268 bond_dev->neigh_setup = slave_dev->neigh_setup;
1284cd3a 1269 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS
1270
1271 bond_dev->type = slave_dev->type;
1272 bond_dev->hard_header_len = slave_dev->hard_header_len;
1273 bond_dev->addr_len = slave_dev->addr_len;
1274
1275 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1276 slave_dev->addr_len);
d90a162a 1277 bond->setup_by_slave = 1;
872254dd
MS
1278}
1279
1da177e4 1280/* enslave device <slave> to bond device <master> */
a77b5325 1281int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT
1282{
1283 struct bonding *bond = bond_dev->priv;
1284 struct slave *new_slave = NULL;
1285 struct dev_mc_list *dmi;
1286 struct sockaddr addr;
1287 int link_reporting;
1288 int old_features = bond_dev->features;
1289 int res = 0;
1290
552709d5 1291 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1292 slave_dev->do_ioctl == NULL) {
1da177e4 1293 printk(KERN_WARNING DRV_NAME
4e0952c7
MW
1294 ": %s: Warning: no link monitoring support for %s\n",
1295 bond_dev->name, slave_dev->name);
1da177e4
LT
1296 }
1297
1298 /* bond must be initialized by bond_open() before enslaving */
1299 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS
1300 printk(KERN_WARNING DRV_NAME
1301 " %s: master_dev is not up in bond_enslave\n",
1302 bond_dev->name);
1da177e4
LT
1303 }
1304
1305 /* already enslaved */
1306 if (slave_dev->flags & IFF_SLAVE) {
1307 dprintk("Error, Device was already enslaved\n");
1308 return -EBUSY;
1309 }
1310
1311 /* vlan challenged mutual exclusion */
1312 /* no need to lock since we're protected by rtnl_lock */
1313 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1314 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1315 if (!list_empty(&bond->vlan_list)) {
1316 printk(KERN_ERR DRV_NAME
4e0952c7 1317 ": %s: Error: cannot enslave VLAN "
1da177e4 1318 "challenged slave %s on VLAN enabled "
4e0952c7 1319 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT
1320 bond_dev->name);
1321 return -EPERM;
1322 } else {
1323 printk(KERN_WARNING DRV_NAME
4e0952c7 1324 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT
1325 "slave %s. Adding VLANs will be blocked as "
1326 "long as %s is part of bond %s\n",
4e0952c7 1327 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT
1328 bond_dev->name);
1329 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1330 }
1331 } else {
1332 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1333 if (bond->slave_cnt == 0) {
1334 /* First slave, and it is not VLAN challenged,
1335 * so remove the block of adding VLANs over the bond.
1336 */
1337 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1338 }
1339 }
1340
217df670
JV
1341 /*
1342 * Old ifenslave binaries are no longer supported. These can
1343 * be identified with moderate accurary by the state of the slave:
1344 * the current ifenslave will set the interface down prior to
1345 * enslaving it; the old ifenslave will not.
1346 */
1347 if ((slave_dev->flags & IFF_UP)) {
1348 printk(KERN_ERR DRV_NAME ": %s is up. "
1349 "This may be due to an out of date ifenslave.\n",
1350 slave_dev->name);
1351 res = -EPERM;
1352 goto err_undo_flags;
1353 }
1da177e4 1354
872254dd
MS
1355 /* set bonding device ether type by slave - bonding netdevices are
1356 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1357 * there is a need to override some of the type dependent attribs/funcs.
1358 *
1359 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1360 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1361 */
1362 if (bond->slave_cnt == 0) {
1363 if (slave_dev->type != ARPHRD_ETHER)
1364 bond_setup_by_slave(bond_dev, slave_dev);
1365 } else if (bond_dev->type != slave_dev->type) {
1366 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1367 "from other slaves (%d), can not enslave it.\n",
1368 slave_dev->name,
1369 slave_dev->type, bond_dev->type);
1370 res = -EINVAL;
1371 goto err_undo_flags;
1372 }
1373
217df670 1374 if (slave_dev->set_mac_address == NULL) {
2ab82852
MS
1375 if (bond->slave_cnt == 0) {
1376 printk(KERN_WARNING DRV_NAME
dd957c57
JV
1377 ": %s: Warning: The first slave device "
1378 "specified does not support setting the MAC "
1379 "address. Enabling the fail_over_mac option.",
1380 bond_dev->name);
1381 bond->params.fail_over_mac = 1;
1382 } else if (!bond->params.fail_over_mac) {
2ab82852 1383 printk(KERN_ERR DRV_NAME
dd957c57
JV
1384 ": %s: Error: The slave device specified "
1385 "does not support setting the MAC address, "
1386 "but fail_over_mac is not enabled.\n"
2ab82852
MS
1387 , bond_dev->name);
1388 res = -EOPNOTSUPP;
1389 goto err_undo_flags;
1390 }
1da177e4
LT
1391 }
1392
243cb4e5 1393 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT
1394 if (!new_slave) {
1395 res = -ENOMEM;
1396 goto err_undo_flags;
1397 }
1398
1da177e4
LT
1399 /* save slave's original flags before calling
1400 * netdev_set_master and dev_open
1401 */
1402 new_slave->original_flags = slave_dev->flags;
1403
217df670
JV
1404 /*
1405 * Save slave's original ("permanent") mac address for modes
1406 * that need it, and for restoring it upon release, and then
1407 * set it to the master's address
1408 */
1409 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1410
dd957c57 1411 if (!bond->params.fail_over_mac) {
2ab82852
MS
1412 /*
1413 * Set slave to master's mac address. The application already
1414 * set the master's mac address to that of the first slave
1415 */
1416 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1417 addr.sa_family = slave_dev->type;
1418 res = dev_set_mac_address(slave_dev, &addr);
1419 if (res) {
1420 dprintk("Error %d calling set_mac_address\n", res);
1421 goto err_free;
1422 }
217df670 1423 }
1da177e4 1424
c2edacf8
JV
1425 res = netdev_set_master(slave_dev, bond_dev);
1426 if (res) {
1427 dprintk("Error %d calling netdev_set_master\n", res);
569f0c4d 1428 goto err_restore_mac;
c2edacf8 1429 }
217df670
JV
1430 /* open the slave since the application closed it */
1431 res = dev_open(slave_dev);
1432 if (res) {
1433 dprintk("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1434 goto err_unset_master;
1da177e4
LT
1435 }
1436
1da177e4 1437 new_slave->dev = slave_dev;
0b680e75 1438 slave_dev->priv_flags |= IFF_BONDING;
1da177e4
LT
1439
1440 if ((bond->params.mode == BOND_MODE_TLB) ||
1441 (bond->params.mode == BOND_MODE_ALB)) {
1442 /* bond_alb_init_slave() must be called before all other stages since
1443 * it might fail and we do not want to have to undo everything
1444 */
1445 res = bond_alb_init_slave(bond, new_slave);
1446 if (res) {
569f0c4d 1447 goto err_close;
1da177e4
LT
1448 }
1449 }
1450
1451 /* If the mode USES_PRIMARY, then the new slave gets the
1452 * master's promisc (and mc) settings only if it becomes the
1453 * curr_active_slave, and that is taken care of later when calling
1454 * bond_change_active()
1455 */
1456 if (!USES_PRIMARY(bond->params.mode)) {
1457 /* set promiscuity level to new slave */
1458 if (bond_dev->flags & IFF_PROMISC) {
1459 dev_set_promiscuity(slave_dev, 1);
1460 }
1461
1462 /* set allmulti level to new slave */
1463 if (bond_dev->flags & IFF_ALLMULTI) {
1464 dev_set_allmulti(slave_dev, 1);
1465 }
1466
80ee5ad2 1467 netif_tx_lock_bh(bond_dev);
1da177e4
LT
1468 /* upload master's mc_list to new slave */
1469 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1470 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1471 }
80ee5ad2 1472 netif_tx_unlock_bh(bond_dev);
1da177e4
LT
1473 }
1474
1475 if (bond->params.mode == BOND_MODE_8023AD) {
1476 /* add lacpdu mc addr to mc list */
1477 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1478
1479 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1480 }
1481
1482 bond_add_vlans_on_slave(bond, slave_dev);
1483
1484 write_lock_bh(&bond->lock);
1485
1486 bond_attach_slave(bond, new_slave);
1487
1488 new_slave->delay = 0;
1489 new_slave->link_failure_count = 0;
1490
8531c5ff
AK
1491 bond_compute_features(bond);
1492
f5b2b966
JV
1493 new_slave->last_arp_rx = jiffies;
1494
1da177e4
LT
1495 if (bond->params.miimon && !bond->params.use_carrier) {
1496 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1497
1498 if ((link_reporting == -1) && !bond->params.arp_interval) {
1499 /*
1500 * miimon is set but a bonded network driver
1501 * does not support ETHTOOL/MII and
1502 * arp_interval is not set. Note: if
1503 * use_carrier is enabled, we will never go
1504 * here (because netif_carrier is always
1505 * supported); thus, we don't need to change
1506 * the messages for netif_carrier.
1507 */
1508 printk(KERN_WARNING DRV_NAME
4e0952c7 1509 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT
1510 "available for interface %s, and "
1511 "arp_interval/arp_ip_target module parameters "
1512 "not specified, thus bonding will not detect "
1513 "link failures! see bonding.txt for details.\n",
4e0952c7 1514 bond_dev->name, slave_dev->name);
1da177e4
LT
1515 } else if (link_reporting == -1) {
1516 /* unable get link status using mii/ethtool */
1517 printk(KERN_WARNING DRV_NAME
4e0952c7 1518 ": %s: Warning: can't get link status from "
1da177e4
LT
1519 "interface %s; the network driver associated "
1520 "with this interface does not support MII or "
1521 "ETHTOOL link status reporting, thus miimon "
1522 "has no effect on this interface.\n",
4e0952c7 1523 bond_dev->name, slave_dev->name);
1da177e4
LT
1524 }
1525 }
1526
1527 /* check for initial state */
1528 if (!bond->params.miimon ||
1529 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1530 if (bond->params.updelay) {
1531 dprintk("Initial state of slave_dev is "
1532 "BOND_LINK_BACK\n");
1533 new_slave->link = BOND_LINK_BACK;
1534 new_slave->delay = bond->params.updelay;
1535 } else {
1536 dprintk("Initial state of slave_dev is "
1537 "BOND_LINK_UP\n");
1538 new_slave->link = BOND_LINK_UP;
1539 }
1540 new_slave->jiffies = jiffies;
1541 } else {
1542 dprintk("Initial state of slave_dev is "
1543 "BOND_LINK_DOWN\n");
1544 new_slave->link = BOND_LINK_DOWN;
1545 }
1546
1547 if (bond_update_speed_duplex(new_slave) &&
1548 (new_slave->link != BOND_LINK_DOWN)) {
1549 printk(KERN_WARNING DRV_NAME
4e0952c7 1550 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1551 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1552 bond_dev->name, new_slave->dev->name);
1da177e4
LT
1553
1554 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW
1555 printk(KERN_WARNING DRV_NAME
1556 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1557 "support in base driver for proper aggregator "
4e0952c7 1558 "selection.\n", bond_dev->name);
1da177e4
LT
1559 }
1560 }
1561
1562 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1563 /* if there is a primary slave, remember it */
1564 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1565 bond->primary_slave = new_slave;
1566 }
1567 }
1568
1569 switch (bond->params.mode) {
1570 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV
1571 bond_set_slave_inactive_flags(new_slave);
1572 bond_select_active_slave(bond);
1da177e4
LT
1573 break;
1574 case BOND_MODE_8023AD:
1575 /* in 802.3ad mode, the internal mechanism
1576 * will activate the slaves in the selected
1577 * aggregator
1578 */
1579 bond_set_slave_inactive_flags(new_slave);
1580 /* if this is the first slave */
1581 if (bond->slave_cnt == 1) {
1582 SLAVE_AD_INFO(new_slave).id = 1;
1583 /* Initialize AD with the number of times that the AD timer is called in 1 second
1584 * can be called only after the mac address of the bond is set
1585 */
1586 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1587 bond->params.lacp_fast);
1588 } else {
1589 SLAVE_AD_INFO(new_slave).id =
1590 SLAVE_AD_INFO(new_slave->prev).id + 1;
1591 }
1592
1593 bond_3ad_bind_slave(new_slave);
1594 break;
1595 case BOND_MODE_TLB:
1596 case BOND_MODE_ALB:
1597 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1598 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT
1599 break;
1600 default:
1601 dprintk("This slave is always active in trunk mode\n");
1602
1603 /* always active in trunk mode */
1604 new_slave->state = BOND_STATE_ACTIVE;
1605
1606 /* In trunking mode there is little meaning to curr_active_slave
1607 * anyway (it holds no special properties of the bond device),
1608 * so we can change it without calling change_active_interface()
1609 */
1610 if (!bond->curr_active_slave) {
1611 bond->curr_active_slave = new_slave;
1612 }
1613 break;
1614 } /* switch(bond_mode) */
1615
ff59c456
JV
1616 bond_set_carrier(bond);
1617
1da177e4
LT
1618 write_unlock_bh(&bond->lock);
1619
b76cdba9
MW
1620 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1621 if (res)
569f0c4d 1622 goto err_close;
b76cdba9 1623
1da177e4
LT
1624 printk(KERN_INFO DRV_NAME
1625 ": %s: enslaving %s as a%s interface with a%s link.\n",
1626 bond_dev->name, slave_dev->name,
1627 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1628 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1629
1630 /* enslave is successful */
1631 return 0;
1632
1633/* Undo stages on error */
1da177e4
LT
1634err_close:
1635 dev_close(slave_dev);
1636
569f0c4d
JV
1637err_unset_master:
1638 netdev_set_master(slave_dev, NULL);
1639
1da177e4 1640err_restore_mac:
dd957c57 1641 if (!bond->params.fail_over_mac) {
2ab82852
MS
1642 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1643 addr.sa_family = slave_dev->type;
1644 dev_set_mac_address(slave_dev, &addr);
1645 }
1da177e4
LT
1646
1647err_free:
1648 kfree(new_slave);
1649
1650err_undo_flags:
1651 bond_dev->features = old_features;
8531c5ff 1652
1da177e4
LT
1653 return res;
1654}
1655
1656/*
1657 * Try to release the slave device <slave> from the bond device <master>
1658 * It is legal to access curr_active_slave without a lock because all the function
1659 * is write-locked.
1660 *
1661 * The rules for slave state should be:
1662 * for Active/Backup:
1663 * Active stays on all backups go down
1664 * for Bonded connections:
1665 * The first up interface should be left on and all others downed.
1666 */
a77b5325 1667int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT
1668{
1669 struct bonding *bond = bond_dev->priv;
1670 struct slave *slave, *oldcurrent;
1671 struct sockaddr addr;
1672 int mac_addr_differ;
0795af57 1673 DECLARE_MAC_BUF(mac);
1da177e4
LT
1674
1675 /* slave is not a slave or master is not master of this slave */
1676 if (!(slave_dev->flags & IFF_SLAVE) ||
1677 (slave_dev->master != bond_dev)) {
1678 printk(KERN_ERR DRV_NAME
4e0952c7 1679 ": %s: Error: cannot release %s.\n",
1da177e4
LT
1680 bond_dev->name, slave_dev->name);
1681 return -EINVAL;
1682 }
1683
1684 write_lock_bh(&bond->lock);
1685
1686 slave = bond_get_slave_by_dev(bond, slave_dev);
1687 if (!slave) {
1688 /* not a slave of this bond */
1689 printk(KERN_INFO DRV_NAME
1690 ": %s: %s not enslaved\n",
1691 bond_dev->name, slave_dev->name);
f5e2a7b2 1692 write_unlock_bh(&bond->lock);
1da177e4
LT
1693 return -EINVAL;
1694 }
1695
1696 mac_addr_differ = memcmp(bond_dev->dev_addr,
1697 slave->perm_hwaddr,
1698 ETH_ALEN);
1699 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1700 printk(KERN_WARNING DRV_NAME
0795af57
JP
1701 ": %s: Warning: the permanent HWaddr of %s - "
1702 "%s - is still in use by %s. "
1703 "Set the HWaddr of %s to a different address "
1704 "to avoid conflicts.\n",
4e0952c7 1705 bond_dev->name,
1da177e4 1706 slave_dev->name,
0795af57 1707 print_mac(mac, slave->perm_hwaddr),
1da177e4
LT
1708 bond_dev->name,
1709 slave_dev->name);
1710 }
1711
1712 /* Inform AD package of unbinding of slave. */
1713 if (bond->params.mode == BOND_MODE_8023AD) {
1714 /* must be called before the slave is
1715 * detached from the list
1716 */
1717 bond_3ad_unbind_slave(slave);
1718 }
1719
1720 printk(KERN_INFO DRV_NAME
1721 ": %s: releasing %s interface %s\n",
1722 bond_dev->name,
1723 (slave->state == BOND_STATE_ACTIVE)
1724 ? "active" : "backup",
1725 slave_dev->name);
1726
1727 oldcurrent = bond->curr_active_slave;
1728
1729 bond->current_arp_slave = NULL;
1730
1731 /* release the slave from its bond */
1732 bond_detach_slave(bond, slave);
1733
8531c5ff
AK
1734 bond_compute_features(bond);
1735
1da177e4
LT
1736 if (bond->primary_slave == slave) {
1737 bond->primary_slave = NULL;
1738 }
1739
1740 if (oldcurrent == slave) {
1741 bond_change_active_slave(bond, NULL);
1742 }
1743
1744 if ((bond->params.mode == BOND_MODE_TLB) ||
1745 (bond->params.mode == BOND_MODE_ALB)) {
1746 /* Must be called only after the slave has been
1747 * detached from the list and the curr_active_slave
1748 * has been cleared (if our_slave == old_current),
1749 * but before a new active slave is selected.
1750 */
2543331d 1751 write_unlock_bh(&bond->lock);
1da177e4 1752 bond_alb_deinit_slave(bond, slave);
2543331d 1753 write_lock_bh(&bond->lock);
1da177e4
LT
1754 }
1755
059fe7a5
JV
1756 if (oldcurrent == slave) {
1757 /*
1758 * Note that we hold RTNL over this sequence, so there
1759 * is no concern that another slave add/remove event
1760 * will interfere.
1761 */
1762 write_unlock_bh(&bond->lock);
1763 read_lock(&bond->lock);
1764 write_lock_bh(&bond->curr_slave_lock);
1765
1da177e4
LT
1766 bond_select_active_slave(bond);
1767
059fe7a5
JV
1768 write_unlock_bh(&bond->curr_slave_lock);
1769 read_unlock(&bond->lock);
1770 write_lock_bh(&bond->lock);
1771 }
1772
1da177e4 1773 if (bond->slave_cnt == 0) {
ff59c456
JV
1774 bond_set_carrier(bond);
1775
1da177e4
LT
1776 /* if the last slave was removed, zero the mac address
1777 * of the master so it will be set by the application
1778 * to the mac address of the first slave
1779 */
1780 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1781
1782 if (list_empty(&bond->vlan_list)) {
1783 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1784 } else {
1785 printk(KERN_WARNING DRV_NAME
4e0952c7 1786 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1787 "still has VLANs.\n",
4e0952c7 1788 bond_dev->name, bond_dev->name);
1da177e4 1789 printk(KERN_WARNING DRV_NAME
4e0952c7
MW
1790 ": %s: When re-adding slaves, make sure the bond's "
1791 "HW address matches its VLANs'.\n",
1792 bond_dev->name);
1da177e4
LT
1793 }
1794 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1795 !bond_has_challenged_slaves(bond)) {
1796 printk(KERN_INFO DRV_NAME
4e0952c7 1797 ": %s: last VLAN challenged slave %s "
1da177e4 1798 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1799 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT
1800 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1801 }
1802
1803 write_unlock_bh(&bond->lock);
1804
b76cdba9
MW
1805 /* must do this from outside any spinlocks */
1806 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1807
1da177e4
LT
1808 bond_del_vlans_from_slave(bond, slave_dev);
1809
1810 /* If the mode USES_PRIMARY, then we should only remove its
1811 * promisc and mc settings if it was the curr_active_slave, but that was
1812 * already taken care of above when we detached the slave
1813 */
1814 if (!USES_PRIMARY(bond->params.mode)) {
1815 /* unset promiscuity level from slave */
1816 if (bond_dev->flags & IFF_PROMISC) {
1817 dev_set_promiscuity(slave_dev, -1);
1818 }
1819
1820 /* unset allmulti level from slave */
1821 if (bond_dev->flags & IFF_ALLMULTI) {
1822 dev_set_allmulti(slave_dev, -1);
1823 }
1824
1825 /* flush master's mc_list from slave */
80ee5ad2 1826 netif_tx_lock_bh(bond_dev);
1da177e4 1827 bond_mc_list_flush(bond_dev, slave_dev);
80ee5ad2 1828 netif_tx_unlock_bh(bond_dev);
1da177e4
LT
1829 }
1830
1831 netdev_set_master(slave_dev, NULL);
1832
1833 /* close slave before restoring its mac address */
1834 dev_close(slave_dev);
1835
dd957c57 1836 if (!bond->params.fail_over_mac) {
2ab82852
MS
1837 /* restore original ("permanent") mac address */
1838 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1839 addr.sa_family = slave_dev->type;
1840 dev_set_mac_address(slave_dev, &addr);
1841 }
1da177e4 1842
8f903c70 1843 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV
1844 IFF_SLAVE_INACTIVE | IFF_BONDING |
1845 IFF_SLAVE_NEEDARP);
1da177e4
LT
1846
1847 kfree(slave);
1848
1849 return 0; /* deletion OK */
1850}
1851
d90a162a
MS
1852/*
1853* Destroy a bonding device.
1854* Must be under rtnl_lock when this function is called.
1855*/
1856void bond_destroy(struct bonding *bond)
1857{
1858 bond_deinit(bond->dev);
1859 bond_destroy_sysfs_entry(bond);
8cbdeec6 1860 unregister_netdevice(bond->dev);
d90a162a
MS
1861}
1862
1863/*
1864* First release a slave and than destroy the bond if no more slaves iare left.
1865* Must be under rtnl_lock when this function is called.
1866*/
1867int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1868{
1869 struct bonding *bond = bond_dev->priv;
1870 int ret;
1871
1872 ret = bond_release(bond_dev, slave_dev);
1873 if ((ret == 0) && (bond->slave_cnt == 0)) {
1874 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1875 bond_dev->name, bond_dev->name);
1876 bond_destroy(bond);
1877 }
1878 return ret;
1879}
1880
1da177e4
LT
1881/*
1882 * This function releases all slaves.
1883 */
1884static int bond_release_all(struct net_device *bond_dev)
1885{
1886 struct bonding *bond = bond_dev->priv;
1887 struct slave *slave;
1888 struct net_device *slave_dev;
1889 struct sockaddr addr;
1890
1891 write_lock_bh(&bond->lock);
1892
ff59c456
JV
1893 netif_carrier_off(bond_dev);
1894
1da177e4
LT
1895 if (bond->slave_cnt == 0) {
1896 goto out;
1897 }
1898
1899 bond->current_arp_slave = NULL;
1900 bond->primary_slave = NULL;
1901 bond_change_active_slave(bond, NULL);
1902
1903 while ((slave = bond->first_slave) != NULL) {
1904 /* Inform AD package of unbinding of slave
1905 * before slave is detached from the list.
1906 */
1907 if (bond->params.mode == BOND_MODE_8023AD) {
1908 bond_3ad_unbind_slave(slave);
1909 }
1910
1911 slave_dev = slave->dev;
1912 bond_detach_slave(bond, slave);
1913
2543331d
JV
1914 /* now that the slave is detached, unlock and perform
1915 * all the undo steps that should not be called from
1916 * within a lock.
1917 */
1918 write_unlock_bh(&bond->lock);
1919
1da177e4
LT
1920 if ((bond->params.mode == BOND_MODE_TLB) ||
1921 (bond->params.mode == BOND_MODE_ALB)) {
1922 /* must be called only after the slave
1923 * has been detached from the list
1924 */
1925 bond_alb_deinit_slave(bond, slave);
1926 }
1927
8531c5ff
AK
1928 bond_compute_features(bond);
1929
b76cdba9 1930 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT
1931 bond_del_vlans_from_slave(bond, slave_dev);
1932
1933 /* If the mode USES_PRIMARY, then we should only remove its
1934 * promisc and mc settings if it was the curr_active_slave, but that was
1935 * already taken care of above when we detached the slave
1936 */
1937 if (!USES_PRIMARY(bond->params.mode)) {
1938 /* unset promiscuity level from slave */
1939 if (bond_dev->flags & IFF_PROMISC) {
1940 dev_set_promiscuity(slave_dev, -1);
1941 }
1942
1943 /* unset allmulti level from slave */
1944 if (bond_dev->flags & IFF_ALLMULTI) {
1945 dev_set_allmulti(slave_dev, -1);
1946 }
1947
1948 /* flush master's mc_list from slave */
80ee5ad2 1949 netif_tx_lock_bh(bond_dev);
1da177e4 1950 bond_mc_list_flush(bond_dev, slave_dev);
80ee5ad2 1951 netif_tx_unlock_bh(bond_dev);
1da177e4
LT
1952 }
1953
1954 netdev_set_master(slave_dev, NULL);
1955
1956 /* close slave before restoring its mac address */
1957 dev_close(slave_dev);
1958
dd957c57 1959 if (!bond->params.fail_over_mac) {
2ab82852
MS
1960 /* restore original ("permanent") mac address*/
1961 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1962 addr.sa_family = slave_dev->type;
1963 dev_set_mac_address(slave_dev, &addr);
1964 }
1da177e4 1965
8f903c70
JV
1966 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1967 IFF_SLAVE_INACTIVE);
1da177e4
LT
1968
1969 kfree(slave);
1970
1971 /* re-acquire the lock before getting the next slave */
1972 write_lock_bh(&bond->lock);
1973 }
1974
1975 /* zero the mac address of the master so it will be
1976 * set by the application to the mac address of the
1977 * first slave
1978 */
1979 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1980
1981 if (list_empty(&bond->vlan_list)) {
1982 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1983 } else {
1984 printk(KERN_WARNING DRV_NAME
4e0952c7 1985 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1986 "still has VLANs.\n",
4e0952c7 1987 bond_dev->name, bond_dev->name);
1da177e4 1988 printk(KERN_WARNING DRV_NAME
4e0952c7
MW
1989 ": %s: When re-adding slaves, make sure the bond's "
1990 "HW address matches its VLANs'.\n",
1991 bond_dev->name);
1da177e4
LT
1992 }
1993
1994 printk(KERN_INFO DRV_NAME
1995 ": %s: released all slaves\n",
1996 bond_dev->name);
1997
1998out:
1999 write_unlock_bh(&bond->lock);
2000
2001 return 0;
2002}
2003
2004/*
2005 * This function changes the active slave to slave <slave_dev>.
2006 * It returns -EINVAL in the following cases.
2007 * - <slave_dev> is not found in the list.
2008 * - There is not active slave now.
2009 * - <slave_dev> is already active.
2010 * - The link state of <slave_dev> is not BOND_LINK_UP.
2011 * - <slave_dev> is not running.
2012 * In these cases, this fuction does nothing.
2013 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2014 */
2015static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2016{
2017 struct bonding *bond = bond_dev->priv;
2018 struct slave *old_active = NULL;
2019 struct slave *new_active = NULL;
2020 int res = 0;
2021
2022 if (!USES_PRIMARY(bond->params.mode)) {
2023 return -EINVAL;
2024 }
2025
2026 /* Verify that master_dev is indeed the master of slave_dev */
2027 if (!(slave_dev->flags & IFF_SLAVE) ||
2028 (slave_dev->master != bond_dev)) {
2029 return -EINVAL;
2030 }
2031
059fe7a5 2032 read_lock(&bond->lock);
1da177e4 2033
059fe7a5 2034 read_lock(&bond->curr_slave_lock);
1da177e4 2035 old_active = bond->curr_active_slave;
059fe7a5
JV
2036 read_unlock(&bond->curr_slave_lock);
2037
1da177e4
LT
2038 new_active = bond_get_slave_by_dev(bond, slave_dev);
2039
2040 /*
2041 * Changing to the current active: do nothing; return success.
2042 */
2043 if (new_active && (new_active == old_active)) {
059fe7a5 2044 read_unlock(&bond->lock);
1da177e4
LT
2045 return 0;
2046 }
2047
2048 if ((new_active) &&
2049 (old_active) &&
2050 (new_active->link == BOND_LINK_UP) &&
2051 IS_UP(new_active->dev)) {
059fe7a5 2052 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2053 bond_change_active_slave(bond, new_active);
059fe7a5 2054 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT
2055 } else {
2056 res = -EINVAL;
2057 }
2058
059fe7a5 2059 read_unlock(&bond->lock);
1da177e4
LT
2060
2061 return res;
2062}
2063
1da177e4
LT
2064static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2065{
2066 struct bonding *bond = bond_dev->priv;
2067
2068 info->bond_mode = bond->params.mode;
2069 info->miimon = bond->params.miimon;
2070
6603a6f2 2071 read_lock(&bond->lock);
1da177e4 2072 info->num_slaves = bond->slave_cnt;
6603a6f2 2073 read_unlock(&bond->lock);
1da177e4
LT
2074
2075 return 0;
2076}
2077
2078static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2079{
2080 struct bonding *bond = bond_dev->priv;
2081 struct slave *slave;
2082 int i, found = 0;
2083
2084 if (info->slave_id < 0) {
2085 return -ENODEV;
2086 }
2087
6603a6f2 2088 read_lock(&bond->lock);
1da177e4
LT
2089
2090 bond_for_each_slave(bond, slave, i) {
2091 if (i == (int)info->slave_id) {
2092 found = 1;
2093 break;
2094 }
2095 }
2096
6603a6f2 2097 read_unlock(&bond->lock);
1da177e4
LT
2098
2099 if (found) {
2100 strcpy(info->slave_name, slave->dev->name);
2101 info->link = slave->link;
2102 info->state = slave->state;
2103 info->link_failure_count = slave->link_failure_count;
2104 } else {
2105 return -ENODEV;
2106 }
2107
2108 return 0;
2109}
2110
2111/*-------------------------------- Monitoring -------------------------------*/
2112
0b0eef66
JV
2113/*
2114 * if !have_locks, return nonzero if a failover is necessary. if
2115 * have_locks, do whatever failover activities are needed.
2116 *
2117 * This is to separate the inspection and failover steps for locking
2118 * purposes; failover requires rtnl, but acquiring it for every
2119 * inspection is undesirable, so a wrapper first does inspection, and
2120 * the acquires the necessary locks and calls again to perform
2121 * failover if needed. Since all locks are dropped, a complete
2122 * restart is needed between calls.
2123 */
2124static int __bond_mii_monitor(struct bonding *bond, int have_locks)
1da177e4 2125{
1da177e4
LT
2126 struct slave *slave, *oldcurrent;
2127 int do_failover = 0;
1da177e4
LT
2128 int i;
2129
0b0eef66 2130 if (bond->slave_cnt == 0)
1da177e4 2131 goto out;
1da177e4
LT
2132
2133 /* we will try to read the link status of each of our slaves, and
2134 * set their IFF_RUNNING flag appropriately. For each slave not
2135 * supporting MII status, we won't do anything so that a user-space
2136 * program could monitor the link itself if needed.
2137 */
2138
1053f62c
MS
2139 if (bond->send_grat_arp) {
2140 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2141 &bond->curr_active_slave->dev->state))
2142 dprintk("Needs to send gratuitous arp but not yet\n");
2143 else {
2144 dprintk("sending delayed gratuitous arp on on %s\n",
2145 bond->curr_active_slave->dev->name);
2146 bond_send_gratuitous_arp(bond);
2147 bond->send_grat_arp = 0;
2148 }
2149 }
1da177e4
LT
2150 read_lock(&bond->curr_slave_lock);
2151 oldcurrent = bond->curr_active_slave;
2152 read_unlock(&bond->curr_slave_lock);
2153
2154 bond_for_each_slave(bond, slave, i) {
2155 struct net_device *slave_dev = slave->dev;
2156 int link_state;
2157 u16 old_speed = slave->speed;
2158 u8 old_duplex = slave->duplex;
2159
2160 link_state = bond_check_dev_link(bond, slave_dev, 0);
2161
2162 switch (slave->link) {
2163 case BOND_LINK_UP: /* the link was up */
2164 if (link_state == BMSR_LSTATUS) {
059fe7a5
JV
2165 if (!oldcurrent) {
2166 if (!have_locks)
2167 return 1;
2168 do_failover = 1;
2169 }
1da177e4
LT
2170 break;
2171 } else { /* link going down */
2172 slave->link = BOND_LINK_FAIL;
2173 slave->delay = bond->params.downdelay;
2174
2175 if (slave->link_failure_count < UINT_MAX) {
2176 slave->link_failure_count++;
2177 }
2178
2179 if (bond->params.downdelay) {
2180 printk(KERN_INFO DRV_NAME
2181 ": %s: link status down for %s "
2182 "interface %s, disabling it in "
2183 "%d ms.\n",
1b76b316 2184 bond->dev->name,
1da177e4
LT
2185 IS_UP(slave_dev)
2186 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2187 ? ((slave == oldcurrent)
2188 ? "active " : "backup ")
2189 : "")
2190 : "idle ",
2191 slave_dev->name,
2192 bond->params.downdelay * bond->params.miimon);
2193 }
2194 }
2195 /* no break ! fall through the BOND_LINK_FAIL test to
2196 ensure proper action to be taken
2197 */
2198 case BOND_LINK_FAIL: /* the link has just gone down */
2199 if (link_state != BMSR_LSTATUS) {
2200 /* link stays down */
2201 if (slave->delay <= 0) {
0b0eef66
JV
2202 if (!have_locks)
2203 return 1;
2204
1da177e4
LT
2205 /* link down for too long time */
2206 slave->link = BOND_LINK_DOWN;
2207
2208 /* in active/backup mode, we must
2209 * completely disable this interface
2210 */
2211 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2212 (bond->params.mode == BOND_MODE_8023AD)) {
2213 bond_set_slave_inactive_flags(slave);
2214 }
2215
2216 printk(KERN_INFO DRV_NAME
2217 ": %s: link status definitely "
2218 "down for interface %s, "
2219 "disabling it\n",
1b76b316 2220 bond->dev->name,
1da177e4
LT
2221 slave_dev->name);
2222
2223 /* notify ad that the link status has changed */
2224 if (bond->params.mode == BOND_MODE_8023AD) {
2225 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2226 }
2227
2228 if ((bond->params.mode == BOND_MODE_TLB) ||
2229 (bond->params.mode == BOND_MODE_ALB)) {
2230 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2231 }
2232
2233 if (slave == oldcurrent) {
2234 do_failover = 1;
2235 }
2236 } else {
2237 slave->delay--;
2238 }
2239 } else {
2240 /* link up again */
2241 slave->link = BOND_LINK_UP;
2242 slave->jiffies = jiffies;
2243 printk(KERN_INFO DRV_NAME
2244 ": %s: link status up again after %d "
2245 "ms for interface %s.\n",
1b76b316 2246 bond->dev->name,
1da177e4
LT
2247 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2248 slave_dev->name);
2249 }
2250 break;
2251 case BOND_LINK_DOWN: /* the link was down */
2252 if (link_state != BMSR_LSTATUS) {
2253 /* the link stays down, nothing more to do */
2254 break;
2255 } else { /* link going up */
2256 slave->link = BOND_LINK_BACK;
2257 slave->delay = bond->params.updelay;
2258
2259 if (bond->params.updelay) {
2260 /* if updelay == 0, no need to
2261 advertise about a 0 ms delay */
2262 printk(KERN_INFO DRV_NAME
2263 ": %s: link status up for "
2264 "interface %s, enabling it "
2265 "in %d ms.\n",
1b76b316 2266 bond->dev->name,
1da177e4
LT
2267 slave_dev->name,
2268 bond->params.updelay * bond->params.miimon);
2269 }
2270 }
2271 /* no break ! fall through the BOND_LINK_BACK state in
2272 case there's something to do.
2273 */
2274 case BOND_LINK_BACK: /* the link has just come back */
2275 if (link_state != BMSR_LSTATUS) {
2276 /* link down again */
2277 slave->link = BOND_LINK_DOWN;
2278
2279 printk(KERN_INFO DRV_NAME
2280 ": %s: link status down again after %d "
2281 "ms for interface %s.\n",
1b76b316 2282 bond->dev->name,
1da177e4
LT
2283 (bond->params.updelay - slave->delay) * bond->params.miimon,
2284 slave_dev->name);
2285 } else {
2286 /* link stays up */
2287 if (slave->delay == 0) {
0b0eef66
JV
2288 if (!have_locks)
2289 return 1;
2290
1da177e4
LT
2291 /* now the link has been up for long time enough */
2292 slave->link = BOND_LINK_UP;
2293 slave->jiffies = jiffies;
2294
2295 if (bond->params.mode == BOND_MODE_8023AD) {
2296 /* prevent it from being the active one */
2297 slave->state = BOND_STATE_BACKUP;
2298 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2299 /* make it immediately active */
2300 slave->state = BOND_STATE_ACTIVE;
2301 } else if (slave != bond->primary_slave) {
2302 /* prevent it from being the active one */
2303 slave->state = BOND_STATE_BACKUP;
2304 }
2305
2306 printk(KERN_INFO DRV_NAME
2307 ": %s: link status definitely "
2308 "up for interface %s.\n",
1b76b316 2309 bond->dev->name,
1da177e4
LT
2310 slave_dev->name);
2311
2312 /* notify ad that the link status has changed */
2313 if (bond->params.mode == BOND_MODE_8023AD) {
2314 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2315 }
2316
2317 if ((bond->params.mode == BOND_MODE_TLB) ||
2318 (bond->params.mode == BOND_MODE_ALB)) {
2319 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2320 }
2321
2322 if ((!oldcurrent) ||
2323 (slave == bond->primary_slave)) {
2324 do_failover = 1;
2325 }
2326 } else {
2327 slave->delay--;
2328 }
2329 }
2330 break;
2331 default:
2332 /* Should not happen */
4e0952c7
MW
2333 printk(KERN_ERR DRV_NAME
2334 ": %s: Error: %s Illegal value (link=%d)\n",
1b76b316 2335 bond->dev->name,
4e0952c7
MW
2336 slave->dev->name,
2337 slave->link);
1da177e4
LT
2338 goto out;
2339 } /* end of switch (slave->link) */
2340
2341 bond_update_speed_duplex(slave);
2342
2343 if (bond->params.mode == BOND_MODE_8023AD) {
2344 if (old_speed != slave->speed) {
2345 bond_3ad_adapter_speed_changed(slave);
2346 }
2347
2348 if (old_duplex != slave->duplex) {
2349 bond_3ad_adapter_duplex_changed(slave);
2350 }
2351 }
2352
2353 } /* end of for */
2354
2355 if (do_failover) {
059fe7a5
JV
2356 ASSERT_RTNL();
2357
2358 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT
2359
2360 bond_select_active_slave(bond);
2361
059fe7a5
JV
2362 write_unlock_bh(&bond->curr_slave_lock);
2363
ff59c456
JV
2364 } else
2365 bond_set_carrier(bond);
1da177e4 2366
1da177e4 2367out:
0b0eef66 2368 return 0;
1da177e4
LT
2369}
2370
0b0eef66
JV
2371/*
2372 * bond_mii_monitor
2373 *
2374 * Really a wrapper that splits the mii monitor into two phases: an
2375 * inspection, then (if inspection indicates something needs to be
2376 * done) an acquisition of appropriate locks followed by another pass
2377 * to implement whatever link state changes are indicated.
2378 */
2379void bond_mii_monitor(struct work_struct *work)
2380{
2381 struct bonding *bond = container_of(work, struct bonding,
2382 mii_work.work);
2383 unsigned long delay;
2384
2385 read_lock(&bond->lock);
2386 if (bond->kill_timers) {
2387 read_unlock(&bond->lock);
2388 return;
2389 }
2390 if (__bond_mii_monitor(bond, 0)) {
2391 read_unlock(&bond->lock);
2392 rtnl_lock();
2393 read_lock(&bond->lock);
2394 __bond_mii_monitor(bond, 1);
5655662d
JV
2395 read_unlock(&bond->lock);
2396 rtnl_unlock(); /* might sleep, hold no other locks */
2397 read_lock(&bond->lock);
0b0eef66
JV
2398 }
2399
5ce0da8f 2400 delay = msecs_to_jiffies(bond->params.miimon);
0b0eef66
JV
2401 read_unlock(&bond->lock);
2402 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2403}
c3ade5ca 2404
d3bb52b0 2405static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV
2406{
2407 struct in_device *idev;
2408 struct in_ifaddr *ifa;
a144ea4b 2409 __be32 addr = 0;
c3ade5ca
JV
2410
2411 if (!dev)
2412 return 0;
2413
2414 rcu_read_lock();
e5ed6399 2415 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV
2416 if (!idev)
2417 goto out;
2418
2419 ifa = idev->ifa_list;
2420 if (!ifa)
2421 goto out;
2422
2423 addr = ifa->ifa_local;
2424out:
2425 rcu_read_unlock();
2426 return addr;
2427}
2428
d3bb52b0 2429static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966
JV
2430{
2431 struct vlan_entry *vlan, *vlan_next;
2432
2433 if (ip == bond->master_ip)
2434 return 1;
2435
2436 if (list_empty(&bond->vlan_list))
2437 return 0;
2438
2439 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2440 vlan_list) {
2441 if (ip == vlan->vlan_ip)
2442 return 1;
2443 }
2444
2445 return 0;
2446}
2447
c3ade5ca
JV
2448/*
2449 * We go to the (large) trouble of VLAN tagging ARP frames because
2450 * switches in VLAN mode (especially if ports are configured as
2451 * "native" to a VLAN) might not pass non-tagged frames.
2452 */
d3bb52b0 2453static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV
2454{
2455 struct sk_buff *skb;
2456
2457 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2458 slave_dev->name, dest_ip, src_ip, vlan_id);
2459
2460 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2461 NULL, slave_dev->dev_addr, NULL);
2462
2463 if (!skb) {
2464 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2465 return;
2466 }
2467 if (vlan_id) {
2468 skb = vlan_put_tag(skb, vlan_id);
2469 if (!skb) {
2470 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2471 return;
2472 }
2473 }
2474 arp_xmit(skb);
2475}
2476
2477
1da177e4
LT
2478static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2479{
c3ade5ca 2480 int i, vlan_id, rv;
d3bb52b0 2481 __be32 *targets = bond->params.arp_targets;
c3ade5ca
JV
2482 struct vlan_entry *vlan, *vlan_next;
2483 struct net_device *vlan_dev;
2484 struct flowi fl;
2485 struct rtable *rt;
1da177e4 2486
6b780567
MW
2487 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2488 if (!targets[i])
2489 continue;
c3ade5ca
JV
2490 dprintk("basa: target %x\n", targets[i]);
2491 if (list_empty(&bond->vlan_list)) {
2492 dprintk("basa: empty vlan: arp_send\n");
2493 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2494 bond->master_ip, 0);
2495 continue;
2496 }
2497
2498 /*
2499 * If VLANs are configured, we do a route lookup to
2500 * determine which VLAN interface would be used, so we
2501 * can tag the ARP with the proper VLAN tag.
2502 */
2503 memset(&fl, 0, sizeof(fl));
2504 fl.fl4_dst = targets[i];
2505 fl.fl4_tos = RTO_ONLINK;
2506
f206351a 2507 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV
2508 if (rv) {
2509 if (net_ratelimit()) {
2510 printk(KERN_WARNING DRV_NAME
2511 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2512 bond->dev->name, NIPQUAD(fl.fl4_dst));
2513 }
2514 continue;
2515 }
2516
2517 /*
2518 * This target is not on a VLAN
2519 */
2520 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2521 ip_rt_put(rt);
c3ade5ca
JV
2522 dprintk("basa: rtdev == bond->dev: arp_send\n");
2523 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2524 bond->master_ip, 0);
2525 continue;
2526 }
2527
2528 vlan_id = 0;
2529 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2530 vlan_list) {
5c15bdec 2531 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV
2532 if (vlan_dev == rt->u.dst.dev) {
2533 vlan_id = vlan->vlan_id;
2534 dprintk("basa: vlan match on %s %d\n",
2535 vlan_dev->name, vlan_id);
2536 break;
2537 }
2538 }
2539
2540 if (vlan_id) {
ed4b9f80 2541 ip_rt_put(rt);
c3ade5ca
JV
2542 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2543 vlan->vlan_ip, vlan_id);
2544 continue;
2545 }
2546
2547 if (net_ratelimit()) {
2548 printk(KERN_WARNING DRV_NAME
2549 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2550 bond->dev->name, NIPQUAD(fl.fl4_dst),
2551 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2552 }
ed4b9f80 2553 ip_rt_put(rt);
c3ade5ca
JV
2554 }
2555}
2556
2557/*
2558 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2559 * for each VLAN above us.
2560 */
2561static void bond_send_gratuitous_arp(struct bonding *bond)
2562{
2563 struct slave *slave = bond->curr_active_slave;
2564 struct vlan_entry *vlan;
2565 struct net_device *vlan_dev;
2566
2567 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2568 slave ? slave->dev->name : "NULL");
2569 if (!slave)
2570 return;
2571
2572 if (bond->master_ip) {
2573 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2574 bond->master_ip, 0);
c3ade5ca
JV
2575 }
2576
2577 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2578 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV
2579 if (vlan->vlan_ip) {
2580 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2581 vlan->vlan_ip, vlan->vlan_id);
2582 }
1da177e4
LT
2583 }
2584}
2585
d3bb52b0 2586static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV
2587{
2588 int i;
d3bb52b0 2589 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV
2590
2591 targets = bond->params.arp_targets;
2592 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2593 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2594 "%u.%u.%u.%u bhti(tip) %d\n",
2595 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2596 bond_has_this_ip(bond, tip));
2597 if (sip == targets[i]) {
2598 if (bond_has_this_ip(bond, tip))
2599 slave->last_arp_rx = jiffies;
2600 return;
2601 }
2602 }
2603}
2604
2605static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2606{
2607 struct arphdr *arp;
2608 struct slave *slave;
2609 struct bonding *bond;
2610 unsigned char *arp_ptr;
d3bb52b0 2611 __be32 sip, tip;
f5b2b966 2612
c346dca1 2613 if (dev_net(dev) != &init_net)
e730c155
EB
2614 goto out;
2615
f5b2b966
JV
2616 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2617 goto out;
2618
2619 bond = dev->priv;
2620 read_lock(&bond->lock);
2621
2622 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2623 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2624 orig_dev ? orig_dev->name : "NULL");
2625
2626 slave = bond_get_slave_by_dev(bond, orig_dev);
2627 if (!slave || !slave_do_arp_validate(bond, slave))
2628 goto out_unlock;
2629
988b7050 2630 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV
2631 goto out_unlock;
2632
d0a92be0 2633 arp = arp_hdr(skb);
f5b2b966
JV
2634 if (arp->ar_hln != dev->addr_len ||
2635 skb->pkt_type == PACKET_OTHERHOST ||
2636 skb->pkt_type == PACKET_LOOPBACK ||
2637 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2638 arp->ar_pro != htons(ETH_P_IP) ||
2639 arp->ar_pln != 4)
2640 goto out_unlock;
2641
2642 arp_ptr = (unsigned char *)(arp + 1);
2643 arp_ptr += dev->addr_len;
2644 memcpy(&sip, arp_ptr, 4);
2645 arp_ptr += 4 + dev->addr_len;
2646 memcpy(&tip, arp_ptr, 4);
2647
2648 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2649 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2650 slave->state, bond->params.arp_validate,
2651 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2652
2653 /*
2654 * Backup slaves won't see the ARP reply, but do come through
2655 * here for each ARP probe (so we swap the sip/tip to validate
2656 * the probe). In a "redundant switch, common router" type of
2657 * configuration, the ARP probe will (hopefully) travel from
2658 * the active, through one switch, the router, then the other
2659 * switch before reaching the backup.
2660 */
2661 if (slave->state == BOND_STATE_ACTIVE)
2662 bond_validate_arp(bond, slave, sip, tip);
2663 else
2664 bond_validate_arp(bond, slave, tip, sip);
2665
2666out_unlock:
2667 read_unlock(&bond->lock);
2668out:
2669 dev_kfree_skb(skb);
2670 return NET_RX_SUCCESS;
2671}
2672
1da177e4
LT
2673/*
2674 * this function is called regularly to monitor each slave's link
2675 * ensuring that traffic is being sent and received when arp monitoring
2676 * is used in load-balancing mode. if the adapter has been dormant, then an
2677 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2678 * arp monitoring in active backup mode.
2679 */
1b76b316 2680void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2681{
1b76b316
JV
2682 struct bonding *bond = container_of(work, struct bonding,
2683 arp_work.work);
1da177e4
LT
2684 struct slave *slave, *oldcurrent;
2685 int do_failover = 0;
2686 int delta_in_ticks;
2687 int i;
2688
2689 read_lock(&bond->lock);
2690
5ce0da8f 2691 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT
2692
2693 if (bond->kill_timers) {
2694 goto out;
2695 }
2696
2697 if (bond->slave_cnt == 0) {
2698 goto re_arm;
2699 }
2700
2701 read_lock(&bond->curr_slave_lock);
2702 oldcurrent = bond->curr_active_slave;
2703 read_unlock(&bond->curr_slave_lock);
2704
2705 /* see if any of the previous devices are up now (i.e. they have
2706 * xmt and rcv traffic). the curr_active_slave does not come into
2707 * the picture unless it is null. also, slave->jiffies is not needed
2708 * here because we send an arp on each slave and give a slave as
2709 * long as it needs to get the tx/rx within the delta.
2710 * TODO: what about up/down delay in arp mode? it wasn't here before
2711 * so it can wait
2712 */
2713 bond_for_each_slave(bond, slave, i) {
2714 if (slave->link != BOND_LINK_UP) {
b63bb739
DS
2715 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2716 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT
2717
2718 slave->link = BOND_LINK_UP;
2719 slave->state = BOND_STATE_ACTIVE;
2720
2721 /* primary_slave has no meaning in round-robin
2722 * mode. the window of a slave being up and
2723 * curr_active_slave being null after enslaving
2724 * is closed.
2725 */
2726 if (!oldcurrent) {
2727 printk(KERN_INFO DRV_NAME
2728 ": %s: link status definitely "
2729 "up for interface %s, ",
1b76b316 2730 bond->dev->name,
1da177e4
LT
2731 slave->dev->name);
2732 do_failover = 1;
2733 } else {
2734 printk(KERN_INFO DRV_NAME
2735 ": %s: interface %s is now up\n",
1b76b316 2736 bond->dev->name,
1da177e4
LT
2737 slave->dev->name);
2738 }
2739 }
2740 } else {
2741 /* slave->link == BOND_LINK_UP */
2742
2743 /* not all switches will respond to an arp request
2744 * when the source ip is 0, so don't take the link down
2745 * if we don't know our ip yet
2746 */
b63bb739 2747 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2748 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT
2749
2750 slave->link = BOND_LINK_DOWN;
2751 slave->state = BOND_STATE_BACKUP;
2752
2753 if (slave->link_failure_count < UINT_MAX) {
2754 slave->link_failure_count++;
2755 }
2756
2757 printk(KERN_INFO DRV_NAME
2758 ": %s: interface %s is now down.\n",
1b76b316 2759 bond->dev->name,
1da177e4
LT
2760 slave->dev->name);
2761
2762 if (slave == oldcurrent) {
2763 do_failover = 1;
2764 }
2765 }
2766 }
2767
2768 /* note: if switch is in round-robin mode, all links
2769 * must tx arp to ensure all links rx an arp - otherwise
2770 * links may oscillate or not come up at all; if switch is
2771 * in something like xor mode, there is nothing we can
2772 * do - all replies will be rx'ed on same link causing slaves
2773 * to be unstable during low/no traffic periods
2774 */
2775 if (IS_UP(slave->dev)) {
2776 bond_arp_send_all(bond, slave);
2777 }
2778 }
2779
2780 if (do_failover) {
059fe7a5 2781 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT
2782
2783 bond_select_active_slave(bond);
2784
059fe7a5 2785 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT
2786 }
2787
2788re_arm:
1b76b316
JV
2789 if (bond->params.arp_interval)
2790 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT
2791out:
2792 read_unlock(&bond->lock);
2793}
2794
2795/*
2796 * When using arp monitoring in active-backup mode, this function is
2797 * called to determine if any backup slaves have went down or a new
2798 * current slave needs to be found.
2799 * The backup slaves never generate traffic, they are considered up by merely
2800 * receiving traffic. If the current slave goes down, each backup slave will
2801 * be given the opportunity to tx/rx an arp before being taken down - this
2802 * prevents all slaves from being taken down due to the current slave not
2803 * sending any traffic for the backups to receive. The arps are not necessarily
2804 * necessary, any tx and rx traffic will keep the current slave up. While any
2805 * rx traffic will keep the backup slaves up, the current slave is responsible
2806 * for generating traffic to keep them up regardless of any other traffic they
2807 * may have received.
2808 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2809 */
1b76b316 2810void bond_activebackup_arp_mon(struct work_struct *work)
1da177e4 2811{
1b76b316
JV
2812 struct bonding *bond = container_of(work, struct bonding,
2813 arp_work.work);
1da177e4
LT
2814 struct slave *slave;
2815 int delta_in_ticks;
2816 int i;
2817
2818 read_lock(&bond->lock);
2819
5ce0da8f 2820 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT
2821
2822 if (bond->kill_timers) {
2823 goto out;
2824 }
2825
2826 if (bond->slave_cnt == 0) {
2827 goto re_arm;
2828 }
2829
2830 /* determine if any slave has come up or any backup slave has
2831 * gone down
2832 * TODO: what about up/down delay in arp mode? it wasn't here before
2833 * so it can wait
2834 */
2835 bond_for_each_slave(bond, slave, i) {
2836 if (slave->link != BOND_LINK_UP) {
b63bb739
DS
2837 if (time_before_eq(jiffies,
2838 slave_last_rx(bond, slave) + delta_in_ticks)) {
1da177e4
LT
2839
2840 slave->link = BOND_LINK_UP;
2841
059fe7a5 2842 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT
2843
2844 if ((!bond->curr_active_slave) &&
b63bb739 2845 time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks)) {
1da177e4
LT
2846 bond_change_active_slave(bond, slave);
2847 bond->current_arp_slave = NULL;
2848 } else if (bond->curr_active_slave != slave) {
2849 /* this slave has just come up but we
2850 * already have a current slave; this
2851 * can also happen if bond_enslave adds
2852 * a new slave that is up while we are
2853 * searching for a new slave
2854 */
2855 bond_set_slave_inactive_flags(slave);
2856 bond->current_arp_slave = NULL;
2857 }
2858
ff59c456
JV
2859 bond_set_carrier(bond);
2860
1da177e4
LT
2861 if (slave == bond->curr_active_slave) {
2862 printk(KERN_INFO DRV_NAME
2863 ": %s: %s is up and now the "
2864 "active interface\n",
1b76b316 2865 bond->dev->name,
1da177e4 2866 slave->dev->name);
ff59c456 2867 netif_carrier_on(bond->dev);
1da177e4
LT
2868 } else {
2869 printk(KERN_INFO DRV_NAME
2870 ": %s: backup interface %s is "
2871 "now up\n",
1b76b316 2872 bond->dev->name,
1da177e4
LT
2873 slave->dev->name);
2874 }
2875
059fe7a5 2876 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT
2877 }
2878 } else {
2879 read_lock(&bond->curr_slave_lock);
2880
2881 if ((slave != bond->curr_active_slave) &&
2882 (!bond->current_arp_slave) &&
4b8a9239 2883 (time_after_eq(jiffies, slave_last_rx(bond, slave) + 3*delta_in_ticks))) {
1da177e4
LT
2884 /* a backup slave has gone down; three times
2885 * the delta allows the current slave to be
2886 * taken out before the backup slave.
2887 * note: a non-null current_arp_slave indicates
2888 * the curr_active_slave went down and we are
2889 * searching for a new one; under this
2890 * condition we only take the curr_active_slave
2891 * down - this gives each slave a chance to
2892 * tx/rx traffic before being taken out
2893 */
2894
2895 read_unlock(&bond->curr_slave_lock);
2896
2897 slave->link = BOND_LINK_DOWN;
2898
2899 if (slave->link_failure_count < UINT_MAX) {
2900 slave->link_failure_count++;
2901 }
2902
2903 bond_set_slave_inactive_flags(slave);
2904
2905 printk(KERN_INFO DRV_NAME
2906 ": %s: backup interface %s is now down\n",
1b76b316 2907 bond->dev->name,
1da177e4
LT
2908 slave->dev->name);
2909 } else {
2910 read_unlock(&bond->curr_slave_lock);
2911 }
2912 }
2913 }
2914
2915 read_lock(&bond->curr_slave_lock);
2916 slave = bond->curr_active_slave;
2917 read_unlock(&bond->curr_slave_lock);
2918
2919 if (slave) {
2920 /* if we have sent traffic in the past 2*arp_intervals but
2921 * haven't xmit and rx traffic in that time interval, select
2922 * a different slave. slave->jiffies is only updated when
2923 * a slave first becomes the curr_active_slave - not necessarily
2924 * after every arp; this ensures the slave has a full 2*delta
2925 * before being taken out. if a primary is being used, check
2926 * if it is up and needs to take over as the curr_active_slave
2927 */
b63bb739 2928 if ((time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2929 (time_after_eq(jiffies, slave_last_rx(bond, slave) + 2*delta_in_ticks))) &&
b63bb739 2930 time_after_eq(jiffies, slave->jiffies + 2*delta_in_ticks)) {
1da177e4
LT
2931
2932 slave->link = BOND_LINK_DOWN;
2933
2934 if (slave->link_failure_count < UINT_MAX) {
2935 slave->link_failure_count++;
2936 }
2937
2938 printk(KERN_INFO DRV_NAME
2939 ": %s: link status down for active interface "
2940 "%s, disabling it\n",
1b76b316 2941 bond->dev->name,
1da177e4
LT
2942 slave->dev->name);
2943
059fe7a5 2944 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT
2945
2946 bond_select_active_slave(bond);
2947 slave = bond->curr_active_slave;
2948
059fe7a5
JV
2949 write_unlock_bh(&bond->curr_slave_lock);
2950
1da177e4
LT
2951 bond->current_arp_slave = slave;
2952
2953 if (slave) {
2954 slave->jiffies = jiffies;
2955 }
2956 } else if ((bond->primary_slave) &&
2957 (bond->primary_slave != slave) &&
2958 (bond->primary_slave->link == BOND_LINK_UP)) {
2959 /* at this point, slave is the curr_active_slave */
2960 printk(KERN_INFO DRV_NAME
2961 ": %s: changing from interface %s to primary "
2962 "interface %s\n",
1b76b316 2963 bond->dev->name,
1da177e4
LT
2964 slave->dev->name,
2965 bond->primary_slave->dev->name);
2966
2967 /* primary is up so switch to it */
059fe7a5 2968 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2969 bond_change_active_slave(bond, bond->primary_slave);
059fe7a5
JV
2970 write_unlock_bh(&bond->curr_slave_lock);
2971
1da177e4
LT
2972 slave = bond->primary_slave;
2973 slave->jiffies = jiffies;
2974 } else {
2975 bond->current_arp_slave = NULL;
2976 }
2977
2978 /* the current slave must tx an arp to ensure backup slaves
2979 * rx traffic
2980 */
4b8a9239 2981 if (slave && IS_UP(slave->dev))
1da177e4 2982 bond_arp_send_all(bond, slave);
1da177e4
LT
2983 }
2984
2985 /* if we don't have a curr_active_slave, search for the next available
2986 * backup slave from the current_arp_slave and make it the candidate
2987 * for becoming the curr_active_slave
2988 */
2989 if (!slave) {
2990 if (!bond->current_arp_slave) {
2991 bond->current_arp_slave = bond->first_slave;
2992 }
2993
2994 if (bond->current_arp_slave) {
2995 bond_set_slave_inactive_flags(bond->current_arp_slave);
2996
2997 /* search for next candidate */
2f872f04 2998 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
1da177e4
LT
2999 if (IS_UP(slave->dev)) {
3000 slave->link = BOND_LINK_BACK;
3001 bond_set_slave_active_flags(slave);
3002 bond_arp_send_all(bond, slave);
3003 slave->jiffies = jiffies;
3004 bond->current_arp_slave = slave;
3005 break;
3006 }
3007
3008 /* if the link state is up at this point, we
3009 * mark it down - this can happen if we have
3010 * simultaneous link failures and
3011 * reselect_active_interface doesn't make this
3012 * one the current slave so it is still marked
3013 * up when it is actually down
3014 */
3015 if (slave->link == BOND_LINK_UP) {
3016 slave->link = BOND_LINK_DOWN;
3017 if (slave->link_failure_count < UINT_MAX) {
3018 slave->link_failure_count++;
3019 }
3020
3021 bond_set_slave_inactive_flags(slave);
3022
3023 printk(KERN_INFO DRV_NAME
3024 ": %s: backup interface %s is "
3025 "now down.\n",
1b76b316 3026 bond->dev->name,
1da177e4
LT
3027 slave->dev->name);
3028 }
3029 }
3030 }
3031 }
3032
3033re_arm:
3034 if (bond->params.arp_interval) {
1b76b316 3035 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT
3036 }
3037out:
3038 read_unlock(&bond->lock);
3039}
3040
3041/*------------------------------ proc/seq_file-------------------------------*/
3042
3043#ifdef CONFIG_PROC_FS
3044
1da177e4
LT
3045static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3046{
3047 struct bonding *bond = seq->private;
3048 loff_t off = 0;
3049 struct slave *slave;
3050 int i;
3051
3052 /* make sure the bond won't be taken away */
3053 read_lock(&dev_base_lock);
6603a6f2 3054 read_lock(&bond->lock);
1da177e4
LT
3055
3056 if (*pos == 0) {
3057 return SEQ_START_TOKEN;
3058 }
3059
3060 bond_for_each_slave(bond, slave, i) {
3061 if (++off == *pos) {
3062 return slave;
3063 }
3064 }
3065
3066 return NULL;
3067}
3068
3069static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3070{
3071 struct bonding *bond = seq->private;
3072 struct slave *slave = v;
3073
3074 ++*pos;
3075 if (v == SEQ_START_TOKEN) {
3076 return bond->first_slave;
3077 }
3078
3079 slave = slave->next;
3080
3081 return (slave == bond->first_slave) ? NULL : slave;
3082}
3083
3084static void bond_info_seq_stop(struct seq_file *seq, void *v)
3085{
3086 struct bonding *bond = seq->private;
3087
6603a6f2 3088 read_unlock(&bond->lock);
1da177e4
LT
3089 read_unlock(&dev_base_lock);
3090}
3091
3092static void bond_info_show_master(struct seq_file *seq)
3093{
3094 struct bonding *bond = seq->private;
3095 struct slave *curr;
4756b02f
MW
3096 int i;
3097 u32 target;
1da177e4
LT
3098
3099 read_lock(&bond->curr_slave_lock);
3100 curr = bond->curr_active_slave;
3101 read_unlock(&bond->curr_slave_lock);
3102
dd957c57 3103 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT
3104 bond_mode_name(bond->params.mode));
3105
dd957c57
JV
3106 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3107 bond->params.fail_over_mac)
3108 seq_printf(seq, " (fail_over_mac)");
3109
3110 seq_printf(seq, "\n");
3111
c61b75ad
MW
3112 if (bond->params.mode == BOND_MODE_XOR ||
3113 bond->params.mode == BOND_MODE_8023AD) {
3114 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3115 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3116 bond->params.xmit_policy);
3117 }
3118
1da177e4
LT
3119 if (USES_PRIMARY(bond->params.mode)) {
3120 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW
3121 (bond->primary_slave) ?
3122 bond->primary_slave->dev->name : "None");
1da177e4
LT
3123
3124 seq_printf(seq, "Currently Active Slave: %s\n",
3125 (curr) ? curr->dev->name : "None");
3126 }
3127
ff59c456
JV
3128 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3129 "up" : "down");
1da177e4
LT
3130 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3131 seq_printf(seq, "Up Delay (ms): %d\n",
3132 bond->params.updelay * bond->params.miimon);
3133 seq_printf(seq, "Down Delay (ms): %d\n",
3134 bond->params.downdelay * bond->params.miimon);
3135
4756b02f
MW
3136
3137 /* ARP information */
3138 if(bond->params.arp_interval > 0) {
3139 int printed=0;
3140 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3141 bond->params.arp_interval);
3142
3143 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3144
3145 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3146 if (!bond->params.arp_targets[i])
3147 continue;
3148 if (printed)
3149 seq_printf(seq, ",");
3150 target = ntohl(bond->params.arp_targets[i]);
3151 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3152 printed = 1;
3153 }
3154 seq_printf(seq, "\n");
3155 }
3156
1da177e4
LT
3157 if (bond->params.mode == BOND_MODE_8023AD) {
3158 struct ad_info ad_info;
0795af57 3159 DECLARE_MAC_BUF(mac);
1da177e4
LT
3160
3161 seq_puts(seq, "\n802.3ad info\n");
3162 seq_printf(seq, "LACP rate: %s\n",
3163 (bond->params.lacp_fast) ? "fast" : "slow");
3164
3165 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3166 seq_printf(seq, "bond %s has no active aggregator\n",
3167 bond->dev->name);
3168 } else {
3169 seq_printf(seq, "Active Aggregator Info:\n");
3170
3171 seq_printf(seq, "\tAggregator ID: %d\n",
3172 ad_info.aggregator_id);
3173 seq_printf(seq, "\tNumber of ports: %d\n",
3174 ad_info.ports);
3175 seq_printf(seq, "\tActor Key: %d\n",
3176 ad_info.actor_key);
3177 seq_printf(seq, "\tPartner Key: %d\n",
3178 ad_info.partner_key);
0795af57
JP
3179 seq_printf(seq, "\tPartner Mac Address: %s\n",
3180 print_mac(mac, ad_info.partner_system));
1da177e4
LT
3181 }
3182 }
3183}
3184
3185static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3186{
3187 struct bonding *bond = seq->private;
0795af57 3188 DECLARE_MAC_BUF(mac);
1da177e4
LT
3189
3190 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3191 seq_printf(seq, "MII Status: %s\n",
3192 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3193 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT
3194 slave->link_failure_count);
3195
217df670 3196 seq_printf(seq,
0795af57
JP
3197 "Permanent HW addr: %s\n",
3198 print_mac(mac, slave->perm_hwaddr));
1da177e4
LT
3199
3200 if (bond->params.mode == BOND_MODE_8023AD) {
3201 const struct aggregator *agg
3202 = SLAVE_AD_INFO(slave).port.aggregator;
3203
3204 if (agg) {
3205 seq_printf(seq, "Aggregator ID: %d\n",
3206 agg->aggregator_identifier);
3207 } else {
3208 seq_puts(seq, "Aggregator ID: N/A\n");
3209 }
3210 }
3211}
3212
3213static int bond_info_seq_show(struct seq_file *seq, void *v)
3214{
3215 if (v == SEQ_START_TOKEN) {
3216 seq_printf(seq, "%s\n", version);
3217 bond_info_show_master(seq);
3218 } else {
3219 bond_info_show_slave(seq, v);
3220 }
3221
3222 return 0;
3223}
3224
3225static struct seq_operations bond_info_seq_ops = {
3226 .start = bond_info_seq_start,
3227 .next = bond_info_seq_next,
3228 .stop = bond_info_seq_stop,
3229 .show = bond_info_seq_show,
3230};
3231
3232static int bond_info_open(struct inode *inode, struct file *file)
3233{
3234 struct seq_file *seq;
3235 struct proc_dir_entry *proc;
3236 int res;
3237
3238 res = seq_open(file, &bond_info_seq_ops);
3239 if (!res) {
3240 /* recover the pointer buried in proc_dir_entry data */
3241 seq = file->private_data;
3242 proc = PDE(inode);
3243 seq->private = proc->data;
3244 }
3245
3246 return res;
3247}
3248
d54b1fdb 3249static const struct file_operations bond_info_fops = {
1da177e4
LT
3250 .owner = THIS_MODULE,
3251 .open = bond_info_open,
3252 .read = seq_read,
3253 .llseek = seq_lseek,
3254 .release = seq_release,
3255};
3256
3257static int bond_create_proc_entry(struct bonding *bond)
3258{
3259 struct net_device *bond_dev = bond->dev;
3260
3261 if (bond_proc_dir) {
a95609cb
DL
3262 bond->proc_entry = proc_create_data(bond_dev->name,
3263 S_IRUGO, bond_proc_dir,
3264 &bond_info_fops, bond);
1da177e4
LT
3265 if (bond->proc_entry == NULL) {
3266 printk(KERN_WARNING DRV_NAME
3267 ": Warning: Cannot create /proc/net/%s/%s\n",
3268 DRV_NAME, bond_dev->name);
3269 } else {
1da177e4
LT
3270 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3271 }
3272 }
3273
3274 return 0;
3275}
3276
3277static void bond_remove_proc_entry(struct bonding *bond)
3278{
3279 if (bond_proc_dir && bond->proc_entry) {
3280 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3281 memset(bond->proc_file_name, 0, IFNAMSIZ);
3282 bond->proc_entry = NULL;
3283 }
3284}
3285
3286/* Create the bonding directory under /proc/net, if doesn't exist yet.
3287 * Caller must hold rtnl_lock.
3288 */
3289static void bond_create_proc_dir(void)
3290{
3291 int len = strlen(DRV_NAME);
3292
457c4cbc 3293 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT
3294 bond_proc_dir = bond_proc_dir->next) {
3295 if ((bond_proc_dir->namelen == len) &&
3296 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3297 break;
3298 }
3299 }
3300
3301 if (!bond_proc_dir) {
457c4cbc 3302 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT
3303 if (bond_proc_dir) {
3304 bond_proc_dir->owner = THIS_MODULE;
3305 } else {
3306 printk(KERN_WARNING DRV_NAME
3307 ": Warning: cannot create /proc/net/%s\n",
3308 DRV_NAME);
3309 }
3310 }
3311}
3312
3313/* Destroy the bonding directory under /proc/net, if empty.
3314 * Caller must hold rtnl_lock.
3315 */
3316static void bond_destroy_proc_dir(void)
3317{
3318 struct proc_dir_entry *de;
3319
3320 if (!bond_proc_dir) {
3321 return;
3322 }
3323
3324 /* verify that the /proc dir is empty */
3325 for (de = bond_proc_dir->subdir; de; de = de->next) {
3326 /* ignore . and .. */
3327 if (*(de->name) != '.') {
3328 break;
3329 }
3330 }
3331
3332 if (de) {
3333 if (bond_proc_dir->owner == THIS_MODULE) {
3334 bond_proc_dir->owner = NULL;
3335 }
3336 } else {
457c4cbc 3337 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT
3338 bond_proc_dir = NULL;
3339 }
3340}
3341#endif /* CONFIG_PROC_FS */
3342
3343/*-------------------------- netdev event handling --------------------------*/
3344
3345/*
3346 * Change device name
3347 */
3348static int bond_event_changename(struct bonding *bond)
3349{
3350#ifdef CONFIG_PROC_FS
3351 bond_remove_proc_entry(bond);
3352 bond_create_proc_entry(bond);
3353#endif
b76cdba9
MW
3354 down_write(&(bonding_rwsem));
3355 bond_destroy_sysfs_entry(bond);
3356 bond_create_sysfs_entry(bond);
3357 up_write(&(bonding_rwsem));
1da177e4
LT
3358 return NOTIFY_DONE;
3359}
3360
3361static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3362{
3363 struct bonding *event_bond = bond_dev->priv;
3364
3365 switch (event) {
3366 case NETDEV_CHANGENAME:
3367 return bond_event_changename(event_bond);
3368 case NETDEV_UNREGISTER:
3b96c858 3369 bond_release_all(event_bond->dev);
1da177e4
LT
3370 break;
3371 default:
3372 break;
3373 }
3374
3375 return NOTIFY_DONE;
3376}
3377
3378static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3379{
3380 struct net_device *bond_dev = slave_dev->master;
8531c5ff 3381 struct bonding *bond = bond_dev->priv;
1da177e4
LT
3382
3383 switch (event) {
3384 case NETDEV_UNREGISTER:
3385 if (bond_dev) {
1284cd3a
JV
3386 if (bond->setup_by_slave)
3387 bond_release_and_destroy(bond_dev, slave_dev);
3388 else
3389 bond_release(bond_dev, slave_dev);
1da177e4
LT
3390 }
3391 break;
3392 case NETDEV_CHANGE:
3393 /*
3394 * TODO: is this what we get if somebody
3395 * sets up a hierarchical bond, then rmmod's
3396 * one of the slave bonding devices?
3397 */
3398 break;
3399 case NETDEV_DOWN:
3400 /*
3401 * ... Or is it this?
3402 */
3403 break;
3404 case NETDEV_CHANGEMTU:
3405 /*
3406 * TODO: Should slaves be allowed to
3407 * independently alter their MTU? For
3408 * an active-backup bond, slaves need
3409 * not be the same type of device, so
3410 * MTUs may vary. For other modes,
3411 * slaves arguably should have the
3412 * same MTUs. To do this, we'd need to
3413 * take over the slave's change_mtu
3414 * function for the duration of their
3415 * servitude.
3416 */
3417 break;
3418 case NETDEV_CHANGENAME:
3419 /*
3420 * TODO: handle changing the primary's name
3421 */
3422 break;
8531c5ff
AK
3423 case NETDEV_FEAT_CHANGE:
3424 bond_compute_features(bond);
3425 break;
1da177e4
LT
3426 default:
3427 break;
3428 }
3429
3430 return NOTIFY_DONE;
3431}
3432
3433/*
3434 * bond_netdev_event: handle netdev notifier chain events.
3435 *
3436 * This function receives events for the netdev chain. The caller (an
e041c683 3437 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT
3438 * locks for us to safely manipulate the slave devices (RTNL lock,
3439 * dev_probe_lock).
3440 */
3441static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3442{
3443 struct net_device *event_dev = (struct net_device *)ptr;
3444
c346dca1 3445 if (dev_net(event_dev) != &init_net)
e9dc8653
EB
3446 return NOTIFY_DONE;
3447
1da177e4
LT
3448 dprintk("event_dev: %s, event: %lx\n",
3449 (event_dev ? event_dev->name : "None"),
3450 event);
3451
0b680e75
JV
3452 if (!(event_dev->priv_flags & IFF_BONDING))
3453 return NOTIFY_DONE;
3454
1da177e4
LT
3455 if (event_dev->flags & IFF_MASTER) {
3456 dprintk("IFF_MASTER\n");
3457 return bond_master_netdev_event(event, event_dev);
3458 }
3459
3460 if (event_dev->flags & IFF_SLAVE) {
3461 dprintk("IFF_SLAVE\n");
3462 return bond_slave_netdev_event(event, event_dev);
3463 }
3464
3465 return NOTIFY_DONE;
3466}
3467
c3ade5ca
JV
3468/*
3469 * bond_inetaddr_event: handle inetaddr notifier chain events.
3470 *
3471 * We keep track of device IPs primarily to use as source addresses in
3472 * ARP monitor probes (rather than spewing out broadcasts all the time).
3473 *
3474 * We track one IP for the main device (if it has one), plus one per VLAN.
3475 */
3476static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3477{
3478 struct in_ifaddr *ifa = ptr;
3479 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3480 struct bonding *bond, *bond_next;
3481 struct vlan_entry *vlan, *vlan_next;
3482
c346dca1 3483 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL
3484 return NOTIFY_DONE;
3485
c3ade5ca
JV
3486 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3487 if (bond->dev == event_dev) {
3488 switch (event) {
3489 case NETDEV_UP:
3490 bond->master_ip = ifa->ifa_local;
3491 return NOTIFY_OK;
3492 case NETDEV_DOWN:
3493 bond->master_ip = bond_glean_dev_ip(bond->dev);
3494 return NOTIFY_OK;
3495 default:
3496 return NOTIFY_DONE;
3497 }
3498 }
3499
3500 if (list_empty(&bond->vlan_list))
3501 continue;
3502
3503 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3504 vlan_list) {
5c15bdec 3505 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV
3506 if (vlan_dev == event_dev) {
3507 switch (event) {
3508 case NETDEV_UP:
3509 vlan->vlan_ip = ifa->ifa_local;
3510 return NOTIFY_OK;
3511 case NETDEV_DOWN:
3512 vlan->vlan_ip =
3513 bond_glean_dev_ip(vlan_dev);
3514 return NOTIFY_OK;
3515 default:
3516 return NOTIFY_DONE;
3517 }
3518 }
3519 }
3520 }
3521 return NOTIFY_DONE;
3522}
3523
1da177e4
LT
3524static struct notifier_block bond_netdev_notifier = {
3525 .notifier_call = bond_netdev_event,
3526};
3527
c3ade5ca
JV
3528static struct notifier_block bond_inetaddr_notifier = {
3529 .notifier_call = bond_inetaddr_event,
3530};
3531
1da177e4
LT
3532/*-------------------------- Packet type handling ---------------------------*/
3533
3534/* register to receive lacpdus on a bond */
3535static void bond_register_lacpdu(struct bonding *bond)
3536{
3537 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3538
3539 /* initialize packet type */
3540 pk_type->type = PKT_TYPE_LACPDU;
3541 pk_type->dev = bond->dev;
3542 pk_type->func = bond_3ad_lacpdu_recv;
3543
3544 dev_add_pack(pk_type);
3545}
3546
3547/* unregister to receive lacpdus on a bond */
3548static void bond_unregister_lacpdu(struct bonding *bond)
3549{
3550 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3551}
3552
f5b2b966
JV
3553void bond_register_arp(struct bonding *bond)
3554{
3555 struct packet_type *pt = &bond->arp_mon_pt;
3556
c4f283b1
JV
3557 if (pt->type)
3558 return;
3559
f5b2b966 3560 pt->type = htons(ETH_P_ARP);
e245cb71 3561 pt->dev = bond->dev;
f5b2b966
JV
3562 pt->func = bond_arp_rcv;
3563 dev_add_pack(pt);
3564}
3565
3566void bond_unregister_arp(struct bonding *bond)
3567{
c4f283b1
JV
3568 struct packet_type *pt = &bond->arp_mon_pt;
3569
3570 dev_remove_pack(pt);
3571 pt->type = 0;
f5b2b966
JV
3572}
3573
169a3e66
JV
3574/*---------------------------- Hashing Policies -----------------------------*/
3575
6f6652be
JV
3576/*
3577 * Hash for the output device based upon layer 2 and layer 3 data. If
3578 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3579 */
3580static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3581 struct net_device *bond_dev, int count)
3582{
3583 struct ethhdr *data = (struct ethhdr *)skb->data;
3584 struct iphdr *iph = ip_hdr(skb);
3585
3586 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3587 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3588 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3589 }
3590
3591 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3592}
3593
169a3e66 3594/*
59c51591 3595 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV
3596 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3597 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3598 */
3599static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3600 struct net_device *bond_dev, int count)
3601{
3602 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3603 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3604 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV
3605 int layer4_xor = 0;
3606
3607 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3608 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3609 (iph->protocol == IPPROTO_TCP ||
3610 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3611 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV
3612 }
3613 return (layer4_xor ^
3614 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3615
3616 }
3617
3618 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3619}
3620
3621/*
3622 * Hash for the output device based upon layer 2 data
3623 */
3624static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3625 struct net_device *bond_dev, int count)
3626{
3627 struct ethhdr *data = (struct ethhdr *)skb->data;
3628
3629 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3630}
3631
1da177e4
LT
3632/*-------------------------- Device entry points ----------------------------*/
3633
3634static int bond_open(struct net_device *bond_dev)
3635{
3636 struct bonding *bond = bond_dev->priv;
1da177e4
LT
3637
3638 bond->kill_timers = 0;
3639
3640 if ((bond->params.mode == BOND_MODE_TLB) ||
3641 (bond->params.mode == BOND_MODE_ALB)) {
1da177e4
LT
3642 /* bond_alb_initialize must be called before the timer
3643 * is started.
3644 */
3645 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3646 /* something went wrong - fail the open operation */
3647 return -1;
3648 }
3649
1b76b316
JV
3650 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3651 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT
3652 }
3653
3654 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV
3655 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3656 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT
3657 }
3658
3659 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV
3660 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3661 INIT_DELAYED_WORK(&bond->arp_work,
3662 bond_activebackup_arp_mon);
3663 else
3664 INIT_DELAYED_WORK(&bond->arp_work,
3665 bond_loadbalance_arp_mon);
3666
3667 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV
3668 if (bond->params.arp_validate)
3669 bond_register_arp(bond);
1da177e4
LT
3670 }
3671
3672 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3673 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3674 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT
3675 /* register to receive LACPDUs */
3676 bond_register_lacpdu(bond);
3677 }
3678
3679 return 0;
3680}
3681
3682static int bond_close(struct net_device *bond_dev)
3683{
3684 struct bonding *bond = bond_dev->priv;
3685
3686 if (bond->params.mode == BOND_MODE_8023AD) {
3687 /* Unregister the receive of LACPDUs */
3688 bond_unregister_lacpdu(bond);
3689 }
3690
f5b2b966
JV
3691 if (bond->params.arp_validate)
3692 bond_unregister_arp(bond);
3693
1da177e4
LT
3694 write_lock_bh(&bond->lock);
3695
1da177e4
LT
3696
3697 /* signal timers not to re-arm */
3698 bond->kill_timers = 1;
3699
3700 write_unlock_bh(&bond->lock);
3701
1da177e4 3702 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3703 cancel_delayed_work(&bond->mii_work);
1da177e4
LT
3704 }
3705
3706 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3707 cancel_delayed_work(&bond->arp_work);
1da177e4
LT
3708 }
3709
3710 switch (bond->params.mode) {
3711 case BOND_MODE_8023AD:
1b76b316 3712 cancel_delayed_work(&bond->ad_work);
1da177e4
LT
3713 break;
3714 case BOND_MODE_TLB:
3715 case BOND_MODE_ALB:
1b76b316 3716 cancel_delayed_work(&bond->alb_work);
1da177e4
LT
3717 break;
3718 default:
3719 break;
3720 }
3721
1da177e4
LT
3722
3723 if ((bond->params.mode == BOND_MODE_TLB) ||
3724 (bond->params.mode == BOND_MODE_ALB)) {
3725 /* Must be called only after all
3726 * slaves have been released
3727 */
3728 bond_alb_deinitialize(bond);
3729 }
3730
3731 return 0;
3732}
3733
3734static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3735{
3736 struct bonding *bond = bond_dev->priv;
3737 struct net_device_stats *stats = &(bond->stats), *sstats;
2439f9eb 3738 struct net_device_stats local_stats;
1da177e4
LT
3739 struct slave *slave;
3740 int i;
3741
2439f9eb 3742 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT
3743
3744 read_lock_bh(&bond->lock);
3745
3746 bond_for_each_slave(bond, slave, i) {
c45d286e 3747 sstats = slave->dev->get_stats(slave->dev);
2439f9eb
AG
3748 local_stats.rx_packets += sstats->rx_packets;
3749 local_stats.rx_bytes += sstats->rx_bytes;
3750 local_stats.rx_errors += sstats->rx_errors;
3751 local_stats.rx_dropped += sstats->rx_dropped;
3752
3753 local_stats.tx_packets += sstats->tx_packets;
3754 local_stats.tx_bytes += sstats->tx_bytes;
3755 local_stats.tx_errors += sstats->tx_errors;
3756 local_stats.tx_dropped += sstats->tx_dropped;
3757
3758 local_stats.multicast += sstats->multicast;
3759 local_stats.collisions += sstats->collisions;
3760
3761 local_stats.rx_length_errors += sstats->rx_length_errors;
3762 local_stats.rx_over_errors += sstats->rx_over_errors;
3763 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3764 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3765 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3766 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3767
3768 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3769 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3770 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3771 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3772 local_stats.tx_window_errors += sstats->tx_window_errors;
3773 }
3774
3775 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT
3776
3777 read_unlock_bh(&bond->lock);
3778
3779 return stats;
3780}
3781
3782static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3783{
3784 struct net_device *slave_dev = NULL;
3785 struct ifbond k_binfo;
3786 struct ifbond __user *u_binfo = NULL;
3787 struct ifslave k_sinfo;
3788 struct ifslave __user *u_sinfo = NULL;
3789 struct mii_ioctl_data *mii = NULL;
1da177e4
LT
3790 int res = 0;
3791
3792 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3793 bond_dev->name, cmd);
3794
3795 switch (cmd) {
1da177e4
LT
3796 case SIOCGMIIPHY:
3797 mii = if_mii(ifr);
3798 if (!mii) {
3799 return -EINVAL;
3800 }
3801 mii->phy_id = 0;
3802 /* Fall Through */
3803 case SIOCGMIIREG:
3804 /*
3805 * We do this again just in case we were called by SIOCGMIIREG
3806 * instead of SIOCGMIIPHY.
3807 */
3808 mii = if_mii(ifr);
3809 if (!mii) {
3810 return -EINVAL;
3811 }
3812
3813 if (mii->reg_num == 1) {
3814 struct bonding *bond = bond_dev->priv;
3815 mii->val_out = 0;
6603a6f2 3816 read_lock(&bond->lock);
1da177e4 3817 read_lock(&bond->curr_slave_lock);
4e140079 3818 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT
3819 mii->val_out = BMSR_LSTATUS;
3820 }
3821 read_unlock(&bond->curr_slave_lock);
6603a6f2 3822 read_unlock(&bond->lock);
1da177e4
LT
3823 }
3824
3825 return 0;
3826 case BOND_INFO_QUERY_OLD:
3827 case SIOCBONDINFOQUERY:
3828 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3829
3830 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3831 return -EFAULT;
3832 }
3833
3834 res = bond_info_query(bond_dev, &k_binfo);
3835 if (res == 0) {
3836 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3837 return -EFAULT;
3838 }
3839 }
3840
3841 return res;
3842 case BOND_SLAVE_INFO_QUERY_OLD:
3843 case SIOCBONDSLAVEINFOQUERY:
3844 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3845
3846 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3847 return -EFAULT;
3848 }
3849
3850 res = bond_slave_info_query(bond_dev, &k_sinfo);
3851 if (res == 0) {
3852 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3853 return -EFAULT;
3854 }
3855 }
3856
3857 return res;
3858 default:
3859 /* Go on */
3860 break;
3861 }
3862
3863 if (!capable(CAP_NET_ADMIN)) {
3864 return -EPERM;
3865 }
3866
b76cdba9 3867 down_write(&(bonding_rwsem));
881d966b 3868 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT
3869
3870 dprintk("slave_dev=%p: \n", slave_dev);
3871
3872 if (!slave_dev) {
3873 res = -ENODEV;
3874 } else {
3875 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3876 switch (cmd) {
3877 case BOND_ENSLAVE_OLD:
3878 case SIOCBONDENSLAVE:
3879 res = bond_enslave(bond_dev, slave_dev);
3880 break;
3881 case BOND_RELEASE_OLD:
3882 case SIOCBONDRELEASE:
3883 res = bond_release(bond_dev, slave_dev);
3884 break;
3885 case BOND_SETHWADDR_OLD:
3886 case SIOCBONDSETHWADDR:
3887 res = bond_sethwaddr(bond_dev, slave_dev);
3888 break;
3889 case BOND_CHANGE_ACTIVE_OLD:
3890 case SIOCBONDCHANGEACTIVE:
3891 res = bond_ioctl_change_active(bond_dev, slave_dev);
3892 break;
3893 default:
3894 res = -EOPNOTSUPP;
3895 }
3896
3897 dev_put(slave_dev);
3898 }
3899
b76cdba9 3900 up_write(&(bonding_rwsem));
1da177e4
LT
3901 return res;
3902}
3903
3904static void bond_set_multicast_list(struct net_device *bond_dev)
3905{
3906 struct bonding *bond = bond_dev->priv;
3907 struct dev_mc_list *dmi;
3908
1da177e4
LT
3909 /*
3910 * Do promisc before checking multicast_mode
3911 */
3912 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3913 bond_set_promiscuity(bond, 1);
3914 }
3915
3916 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3917 bond_set_promiscuity(bond, -1);
3918 }
3919
3920 /* set allmulti flag to slaves */
3921 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3922 bond_set_allmulti(bond, 1);
3923 }
3924
3925 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3926 bond_set_allmulti(bond, -1);
3927 }
3928
80ee5ad2
JV
3929 read_lock(&bond->lock);
3930
1da177e4
LT
3931 bond->flags = bond_dev->flags;
3932
3933 /* looking for addresses to add to slaves' mc list */
3934 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3935 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3936 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3937 }
3938 }
3939
3940 /* looking for addresses to delete from slaves' list */
3941 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3942 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3943 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3944 }
3945 }
3946
3947 /* save master's multicast list */
3948 bond_mc_list_destroy(bond);
3949 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3950
80ee5ad2 3951 read_unlock(&bond->lock);
1da177e4
LT
3952}
3953
3954/*
3955 * Change the MTU of all of a master's slaves to match the master
3956 */
3957static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3958{
3959 struct bonding *bond = bond_dev->priv;
3960 struct slave *slave, *stop_at;
3961 int res = 0;
3962 int i;
3963
3964 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3965 (bond_dev ? bond_dev->name : "None"), new_mtu);
3966
3967 /* Can't hold bond->lock with bh disabled here since
3968 * some base drivers panic. On the other hand we can't
3969 * hold bond->lock without bh disabled because we'll
3970 * deadlock. The only solution is to rely on the fact
3971 * that we're under rtnl_lock here, and the slaves
3972 * list won't change. This doesn't solve the problem
3973 * of setting the slave's MTU while it is
3974 * transmitting, but the assumption is that the base
3975 * driver can handle that.
3976 *
3977 * TODO: figure out a way to safely iterate the slaves
3978 * list, but without holding a lock around the actual
3979 * call to the base driver.
3980 */
3981
3982 bond_for_each_slave(bond, slave, i) {
3983 dprintk("s %p s->p %p c_m %p\n", slave,
3984 slave->prev, slave->dev->change_mtu);
e944ef79 3985
1da177e4
LT
3986 res = dev_set_mtu(slave->dev, new_mtu);
3987
3988 if (res) {
3989 /* If we failed to set the slave's mtu to the new value
3990 * we must abort the operation even in ACTIVE_BACKUP
3991 * mode, because if we allow the backup slaves to have
3992 * different mtu values than the active slave we'll
3993 * need to change their mtu when doing a failover. That
3994 * means changing their mtu from timer context, which
3995 * is probably not a good idea.
3996 */
3997 dprintk("err %d %s\n", res, slave->dev->name);
3998 goto unwind;
3999 }
4000 }
4001
4002 bond_dev->mtu = new_mtu;
4003
4004 return 0;
4005
4006unwind:
4007 /* unwind from head to the slave that failed */
4008 stop_at = slave;
4009 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4010 int tmp_res;
4011
4012 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4013 if (tmp_res) {
4014 dprintk("unwind err %d dev %s\n", tmp_res,
4015 slave->dev->name);
4016 }
4017 }
4018
4019 return res;
4020}
4021
4022/*
4023 * Change HW address
4024 *
4025 * Note that many devices must be down to change the HW address, and
4026 * downing the master releases all slaves. We can make bonds full of
4027 * bonding devices to test this, however.
4028 */
4029static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4030{
4031 struct bonding *bond = bond_dev->priv;
4032 struct sockaddr *sa = addr, tmp_sa;
4033 struct slave *slave, *stop_at;
4034 int res = 0;
4035 int i;
4036
4037 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4038
dd957c57
JV
4039 /*
4040 * If fail_over_mac is enabled, do nothing and return success.
4041 * Returning an error causes ifenslave to fail.
4042 */
4043 if (bond->params.fail_over_mac)
4044 return 0;
2ab82852 4045
1da177e4
LT
4046 if (!is_valid_ether_addr(sa->sa_data)) {
4047 return -EADDRNOTAVAIL;
4048 }
4049
4050 /* Can't hold bond->lock with bh disabled here since
4051 * some base drivers panic. On the other hand we can't
4052 * hold bond->lock without bh disabled because we'll
4053 * deadlock. The only solution is to rely on the fact
4054 * that we're under rtnl_lock here, and the slaves
4055 * list won't change. This doesn't solve the problem
4056 * of setting the slave's hw address while it is
4057 * transmitting, but the assumption is that the base
4058 * driver can handle that.
4059 *
4060 * TODO: figure out a way to safely iterate the slaves
4061 * list, but without holding a lock around the actual
4062 * call to the base driver.
4063 */
4064
4065 bond_for_each_slave(bond, slave, i) {
4066 dprintk("slave %p %s\n", slave, slave->dev->name);
4067
4068 if (slave->dev->set_mac_address == NULL) {
4069 res = -EOPNOTSUPP;
4070 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4071 goto unwind;
4072 }
4073
4074 res = dev_set_mac_address(slave->dev, addr);
4075 if (res) {
4076 /* TODO: consider downing the slave
4077 * and retry ?
4078 * User should expect communications
4079 * breakage anyway until ARP finish
4080 * updating, so...
4081 */
4082 dprintk("err %d %s\n", res, slave->dev->name);
4083 goto unwind;
4084 }
4085 }
4086
4087 /* success */
4088 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4089 return 0;
4090
4091unwind:
4092 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4093 tmp_sa.sa_family = bond_dev->type;
4094
4095 /* unwind from head to the slave that failed */
4096 stop_at = slave;
4097 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4098 int tmp_res;
4099
4100 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4101 if (tmp_res) {
4102 dprintk("unwind err %d dev %s\n", tmp_res,
4103 slave->dev->name);
4104 }
4105 }
4106
4107 return res;
4108}
4109
4110static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4111{
4112 struct bonding *bond = bond_dev->priv;
4113 struct slave *slave, *start_at;
cf5f9044 4114 int i, slave_no, res = 1;
1da177e4
LT
4115
4116 read_lock(&bond->lock);
4117
4118 if (!BOND_IS_OK(bond)) {
4119 goto out;
4120 }
4121
cf5f9044
JV
4122 /*
4123 * Concurrent TX may collide on rr_tx_counter; we accept that
4124 * as being rare enough not to justify using an atomic op here
4125 */
4126 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4127
cf5f9044
JV
4128 bond_for_each_slave(bond, slave, i) {
4129 slave_no--;
4130 if (slave_no < 0) {
4131 break;
4132 }
1da177e4
LT
4133 }
4134
cf5f9044 4135 start_at = slave;
1da177e4
LT
4136 bond_for_each_slave_from(bond, slave, i, start_at) {
4137 if (IS_UP(slave->dev) &&
4138 (slave->link == BOND_LINK_UP) &&
4139 (slave->state == BOND_STATE_ACTIVE)) {
4140 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT
4141 break;
4142 }
4143 }
4144
1da177e4
LT
4145out:
4146 if (res) {
4147 /* no suitable interface, frame not sent */
4148 dev_kfree_skb(skb);
4149 }
4150 read_unlock(&bond->lock);
4151 return 0;
4152}
4153
075897ce 4154
1da177e4
LT
4155/*
4156 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4157 * the bond has a usable interface.
4158 */
4159static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4160{
4161 struct bonding *bond = bond_dev->priv;
4162 int res = 1;
4163
1da177e4
LT
4164 read_lock(&bond->lock);
4165 read_lock(&bond->curr_slave_lock);
4166
4167 if (!BOND_IS_OK(bond)) {
4168 goto out;
4169 }
4170
075897ce
JL
4171 if (!bond->curr_active_slave)
4172 goto out;
4173
075897ce
JL
4174 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4175
1da177e4
LT
4176out:
4177 if (res) {
4178 /* no suitable interface, frame not sent */
4179 dev_kfree_skb(skb);
4180 }
4181 read_unlock(&bond->curr_slave_lock);
4182 read_unlock(&bond->lock);
4183 return 0;
4184}
4185
4186/*
169a3e66
JV
4187 * In bond_xmit_xor() , we determine the output device by using a pre-
4188 * determined xmit_hash_policy(), If the selected device is not enabled,
4189 * find the next active slave.
1da177e4
LT
4190 */
4191static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4192{
4193 struct bonding *bond = bond_dev->priv;
1da177e4
LT
4194 struct slave *slave, *start_at;
4195 int slave_no;
4196 int i;
4197 int res = 1;
4198
4199 read_lock(&bond->lock);
4200
4201 if (!BOND_IS_OK(bond)) {
4202 goto out;
4203 }
4204
169a3e66 4205 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT
4206
4207 bond_for_each_slave(bond, slave, i) {
4208 slave_no--;
4209 if (slave_no < 0) {
4210 break;
4211 }
4212 }
4213
4214 start_at = slave;
4215
4216 bond_for_each_slave_from(bond, slave, i, start_at) {
4217 if (IS_UP(slave->dev) &&
4218 (slave->link == BOND_LINK_UP) &&
4219 (slave->state == BOND_STATE_ACTIVE)) {
4220 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4221 break;
4222 }
4223 }
4224
4225out:
4226 if (res) {
4227 /* no suitable interface, frame not sent */
4228 dev_kfree_skb(skb);
4229 }
4230 read_unlock(&bond->lock);
4231 return 0;
4232}
4233
4234/*
4235 * in broadcast mode, we send everything to all usable interfaces.
4236 */
4237static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4238{
4239 struct bonding *bond = bond_dev->priv;
4240 struct slave *slave, *start_at;
4241 struct net_device *tx_dev = NULL;
4242 int i;
4243 int res = 1;
4244
4245 read_lock(&bond->lock);
4246
4247 if (!BOND_IS_OK(bond)) {
4248 goto out;
4249 }
4250
4251 read_lock(&bond->curr_slave_lock);
4252 start_at = bond->curr_active_slave;
4253 read_unlock(&bond->curr_slave_lock);
4254
4255 if (!start_at) {
4256 goto out;
4257 }
4258
4259 bond_for_each_slave_from(bond, slave, i, start_at) {
4260 if (IS_UP(slave->dev) &&
4261 (slave->link == BOND_LINK_UP) &&
4262 (slave->state == BOND_STATE_ACTIVE)) {
4263 if (tx_dev) {
4264 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4265 if (!skb2) {
4266 printk(KERN_ERR DRV_NAME
4e0952c7
MW
4267 ": %s: Error: bond_xmit_broadcast(): "
4268 "skb_clone() failed\n",
4269 bond_dev->name);
1da177e4
LT
4270 continue;
4271 }
4272
4273 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4274 if (res) {
4275 dev_kfree_skb(skb2);
4276 continue;
4277 }
4278 }
4279 tx_dev = slave->dev;
4280 }
4281 }
4282
4283 if (tx_dev) {
4284 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4285 }
4286
4287out:
4288 if (res) {
4289 /* no suitable interface, frame not sent */
4290 dev_kfree_skb(skb);
4291 }
4292 /* frame sent to all suitable interfaces */
4293 read_unlock(&bond->lock);
4294 return 0;
4295}
4296
4297/*------------------------- Device initialization ---------------------------*/
4298
6f6652be
JV
4299static void bond_set_xmit_hash_policy(struct bonding *bond)
4300{
4301 switch (bond->params.xmit_policy) {
4302 case BOND_XMIT_POLICY_LAYER23:
4303 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4304 break;
4305 case BOND_XMIT_POLICY_LAYER34:
4306 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4307 break;
4308 case BOND_XMIT_POLICY_LAYER2:
4309 default:
4310 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4311 break;
4312 }
4313}
4314
1da177e4
LT
4315/*
4316 * set bond mode specific net device operations
4317 */
a77b5325 4318void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4319{
169a3e66
JV
4320 struct net_device *bond_dev = bond->dev;
4321
1da177e4
LT
4322 switch (mode) {
4323 case BOND_MODE_ROUNDROBIN:
4324 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4325 break;
4326 case BOND_MODE_ACTIVEBACKUP:
4327 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4328 break;
4329 case BOND_MODE_XOR:
4330 bond_dev->hard_start_xmit = bond_xmit_xor;
6f6652be 4331 bond_set_xmit_hash_policy(bond);
1da177e4
LT
4332 break;
4333 case BOND_MODE_BROADCAST:
4334 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4335 break;
4336 case BOND_MODE_8023AD:
8f903c70 4337 bond_set_master_3ad_flags(bond);
1da177e4 4338 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
6f6652be 4339 bond_set_xmit_hash_policy(bond);
1da177e4 4340 break;
1da177e4 4341 case BOND_MODE_ALB:
8f903c70
JV
4342 bond_set_master_alb_flags(bond);
4343 /* FALLTHRU */
4344 case BOND_MODE_TLB:
1da177e4
LT
4345 bond_dev->hard_start_xmit = bond_alb_xmit;
4346 bond_dev->set_mac_address = bond_alb_set_mac_address;
4347 break;
4348 default:
4349 /* Should never happen, mode already checked */
4350 printk(KERN_ERR DRV_NAME
4e0952c7
MW
4351 ": %s: Error: Unknown bonding mode %d\n",
4352 bond_dev->name,
1da177e4
LT
4353 mode);
4354 break;
4355 }
4356}
4357
217df670
JV
4358static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4359 struct ethtool_drvinfo *drvinfo)
4360{
4361 strncpy(drvinfo->driver, DRV_NAME, 32);
4362 strncpy(drvinfo->version, DRV_VERSION, 32);
4363 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4364}
4365
7282d491 4366static const struct ethtool_ops bond_ethtool_ops = {
217df670 4367 .get_drvinfo = bond_ethtool_get_drvinfo,
8531c5ff
AK
4368};
4369
1da177e4
LT
4370/*
4371 * Does not allocate but creates a /proc entry.
4372 * Allowed to fail.
4373 */
3c535952 4374static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4
LT
4375{
4376 struct bonding *bond = bond_dev->priv;
4377
4378 dprintk("Begin bond_init for %s\n", bond_dev->name);
4379
4380 /* initialize rwlocks */
4381 rwlock_init(&bond->lock);
4382 rwlock_init(&bond->curr_slave_lock);
4383
4384 bond->params = *params; /* copy params struct */
4385
1b76b316
JV
4386 bond->wq = create_singlethread_workqueue(bond_dev->name);
4387 if (!bond->wq)
4388 return -ENOMEM;
4389
1da177e4
LT
4390 /* Initialize pointers */
4391 bond->first_slave = NULL;
4392 bond->curr_active_slave = NULL;
4393 bond->current_arp_slave = NULL;
4394 bond->primary_slave = NULL;
4395 bond->dev = bond_dev;
1053f62c 4396 bond->send_grat_arp = 0;
d90a162a 4397 bond->setup_by_slave = 0;
1da177e4
LT
4398 INIT_LIST_HEAD(&bond->vlan_list);
4399
4400 /* Initialize the device entry points */
4401 bond_dev->open = bond_open;
4402 bond_dev->stop = bond_close;
4403 bond_dev->get_stats = bond_get_stats;
4404 bond_dev->do_ioctl = bond_do_ioctl;
8531c5ff 4405 bond_dev->ethtool_ops = &bond_ethtool_ops;
1da177e4
LT
4406 bond_dev->set_multicast_list = bond_set_multicast_list;
4407 bond_dev->change_mtu = bond_change_mtu;
4408 bond_dev->set_mac_address = bond_set_mac_address;
3a1521b7 4409 bond_dev->validate_addr = NULL;
1da177e4 4410
169a3e66 4411 bond_set_mode_ops(bond, bond->params.mode);
1da177e4
LT
4412
4413 bond_dev->destructor = free_netdev;
4414
4415 /* Initialize the device options */
4416 bond_dev->tx_queue_len = 0;
4417 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4418 bond_dev->priv_flags |= IFF_BONDING;
1da177e4
LT
4419
4420 /* At first, we block adding VLANs. That's the only way to
4421 * prevent problems that occur when adding VLANs over an
4422 * empty bond. The block will be removed once non-challenged
4423 * slaves are enslaved.
4424 */
4425 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4426
932ff279 4427 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT
4428 * transmitting */
4429 bond_dev->features |= NETIF_F_LLTX;
4430
4431 /* By default, we declare the bond to be fully
4432 * VLAN hardware accelerated capable. Special
4433 * care is taken in the various xmit functions
4434 * when there are slaves that are not hw accel
4435 * capable
4436 */
4437 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4438 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4439 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4440 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4441 NETIF_F_HW_VLAN_RX |
4442 NETIF_F_HW_VLAN_FILTER);
4443
4444#ifdef CONFIG_PROC_FS
4445 bond_create_proc_entry(bond);
4446#endif
1da177e4
LT
4447 list_add_tail(&bond->bond_list, &bond_dev_list);
4448
4449 return 0;
4450}
4451
4452/* De-initialize device specific data.
4453 * Caller must hold rtnl_lock.
4454 */
c50b85d0 4455static void bond_deinit(struct net_device *bond_dev)
1da177e4
LT
4456{
4457 struct bonding *bond = bond_dev->priv;
4458
4459 list_del(&bond->bond_list);
4460
4461#ifdef CONFIG_PROC_FS
4462 bond_remove_proc_entry(bond);
4463#endif
4464}
4465
fdaea7a9
JV
4466static void bond_work_cancel_all(struct bonding *bond)
4467{
4468 write_lock_bh(&bond->lock);
4469 bond->kill_timers = 1;
4470 write_unlock_bh(&bond->lock);
4471
4472 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4473 cancel_delayed_work(&bond->mii_work);
4474
4475 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4476 cancel_delayed_work(&bond->arp_work);
4477
4478 if (bond->params.mode == BOND_MODE_ALB &&
4479 delayed_work_pending(&bond->alb_work))
4480 cancel_delayed_work(&bond->alb_work);
4481
4482 if (bond->params.mode == BOND_MODE_8023AD &&
4483 delayed_work_pending(&bond->ad_work))
4484 cancel_delayed_work(&bond->ad_work);
4485}
4486
1da177e4
LT
4487/* Unregister and free all bond devices.
4488 * Caller must hold rtnl_lock.
4489 */
4490static void bond_free_all(void)
4491{
4492 struct bonding *bond, *nxt;
4493
4494 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4495 struct net_device *bond_dev = bond->dev;
4496
fdaea7a9 4497 bond_work_cancel_all(bond);
80ee5ad2 4498 netif_tx_lock_bh(bond_dev);
70298705 4499 bond_mc_list_destroy(bond);
80ee5ad2 4500 netif_tx_unlock_bh(bond_dev);
70298705 4501 /* Release the bonded slaves */
4502 bond_release_all(bond_dev);
92b41daa 4503 bond_destroy(bond);
1da177e4
LT
4504 }
4505
4506#ifdef CONFIG_PROC_FS
4507 bond_destroy_proc_dir();
4508#endif
4509}
4510
4511/*------------------------- Module initialization ---------------------------*/
4512
4513/*
4514 * Convert string input module parms. Accept either the
ece95f7f
JV
4515 * number of the mode or its string name. A bit complicated because
4516 * some mode names are substrings of other names, and calls from sysfs
4517 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4518 */
ece95f7f 4519int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4520{
ece95f7f 4521 int mode = -1, i, rv;
a42e534f 4522 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4523
a42e534f
JV
4524 for (p = (char *)buf; *p; p++)
4525 if (!(isdigit(*p) || isspace(*p)))
4526 break;
4527
4528 if (*p)
ece95f7f 4529 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV
4530 else
4531 rv = sscanf(buf, "%d", &mode);
4532
4533 if (!rv)
4534 return -1;
1da177e4
LT
4535
4536 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV
4537 if (mode == tbl[i].mode)
4538 return tbl[i].mode;
4539 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4540 return tbl[i].mode;
1da177e4
LT
4541 }
4542
4543 return -1;
4544}
4545
4546static int bond_check_params(struct bond_params *params)
4547{
f5b2b966
JV
4548 int arp_validate_value;
4549
1da177e4
LT
4550 /*
4551 * Convert string parameters.
4552 */
4553 if (mode) {
4554 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4555 if (bond_mode == -1) {
4556 printk(KERN_ERR DRV_NAME
4557 ": Error: Invalid bonding mode \"%s\"\n",
4558 mode == NULL ? "NULL" : mode);
4559 return -EINVAL;
4560 }
4561 }
4562
169a3e66
JV
4563 if (xmit_hash_policy) {
4564 if ((bond_mode != BOND_MODE_XOR) &&
4565 (bond_mode != BOND_MODE_8023AD)) {
4566 printk(KERN_INFO DRV_NAME
4567 ": xor_mode param is irrelevant in mode %s\n",
4568 bond_mode_name(bond_mode));
4569 } else {
4570 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4571 xmit_hashtype_tbl);
4572 if (xmit_hashtype == -1) {
4573 printk(KERN_ERR DRV_NAME
4574 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4575 xmit_hash_policy == NULL ? "NULL" :
4576 xmit_hash_policy);
4577 return -EINVAL;
4578 }
4579 }
4580 }
4581
1da177e4
LT
4582 if (lacp_rate) {
4583 if (bond_mode != BOND_MODE_8023AD) {
4584 printk(KERN_INFO DRV_NAME
4585 ": lacp_rate param is irrelevant in mode %s\n",
4586 bond_mode_name(bond_mode));
4587 } else {
4588 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4589 if (lacp_fast == -1) {
4590 printk(KERN_ERR DRV_NAME
4591 ": Error: Invalid lacp rate \"%s\"\n",
4592 lacp_rate == NULL ? "NULL" : lacp_rate);
4593 return -EINVAL;
4594 }
4595 }
4596 }
4597
4598 if (max_bonds < 1 || max_bonds > INT_MAX) {
4599 printk(KERN_WARNING DRV_NAME
4600 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4601 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
1da177e4
LT
4602 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4603 max_bonds = BOND_DEFAULT_MAX_BONDS;
4604 }
4605
4606 if (miimon < 0) {
4607 printk(KERN_WARNING DRV_NAME
4608 ": Warning: miimon module parameter (%d), "
4609 "not in range 0-%d, so it was reset to %d\n",
4610 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4611 miimon = BOND_LINK_MON_INTERV;
4612 }
4613
4614 if (updelay < 0) {
4615 printk(KERN_WARNING DRV_NAME
4616 ": Warning: updelay module parameter (%d), "
4617 "not in range 0-%d, so it was reset to 0\n",
4618 updelay, INT_MAX);
4619 updelay = 0;
4620 }
4621
4622 if (downdelay < 0) {
4623 printk(KERN_WARNING DRV_NAME
4624 ": Warning: downdelay module parameter (%d), "
4625 "not in range 0-%d, so it was reset to 0\n",
4626 downdelay, INT_MAX);
4627 downdelay = 0;
4628 }
4629
4630 if ((use_carrier != 0) && (use_carrier != 1)) {
4631 printk(KERN_WARNING DRV_NAME
4632 ": Warning: use_carrier module parameter (%d), "
4633 "not of valid value (0/1), so it was set to 1\n",
4634 use_carrier);
4635 use_carrier = 1;
4636 }
4637
4638 /* reset values for 802.3ad */
4639 if (bond_mode == BOND_MODE_8023AD) {
4640 if (!miimon) {
4641 printk(KERN_WARNING DRV_NAME
4642 ": Warning: miimon must be specified, "
4643 "otherwise bonding will not detect link "
4644 "failure, speed and duplex which are "
4645 "essential for 802.3ad operation\n");
4646 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4647 miimon = 100;
4648 }
4649 }
4650
4651 /* reset values for TLB/ALB */
4652 if ((bond_mode == BOND_MODE_TLB) ||
4653 (bond_mode == BOND_MODE_ALB)) {
4654 if (!miimon) {
4655 printk(KERN_WARNING DRV_NAME
4656 ": Warning: miimon must be specified, "
4657 "otherwise bonding will not detect link "
4658 "failure and link speed which are essential "
4659 "for TLB/ALB load balancing\n");
4660 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4661 miimon = 100;
4662 }
4663 }
4664
4665 if (bond_mode == BOND_MODE_ALB) {
4666 printk(KERN_NOTICE DRV_NAME
4667 ": In ALB mode you might experience client "
4668 "disconnections upon reconnection of a link if the "
4669 "bonding module updelay parameter (%d msec) is "
4670 "incompatible with the forwarding delay time of the "
4671 "switch\n",
4672 updelay);
4673 }
4674
4675 if (!miimon) {
4676 if (updelay || downdelay) {
4677 /* just warn the user the up/down delay will have
4678 * no effect since miimon is zero...
4679 */
4680 printk(KERN_WARNING DRV_NAME
4681 ": Warning: miimon module parameter not set "
4682 "and updelay (%d) or downdelay (%d) module "
4683 "parameter is set; updelay and downdelay have "
4684 "no effect unless miimon is set\n",
4685 updelay, downdelay);
4686 }
4687 } else {
4688 /* don't allow arp monitoring */
4689 if (arp_interval) {
4690 printk(KERN_WARNING DRV_NAME
4691 ": Warning: miimon (%d) and arp_interval (%d) "
4692 "can't be used simultaneously, disabling ARP "
4693 "monitoring\n",
4694 miimon, arp_interval);
4695 arp_interval = 0;
4696 }
4697
4698 if ((updelay % miimon) != 0) {
4699 printk(KERN_WARNING DRV_NAME
4700 ": Warning: updelay (%d) is not a multiple "
4701 "of miimon (%d), updelay rounded to %d ms\n",
4702 updelay, miimon, (updelay / miimon) * miimon);
4703 }
4704
4705 updelay /= miimon;
4706
4707 if ((downdelay % miimon) != 0) {
4708 printk(KERN_WARNING DRV_NAME
4709 ": Warning: downdelay (%d) is not a multiple "
4710 "of miimon (%d), downdelay rounded to %d ms\n",
4711 downdelay, miimon,
4712 (downdelay / miimon) * miimon);
4713 }
4714
4715 downdelay /= miimon;
4716 }
4717
4718 if (arp_interval < 0) {
4719 printk(KERN_WARNING DRV_NAME
4720 ": Warning: arp_interval module parameter (%d) "
4721 ", not in range 0-%d, so it was reset to %d\n",
4722 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4723 arp_interval = BOND_LINK_ARP_INTERV;
4724 }
4725
4726 for (arp_ip_count = 0;
4727 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4728 arp_ip_count++) {
4729 /* not complete check, but should be good enough to
4730 catch mistakes */
4731 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4732 printk(KERN_WARNING DRV_NAME
4733 ": Warning: bad arp_ip_target module parameter "
4734 "(%s), ARP monitoring will not be performed\n",
4735 arp_ip_target[arp_ip_count]);
4736 arp_interval = 0;
4737 } else {
d3bb52b0 4738 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT
4739 arp_target[arp_ip_count] = ip;
4740 }
4741 }
4742
4743 if (arp_interval && !arp_ip_count) {
4744 /* don't allow arping if no arp_ip_target given... */
4745 printk(KERN_WARNING DRV_NAME
4746 ": Warning: arp_interval module parameter (%d) "
4747 "specified without providing an arp_ip_target "
4748 "parameter, arp_interval was reset to 0\n",
4749 arp_interval);
4750 arp_interval = 0;
4751 }
4752
f5b2b966
JV
4753 if (arp_validate) {
4754 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4755 printk(KERN_ERR DRV_NAME
4756 ": arp_validate only supported in active-backup mode\n");
4757 return -EINVAL;
4758 }
4759 if (!arp_interval) {
4760 printk(KERN_ERR DRV_NAME
4761 ": arp_validate requires arp_interval\n");
4762 return -EINVAL;
4763 }
4764
4765 arp_validate_value = bond_parse_parm(arp_validate,
4766 arp_validate_tbl);
4767 if (arp_validate_value == -1) {
4768 printk(KERN_ERR DRV_NAME
4769 ": Error: invalid arp_validate \"%s\"\n",
4770 arp_validate == NULL ? "NULL" : arp_validate);
4771 return -EINVAL;
4772 }
4773 } else
4774 arp_validate_value = 0;
4775
1da177e4
LT
4776 if (miimon) {
4777 printk(KERN_INFO DRV_NAME
4778 ": MII link monitoring set to %d ms\n",
4779 miimon);
4780 } else if (arp_interval) {
4781 int i;
4782
4783 printk(KERN_INFO DRV_NAME
f5b2b966
JV
4784 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4785 arp_interval,
4786 arp_validate_tbl[arp_validate_value].modename,
4787 arp_ip_count);
1da177e4
LT
4788
4789 for (i = 0; i < arp_ip_count; i++)
4790 printk (" %s", arp_ip_target[i]);
4791
4792 printk("\n");
4793
4794 } else {
4795 /* miimon and arp_interval not set, we need one so things
4796 * work as expected, see bonding.txt for details
4797 */
4798 printk(KERN_WARNING DRV_NAME
4799 ": Warning: either miimon or arp_interval and "
4800 "arp_ip_target module parameters must be specified, "
4801 "otherwise bonding will not detect link failures! see "
4802 "bonding.txt for details.\n");
4803 }
4804
4805 if (primary && !USES_PRIMARY(bond_mode)) {
4806 /* currently, using a primary only makes sense
4807 * in active backup, TLB or ALB modes
4808 */
4809 printk(KERN_WARNING DRV_NAME
4810 ": Warning: %s primary device specified but has no "
4811 "effect in %s mode\n",
4812 primary, bond_mode_name(bond_mode));
4813 primary = NULL;
4814 }
4815
dd957c57
JV
4816 if (fail_over_mac && (bond_mode != BOND_MODE_ACTIVEBACKUP))
4817 printk(KERN_WARNING DRV_NAME
4818 ": Warning: fail_over_mac only affects "
4819 "active-backup mode.\n");
4820
1da177e4
LT
4821 /* fill params struct with the proper values */
4822 params->mode = bond_mode;
169a3e66 4823 params->xmit_policy = xmit_hashtype;
1da177e4
LT
4824 params->miimon = miimon;
4825 params->arp_interval = arp_interval;
f5b2b966 4826 params->arp_validate = arp_validate_value;
1da177e4
LT
4827 params->updelay = updelay;
4828 params->downdelay = downdelay;
4829 params->use_carrier = use_carrier;
4830 params->lacp_fast = lacp_fast;
4831 params->primary[0] = 0;
dd957c57 4832 params->fail_over_mac = fail_over_mac;
1da177e4
LT
4833
4834 if (primary) {
4835 strncpy(params->primary, primary, IFNAMSIZ);
4836 params->primary[IFNAMSIZ - 1] = 0;
4837 }
4838
4839 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4840
4841 return 0;
4842}
4843
0daa2303
PZ
4844static struct lock_class_key bonding_netdev_xmit_lock_key;
4845
dfe60397 4846/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 4847 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW
4848 * Caller must NOT hold rtnl_lock; we need to release it here before we
4849 * set up our sysfs entries.
4850 */
0dd646fe 4851int bond_create(char *name, struct bond_params *params)
dfe60397
MW
4852{
4853 struct net_device *bond_dev;
027ea041 4854 struct bonding *bond, *nxt;
dfe60397
MW
4855 int res;
4856
4857 rtnl_lock();
027ea041
JV
4858 down_write(&bonding_rwsem);
4859
4860 /* Check to see if the bond already exists. */
4fe4763c
JV
4861 if (name) {
4862 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list)
4863 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
4864 printk(KERN_ERR DRV_NAME
027ea041 4865 ": cannot add bond %s; it already exists\n",
4fe4763c
JV
4866 name);
4867 res = -EPERM;
4868 goto out_rtnl;
4869 }
4870 }
027ea041 4871
e4b91c48
JV
4872 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4873 ether_setup);
dfe60397
MW
4874 if (!bond_dev) {
4875 printk(KERN_ERR DRV_NAME
4876 ": %s: eek! can't alloc netdev!\n",
4877 name);
4878 res = -ENOMEM;
4879 goto out_rtnl;
4880 }
4881
e4b91c48
JV
4882 if (!name) {
4883 res = dev_alloc_name(bond_dev, "bond%d");
4884 if (res < 0)
4885 goto out_netdev;
4886 }
4887
dfe60397
MW
4888 /* bond_init() must be called after dev_alloc_name() (for the
4889 * /proc files), but before register_netdevice(), because we
4890 * need to set function pointers.
4891 */
4892
4893 res = bond_init(bond_dev, params);
4894 if (res < 0) {
4895 goto out_netdev;
4896 }
4897
dfe60397
MW
4898 res = register_netdevice(bond_dev);
4899 if (res < 0) {
4900 goto out_bond;
4901 }
0daa2303
PZ
4902
4903 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4904
ff59c456
JV
4905 netif_carrier_off(bond_dev);
4906
027ea041 4907 up_write(&bonding_rwsem);
dfe60397 4908 rtnl_unlock(); /* allows sysfs registration of net device */
b76cdba9 4909 res = bond_create_sysfs_entry(bond_dev->priv);
09c89279
JV
4910 if (res < 0) {
4911 rtnl_lock();
027ea041 4912 down_write(&bonding_rwsem);
822973ba
PE
4913 bond_deinit(bond_dev);
4914 unregister_netdevice(bond_dev);
4915 goto out_rtnl;
09c89279
JV
4916 }
4917
4918 return 0;
4919
dfe60397
MW
4920out_bond:
4921 bond_deinit(bond_dev);
4922out_netdev:
4923 free_netdev(bond_dev);
4924out_rtnl:
027ea041 4925 up_write(&bonding_rwsem);
dfe60397 4926 rtnl_unlock();
dfe60397
MW
4927 return res;
4928}
4929
1da177e4
LT
4930static int __init bonding_init(void)
4931{
1da177e4
LT
4932 int i;
4933 int res;
1b76b316 4934 struct bonding *bond, *nxt;
1da177e4
LT
4935
4936 printk(KERN_INFO "%s", version);
4937
dfe60397 4938 res = bond_check_params(&bonding_defaults);
1da177e4 4939 if (res) {
dfe60397 4940 goto out;
1da177e4
LT
4941 }
4942
1da177e4
LT
4943#ifdef CONFIG_PROC_FS
4944 bond_create_proc_dir();
4945#endif
027ea041
JV
4946
4947 init_rwsem(&bonding_rwsem);
4948
1da177e4 4949 for (i = 0; i < max_bonds; i++) {
0dd646fe 4950 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW
4951 if (res)
4952 goto err;
1da177e4
LT
4953 }
4954
b76cdba9
MW
4955 res = bond_create_sysfs();
4956 if (res)
4957 goto err;
4958
1da177e4 4959 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 4960 register_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 4961
dfe60397
MW
4962 goto out;
4963err:
1b76b316
JV
4964 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4965 bond_work_cancel_all(bond);
4966 destroy_workqueue(bond->wq);
4967 }
4968
ae68c398
PE
4969 bond_destroy_sysfs();
4970
40abc270 4971 rtnl_lock();
1da177e4 4972 bond_free_all();
1da177e4 4973 rtnl_unlock();
dfe60397 4974out:
1da177e4 4975 return res;
dfe60397 4976
1da177e4
LT
4977}
4978
4979static void __exit bonding_exit(void)
4980{
4981 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 4982 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 4983
ae68c398
PE
4984 bond_destroy_sysfs();
4985
1da177e4
LT
4986 rtnl_lock();
4987 bond_free_all();
4988 rtnl_unlock();
4989}
4990
4991module_init(bonding_init);
4992module_exit(bonding_exit);
4993MODULE_LICENSE("GPL");
4994MODULE_VERSION(DRV_VERSION);
4995MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4996MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4997MODULE_SUPPORTED_DEVICE("most ethernet devices");
4998
4999/*
5000 * Local variables:
5001 * c-indent-level: 8
5002 * c-basic-offset: 8
5003 * tab-width: 8
5004 * End:
5005 */
5006