blob: e0f51afec778ef00bc2d7ecf2c3e28bf454bd549 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 */
33
34//#define BONDING_DEBUG 1
35
36#include <linux/config.h>
37#include <linux/kernel.h>
38#include <linux/module.h>
39#include <linux/sched.h>
40#include <linux/types.h>
41#include <linux/fcntl.h>
42#include <linux/interrupt.h>
43#include <linux/ptrace.h>
44#include <linux/ioport.h>
45#include <linux/in.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040046#include <net/ip.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/ip.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040048#include <linux/tcp.h>
49#include <linux/udp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/slab.h>
51#include <linux/string.h>
52#include <linux/init.h>
53#include <linux/timer.h>
54#include <linux/socket.h>
55#include <linux/ctype.h>
56#include <linux/inet.h>
57#include <linux/bitops.h>
58#include <asm/system.h>
59#include <asm/io.h>
60#include <asm/dma.h>
61#include <asm/uaccess.h>
62#include <linux/errno.h>
63#include <linux/netdevice.h>
64#include <linux/inetdevice.h>
65#include <linux/etherdevice.h>
66#include <linux/skbuff.h>
67#include <net/sock.h>
68#include <linux/rtnetlink.h>
69#include <linux/proc_fs.h>
70#include <linux/seq_file.h>
71#include <linux/smp.h>
72#include <linux/if_ether.h>
73#include <net/arp.h>
74#include <linux/mii.h>
75#include <linux/ethtool.h>
76#include <linux/if_vlan.h>
77#include <linux/if_bonding.h>
Jay Vosburghc3ade5c2005-06-26 17:52:20 -040078#include <net/route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#include "bonding.h"
80#include "bond_3ad.h"
81#include "bond_alb.h"
82
83/*---------------------------- Module parameters ----------------------------*/
84
85/* monitor all links that often (in milliseconds). <=0 disables monitoring */
86#define BOND_LINK_MON_INTERV 0
87#define BOND_LINK_ARP_INTERV 0
88
89static int max_bonds = BOND_DEFAULT_MAX_BONDS;
90static int miimon = BOND_LINK_MON_INTERV;
91static int updelay = 0;
92static int downdelay = 0;
93static int use_carrier = 1;
94static char *mode = NULL;
95static char *primary = NULL;
96static char *lacp_rate = NULL;
Jay Vosburgh169a3e62005-06-26 17:54:11 -040097static char *xmit_hash_policy = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098static int arp_interval = BOND_LINK_ARP_INTERV;
99static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
Mitch Williams12479f92005-11-09 10:35:44 -0800100struct bond_params bonding_defaults;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102module_param(max_bonds, int, 0);
103MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
104module_param(miimon, int, 0);
105MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
106module_param(updelay, int, 0);
107MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
108module_param(downdelay, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800109MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
110 "in milliseconds");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111module_param(use_carrier, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800112MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
113 "0 for off, 1 for on (default)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114module_param(mode, charp, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800115MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
116 "1 for active-backup, 2 for balance-xor, "
117 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
118 "6 for balance-alb");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119module_param(primary, charp, 0);
120MODULE_PARM_DESC(primary, "Primary network device to use");
121module_param(lacp_rate, charp, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800122MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
123 "(slow/fast)");
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400124module_param(xmit_hash_policy, charp, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800125MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
126 ", 1 for layer 3+4");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127module_param(arp_interval, int, 0);
128MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
129module_param_array(arp_ip_target, charp, NULL, 0);
130MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
131
132/*----------------------------- Global variables ----------------------------*/
133
134static const char *version =
135 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
136
Mitch Williams12479f92005-11-09 10:35:44 -0800137LIST_HEAD(bond_dev_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
139#ifdef CONFIG_PROC_FS
140static struct proc_dir_entry *bond_proc_dir = NULL;
141#endif
142
Mitch Williamsb76cdba2005-11-09 10:36:41 -0800143extern struct rw_semaphore bonding_rwsem;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
145static int arp_ip_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146static int bond_mode = BOND_MODE_ROUNDROBIN;
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400147static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148static int lacp_fast = 0;
Jay Vosburgh217df672005-09-26 16:11:50 -0700149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Mitch Williams12479f92005-11-09 10:35:44 -0800151struct bond_parm_tbl bond_lacp_tbl[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152{ "slow", AD_LACP_SLOW},
153{ "fast", AD_LACP_FAST},
154{ NULL, -1},
155};
156
Mitch Williams12479f92005-11-09 10:35:44 -0800157struct bond_parm_tbl bond_mode_tbl[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{ "balance-rr", BOND_MODE_ROUNDROBIN},
159{ "active-backup", BOND_MODE_ACTIVEBACKUP},
160{ "balance-xor", BOND_MODE_XOR},
161{ "broadcast", BOND_MODE_BROADCAST},
162{ "802.3ad", BOND_MODE_8023AD},
163{ "balance-tlb", BOND_MODE_TLB},
164{ "balance-alb", BOND_MODE_ALB},
165{ NULL, -1},
166};
167
Mitch Williams12479f92005-11-09 10:35:44 -0800168struct bond_parm_tbl xmit_hashtype_tbl[] = {
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400169{ "layer2", BOND_XMIT_POLICY_LAYER2},
170{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
171{ NULL, -1},
172};
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174/*-------------------------- Forward declarations ---------------------------*/
175
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400176static void bond_send_gratuitous_arp(struct bonding *bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178/*---------------------------- General routines -----------------------------*/
179
Mitch Williamsa77b5322005-11-09 10:35:51 -0800180const char *bond_mode_name(int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181{
182 switch (mode) {
183 case BOND_MODE_ROUNDROBIN :
184 return "load balancing (round-robin)";
185 case BOND_MODE_ACTIVEBACKUP :
186 return "fault-tolerance (active-backup)";
187 case BOND_MODE_XOR :
188 return "load balancing (xor)";
189 case BOND_MODE_BROADCAST :
190 return "fault-tolerance (broadcast)";
191 case BOND_MODE_8023AD:
192 return "IEEE 802.3ad Dynamic link aggregation";
193 case BOND_MODE_TLB:
194 return "transmit load balancing";
195 case BOND_MODE_ALB:
196 return "adaptive load balancing";
197 default:
198 return "unknown";
199 }
200}
201
202/*---------------------------------- VLAN -----------------------------------*/
203
204/**
205 * bond_add_vlan - add a new vlan id on bond
206 * @bond: bond that got the notification
207 * @vlan_id: the vlan id to add
208 *
209 * Returns -ENOMEM if allocation failed.
210 */
211static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
212{
213 struct vlan_entry *vlan;
214
215 dprintk("bond: %s, vlan id %d\n",
216 (bond ? bond->dev->name: "None"), vlan_id);
217
218 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
219 if (!vlan) {
220 return -ENOMEM;
221 }
222
223 INIT_LIST_HEAD(&vlan->vlan_list);
224 vlan->vlan_id = vlan_id;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400225 vlan->vlan_ip = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
227 write_lock_bh(&bond->lock);
228
229 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
230
231 write_unlock_bh(&bond->lock);
232
233 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
234
235 return 0;
236}
237
238/**
239 * bond_del_vlan - delete a vlan id from bond
240 * @bond: bond that got the notification
241 * @vlan_id: the vlan id to delete
242 *
243 * returns -ENODEV if @vlan_id was not found in @bond.
244 */
245static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
246{
247 struct vlan_entry *vlan, *next;
248 int res = -ENODEV;
249
250 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
251
252 write_lock_bh(&bond->lock);
253
254 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
255 if (vlan->vlan_id == vlan_id) {
256 list_del(&vlan->vlan_list);
257
258 if ((bond->params.mode == BOND_MODE_TLB) ||
259 (bond->params.mode == BOND_MODE_ALB)) {
260 bond_alb_clear_vlan(bond, vlan_id);
261 }
262
263 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
264 bond->dev->name);
265
266 kfree(vlan);
267
268 if (list_empty(&bond->vlan_list) &&
269 (bond->slave_cnt == 0)) {
270 /* Last VLAN removed and no slaves, so
271 * restore block on adding VLANs. This will
272 * be removed once new slaves that are not
273 * VLAN challenged will be added.
274 */
275 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
276 }
277
278 res = 0;
279 goto out;
280 }
281 }
282
283 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
284 bond->dev->name);
285
286out:
287 write_unlock_bh(&bond->lock);
288 return res;
289}
290
291/**
292 * bond_has_challenged_slaves
293 * @bond: the bond we're working on
294 *
295 * Searches the slave list. Returns 1 if a vlan challenged slave
296 * was found, 0 otherwise.
297 *
298 * Assumes bond->lock is held.
299 */
300static int bond_has_challenged_slaves(struct bonding *bond)
301{
302 struct slave *slave;
303 int i;
304
305 bond_for_each_slave(bond, slave, i) {
306 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
307 dprintk("found VLAN challenged slave - %s\n",
308 slave->dev->name);
309 return 1;
310 }
311 }
312
313 dprintk("no VLAN challenged slaves found\n");
314 return 0;
315}
316
317/**
318 * bond_next_vlan - safely skip to the next item in the vlans list.
319 * @bond: the bond we're working on
320 * @curr: item we're advancing from
321 *
322 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
323 * or @curr->next otherwise (even if it is @curr itself again).
324 *
325 * Caller must hold bond->lock
326 */
327struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
328{
329 struct vlan_entry *next, *last;
330
331 if (list_empty(&bond->vlan_list)) {
332 return NULL;
333 }
334
335 if (!curr) {
336 next = list_entry(bond->vlan_list.next,
337 struct vlan_entry, vlan_list);
338 } else {
339 last = list_entry(bond->vlan_list.prev,
340 struct vlan_entry, vlan_list);
341 if (last == curr) {
342 next = list_entry(bond->vlan_list.next,
343 struct vlan_entry, vlan_list);
344 } else {
345 next = list_entry(curr->vlan_list.next,
346 struct vlan_entry, vlan_list);
347 }
348 }
349
350 return next;
351}
352
353/**
354 * bond_dev_queue_xmit - Prepare skb for xmit.
355 *
356 * @bond: bond device that got this skb for tx.
357 * @skb: hw accel VLAN tagged skb to transmit
358 * @slave_dev: slave that is supposed to xmit this skbuff
359 *
360 * When the bond gets an skb to transmit that is
361 * already hardware accelerated VLAN tagged, and it
362 * needs to relay this skb to a slave that is not
363 * hw accel capable, the skb needs to be "unaccelerated",
364 * i.e. strip the hwaccel tag and re-insert it as part
365 * of the payload.
366 */
367int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
368{
369 unsigned short vlan_id;
370
371 if (!list_empty(&bond->vlan_list) &&
372 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
373 vlan_get_tag(skb, &vlan_id) == 0) {
374 skb->dev = slave_dev;
375 skb = vlan_put_tag(skb, vlan_id);
376 if (!skb) {
377 /* vlan_put_tag() frees the skb in case of error,
378 * so return success here so the calling functions
379 * won't attempt to free is again.
380 */
381 return 0;
382 }
383 } else {
384 skb->dev = slave_dev;
385 }
386
387 skb->priority = 1;
388 dev_queue_xmit(skb);
389
390 return 0;
391}
392
393/*
394 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
395 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
396 * lock because:
397 * a. This operation is performed in IOCTL context,
398 * b. The operation is protected by the RTNL semaphore in the 8021q code,
399 * c. Holding a lock with BH disabled while directly calling a base driver
400 * entry point is generally a BAD idea.
401 *
402 * The design of synchronization/protection for this operation in the 8021q
403 * module is good for one or more VLAN devices over a single physical device
404 * and cannot be extended for a teaming solution like bonding, so there is a
405 * potential race condition here where a net device from the vlan group might
406 * be referenced (either by a base driver or the 8021q code) while it is being
407 * removed from the system. However, it turns out we're not making matters
408 * worse, and if it works for regular VLAN usage it will work here too.
409*/
410
411/**
412 * bond_vlan_rx_register - Propagates registration to slaves
413 * @bond_dev: bonding net device that got called
414 * @grp: vlan group being registered
415 */
416static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
417{
418 struct bonding *bond = bond_dev->priv;
419 struct slave *slave;
420 int i;
421
422 bond->vlgrp = grp;
423
424 bond_for_each_slave(bond, slave, i) {
425 struct net_device *slave_dev = slave->dev;
426
427 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
428 slave_dev->vlan_rx_register) {
429 slave_dev->vlan_rx_register(slave_dev, grp);
430 }
431 }
432}
433
434/**
435 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
436 * @bond_dev: bonding net device that got called
437 * @vid: vlan id being added
438 */
439static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
440{
441 struct bonding *bond = bond_dev->priv;
442 struct slave *slave;
443 int i, res;
444
445 bond_for_each_slave(bond, slave, i) {
446 struct net_device *slave_dev = slave->dev;
447
448 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
449 slave_dev->vlan_rx_add_vid) {
450 slave_dev->vlan_rx_add_vid(slave_dev, vid);
451 }
452 }
453
454 res = bond_add_vlan(bond, vid);
455 if (res) {
456 printk(KERN_ERR DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -0800457 ": %s: Error: Failed to add vlan id %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 bond_dev->name, vid);
459 }
460}
461
462/**
463 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
464 * @bond_dev: bonding net device that got called
465 * @vid: vlan id being removed
466 */
467static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
468{
469 struct bonding *bond = bond_dev->priv;
470 struct slave *slave;
471 struct net_device *vlan_dev;
472 int i, res;
473
474 bond_for_each_slave(bond, slave, i) {
475 struct net_device *slave_dev = slave->dev;
476
477 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
478 slave_dev->vlan_rx_kill_vid) {
479 /* Save and then restore vlan_dev in the grp array,
480 * since the slave's driver might clear it.
481 */
482 vlan_dev = bond->vlgrp->vlan_devices[vid];
483 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
484 bond->vlgrp->vlan_devices[vid] = vlan_dev;
485 }
486 }
487
488 res = bond_del_vlan(bond, vid);
489 if (res) {
490 printk(KERN_ERR DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -0800491 ": %s: Error: Failed to remove vlan id %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 bond_dev->name, vid);
493 }
494}
495
496static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
497{
498 struct vlan_entry *vlan;
499
500 write_lock_bh(&bond->lock);
501
502 if (list_empty(&bond->vlan_list)) {
503 goto out;
504 }
505
506 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
507 slave_dev->vlan_rx_register) {
508 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
509 }
510
511 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
512 !(slave_dev->vlan_rx_add_vid)) {
513 goto out;
514 }
515
516 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
517 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
518 }
519
520out:
521 write_unlock_bh(&bond->lock);
522}
523
524static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
525{
526 struct vlan_entry *vlan;
527 struct net_device *vlan_dev;
528
529 write_lock_bh(&bond->lock);
530
531 if (list_empty(&bond->vlan_list)) {
532 goto out;
533 }
534
535 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
536 !(slave_dev->vlan_rx_kill_vid)) {
537 goto unreg;
538 }
539
540 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
541 /* Save and then restore vlan_dev in the grp array,
542 * since the slave's driver might clear it.
543 */
544 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
545 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
546 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
547 }
548
549unreg:
550 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
551 slave_dev->vlan_rx_register) {
552 slave_dev->vlan_rx_register(slave_dev, NULL);
553 }
554
555out:
556 write_unlock_bh(&bond->lock);
557}
558
559/*------------------------------- Link status -------------------------------*/
560
561/*
562 * Get link speed and duplex from the slave's base driver
563 * using ethtool. If for some reason the call fails or the
564 * values are invalid, fake speed and duplex to 100/Full
565 * and return error.
566 */
567static int bond_update_speed_duplex(struct slave *slave)
568{
569 struct net_device *slave_dev = slave->dev;
570 static int (* ioctl)(struct net_device *, struct ifreq *, int);
571 struct ifreq ifr;
572 struct ethtool_cmd etool;
573
574 /* Fake speed and duplex */
575 slave->speed = SPEED_100;
576 slave->duplex = DUPLEX_FULL;
577
578 if (slave_dev->ethtool_ops) {
Eric Sesterhenn6a986ce2006-01-20 23:30:01 +0300579 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 if (!slave_dev->ethtool_ops->get_settings) {
582 return -1;
583 }
584
585 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
586 if (res < 0) {
587 return -1;
588 }
589
590 goto verify;
591 }
592
593 ioctl = slave_dev->do_ioctl;
594 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
595 etool.cmd = ETHTOOL_GSET;
596 ifr.ifr_data = (char*)&etool;
597 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
598 return -1;
599 }
600
601verify:
602 switch (etool.speed) {
603 case SPEED_10:
604 case SPEED_100:
605 case SPEED_1000:
606 break;
607 default:
608 return -1;
609 }
610
611 switch (etool.duplex) {
612 case DUPLEX_FULL:
613 case DUPLEX_HALF:
614 break;
615 default:
616 return -1;
617 }
618
619 slave->speed = etool.speed;
620 slave->duplex = etool.duplex;
621
622 return 0;
623}
624
625/*
626 * if <dev> supports MII link status reporting, check its link status.
627 *
628 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
629 * depening upon the setting of the use_carrier parameter.
630 *
631 * Return either BMSR_LSTATUS, meaning that the link is up (or we
632 * can't tell and just pretend it is), or 0, meaning that the link is
633 * down.
634 *
635 * If reporting is non-zero, instead of faking link up, return -1 if
636 * both ETHTOOL and MII ioctls fail (meaning the device does not
637 * support them). If use_carrier is set, return whatever it says.
638 * It'd be nice if there was a good way to tell if a driver supports
639 * netif_carrier, but there really isn't.
640 */
641static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
642{
643 static int (* ioctl)(struct net_device *, struct ifreq *, int);
644 struct ifreq ifr;
645 struct mii_ioctl_data *mii;
646 struct ethtool_value etool;
647
648 if (bond->params.use_carrier) {
649 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
650 }
651
652 ioctl = slave_dev->do_ioctl;
653 if (ioctl) {
654 /* TODO: set pointer to correct ioctl on a per team member */
655 /* bases to make this more efficient. that is, once */
656 /* we determine the correct ioctl, we will always */
657 /* call it and not the others for that team */
658 /* member. */
659
660 /*
661 * We cannot assume that SIOCGMIIPHY will also read a
662 * register; not all network drivers (e.g., e100)
663 * support that.
664 */
665
666 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
667 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
668 mii = if_mii(&ifr);
669 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
670 mii->reg_num = MII_BMSR;
671 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
672 return (mii->val_out & BMSR_LSTATUS);
673 }
674 }
675 }
676
677 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
678 /* for a period of time so we attempt to get link status */
679 /* from it last if the above MII ioctls fail... */
680 if (slave_dev->ethtool_ops) {
681 if (slave_dev->ethtool_ops->get_link) {
682 u32 link;
683
684 link = slave_dev->ethtool_ops->get_link(slave_dev);
685
686 return link ? BMSR_LSTATUS : 0;
687 }
688 }
689
690 if (ioctl) {
691 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
692 etool.cmd = ETHTOOL_GLINK;
693 ifr.ifr_data = (char*)&etool;
694 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
695 if (etool.data == 1) {
696 return BMSR_LSTATUS;
697 } else {
698 dprintk("SIOCETHTOOL shows link down\n");
699 return 0;
700 }
701 }
702 }
703
704 /*
705 * If reporting, report that either there's no dev->do_ioctl,
706 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
707 * cannot report link status). If not reporting, pretend
708 * we're ok.
709 */
710 return (reporting ? -1 : BMSR_LSTATUS);
711}
712
713/*----------------------------- Multicast list ------------------------------*/
714
715/*
716 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
717 */
718static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
719{
720 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
721 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
722}
723
724/*
725 * returns dmi entry if found, NULL otherwise
726 */
727static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
728{
729 struct dev_mc_list *idmi;
730
731 for (idmi = mc_list; idmi; idmi = idmi->next) {
732 if (bond_is_dmi_same(dmi, idmi)) {
733 return idmi;
734 }
735 }
736
737 return NULL;
738}
739
740/*
741 * Push the promiscuity flag down to appropriate slaves
742 */
743static void bond_set_promiscuity(struct bonding *bond, int inc)
744{
745 if (USES_PRIMARY(bond->params.mode)) {
746 /* write lock already acquired */
747 if (bond->curr_active_slave) {
748 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
749 }
750 } else {
751 struct slave *slave;
752 int i;
753 bond_for_each_slave(bond, slave, i) {
754 dev_set_promiscuity(slave->dev, inc);
755 }
756 }
757}
758
759/*
760 * Push the allmulti flag down to all slaves
761 */
762static void bond_set_allmulti(struct bonding *bond, int inc)
763{
764 if (USES_PRIMARY(bond->params.mode)) {
765 /* write lock already acquired */
766 if (bond->curr_active_slave) {
767 dev_set_allmulti(bond->curr_active_slave->dev, inc);
768 }
769 } else {
770 struct slave *slave;
771 int i;
772 bond_for_each_slave(bond, slave, i) {
773 dev_set_allmulti(slave->dev, inc);
774 }
775 }
776}
777
778/*
779 * Add a Multicast address to slaves
780 * according to mode
781 */
782static void bond_mc_add(struct bonding *bond, void *addr, int alen)
783{
784 if (USES_PRIMARY(bond->params.mode)) {
785 /* write lock already acquired */
786 if (bond->curr_active_slave) {
787 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
788 }
789 } else {
790 struct slave *slave;
791 int i;
792 bond_for_each_slave(bond, slave, i) {
793 dev_mc_add(slave->dev, addr, alen, 0);
794 }
795 }
796}
797
798/*
799 * Remove a multicast address from slave
800 * according to mode
801 */
802static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
803{
804 if (USES_PRIMARY(bond->params.mode)) {
805 /* write lock already acquired */
806 if (bond->curr_active_slave) {
807 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
808 }
809 } else {
810 struct slave *slave;
811 int i;
812 bond_for_each_slave(bond, slave, i) {
813 dev_mc_delete(slave->dev, addr, alen, 0);
814 }
815 }
816}
817
818/*
819 * Totally destroys the mc_list in bond
820 */
821static void bond_mc_list_destroy(struct bonding *bond)
822{
823 struct dev_mc_list *dmi;
824
825 dmi = bond->mc_list;
826 while (dmi) {
827 bond->mc_list = dmi->next;
828 kfree(dmi);
829 dmi = bond->mc_list;
830 }
831}
832
833/*
834 * Copy all the Multicast addresses from src to the bonding device dst
835 */
Randy Dunlapde54f392005-10-04 22:39:41 -0700836static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
Al Virodd0fc662005-10-07 07:46:04 +0100837 gfp_t gfp_flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838{
839 struct dev_mc_list *dmi, *new_dmi;
840
841 for (dmi = mc_list; dmi; dmi = dmi->next) {
Randy Dunlapde54f392005-10-04 22:39:41 -0700842 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
844 if (!new_dmi) {
845 /* FIXME: Potential memory leak !!! */
846 return -ENOMEM;
847 }
848
849 new_dmi->next = bond->mc_list;
850 bond->mc_list = new_dmi;
851 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
852 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
853 new_dmi->dmi_users = dmi->dmi_users;
854 new_dmi->dmi_gusers = dmi->dmi_gusers;
855 }
856
857 return 0;
858}
859
860/*
861 * flush all members of flush->mc_list from device dev->mc_list
862 */
863static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
864{
865 struct bonding *bond = bond_dev->priv;
866 struct dev_mc_list *dmi;
867
868 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
869 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
870 }
871
872 if (bond->params.mode == BOND_MODE_8023AD) {
873 /* del lacpdu mc addr from mc list */
874 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
875
876 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
877 }
878}
879
880/*--------------------------- Active slave change ---------------------------*/
881
882/*
883 * Update the mc list and multicast-related flags for the new and
884 * old active slaves (if any) according to the multicast mode, and
885 * promiscuous flags unconditionally.
886 */
887static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
888{
889 struct dev_mc_list *dmi;
890
891 if (!USES_PRIMARY(bond->params.mode)) {
892 /* nothing to do - mc list is already up-to-date on
893 * all slaves
894 */
895 return;
896 }
897
898 if (old_active) {
899 if (bond->dev->flags & IFF_PROMISC) {
900 dev_set_promiscuity(old_active->dev, -1);
901 }
902
903 if (bond->dev->flags & IFF_ALLMULTI) {
904 dev_set_allmulti(old_active->dev, -1);
905 }
906
907 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
908 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
909 }
910 }
911
912 if (new_active) {
913 if (bond->dev->flags & IFF_PROMISC) {
914 dev_set_promiscuity(new_active->dev, 1);
915 }
916
917 if (bond->dev->flags & IFF_ALLMULTI) {
918 dev_set_allmulti(new_active->dev, 1);
919 }
920
921 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
922 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
923 }
924 }
925}
926
927/**
928 * find_best_interface - select the best available slave to be the active one
929 * @bond: our bonding struct
930 *
931 * Warning: Caller must hold curr_slave_lock for writing.
932 */
933static struct slave *bond_find_best_slave(struct bonding *bond)
934{
935 struct slave *new_active, *old_active;
936 struct slave *bestslave = NULL;
937 int mintime = bond->params.updelay;
938 int i;
939
940 new_active = old_active = bond->curr_active_slave;
941
942 if (!new_active) { /* there were no active slaves left */
943 if (bond->slave_cnt > 0) { /* found one slave */
944 new_active = bond->first_slave;
945 } else {
946 return NULL; /* still no slave, return NULL */
947 }
948 }
949
950 /* first try the primary link; if arping, a link must tx/rx traffic
951 * before it can be considered the curr_active_slave - also, we would skip
952 * slaves between the curr_active_slave and primary_slave that may be up
953 * and able to arp
954 */
955 if ((bond->primary_slave) &&
956 (!bond->params.arp_interval) &&
957 (IS_UP(bond->primary_slave->dev))) {
958 new_active = bond->primary_slave;
959 }
960
961 /* remember where to stop iterating over the slaves */
962 old_active = new_active;
963
964 bond_for_each_slave_from(bond, new_active, i, old_active) {
965 if (IS_UP(new_active->dev)) {
966 if (new_active->link == BOND_LINK_UP) {
967 return new_active;
968 } else if (new_active->link == BOND_LINK_BACK) {
969 /* link up, but waiting for stabilization */
970 if (new_active->delay < mintime) {
971 mintime = new_active->delay;
972 bestslave = new_active;
973 }
974 }
975 }
976 }
977
978 return bestslave;
979}
980
981/**
982 * change_active_interface - change the active slave into the specified one
983 * @bond: our bonding struct
984 * @new: the new slave to make the active one
985 *
986 * Set the new slave to the bond's settings and unset them on the old
987 * curr_active_slave.
988 * Setting include flags, mc-list, promiscuity, allmulti, etc.
989 *
990 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
991 * because it is apparently the best available slave we have, even though its
992 * updelay hasn't timed out yet.
993 *
994 * Warning: Caller must hold curr_slave_lock for writing.
995 */
Mitch Williamsa77b5322005-11-09 10:35:51 -0800996void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997{
998 struct slave *old_active = bond->curr_active_slave;
999
1000 if (old_active == new_active) {
1001 return;
1002 }
1003
1004 if (new_active) {
1005 if (new_active->link == BOND_LINK_BACK) {
1006 if (USES_PRIMARY(bond->params.mode)) {
1007 printk(KERN_INFO DRV_NAME
1008 ": %s: making interface %s the new "
1009 "active one %d ms earlier.\n",
1010 bond->dev->name, new_active->dev->name,
1011 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1012 }
1013
1014 new_active->delay = 0;
1015 new_active->link = BOND_LINK_UP;
1016 new_active->jiffies = jiffies;
1017
1018 if (bond->params.mode == BOND_MODE_8023AD) {
1019 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1020 }
1021
1022 if ((bond->params.mode == BOND_MODE_TLB) ||
1023 (bond->params.mode == BOND_MODE_ALB)) {
1024 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1025 }
1026 } else {
1027 if (USES_PRIMARY(bond->params.mode)) {
1028 printk(KERN_INFO DRV_NAME
1029 ": %s: making interface %s the new "
1030 "active one.\n",
1031 bond->dev->name, new_active->dev->name);
1032 }
1033 }
1034 }
1035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 if (USES_PRIMARY(bond->params.mode)) {
1037 bond_mc_swap(bond, new_active, old_active);
1038 }
1039
1040 if ((bond->params.mode == BOND_MODE_TLB) ||
1041 (bond->params.mode == BOND_MODE_ALB)) {
1042 bond_alb_handle_active_change(bond, new_active);
1043 } else {
1044 bond->curr_active_slave = new_active;
1045 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001046
1047 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1048 if (old_active) {
1049 bond_set_slave_inactive_flags(old_active);
1050 }
1051
1052 if (new_active) {
1053 bond_set_slave_active_flags(new_active);
1054 }
1055 bond_send_gratuitous_arp(bond);
1056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057}
1058
1059/**
1060 * bond_select_active_slave - select a new active slave, if needed
1061 * @bond: our bonding struct
1062 *
1063 * This functions shoud be called when one of the following occurs:
1064 * - The old curr_active_slave has been released or lost its link.
1065 * - The primary_slave has got its link back.
1066 * - A slave has got its link back and there's no old curr_active_slave.
1067 *
1068 * Warning: Caller must hold curr_slave_lock for writing.
1069 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001070void bond_select_active_slave(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
1072 struct slave *best_slave;
1073
1074 best_slave = bond_find_best_slave(bond);
1075 if (best_slave != bond->curr_active_slave) {
1076 bond_change_active_slave(bond, best_slave);
1077 }
1078}
1079
1080/*--------------------------- slave list handling ---------------------------*/
1081
1082/*
1083 * This function attaches the slave to the end of list.
1084 *
1085 * bond->lock held for writing by caller.
1086 */
1087static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1088{
1089 if (bond->first_slave == NULL) { /* attaching the first slave */
1090 new_slave->next = new_slave;
1091 new_slave->prev = new_slave;
1092 bond->first_slave = new_slave;
1093 } else {
1094 new_slave->next = bond->first_slave;
1095 new_slave->prev = bond->first_slave->prev;
1096 new_slave->next->prev = new_slave;
1097 new_slave->prev->next = new_slave;
1098 }
1099
1100 bond->slave_cnt++;
1101}
1102
1103/*
1104 * This function detaches the slave from the list.
1105 * WARNING: no check is made to verify if the slave effectively
1106 * belongs to <bond>.
1107 * Nothing is freed on return, structures are just unchained.
1108 * If any slave pointer in bond was pointing to <slave>,
1109 * it should be changed by the calling function.
1110 *
1111 * bond->lock held for writing by caller.
1112 */
1113static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1114{
1115 if (slave->next) {
1116 slave->next->prev = slave->prev;
1117 }
1118
1119 if (slave->prev) {
1120 slave->prev->next = slave->next;
1121 }
1122
1123 if (bond->first_slave == slave) { /* slave is the first slave */
1124 if (bond->slave_cnt > 1) { /* there are more slave */
1125 bond->first_slave = slave->next;
1126 } else {
1127 bond->first_slave = NULL; /* slave was the last one */
1128 }
1129 }
1130
1131 slave->next = NULL;
1132 slave->prev = NULL;
1133 bond->slave_cnt--;
1134}
1135
1136/*---------------------------------- IOCTL ----------------------------------*/
1137
Mitch Williamsa77b5322005-11-09 10:35:51 -08001138int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139{
1140 dprintk("bond_dev=%p\n", bond_dev);
1141 dprintk("slave_dev=%p\n", slave_dev);
1142 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1143 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1144 return 0;
1145}
1146
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001147#define BOND_INTERSECT_FEATURES \
Jay Vosburgha0de3ad2006-01-30 15:40:59 -08001148 (NETIF_F_SG|NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM|\
1149 NETIF_F_TSO|NETIF_F_UFO)
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001150
1151/*
Jay Vosburgh8e3babc2005-11-04 18:45:45 -08001152 * Compute the common dev->feature set available to all slaves. Some
1153 * feature bits are managed elsewhere, so preserve feature bits set on
1154 * master device that are not part of the examined set.
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001155 */
1156static int bond_compute_features(struct bonding *bond)
1157{
Jay Vosburgh8e3babc2005-11-04 18:45:45 -08001158 unsigned long features = BOND_INTERSECT_FEATURES;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001159 struct slave *slave;
1160 struct net_device *bond_dev = bond->dev;
Jay Vosburgh8e3babc2005-11-04 18:45:45 -08001161 int i;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001162
Jay Vosburgh8e3babc2005-11-04 18:45:45 -08001163 bond_for_each_slave(bond, slave, i)
1164 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001165
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001166 if ((features & NETIF_F_SG) &&
Jay Vosburgh8e3babc2005-11-04 18:45:45 -08001167 !(features & (NETIF_F_IP_CSUM |
1168 NETIF_F_NO_CSUM |
1169 NETIF_F_HW_CSUM)))
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001170 features &= ~NETIF_F_SG;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001171
Jay Vosburgha0de3ad2006-01-30 15:40:59 -08001172 /*
1173 * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1174 * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1175 * implies that all slaves also support scatter-gather
1176 * (NETIF_F_SG), which implies that features also includes
1177 * NETIF_F_SG. So no need to check whether we have an
1178 * illegal combination of NETIF_F_{TSO,UFO} and
1179 * !NETIF_F_SG
1180 */
1181
Jay Vosburgh8e3babc2005-11-04 18:45:45 -08001182 features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001183 bond_dev->features = features;
1184
1185 return 0;
1186}
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188/* enslave device <slave> to bond device <master> */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001189int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
1191 struct bonding *bond = bond_dev->priv;
1192 struct slave *new_slave = NULL;
1193 struct dev_mc_list *dmi;
1194 struct sockaddr addr;
1195 int link_reporting;
1196 int old_features = bond_dev->features;
1197 int res = 0;
1198
nsxfreddy@gmail.com552709d2005-09-21 14:18:04 -05001199 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1200 slave_dev->do_ioctl == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001202 ": %s: Warning: no link monitoring support for %s\n",
1203 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
1205
1206 /* bond must be initialized by bond_open() before enslaving */
1207 if (!(bond_dev->flags & IFF_UP)) {
1208 dprintk("Error, master_dev is not up\n");
1209 return -EPERM;
1210 }
1211
1212 /* already enslaved */
1213 if (slave_dev->flags & IFF_SLAVE) {
1214 dprintk("Error, Device was already enslaved\n");
1215 return -EBUSY;
1216 }
1217
1218 /* vlan challenged mutual exclusion */
1219 /* no need to lock since we're protected by rtnl_lock */
1220 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1221 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1222 if (!list_empty(&bond->vlan_list)) {
1223 printk(KERN_ERR DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001224 ": %s: Error: cannot enslave VLAN "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 "challenged slave %s on VLAN enabled "
Mitch Williams4e0952c2005-11-09 10:34:57 -08001226 "bond %s\n", bond_dev->name, slave_dev->name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 bond_dev->name);
1228 return -EPERM;
1229 } else {
1230 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001231 ": %s: Warning: enslaved VLAN challenged "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 "slave %s. Adding VLANs will be blocked as "
1233 "long as %s is part of bond %s\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001234 bond_dev->name, slave_dev->name, slave_dev->name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 bond_dev->name);
1236 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1237 }
1238 } else {
1239 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1240 if (bond->slave_cnt == 0) {
1241 /* First slave, and it is not VLAN challenged,
1242 * so remove the block of adding VLANs over the bond.
1243 */
1244 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1245 }
1246 }
1247
Jay Vosburgh217df672005-09-26 16:11:50 -07001248 /*
1249 * Old ifenslave binaries are no longer supported. These can
1250 * be identified with moderate accurary by the state of the slave:
1251 * the current ifenslave will set the interface down prior to
1252 * enslaving it; the old ifenslave will not.
1253 */
1254 if ((slave_dev->flags & IFF_UP)) {
1255 printk(KERN_ERR DRV_NAME ": %s is up. "
1256 "This may be due to an out of date ifenslave.\n",
1257 slave_dev->name);
1258 res = -EPERM;
1259 goto err_undo_flags;
1260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
Jay Vosburgh217df672005-09-26 16:11:50 -07001262 if (slave_dev->set_mac_address == NULL) {
1263 printk(KERN_ERR DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001264 ": %s: Error: The slave device you specified does "
1265 "not support setting the MAC address. "
1266 "Your kernel likely does not support slave "
1267 "devices.\n", bond_dev->name);
1268 res = -EOPNOTSUPP;
Jay Vosburgh217df672005-09-26 16:11:50 -07001269 goto err_undo_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 }
1271
1272 new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1273 if (!new_slave) {
1274 res = -ENOMEM;
1275 goto err_undo_flags;
1276 }
1277
1278 memset(new_slave, 0, sizeof(struct slave));
1279
1280 /* save slave's original flags before calling
1281 * netdev_set_master and dev_open
1282 */
1283 new_slave->original_flags = slave_dev->flags;
1284
Jay Vosburgh217df672005-09-26 16:11:50 -07001285 /*
1286 * Save slave's original ("permanent") mac address for modes
1287 * that need it, and for restoring it upon release, and then
1288 * set it to the master's address
1289 */
1290 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
Jay Vosburgh217df672005-09-26 16:11:50 -07001292 /*
1293 * Set slave to master's mac address. The application already
1294 * set the master's mac address to that of the first slave
1295 */
1296 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1297 addr.sa_family = slave_dev->type;
1298 res = dev_set_mac_address(slave_dev, &addr);
1299 if (res) {
1300 dprintk("Error %d calling set_mac_address\n", res);
1301 goto err_free;
1302 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Jay Vosburgh217df672005-09-26 16:11:50 -07001304 /* open the slave since the application closed it */
1305 res = dev_open(slave_dev);
1306 if (res) {
1307 dprintk("Openning slave %s failed\n", slave_dev->name);
1308 goto err_restore_mac;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 }
1310
1311 res = netdev_set_master(slave_dev, bond_dev);
1312 if (res) {
1313 dprintk("Error %d calling netdev_set_master\n", res);
Jay Vosburgh217df672005-09-26 16:11:50 -07001314 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 }
1316
1317 new_slave->dev = slave_dev;
1318
1319 if ((bond->params.mode == BOND_MODE_TLB) ||
1320 (bond->params.mode == BOND_MODE_ALB)) {
1321 /* bond_alb_init_slave() must be called before all other stages since
1322 * it might fail and we do not want to have to undo everything
1323 */
1324 res = bond_alb_init_slave(bond, new_slave);
1325 if (res) {
1326 goto err_unset_master;
1327 }
1328 }
1329
1330 /* If the mode USES_PRIMARY, then the new slave gets the
1331 * master's promisc (and mc) settings only if it becomes the
1332 * curr_active_slave, and that is taken care of later when calling
1333 * bond_change_active()
1334 */
1335 if (!USES_PRIMARY(bond->params.mode)) {
1336 /* set promiscuity level to new slave */
1337 if (bond_dev->flags & IFF_PROMISC) {
1338 dev_set_promiscuity(slave_dev, 1);
1339 }
1340
1341 /* set allmulti level to new slave */
1342 if (bond_dev->flags & IFF_ALLMULTI) {
1343 dev_set_allmulti(slave_dev, 1);
1344 }
1345
1346 /* upload master's mc_list to new slave */
1347 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1348 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1349 }
1350 }
1351
1352 if (bond->params.mode == BOND_MODE_8023AD) {
1353 /* add lacpdu mc addr to mc list */
1354 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1355
1356 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1357 }
1358
1359 bond_add_vlans_on_slave(bond, slave_dev);
1360
1361 write_lock_bh(&bond->lock);
1362
1363 bond_attach_slave(bond, new_slave);
1364
1365 new_slave->delay = 0;
1366 new_slave->link_failure_count = 0;
1367
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001368 bond_compute_features(bond);
1369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 if (bond->params.miimon && !bond->params.use_carrier) {
1371 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1372
1373 if ((link_reporting == -1) && !bond->params.arp_interval) {
1374 /*
1375 * miimon is set but a bonded network driver
1376 * does not support ETHTOOL/MII and
1377 * arp_interval is not set. Note: if
1378 * use_carrier is enabled, we will never go
1379 * here (because netif_carrier is always
1380 * supported); thus, we don't need to change
1381 * the messages for netif_carrier.
1382 */
1383 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001384 ": %s: Warning: MII and ETHTOOL support not "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 "available for interface %s, and "
1386 "arp_interval/arp_ip_target module parameters "
1387 "not specified, thus bonding will not detect "
1388 "link failures! see bonding.txt for details.\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001389 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 } else if (link_reporting == -1) {
1391 /* unable get link status using mii/ethtool */
1392 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001393 ": %s: Warning: can't get link status from "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 "interface %s; the network driver associated "
1395 "with this interface does not support MII or "
1396 "ETHTOOL link status reporting, thus miimon "
1397 "has no effect on this interface.\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001398 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 }
1400 }
1401
1402 /* check for initial state */
1403 if (!bond->params.miimon ||
1404 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1405 if (bond->params.updelay) {
1406 dprintk("Initial state of slave_dev is "
1407 "BOND_LINK_BACK\n");
1408 new_slave->link = BOND_LINK_BACK;
1409 new_slave->delay = bond->params.updelay;
1410 } else {
1411 dprintk("Initial state of slave_dev is "
1412 "BOND_LINK_UP\n");
1413 new_slave->link = BOND_LINK_UP;
1414 }
1415 new_slave->jiffies = jiffies;
1416 } else {
1417 dprintk("Initial state of slave_dev is "
1418 "BOND_LINK_DOWN\n");
1419 new_slave->link = BOND_LINK_DOWN;
1420 }
1421
1422 if (bond_update_speed_duplex(new_slave) &&
1423 (new_slave->link != BOND_LINK_DOWN)) {
1424 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001425 ": %s: Warning: failed to get speed and duplex from %s, "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 "assumed to be 100Mb/sec and Full.\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001427 bond_dev->name, new_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 if (bond->params.mode == BOND_MODE_8023AD) {
Mitch Williams4e0952c2005-11-09 10:34:57 -08001430 printk(KERN_WARNING DRV_NAME
1431 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 "support in base driver for proper aggregator "
Mitch Williams4e0952c2005-11-09 10:34:57 -08001433 "selection.\n", bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 }
1435 }
1436
1437 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1438 /* if there is a primary slave, remember it */
1439 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1440 bond->primary_slave = new_slave;
1441 }
1442 }
1443
1444 switch (bond->params.mode) {
1445 case BOND_MODE_ACTIVEBACKUP:
1446 /* if we're in active-backup mode, we need one and only one active
1447 * interface. The backup interfaces will have their NOARP flag set
1448 * because we need them to be completely deaf and not to respond to
1449 * any ARP request on the network to avoid fooling a switch. Thus,
1450 * since we guarantee that curr_active_slave always point to the last
1451 * usable interface, we just have to verify this interface's flag.
1452 */
1453 if (((!bond->curr_active_slave) ||
1454 (bond->curr_active_slave->dev->flags & IFF_NOARP)) &&
1455 (new_slave->link != BOND_LINK_DOWN)) {
1456 dprintk("This is the first active slave\n");
1457 /* first slave or no active slave yet, and this link
1458 is OK, so make this interface the active one */
1459 bond_change_active_slave(bond, new_slave);
1460 } else {
1461 dprintk("This is just a backup slave\n");
1462 bond_set_slave_inactive_flags(new_slave);
1463 }
1464 break;
1465 case BOND_MODE_8023AD:
1466 /* in 802.3ad mode, the internal mechanism
1467 * will activate the slaves in the selected
1468 * aggregator
1469 */
1470 bond_set_slave_inactive_flags(new_slave);
1471 /* if this is the first slave */
1472 if (bond->slave_cnt == 1) {
1473 SLAVE_AD_INFO(new_slave).id = 1;
1474 /* Initialize AD with the number of times that the AD timer is called in 1 second
1475 * can be called only after the mac address of the bond is set
1476 */
1477 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1478 bond->params.lacp_fast);
1479 } else {
1480 SLAVE_AD_INFO(new_slave).id =
1481 SLAVE_AD_INFO(new_slave->prev).id + 1;
1482 }
1483
1484 bond_3ad_bind_slave(new_slave);
1485 break;
1486 case BOND_MODE_TLB:
1487 case BOND_MODE_ALB:
1488 new_slave->state = BOND_STATE_ACTIVE;
1489 if ((!bond->curr_active_slave) &&
1490 (new_slave->link != BOND_LINK_DOWN)) {
1491 /* first slave or no active slave yet, and this link
1492 * is OK, so make this interface the active one
1493 */
1494 bond_change_active_slave(bond, new_slave);
1495 }
1496 break;
1497 default:
1498 dprintk("This slave is always active in trunk mode\n");
1499
1500 /* always active in trunk mode */
1501 new_slave->state = BOND_STATE_ACTIVE;
1502
1503 /* In trunking mode there is little meaning to curr_active_slave
1504 * anyway (it holds no special properties of the bond device),
1505 * so we can change it without calling change_active_interface()
1506 */
1507 if (!bond->curr_active_slave) {
1508 bond->curr_active_slave = new_slave;
1509 }
1510 break;
1511 } /* switch(bond_mode) */
1512
1513 write_unlock_bh(&bond->lock);
1514
Mitch Williamsb76cdba2005-11-09 10:36:41 -08001515 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1516 if (res)
1517 goto err_unset_master;
1518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 printk(KERN_INFO DRV_NAME
1520 ": %s: enslaving %s as a%s interface with a%s link.\n",
1521 bond_dev->name, slave_dev->name,
1522 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1523 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1524
1525 /* enslave is successful */
1526 return 0;
1527
1528/* Undo stages on error */
1529err_unset_master:
1530 netdev_set_master(slave_dev, NULL);
1531
1532err_close:
1533 dev_close(slave_dev);
1534
1535err_restore_mac:
1536 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1537 addr.sa_family = slave_dev->type;
1538 dev_set_mac_address(slave_dev, &addr);
1539
1540err_free:
1541 kfree(new_slave);
1542
1543err_undo_flags:
1544 bond_dev->features = old_features;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001545
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 return res;
1547}
1548
1549/*
1550 * Try to release the slave device <slave> from the bond device <master>
1551 * It is legal to access curr_active_slave without a lock because all the function
1552 * is write-locked.
1553 *
1554 * The rules for slave state should be:
1555 * for Active/Backup:
1556 * Active stays on all backups go down
1557 * for Bonded connections:
1558 * The first up interface should be left on and all others downed.
1559 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001560int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
1562 struct bonding *bond = bond_dev->priv;
1563 struct slave *slave, *oldcurrent;
1564 struct sockaddr addr;
1565 int mac_addr_differ;
1566
1567 /* slave is not a slave or master is not master of this slave */
1568 if (!(slave_dev->flags & IFF_SLAVE) ||
1569 (slave_dev->master != bond_dev)) {
1570 printk(KERN_ERR DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001571 ": %s: Error: cannot release %s.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 bond_dev->name, slave_dev->name);
1573 return -EINVAL;
1574 }
1575
1576 write_lock_bh(&bond->lock);
1577
1578 slave = bond_get_slave_by_dev(bond, slave_dev);
1579 if (!slave) {
1580 /* not a slave of this bond */
1581 printk(KERN_INFO DRV_NAME
1582 ": %s: %s not enslaved\n",
1583 bond_dev->name, slave_dev->name);
1584 return -EINVAL;
1585 }
1586
1587 mac_addr_differ = memcmp(bond_dev->dev_addr,
1588 slave->perm_hwaddr,
1589 ETH_ALEN);
1590 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1591 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001592 ": %s: Warning: the permanent HWaddr of %s "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1594 "still in use by %s. Set the HWaddr of "
1595 "%s to a different address to avoid "
1596 "conflicts.\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001597 bond_dev->name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 slave_dev->name,
1599 slave->perm_hwaddr[0],
1600 slave->perm_hwaddr[1],
1601 slave->perm_hwaddr[2],
1602 slave->perm_hwaddr[3],
1603 slave->perm_hwaddr[4],
1604 slave->perm_hwaddr[5],
1605 bond_dev->name,
1606 slave_dev->name);
1607 }
1608
1609 /* Inform AD package of unbinding of slave. */
1610 if (bond->params.mode == BOND_MODE_8023AD) {
1611 /* must be called before the slave is
1612 * detached from the list
1613 */
1614 bond_3ad_unbind_slave(slave);
1615 }
1616
1617 printk(KERN_INFO DRV_NAME
1618 ": %s: releasing %s interface %s\n",
1619 bond_dev->name,
1620 (slave->state == BOND_STATE_ACTIVE)
1621 ? "active" : "backup",
1622 slave_dev->name);
1623
1624 oldcurrent = bond->curr_active_slave;
1625
1626 bond->current_arp_slave = NULL;
1627
1628 /* release the slave from its bond */
1629 bond_detach_slave(bond, slave);
1630
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001631 bond_compute_features(bond);
1632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 if (bond->primary_slave == slave) {
1634 bond->primary_slave = NULL;
1635 }
1636
1637 if (oldcurrent == slave) {
1638 bond_change_active_slave(bond, NULL);
1639 }
1640
1641 if ((bond->params.mode == BOND_MODE_TLB) ||
1642 (bond->params.mode == BOND_MODE_ALB)) {
1643 /* Must be called only after the slave has been
1644 * detached from the list and the curr_active_slave
1645 * has been cleared (if our_slave == old_current),
1646 * but before a new active slave is selected.
1647 */
1648 bond_alb_deinit_slave(bond, slave);
1649 }
1650
1651 if (oldcurrent == slave) {
1652 bond_select_active_slave(bond);
1653
1654 if (!bond->curr_active_slave) {
1655 printk(KERN_INFO DRV_NAME
1656 ": %s: now running without any active "
1657 "interface !\n",
1658 bond_dev->name);
1659 }
1660 }
1661
1662 if (bond->slave_cnt == 0) {
1663 /* if the last slave was removed, zero the mac address
1664 * of the master so it will be set by the application
1665 * to the mac address of the first slave
1666 */
1667 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1668
1669 if (list_empty(&bond->vlan_list)) {
1670 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1671 } else {
1672 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001673 ": %s: Warning: clearing HW address of %s while it "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 "still has VLANs.\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001675 bond_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001677 ": %s: When re-adding slaves, make sure the bond's "
1678 "HW address matches its VLANs'.\n",
1679 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 }
1681 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1682 !bond_has_challenged_slaves(bond)) {
1683 printk(KERN_INFO DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001684 ": %s: last VLAN challenged slave %s "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 "left bond %s. VLAN blocking is removed\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001686 bond_dev->name, slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1688 }
1689
1690 write_unlock_bh(&bond->lock);
1691
Mitch Williamsb76cdba2005-11-09 10:36:41 -08001692 /* must do this from outside any spinlocks */
1693 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1694
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 bond_del_vlans_from_slave(bond, slave_dev);
1696
1697 /* If the mode USES_PRIMARY, then we should only remove its
1698 * promisc and mc settings if it was the curr_active_slave, but that was
1699 * already taken care of above when we detached the slave
1700 */
1701 if (!USES_PRIMARY(bond->params.mode)) {
1702 /* unset promiscuity level from slave */
1703 if (bond_dev->flags & IFF_PROMISC) {
1704 dev_set_promiscuity(slave_dev, -1);
1705 }
1706
1707 /* unset allmulti level from slave */
1708 if (bond_dev->flags & IFF_ALLMULTI) {
1709 dev_set_allmulti(slave_dev, -1);
1710 }
1711
1712 /* flush master's mc_list from slave */
1713 bond_mc_list_flush(bond_dev, slave_dev);
1714 }
1715
1716 netdev_set_master(slave_dev, NULL);
1717
1718 /* close slave before restoring its mac address */
1719 dev_close(slave_dev);
1720
Jay Vosburgh217df672005-09-26 16:11:50 -07001721 /* restore original ("permanent") mac address */
1722 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1723 addr.sa_family = slave_dev->type;
1724 dev_set_mac_address(slave_dev, &addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725
1726 /* restore the original state of the
1727 * IFF_NOARP flag that might have been
1728 * set by bond_set_slave_inactive_flags()
1729 */
1730 if ((slave->original_flags & IFF_NOARP) == 0) {
1731 slave_dev->flags &= ~IFF_NOARP;
1732 }
1733
1734 kfree(slave);
1735
1736 return 0; /* deletion OK */
1737}
1738
1739/*
1740 * This function releases all slaves.
1741 */
1742static int bond_release_all(struct net_device *bond_dev)
1743{
1744 struct bonding *bond = bond_dev->priv;
1745 struct slave *slave;
1746 struct net_device *slave_dev;
1747 struct sockaddr addr;
1748
1749 write_lock_bh(&bond->lock);
1750
1751 if (bond->slave_cnt == 0) {
1752 goto out;
1753 }
1754
1755 bond->current_arp_slave = NULL;
1756 bond->primary_slave = NULL;
1757 bond_change_active_slave(bond, NULL);
1758
1759 while ((slave = bond->first_slave) != NULL) {
1760 /* Inform AD package of unbinding of slave
1761 * before slave is detached from the list.
1762 */
1763 if (bond->params.mode == BOND_MODE_8023AD) {
1764 bond_3ad_unbind_slave(slave);
1765 }
1766
1767 slave_dev = slave->dev;
1768 bond_detach_slave(bond, slave);
1769
1770 if ((bond->params.mode == BOND_MODE_TLB) ||
1771 (bond->params.mode == BOND_MODE_ALB)) {
1772 /* must be called only after the slave
1773 * has been detached from the list
1774 */
1775 bond_alb_deinit_slave(bond, slave);
1776 }
1777
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001778 bond_compute_features(bond);
1779
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 /* now that the slave is detached, unlock and perform
1781 * all the undo steps that should not be called from
1782 * within a lock.
1783 */
1784 write_unlock_bh(&bond->lock);
1785
Mitch Williamsb76cdba2005-11-09 10:36:41 -08001786 bond_destroy_slave_symlinks(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 bond_del_vlans_from_slave(bond, slave_dev);
1788
1789 /* If the mode USES_PRIMARY, then we should only remove its
1790 * promisc and mc settings if it was the curr_active_slave, but that was
1791 * already taken care of above when we detached the slave
1792 */
1793 if (!USES_PRIMARY(bond->params.mode)) {
1794 /* unset promiscuity level from slave */
1795 if (bond_dev->flags & IFF_PROMISC) {
1796 dev_set_promiscuity(slave_dev, -1);
1797 }
1798
1799 /* unset allmulti level from slave */
1800 if (bond_dev->flags & IFF_ALLMULTI) {
1801 dev_set_allmulti(slave_dev, -1);
1802 }
1803
1804 /* flush master's mc_list from slave */
1805 bond_mc_list_flush(bond_dev, slave_dev);
1806 }
1807
1808 netdev_set_master(slave_dev, NULL);
1809
1810 /* close slave before restoring its mac address */
1811 dev_close(slave_dev);
1812
Jay Vosburgh217df672005-09-26 16:11:50 -07001813 /* restore original ("permanent") mac address*/
1814 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1815 addr.sa_family = slave_dev->type;
1816 dev_set_mac_address(slave_dev, &addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
1818 /* restore the original state of the IFF_NOARP flag that might have
1819 * been set by bond_set_slave_inactive_flags()
1820 */
1821 if ((slave->original_flags & IFF_NOARP) == 0) {
1822 slave_dev->flags &= ~IFF_NOARP;
1823 }
1824
1825 kfree(slave);
1826
1827 /* re-acquire the lock before getting the next slave */
1828 write_lock_bh(&bond->lock);
1829 }
1830
1831 /* zero the mac address of the master so it will be
1832 * set by the application to the mac address of the
1833 * first slave
1834 */
1835 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1836
1837 if (list_empty(&bond->vlan_list)) {
1838 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1839 } else {
1840 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001841 ": %s: Warning: clearing HW address of %s while it "
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 "still has VLANs.\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001843 bond_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 printk(KERN_WARNING DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08001845 ": %s: When re-adding slaves, make sure the bond's "
1846 "HW address matches its VLANs'.\n",
1847 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 }
1849
1850 printk(KERN_INFO DRV_NAME
1851 ": %s: released all slaves\n",
1852 bond_dev->name);
1853
1854out:
1855 write_unlock_bh(&bond->lock);
1856
1857 return 0;
1858}
1859
1860/*
1861 * This function changes the active slave to slave <slave_dev>.
1862 * It returns -EINVAL in the following cases.
1863 * - <slave_dev> is not found in the list.
1864 * - There is not active slave now.
1865 * - <slave_dev> is already active.
1866 * - The link state of <slave_dev> is not BOND_LINK_UP.
1867 * - <slave_dev> is not running.
1868 * In these cases, this fuction does nothing.
1869 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1870 */
1871static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1872{
1873 struct bonding *bond = bond_dev->priv;
1874 struct slave *old_active = NULL;
1875 struct slave *new_active = NULL;
1876 int res = 0;
1877
1878 if (!USES_PRIMARY(bond->params.mode)) {
1879 return -EINVAL;
1880 }
1881
1882 /* Verify that master_dev is indeed the master of slave_dev */
1883 if (!(slave_dev->flags & IFF_SLAVE) ||
1884 (slave_dev->master != bond_dev)) {
1885 return -EINVAL;
1886 }
1887
1888 write_lock_bh(&bond->lock);
1889
1890 old_active = bond->curr_active_slave;
1891 new_active = bond_get_slave_by_dev(bond, slave_dev);
1892
1893 /*
1894 * Changing to the current active: do nothing; return success.
1895 */
1896 if (new_active && (new_active == old_active)) {
1897 write_unlock_bh(&bond->lock);
1898 return 0;
1899 }
1900
1901 if ((new_active) &&
1902 (old_active) &&
1903 (new_active->link == BOND_LINK_UP) &&
1904 IS_UP(new_active->dev)) {
1905 bond_change_active_slave(bond, new_active);
1906 } else {
1907 res = -EINVAL;
1908 }
1909
1910 write_unlock_bh(&bond->lock);
1911
1912 return res;
1913}
1914
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1916{
1917 struct bonding *bond = bond_dev->priv;
1918
1919 info->bond_mode = bond->params.mode;
1920 info->miimon = bond->params.miimon;
1921
1922 read_lock_bh(&bond->lock);
1923 info->num_slaves = bond->slave_cnt;
1924 read_unlock_bh(&bond->lock);
1925
1926 return 0;
1927}
1928
1929static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1930{
1931 struct bonding *bond = bond_dev->priv;
1932 struct slave *slave;
1933 int i, found = 0;
1934
1935 if (info->slave_id < 0) {
1936 return -ENODEV;
1937 }
1938
1939 read_lock_bh(&bond->lock);
1940
1941 bond_for_each_slave(bond, slave, i) {
1942 if (i == (int)info->slave_id) {
1943 found = 1;
1944 break;
1945 }
1946 }
1947
1948 read_unlock_bh(&bond->lock);
1949
1950 if (found) {
1951 strcpy(info->slave_name, slave->dev->name);
1952 info->link = slave->link;
1953 info->state = slave->state;
1954 info->link_failure_count = slave->link_failure_count;
1955 } else {
1956 return -ENODEV;
1957 }
1958
1959 return 0;
1960}
1961
1962/*-------------------------------- Monitoring -------------------------------*/
1963
1964/* this function is called regularly to monitor each slave's link. */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001965void bond_mii_monitor(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966{
1967 struct bonding *bond = bond_dev->priv;
1968 struct slave *slave, *oldcurrent;
1969 int do_failover = 0;
1970 int delta_in_ticks;
1971 int i;
1972
1973 read_lock(&bond->lock);
1974
1975 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
1976
1977 if (bond->kill_timers) {
1978 goto out;
1979 }
1980
1981 if (bond->slave_cnt == 0) {
1982 goto re_arm;
1983 }
1984
1985 /* we will try to read the link status of each of our slaves, and
1986 * set their IFF_RUNNING flag appropriately. For each slave not
1987 * supporting MII status, we won't do anything so that a user-space
1988 * program could monitor the link itself if needed.
1989 */
1990
1991 read_lock(&bond->curr_slave_lock);
1992 oldcurrent = bond->curr_active_slave;
1993 read_unlock(&bond->curr_slave_lock);
1994
1995 bond_for_each_slave(bond, slave, i) {
1996 struct net_device *slave_dev = slave->dev;
1997 int link_state;
1998 u16 old_speed = slave->speed;
1999 u8 old_duplex = slave->duplex;
2000
2001 link_state = bond_check_dev_link(bond, slave_dev, 0);
2002
2003 switch (slave->link) {
2004 case BOND_LINK_UP: /* the link was up */
2005 if (link_state == BMSR_LSTATUS) {
2006 /* link stays up, nothing more to do */
2007 break;
2008 } else { /* link going down */
2009 slave->link = BOND_LINK_FAIL;
2010 slave->delay = bond->params.downdelay;
2011
2012 if (slave->link_failure_count < UINT_MAX) {
2013 slave->link_failure_count++;
2014 }
2015
2016 if (bond->params.downdelay) {
2017 printk(KERN_INFO DRV_NAME
2018 ": %s: link status down for %s "
2019 "interface %s, disabling it in "
2020 "%d ms.\n",
2021 bond_dev->name,
2022 IS_UP(slave_dev)
2023 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2024 ? ((slave == oldcurrent)
2025 ? "active " : "backup ")
2026 : "")
2027 : "idle ",
2028 slave_dev->name,
2029 bond->params.downdelay * bond->params.miimon);
2030 }
2031 }
2032 /* no break ! fall through the BOND_LINK_FAIL test to
2033 ensure proper action to be taken
2034 */
2035 case BOND_LINK_FAIL: /* the link has just gone down */
2036 if (link_state != BMSR_LSTATUS) {
2037 /* link stays down */
2038 if (slave->delay <= 0) {
2039 /* link down for too long time */
2040 slave->link = BOND_LINK_DOWN;
2041
2042 /* in active/backup mode, we must
2043 * completely disable this interface
2044 */
2045 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2046 (bond->params.mode == BOND_MODE_8023AD)) {
2047 bond_set_slave_inactive_flags(slave);
2048 }
2049
2050 printk(KERN_INFO DRV_NAME
2051 ": %s: link status definitely "
2052 "down for interface %s, "
2053 "disabling it\n",
2054 bond_dev->name,
2055 slave_dev->name);
2056
2057 /* notify ad that the link status has changed */
2058 if (bond->params.mode == BOND_MODE_8023AD) {
2059 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2060 }
2061
2062 if ((bond->params.mode == BOND_MODE_TLB) ||
2063 (bond->params.mode == BOND_MODE_ALB)) {
2064 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2065 }
2066
2067 if (slave == oldcurrent) {
2068 do_failover = 1;
2069 }
2070 } else {
2071 slave->delay--;
2072 }
2073 } else {
2074 /* link up again */
2075 slave->link = BOND_LINK_UP;
2076 slave->jiffies = jiffies;
2077 printk(KERN_INFO DRV_NAME
2078 ": %s: link status up again after %d "
2079 "ms for interface %s.\n",
2080 bond_dev->name,
2081 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2082 slave_dev->name);
2083 }
2084 break;
2085 case BOND_LINK_DOWN: /* the link was down */
2086 if (link_state != BMSR_LSTATUS) {
2087 /* the link stays down, nothing more to do */
2088 break;
2089 } else { /* link going up */
2090 slave->link = BOND_LINK_BACK;
2091 slave->delay = bond->params.updelay;
2092
2093 if (bond->params.updelay) {
2094 /* if updelay == 0, no need to
2095 advertise about a 0 ms delay */
2096 printk(KERN_INFO DRV_NAME
2097 ": %s: link status up for "
2098 "interface %s, enabling it "
2099 "in %d ms.\n",
2100 bond_dev->name,
2101 slave_dev->name,
2102 bond->params.updelay * bond->params.miimon);
2103 }
2104 }
2105 /* no break ! fall through the BOND_LINK_BACK state in
2106 case there's something to do.
2107 */
2108 case BOND_LINK_BACK: /* the link has just come back */
2109 if (link_state != BMSR_LSTATUS) {
2110 /* link down again */
2111 slave->link = BOND_LINK_DOWN;
2112
2113 printk(KERN_INFO DRV_NAME
2114 ": %s: link status down again after %d "
2115 "ms for interface %s.\n",
2116 bond_dev->name,
2117 (bond->params.updelay - slave->delay) * bond->params.miimon,
2118 slave_dev->name);
2119 } else {
2120 /* link stays up */
2121 if (slave->delay == 0) {
2122 /* now the link has been up for long time enough */
2123 slave->link = BOND_LINK_UP;
2124 slave->jiffies = jiffies;
2125
2126 if (bond->params.mode == BOND_MODE_8023AD) {
2127 /* prevent it from being the active one */
2128 slave->state = BOND_STATE_BACKUP;
2129 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2130 /* make it immediately active */
2131 slave->state = BOND_STATE_ACTIVE;
2132 } else if (slave != bond->primary_slave) {
2133 /* prevent it from being the active one */
2134 slave->state = BOND_STATE_BACKUP;
2135 }
2136
2137 printk(KERN_INFO DRV_NAME
2138 ": %s: link status definitely "
2139 "up for interface %s.\n",
2140 bond_dev->name,
2141 slave_dev->name);
2142
2143 /* notify ad that the link status has changed */
2144 if (bond->params.mode == BOND_MODE_8023AD) {
2145 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2146 }
2147
2148 if ((bond->params.mode == BOND_MODE_TLB) ||
2149 (bond->params.mode == BOND_MODE_ALB)) {
2150 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2151 }
2152
2153 if ((!oldcurrent) ||
2154 (slave == bond->primary_slave)) {
2155 do_failover = 1;
2156 }
2157 } else {
2158 slave->delay--;
2159 }
2160 }
2161 break;
2162 default:
2163 /* Should not happen */
Mitch Williams4e0952c2005-11-09 10:34:57 -08002164 printk(KERN_ERR DRV_NAME
2165 ": %s: Error: %s Illegal value (link=%d)\n",
2166 bond_dev->name,
2167 slave->dev->name,
2168 slave->link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 goto out;
2170 } /* end of switch (slave->link) */
2171
2172 bond_update_speed_duplex(slave);
2173
2174 if (bond->params.mode == BOND_MODE_8023AD) {
2175 if (old_speed != slave->speed) {
2176 bond_3ad_adapter_speed_changed(slave);
2177 }
2178
2179 if (old_duplex != slave->duplex) {
2180 bond_3ad_adapter_duplex_changed(slave);
2181 }
2182 }
2183
2184 } /* end of for */
2185
2186 if (do_failover) {
2187 write_lock(&bond->curr_slave_lock);
2188
2189 bond_select_active_slave(bond);
2190
2191 if (oldcurrent && !bond->curr_active_slave) {
2192 printk(KERN_INFO DRV_NAME
2193 ": %s: now running without any active "
2194 "interface !\n",
2195 bond_dev->name);
2196 }
2197
2198 write_unlock(&bond->curr_slave_lock);
2199 }
2200
2201re_arm:
2202 if (bond->params.miimon) {
2203 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2204 }
2205out:
2206 read_unlock(&bond->lock);
2207}
2208
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002209
2210static u32 bond_glean_dev_ip(struct net_device *dev)
2211{
2212 struct in_device *idev;
2213 struct in_ifaddr *ifa;
2214 u32 addr = 0;
2215
2216 if (!dev)
2217 return 0;
2218
2219 rcu_read_lock();
Herbert Xue5ed6392005-10-03 14:35:55 -07002220 idev = __in_dev_get_rcu(dev);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002221 if (!idev)
2222 goto out;
2223
2224 ifa = idev->ifa_list;
2225 if (!ifa)
2226 goto out;
2227
2228 addr = ifa->ifa_local;
2229out:
2230 rcu_read_unlock();
2231 return addr;
2232}
2233
2234static int bond_has_ip(struct bonding *bond)
2235{
2236 struct vlan_entry *vlan, *vlan_next;
2237
2238 if (bond->master_ip)
2239 return 1;
2240
2241 if (list_empty(&bond->vlan_list))
2242 return 0;
2243
2244 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2245 vlan_list) {
2246 if (vlan->vlan_ip)
2247 return 1;
2248 }
2249
2250 return 0;
2251}
2252
2253/*
2254 * We go to the (large) trouble of VLAN tagging ARP frames because
2255 * switches in VLAN mode (especially if ports are configured as
2256 * "native" to a VLAN) might not pass non-tagged frames.
2257 */
2258static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2259{
2260 struct sk_buff *skb;
2261
2262 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2263 slave_dev->name, dest_ip, src_ip, vlan_id);
2264
2265 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2266 NULL, slave_dev->dev_addr, NULL);
2267
2268 if (!skb) {
2269 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2270 return;
2271 }
2272 if (vlan_id) {
2273 skb = vlan_put_tag(skb, vlan_id);
2274 if (!skb) {
2275 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2276 return;
2277 }
2278 }
2279 arp_xmit(skb);
2280}
2281
2282
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2284{
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002285 int i, vlan_id, rv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 u32 *targets = bond->params.arp_targets;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002287 struct vlan_entry *vlan, *vlan_next;
2288 struct net_device *vlan_dev;
2289 struct flowi fl;
2290 struct rtable *rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
Mitch Williams6b780562005-11-09 10:36:19 -08002292 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2293 if (!targets[i])
2294 continue;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002295 dprintk("basa: target %x\n", targets[i]);
2296 if (list_empty(&bond->vlan_list)) {
2297 dprintk("basa: empty vlan: arp_send\n");
2298 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2299 bond->master_ip, 0);
2300 continue;
2301 }
2302
2303 /*
2304 * If VLANs are configured, we do a route lookup to
2305 * determine which VLAN interface would be used, so we
2306 * can tag the ARP with the proper VLAN tag.
2307 */
2308 memset(&fl, 0, sizeof(fl));
2309 fl.fl4_dst = targets[i];
2310 fl.fl4_tos = RTO_ONLINK;
2311
2312 rv = ip_route_output_key(&rt, &fl);
2313 if (rv) {
2314 if (net_ratelimit()) {
2315 printk(KERN_WARNING DRV_NAME
2316 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2317 bond->dev->name, NIPQUAD(fl.fl4_dst));
2318 }
2319 continue;
2320 }
2321
2322 /*
2323 * This target is not on a VLAN
2324 */
2325 if (rt->u.dst.dev == bond->dev) {
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002326 ip_rt_put(rt);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002327 dprintk("basa: rtdev == bond->dev: arp_send\n");
2328 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2329 bond->master_ip, 0);
2330 continue;
2331 }
2332
2333 vlan_id = 0;
2334 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2335 vlan_list) {
2336 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2337 if (vlan_dev == rt->u.dst.dev) {
2338 vlan_id = vlan->vlan_id;
2339 dprintk("basa: vlan match on %s %d\n",
2340 vlan_dev->name, vlan_id);
2341 break;
2342 }
2343 }
2344
2345 if (vlan_id) {
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002346 ip_rt_put(rt);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002347 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2348 vlan->vlan_ip, vlan_id);
2349 continue;
2350 }
2351
2352 if (net_ratelimit()) {
2353 printk(KERN_WARNING DRV_NAME
2354 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2355 bond->dev->name, NIPQUAD(fl.fl4_dst),
2356 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2357 }
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002358 ip_rt_put(rt);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002359 }
2360}
2361
2362/*
2363 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2364 * for each VLAN above us.
2365 */
2366static void bond_send_gratuitous_arp(struct bonding *bond)
2367{
2368 struct slave *slave = bond->curr_active_slave;
2369 struct vlan_entry *vlan;
2370 struct net_device *vlan_dev;
2371
2372 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2373 slave ? slave->dev->name : "NULL");
2374 if (!slave)
2375 return;
2376
2377 if (bond->master_ip) {
2378 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2379 bond->master_ip, 0);
2380 }
2381
2382 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2383 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2384 if (vlan->vlan_ip) {
2385 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2386 vlan->vlan_ip, vlan->vlan_id);
2387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 }
2389}
2390
2391/*
2392 * this function is called regularly to monitor each slave's link
2393 * ensuring that traffic is being sent and received when arp monitoring
2394 * is used in load-balancing mode. if the adapter has been dormant, then an
2395 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2396 * arp monitoring in active backup mode.
2397 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08002398void bond_loadbalance_arp_mon(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399{
2400 struct bonding *bond = bond_dev->priv;
2401 struct slave *slave, *oldcurrent;
2402 int do_failover = 0;
2403 int delta_in_ticks;
2404 int i;
2405
2406 read_lock(&bond->lock);
2407
2408 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2409
2410 if (bond->kill_timers) {
2411 goto out;
2412 }
2413
2414 if (bond->slave_cnt == 0) {
2415 goto re_arm;
2416 }
2417
2418 read_lock(&bond->curr_slave_lock);
2419 oldcurrent = bond->curr_active_slave;
2420 read_unlock(&bond->curr_slave_lock);
2421
2422 /* see if any of the previous devices are up now (i.e. they have
2423 * xmt and rcv traffic). the curr_active_slave does not come into
2424 * the picture unless it is null. also, slave->jiffies is not needed
2425 * here because we send an arp on each slave and give a slave as
2426 * long as it needs to get the tx/rx within the delta.
2427 * TODO: what about up/down delay in arp mode? it wasn't here before
2428 * so it can wait
2429 */
2430 bond_for_each_slave(bond, slave, i) {
2431 if (slave->link != BOND_LINK_UP) {
2432 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2433 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2434
2435 slave->link = BOND_LINK_UP;
2436 slave->state = BOND_STATE_ACTIVE;
2437
2438 /* primary_slave has no meaning in round-robin
2439 * mode. the window of a slave being up and
2440 * curr_active_slave being null after enslaving
2441 * is closed.
2442 */
2443 if (!oldcurrent) {
2444 printk(KERN_INFO DRV_NAME
2445 ": %s: link status definitely "
2446 "up for interface %s, ",
2447 bond_dev->name,
2448 slave->dev->name);
2449 do_failover = 1;
2450 } else {
2451 printk(KERN_INFO DRV_NAME
2452 ": %s: interface %s is now up\n",
2453 bond_dev->name,
2454 slave->dev->name);
2455 }
2456 }
2457 } else {
2458 /* slave->link == BOND_LINK_UP */
2459
2460 /* not all switches will respond to an arp request
2461 * when the source ip is 0, so don't take the link down
2462 * if we don't know our ip yet
2463 */
2464 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2465 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002466 bond_has_ip(bond))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
2468 slave->link = BOND_LINK_DOWN;
2469 slave->state = BOND_STATE_BACKUP;
2470
2471 if (slave->link_failure_count < UINT_MAX) {
2472 slave->link_failure_count++;
2473 }
2474
2475 printk(KERN_INFO DRV_NAME
2476 ": %s: interface %s is now down.\n",
2477 bond_dev->name,
2478 slave->dev->name);
2479
2480 if (slave == oldcurrent) {
2481 do_failover = 1;
2482 }
2483 }
2484 }
2485
2486 /* note: if switch is in round-robin mode, all links
2487 * must tx arp to ensure all links rx an arp - otherwise
2488 * links may oscillate or not come up at all; if switch is
2489 * in something like xor mode, there is nothing we can
2490 * do - all replies will be rx'ed on same link causing slaves
2491 * to be unstable during low/no traffic periods
2492 */
2493 if (IS_UP(slave->dev)) {
2494 bond_arp_send_all(bond, slave);
2495 }
2496 }
2497
2498 if (do_failover) {
2499 write_lock(&bond->curr_slave_lock);
2500
2501 bond_select_active_slave(bond);
2502
2503 if (oldcurrent && !bond->curr_active_slave) {
2504 printk(KERN_INFO DRV_NAME
2505 ": %s: now running without any active "
2506 "interface !\n",
2507 bond_dev->name);
2508 }
2509
2510 write_unlock(&bond->curr_slave_lock);
2511 }
2512
2513re_arm:
2514 if (bond->params.arp_interval) {
2515 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2516 }
2517out:
2518 read_unlock(&bond->lock);
2519}
2520
2521/*
2522 * When using arp monitoring in active-backup mode, this function is
2523 * called to determine if any backup slaves have went down or a new
2524 * current slave needs to be found.
2525 * The backup slaves never generate traffic, they are considered up by merely
2526 * receiving traffic. If the current slave goes down, each backup slave will
2527 * be given the opportunity to tx/rx an arp before being taken down - this
2528 * prevents all slaves from being taken down due to the current slave not
2529 * sending any traffic for the backups to receive. The arps are not necessarily
2530 * necessary, any tx and rx traffic will keep the current slave up. While any
2531 * rx traffic will keep the backup slaves up, the current slave is responsible
2532 * for generating traffic to keep them up regardless of any other traffic they
2533 * may have received.
2534 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2535 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08002536void bond_activebackup_arp_mon(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537{
2538 struct bonding *bond = bond_dev->priv;
2539 struct slave *slave;
2540 int delta_in_ticks;
2541 int i;
2542
2543 read_lock(&bond->lock);
2544
2545 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2546
2547 if (bond->kill_timers) {
2548 goto out;
2549 }
2550
2551 if (bond->slave_cnt == 0) {
2552 goto re_arm;
2553 }
2554
2555 /* determine if any slave has come up or any backup slave has
2556 * gone down
2557 * TODO: what about up/down delay in arp mode? it wasn't here before
2558 * so it can wait
2559 */
2560 bond_for_each_slave(bond, slave, i) {
2561 if (slave->link != BOND_LINK_UP) {
2562 if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
2563
2564 slave->link = BOND_LINK_UP;
2565
2566 write_lock(&bond->curr_slave_lock);
2567
2568 if ((!bond->curr_active_slave) &&
2569 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2570 bond_change_active_slave(bond, slave);
2571 bond->current_arp_slave = NULL;
2572 } else if (bond->curr_active_slave != slave) {
2573 /* this slave has just come up but we
2574 * already have a current slave; this
2575 * can also happen if bond_enslave adds
2576 * a new slave that is up while we are
2577 * searching for a new slave
2578 */
2579 bond_set_slave_inactive_flags(slave);
2580 bond->current_arp_slave = NULL;
2581 }
2582
2583 if (slave == bond->curr_active_slave) {
2584 printk(KERN_INFO DRV_NAME
2585 ": %s: %s is up and now the "
2586 "active interface\n",
2587 bond_dev->name,
2588 slave->dev->name);
2589 } else {
2590 printk(KERN_INFO DRV_NAME
2591 ": %s: backup interface %s is "
2592 "now up\n",
2593 bond_dev->name,
2594 slave->dev->name);
2595 }
2596
2597 write_unlock(&bond->curr_slave_lock);
2598 }
2599 } else {
2600 read_lock(&bond->curr_slave_lock);
2601
2602 if ((slave != bond->curr_active_slave) &&
2603 (!bond->current_arp_slave) &&
2604 (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002605 bond_has_ip(bond))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 /* a backup slave has gone down; three times
2607 * the delta allows the current slave to be
2608 * taken out before the backup slave.
2609 * note: a non-null current_arp_slave indicates
2610 * the curr_active_slave went down and we are
2611 * searching for a new one; under this
2612 * condition we only take the curr_active_slave
2613 * down - this gives each slave a chance to
2614 * tx/rx traffic before being taken out
2615 */
2616
2617 read_unlock(&bond->curr_slave_lock);
2618
2619 slave->link = BOND_LINK_DOWN;
2620
2621 if (slave->link_failure_count < UINT_MAX) {
2622 slave->link_failure_count++;
2623 }
2624
2625 bond_set_slave_inactive_flags(slave);
2626
2627 printk(KERN_INFO DRV_NAME
2628 ": %s: backup interface %s is now down\n",
2629 bond_dev->name,
2630 slave->dev->name);
2631 } else {
2632 read_unlock(&bond->curr_slave_lock);
2633 }
2634 }
2635 }
2636
2637 read_lock(&bond->curr_slave_lock);
2638 slave = bond->curr_active_slave;
2639 read_unlock(&bond->curr_slave_lock);
2640
2641 if (slave) {
2642 /* if we have sent traffic in the past 2*arp_intervals but
2643 * haven't xmit and rx traffic in that time interval, select
2644 * a different slave. slave->jiffies is only updated when
2645 * a slave first becomes the curr_active_slave - not necessarily
2646 * after every arp; this ensures the slave has a full 2*delta
2647 * before being taken out. if a primary is being used, check
2648 * if it is up and needs to take over as the curr_active_slave
2649 */
2650 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002651 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2652 bond_has_ip(bond))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2654
2655 slave->link = BOND_LINK_DOWN;
2656
2657 if (slave->link_failure_count < UINT_MAX) {
2658 slave->link_failure_count++;
2659 }
2660
2661 printk(KERN_INFO DRV_NAME
2662 ": %s: link status down for active interface "
2663 "%s, disabling it\n",
2664 bond_dev->name,
2665 slave->dev->name);
2666
2667 write_lock(&bond->curr_slave_lock);
2668
2669 bond_select_active_slave(bond);
2670 slave = bond->curr_active_slave;
2671
2672 write_unlock(&bond->curr_slave_lock);
2673
2674 bond->current_arp_slave = slave;
2675
2676 if (slave) {
2677 slave->jiffies = jiffies;
2678 }
2679 } else if ((bond->primary_slave) &&
2680 (bond->primary_slave != slave) &&
2681 (bond->primary_slave->link == BOND_LINK_UP)) {
2682 /* at this point, slave is the curr_active_slave */
2683 printk(KERN_INFO DRV_NAME
2684 ": %s: changing from interface %s to primary "
2685 "interface %s\n",
2686 bond_dev->name,
2687 slave->dev->name,
2688 bond->primary_slave->dev->name);
2689
2690 /* primary is up so switch to it */
2691 write_lock(&bond->curr_slave_lock);
2692 bond_change_active_slave(bond, bond->primary_slave);
2693 write_unlock(&bond->curr_slave_lock);
2694
2695 slave = bond->primary_slave;
2696 slave->jiffies = jiffies;
2697 } else {
2698 bond->current_arp_slave = NULL;
2699 }
2700
2701 /* the current slave must tx an arp to ensure backup slaves
2702 * rx traffic
2703 */
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002704 if (slave && bond_has_ip(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 bond_arp_send_all(bond, slave);
2706 }
2707 }
2708
2709 /* if we don't have a curr_active_slave, search for the next available
2710 * backup slave from the current_arp_slave and make it the candidate
2711 * for becoming the curr_active_slave
2712 */
2713 if (!slave) {
2714 if (!bond->current_arp_slave) {
2715 bond->current_arp_slave = bond->first_slave;
2716 }
2717
2718 if (bond->current_arp_slave) {
2719 bond_set_slave_inactive_flags(bond->current_arp_slave);
2720
2721 /* search for next candidate */
Jay Vosburgh2f872f02005-05-26 12:56:59 -07002722 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 if (IS_UP(slave->dev)) {
2724 slave->link = BOND_LINK_BACK;
2725 bond_set_slave_active_flags(slave);
2726 bond_arp_send_all(bond, slave);
2727 slave->jiffies = jiffies;
2728 bond->current_arp_slave = slave;
2729 break;
2730 }
2731
2732 /* if the link state is up at this point, we
2733 * mark it down - this can happen if we have
2734 * simultaneous link failures and
2735 * reselect_active_interface doesn't make this
2736 * one the current slave so it is still marked
2737 * up when it is actually down
2738 */
2739 if (slave->link == BOND_LINK_UP) {
2740 slave->link = BOND_LINK_DOWN;
2741 if (slave->link_failure_count < UINT_MAX) {
2742 slave->link_failure_count++;
2743 }
2744
2745 bond_set_slave_inactive_flags(slave);
2746
2747 printk(KERN_INFO DRV_NAME
2748 ": %s: backup interface %s is "
2749 "now down.\n",
2750 bond_dev->name,
2751 slave->dev->name);
2752 }
2753 }
2754 }
2755 }
2756
2757re_arm:
2758 if (bond->params.arp_interval) {
2759 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2760 }
2761out:
2762 read_unlock(&bond->lock);
2763}
2764
2765/*------------------------------ proc/seq_file-------------------------------*/
2766
2767#ifdef CONFIG_PROC_FS
2768
2769#define SEQ_START_TOKEN ((void *)1)
2770
2771static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2772{
2773 struct bonding *bond = seq->private;
2774 loff_t off = 0;
2775 struct slave *slave;
2776 int i;
2777
2778 /* make sure the bond won't be taken away */
2779 read_lock(&dev_base_lock);
2780 read_lock_bh(&bond->lock);
2781
2782 if (*pos == 0) {
2783 return SEQ_START_TOKEN;
2784 }
2785
2786 bond_for_each_slave(bond, slave, i) {
2787 if (++off == *pos) {
2788 return slave;
2789 }
2790 }
2791
2792 return NULL;
2793}
2794
2795static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2796{
2797 struct bonding *bond = seq->private;
2798 struct slave *slave = v;
2799
2800 ++*pos;
2801 if (v == SEQ_START_TOKEN) {
2802 return bond->first_slave;
2803 }
2804
2805 slave = slave->next;
2806
2807 return (slave == bond->first_slave) ? NULL : slave;
2808}
2809
2810static void bond_info_seq_stop(struct seq_file *seq, void *v)
2811{
2812 struct bonding *bond = seq->private;
2813
2814 read_unlock_bh(&bond->lock);
2815 read_unlock(&dev_base_lock);
2816}
2817
2818static void bond_info_show_master(struct seq_file *seq)
2819{
2820 struct bonding *bond = seq->private;
2821 struct slave *curr;
Mitch Williams4756b022005-11-09 10:36:25 -08002822 int i;
2823 u32 target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824
2825 read_lock(&bond->curr_slave_lock);
2826 curr = bond->curr_active_slave;
2827 read_unlock(&bond->curr_slave_lock);
2828
2829 seq_printf(seq, "Bonding Mode: %s\n",
2830 bond_mode_name(bond->params.mode));
2831
Mitch Williamsc61b75a2005-11-09 10:35:13 -08002832 if (bond->params.mode == BOND_MODE_XOR ||
2833 bond->params.mode == BOND_MODE_8023AD) {
2834 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2835 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2836 bond->params.xmit_policy);
2837 }
2838
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 if (USES_PRIMARY(bond->params.mode)) {
2840 seq_printf(seq, "Primary Slave: %s\n",
Mitch Williams0f418b22005-11-09 10:35:21 -08002841 (bond->primary_slave) ?
2842 bond->primary_slave->dev->name : "None");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843
2844 seq_printf(seq, "Currently Active Slave: %s\n",
2845 (curr) ? curr->dev->name : "None");
2846 }
2847
2848 seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down");
2849 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
2850 seq_printf(seq, "Up Delay (ms): %d\n",
2851 bond->params.updelay * bond->params.miimon);
2852 seq_printf(seq, "Down Delay (ms): %d\n",
2853 bond->params.downdelay * bond->params.miimon);
2854
Mitch Williams4756b022005-11-09 10:36:25 -08002855
2856 /* ARP information */
2857 if(bond->params.arp_interval > 0) {
2858 int printed=0;
2859 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
2860 bond->params.arp_interval);
2861
2862 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
2863
2864 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
2865 if (!bond->params.arp_targets[i])
2866 continue;
2867 if (printed)
2868 seq_printf(seq, ",");
2869 target = ntohl(bond->params.arp_targets[i]);
2870 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
2871 printed = 1;
2872 }
2873 seq_printf(seq, "\n");
2874 }
2875
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 if (bond->params.mode == BOND_MODE_8023AD) {
2877 struct ad_info ad_info;
2878
2879 seq_puts(seq, "\n802.3ad info\n");
2880 seq_printf(seq, "LACP rate: %s\n",
2881 (bond->params.lacp_fast) ? "fast" : "slow");
2882
2883 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
2884 seq_printf(seq, "bond %s has no active aggregator\n",
2885 bond->dev->name);
2886 } else {
2887 seq_printf(seq, "Active Aggregator Info:\n");
2888
2889 seq_printf(seq, "\tAggregator ID: %d\n",
2890 ad_info.aggregator_id);
2891 seq_printf(seq, "\tNumber of ports: %d\n",
2892 ad_info.ports);
2893 seq_printf(seq, "\tActor Key: %d\n",
2894 ad_info.actor_key);
2895 seq_printf(seq, "\tPartner Key: %d\n",
2896 ad_info.partner_key);
2897 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2898 ad_info.partner_system[0],
2899 ad_info.partner_system[1],
2900 ad_info.partner_system[2],
2901 ad_info.partner_system[3],
2902 ad_info.partner_system[4],
2903 ad_info.partner_system[5]);
2904 }
2905 }
2906}
2907
2908static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
2909{
2910 struct bonding *bond = seq->private;
2911
2912 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
2913 seq_printf(seq, "MII Status: %s\n",
2914 (slave->link == BOND_LINK_UP) ? "up" : "down");
2915 seq_printf(seq, "Link Failure Count: %d\n",
2916 slave->link_failure_count);
2917
Jay Vosburgh217df672005-09-26 16:11:50 -07002918 seq_printf(seq,
2919 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
2920 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
2921 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
2922 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923
2924 if (bond->params.mode == BOND_MODE_8023AD) {
2925 const struct aggregator *agg
2926 = SLAVE_AD_INFO(slave).port.aggregator;
2927
2928 if (agg) {
2929 seq_printf(seq, "Aggregator ID: %d\n",
2930 agg->aggregator_identifier);
2931 } else {
2932 seq_puts(seq, "Aggregator ID: N/A\n");
2933 }
2934 }
2935}
2936
2937static int bond_info_seq_show(struct seq_file *seq, void *v)
2938{
2939 if (v == SEQ_START_TOKEN) {
2940 seq_printf(seq, "%s\n", version);
2941 bond_info_show_master(seq);
2942 } else {
2943 bond_info_show_slave(seq, v);
2944 }
2945
2946 return 0;
2947}
2948
2949static struct seq_operations bond_info_seq_ops = {
2950 .start = bond_info_seq_start,
2951 .next = bond_info_seq_next,
2952 .stop = bond_info_seq_stop,
2953 .show = bond_info_seq_show,
2954};
2955
2956static int bond_info_open(struct inode *inode, struct file *file)
2957{
2958 struct seq_file *seq;
2959 struct proc_dir_entry *proc;
2960 int res;
2961
2962 res = seq_open(file, &bond_info_seq_ops);
2963 if (!res) {
2964 /* recover the pointer buried in proc_dir_entry data */
2965 seq = file->private_data;
2966 proc = PDE(inode);
2967 seq->private = proc->data;
2968 }
2969
2970 return res;
2971}
2972
2973static struct file_operations bond_info_fops = {
2974 .owner = THIS_MODULE,
2975 .open = bond_info_open,
2976 .read = seq_read,
2977 .llseek = seq_lseek,
2978 .release = seq_release,
2979};
2980
2981static int bond_create_proc_entry(struct bonding *bond)
2982{
2983 struct net_device *bond_dev = bond->dev;
2984
2985 if (bond_proc_dir) {
2986 bond->proc_entry = create_proc_entry(bond_dev->name,
2987 S_IRUGO,
2988 bond_proc_dir);
2989 if (bond->proc_entry == NULL) {
2990 printk(KERN_WARNING DRV_NAME
2991 ": Warning: Cannot create /proc/net/%s/%s\n",
2992 DRV_NAME, bond_dev->name);
2993 } else {
2994 bond->proc_entry->data = bond;
2995 bond->proc_entry->proc_fops = &bond_info_fops;
2996 bond->proc_entry->owner = THIS_MODULE;
2997 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
2998 }
2999 }
3000
3001 return 0;
3002}
3003
3004static void bond_remove_proc_entry(struct bonding *bond)
3005{
3006 if (bond_proc_dir && bond->proc_entry) {
3007 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3008 memset(bond->proc_file_name, 0, IFNAMSIZ);
3009 bond->proc_entry = NULL;
3010 }
3011}
3012
3013/* Create the bonding directory under /proc/net, if doesn't exist yet.
3014 * Caller must hold rtnl_lock.
3015 */
3016static void bond_create_proc_dir(void)
3017{
3018 int len = strlen(DRV_NAME);
3019
3020 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3021 bond_proc_dir = bond_proc_dir->next) {
3022 if ((bond_proc_dir->namelen == len) &&
3023 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3024 break;
3025 }
3026 }
3027
3028 if (!bond_proc_dir) {
3029 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3030 if (bond_proc_dir) {
3031 bond_proc_dir->owner = THIS_MODULE;
3032 } else {
3033 printk(KERN_WARNING DRV_NAME
3034 ": Warning: cannot create /proc/net/%s\n",
3035 DRV_NAME);
3036 }
3037 }
3038}
3039
3040/* Destroy the bonding directory under /proc/net, if empty.
3041 * Caller must hold rtnl_lock.
3042 */
3043static void bond_destroy_proc_dir(void)
3044{
3045 struct proc_dir_entry *de;
3046
3047 if (!bond_proc_dir) {
3048 return;
3049 }
3050
3051 /* verify that the /proc dir is empty */
3052 for (de = bond_proc_dir->subdir; de; de = de->next) {
3053 /* ignore . and .. */
3054 if (*(de->name) != '.') {
3055 break;
3056 }
3057 }
3058
3059 if (de) {
3060 if (bond_proc_dir->owner == THIS_MODULE) {
3061 bond_proc_dir->owner = NULL;
3062 }
3063 } else {
3064 remove_proc_entry(DRV_NAME, proc_net);
3065 bond_proc_dir = NULL;
3066 }
3067}
3068#endif /* CONFIG_PROC_FS */
3069
3070/*-------------------------- netdev event handling --------------------------*/
3071
3072/*
3073 * Change device name
3074 */
3075static int bond_event_changename(struct bonding *bond)
3076{
3077#ifdef CONFIG_PROC_FS
3078 bond_remove_proc_entry(bond);
3079 bond_create_proc_entry(bond);
3080#endif
Mitch Williamsb76cdba2005-11-09 10:36:41 -08003081 down_write(&(bonding_rwsem));
3082 bond_destroy_sysfs_entry(bond);
3083 bond_create_sysfs_entry(bond);
3084 up_write(&(bonding_rwsem));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085 return NOTIFY_DONE;
3086}
3087
3088static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3089{
3090 struct bonding *event_bond = bond_dev->priv;
3091
3092 switch (event) {
3093 case NETDEV_CHANGENAME:
3094 return bond_event_changename(event_bond);
3095 case NETDEV_UNREGISTER:
3096 /*
3097 * TODO: remove a bond from the list?
3098 */
3099 break;
3100 default:
3101 break;
3102 }
3103
3104 return NOTIFY_DONE;
3105}
3106
3107static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3108{
3109 struct net_device *bond_dev = slave_dev->master;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04003110 struct bonding *bond = bond_dev->priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111
3112 switch (event) {
3113 case NETDEV_UNREGISTER:
3114 if (bond_dev) {
3115 bond_release(bond_dev, slave_dev);
3116 }
3117 break;
3118 case NETDEV_CHANGE:
3119 /*
3120 * TODO: is this what we get if somebody
3121 * sets up a hierarchical bond, then rmmod's
3122 * one of the slave bonding devices?
3123 */
3124 break;
3125 case NETDEV_DOWN:
3126 /*
3127 * ... Or is it this?
3128 */
3129 break;
3130 case NETDEV_CHANGEMTU:
3131 /*
3132 * TODO: Should slaves be allowed to
3133 * independently alter their MTU? For
3134 * an active-backup bond, slaves need
3135 * not be the same type of device, so
3136 * MTUs may vary. For other modes,
3137 * slaves arguably should have the
3138 * same MTUs. To do this, we'd need to
3139 * take over the slave's change_mtu
3140 * function for the duration of their
3141 * servitude.
3142 */
3143 break;
3144 case NETDEV_CHANGENAME:
3145 /*
3146 * TODO: handle changing the primary's name
3147 */
3148 break;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04003149 case NETDEV_FEAT_CHANGE:
3150 bond_compute_features(bond);
3151 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 default:
3153 break;
3154 }
3155
3156 return NOTIFY_DONE;
3157}
3158
3159/*
3160 * bond_netdev_event: handle netdev notifier chain events.
3161 *
3162 * This function receives events for the netdev chain. The caller (an
3163 * ioctl handler calling notifier_call_chain) holds the necessary
3164 * locks for us to safely manipulate the slave devices (RTNL lock,
3165 * dev_probe_lock).
3166 */
3167static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3168{
3169 struct net_device *event_dev = (struct net_device *)ptr;
3170
3171 dprintk("event_dev: %s, event: %lx\n",
3172 (event_dev ? event_dev->name : "None"),
3173 event);
3174
3175 if (event_dev->flags & IFF_MASTER) {
3176 dprintk("IFF_MASTER\n");
3177 return bond_master_netdev_event(event, event_dev);
3178 }
3179
3180 if (event_dev->flags & IFF_SLAVE) {
3181 dprintk("IFF_SLAVE\n");
3182 return bond_slave_netdev_event(event, event_dev);
3183 }
3184
3185 return NOTIFY_DONE;
3186}
3187
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003188/*
3189 * bond_inetaddr_event: handle inetaddr notifier chain events.
3190 *
3191 * We keep track of device IPs primarily to use as source addresses in
3192 * ARP monitor probes (rather than spewing out broadcasts all the time).
3193 *
3194 * We track one IP for the main device (if it has one), plus one per VLAN.
3195 */
3196static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3197{
3198 struct in_ifaddr *ifa = ptr;
3199 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3200 struct bonding *bond, *bond_next;
3201 struct vlan_entry *vlan, *vlan_next;
3202
3203 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3204 if (bond->dev == event_dev) {
3205 switch (event) {
3206 case NETDEV_UP:
3207 bond->master_ip = ifa->ifa_local;
3208 return NOTIFY_OK;
3209 case NETDEV_DOWN:
3210 bond->master_ip = bond_glean_dev_ip(bond->dev);
3211 return NOTIFY_OK;
3212 default:
3213 return NOTIFY_DONE;
3214 }
3215 }
3216
3217 if (list_empty(&bond->vlan_list))
3218 continue;
3219
3220 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3221 vlan_list) {
3222 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3223 if (vlan_dev == event_dev) {
3224 switch (event) {
3225 case NETDEV_UP:
3226 vlan->vlan_ip = ifa->ifa_local;
3227 return NOTIFY_OK;
3228 case NETDEV_DOWN:
3229 vlan->vlan_ip =
3230 bond_glean_dev_ip(vlan_dev);
3231 return NOTIFY_OK;
3232 default:
3233 return NOTIFY_DONE;
3234 }
3235 }
3236 }
3237 }
3238 return NOTIFY_DONE;
3239}
3240
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241static struct notifier_block bond_netdev_notifier = {
3242 .notifier_call = bond_netdev_event,
3243};
3244
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003245static struct notifier_block bond_inetaddr_notifier = {
3246 .notifier_call = bond_inetaddr_event,
3247};
3248
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249/*-------------------------- Packet type handling ---------------------------*/
3250
3251/* register to receive lacpdus on a bond */
3252static void bond_register_lacpdu(struct bonding *bond)
3253{
3254 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3255
3256 /* initialize packet type */
3257 pk_type->type = PKT_TYPE_LACPDU;
3258 pk_type->dev = bond->dev;
3259 pk_type->func = bond_3ad_lacpdu_recv;
3260
3261 dev_add_pack(pk_type);
3262}
3263
3264/* unregister to receive lacpdus on a bond */
3265static void bond_unregister_lacpdu(struct bonding *bond)
3266{
3267 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3268}
3269
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003270/*---------------------------- Hashing Policies -----------------------------*/
3271
3272/*
3273 * Hash for the the output device based upon layer 3 and layer 4 data. If
3274 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3275 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3276 */
3277static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3278 struct net_device *bond_dev, int count)
3279{
3280 struct ethhdr *data = (struct ethhdr *)skb->data;
3281 struct iphdr *iph = skb->nh.iph;
3282 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3283 int layer4_xor = 0;
3284
3285 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3286 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3287 (iph->protocol == IPPROTO_TCP ||
3288 iph->protocol == IPPROTO_UDP)) {
3289 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3290 }
3291 return (layer4_xor ^
3292 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3293
3294 }
3295
3296 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3297}
3298
3299/*
3300 * Hash for the output device based upon layer 2 data
3301 */
3302static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3303 struct net_device *bond_dev, int count)
3304{
3305 struct ethhdr *data = (struct ethhdr *)skb->data;
3306
3307 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3308}
3309
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310/*-------------------------- Device entry points ----------------------------*/
3311
3312static int bond_open(struct net_device *bond_dev)
3313{
3314 struct bonding *bond = bond_dev->priv;
3315 struct timer_list *mii_timer = &bond->mii_timer;
3316 struct timer_list *arp_timer = &bond->arp_timer;
3317
3318 bond->kill_timers = 0;
3319
3320 if ((bond->params.mode == BOND_MODE_TLB) ||
3321 (bond->params.mode == BOND_MODE_ALB)) {
3322 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3323
3324 /* bond_alb_initialize must be called before the timer
3325 * is started.
3326 */
3327 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3328 /* something went wrong - fail the open operation */
3329 return -1;
3330 }
3331
3332 init_timer(alb_timer);
3333 alb_timer->expires = jiffies + 1;
3334 alb_timer->data = (unsigned long)bond;
3335 alb_timer->function = (void *)&bond_alb_monitor;
3336 add_timer(alb_timer);
3337 }
3338
3339 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3340 init_timer(mii_timer);
3341 mii_timer->expires = jiffies + 1;
3342 mii_timer->data = (unsigned long)bond_dev;
3343 mii_timer->function = (void *)&bond_mii_monitor;
3344 add_timer(mii_timer);
3345 }
3346
3347 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3348 init_timer(arp_timer);
3349 arp_timer->expires = jiffies + 1;
3350 arp_timer->data = (unsigned long)bond_dev;
3351 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3352 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3353 } else {
3354 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3355 }
3356 add_timer(arp_timer);
3357 }
3358
3359 if (bond->params.mode == BOND_MODE_8023AD) {
3360 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3361 init_timer(ad_timer);
3362 ad_timer->expires = jiffies + 1;
3363 ad_timer->data = (unsigned long)bond;
3364 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3365 add_timer(ad_timer);
3366
3367 /* register to receive LACPDUs */
3368 bond_register_lacpdu(bond);
3369 }
3370
3371 return 0;
3372}
3373
3374static int bond_close(struct net_device *bond_dev)
3375{
3376 struct bonding *bond = bond_dev->priv;
3377
3378 if (bond->params.mode == BOND_MODE_8023AD) {
3379 /* Unregister the receive of LACPDUs */
3380 bond_unregister_lacpdu(bond);
3381 }
3382
3383 write_lock_bh(&bond->lock);
3384
3385 bond_mc_list_destroy(bond);
3386
3387 /* signal timers not to re-arm */
3388 bond->kill_timers = 1;
3389
3390 write_unlock_bh(&bond->lock);
3391
3392 /* del_timer_sync must run without holding the bond->lock
3393 * because a running timer might be trying to hold it too
3394 */
3395
3396 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3397 del_timer_sync(&bond->mii_timer);
3398 }
3399
3400 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3401 del_timer_sync(&bond->arp_timer);
3402 }
3403
3404 switch (bond->params.mode) {
3405 case BOND_MODE_8023AD:
3406 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3407 break;
3408 case BOND_MODE_TLB:
3409 case BOND_MODE_ALB:
3410 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3411 break;
3412 default:
3413 break;
3414 }
3415
3416 /* Release the bonded slaves */
3417 bond_release_all(bond_dev);
3418
3419 if ((bond->params.mode == BOND_MODE_TLB) ||
3420 (bond->params.mode == BOND_MODE_ALB)) {
3421 /* Must be called only after all
3422 * slaves have been released
3423 */
3424 bond_alb_deinitialize(bond);
3425 }
3426
3427 return 0;
3428}
3429
3430static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3431{
3432 struct bonding *bond = bond_dev->priv;
3433 struct net_device_stats *stats = &(bond->stats), *sstats;
3434 struct slave *slave;
3435 int i;
3436
3437 memset(stats, 0, sizeof(struct net_device_stats));
3438
3439 read_lock_bh(&bond->lock);
3440
3441 bond_for_each_slave(bond, slave, i) {
3442 sstats = slave->dev->get_stats(slave->dev);
3443
3444 stats->rx_packets += sstats->rx_packets;
3445 stats->rx_bytes += sstats->rx_bytes;
3446 stats->rx_errors += sstats->rx_errors;
3447 stats->rx_dropped += sstats->rx_dropped;
3448
3449 stats->tx_packets += sstats->tx_packets;
3450 stats->tx_bytes += sstats->tx_bytes;
3451 stats->tx_errors += sstats->tx_errors;
3452 stats->tx_dropped += sstats->tx_dropped;
3453
3454 stats->multicast += sstats->multicast;
3455 stats->collisions += sstats->collisions;
3456
3457 stats->rx_length_errors += sstats->rx_length_errors;
3458 stats->rx_over_errors += sstats->rx_over_errors;
3459 stats->rx_crc_errors += sstats->rx_crc_errors;
3460 stats->rx_frame_errors += sstats->rx_frame_errors;
3461 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3462 stats->rx_missed_errors += sstats->rx_missed_errors;
3463
3464 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3465 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3466 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3467 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3468 stats->tx_window_errors += sstats->tx_window_errors;
3469 }
3470
3471 read_unlock_bh(&bond->lock);
3472
3473 return stats;
3474}
3475
3476static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3477{
3478 struct net_device *slave_dev = NULL;
3479 struct ifbond k_binfo;
3480 struct ifbond __user *u_binfo = NULL;
3481 struct ifslave k_sinfo;
3482 struct ifslave __user *u_sinfo = NULL;
3483 struct mii_ioctl_data *mii = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 int res = 0;
3485
3486 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3487 bond_dev->name, cmd);
3488
3489 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490 case SIOCGMIIPHY:
3491 mii = if_mii(ifr);
3492 if (!mii) {
3493 return -EINVAL;
3494 }
3495 mii->phy_id = 0;
3496 /* Fall Through */
3497 case SIOCGMIIREG:
3498 /*
3499 * We do this again just in case we were called by SIOCGMIIREG
3500 * instead of SIOCGMIIPHY.
3501 */
3502 mii = if_mii(ifr);
3503 if (!mii) {
3504 return -EINVAL;
3505 }
3506
3507 if (mii->reg_num == 1) {
3508 struct bonding *bond = bond_dev->priv;
3509 mii->val_out = 0;
3510 read_lock_bh(&bond->lock);
3511 read_lock(&bond->curr_slave_lock);
3512 if (bond->curr_active_slave) {
3513 mii->val_out = BMSR_LSTATUS;
3514 }
3515 read_unlock(&bond->curr_slave_lock);
3516 read_unlock_bh(&bond->lock);
3517 }
3518
3519 return 0;
3520 case BOND_INFO_QUERY_OLD:
3521 case SIOCBONDINFOQUERY:
3522 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3523
3524 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3525 return -EFAULT;
3526 }
3527
3528 res = bond_info_query(bond_dev, &k_binfo);
3529 if (res == 0) {
3530 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3531 return -EFAULT;
3532 }
3533 }
3534
3535 return res;
3536 case BOND_SLAVE_INFO_QUERY_OLD:
3537 case SIOCBONDSLAVEINFOQUERY:
3538 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3539
3540 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3541 return -EFAULT;
3542 }
3543
3544 res = bond_slave_info_query(bond_dev, &k_sinfo);
3545 if (res == 0) {
3546 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3547 return -EFAULT;
3548 }
3549 }
3550
3551 return res;
3552 default:
3553 /* Go on */
3554 break;
3555 }
3556
3557 if (!capable(CAP_NET_ADMIN)) {
3558 return -EPERM;
3559 }
3560
Mitch Williamsb76cdba2005-11-09 10:36:41 -08003561 down_write(&(bonding_rwsem));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 slave_dev = dev_get_by_name(ifr->ifr_slave);
3563
3564 dprintk("slave_dev=%p: \n", slave_dev);
3565
3566 if (!slave_dev) {
3567 res = -ENODEV;
3568 } else {
3569 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3570 switch (cmd) {
3571 case BOND_ENSLAVE_OLD:
3572 case SIOCBONDENSLAVE:
3573 res = bond_enslave(bond_dev, slave_dev);
3574 break;
3575 case BOND_RELEASE_OLD:
3576 case SIOCBONDRELEASE:
3577 res = bond_release(bond_dev, slave_dev);
3578 break;
3579 case BOND_SETHWADDR_OLD:
3580 case SIOCBONDSETHWADDR:
3581 res = bond_sethwaddr(bond_dev, slave_dev);
3582 break;
3583 case BOND_CHANGE_ACTIVE_OLD:
3584 case SIOCBONDCHANGEACTIVE:
3585 res = bond_ioctl_change_active(bond_dev, slave_dev);
3586 break;
3587 default:
3588 res = -EOPNOTSUPP;
3589 }
3590
3591 dev_put(slave_dev);
3592 }
3593
Mitch Williamsb76cdba2005-11-09 10:36:41 -08003594 up_write(&(bonding_rwsem));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595 return res;
3596}
3597
3598static void bond_set_multicast_list(struct net_device *bond_dev)
3599{
3600 struct bonding *bond = bond_dev->priv;
3601 struct dev_mc_list *dmi;
3602
3603 write_lock_bh(&bond->lock);
3604
3605 /*
3606 * Do promisc before checking multicast_mode
3607 */
3608 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3609 bond_set_promiscuity(bond, 1);
3610 }
3611
3612 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3613 bond_set_promiscuity(bond, -1);
3614 }
3615
3616 /* set allmulti flag to slaves */
3617 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3618 bond_set_allmulti(bond, 1);
3619 }
3620
3621 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3622 bond_set_allmulti(bond, -1);
3623 }
3624
3625 bond->flags = bond_dev->flags;
3626
3627 /* looking for addresses to add to slaves' mc list */
3628 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3629 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3630 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3631 }
3632 }
3633
3634 /* looking for addresses to delete from slaves' list */
3635 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3636 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3637 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3638 }
3639 }
3640
3641 /* save master's multicast list */
3642 bond_mc_list_destroy(bond);
3643 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3644
3645 write_unlock_bh(&bond->lock);
3646}
3647
3648/*
3649 * Change the MTU of all of a master's slaves to match the master
3650 */
3651static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3652{
3653 struct bonding *bond = bond_dev->priv;
3654 struct slave *slave, *stop_at;
3655 int res = 0;
3656 int i;
3657
3658 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3659 (bond_dev ? bond_dev->name : "None"), new_mtu);
3660
3661 /* Can't hold bond->lock with bh disabled here since
3662 * some base drivers panic. On the other hand we can't
3663 * hold bond->lock without bh disabled because we'll
3664 * deadlock. The only solution is to rely on the fact
3665 * that we're under rtnl_lock here, and the slaves
3666 * list won't change. This doesn't solve the problem
3667 * of setting the slave's MTU while it is
3668 * transmitting, but the assumption is that the base
3669 * driver can handle that.
3670 *
3671 * TODO: figure out a way to safely iterate the slaves
3672 * list, but without holding a lock around the actual
3673 * call to the base driver.
3674 */
3675
3676 bond_for_each_slave(bond, slave, i) {
3677 dprintk("s %p s->p %p c_m %p\n", slave,
3678 slave->prev, slave->dev->change_mtu);
Mitch Williamse944ef72005-11-09 10:36:50 -08003679
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 res = dev_set_mtu(slave->dev, new_mtu);
3681
3682 if (res) {
3683 /* If we failed to set the slave's mtu to the new value
3684 * we must abort the operation even in ACTIVE_BACKUP
3685 * mode, because if we allow the backup slaves to have
3686 * different mtu values than the active slave we'll
3687 * need to change their mtu when doing a failover. That
3688 * means changing their mtu from timer context, which
3689 * is probably not a good idea.
3690 */
3691 dprintk("err %d %s\n", res, slave->dev->name);
3692 goto unwind;
3693 }
3694 }
3695
3696 bond_dev->mtu = new_mtu;
3697
3698 return 0;
3699
3700unwind:
3701 /* unwind from head to the slave that failed */
3702 stop_at = slave;
3703 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3704 int tmp_res;
3705
3706 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3707 if (tmp_res) {
3708 dprintk("unwind err %d dev %s\n", tmp_res,
3709 slave->dev->name);
3710 }
3711 }
3712
3713 return res;
3714}
3715
3716/*
3717 * Change HW address
3718 *
3719 * Note that many devices must be down to change the HW address, and
3720 * downing the master releases all slaves. We can make bonds full of
3721 * bonding devices to test this, however.
3722 */
3723static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3724{
3725 struct bonding *bond = bond_dev->priv;
3726 struct sockaddr *sa = addr, tmp_sa;
3727 struct slave *slave, *stop_at;
3728 int res = 0;
3729 int i;
3730
3731 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3732
3733 if (!is_valid_ether_addr(sa->sa_data)) {
3734 return -EADDRNOTAVAIL;
3735 }
3736
3737 /* Can't hold bond->lock with bh disabled here since
3738 * some base drivers panic. On the other hand we can't
3739 * hold bond->lock without bh disabled because we'll
3740 * deadlock. The only solution is to rely on the fact
3741 * that we're under rtnl_lock here, and the slaves
3742 * list won't change. This doesn't solve the problem
3743 * of setting the slave's hw address while it is
3744 * transmitting, but the assumption is that the base
3745 * driver can handle that.
3746 *
3747 * TODO: figure out a way to safely iterate the slaves
3748 * list, but without holding a lock around the actual
3749 * call to the base driver.
3750 */
3751
3752 bond_for_each_slave(bond, slave, i) {
3753 dprintk("slave %p %s\n", slave, slave->dev->name);
3754
3755 if (slave->dev->set_mac_address == NULL) {
3756 res = -EOPNOTSUPP;
3757 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3758 goto unwind;
3759 }
3760
3761 res = dev_set_mac_address(slave->dev, addr);
3762 if (res) {
3763 /* TODO: consider downing the slave
3764 * and retry ?
3765 * User should expect communications
3766 * breakage anyway until ARP finish
3767 * updating, so...
3768 */
3769 dprintk("err %d %s\n", res, slave->dev->name);
3770 goto unwind;
3771 }
3772 }
3773
3774 /* success */
3775 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3776 return 0;
3777
3778unwind:
3779 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3780 tmp_sa.sa_family = bond_dev->type;
3781
3782 /* unwind from head to the slave that failed */
3783 stop_at = slave;
3784 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3785 int tmp_res;
3786
3787 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3788 if (tmp_res) {
3789 dprintk("unwind err %d dev %s\n", tmp_res,
3790 slave->dev->name);
3791 }
3792 }
3793
3794 return res;
3795}
3796
3797static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3798{
3799 struct bonding *bond = bond_dev->priv;
3800 struct slave *slave, *start_at;
3801 int i;
3802 int res = 1;
3803
3804 read_lock(&bond->lock);
3805
3806 if (!BOND_IS_OK(bond)) {
3807 goto out;
3808 }
3809
3810 read_lock(&bond->curr_slave_lock);
3811 slave = start_at = bond->curr_active_slave;
3812 read_unlock(&bond->curr_slave_lock);
3813
3814 if (!slave) {
3815 goto out;
3816 }
3817
3818 bond_for_each_slave_from(bond, slave, i, start_at) {
3819 if (IS_UP(slave->dev) &&
3820 (slave->link == BOND_LINK_UP) &&
3821 (slave->state == BOND_STATE_ACTIVE)) {
3822 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3823
3824 write_lock(&bond->curr_slave_lock);
3825 bond->curr_active_slave = slave->next;
3826 write_unlock(&bond->curr_slave_lock);
3827
3828 break;
3829 }
3830 }
3831
3832
3833out:
3834 if (res) {
3835 /* no suitable interface, frame not sent */
3836 dev_kfree_skb(skb);
3837 }
3838 read_unlock(&bond->lock);
3839 return 0;
3840}
3841
John W. Linville075897c2005-09-28 17:50:53 -04003842static void bond_activebackup_xmit_copy(struct sk_buff *skb,
John W. Linvilledf498982005-10-18 21:30:58 -04003843 struct bonding *bond,
3844 struct slave *slave)
John W. Linville075897c2005-09-28 17:50:53 -04003845{
3846 struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
3847 struct ethhdr *eth_data;
3848 u8 *hwaddr;
3849 int res;
3850
3851 if (!skb2) {
3852 printk(KERN_ERR DRV_NAME ": Error: "
3853 "bond_activebackup_xmit_copy(): skb_copy() failed\n");
3854 return;
3855 }
3856
3857 skb2->mac.raw = (unsigned char *)skb2->data;
3858 eth_data = eth_hdr(skb2);
3859
John W. Linvilledf498982005-10-18 21:30:58 -04003860 /* Pick an appropriate source MAC address
3861 * -- use slave's perm MAC addr, unless used by bond
3862 * -- otherwise, borrow active slave's perm MAC addr
3863 * since that will not be used
3864 */
John W. Linville075897c2005-09-28 17:50:53 -04003865 hwaddr = slave->perm_hwaddr;
3866 if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
3867 hwaddr = bond->curr_active_slave->perm_hwaddr;
3868
3869 /* Set source MAC address appropriately */
3870 memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
3871
3872 res = bond_dev_queue_xmit(bond, skb2, slave->dev);
3873 if (res)
3874 dev_kfree_skb(skb2);
3875
3876 return;
3877}
3878
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879/*
3880 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3881 * the bond has a usable interface.
3882 */
3883static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3884{
3885 struct bonding *bond = bond_dev->priv;
3886 int res = 1;
3887
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888 read_lock(&bond->lock);
3889 read_lock(&bond->curr_slave_lock);
3890
3891 if (!BOND_IS_OK(bond)) {
3892 goto out;
3893 }
3894
John W. Linville075897c2005-09-28 17:50:53 -04003895 if (!bond->curr_active_slave)
3896 goto out;
3897
3898 /* Xmit IGMP frames on all slaves to ensure rapid fail-over
3899 for multicast traffic on snooping switches */
3900 if (skb->protocol == __constant_htons(ETH_P_IP) &&
3901 skb->nh.iph->protocol == IPPROTO_IGMP) {
3902 struct slave *slave, *active_slave;
3903 int i;
3904
3905 active_slave = bond->curr_active_slave;
3906 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
3907 active_slave->prev)
3908 if (IS_UP(slave->dev) &&
3909 (slave->link == BOND_LINK_UP))
3910 bond_activebackup_xmit_copy(skb, bond, slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 }
3912
John W. Linville075897c2005-09-28 17:50:53 -04003913 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
3914
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915out:
3916 if (res) {
3917 /* no suitable interface, frame not sent */
3918 dev_kfree_skb(skb);
3919 }
3920 read_unlock(&bond->curr_slave_lock);
3921 read_unlock(&bond->lock);
3922 return 0;
3923}
3924
3925/*
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003926 * In bond_xmit_xor() , we determine the output device by using a pre-
3927 * determined xmit_hash_policy(), If the selected device is not enabled,
3928 * find the next active slave.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003929 */
3930static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3931{
3932 struct bonding *bond = bond_dev->priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933 struct slave *slave, *start_at;
3934 int slave_no;
3935 int i;
3936 int res = 1;
3937
3938 read_lock(&bond->lock);
3939
3940 if (!BOND_IS_OK(bond)) {
3941 goto out;
3942 }
3943
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003944 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003945
3946 bond_for_each_slave(bond, slave, i) {
3947 slave_no--;
3948 if (slave_no < 0) {
3949 break;
3950 }
3951 }
3952
3953 start_at = slave;
3954
3955 bond_for_each_slave_from(bond, slave, i, start_at) {
3956 if (IS_UP(slave->dev) &&
3957 (slave->link == BOND_LINK_UP) &&
3958 (slave->state == BOND_STATE_ACTIVE)) {
3959 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3960 break;
3961 }
3962 }
3963
3964out:
3965 if (res) {
3966 /* no suitable interface, frame not sent */
3967 dev_kfree_skb(skb);
3968 }
3969 read_unlock(&bond->lock);
3970 return 0;
3971}
3972
3973/*
3974 * in broadcast mode, we send everything to all usable interfaces.
3975 */
3976static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3977{
3978 struct bonding *bond = bond_dev->priv;
3979 struct slave *slave, *start_at;
3980 struct net_device *tx_dev = NULL;
3981 int i;
3982 int res = 1;
3983
3984 read_lock(&bond->lock);
3985
3986 if (!BOND_IS_OK(bond)) {
3987 goto out;
3988 }
3989
3990 read_lock(&bond->curr_slave_lock);
3991 start_at = bond->curr_active_slave;
3992 read_unlock(&bond->curr_slave_lock);
3993
3994 if (!start_at) {
3995 goto out;
3996 }
3997
3998 bond_for_each_slave_from(bond, slave, i, start_at) {
3999 if (IS_UP(slave->dev) &&
4000 (slave->link == BOND_LINK_UP) &&
4001 (slave->state == BOND_STATE_ACTIVE)) {
4002 if (tx_dev) {
4003 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4004 if (!skb2) {
4005 printk(KERN_ERR DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08004006 ": %s: Error: bond_xmit_broadcast(): "
4007 "skb_clone() failed\n",
4008 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009 continue;
4010 }
4011
4012 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4013 if (res) {
4014 dev_kfree_skb(skb2);
4015 continue;
4016 }
4017 }
4018 tx_dev = slave->dev;
4019 }
4020 }
4021
4022 if (tx_dev) {
4023 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4024 }
4025
4026out:
4027 if (res) {
4028 /* no suitable interface, frame not sent */
4029 dev_kfree_skb(skb);
4030 }
4031 /* frame sent to all suitable interfaces */
4032 read_unlock(&bond->lock);
4033 return 0;
4034}
4035
4036/*------------------------- Device initialization ---------------------------*/
4037
4038/*
4039 * set bond mode specific net device operations
4040 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08004041void bond_set_mode_ops(struct bonding *bond, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042{
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004043 struct net_device *bond_dev = bond->dev;
4044
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045 switch (mode) {
4046 case BOND_MODE_ROUNDROBIN:
4047 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4048 break;
4049 case BOND_MODE_ACTIVEBACKUP:
4050 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4051 break;
4052 case BOND_MODE_XOR:
4053 bond_dev->hard_start_xmit = bond_xmit_xor;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004054 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4055 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4056 else
4057 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058 break;
4059 case BOND_MODE_BROADCAST:
4060 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4061 break;
4062 case BOND_MODE_8023AD:
4063 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004064 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4065 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4066 else
4067 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004068 break;
4069 case BOND_MODE_TLB:
4070 case BOND_MODE_ALB:
4071 bond_dev->hard_start_xmit = bond_alb_xmit;
4072 bond_dev->set_mac_address = bond_alb_set_mac_address;
4073 break;
4074 default:
4075 /* Should never happen, mode already checked */
4076 printk(KERN_ERR DRV_NAME
Mitch Williams4e0952c2005-11-09 10:34:57 -08004077 ": %s: Error: Unknown bonding mode %d\n",
4078 bond_dev->name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079 mode);
4080 break;
4081 }
4082}
4083
Jay Vosburgh217df672005-09-26 16:11:50 -07004084static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4085 struct ethtool_drvinfo *drvinfo)
4086{
4087 strncpy(drvinfo->driver, DRV_NAME, 32);
4088 strncpy(drvinfo->version, DRV_VERSION, 32);
4089 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4090}
4091
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004092static struct ethtool_ops bond_ethtool_ops = {
4093 .get_tx_csum = ethtool_op_get_tx_csum,
Jay Vosburgha0de3ad2006-01-30 15:40:59 -08004094 .get_tso = ethtool_op_get_tso,
4095 .get_ufo = ethtool_op_get_ufo,
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004096 .get_sg = ethtool_op_get_sg,
Jay Vosburgh217df672005-09-26 16:11:50 -07004097 .get_drvinfo = bond_ethtool_get_drvinfo,
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004098};
4099
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100/*
4101 * Does not allocate but creates a /proc entry.
4102 * Allowed to fail.
4103 */
Mitch Williams3c535952005-11-09 10:36:11 -08004104static int bond_init(struct net_device *bond_dev, struct bond_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105{
4106 struct bonding *bond = bond_dev->priv;
4107
4108 dprintk("Begin bond_init for %s\n", bond_dev->name);
4109
4110 /* initialize rwlocks */
4111 rwlock_init(&bond->lock);
4112 rwlock_init(&bond->curr_slave_lock);
4113
4114 bond->params = *params; /* copy params struct */
4115
4116 /* Initialize pointers */
4117 bond->first_slave = NULL;
4118 bond->curr_active_slave = NULL;
4119 bond->current_arp_slave = NULL;
4120 bond->primary_slave = NULL;
4121 bond->dev = bond_dev;
4122 INIT_LIST_HEAD(&bond->vlan_list);
4123
4124 /* Initialize the device entry points */
4125 bond_dev->open = bond_open;
4126 bond_dev->stop = bond_close;
4127 bond_dev->get_stats = bond_get_stats;
4128 bond_dev->do_ioctl = bond_do_ioctl;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004129 bond_dev->ethtool_ops = &bond_ethtool_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130 bond_dev->set_multicast_list = bond_set_multicast_list;
4131 bond_dev->change_mtu = bond_change_mtu;
4132 bond_dev->set_mac_address = bond_set_mac_address;
4133
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004134 bond_set_mode_ops(bond, bond->params.mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004135
4136 bond_dev->destructor = free_netdev;
4137
4138 /* Initialize the device options */
4139 bond_dev->tx_queue_len = 0;
4140 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4141
4142 /* At first, we block adding VLANs. That's the only way to
4143 * prevent problems that occur when adding VLANs over an
4144 * empty bond. The block will be removed once non-challenged
4145 * slaves are enslaved.
4146 */
4147 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4148
4149 /* don't acquire bond device's xmit_lock when
4150 * transmitting */
4151 bond_dev->features |= NETIF_F_LLTX;
4152
4153 /* By default, we declare the bond to be fully
4154 * VLAN hardware accelerated capable. Special
4155 * care is taken in the various xmit functions
4156 * when there are slaves that are not hw accel
4157 * capable
4158 */
4159 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4160 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4161 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4162 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4163 NETIF_F_HW_VLAN_RX |
4164 NETIF_F_HW_VLAN_FILTER);
4165
4166#ifdef CONFIG_PROC_FS
4167 bond_create_proc_entry(bond);
4168#endif
4169
4170 list_add_tail(&bond->bond_list, &bond_dev_list);
4171
4172 return 0;
4173}
4174
4175/* De-initialize device specific data.
4176 * Caller must hold rtnl_lock.
4177 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08004178void bond_deinit(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179{
4180 struct bonding *bond = bond_dev->priv;
4181
4182 list_del(&bond->bond_list);
4183
4184#ifdef CONFIG_PROC_FS
4185 bond_remove_proc_entry(bond);
4186#endif
4187}
4188
4189/* Unregister and free all bond devices.
4190 * Caller must hold rtnl_lock.
4191 */
4192static void bond_free_all(void)
4193{
4194 struct bonding *bond, *nxt;
4195
4196 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4197 struct net_device *bond_dev = bond->dev;
4198
4199 unregister_netdevice(bond_dev);
4200 bond_deinit(bond_dev);
4201 }
4202
4203#ifdef CONFIG_PROC_FS
4204 bond_destroy_proc_dir();
4205#endif
4206}
4207
4208/*------------------------- Module initialization ---------------------------*/
4209
4210/*
4211 * Convert string input module parms. Accept either the
4212 * number of the mode or its string name.
4213 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08004214int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215{
4216 int i;
4217
4218 for (i = 0; tbl[i].modename; i++) {
4219 if ((isdigit(*mode_arg) &&
4220 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4221 (strncmp(mode_arg, tbl[i].modename,
4222 strlen(tbl[i].modename)) == 0)) {
4223 return tbl[i].mode;
4224 }
4225 }
4226
4227 return -1;
4228}
4229
4230static int bond_check_params(struct bond_params *params)
4231{
4232 /*
4233 * Convert string parameters.
4234 */
4235 if (mode) {
4236 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4237 if (bond_mode == -1) {
4238 printk(KERN_ERR DRV_NAME
4239 ": Error: Invalid bonding mode \"%s\"\n",
4240 mode == NULL ? "NULL" : mode);
4241 return -EINVAL;
4242 }
4243 }
4244
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004245 if (xmit_hash_policy) {
4246 if ((bond_mode != BOND_MODE_XOR) &&
4247 (bond_mode != BOND_MODE_8023AD)) {
4248 printk(KERN_INFO DRV_NAME
4249 ": xor_mode param is irrelevant in mode %s\n",
4250 bond_mode_name(bond_mode));
4251 } else {
4252 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4253 xmit_hashtype_tbl);
4254 if (xmit_hashtype == -1) {
4255 printk(KERN_ERR DRV_NAME
4256 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4257 xmit_hash_policy == NULL ? "NULL" :
4258 xmit_hash_policy);
4259 return -EINVAL;
4260 }
4261 }
4262 }
4263
Linus Torvalds1da177e2005-04-16 15:20:36 -07004264 if (lacp_rate) {
4265 if (bond_mode != BOND_MODE_8023AD) {
4266 printk(KERN_INFO DRV_NAME
4267 ": lacp_rate param is irrelevant in mode %s\n",
4268 bond_mode_name(bond_mode));
4269 } else {
4270 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4271 if (lacp_fast == -1) {
4272 printk(KERN_ERR DRV_NAME
4273 ": Error: Invalid lacp rate \"%s\"\n",
4274 lacp_rate == NULL ? "NULL" : lacp_rate);
4275 return -EINVAL;
4276 }
4277 }
4278 }
4279
4280 if (max_bonds < 1 || max_bonds > INT_MAX) {
4281 printk(KERN_WARNING DRV_NAME
4282 ": Warning: max_bonds (%d) not in range %d-%d, so it "
Mitch Williams4e0952c2005-11-09 10:34:57 -08004283 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004284 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4285 max_bonds = BOND_DEFAULT_MAX_BONDS;
4286 }
4287
4288 if (miimon < 0) {
4289 printk(KERN_WARNING DRV_NAME
4290 ": Warning: miimon module parameter (%d), "
4291 "not in range 0-%d, so it was reset to %d\n",
4292 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4293 miimon = BOND_LINK_MON_INTERV;
4294 }
4295
4296 if (updelay < 0) {
4297 printk(KERN_WARNING DRV_NAME
4298 ": Warning: updelay module parameter (%d), "
4299 "not in range 0-%d, so it was reset to 0\n",
4300 updelay, INT_MAX);
4301 updelay = 0;
4302 }
4303
4304 if (downdelay < 0) {
4305 printk(KERN_WARNING DRV_NAME
4306 ": Warning: downdelay module parameter (%d), "
4307 "not in range 0-%d, so it was reset to 0\n",
4308 downdelay, INT_MAX);
4309 downdelay = 0;
4310 }
4311
4312 if ((use_carrier != 0) && (use_carrier != 1)) {
4313 printk(KERN_WARNING DRV_NAME
4314 ": Warning: use_carrier module parameter (%d), "
4315 "not of valid value (0/1), so it was set to 1\n",
4316 use_carrier);
4317 use_carrier = 1;
4318 }
4319
4320 /* reset values for 802.3ad */
4321 if (bond_mode == BOND_MODE_8023AD) {
4322 if (!miimon) {
4323 printk(KERN_WARNING DRV_NAME
4324 ": Warning: miimon must be specified, "
4325 "otherwise bonding will not detect link "
4326 "failure, speed and duplex which are "
4327 "essential for 802.3ad operation\n");
4328 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4329 miimon = 100;
4330 }
4331 }
4332
4333 /* reset values for TLB/ALB */
4334 if ((bond_mode == BOND_MODE_TLB) ||
4335 (bond_mode == BOND_MODE_ALB)) {
4336 if (!miimon) {
4337 printk(KERN_WARNING DRV_NAME
4338 ": Warning: miimon must be specified, "
4339 "otherwise bonding will not detect link "
4340 "failure and link speed which are essential "
4341 "for TLB/ALB load balancing\n");
4342 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4343 miimon = 100;
4344 }
4345 }
4346
4347 if (bond_mode == BOND_MODE_ALB) {
4348 printk(KERN_NOTICE DRV_NAME
4349 ": In ALB mode you might experience client "
4350 "disconnections upon reconnection of a link if the "
4351 "bonding module updelay parameter (%d msec) is "
4352 "incompatible with the forwarding delay time of the "
4353 "switch\n",
4354 updelay);
4355 }
4356
4357 if (!miimon) {
4358 if (updelay || downdelay) {
4359 /* just warn the user the up/down delay will have
4360 * no effect since miimon is zero...
4361 */
4362 printk(KERN_WARNING DRV_NAME
4363 ": Warning: miimon module parameter not set "
4364 "and updelay (%d) or downdelay (%d) module "
4365 "parameter is set; updelay and downdelay have "
4366 "no effect unless miimon is set\n",
4367 updelay, downdelay);
4368 }
4369 } else {
4370 /* don't allow arp monitoring */
4371 if (arp_interval) {
4372 printk(KERN_WARNING DRV_NAME
4373 ": Warning: miimon (%d) and arp_interval (%d) "
4374 "can't be used simultaneously, disabling ARP "
4375 "monitoring\n",
4376 miimon, arp_interval);
4377 arp_interval = 0;
4378 }
4379
4380 if ((updelay % miimon) != 0) {
4381 printk(KERN_WARNING DRV_NAME
4382 ": Warning: updelay (%d) is not a multiple "
4383 "of miimon (%d), updelay rounded to %d ms\n",
4384 updelay, miimon, (updelay / miimon) * miimon);
4385 }
4386
4387 updelay /= miimon;
4388
4389 if ((downdelay % miimon) != 0) {
4390 printk(KERN_WARNING DRV_NAME
4391 ": Warning: downdelay (%d) is not a multiple "
4392 "of miimon (%d), downdelay rounded to %d ms\n",
4393 downdelay, miimon,
4394 (downdelay / miimon) * miimon);
4395 }
4396
4397 downdelay /= miimon;
4398 }
4399
4400 if (arp_interval < 0) {
4401 printk(KERN_WARNING DRV_NAME
4402 ": Warning: arp_interval module parameter (%d) "
4403 ", not in range 0-%d, so it was reset to %d\n",
4404 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4405 arp_interval = BOND_LINK_ARP_INTERV;
4406 }
4407
4408 for (arp_ip_count = 0;
4409 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4410 arp_ip_count++) {
4411 /* not complete check, but should be good enough to
4412 catch mistakes */
4413 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4414 printk(KERN_WARNING DRV_NAME
4415 ": Warning: bad arp_ip_target module parameter "
4416 "(%s), ARP monitoring will not be performed\n",
4417 arp_ip_target[arp_ip_count]);
4418 arp_interval = 0;
4419 } else {
4420 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4421 arp_target[arp_ip_count] = ip;
4422 }
4423 }
4424
4425 if (arp_interval && !arp_ip_count) {
4426 /* don't allow arping if no arp_ip_target given... */
4427 printk(KERN_WARNING DRV_NAME
4428 ": Warning: arp_interval module parameter (%d) "
4429 "specified without providing an arp_ip_target "
4430 "parameter, arp_interval was reset to 0\n",
4431 arp_interval);
4432 arp_interval = 0;
4433 }
4434
4435 if (miimon) {
4436 printk(KERN_INFO DRV_NAME
4437 ": MII link monitoring set to %d ms\n",
4438 miimon);
4439 } else if (arp_interval) {
4440 int i;
4441
4442 printk(KERN_INFO DRV_NAME
4443 ": ARP monitoring set to %d ms with %d target(s):",
4444 arp_interval, arp_ip_count);
4445
4446 for (i = 0; i < arp_ip_count; i++)
4447 printk (" %s", arp_ip_target[i]);
4448
4449 printk("\n");
4450
4451 } else {
4452 /* miimon and arp_interval not set, we need one so things
4453 * work as expected, see bonding.txt for details
4454 */
4455 printk(KERN_WARNING DRV_NAME
4456 ": Warning: either miimon or arp_interval and "
4457 "arp_ip_target module parameters must be specified, "
4458 "otherwise bonding will not detect link failures! see "
4459 "bonding.txt for details.\n");
4460 }
4461
4462 if (primary && !USES_PRIMARY(bond_mode)) {
4463 /* currently, using a primary only makes sense
4464 * in active backup, TLB or ALB modes
4465 */
4466 printk(KERN_WARNING DRV_NAME
4467 ": Warning: %s primary device specified but has no "
4468 "effect in %s mode\n",
4469 primary, bond_mode_name(bond_mode));
4470 primary = NULL;
4471 }
4472
4473 /* fill params struct with the proper values */
4474 params->mode = bond_mode;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004475 params->xmit_policy = xmit_hashtype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004476 params->miimon = miimon;
4477 params->arp_interval = arp_interval;
4478 params->updelay = updelay;
4479 params->downdelay = downdelay;
4480 params->use_carrier = use_carrier;
4481 params->lacp_fast = lacp_fast;
4482 params->primary[0] = 0;
4483
4484 if (primary) {
4485 strncpy(params->primary, primary, IFNAMSIZ);
4486 params->primary[IFNAMSIZ - 1] = 0;
4487 }
4488
4489 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4490
4491 return 0;
4492}
4493
Mitch Williamsdfe60392005-11-09 10:36:04 -08004494/* Create a new bond based on the specified name and bonding parameters.
4495 * Caller must NOT hold rtnl_lock; we need to release it here before we
4496 * set up our sysfs entries.
4497 */
4498int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4499{
4500 struct net_device *bond_dev;
4501 int res;
4502
4503 rtnl_lock();
4504 bond_dev = alloc_netdev(sizeof(struct bonding), name, ether_setup);
4505 if (!bond_dev) {
4506 printk(KERN_ERR DRV_NAME
4507 ": %s: eek! can't alloc netdev!\n",
4508 name);
4509 res = -ENOMEM;
4510 goto out_rtnl;
4511 }
4512
4513 /* bond_init() must be called after dev_alloc_name() (for the
4514 * /proc files), but before register_netdevice(), because we
4515 * need to set function pointers.
4516 */
4517
4518 res = bond_init(bond_dev, params);
4519 if (res < 0) {
4520 goto out_netdev;
4521 }
4522
4523 SET_MODULE_OWNER(bond_dev);
4524
4525 res = register_netdevice(bond_dev);
4526 if (res < 0) {
4527 goto out_bond;
4528 }
4529 if (newbond)
4530 *newbond = bond_dev->priv;
4531
4532 rtnl_unlock(); /* allows sysfs registration of net device */
Mitch Williamsb76cdba2005-11-09 10:36:41 -08004533 res = bond_create_sysfs_entry(bond_dev->priv);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004534 goto done;
4535out_bond:
4536 bond_deinit(bond_dev);
4537out_netdev:
4538 free_netdev(bond_dev);
4539out_rtnl:
4540 rtnl_unlock();
4541done:
4542 return res;
4543}
4544
Linus Torvalds1da177e2005-04-16 15:20:36 -07004545static int __init bonding_init(void)
4546{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004547 int i;
4548 int res;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004549 char new_bond_name[8]; /* Enough room for 999 bonds at init. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004550
4551 printk(KERN_INFO "%s", version);
4552
Mitch Williamsdfe60392005-11-09 10:36:04 -08004553 res = bond_check_params(&bonding_defaults);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004554 if (res) {
Mitch Williamsdfe60392005-11-09 10:36:04 -08004555 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004556 }
4557
Linus Torvalds1da177e2005-04-16 15:20:36 -07004558#ifdef CONFIG_PROC_FS
4559 bond_create_proc_dir();
4560#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004561 for (i = 0; i < max_bonds; i++) {
Mitch Williamsdfe60392005-11-09 10:36:04 -08004562 sprintf(new_bond_name, "bond%d",i);
4563 res = bond_create(new_bond_name,&bonding_defaults, NULL);
4564 if (res)
4565 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004566 }
4567
Mitch Williamsb76cdba2005-11-09 10:36:41 -08004568 res = bond_create_sysfs();
4569 if (res)
4570 goto err;
4571
Linus Torvalds1da177e2005-04-16 15:20:36 -07004572 register_netdevice_notifier(&bond_netdev_notifier);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04004573 register_inetaddr_notifier(&bond_inetaddr_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004574
Mitch Williamsdfe60392005-11-09 10:36:04 -08004575 goto out;
4576err:
Florin Malita40abc272005-09-18 00:24:12 -07004577 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 bond_free_all();
Mitch Williamsb76cdba2005-11-09 10:36:41 -08004579 bond_destroy_sysfs();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004580 rtnl_unlock();
Mitch Williamsdfe60392005-11-09 10:36:04 -08004581out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004582 return res;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004583
Linus Torvalds1da177e2005-04-16 15:20:36 -07004584}
4585
4586static void __exit bonding_exit(void)
4587{
4588 unregister_netdevice_notifier(&bond_netdev_notifier);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04004589 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004590
4591 rtnl_lock();
4592 bond_free_all();
Mitch Williamsb76cdba2005-11-09 10:36:41 -08004593 bond_destroy_sysfs();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004594 rtnl_unlock();
4595}
4596
4597module_init(bonding_init);
4598module_exit(bonding_exit);
4599MODULE_LICENSE("GPL");
4600MODULE_VERSION(DRV_VERSION);
4601MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4602MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4603MODULE_SUPPORTED_DEVICE("most ethernet devices");
4604
4605/*
4606 * Local variables:
4607 * c-indent-level: 8
4608 * c-basic-offset: 8
4609 * tab-width: 8
4610 * End:
4611 */
4612