]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/bonding/bond_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ps3: gigabit ethernet driver for PS3, take3
[net-next-2.6.git] / drivers / net / bonding / bond_main.c
CommitLineData
1da177e4
LT		 1/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
1da177e4
LT		 32 */
33
34//#define BONDING_DEBUG 1
35
1da177e4
LT		 36#include <linux/kernel.h>
37#include <linux/module.h>
1da177e4
LT		 38#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
169a3e66 44#include <net/ip.h>
1da177e4 45#include <linux/ip.h>
169a3e66
JV		 46#include <linux/tcp.h>
47#include <linux/udp.h>
1da177e4
LT		 48#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
56#include <asm/system.h>
57#include <asm/io.h>
58#include <asm/dma.h>
59#include <asm/uaccess.h>
60#include <linux/errno.h>
61#include <linux/netdevice.h>
62#include <linux/inetdevice.h>
a816c7c7 63#include <linux/igmp.h>
1da177e4
LT		 64#include <linux/etherdevice.h>
65#include <linux/skbuff.h>
66#include <net/sock.h>
67#include <linux/rtnetlink.h>
68#include <linux/proc_fs.h>
69#include <linux/seq_file.h>
70#include <linux/smp.h>
71#include <linux/if_ether.h>
72#include <net/arp.h>
73#include <linux/mii.h>
74#include <linux/ethtool.h>
75#include <linux/if_vlan.h>
76#include <linux/if_bonding.h>
c3ade5ca 77#include <net/route.h>
1da177e4
LT		 78#include "bonding.h"
79#include "bond_3ad.h"
80#include "bond_alb.h"
81
82/*---------------------------- Module parameters ----------------------------*/
83
84/* monitor all links that often (in milliseconds). <=0 disables monitoring */
85#define BOND_LINK_MON_INTERV 0
86#define BOND_LINK_ARP_INTERV 0
87
88static int max_bonds = BOND_DEFAULT_MAX_BONDS;
89static int miimon = BOND_LINK_MON_INTERV;
90static int updelay = 0;
91static int downdelay = 0;
92static int use_carrier = 1;
93static char *mode = NULL;
94static char *primary = NULL;
95static char *lacp_rate = NULL;
169a3e66 96static char *xmit_hash_policy = NULL;
1da177e4
LT		 97static int arp_interval = BOND_LINK_ARP_INTERV;
98static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
f5b2b966 99static char *arp_validate = NULL;
12479f9a 100struct bond_params bonding_defaults;
1da177e4
LT		 101
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);
2ac47660
MW		 109MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
110 "in milliseconds");
1da177e4 111module_param(use_carrier, int, 0);
2ac47660
MW		 112MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
113 "0 for off, 1 for on (default)");
1da177e4 114module_param(mode, charp, 0);
2ac47660
MW		 115MODULE_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");
1da177e4
LT		 119module_param(primary, charp, 0);
120MODULE_PARM_DESC(primary, "Primary network device to use");
121module_param(lacp_rate, charp, 0);
2ac47660
MW		 122MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
123 "(slow/fast)");
169a3e66 124module_param(xmit_hash_policy, charp, 0);
2ac47660
MW		 125MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
126 ", 1 for layer 3+4");
1da177e4
LT		 127module_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");
f5b2b966
JV		 131module_param(arp_validate, charp, 0);
132MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
1da177e4
LT		 133
134/*----------------------------- Global variables ----------------------------*/
135
f71e1309 136static const char * const version =
1da177e4
LT		 137 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
138
12479f9a 139LIST_HEAD(bond_dev_list);
1da177e4
LT		 140
141#ifdef CONFIG_PROC_FS
142static struct proc_dir_entry *bond_proc_dir = NULL;
143#endif
144
b76cdba9 145extern struct rw_semaphore bonding_rwsem;
1da177e4
LT		 146static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
147static int arp_ip_count = 0;
1da177e4 148static int bond_mode = BOND_MODE_ROUNDROBIN;
169a3e66 149static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
1da177e4 150static int lacp_fast = 0;
217df670 151
1da177e4 152
12479f9a 153struct bond_parm_tbl bond_lacp_tbl[] = {
1da177e4
LT		 154{ "slow", AD_LACP_SLOW},
155{ "fast", AD_LACP_FAST},
156{ NULL, -1},
157};
158
12479f9a 159struct bond_parm_tbl bond_mode_tbl[] = {
1da177e4
LT		 160{ "balance-rr", BOND_MODE_ROUNDROBIN},
161{ "active-backup", BOND_MODE_ACTIVEBACKUP},
162{ "balance-xor", BOND_MODE_XOR},
163{ "broadcast", BOND_MODE_BROADCAST},
164{ "802.3ad", BOND_MODE_8023AD},
165{ "balance-tlb", BOND_MODE_TLB},
166{ "balance-alb", BOND_MODE_ALB},
167{ NULL, -1},
168};
169
12479f9a 170struct bond_parm_tbl xmit_hashtype_tbl[] = {
169a3e66
JV		 171{ "layer2", BOND_XMIT_POLICY_LAYER2},
172{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
173{ NULL, -1},
174};
175
f5b2b966
JV		 176struct bond_parm_tbl arp_validate_tbl[] = {
177{ "none", BOND_ARP_VALIDATE_NONE},
178{ "active", BOND_ARP_VALIDATE_ACTIVE},
179{ "backup", BOND_ARP_VALIDATE_BACKUP},
180{ "all", BOND_ARP_VALIDATE_ALL},
181{ NULL, -1},
182};
183
1da177e4
LT		 184/*-------------------------- Forward declarations ---------------------------*/
185
c3ade5ca 186static void bond_send_gratuitous_arp(struct bonding *bond);
1da177e4
LT		 187
188/*---------------------------- General routines -----------------------------*/
189
a77b5325 190const char *bond_mode_name(int mode)
1da177e4
LT		 191{
192 switch (mode) {
193 case BOND_MODE_ROUNDROBIN :
194 return "load balancing (round-robin)";
195 case BOND_MODE_ACTIVEBACKUP :
196 return "fault-tolerance (active-backup)";
197 case BOND_MODE_XOR :
198 return "load balancing (xor)";
199 case BOND_MODE_BROADCAST :
200 return "fault-tolerance (broadcast)";
201 case BOND_MODE_8023AD:
202 return "IEEE 802.3ad Dynamic link aggregation";
203 case BOND_MODE_TLB:
204 return "transmit load balancing";
205 case BOND_MODE_ALB:
206 return "adaptive load balancing";
207 default:
208 return "unknown";
209 }
210}
211
212/*---------------------------------- VLAN -----------------------------------*/
213
214/**
215 * bond_add_vlan - add a new vlan id on bond
216 * @bond: bond that got the notification
217 * @vlan_id: the vlan id to add
218 *
219 * Returns -ENOMEM if allocation failed.
220 */
221static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
222{
223 struct vlan_entry *vlan;
224
225 dprintk("bond: %s, vlan id %d\n",
226 (bond ? bond->dev->name: "None"), vlan_id);
227
228 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
229 if (!vlan) {
230 return -ENOMEM;
231 }
232
233 INIT_LIST_HEAD(&vlan->vlan_list);
234 vlan->vlan_id = vlan_id;
c3ade5ca 235 vlan->vlan_ip = 0;
1da177e4
LT		 236
237 write_lock_bh(&bond->lock);
238
239 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
240
241 write_unlock_bh(&bond->lock);
242
243 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
244
245 return 0;
246}
247
248/**
249 * bond_del_vlan - delete a vlan id from bond
250 * @bond: bond that got the notification
251 * @vlan_id: the vlan id to delete
252 *
253 * returns -ENODEV if @vlan_id was not found in @bond.
254 */
255static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
256{
257 struct vlan_entry *vlan, *next;
258 int res = -ENODEV;
259
260 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
261
262 write_lock_bh(&bond->lock);
263
264 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
265 if (vlan->vlan_id == vlan_id) {
266 list_del(&vlan->vlan_list);
267
268 if ((bond->params.mode == BOND_MODE_TLB) ||
269 (bond->params.mode == BOND_MODE_ALB)) {
270 bond_alb_clear_vlan(bond, vlan_id);
271 }
272
273 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
274 bond->dev->name);
275
276 kfree(vlan);
277
278 if (list_empty(&bond->vlan_list) &&
279 (bond->slave_cnt == 0)) {
280 /* Last VLAN removed and no slaves, so
281 * restore block on adding VLANs. This will
282 * be removed once new slaves that are not
283 * VLAN challenged will be added.
284 */
285 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
286 }
287
288 res = 0;
289 goto out;
290 }
291 }
292
293 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
294 bond->dev->name);
295
296out:
297 write_unlock_bh(&bond->lock);
298 return res;
299}
300
301/**
302 * bond_has_challenged_slaves
303 * @bond: the bond we're working on
304 *
305 * Searches the slave list. Returns 1 if a vlan challenged slave
306 * was found, 0 otherwise.
307 *
308 * Assumes bond->lock is held.
309 */
310static int bond_has_challenged_slaves(struct bonding *bond)
311{
312 struct slave *slave;
313 int i;
314
315 bond_for_each_slave(bond, slave, i) {
316 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
317 dprintk("found VLAN challenged slave - %s\n",
318 slave->dev->name);
319 return 1;
320 }
321 }
322
323 dprintk("no VLAN challenged slaves found\n");
324 return 0;
325}
326
327/**
328 * bond_next_vlan - safely skip to the next item in the vlans list.
329 * @bond: the bond we're working on
330 * @curr: item we're advancing from
331 *
332 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
333 * or @curr->next otherwise (even if it is @curr itself again).
334 *
335 * Caller must hold bond->lock
336 */
337struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
338{
339 struct vlan_entry *next, *last;
340
341 if (list_empty(&bond->vlan_list)) {
342 return NULL;
343 }
344
345 if (!curr) {
346 next = list_entry(bond->vlan_list.next,
347 struct vlan_entry, vlan_list);
348 } else {
349 last = list_entry(bond->vlan_list.prev,
350 struct vlan_entry, vlan_list);
351 if (last == curr) {
352 next = list_entry(bond->vlan_list.next,
353 struct vlan_entry, vlan_list);
354 } else {
355 next = list_entry(curr->vlan_list.next,
356 struct vlan_entry, vlan_list);
357 }
358 }
359
360 return next;
361}
362
363/**
364 * bond_dev_queue_xmit - Prepare skb for xmit.
365 *
366 * @bond: bond device that got this skb for tx.
367 * @skb: hw accel VLAN tagged skb to transmit
368 * @slave_dev: slave that is supposed to xmit this skbuff
369 *
370 * When the bond gets an skb to transmit that is
371 * already hardware accelerated VLAN tagged, and it
372 * needs to relay this skb to a slave that is not
373 * hw accel capable, the skb needs to be "unaccelerated",
374 * i.e. strip the hwaccel tag and re-insert it as part
375 * of the payload.
376 */
377int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
378{
379 unsigned short vlan_id;
380
381 if (!list_empty(&bond->vlan_list) &&
382 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
383 vlan_get_tag(skb, &vlan_id) == 0) {
384 skb->dev = slave_dev;
385 skb = vlan_put_tag(skb, vlan_id);
386 if (!skb) {
387 /* vlan_put_tag() frees the skb in case of error,
388 * so return success here so the calling functions
389 * won't attempt to free is again.
390 */
391 return 0;
392 }
393 } else {
394 skb->dev = slave_dev;
395 }
396
397 skb->priority = 1;
398 dev_queue_xmit(skb);
399
400 return 0;
401}
402
403/*
404 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
405 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
406 * lock because:
407 * a. This operation is performed in IOCTL context,
408 * b. The operation is protected by the RTNL semaphore in the 8021q code,
409 * c. Holding a lock with BH disabled while directly calling a base driver
410 * entry point is generally a BAD idea.
411 *
412 * The design of synchronization/protection for this operation in the 8021q
413 * module is good for one or more VLAN devices over a single physical device
414 * and cannot be extended for a teaming solution like bonding, so there is a
415 * potential race condition here where a net device from the vlan group might
416 * be referenced (either by a base driver or the 8021q code) while it is being
417 * removed from the system. However, it turns out we're not making matters
418 * worse, and if it works for regular VLAN usage it will work here too.
419*/
420
421/**
422 * bond_vlan_rx_register - Propagates registration to slaves
423 * @bond_dev: bonding net device that got called
424 * @grp: vlan group being registered
425 */
426static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
427{
428 struct bonding *bond = bond_dev->priv;
429 struct slave *slave;
430 int i;
431
432 bond->vlgrp = grp;
433
434 bond_for_each_slave(bond, slave, i) {
435 struct net_device *slave_dev = slave->dev;
436
437 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
438 slave_dev->vlan_rx_register) {
439 slave_dev->vlan_rx_register(slave_dev, grp);
440 }
441 }
442}
443
444/**
445 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
446 * @bond_dev: bonding net device that got called
447 * @vid: vlan id being added
448 */
449static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
450{
451 struct bonding *bond = bond_dev->priv;
452 struct slave *slave;
453 int i, res;
454
455 bond_for_each_slave(bond, slave, i) {
456 struct net_device *slave_dev = slave->dev;
457
458 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
459 slave_dev->vlan_rx_add_vid) {
460 slave_dev->vlan_rx_add_vid(slave_dev, vid);
461 }
462 }
463
464 res = bond_add_vlan(bond, vid);
465 if (res) {
466 printk(KERN_ERR DRV_NAME
4e0952c7 467 ": %s: Error: Failed to add vlan id %d\n",
1da177e4
LT		 468 bond_dev->name, vid);
469 }
470}
471
472/**
473 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
474 * @bond_dev: bonding net device that got called
475 * @vid: vlan id being removed
476 */
477static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
478{
479 struct bonding *bond = bond_dev->priv;
480 struct slave *slave;
481 struct net_device *vlan_dev;
482 int i, res;
483
484 bond_for_each_slave(bond, slave, i) {
485 struct net_device *slave_dev = slave->dev;
486
487 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
488 slave_dev->vlan_rx_kill_vid) {
489 /* Save and then restore vlan_dev in the grp array,
490 * since the slave's driver might clear it.
491 */
5c15bdec 492 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
1da177e4 493 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
5c15bdec 494 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
1da177e4
LT		 495 }
496 }
497
498 res = bond_del_vlan(bond, vid);
499 if (res) {
500 printk(KERN_ERR DRV_NAME
4e0952c7 501 ": %s: Error: Failed to remove vlan id %d\n",
1da177e4
LT		 502 bond_dev->name, vid);
503 }
504}
505
506static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
507{
508 struct vlan_entry *vlan;
509
510 write_lock_bh(&bond->lock);
511
512 if (list_empty(&bond->vlan_list)) {
513 goto out;
514 }
515
516 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
517 slave_dev->vlan_rx_register) {
518 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
519 }
520
521 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
522 !(slave_dev->vlan_rx_add_vid)) {
523 goto out;
524 }
525
526 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
527 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
528 }
529
530out:
531 write_unlock_bh(&bond->lock);
532}
533
534static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
535{
536 struct vlan_entry *vlan;
537 struct net_device *vlan_dev;
538
539 write_lock_bh(&bond->lock);
540
541 if (list_empty(&bond->vlan_list)) {
542 goto out;
543 }
544
545 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
546 !(slave_dev->vlan_rx_kill_vid)) {
547 goto unreg;
548 }
549
550 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
551 /* Save and then restore vlan_dev in the grp array,
552 * since the slave's driver might clear it.
553 */
5c15bdec 554 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
1da177e4 555 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
5c15bdec 556 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
1da177e4
LT		 557 }
558
559unreg:
560 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
561 slave_dev->vlan_rx_register) {
562 slave_dev->vlan_rx_register(slave_dev, NULL);
563 }
564
565out:
566 write_unlock_bh(&bond->lock);
567}
568
569/*------------------------------- Link status -------------------------------*/
570
ff59c456
JV		 571/*
572 * Set the carrier state for the master according to the state of its
573 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
574 * do special 802.3ad magic.
575 *
576 * Returns zero if carrier state does not change, nonzero if it does.
577 */
578static int bond_set_carrier(struct bonding *bond)
579{
580 struct slave *slave;
581 int i;
582
583 if (bond->slave_cnt == 0)
584 goto down;
585
586 if (bond->params.mode == BOND_MODE_8023AD)
587 return bond_3ad_set_carrier(bond);
588
589 bond_for_each_slave(bond, slave, i) {
590 if (slave->link == BOND_LINK_UP) {
591 if (!netif_carrier_ok(bond->dev)) {
592 netif_carrier_on(bond->dev);
593 return 1;
594 }
595 return 0;
596 }
597 }
598
599down:
600 if (netif_carrier_ok(bond->dev)) {
601 netif_carrier_off(bond->dev);
602 return 1;
603 }
604 return 0;
605}
606
1da177e4
LT		 607/*
608 * Get link speed and duplex from the slave's base driver
609 * using ethtool. If for some reason the call fails or the
610 * values are invalid, fake speed and duplex to 100/Full
611 * and return error.
612 */
613static int bond_update_speed_duplex(struct slave *slave)
614{
615 struct net_device *slave_dev = slave->dev;
616 static int (* ioctl)(struct net_device *, struct ifreq *, int);
617 struct ifreq ifr;
618 struct ethtool_cmd etool;
619
620 /* Fake speed and duplex */
621 slave->speed = SPEED_100;
622 slave->duplex = DUPLEX_FULL;
623
624 if (slave_dev->ethtool_ops) {
6a986ce4 625 int res;
1da177e4
LT		 626
627 if (!slave_dev->ethtool_ops->get_settings) {
628 return -1;
629 }
630
631 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
632 if (res < 0) {
633 return -1;
634 }
635
636 goto verify;
637 }
638
639 ioctl = slave_dev->do_ioctl;
640 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
641 etool.cmd = ETHTOOL_GSET;
642 ifr.ifr_data = (char*)&etool;
643 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
644 return -1;
645 }
646
647verify:
648 switch (etool.speed) {
649 case SPEED_10:
650 case SPEED_100:
651 case SPEED_1000:
94dbffd5 652 case SPEED_10000:
1da177e4
LT		 653 break;
654 default:
655 return -1;
656 }
657
658 switch (etool.duplex) {
659 case DUPLEX_FULL:
660 case DUPLEX_HALF:
661 break;
662 default:
663 return -1;
664 }
665
666 slave->speed = etool.speed;
667 slave->duplex = etool.duplex;
668
669 return 0;
670}
671
672/*
673 * if <dev> supports MII link status reporting, check its link status.
674 *
675 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
676 * depening upon the setting of the use_carrier parameter.
677 *
678 * Return either BMSR_LSTATUS, meaning that the link is up (or we
679 * can't tell and just pretend it is), or 0, meaning that the link is
680 * down.
681 *
682 * If reporting is non-zero, instead of faking link up, return -1 if
683 * both ETHTOOL and MII ioctls fail (meaning the device does not
684 * support them). If use_carrier is set, return whatever it says.
685 * It'd be nice if there was a good way to tell if a driver supports
686 * netif_carrier, but there really isn't.
687 */
688static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
689{
690 static int (* ioctl)(struct net_device *, struct ifreq *, int);
691 struct ifreq ifr;
692 struct mii_ioctl_data *mii;
693 struct ethtool_value etool;
694
695 if (bond->params.use_carrier) {
696 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
697 }
698
699 ioctl = slave_dev->do_ioctl;
700 if (ioctl) {
701 /* TODO: set pointer to correct ioctl on a per team member */
702 /* bases to make this more efficient. that is, once */
703 /* we determine the correct ioctl, we will always */
704 /* call it and not the others for that team */
705 /* member. */
706
707 /*
708 * We cannot assume that SIOCGMIIPHY will also read a
709 * register; not all network drivers (e.g., e100)
710 * support that.
711 */
712
713 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
714 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
715 mii = if_mii(&ifr);
716 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
717 mii->reg_num = MII_BMSR;
718 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
719 return (mii->val_out & BMSR_LSTATUS);
720 }
721 }
722 }
723
724 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
725 /* for a period of time so we attempt to get link status */
726 /* from it last if the above MII ioctls fail... */
727 if (slave_dev->ethtool_ops) {
728 if (slave_dev->ethtool_ops->get_link) {
729 u32 link;
730
731 link = slave_dev->ethtool_ops->get_link(slave_dev);
732
733 return link ? BMSR_LSTATUS : 0;
734 }
735 }
736
737 if (ioctl) {
738 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
739 etool.cmd = ETHTOOL_GLINK;
740 ifr.ifr_data = (char*)&etool;
741 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
742 if (etool.data == 1) {
743 return BMSR_LSTATUS;
744 } else {
745 dprintk("SIOCETHTOOL shows link down\n");
746 return 0;
747 }
748 }
749 }
750
751 /*
752 * If reporting, report that either there's no dev->do_ioctl,
753 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
754 * cannot report link status). If not reporting, pretend
755 * we're ok.
756 */
757 return (reporting ? -1 : BMSR_LSTATUS);
758}
759
760/*----------------------------- Multicast list ------------------------------*/
761
762/*
763 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
764 */
765static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
766{
767 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
768 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
769}
770
771/*
772 * returns dmi entry if found, NULL otherwise
773 */
774static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
775{
776 struct dev_mc_list *idmi;
777
778 for (idmi = mc_list; idmi; idmi = idmi->next) {
779 if (bond_is_dmi_same(dmi, idmi)) {
780 return idmi;
781 }
782 }
783
784 return NULL;
785}
786
787/*
788 * Push the promiscuity flag down to appropriate slaves
789 */
790static void bond_set_promiscuity(struct bonding *bond, int inc)
791{
792 if (USES_PRIMARY(bond->params.mode)) {
793 /* write lock already acquired */
794 if (bond->curr_active_slave) {
795 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
796 }
797 } else {
798 struct slave *slave;
799 int i;
800 bond_for_each_slave(bond, slave, i) {
801 dev_set_promiscuity(slave->dev, inc);
802 }
803 }
804}
805
806/*
807 * Push the allmulti flag down to all slaves
808 */
809static void bond_set_allmulti(struct bonding *bond, int inc)
810{
811 if (USES_PRIMARY(bond->params.mode)) {
812 /* write lock already acquired */
813 if (bond->curr_active_slave) {
814 dev_set_allmulti(bond->curr_active_slave->dev, inc);
815 }
816 } else {
817 struct slave *slave;
818 int i;
819 bond_for_each_slave(bond, slave, i) {
820 dev_set_allmulti(slave->dev, inc);
821 }
822 }
823}
824
825/*
826 * Add a Multicast address to slaves
827 * according to mode
828 */
829static void bond_mc_add(struct bonding *bond, void *addr, int alen)
830{
831 if (USES_PRIMARY(bond->params.mode)) {
832 /* write lock already acquired */
833 if (bond->curr_active_slave) {
834 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
835 }
836 } else {
837 struct slave *slave;
838 int i;
839 bond_for_each_slave(bond, slave, i) {
840 dev_mc_add(slave->dev, addr, alen, 0);
841 }
842 }
843}
844
845/*
846 * Remove a multicast address from slave
847 * according to mode
848 */
849static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
850{
851 if (USES_PRIMARY(bond->params.mode)) {
852 /* write lock already acquired */
853 if (bond->curr_active_slave) {
854 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
855 }
856 } else {
857 struct slave *slave;
858 int i;
859 bond_for_each_slave(bond, slave, i) {
860 dev_mc_delete(slave->dev, addr, alen, 0);
861 }
862 }
863}
864
a816c7c7
JV		 865
866/*
867 * Retrieve the list of registered multicast addresses for the bonding
868 * device and retransmit an IGMP JOIN request to the current active
869 * slave.
870 */
871static void bond_resend_igmp_join_requests(struct bonding *bond)
872{
873 struct in_device *in_dev;
874 struct ip_mc_list *im;
875
876 rcu_read_lock();
877 in_dev = __in_dev_get_rcu(bond->dev);
878 if (in_dev) {
879 for (im = in_dev->mc_list; im; im = im->next) {
880 ip_mc_rejoin_group(im);
881 }
882 }
883
884 rcu_read_unlock();
885}
886
1da177e4
LT		 887/*
888 * Totally destroys the mc_list in bond
889 */
890static void bond_mc_list_destroy(struct bonding *bond)
891{
892 struct dev_mc_list *dmi;
893
894 dmi = bond->mc_list;
895 while (dmi) {
896 bond->mc_list = dmi->next;
897 kfree(dmi);
898 dmi = bond->mc_list;
899 }
a816c7c7 900 bond->mc_list = NULL;
1da177e4
LT		 901}
902
903/*
904 * Copy all the Multicast addresses from src to the bonding device dst
905 */
de54f390 906static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 907 gfp_t gfp_flag)
1da177e4
LT		 908{
909 struct dev_mc_list *dmi, *new_dmi;
910
911 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 912 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT		 913
914 if (!new_dmi) {
915 /* FIXME: Potential memory leak !!! */
916 return -ENOMEM;
917 }
918
919 new_dmi->next = bond->mc_list;
920 bond->mc_list = new_dmi;
921 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
922 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
923 new_dmi->dmi_users = dmi->dmi_users;
924 new_dmi->dmi_gusers = dmi->dmi_gusers;
925 }
926
927 return 0;
928}
929
930/*
931 * flush all members of flush->mc_list from device dev->mc_list
932 */
933static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
934{
935 struct bonding *bond = bond_dev->priv;
936 struct dev_mc_list *dmi;
937
938 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
939 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
940 }
941
942 if (bond->params.mode == BOND_MODE_8023AD) {
943 /* del lacpdu mc addr from mc list */
944 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
945
946 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
947 }
948}
949
950/*--------------------------- Active slave change ---------------------------*/
951
952/*
953 * Update the mc list and multicast-related flags for the new and
954 * old active slaves (if any) according to the multicast mode, and
955 * promiscuous flags unconditionally.
956 */
957static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
958{
959 struct dev_mc_list *dmi;
960
961 if (!USES_PRIMARY(bond->params.mode)) {
962 /* nothing to do - mc list is already up-to-date on
963 * all slaves
964 */
965 return;
966 }
967
968 if (old_active) {
969 if (bond->dev->flags & IFF_PROMISC) {
970 dev_set_promiscuity(old_active->dev, -1);
971 }
972
973 if (bond->dev->flags & IFF_ALLMULTI) {
974 dev_set_allmulti(old_active->dev, -1);
975 }
976
977 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
978 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
979 }
980 }
981
982 if (new_active) {
983 if (bond->dev->flags & IFF_PROMISC) {
984 dev_set_promiscuity(new_active->dev, 1);
985 }
986
987 if (bond->dev->flags & IFF_ALLMULTI) {
988 dev_set_allmulti(new_active->dev, 1);
989 }
990
991 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
992 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
993 }
a816c7c7 994 bond_resend_igmp_join_requests(bond);
1da177e4
LT		 995 }
996}
997
998/**
999 * find_best_interface - select the best available slave to be the active one
1000 * @bond: our bonding struct
1001 *
1002 * Warning: Caller must hold curr_slave_lock for writing.
1003 */
1004static struct slave *bond_find_best_slave(struct bonding *bond)
1005{
1006 struct slave *new_active, *old_active;
1007 struct slave *bestslave = NULL;
1008 int mintime = bond->params.updelay;
1009 int i;
1010
1011 new_active = old_active = bond->curr_active_slave;
1012
1013 if (!new_active) { /* there were no active slaves left */
1014 if (bond->slave_cnt > 0) { /* found one slave */
1015 new_active = bond->first_slave;
1016 } else {
1017 return NULL; /* still no slave, return NULL */
1018 }
1019 }
1020
1021 /* first try the primary link; if arping, a link must tx/rx traffic
1022 * before it can be considered the curr_active_slave - also, we would skip
1023 * slaves between the curr_active_slave and primary_slave that may be up
1024 * and able to arp
1025 */
1026 if ((bond->primary_slave) &&
1027 (!bond->params.arp_interval) &&
1028 (IS_UP(bond->primary_slave->dev))) {
1029 new_active = bond->primary_slave;
1030 }
1031
1032 /* remember where to stop iterating over the slaves */
1033 old_active = new_active;
1034
1035 bond_for_each_slave_from(bond, new_active, i, old_active) {
1036 if (IS_UP(new_active->dev)) {
1037 if (new_active->link == BOND_LINK_UP) {
1038 return new_active;
1039 } else if (new_active->link == BOND_LINK_BACK) {
1040 /* link up, but waiting for stabilization */
1041 if (new_active->delay < mintime) {
1042 mintime = new_active->delay;
1043 bestslave = new_active;
1044 }
1045 }
1046 }
1047 }
1048
1049 return bestslave;
1050}
1051
1052/**
1053 * change_active_interface - change the active slave into the specified one
1054 * @bond: our bonding struct
1055 * @new: the new slave to make the active one
1056 *
1057 * Set the new slave to the bond's settings and unset them on the old
1058 * curr_active_slave.
1059 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1060 *
1061 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1062 * because it is apparently the best available slave we have, even though its
1063 * updelay hasn't timed out yet.
1064 *
1065 * Warning: Caller must hold curr_slave_lock for writing.
1066 */
a77b5325 1067void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1068{
1069 struct slave *old_active = bond->curr_active_slave;
1070
1071 if (old_active == new_active) {
1072 return;
1073 }
1074
1075 if (new_active) {
1076 if (new_active->link == BOND_LINK_BACK) {
1077 if (USES_PRIMARY(bond->params.mode)) {
1078 printk(KERN_INFO DRV_NAME
1079 ": %s: making interface %s the new "
1080 "active one %d ms earlier.\n",
1081 bond->dev->name, new_active->dev->name,
1082 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1083 }
1084
1085 new_active->delay = 0;
1086 new_active->link = BOND_LINK_UP;
1087 new_active->jiffies = jiffies;
1088
1089 if (bond->params.mode == BOND_MODE_8023AD) {
1090 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1091 }
1092
1093 if ((bond->params.mode == BOND_MODE_TLB) ||
1094 (bond->params.mode == BOND_MODE_ALB)) {
1095 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1096 }
1097 } else {
1098 if (USES_PRIMARY(bond->params.mode)) {
1099 printk(KERN_INFO DRV_NAME
1100 ": %s: making interface %s the new "
1101 "active one.\n",
1102 bond->dev->name, new_active->dev->name);
1103 }
1104 }
1105 }
1106
1da177e4
LT		 1107 if (USES_PRIMARY(bond->params.mode)) {
1108 bond_mc_swap(bond, new_active, old_active);
1109 }
1110
1111 if ((bond->params.mode == BOND_MODE_TLB) ||
1112 (bond->params.mode == BOND_MODE_ALB)) {
1113 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1114 if (old_active)
1115 bond_set_slave_inactive_flags(old_active);
1116 if (new_active)
1117 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1118 } else {
1119 bond->curr_active_slave = new_active;
1120 }
c3ade5ca
JV		 1121
1122 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1123 if (old_active) {
1124 bond_set_slave_inactive_flags(old_active);
1125 }
1126
1127 if (new_active) {
1128 bond_set_slave_active_flags(new_active);
1129 }
1130 bond_send_gratuitous_arp(bond);
1131 }
1da177e4
LT		 1132}
1133
1134/**
1135 * bond_select_active_slave - select a new active slave, if needed
1136 * @bond: our bonding struct
1137 *
1138 * This functions shoud be called when one of the following occurs:
1139 * - The old curr_active_slave has been released or lost its link.
1140 * - The primary_slave has got its link back.
1141 * - A slave has got its link back and there's no old curr_active_slave.
1142 *
1143 * Warning: Caller must hold curr_slave_lock for writing.
1144 */
a77b5325 1145void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1146{
1147 struct slave *best_slave;
ff59c456 1148 int rv;
1da177e4
LT		 1149
1150 best_slave = bond_find_best_slave(bond);
1151 if (best_slave != bond->curr_active_slave) {
1152 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1153 rv = bond_set_carrier(bond);
1154 if (!rv)
1155 return;
1156
1157 if (netif_carrier_ok(bond->dev)) {
1158 printk(KERN_INFO DRV_NAME
1159 ": %s: first active interface up!\n",
1160 bond->dev->name);
1161 } else {
1162 printk(KERN_INFO DRV_NAME ": %s: "
1163 "now running without any active interface !\n",
1164 bond->dev->name);
1165 }
1da177e4
LT		 1166 }
1167}
1168
1169/*--------------------------- slave list handling ---------------------------*/
1170
1171/*
1172 * This function attaches the slave to the end of list.
1173 *
1174 * bond->lock held for writing by caller.
1175 */
1176static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1177{
1178 if (bond->first_slave == NULL) { /* attaching the first slave */
1179 new_slave->next = new_slave;
1180 new_slave->prev = new_slave;
1181 bond->first_slave = new_slave;
1182 } else {
1183 new_slave->next = bond->first_slave;
1184 new_slave->prev = bond->first_slave->prev;
1185 new_slave->next->prev = new_slave;
1186 new_slave->prev->next = new_slave;
1187 }
1188
1189 bond->slave_cnt++;
1190}
1191
1192/*
1193 * This function detaches the slave from the list.
1194 * WARNING: no check is made to verify if the slave effectively
1195 * belongs to <bond>.
1196 * Nothing is freed on return, structures are just unchained.
1197 * If any slave pointer in bond was pointing to <slave>,
1198 * it should be changed by the calling function.
1199 *
1200 * bond->lock held for writing by caller.
1201 */
1202static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1203{
1204 if (slave->next) {
1205 slave->next->prev = slave->prev;
1206 }
1207
1208 if (slave->prev) {
1209 slave->prev->next = slave->next;
1210 }
1211
1212 if (bond->first_slave == slave) { /* slave is the first slave */
1213 if (bond->slave_cnt > 1) { /* there are more slave */
1214 bond->first_slave = slave->next;
1215 } else {
1216 bond->first_slave = NULL; /* slave was the last one */
1217 }
1218 }
1219
1220 slave->next = NULL;
1221 slave->prev = NULL;
1222 bond->slave_cnt--;
1223}
1224
1225/*---------------------------------- IOCTL ----------------------------------*/
1226
a77b5325 1227int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1228{
1229 dprintk("bond_dev=%p\n", bond_dev);
1230 dprintk("slave_dev=%p\n", slave_dev);
1231 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1232 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1233 return 0;
1234}
1235
8531c5ff 1236#define BOND_INTERSECT_FEATURES \
8648b305 1237 (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_TSO | NETIF_F_UFO)
8531c5ff
AK		 1238
1239/*
8e3babcd
JV		 1240 * Compute the common dev->feature set available to all slaves. Some
1241 * feature bits are managed elsewhere, so preserve feature bits set on
1242 * master device that are not part of the examined set.
8531c5ff
AK		 1243 */
1244static int bond_compute_features(struct bonding *bond)
1245{
8e3babcd 1246 unsigned long features = BOND_INTERSECT_FEATURES;
8531c5ff
AK		 1247 struct slave *slave;
1248 struct net_device *bond_dev = bond->dev;
54ef3137 1249 unsigned short max_hard_header_len = ETH_HLEN;
8e3babcd 1250 int i;
8531c5ff 1251
54ef3137 1252 bond_for_each_slave(bond, slave, i) {
8e3babcd 1253 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
54ef3137
JV		 1254 if (slave->dev->hard_header_len > max_hard_header_len)
1255 max_hard_header_len = slave->dev->hard_header_len;
1256 }
8531c5ff 1257
8531c5ff 1258 if ((features & NETIF_F_SG) &&
8648b305 1259 !(features & NETIF_F_ALL_CSUM))
8531c5ff 1260 features &= ~NETIF_F_SG;
8531c5ff 1261
a0de3adf
JV		 1262 /*
1263 * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1264 * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1265 * implies that all slaves also support scatter-gather
1266 * (NETIF_F_SG), which implies that features also includes
1267 * NETIF_F_SG. So no need to check whether we have an
1268 * illegal combination of NETIF_F_{TSO,UFO} and
1269 * !NETIF_F_SG
1270 */
1271
8e3babcd 1272 features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
8531c5ff 1273 bond_dev->features = features;
54ef3137 1274 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1275
1276 return 0;
1277}
1278
1da177e4 1279/* enslave device <slave> to bond device <master> */
a77b5325 1280int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1281{
1282 struct bonding *bond = bond_dev->priv;
1283 struct slave *new_slave = NULL;
1284 struct dev_mc_list *dmi;
1285 struct sockaddr addr;
1286 int link_reporting;
1287 int old_features = bond_dev->features;
1288 int res = 0;
1289
552709d5 1290 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1291 slave_dev->do_ioctl == NULL) {
1da177e4 1292 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1293 ": %s: Warning: no link monitoring support for %s\n",
1294 bond_dev->name, slave_dev->name);
1da177e4
LT		 1295 }
1296
1297 /* bond must be initialized by bond_open() before enslaving */
1298 if (!(bond_dev->flags & IFF_UP)) {
1299 dprintk("Error, master_dev is not up\n");
1300 return -EPERM;
1301 }
1302
1303 /* already enslaved */
1304 if (slave_dev->flags & IFF_SLAVE) {
1305 dprintk("Error, Device was already enslaved\n");
1306 return -EBUSY;
1307 }
1308
1309 /* vlan challenged mutual exclusion */
1310 /* no need to lock since we're protected by rtnl_lock */
1311 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1312 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1313 if (!list_empty(&bond->vlan_list)) {
1314 printk(KERN_ERR DRV_NAME
4e0952c7 1315 ": %s: Error: cannot enslave VLAN "
1da177e4 1316 "challenged slave %s on VLAN enabled "
4e0952c7 1317 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1318 bond_dev->name);
1319 return -EPERM;
1320 } else {
1321 printk(KERN_WARNING DRV_NAME
4e0952c7 1322 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1323 "slave %s. Adding VLANs will be blocked as "
1324 "long as %s is part of bond %s\n",
4e0952c7 1325 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1326 bond_dev->name);
1327 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1328 }
1329 } else {
1330 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1331 if (bond->slave_cnt == 0) {
1332 /* First slave, and it is not VLAN challenged,
1333 * so remove the block of adding VLANs over the bond.
1334 */
1335 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1336 }
1337 }
1338
217df670
JV		 1339 /*
1340 * Old ifenslave binaries are no longer supported. These can
1341 * be identified with moderate accurary by the state of the slave:
1342 * the current ifenslave will set the interface down prior to
1343 * enslaving it; the old ifenslave will not.
1344 */
1345 if ((slave_dev->flags & IFF_UP)) {
1346 printk(KERN_ERR DRV_NAME ": %s is up. "
1347 "This may be due to an out of date ifenslave.\n",
1348 slave_dev->name);
1349 res = -EPERM;
1350 goto err_undo_flags;
1351 }
1da177e4 1352
217df670
JV		 1353 if (slave_dev->set_mac_address == NULL) {
1354 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 1355 ": %s: Error: The slave device you specified does "
1356 "not support setting the MAC address. "
1357 "Your kernel likely does not support slave "
1358 "devices.\n", bond_dev->name);
1359 res = -EOPNOTSUPP;
217df670 1360 goto err_undo_flags;
1da177e4
LT		 1361 }
1362
243cb4e5 1363 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1364 if (!new_slave) {
1365 res = -ENOMEM;
1366 goto err_undo_flags;
1367 }
1368
1da177e4
LT		 1369 /* save slave's original flags before calling
1370 * netdev_set_master and dev_open
1371 */
1372 new_slave->original_flags = slave_dev->flags;
1373
217df670
JV		 1374 /*
1375 * Save slave's original ("permanent") mac address for modes
1376 * that need it, and for restoring it upon release, and then
1377 * set it to the master's address
1378 */
1379 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1380
217df670
JV		 1381 /*
1382 * Set slave to master's mac address. The application already
1383 * set the master's mac address to that of the first slave
1384 */
1385 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1386 addr.sa_family = slave_dev->type;
1387 res = dev_set_mac_address(slave_dev, &addr);
1388 if (res) {
1389 dprintk("Error %d calling set_mac_address\n", res);
1390 goto err_free;
1391 }
1da177e4 1392
c2edacf8
JV		 1393 res = netdev_set_master(slave_dev, bond_dev);
1394 if (res) {
1395 dprintk("Error %d calling netdev_set_master\n", res);
1396 goto err_close;
1397 }
217df670
JV		 1398 /* open the slave since the application closed it */
1399 res = dev_open(slave_dev);
1400 if (res) {
1401 dprintk("Openning slave %s failed\n", slave_dev->name);
1402 goto err_restore_mac;
1da177e4
LT		 1403 }
1404
1da177e4 1405 new_slave->dev = slave_dev;
0b680e75 1406 slave_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 1407
1408 if ((bond->params.mode == BOND_MODE_TLB) ||
1409 (bond->params.mode == BOND_MODE_ALB)) {
1410 /* bond_alb_init_slave() must be called before all other stages since
1411 * it might fail and we do not want to have to undo everything
1412 */
1413 res = bond_alb_init_slave(bond, new_slave);
1414 if (res) {
1415 goto err_unset_master;
1416 }
1417 }
1418
1419 /* If the mode USES_PRIMARY, then the new slave gets the
1420 * master's promisc (and mc) settings only if it becomes the
1421 * curr_active_slave, and that is taken care of later when calling
1422 * bond_change_active()
1423 */
1424 if (!USES_PRIMARY(bond->params.mode)) {
1425 /* set promiscuity level to new slave */
1426 if (bond_dev->flags & IFF_PROMISC) {
1427 dev_set_promiscuity(slave_dev, 1);
1428 }
1429
1430 /* set allmulti level to new slave */
1431 if (bond_dev->flags & IFF_ALLMULTI) {
1432 dev_set_allmulti(slave_dev, 1);
1433 }
1434
1435 /* upload master's mc_list to new slave */
1436 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1437 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1438 }
1439 }
1440
1441 if (bond->params.mode == BOND_MODE_8023AD) {
1442 /* add lacpdu mc addr to mc list */
1443 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1444
1445 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1446 }
1447
1448 bond_add_vlans_on_slave(bond, slave_dev);
1449
1450 write_lock_bh(&bond->lock);
1451
1452 bond_attach_slave(bond, new_slave);
1453
1454 new_slave->delay = 0;
1455 new_slave->link_failure_count = 0;
1456
8531c5ff
AK		 1457 bond_compute_features(bond);
1458
f5b2b966
JV		 1459 new_slave->last_arp_rx = jiffies;
1460
1da177e4
LT		 1461 if (bond->params.miimon && !bond->params.use_carrier) {
1462 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1463
1464 if ((link_reporting == -1) && !bond->params.arp_interval) {
1465 /*
1466 * miimon is set but a bonded network driver
1467 * does not support ETHTOOL/MII and
1468 * arp_interval is not set. Note: if
1469 * use_carrier is enabled, we will never go
1470 * here (because netif_carrier is always
1471 * supported); thus, we don't need to change
1472 * the messages for netif_carrier.
1473 */
1474 printk(KERN_WARNING DRV_NAME
4e0952c7 1475 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1476 "available for interface %s, and "
1477 "arp_interval/arp_ip_target module parameters "
1478 "not specified, thus bonding will not detect "
1479 "link failures! see bonding.txt for details.\n",
4e0952c7 1480 bond_dev->name, slave_dev->name);
1da177e4
LT		 1481 } else if (link_reporting == -1) {
1482 /* unable get link status using mii/ethtool */
1483 printk(KERN_WARNING DRV_NAME
4e0952c7 1484 ": %s: Warning: can't get link status from "
1da177e4
LT		 1485 "interface %s; the network driver associated "
1486 "with this interface does not support MII or "
1487 "ETHTOOL link status reporting, thus miimon "
1488 "has no effect on this interface.\n",
4e0952c7 1489 bond_dev->name, slave_dev->name);
1da177e4
LT		 1490 }
1491 }
1492
1493 /* check for initial state */
1494 if (!bond->params.miimon ||
1495 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1496 if (bond->params.updelay) {
1497 dprintk("Initial state of slave_dev is "
1498 "BOND_LINK_BACK\n");
1499 new_slave->link = BOND_LINK_BACK;
1500 new_slave->delay = bond->params.updelay;
1501 } else {
1502 dprintk("Initial state of slave_dev is "
1503 "BOND_LINK_UP\n");
1504 new_slave->link = BOND_LINK_UP;
1505 }
1506 new_slave->jiffies = jiffies;
1507 } else {
1508 dprintk("Initial state of slave_dev is "
1509 "BOND_LINK_DOWN\n");
1510 new_slave->link = BOND_LINK_DOWN;
1511 }
1512
1513 if (bond_update_speed_duplex(new_slave) &&
1514 (new_slave->link != BOND_LINK_DOWN)) {
1515 printk(KERN_WARNING DRV_NAME
4e0952c7 1516 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1517 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1518 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1519
1520 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1521 printk(KERN_WARNING DRV_NAME
1522 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1523 "support in base driver for proper aggregator "
4e0952c7 1524 "selection.\n", bond_dev->name);
1da177e4
LT		 1525 }
1526 }
1527
1528 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1529 /* if there is a primary slave, remember it */
1530 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1531 bond->primary_slave = new_slave;
1532 }
1533 }
1534
1535 switch (bond->params.mode) {
1536 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1537 bond_set_slave_inactive_flags(new_slave);
1538 bond_select_active_slave(bond);
1da177e4
LT		 1539 break;
1540 case BOND_MODE_8023AD:
1541 /* in 802.3ad mode, the internal mechanism
1542 * will activate the slaves in the selected
1543 * aggregator
1544 */
1545 bond_set_slave_inactive_flags(new_slave);
1546 /* if this is the first slave */
1547 if (bond->slave_cnt == 1) {
1548 SLAVE_AD_INFO(new_slave).id = 1;
1549 /* Initialize AD with the number of times that the AD timer is called in 1 second
1550 * can be called only after the mac address of the bond is set
1551 */
1552 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1553 bond->params.lacp_fast);
1554 } else {
1555 SLAVE_AD_INFO(new_slave).id =
1556 SLAVE_AD_INFO(new_slave->prev).id + 1;
1557 }
1558
1559 bond_3ad_bind_slave(new_slave);
1560 break;
1561 case BOND_MODE_TLB:
1562 case BOND_MODE_ALB:
1563 new_slave->state = BOND_STATE_ACTIVE;
1564 if ((!bond->curr_active_slave) &&
1565 (new_slave->link != BOND_LINK_DOWN)) {
1566 /* first slave or no active slave yet, and this link
1567 * is OK, so make this interface the active one
1568 */
1569 bond_change_active_slave(bond, new_slave);
8f903c70
JV		 1570 } else {
1571 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1572 }
1573 break;
1574 default:
1575 dprintk("This slave is always active in trunk mode\n");
1576
1577 /* always active in trunk mode */
1578 new_slave->state = BOND_STATE_ACTIVE;
1579
1580 /* In trunking mode there is little meaning to curr_active_slave
1581 * anyway (it holds no special properties of the bond device),
1582 * so we can change it without calling change_active_interface()
1583 */
1584 if (!bond->curr_active_slave) {
1585 bond->curr_active_slave = new_slave;
1586 }
1587 break;
1588 } /* switch(bond_mode) */
1589
ff59c456
JV		 1590 bond_set_carrier(bond);
1591
1da177e4
LT		 1592 write_unlock_bh(&bond->lock);
1593
b76cdba9
MW		 1594 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1595 if (res)
1596 goto err_unset_master;
1597
1da177e4
LT		 1598 printk(KERN_INFO DRV_NAME
1599 ": %s: enslaving %s as a%s interface with a%s link.\n",
1600 bond_dev->name, slave_dev->name,
1601 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1602 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1603
1604 /* enslave is successful */
1605 return 0;
1606
1607/* Undo stages on error */
1608err_unset_master:
1609 netdev_set_master(slave_dev, NULL);
1610
1611err_close:
1612 dev_close(slave_dev);
1613
1614err_restore_mac:
1615 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1616 addr.sa_family = slave_dev->type;
1617 dev_set_mac_address(slave_dev, &addr);
1618
1619err_free:
1620 kfree(new_slave);
1621
1622err_undo_flags:
1623 bond_dev->features = old_features;
8531c5ff 1624
1da177e4
LT		 1625 return res;
1626}
1627
1628/*
1629 * Try to release the slave device <slave> from the bond device <master>
1630 * It is legal to access curr_active_slave without a lock because all the function
1631 * is write-locked.
1632 *
1633 * The rules for slave state should be:
1634 * for Active/Backup:
1635 * Active stays on all backups go down
1636 * for Bonded connections:
1637 * The first up interface should be left on and all others downed.
1638 */
a77b5325 1639int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1640{
1641 struct bonding *bond = bond_dev->priv;
1642 struct slave *slave, *oldcurrent;
1643 struct sockaddr addr;
1644 int mac_addr_differ;
1645
1646 /* slave is not a slave or master is not master of this slave */
1647 if (!(slave_dev->flags & IFF_SLAVE) ||
1648 (slave_dev->master != bond_dev)) {
1649 printk(KERN_ERR DRV_NAME
4e0952c7 1650 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1651 bond_dev->name, slave_dev->name);
1652 return -EINVAL;
1653 }
1654
1655 write_lock_bh(&bond->lock);
1656
1657 slave = bond_get_slave_by_dev(bond, slave_dev);
1658 if (!slave) {
1659 /* not a slave of this bond */
1660 printk(KERN_INFO DRV_NAME
1661 ": %s: %s not enslaved\n",
1662 bond_dev->name, slave_dev->name);
f5e2a7b2 1663 write_unlock_bh(&bond->lock);
1da177e4
LT		 1664 return -EINVAL;
1665 }
1666
1667 mac_addr_differ = memcmp(bond_dev->dev_addr,
1668 slave->perm_hwaddr,
1669 ETH_ALEN);
1670 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1671 printk(KERN_WARNING DRV_NAME
4e0952c7 1672 ": %s: Warning: the permanent HWaddr of %s "
1da177e4
LT		 1673 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1674 "still in use by %s. Set the HWaddr of "
1675 "%s to a different address to avoid "
1676 "conflicts.\n",
4e0952c7 1677 bond_dev->name,
1da177e4
LT		 1678 slave_dev->name,
1679 slave->perm_hwaddr[0],
1680 slave->perm_hwaddr[1],
1681 slave->perm_hwaddr[2],
1682 slave->perm_hwaddr[3],
1683 slave->perm_hwaddr[4],
1684 slave->perm_hwaddr[5],
1685 bond_dev->name,
1686 slave_dev->name);
1687 }
1688
1689 /* Inform AD package of unbinding of slave. */
1690 if (bond->params.mode == BOND_MODE_8023AD) {
1691 /* must be called before the slave is
1692 * detached from the list
1693 */
1694 bond_3ad_unbind_slave(slave);
1695 }
1696
1697 printk(KERN_INFO DRV_NAME
1698 ": %s: releasing %s interface %s\n",
1699 bond_dev->name,
1700 (slave->state == BOND_STATE_ACTIVE)
1701 ? "active" : "backup",
1702 slave_dev->name);
1703
1704 oldcurrent = bond->curr_active_slave;
1705
1706 bond->current_arp_slave = NULL;
1707
1708 /* release the slave from its bond */
1709 bond_detach_slave(bond, slave);
1710
8531c5ff
AK		 1711 bond_compute_features(bond);
1712
1da177e4
LT		 1713 if (bond->primary_slave == slave) {
1714 bond->primary_slave = NULL;
1715 }
1716
1717 if (oldcurrent == slave) {
1718 bond_change_active_slave(bond, NULL);
1719 }
1720
1721 if ((bond->params.mode == BOND_MODE_TLB) ||
1722 (bond->params.mode == BOND_MODE_ALB)) {
1723 /* Must be called only after the slave has been
1724 * detached from the list and the curr_active_slave
1725 * has been cleared (if our_slave == old_current),
1726 * but before a new active slave is selected.
1727 */
1728 bond_alb_deinit_slave(bond, slave);
1729 }
1730
ff59c456 1731 if (oldcurrent == slave)
1da177e4
LT		 1732 bond_select_active_slave(bond);
1733
1da177e4 1734 if (bond->slave_cnt == 0) {
ff59c456
JV		 1735 bond_set_carrier(bond);
1736
1da177e4
LT		 1737 /* if the last slave was removed, zero the mac address
1738 * of the master so it will be set by the application
1739 * to the mac address of the first slave
1740 */
1741 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1742
1743 if (list_empty(&bond->vlan_list)) {
1744 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1745 } else {
1746 printk(KERN_WARNING DRV_NAME
4e0952c7 1747 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1748 "still has VLANs.\n",
4e0952c7 1749 bond_dev->name, bond_dev->name);
1da177e4 1750 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1751 ": %s: When re-adding slaves, make sure the bond's "
1752 "HW address matches its VLANs'.\n",
1753 bond_dev->name);
1da177e4
LT		 1754 }
1755 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1756 !bond_has_challenged_slaves(bond)) {
1757 printk(KERN_INFO DRV_NAME
4e0952c7 1758 ": %s: last VLAN challenged slave %s "
1da177e4 1759 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1760 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1761 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1762 }
1763
1764 write_unlock_bh(&bond->lock);
1765
b76cdba9
MW		 1766 /* must do this from outside any spinlocks */
1767 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1768
1da177e4
LT		 1769 bond_del_vlans_from_slave(bond, slave_dev);
1770
1771 /* If the mode USES_PRIMARY, then we should only remove its
1772 * promisc and mc settings if it was the curr_active_slave, but that was
1773 * already taken care of above when we detached the slave
1774 */
1775 if (!USES_PRIMARY(bond->params.mode)) {
1776 /* unset promiscuity level from slave */
1777 if (bond_dev->flags & IFF_PROMISC) {
1778 dev_set_promiscuity(slave_dev, -1);
1779 }
1780
1781 /* unset allmulti level from slave */
1782 if (bond_dev->flags & IFF_ALLMULTI) {
1783 dev_set_allmulti(slave_dev, -1);
1784 }
1785
1786 /* flush master's mc_list from slave */
1787 bond_mc_list_flush(bond_dev, slave_dev);
1788 }
1789
1790 netdev_set_master(slave_dev, NULL);
1791
1792 /* close slave before restoring its mac address */
1793 dev_close(slave_dev);
1794
217df670
JV		 1795 /* restore original ("permanent") mac address */
1796 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1797 addr.sa_family = slave_dev->type;
1798 dev_set_mac_address(slave_dev, &addr);
1da177e4 1799
8f903c70 1800 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1801 IFF_SLAVE_INACTIVE | IFF_BONDING |
1802 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1803
1804 kfree(slave);
1805
1806 return 0; /* deletion OK */
1807}
1808
1809/*
1810 * This function releases all slaves.
1811 */
1812static int bond_release_all(struct net_device *bond_dev)
1813{
1814 struct bonding *bond = bond_dev->priv;
1815 struct slave *slave;
1816 struct net_device *slave_dev;
1817 struct sockaddr addr;
1818
1819 write_lock_bh(&bond->lock);
1820
ff59c456
JV		 1821 netif_carrier_off(bond_dev);
1822
1da177e4
LT		 1823 if (bond->slave_cnt == 0) {
1824 goto out;
1825 }
1826
1827 bond->current_arp_slave = NULL;
1828 bond->primary_slave = NULL;
1829 bond_change_active_slave(bond, NULL);
1830
1831 while ((slave = bond->first_slave) != NULL) {
1832 /* Inform AD package of unbinding of slave
1833 * before slave is detached from the list.
1834 */
1835 if (bond->params.mode == BOND_MODE_8023AD) {
1836 bond_3ad_unbind_slave(slave);
1837 }
1838
1839 slave_dev = slave->dev;
1840 bond_detach_slave(bond, slave);
1841
1842 if ((bond->params.mode == BOND_MODE_TLB) ||
1843 (bond->params.mode == BOND_MODE_ALB)) {
1844 /* must be called only after the slave
1845 * has been detached from the list
1846 */
1847 bond_alb_deinit_slave(bond, slave);
1848 }
1849
8531c5ff
AK		 1850 bond_compute_features(bond);
1851
1da177e4
LT		 1852 /* now that the slave is detached, unlock and perform
1853 * all the undo steps that should not be called from
1854 * within a lock.
1855 */
1856 write_unlock_bh(&bond->lock);
1857
b76cdba9 1858 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 1859 bond_del_vlans_from_slave(bond, slave_dev);
1860
1861 /* If the mode USES_PRIMARY, then we should only remove its
1862 * promisc and mc settings if it was the curr_active_slave, but that was
1863 * already taken care of above when we detached the slave
1864 */
1865 if (!USES_PRIMARY(bond->params.mode)) {
1866 /* unset promiscuity level from slave */
1867 if (bond_dev->flags & IFF_PROMISC) {
1868 dev_set_promiscuity(slave_dev, -1);
1869 }
1870
1871 /* unset allmulti level from slave */
1872 if (bond_dev->flags & IFF_ALLMULTI) {
1873 dev_set_allmulti(slave_dev, -1);
1874 }
1875
1876 /* flush master's mc_list from slave */
1877 bond_mc_list_flush(bond_dev, slave_dev);
1878 }
1879
1880 netdev_set_master(slave_dev, NULL);
1881
1882 /* close slave before restoring its mac address */
1883 dev_close(slave_dev);
1884
217df670
JV		 1885 /* restore original ("permanent") mac address*/
1886 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1887 addr.sa_family = slave_dev->type;
1888 dev_set_mac_address(slave_dev, &addr);
1da177e4 1889
8f903c70
JV		 1890 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1891 IFF_SLAVE_INACTIVE);
1da177e4
LT		 1892
1893 kfree(slave);
1894
1895 /* re-acquire the lock before getting the next slave */
1896 write_lock_bh(&bond->lock);
1897 }
1898
1899 /* zero the mac address of the master so it will be
1900 * set by the application to the mac address of the
1901 * first slave
1902 */
1903 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1904
1905 if (list_empty(&bond->vlan_list)) {
1906 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1907 } else {
1908 printk(KERN_WARNING DRV_NAME
4e0952c7 1909 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1910 "still has VLANs.\n",
4e0952c7 1911 bond_dev->name, bond_dev->name);
1da177e4 1912 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1913 ": %s: When re-adding slaves, make sure the bond's "
1914 "HW address matches its VLANs'.\n",
1915 bond_dev->name);
1da177e4
LT		 1916 }
1917
1918 printk(KERN_INFO DRV_NAME
1919 ": %s: released all slaves\n",
1920 bond_dev->name);
1921
1922out:
1923 write_unlock_bh(&bond->lock);
1924
1925 return 0;
1926}
1927
1928/*
1929 * This function changes the active slave to slave <slave_dev>.
1930 * It returns -EINVAL in the following cases.
1931 * - <slave_dev> is not found in the list.
1932 * - There is not active slave now.
1933 * - <slave_dev> is already active.
1934 * - The link state of <slave_dev> is not BOND_LINK_UP.
1935 * - <slave_dev> is not running.
1936 * In these cases, this fuction does nothing.
1937 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1938 */
1939static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1940{
1941 struct bonding *bond = bond_dev->priv;
1942 struct slave *old_active = NULL;
1943 struct slave *new_active = NULL;
1944 int res = 0;
1945
1946 if (!USES_PRIMARY(bond->params.mode)) {
1947 return -EINVAL;
1948 }
1949
1950 /* Verify that master_dev is indeed the master of slave_dev */
1951 if (!(slave_dev->flags & IFF_SLAVE) ||
1952 (slave_dev->master != bond_dev)) {
1953 return -EINVAL;
1954 }
1955
1956 write_lock_bh(&bond->lock);
1957
1958 old_active = bond->curr_active_slave;
1959 new_active = bond_get_slave_by_dev(bond, slave_dev);
1960
1961 /*
1962 * Changing to the current active: do nothing; return success.
1963 */
1964 if (new_active && (new_active == old_active)) {
1965 write_unlock_bh(&bond->lock);
1966 return 0;
1967 }
1968
1969 if ((new_active) &&
1970 (old_active) &&
1971 (new_active->link == BOND_LINK_UP) &&
1972 IS_UP(new_active->dev)) {
1973 bond_change_active_slave(bond, new_active);
1974 } else {
1975 res = -EINVAL;
1976 }
1977
1978 write_unlock_bh(&bond->lock);
1979
1980 return res;
1981}
1982
1da177e4
LT		 1983static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1984{
1985 struct bonding *bond = bond_dev->priv;
1986
1987 info->bond_mode = bond->params.mode;
1988 info->miimon = bond->params.miimon;
1989
1990 read_lock_bh(&bond->lock);
1991 info->num_slaves = bond->slave_cnt;
1992 read_unlock_bh(&bond->lock);
1993
1994 return 0;
1995}
1996
1997static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1998{
1999 struct bonding *bond = bond_dev->priv;
2000 struct slave *slave;
2001 int i, found = 0;
2002
2003 if (info->slave_id < 0) {
2004 return -ENODEV;
2005 }
2006
2007 read_lock_bh(&bond->lock);
2008
2009 bond_for_each_slave(bond, slave, i) {
2010 if (i == (int)info->slave_id) {
2011 found = 1;
2012 break;
2013 }
2014 }
2015
2016 read_unlock_bh(&bond->lock);
2017
2018 if (found) {
2019 strcpy(info->slave_name, slave->dev->name);
2020 info->link = slave->link;
2021 info->state = slave->state;
2022 info->link_failure_count = slave->link_failure_count;
2023 } else {
2024 return -ENODEV;
2025 }
2026
2027 return 0;
2028}
2029
2030/*-------------------------------- Monitoring -------------------------------*/
2031
2032/* this function is called regularly to monitor each slave's link. */
a77b5325 2033void bond_mii_monitor(struct net_device *bond_dev)
1da177e4
LT		 2034{
2035 struct bonding *bond = bond_dev->priv;
2036 struct slave *slave, *oldcurrent;
2037 int do_failover = 0;
2038 int delta_in_ticks;
2039 int i;
2040
2041 read_lock(&bond->lock);
2042
2043 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2044
2045 if (bond->kill_timers) {
2046 goto out;
2047 }
2048
2049 if (bond->slave_cnt == 0) {
2050 goto re_arm;
2051 }
2052
2053 /* we will try to read the link status of each of our slaves, and
2054 * set their IFF_RUNNING flag appropriately. For each slave not
2055 * supporting MII status, we won't do anything so that a user-space
2056 * program could monitor the link itself if needed.
2057 */
2058
2059 read_lock(&bond->curr_slave_lock);
2060 oldcurrent = bond->curr_active_slave;
2061 read_unlock(&bond->curr_slave_lock);
2062
2063 bond_for_each_slave(bond, slave, i) {
2064 struct net_device *slave_dev = slave->dev;
2065 int link_state;
2066 u16 old_speed = slave->speed;
2067 u8 old_duplex = slave->duplex;
2068
2069 link_state = bond_check_dev_link(bond, slave_dev, 0);
2070
2071 switch (slave->link) {
2072 case BOND_LINK_UP: /* the link was up */
2073 if (link_state == BMSR_LSTATUS) {
2074 /* link stays up, nothing more to do */
2075 break;
2076 } else { /* link going down */
2077 slave->link = BOND_LINK_FAIL;
2078 slave->delay = bond->params.downdelay;
2079
2080 if (slave->link_failure_count < UINT_MAX) {
2081 slave->link_failure_count++;
2082 }
2083
2084 if (bond->params.downdelay) {
2085 printk(KERN_INFO DRV_NAME
2086 ": %s: link status down for %s "
2087 "interface %s, disabling it in "
2088 "%d ms.\n",
2089 bond_dev->name,
2090 IS_UP(slave_dev)
2091 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2092 ? ((slave == oldcurrent)
2093 ? "active " : "backup ")
2094 : "")
2095 : "idle ",
2096 slave_dev->name,
2097 bond->params.downdelay * bond->params.miimon);
2098 }
2099 }
2100 /* no break ! fall through the BOND_LINK_FAIL test to
2101 ensure proper action to be taken
2102 */
2103 case BOND_LINK_FAIL: /* the link has just gone down */
2104 if (link_state != BMSR_LSTATUS) {
2105 /* link stays down */
2106 if (slave->delay <= 0) {
2107 /* link down for too long time */
2108 slave->link = BOND_LINK_DOWN;
2109
2110 /* in active/backup mode, we must
2111 * completely disable this interface
2112 */
2113 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2114 (bond->params.mode == BOND_MODE_8023AD)) {
2115 bond_set_slave_inactive_flags(slave);
2116 }
2117
2118 printk(KERN_INFO DRV_NAME
2119 ": %s: link status definitely "
2120 "down for interface %s, "
2121 "disabling it\n",
2122 bond_dev->name,
2123 slave_dev->name);
2124
2125 /* notify ad that the link status has changed */
2126 if (bond->params.mode == BOND_MODE_8023AD) {
2127 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2128 }
2129
2130 if ((bond->params.mode == BOND_MODE_TLB) ||
2131 (bond->params.mode == BOND_MODE_ALB)) {
2132 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2133 }
2134
2135 if (slave == oldcurrent) {
2136 do_failover = 1;
2137 }
2138 } else {
2139 slave->delay--;
2140 }
2141 } else {
2142 /* link up again */
2143 slave->link = BOND_LINK_UP;
2144 slave->jiffies = jiffies;
2145 printk(KERN_INFO DRV_NAME
2146 ": %s: link status up again after %d "
2147 "ms for interface %s.\n",
2148 bond_dev->name,
2149 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2150 slave_dev->name);
2151 }
2152 break;
2153 case BOND_LINK_DOWN: /* the link was down */
2154 if (link_state != BMSR_LSTATUS) {
2155 /* the link stays down, nothing more to do */
2156 break;
2157 } else { /* link going up */
2158 slave->link = BOND_LINK_BACK;
2159 slave->delay = bond->params.updelay;
2160
2161 if (bond->params.updelay) {
2162 /* if updelay == 0, no need to
2163 advertise about a 0 ms delay */
2164 printk(KERN_INFO DRV_NAME
2165 ": %s: link status up for "
2166 "interface %s, enabling it "
2167 "in %d ms.\n",
2168 bond_dev->name,
2169 slave_dev->name,
2170 bond->params.updelay * bond->params.miimon);
2171 }
2172 }
2173 /* no break ! fall through the BOND_LINK_BACK state in
2174 case there's something to do.
2175 */
2176 case BOND_LINK_BACK: /* the link has just come back */
2177 if (link_state != BMSR_LSTATUS) {
2178 /* link down again */
2179 slave->link = BOND_LINK_DOWN;
2180
2181 printk(KERN_INFO DRV_NAME
2182 ": %s: link status down again after %d "
2183 "ms for interface %s.\n",
2184 bond_dev->name,
2185 (bond->params.updelay - slave->delay) * bond->params.miimon,
2186 slave_dev->name);
2187 } else {
2188 /* link stays up */
2189 if (slave->delay == 0) {
2190 /* now the link has been up for long time enough */
2191 slave->link = BOND_LINK_UP;
2192 slave->jiffies = jiffies;
2193
2194 if (bond->params.mode == BOND_MODE_8023AD) {
2195 /* prevent it from being the active one */
2196 slave->state = BOND_STATE_BACKUP;
2197 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2198 /* make it immediately active */
2199 slave->state = BOND_STATE_ACTIVE;
2200 } else if (slave != bond->primary_slave) {
2201 /* prevent it from being the active one */
2202 slave->state = BOND_STATE_BACKUP;
2203 }
2204
2205 printk(KERN_INFO DRV_NAME
2206 ": %s: link status definitely "
2207 "up for interface %s.\n",
2208 bond_dev->name,
2209 slave_dev->name);
2210
2211 /* notify ad that the link status has changed */
2212 if (bond->params.mode == BOND_MODE_8023AD) {
2213 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2214 }
2215
2216 if ((bond->params.mode == BOND_MODE_TLB) ||
2217 (bond->params.mode == BOND_MODE_ALB)) {
2218 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2219 }
2220
2221 if ((!oldcurrent) ||
2222 (slave == bond->primary_slave)) {
2223 do_failover = 1;
2224 }
2225 } else {
2226 slave->delay--;
2227 }
2228 }
2229 break;
2230 default:
2231 /* Should not happen */
4e0952c7
MW		 2232 printk(KERN_ERR DRV_NAME
2233 ": %s: Error: %s Illegal value (link=%d)\n",
2234 bond_dev->name,
2235 slave->dev->name,
2236 slave->link);
1da177e4
LT		 2237 goto out;
2238 } /* end of switch (slave->link) */
2239
2240 bond_update_speed_duplex(slave);
2241
2242 if (bond->params.mode == BOND_MODE_8023AD) {
2243 if (old_speed != slave->speed) {
2244 bond_3ad_adapter_speed_changed(slave);
2245 }
2246
2247 if (old_duplex != slave->duplex) {
2248 bond_3ad_adapter_duplex_changed(slave);
2249 }
2250 }
2251
2252 } /* end of for */
2253
2254 if (do_failover) {
2255 write_lock(&bond->curr_slave_lock);
2256
2257 bond_select_active_slave(bond);
2258
1da177e4 2259 write_unlock(&bond->curr_slave_lock);
ff59c456
JV		 2260 } else
2261 bond_set_carrier(bond);
1da177e4
LT		 2262
2263re_arm:
2264 if (bond->params.miimon) {
2265 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2266 }
2267out:
2268 read_unlock(&bond->lock);
2269}
2270
c3ade5ca
JV		 2271
2272static u32 bond_glean_dev_ip(struct net_device *dev)
2273{
2274 struct in_device *idev;
2275 struct in_ifaddr *ifa;
a144ea4b 2276 __be32 addr = 0;
c3ade5ca
JV		 2277
2278 if (!dev)
2279 return 0;
2280
2281 rcu_read_lock();
e5ed6399 2282 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2283 if (!idev)
2284 goto out;
2285
2286 ifa = idev->ifa_list;
2287 if (!ifa)
2288 goto out;
2289
2290 addr = ifa->ifa_local;
2291out:
2292 rcu_read_unlock();
2293 return addr;
2294}
2295
2296static int bond_has_ip(struct bonding *bond)
2297{
2298 struct vlan_entry *vlan, *vlan_next;
2299
2300 if (bond->master_ip)
2301 return 1;
2302
2303 if (list_empty(&bond->vlan_list))
2304 return 0;
2305
2306 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2307 vlan_list) {
2308 if (vlan->vlan_ip)
2309 return 1;
2310 }
2311
2312 return 0;
2313}
2314
f5b2b966
JV		 2315static int bond_has_this_ip(struct bonding *bond, u32 ip)
2316{
2317 struct vlan_entry *vlan, *vlan_next;
2318
2319 if (ip == bond->master_ip)
2320 return 1;
2321
2322 if (list_empty(&bond->vlan_list))
2323 return 0;
2324
2325 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2326 vlan_list) {
2327 if (ip == vlan->vlan_ip)
2328 return 1;
2329 }
2330
2331 return 0;
2332}
2333
c3ade5ca
JV		 2334/*
2335 * We go to the (large) trouble of VLAN tagging ARP frames because
2336 * switches in VLAN mode (especially if ports are configured as
2337 * "native" to a VLAN) might not pass non-tagged frames.
2338 */
2339static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2340{
2341 struct sk_buff *skb;
2342
2343 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2344 slave_dev->name, dest_ip, src_ip, vlan_id);
2345
2346 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2347 NULL, slave_dev->dev_addr, NULL);
2348
2349 if (!skb) {
2350 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2351 return;
2352 }
2353 if (vlan_id) {
2354 skb = vlan_put_tag(skb, vlan_id);
2355 if (!skb) {
2356 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2357 return;
2358 }
2359 }
2360 arp_xmit(skb);
2361}
2362
2363
1da177e4
LT		 2364static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2365{
c3ade5ca 2366 int i, vlan_id, rv;
1da177e4 2367 u32 *targets = bond->params.arp_targets;
c3ade5ca
JV		 2368 struct vlan_entry *vlan, *vlan_next;
2369 struct net_device *vlan_dev;
2370 struct flowi fl;
2371 struct rtable *rt;
1da177e4 2372
6b780567
MW		 2373 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2374 if (!targets[i])
2375 continue;
c3ade5ca
JV		 2376 dprintk("basa: target %x\n", targets[i]);
2377 if (list_empty(&bond->vlan_list)) {
2378 dprintk("basa: empty vlan: arp_send\n");
2379 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2380 bond->master_ip, 0);
2381 continue;
2382 }
2383
2384 /*
2385 * If VLANs are configured, we do a route lookup to
2386 * determine which VLAN interface would be used, so we
2387 * can tag the ARP with the proper VLAN tag.
2388 */
2389 memset(&fl, 0, sizeof(fl));
2390 fl.fl4_dst = targets[i];
2391 fl.fl4_tos = RTO_ONLINK;
2392
2393 rv = ip_route_output_key(&rt, &fl);
2394 if (rv) {
2395 if (net_ratelimit()) {
2396 printk(KERN_WARNING DRV_NAME
2397 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2398 bond->dev->name, NIPQUAD(fl.fl4_dst));
2399 }
2400 continue;
2401 }
2402
2403 /*
2404 * This target is not on a VLAN
2405 */
2406 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2407 ip_rt_put(rt);
c3ade5ca
JV		 2408 dprintk("basa: rtdev == bond->dev: arp_send\n");
2409 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2410 bond->master_ip, 0);
2411 continue;
2412 }
2413
2414 vlan_id = 0;
2415 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2416 vlan_list) {
5c15bdec 2417 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2418 if (vlan_dev == rt->u.dst.dev) {
2419 vlan_id = vlan->vlan_id;
2420 dprintk("basa: vlan match on %s %d\n",
2421 vlan_dev->name, vlan_id);
2422 break;
2423 }
2424 }
2425
2426 if (vlan_id) {
ed4b9f80 2427 ip_rt_put(rt);
c3ade5ca
JV		 2428 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2429 vlan->vlan_ip, vlan_id);
2430 continue;
2431 }
2432
2433 if (net_ratelimit()) {
2434 printk(KERN_WARNING DRV_NAME
2435 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2436 bond->dev->name, NIPQUAD(fl.fl4_dst),
2437 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2438 }
ed4b9f80 2439 ip_rt_put(rt);
c3ade5ca
JV		 2440 }
2441}
2442
2443/*
2444 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2445 * for each VLAN above us.
2446 */
2447static void bond_send_gratuitous_arp(struct bonding *bond)
2448{
2449 struct slave *slave = bond->curr_active_slave;
2450 struct vlan_entry *vlan;
2451 struct net_device *vlan_dev;
2452
2453 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2454 slave ? slave->dev->name : "NULL");
2455 if (!slave)
2456 return;
2457
2458 if (bond->master_ip) {
2459 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2460 bond->master_ip, 0);
2461 }
2462
2463 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2464 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2465 if (vlan->vlan_ip) {
2466 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2467 vlan->vlan_ip, vlan->vlan_id);
2468 }
1da177e4
LT		 2469 }
2470}
2471
f5b2b966
JV		 2472static void bond_validate_arp(struct bonding *bond, struct slave *slave, u32 sip, u32 tip)
2473{
2474 int i;
2475 u32 *targets = bond->params.arp_targets;
2476
2477 targets = bond->params.arp_targets;
2478 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2479 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2480 "%u.%u.%u.%u bhti(tip) %d\n",
2481 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2482 bond_has_this_ip(bond, tip));
2483 if (sip == targets[i]) {
2484 if (bond_has_this_ip(bond, tip))
2485 slave->last_arp_rx = jiffies;
2486 return;
2487 }
2488 }
2489}
2490
2491static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2492{
2493 struct arphdr *arp;
2494 struct slave *slave;
2495 struct bonding *bond;
2496 unsigned char *arp_ptr;
2497 u32 sip, tip;
2498
2499 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2500 goto out;
2501
2502 bond = dev->priv;
2503 read_lock(&bond->lock);
2504
2505 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2506 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2507 orig_dev ? orig_dev->name : "NULL");
2508
2509 slave = bond_get_slave_by_dev(bond, orig_dev);
2510 if (!slave || !slave_do_arp_validate(bond, slave))
2511 goto out_unlock;
2512
2513 /* ARP header, plus 2 device addresses, plus 2 IP addresses. */
2514 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2515 (2 * dev->addr_len) +
2516 (2 * sizeof(u32)))))
2517 goto out_unlock;
2518
d0a92be0 2519 arp = arp_hdr(skb);
f5b2b966
JV		 2520 if (arp->ar_hln != dev->addr_len ||
2521 skb->pkt_type == PACKET_OTHERHOST ||
2522 skb->pkt_type == PACKET_LOOPBACK ||
2523 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2524 arp->ar_pro != htons(ETH_P_IP) ||
2525 arp->ar_pln != 4)
2526 goto out_unlock;
2527
2528 arp_ptr = (unsigned char *)(arp + 1);
2529 arp_ptr += dev->addr_len;
2530 memcpy(&sip, arp_ptr, 4);
2531 arp_ptr += 4 + dev->addr_len;
2532 memcpy(&tip, arp_ptr, 4);
2533
2534 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2535 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2536 slave->state, bond->params.arp_validate,
2537 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2538
2539 /*
2540 * Backup slaves won't see the ARP reply, but do come through
2541 * here for each ARP probe (so we swap the sip/tip to validate
2542 * the probe). In a "redundant switch, common router" type of
2543 * configuration, the ARP probe will (hopefully) travel from
2544 * the active, through one switch, the router, then the other
2545 * switch before reaching the backup.
2546 */
2547 if (slave->state == BOND_STATE_ACTIVE)
2548 bond_validate_arp(bond, slave, sip, tip);
2549 else
2550 bond_validate_arp(bond, slave, tip, sip);
2551
2552out_unlock:
2553 read_unlock(&bond->lock);
2554out:
2555 dev_kfree_skb(skb);
2556 return NET_RX_SUCCESS;
2557}
2558
1da177e4
LT		 2559/*
2560 * this function is called regularly to monitor each slave's link
2561 * ensuring that traffic is being sent and received when arp monitoring
2562 * is used in load-balancing mode. if the adapter has been dormant, then an
2563 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2564 * arp monitoring in active backup mode.
2565 */
a77b5325 2566void bond_loadbalance_arp_mon(struct net_device *bond_dev)
1da177e4
LT		 2567{
2568 struct bonding *bond = bond_dev->priv;
2569 struct slave *slave, *oldcurrent;
2570 int do_failover = 0;
2571 int delta_in_ticks;
2572 int i;
2573
2574 read_lock(&bond->lock);
2575
2576 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2577
2578 if (bond->kill_timers) {
2579 goto out;
2580 }
2581
2582 if (bond->slave_cnt == 0) {
2583 goto re_arm;
2584 }
2585
2586 read_lock(&bond->curr_slave_lock);
2587 oldcurrent = bond->curr_active_slave;
2588 read_unlock(&bond->curr_slave_lock);
2589
2590 /* see if any of the previous devices are up now (i.e. they have
2591 * xmt and rcv traffic). the curr_active_slave does not come into
2592 * the picture unless it is null. also, slave->jiffies is not needed
2593 * here because we send an arp on each slave and give a slave as
2594 * long as it needs to get the tx/rx within the delta.
2595 * TODO: what about up/down delay in arp mode? it wasn't here before
2596 * so it can wait
2597 */
2598 bond_for_each_slave(bond, slave, i) {
2599 if (slave->link != BOND_LINK_UP) {
2600 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2601 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2602
2603 slave->link = BOND_LINK_UP;
2604 slave->state = BOND_STATE_ACTIVE;
2605
2606 /* primary_slave has no meaning in round-robin
2607 * mode. the window of a slave being up and
2608 * curr_active_slave being null after enslaving
2609 * is closed.
2610 */
2611 if (!oldcurrent) {
2612 printk(KERN_INFO DRV_NAME
2613 ": %s: link status definitely "
2614 "up for interface %s, ",
2615 bond_dev->name,
2616 slave->dev->name);
2617 do_failover = 1;
2618 } else {
2619 printk(KERN_INFO DRV_NAME
2620 ": %s: interface %s is now up\n",
2621 bond_dev->name,
2622 slave->dev->name);
2623 }
2624 }
2625 } else {
2626 /* slave->link == BOND_LINK_UP */
2627
2628 /* not all switches will respond to an arp request
2629 * when the source ip is 0, so don't take the link down
2630 * if we don't know our ip yet
2631 */
2632 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2633 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
c3ade5ca 2634 bond_has_ip(bond))) {
1da177e4
LT		 2635
2636 slave->link = BOND_LINK_DOWN;
2637 slave->state = BOND_STATE_BACKUP;
2638
2639 if (slave->link_failure_count < UINT_MAX) {
2640 slave->link_failure_count++;
2641 }
2642
2643 printk(KERN_INFO DRV_NAME
2644 ": %s: interface %s is now down.\n",
2645 bond_dev->name,
2646 slave->dev->name);
2647
2648 if (slave == oldcurrent) {
2649 do_failover = 1;
2650 }
2651 }
2652 }
2653
2654 /* note: if switch is in round-robin mode, all links
2655 * must tx arp to ensure all links rx an arp - otherwise
2656 * links may oscillate or not come up at all; if switch is
2657 * in something like xor mode, there is nothing we can
2658 * do - all replies will be rx'ed on same link causing slaves
2659 * to be unstable during low/no traffic periods
2660 */
2661 if (IS_UP(slave->dev)) {
2662 bond_arp_send_all(bond, slave);
2663 }
2664 }
2665
2666 if (do_failover) {
2667 write_lock(&bond->curr_slave_lock);
2668
2669 bond_select_active_slave(bond);
2670
1da177e4
LT		 2671 write_unlock(&bond->curr_slave_lock);
2672 }
2673
2674re_arm:
2675 if (bond->params.arp_interval) {
2676 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2677 }
2678out:
2679 read_unlock(&bond->lock);
2680}
2681
2682/*
2683 * When using arp monitoring in active-backup mode, this function is
2684 * called to determine if any backup slaves have went down or a new
2685 * current slave needs to be found.
2686 * The backup slaves never generate traffic, they are considered up by merely
2687 * receiving traffic. If the current slave goes down, each backup slave will
2688 * be given the opportunity to tx/rx an arp before being taken down - this
2689 * prevents all slaves from being taken down due to the current slave not
2690 * sending any traffic for the backups to receive. The arps are not necessarily
2691 * necessary, any tx and rx traffic will keep the current slave up. While any
2692 * rx traffic will keep the backup slaves up, the current slave is responsible
2693 * for generating traffic to keep them up regardless of any other traffic they
2694 * may have received.
2695 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2696 */
a77b5325 2697void bond_activebackup_arp_mon(struct net_device *bond_dev)
1da177e4
LT		 2698{
2699 struct bonding *bond = bond_dev->priv;
2700 struct slave *slave;
2701 int delta_in_ticks;
2702 int i;
2703
2704 read_lock(&bond->lock);
2705
2706 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2707
2708 if (bond->kill_timers) {
2709 goto out;
2710 }
2711
2712 if (bond->slave_cnt == 0) {
2713 goto re_arm;
2714 }
2715
2716 /* determine if any slave has come up or any backup slave has
2717 * gone down
2718 * TODO: what about up/down delay in arp mode? it wasn't here before
2719 * so it can wait
2720 */
2721 bond_for_each_slave(bond, slave, i) {
2722 if (slave->link != BOND_LINK_UP) {
f5b2b966
JV		 2723 if ((jiffies - slave_last_rx(bond, slave)) <=
2724 delta_in_ticks) {
1da177e4
LT		 2725
2726 slave->link = BOND_LINK_UP;
2727
2728 write_lock(&bond->curr_slave_lock);
2729
2730 if ((!bond->curr_active_slave) &&
2731 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2732 bond_change_active_slave(bond, slave);
2733 bond->current_arp_slave = NULL;
2734 } else if (bond->curr_active_slave != slave) {
2735 /* this slave has just come up but we
2736 * already have a current slave; this
2737 * can also happen if bond_enslave adds
2738 * a new slave that is up while we are
2739 * searching for a new slave
2740 */
2741 bond_set_slave_inactive_flags(slave);
2742 bond->current_arp_slave = NULL;
2743 }
2744
ff59c456
JV		 2745 bond_set_carrier(bond);
2746
1da177e4
LT		 2747 if (slave == bond->curr_active_slave) {
2748 printk(KERN_INFO DRV_NAME
2749 ": %s: %s is up and now the "
2750 "active interface\n",
2751 bond_dev->name,
2752 slave->dev->name);
ff59c456 2753 netif_carrier_on(bond->dev);
1da177e4
LT		 2754 } else {
2755 printk(KERN_INFO DRV_NAME
2756 ": %s: backup interface %s is "
2757 "now up\n",
2758 bond_dev->name,
2759 slave->dev->name);
2760 }
2761
2762 write_unlock(&bond->curr_slave_lock);
2763 }
2764 } else {
2765 read_lock(&bond->curr_slave_lock);
2766
2767 if ((slave != bond->curr_active_slave) &&
2768 (!bond->current_arp_slave) &&
f5b2b966 2769 (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
c3ade5ca 2770 bond_has_ip(bond))) {
1da177e4
LT		 2771 /* a backup slave has gone down; three times
2772 * the delta allows the current slave to be
2773 * taken out before the backup slave.
2774 * note: a non-null current_arp_slave indicates
2775 * the curr_active_slave went down and we are
2776 * searching for a new one; under this
2777 * condition we only take the curr_active_slave
2778 * down - this gives each slave a chance to
2779 * tx/rx traffic before being taken out
2780 */
2781
2782 read_unlock(&bond->curr_slave_lock);
2783
2784 slave->link = BOND_LINK_DOWN;
2785
2786 if (slave->link_failure_count < UINT_MAX) {
2787 slave->link_failure_count++;
2788 }
2789
2790 bond_set_slave_inactive_flags(slave);
2791
2792 printk(KERN_INFO DRV_NAME
2793 ": %s: backup interface %s is now down\n",
2794 bond_dev->name,
2795 slave->dev->name);
2796 } else {
2797 read_unlock(&bond->curr_slave_lock);
2798 }
2799 }
2800 }
2801
2802 read_lock(&bond->curr_slave_lock);
2803 slave = bond->curr_active_slave;
2804 read_unlock(&bond->curr_slave_lock);
2805
2806 if (slave) {
2807 /* if we have sent traffic in the past 2*arp_intervals but
2808 * haven't xmit and rx traffic in that time interval, select
2809 * a different slave. slave->jiffies is only updated when
2810 * a slave first becomes the curr_active_slave - not necessarily
2811 * after every arp; this ensures the slave has a full 2*delta
2812 * before being taken out. if a primary is being used, check
2813 * if it is up and needs to take over as the curr_active_slave
2814 */
2815 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
f5b2b966 2816 (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
c3ade5ca 2817 bond_has_ip(bond))) &&
1da177e4
LT		 2818 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2819
2820 slave->link = BOND_LINK_DOWN;
2821
2822 if (slave->link_failure_count < UINT_MAX) {
2823 slave->link_failure_count++;
2824 }
2825
2826 printk(KERN_INFO DRV_NAME
2827 ": %s: link status down for active interface "
2828 "%s, disabling it\n",
2829 bond_dev->name,
2830 slave->dev->name);
2831
2832 write_lock(&bond->curr_slave_lock);
2833
2834 bond_select_active_slave(bond);
2835 slave = bond->curr_active_slave;
2836
2837 write_unlock(&bond->curr_slave_lock);
2838
2839 bond->current_arp_slave = slave;
2840
2841 if (slave) {
2842 slave->jiffies = jiffies;
2843 }
2844 } else if ((bond->primary_slave) &&
2845 (bond->primary_slave != slave) &&
2846 (bond->primary_slave->link == BOND_LINK_UP)) {
2847 /* at this point, slave is the curr_active_slave */
2848 printk(KERN_INFO DRV_NAME
2849 ": %s: changing from interface %s to primary "
2850 "interface %s\n",
2851 bond_dev->name,
2852 slave->dev->name,
2853 bond->primary_slave->dev->name);
2854
2855 /* primary is up so switch to it */
2856 write_lock(&bond->curr_slave_lock);
2857 bond_change_active_slave(bond, bond->primary_slave);
2858 write_unlock(&bond->curr_slave_lock);
2859
2860 slave = bond->primary_slave;
2861 slave->jiffies = jiffies;
2862 } else {
2863 bond->current_arp_slave = NULL;
2864 }
2865
2866 /* the current slave must tx an arp to ensure backup slaves
2867 * rx traffic
2868 */
c3ade5ca 2869 if (slave && bond_has_ip(bond)) {
1da177e4
LT		 2870 bond_arp_send_all(bond, slave);
2871 }
2872 }
2873
2874 /* if we don't have a curr_active_slave, search for the next available
2875 * backup slave from the current_arp_slave and make it the candidate
2876 * for becoming the curr_active_slave
2877 */
2878 if (!slave) {
2879 if (!bond->current_arp_slave) {
2880 bond->current_arp_slave = bond->first_slave;
2881 }
2882
2883 if (bond->current_arp_slave) {
2884 bond_set_slave_inactive_flags(bond->current_arp_slave);
2885
2886 /* search for next candidate */
2f872f04 2887 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
1da177e4
LT		 2888 if (IS_UP(slave->dev)) {
2889 slave->link = BOND_LINK_BACK;
2890 bond_set_slave_active_flags(slave);
2891 bond_arp_send_all(bond, slave);
2892 slave->jiffies = jiffies;
2893 bond->current_arp_slave = slave;
2894 break;
2895 }
2896
2897 /* if the link state is up at this point, we
2898 * mark it down - this can happen if we have
2899 * simultaneous link failures and
2900 * reselect_active_interface doesn't make this
2901 * one the current slave so it is still marked
2902 * up when it is actually down
2903 */
2904 if (slave->link == BOND_LINK_UP) {
2905 slave->link = BOND_LINK_DOWN;
2906 if (slave->link_failure_count < UINT_MAX) {
2907 slave->link_failure_count++;
2908 }
2909
2910 bond_set_slave_inactive_flags(slave);
2911
2912 printk(KERN_INFO DRV_NAME
2913 ": %s: backup interface %s is "
2914 "now down.\n",
2915 bond_dev->name,
2916 slave->dev->name);
2917 }
2918 }
2919 }
2920 }
2921
2922re_arm:
2923 if (bond->params.arp_interval) {
2924 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2925 }
2926out:
2927 read_unlock(&bond->lock);
2928}
2929
2930/*------------------------------ proc/seq_file-------------------------------*/
2931
2932#ifdef CONFIG_PROC_FS
2933
2934#define SEQ_START_TOKEN ((void *)1)
2935
2936static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2937{
2938 struct bonding *bond = seq->private;
2939 loff_t off = 0;
2940 struct slave *slave;
2941 int i;
2942
2943 /* make sure the bond won't be taken away */
2944 read_lock(&dev_base_lock);
2945 read_lock_bh(&bond->lock);
2946
2947 if (*pos == 0) {
2948 return SEQ_START_TOKEN;
2949 }
2950
2951 bond_for_each_slave(bond, slave, i) {
2952 if (++off == *pos) {
2953 return slave;
2954 }
2955 }
2956
2957 return NULL;
2958}
2959
2960static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2961{
2962 struct bonding *bond = seq->private;
2963 struct slave *slave = v;
2964
2965 ++*pos;
2966 if (v == SEQ_START_TOKEN) {
2967 return bond->first_slave;
2968 }
2969
2970 slave = slave->next;
2971
2972 return (slave == bond->first_slave) ? NULL : slave;
2973}
2974
2975static void bond_info_seq_stop(struct seq_file *seq, void *v)
2976{
2977 struct bonding *bond = seq->private;
2978
2979 read_unlock_bh(&bond->lock);
2980 read_unlock(&dev_base_lock);
2981}
2982
2983static void bond_info_show_master(struct seq_file *seq)
2984{
2985 struct bonding *bond = seq->private;
2986 struct slave *curr;
4756b02f
MW		 2987 int i;
2988 u32 target;
1da177e4
LT		 2989
2990 read_lock(&bond->curr_slave_lock);
2991 curr = bond->curr_active_slave;
2992 read_unlock(&bond->curr_slave_lock);
2993
2994 seq_printf(seq, "Bonding Mode: %s\n",
2995 bond_mode_name(bond->params.mode));
2996
c61b75ad
MW		 2997 if (bond->params.mode == BOND_MODE_XOR ||
2998 bond->params.mode == BOND_MODE_8023AD) {
2999 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3000 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3001 bond->params.xmit_policy);
3002 }
3003
1da177e4
LT		 3004 if (USES_PRIMARY(bond->params.mode)) {
3005 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3006 (bond->primary_slave) ?
3007 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3008
3009 seq_printf(seq, "Currently Active Slave: %s\n",
3010 (curr) ? curr->dev->name : "None");
3011 }
3012
ff59c456
JV		 3013 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3014 "up" : "down");
1da177e4
LT		 3015 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3016 seq_printf(seq, "Up Delay (ms): %d\n",
3017 bond->params.updelay * bond->params.miimon);
3018 seq_printf(seq, "Down Delay (ms): %d\n",
3019 bond->params.downdelay * bond->params.miimon);
3020
4756b02f
MW		 3021
3022 /* ARP information */
3023 if(bond->params.arp_interval > 0) {
3024 int printed=0;
3025 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3026 bond->params.arp_interval);
3027
3028 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3029
3030 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3031 if (!bond->params.arp_targets[i])
3032 continue;
3033 if (printed)
3034 seq_printf(seq, ",");
3035 target = ntohl(bond->params.arp_targets[i]);
3036 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3037 printed = 1;
3038 }
3039 seq_printf(seq, "\n");
3040 }
3041
1da177e4
LT		 3042 if (bond->params.mode == BOND_MODE_8023AD) {
3043 struct ad_info ad_info;
3044
3045 seq_puts(seq, "\n802.3ad info\n");
3046 seq_printf(seq, "LACP rate: %s\n",
3047 (bond->params.lacp_fast) ? "fast" : "slow");
3048
3049 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3050 seq_printf(seq, "bond %s has no active aggregator\n",
3051 bond->dev->name);
3052 } else {
3053 seq_printf(seq, "Active Aggregator Info:\n");
3054
3055 seq_printf(seq, "\tAggregator ID: %d\n",
3056 ad_info.aggregator_id);
3057 seq_printf(seq, "\tNumber of ports: %d\n",
3058 ad_info.ports);
3059 seq_printf(seq, "\tActor Key: %d\n",
3060 ad_info.actor_key);
3061 seq_printf(seq, "\tPartner Key: %d\n",
3062 ad_info.partner_key);
3063 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3064 ad_info.partner_system[0],
3065 ad_info.partner_system[1],
3066 ad_info.partner_system[2],
3067 ad_info.partner_system[3],
3068 ad_info.partner_system[4],
3069 ad_info.partner_system[5]);
3070 }
3071 }
3072}
3073
3074static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3075{
3076 struct bonding *bond = seq->private;
3077
3078 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3079 seq_printf(seq, "MII Status: %s\n",
3080 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3081 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3082 slave->link_failure_count);
3083
217df670
JV		 3084 seq_printf(seq,
3085 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3086 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3087 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3088 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
1da177e4
LT		 3089
3090 if (bond->params.mode == BOND_MODE_8023AD) {
3091 const struct aggregator *agg
3092 = SLAVE_AD_INFO(slave).port.aggregator;
3093
3094 if (agg) {
3095 seq_printf(seq, "Aggregator ID: %d\n",
3096 agg->aggregator_identifier);
3097 } else {
3098 seq_puts(seq, "Aggregator ID: N/A\n");
3099 }
3100 }
3101}
3102
3103static int bond_info_seq_show(struct seq_file *seq, void *v)
3104{
3105 if (v == SEQ_START_TOKEN) {
3106 seq_printf(seq, "%s\n", version);
3107 bond_info_show_master(seq);
3108 } else {
3109 bond_info_show_slave(seq, v);
3110 }
3111
3112 return 0;
3113}
3114
3115static struct seq_operations bond_info_seq_ops = {
3116 .start = bond_info_seq_start,
3117 .next = bond_info_seq_next,
3118 .stop = bond_info_seq_stop,
3119 .show = bond_info_seq_show,
3120};
3121
3122static int bond_info_open(struct inode *inode, struct file *file)
3123{
3124 struct seq_file *seq;
3125 struct proc_dir_entry *proc;
3126 int res;
3127
3128 res = seq_open(file, &bond_info_seq_ops);
3129 if (!res) {
3130 /* recover the pointer buried in proc_dir_entry data */
3131 seq = file->private_data;
3132 proc = PDE(inode);
3133 seq->private = proc->data;
3134 }
3135
3136 return res;
3137}
3138
d54b1fdb 3139static const struct file_operations bond_info_fops = {
1da177e4
LT		 3140 .owner = THIS_MODULE,
3141 .open = bond_info_open,
3142 .read = seq_read,
3143 .llseek = seq_lseek,
3144 .release = seq_release,
3145};
3146
3147static int bond_create_proc_entry(struct bonding *bond)
3148{
3149 struct net_device *bond_dev = bond->dev;
3150
3151 if (bond_proc_dir) {
3152 bond->proc_entry = create_proc_entry(bond_dev->name,
3153 S_IRUGO,
3154 bond_proc_dir);
3155 if (bond->proc_entry == NULL) {
3156 printk(KERN_WARNING DRV_NAME
3157 ": Warning: Cannot create /proc/net/%s/%s\n",
3158 DRV_NAME, bond_dev->name);
3159 } else {
3160 bond->proc_entry->data = bond;
3161 bond->proc_entry->proc_fops = &bond_info_fops;
3162 bond->proc_entry->owner = THIS_MODULE;
3163 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3164 }
3165 }
3166
3167 return 0;
3168}
3169
3170static void bond_remove_proc_entry(struct bonding *bond)
3171{
3172 if (bond_proc_dir && bond->proc_entry) {
3173 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3174 memset(bond->proc_file_name, 0, IFNAMSIZ);
3175 bond->proc_entry = NULL;
3176 }
3177}
3178
3179/* Create the bonding directory under /proc/net, if doesn't exist yet.
3180 * Caller must hold rtnl_lock.
3181 */
3182static void bond_create_proc_dir(void)
3183{
3184 int len = strlen(DRV_NAME);
3185
3186 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3187 bond_proc_dir = bond_proc_dir->next) {
3188 if ((bond_proc_dir->namelen == len) &&
3189 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3190 break;
3191 }
3192 }
3193
3194 if (!bond_proc_dir) {
3195 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3196 if (bond_proc_dir) {
3197 bond_proc_dir->owner = THIS_MODULE;
3198 } else {
3199 printk(KERN_WARNING DRV_NAME
3200 ": Warning: cannot create /proc/net/%s\n",
3201 DRV_NAME);
3202 }
3203 }
3204}
3205
3206/* Destroy the bonding directory under /proc/net, if empty.
3207 * Caller must hold rtnl_lock.
3208 */
3209static void bond_destroy_proc_dir(void)
3210{
3211 struct proc_dir_entry *de;
3212
3213 if (!bond_proc_dir) {
3214 return;
3215 }
3216
3217 /* verify that the /proc dir is empty */
3218 for (de = bond_proc_dir->subdir; de; de = de->next) {
3219 /* ignore . and .. */
3220 if (*(de->name) != '.') {
3221 break;
3222 }
3223 }
3224
3225 if (de) {
3226 if (bond_proc_dir->owner == THIS_MODULE) {
3227 bond_proc_dir->owner = NULL;
3228 }
3229 } else {
3230 remove_proc_entry(DRV_NAME, proc_net);
3231 bond_proc_dir = NULL;
3232 }
3233}
3234#endif /* CONFIG_PROC_FS */
3235
3236/*-------------------------- netdev event handling --------------------------*/
3237
3238/*
3239 * Change device name
3240 */
3241static int bond_event_changename(struct bonding *bond)
3242{
3243#ifdef CONFIG_PROC_FS
3244 bond_remove_proc_entry(bond);
3245 bond_create_proc_entry(bond);
3246#endif
b76cdba9
MW		 3247 down_write(&(bonding_rwsem));
3248 bond_destroy_sysfs_entry(bond);
3249 bond_create_sysfs_entry(bond);
3250 up_write(&(bonding_rwsem));
1da177e4
LT		 3251 return NOTIFY_DONE;
3252}
3253
3254static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3255{
3256 struct bonding *event_bond = bond_dev->priv;
3257
3258 switch (event) {
3259 case NETDEV_CHANGENAME:
3260 return bond_event_changename(event_bond);
3261 case NETDEV_UNREGISTER:
3262 /*
3263 * TODO: remove a bond from the list?
3264 */
3265 break;
3266 default:
3267 break;
3268 }
3269
3270 return NOTIFY_DONE;
3271}
3272
3273static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3274{
3275 struct net_device *bond_dev = slave_dev->master;
8531c5ff 3276 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3277
3278 switch (event) {
3279 case NETDEV_UNREGISTER:
3280 if (bond_dev) {
3281 bond_release(bond_dev, slave_dev);
3282 }
3283 break;
3284 case NETDEV_CHANGE:
3285 /*
3286 * TODO: is this what we get if somebody
3287 * sets up a hierarchical bond, then rmmod's
3288 * one of the slave bonding devices?
3289 */
3290 break;
3291 case NETDEV_DOWN:
3292 /*
3293 * ... Or is it this?
3294 */
3295 break;
3296 case NETDEV_CHANGEMTU:
3297 /*
3298 * TODO: Should slaves be allowed to
3299 * independently alter their MTU? For
3300 * an active-backup bond, slaves need
3301 * not be the same type of device, so
3302 * MTUs may vary. For other modes,
3303 * slaves arguably should have the
3304 * same MTUs. To do this, we'd need to
3305 * take over the slave's change_mtu
3306 * function for the duration of their
3307 * servitude.
3308 */
3309 break;
3310 case NETDEV_CHANGENAME:
3311 /*
3312 * TODO: handle changing the primary's name
3313 */
3314 break;
8531c5ff
AK		 3315 case NETDEV_FEAT_CHANGE:
3316 bond_compute_features(bond);
3317 break;
1da177e4
LT		 3318 default:
3319 break;
3320 }
3321
3322 return NOTIFY_DONE;
3323}
3324
3325/*
3326 * bond_netdev_event: handle netdev notifier chain events.
3327 *
3328 * This function receives events for the netdev chain. The caller (an
e041c683 3329 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3330 * locks for us to safely manipulate the slave devices (RTNL lock,
3331 * dev_probe_lock).
3332 */
3333static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3334{
3335 struct net_device *event_dev = (struct net_device *)ptr;
3336
3337 dprintk("event_dev: %s, event: %lx\n",
3338 (event_dev ? event_dev->name : "None"),
3339 event);
3340
0b680e75
JV		 3341 if (!(event_dev->priv_flags & IFF_BONDING))
3342 return NOTIFY_DONE;
3343
1da177e4
LT		 3344 if (event_dev->flags & IFF_MASTER) {
3345 dprintk("IFF_MASTER\n");
3346 return bond_master_netdev_event(event, event_dev);
3347 }
3348
3349 if (event_dev->flags & IFF_SLAVE) {
3350 dprintk("IFF_SLAVE\n");
3351 return bond_slave_netdev_event(event, event_dev);
3352 }
3353
3354 return NOTIFY_DONE;
3355}
3356
c3ade5ca
JV		 3357/*
3358 * bond_inetaddr_event: handle inetaddr notifier chain events.
3359 *
3360 * We keep track of device IPs primarily to use as source addresses in
3361 * ARP monitor probes (rather than spewing out broadcasts all the time).
3362 *
3363 * We track one IP for the main device (if it has one), plus one per VLAN.
3364 */
3365static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3366{
3367 struct in_ifaddr *ifa = ptr;
3368 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3369 struct bonding *bond, *bond_next;
3370 struct vlan_entry *vlan, *vlan_next;
3371
3372 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3373 if (bond->dev == event_dev) {
3374 switch (event) {
3375 case NETDEV_UP:
3376 bond->master_ip = ifa->ifa_local;
3377 return NOTIFY_OK;
3378 case NETDEV_DOWN:
3379 bond->master_ip = bond_glean_dev_ip(bond->dev);
3380 return NOTIFY_OK;
3381 default:
3382 return NOTIFY_DONE;
3383 }
3384 }
3385
3386 if (list_empty(&bond->vlan_list))
3387 continue;
3388
3389 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3390 vlan_list) {
5c15bdec 3391 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3392 if (vlan_dev == event_dev) {
3393 switch (event) {
3394 case NETDEV_UP:
3395 vlan->vlan_ip = ifa->ifa_local;
3396 return NOTIFY_OK;
3397 case NETDEV_DOWN:
3398 vlan->vlan_ip =
3399 bond_glean_dev_ip(vlan_dev);
3400 return NOTIFY_OK;
3401 default:
3402 return NOTIFY_DONE;
3403 }
3404 }
3405 }
3406 }
3407 return NOTIFY_DONE;
3408}
3409
1da177e4
LT		 3410static struct notifier_block bond_netdev_notifier = {
3411 .notifier_call = bond_netdev_event,
3412};
3413
c3ade5ca
JV		 3414static struct notifier_block bond_inetaddr_notifier = {
3415 .notifier_call = bond_inetaddr_event,
3416};
3417
1da177e4
LT		 3418/*-------------------------- Packet type handling ---------------------------*/
3419
3420/* register to receive lacpdus on a bond */
3421static void bond_register_lacpdu(struct bonding *bond)
3422{
3423 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3424
3425 /* initialize packet type */
3426 pk_type->type = PKT_TYPE_LACPDU;
3427 pk_type->dev = bond->dev;
3428 pk_type->func = bond_3ad_lacpdu_recv;
3429
3430 dev_add_pack(pk_type);
3431}
3432
3433/* unregister to receive lacpdus on a bond */
3434static void bond_unregister_lacpdu(struct bonding *bond)
3435{
3436 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3437}
3438
f5b2b966
JV		 3439void bond_register_arp(struct bonding *bond)
3440{
3441 struct packet_type *pt = &bond->arp_mon_pt;
3442
c4f283b1
JV		 3443 if (pt->type)
3444 return;
3445
f5b2b966 3446 pt->type = htons(ETH_P_ARP);
e245cb71 3447 pt->dev = bond->dev;
f5b2b966
JV		 3448 pt->func = bond_arp_rcv;
3449 dev_add_pack(pt);
3450}
3451
3452void bond_unregister_arp(struct bonding *bond)
3453{
c4f283b1
JV		 3454 struct packet_type *pt = &bond->arp_mon_pt;
3455
3456 dev_remove_pack(pt);
3457 pt->type = 0;
f5b2b966
JV		 3458}
3459
169a3e66
JV		 3460/*---------------------------- Hashing Policies -----------------------------*/
3461
3462/*
59c51591 3463 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3464 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3465 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3466 */
3467static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3468 struct net_device *bond_dev, int count)
3469{
3470 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3471 struct iphdr *iph = ip_hdr(skb);
169a3e66
JV		 3472 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3473 int layer4_xor = 0;
3474
3475 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3476 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3477 (iph->protocol == IPPROTO_TCP ||
3478 iph->protocol == IPPROTO_UDP)) {
3479 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3480 }
3481 return (layer4_xor ^
3482 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3483
3484 }
3485
3486 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3487}
3488
3489/*
3490 * Hash for the output device based upon layer 2 data
3491 */
3492static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3493 struct net_device *bond_dev, int count)
3494{
3495 struct ethhdr *data = (struct ethhdr *)skb->data;
3496
3497 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3498}
3499
1da177e4
LT		 3500/*-------------------------- Device entry points ----------------------------*/
3501
3502static int bond_open(struct net_device *bond_dev)
3503{
3504 struct bonding *bond = bond_dev->priv;
3505 struct timer_list *mii_timer = &bond->mii_timer;
3506 struct timer_list *arp_timer = &bond->arp_timer;
3507
3508 bond->kill_timers = 0;
3509
3510 if ((bond->params.mode == BOND_MODE_TLB) ||
3511 (bond->params.mode == BOND_MODE_ALB)) {
3512 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3513
3514 /* bond_alb_initialize must be called before the timer
3515 * is started.
3516 */
3517 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3518 /* something went wrong - fail the open operation */
3519 return -1;
3520 }
3521
3522 init_timer(alb_timer);
3523 alb_timer->expires = jiffies + 1;
3524 alb_timer->data = (unsigned long)bond;
3525 alb_timer->function = (void *)&bond_alb_monitor;
3526 add_timer(alb_timer);
3527 }
3528
3529 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3530 init_timer(mii_timer);
3531 mii_timer->expires = jiffies + 1;
3532 mii_timer->data = (unsigned long)bond_dev;
3533 mii_timer->function = (void *)&bond_mii_monitor;
3534 add_timer(mii_timer);
3535 }
3536
3537 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3538 init_timer(arp_timer);
3539 arp_timer->expires = jiffies + 1;
3540 arp_timer->data = (unsigned long)bond_dev;
3541 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3542 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3543 } else {
3544 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3545 }
f5b2b966
JV		 3546 if (bond->params.arp_validate)
3547 bond_register_arp(bond);
3548
1da177e4
LT		 3549 add_timer(arp_timer);
3550 }
3551
3552 if (bond->params.mode == BOND_MODE_8023AD) {
3553 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3554 init_timer(ad_timer);
3555 ad_timer->expires = jiffies + 1;
3556 ad_timer->data = (unsigned long)bond;
3557 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3558 add_timer(ad_timer);
3559
3560 /* register to receive LACPDUs */
3561 bond_register_lacpdu(bond);
3562 }
3563
3564 return 0;
3565}
3566
3567static int bond_close(struct net_device *bond_dev)
3568{
3569 struct bonding *bond = bond_dev->priv;
3570
3571 if (bond->params.mode == BOND_MODE_8023AD) {
3572 /* Unregister the receive of LACPDUs */
3573 bond_unregister_lacpdu(bond);
3574 }
3575
f5b2b966
JV		 3576 if (bond->params.arp_validate)
3577 bond_unregister_arp(bond);
3578
1da177e4
LT		 3579 write_lock_bh(&bond->lock);
3580
1da177e4
LT		 3581
3582 /* signal timers not to re-arm */
3583 bond->kill_timers = 1;
3584
3585 write_unlock_bh(&bond->lock);
3586
3587 /* del_timer_sync must run without holding the bond->lock
3588 * because a running timer might be trying to hold it too
3589 */
3590
3591 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3592 del_timer_sync(&bond->mii_timer);
3593 }
3594
3595 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3596 del_timer_sync(&bond->arp_timer);
3597 }
3598
3599 switch (bond->params.mode) {
3600 case BOND_MODE_8023AD:
3601 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3602 break;
3603 case BOND_MODE_TLB:
3604 case BOND_MODE_ALB:
3605 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3606 break;
3607 default:
3608 break;
3609 }
3610
1da177e4
LT		 3611
3612 if ((bond->params.mode == BOND_MODE_TLB) ||
3613 (bond->params.mode == BOND_MODE_ALB)) {
3614 /* Must be called only after all
3615 * slaves have been released
3616 */
3617 bond_alb_deinitialize(bond);
3618 }
3619
3620 return 0;
3621}
3622
3623static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3624{
3625 struct bonding *bond = bond_dev->priv;
3626 struct net_device_stats *stats = &(bond->stats), *sstats;
3627 struct slave *slave;
3628 int i;
3629
3630 memset(stats, 0, sizeof(struct net_device_stats));
3631
3632 read_lock_bh(&bond->lock);
3633
3634 bond_for_each_slave(bond, slave, i) {
c45d286e 3635 sstats = slave->dev->get_stats(slave->dev);
5a1b5898
RR		 3636 stats->rx_packets += sstats->rx_packets;
3637 stats->rx_bytes += sstats->rx_bytes;
3638 stats->rx_errors += sstats->rx_errors;
3639 stats->rx_dropped += sstats->rx_dropped;
3640
3641 stats->tx_packets += sstats->tx_packets;
3642 stats->tx_bytes += sstats->tx_bytes;
3643 stats->tx_errors += sstats->tx_errors;
3644 stats->tx_dropped += sstats->tx_dropped;
3645
3646 stats->multicast += sstats->multicast;
3647 stats->collisions += sstats->collisions;
3648
3649 stats->rx_length_errors += sstats->rx_length_errors;
3650 stats->rx_over_errors += sstats->rx_over_errors;
3651 stats->rx_crc_errors += sstats->rx_crc_errors;
3652 stats->rx_frame_errors += sstats->rx_frame_errors;
3653 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3654 stats->rx_missed_errors += sstats->rx_missed_errors;
3655
3656 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3657 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3658 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3659 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3660 stats->tx_window_errors += sstats->tx_window_errors;
1da177e4
LT		 3661 }
3662
3663 read_unlock_bh(&bond->lock);
3664
3665 return stats;
3666}
3667
3668static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3669{
3670 struct net_device *slave_dev = NULL;
3671 struct ifbond k_binfo;
3672 struct ifbond __user *u_binfo = NULL;
3673 struct ifslave k_sinfo;
3674 struct ifslave __user *u_sinfo = NULL;
3675 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3676 int res = 0;
3677
3678 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3679 bond_dev->name, cmd);
3680
3681 switch (cmd) {
1da177e4
LT		 3682 case SIOCGMIIPHY:
3683 mii = if_mii(ifr);
3684 if (!mii) {
3685 return -EINVAL;
3686 }
3687 mii->phy_id = 0;
3688 /* Fall Through */
3689 case SIOCGMIIREG:
3690 /*
3691 * We do this again just in case we were called by SIOCGMIIREG
3692 * instead of SIOCGMIIPHY.
3693 */
3694 mii = if_mii(ifr);
3695 if (!mii) {
3696 return -EINVAL;
3697 }
3698
3699 if (mii->reg_num == 1) {
3700 struct bonding *bond = bond_dev->priv;
3701 mii->val_out = 0;
3702 read_lock_bh(&bond->lock);
3703 read_lock(&bond->curr_slave_lock);
4e140079 3704 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3705 mii->val_out = BMSR_LSTATUS;
3706 }
3707 read_unlock(&bond->curr_slave_lock);
3708 read_unlock_bh(&bond->lock);
3709 }
3710
3711 return 0;
3712 case BOND_INFO_QUERY_OLD:
3713 case SIOCBONDINFOQUERY:
3714 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3715
3716 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3717 return -EFAULT;
3718 }
3719
3720 res = bond_info_query(bond_dev, &k_binfo);
3721 if (res == 0) {
3722 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3723 return -EFAULT;
3724 }
3725 }
3726
3727 return res;
3728 case BOND_SLAVE_INFO_QUERY_OLD:
3729 case SIOCBONDSLAVEINFOQUERY:
3730 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3731
3732 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3733 return -EFAULT;
3734 }
3735
3736 res = bond_slave_info_query(bond_dev, &k_sinfo);
3737 if (res == 0) {
3738 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3739 return -EFAULT;
3740 }
3741 }
3742
3743 return res;
3744 default:
3745 /* Go on */
3746 break;
3747 }
3748
3749 if (!capable(CAP_NET_ADMIN)) {
3750 return -EPERM;
3751 }
3752
b76cdba9 3753 down_write(&(bonding_rwsem));
1da177e4
LT		 3754 slave_dev = dev_get_by_name(ifr->ifr_slave);
3755
3756 dprintk("slave_dev=%p: \n", slave_dev);
3757
3758 if (!slave_dev) {
3759 res = -ENODEV;
3760 } else {
3761 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3762 switch (cmd) {
3763 case BOND_ENSLAVE_OLD:
3764 case SIOCBONDENSLAVE:
3765 res = bond_enslave(bond_dev, slave_dev);
3766 break;
3767 case BOND_RELEASE_OLD:
3768 case SIOCBONDRELEASE:
3769 res = bond_release(bond_dev, slave_dev);
3770 break;
3771 case BOND_SETHWADDR_OLD:
3772 case SIOCBONDSETHWADDR:
3773 res = bond_sethwaddr(bond_dev, slave_dev);
3774 break;
3775 case BOND_CHANGE_ACTIVE_OLD:
3776 case SIOCBONDCHANGEACTIVE:
3777 res = bond_ioctl_change_active(bond_dev, slave_dev);
3778 break;
3779 default:
3780 res = -EOPNOTSUPP;
3781 }
3782
3783 dev_put(slave_dev);
3784 }
3785
b76cdba9 3786 up_write(&(bonding_rwsem));
1da177e4
LT		 3787 return res;
3788}
3789
3790static void bond_set_multicast_list(struct net_device *bond_dev)
3791{
3792 struct bonding *bond = bond_dev->priv;
3793 struct dev_mc_list *dmi;
3794
3795 write_lock_bh(&bond->lock);
3796
3797 /*
3798 * Do promisc before checking multicast_mode
3799 */
3800 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3801 bond_set_promiscuity(bond, 1);
3802 }
3803
3804 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3805 bond_set_promiscuity(bond, -1);
3806 }
3807
3808 /* set allmulti flag to slaves */
3809 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3810 bond_set_allmulti(bond, 1);
3811 }
3812
3813 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3814 bond_set_allmulti(bond, -1);
3815 }
3816
3817 bond->flags = bond_dev->flags;
3818
3819 /* looking for addresses to add to slaves' mc list */
3820 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3821 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3822 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3823 }
3824 }
3825
3826 /* looking for addresses to delete from slaves' list */
3827 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3828 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3829 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3830 }
3831 }
3832
3833 /* save master's multicast list */
3834 bond_mc_list_destroy(bond);
3835 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3836
3837 write_unlock_bh(&bond->lock);
3838}
3839
3840/*
3841 * Change the MTU of all of a master's slaves to match the master
3842 */
3843static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3844{
3845 struct bonding *bond = bond_dev->priv;
3846 struct slave *slave, *stop_at;
3847 int res = 0;
3848 int i;
3849
3850 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3851 (bond_dev ? bond_dev->name : "None"), new_mtu);
3852
3853 /* Can't hold bond->lock with bh disabled here since
3854 * some base drivers panic. On the other hand we can't
3855 * hold bond->lock without bh disabled because we'll
3856 * deadlock. The only solution is to rely on the fact
3857 * that we're under rtnl_lock here, and the slaves
3858 * list won't change. This doesn't solve the problem
3859 * of setting the slave's MTU while it is
3860 * transmitting, but the assumption is that the base
3861 * driver can handle that.
3862 *
3863 * TODO: figure out a way to safely iterate the slaves
3864 * list, but without holding a lock around the actual
3865 * call to the base driver.
3866 */
3867
3868 bond_for_each_slave(bond, slave, i) {
3869 dprintk("s %p s->p %p c_m %p\n", slave,
3870 slave->prev, slave->dev->change_mtu);
e944ef79 3871
1da177e4
LT		 3872 res = dev_set_mtu(slave->dev, new_mtu);
3873
3874 if (res) {
3875 /* If we failed to set the slave's mtu to the new value
3876 * we must abort the operation even in ACTIVE_BACKUP
3877 * mode, because if we allow the backup slaves to have
3878 * different mtu values than the active slave we'll
3879 * need to change their mtu when doing a failover. That
3880 * means changing their mtu from timer context, which
3881 * is probably not a good idea.
3882 */
3883 dprintk("err %d %s\n", res, slave->dev->name);
3884 goto unwind;
3885 }
3886 }
3887
3888 bond_dev->mtu = new_mtu;
3889
3890 return 0;
3891
3892unwind:
3893 /* unwind from head to the slave that failed */
3894 stop_at = slave;
3895 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3896 int tmp_res;
3897
3898 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3899 if (tmp_res) {
3900 dprintk("unwind err %d dev %s\n", tmp_res,
3901 slave->dev->name);
3902 }
3903 }
3904
3905 return res;
3906}
3907
3908/*
3909 * Change HW address
3910 *
3911 * Note that many devices must be down to change the HW address, and
3912 * downing the master releases all slaves. We can make bonds full of
3913 * bonding devices to test this, however.
3914 */
3915static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3916{
3917 struct bonding *bond = bond_dev->priv;
3918 struct sockaddr *sa = addr, tmp_sa;
3919 struct slave *slave, *stop_at;
3920 int res = 0;
3921 int i;
3922
3923 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3924
3925 if (!is_valid_ether_addr(sa->sa_data)) {
3926 return -EADDRNOTAVAIL;
3927 }
3928
3929 /* Can't hold bond->lock with bh disabled here since
3930 * some base drivers panic. On the other hand we can't
3931 * hold bond->lock without bh disabled because we'll
3932 * deadlock. The only solution is to rely on the fact
3933 * that we're under rtnl_lock here, and the slaves
3934 * list won't change. This doesn't solve the problem
3935 * of setting the slave's hw address while it is
3936 * transmitting, but the assumption is that the base
3937 * driver can handle that.
3938 *
3939 * TODO: figure out a way to safely iterate the slaves
3940 * list, but without holding a lock around the actual
3941 * call to the base driver.
3942 */
3943
3944 bond_for_each_slave(bond, slave, i) {
3945 dprintk("slave %p %s\n", slave, slave->dev->name);
3946
3947 if (slave->dev->set_mac_address == NULL) {
3948 res = -EOPNOTSUPP;
3949 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3950 goto unwind;
3951 }
3952
3953 res = dev_set_mac_address(slave->dev, addr);
3954 if (res) {
3955 /* TODO: consider downing the slave
3956 * and retry ?
3957 * User should expect communications
3958 * breakage anyway until ARP finish
3959 * updating, so...
3960 */
3961 dprintk("err %d %s\n", res, slave->dev->name);
3962 goto unwind;
3963 }
3964 }
3965
3966 /* success */
3967 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3968 return 0;
3969
3970unwind:
3971 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3972 tmp_sa.sa_family = bond_dev->type;
3973
3974 /* unwind from head to the slave that failed */
3975 stop_at = slave;
3976 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3977 int tmp_res;
3978
3979 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3980 if (tmp_res) {
3981 dprintk("unwind err %d dev %s\n", tmp_res,
3982 slave->dev->name);
3983 }
3984 }
3985
3986 return res;
3987}
3988
3989static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3990{
3991 struct bonding *bond = bond_dev->priv;
3992 struct slave *slave, *start_at;
3993 int i;
3994 int res = 1;
3995
3996 read_lock(&bond->lock);
3997
3998 if (!BOND_IS_OK(bond)) {
3999 goto out;
4000 }
4001
4002 read_lock(&bond->curr_slave_lock);
4003 slave = start_at = bond->curr_active_slave;
4004 read_unlock(&bond->curr_slave_lock);
4005
4006 if (!slave) {
4007 goto out;
4008 }
4009
4010 bond_for_each_slave_from(bond, slave, i, start_at) {
4011 if (IS_UP(slave->dev) &&
4012 (slave->link == BOND_LINK_UP) &&
4013 (slave->state == BOND_STATE_ACTIVE)) {
4014 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4015
4016 write_lock(&bond->curr_slave_lock);
4017 bond->curr_active_slave = slave->next;
4018 write_unlock(&bond->curr_slave_lock);
4019
4020 break;
4021 }
4022 }
4023
4024
4025out:
4026 if (res) {
4027 /* no suitable interface, frame not sent */
4028 dev_kfree_skb(skb);
4029 }
4030 read_unlock(&bond->lock);
4031 return 0;
4032}
4033
075897ce 4034
1da177e4
LT		 4035/*
4036 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4037 * the bond has a usable interface.
4038 */
4039static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4040{
4041 struct bonding *bond = bond_dev->priv;
4042 int res = 1;
4043
1da177e4
LT		 4044 read_lock(&bond->lock);
4045 read_lock(&bond->curr_slave_lock);
4046
4047 if (!BOND_IS_OK(bond)) {
4048 goto out;
4049 }
4050
075897ce
JL		 4051 if (!bond->curr_active_slave)
4052 goto out;
4053
075897ce
JL		 4054 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4055
1da177e4
LT		 4056out:
4057 if (res) {
4058 /* no suitable interface, frame not sent */
4059 dev_kfree_skb(skb);
4060 }
4061 read_unlock(&bond->curr_slave_lock);
4062 read_unlock(&bond->lock);
4063 return 0;
4064}
4065
4066/*
169a3e66
JV		 4067 * In bond_xmit_xor() , we determine the output device by using a pre-
4068 * determined xmit_hash_policy(), If the selected device is not enabled,
4069 * find the next active slave.
1da177e4
LT		 4070 */
4071static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4072{
4073 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 4074 struct slave *slave, *start_at;
4075 int slave_no;
4076 int i;
4077 int res = 1;
4078
4079 read_lock(&bond->lock);
4080
4081 if (!BOND_IS_OK(bond)) {
4082 goto out;
4083 }
4084
169a3e66 4085 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4086
4087 bond_for_each_slave(bond, slave, i) {
4088 slave_no--;
4089 if (slave_no < 0) {
4090 break;
4091 }
4092 }
4093
4094 start_at = slave;
4095
4096 bond_for_each_slave_from(bond, slave, i, start_at) {
4097 if (IS_UP(slave->dev) &&
4098 (slave->link == BOND_LINK_UP) &&
4099 (slave->state == BOND_STATE_ACTIVE)) {
4100 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4101 break;
4102 }
4103 }
4104
4105out:
4106 if (res) {
4107 /* no suitable interface, frame not sent */
4108 dev_kfree_skb(skb);
4109 }
4110 read_unlock(&bond->lock);
4111 return 0;
4112}
4113
4114/*
4115 * in broadcast mode, we send everything to all usable interfaces.
4116 */
4117static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4118{
4119 struct bonding *bond = bond_dev->priv;
4120 struct slave *slave, *start_at;
4121 struct net_device *tx_dev = NULL;
4122 int i;
4123 int res = 1;
4124
4125 read_lock(&bond->lock);
4126
4127 if (!BOND_IS_OK(bond)) {
4128 goto out;
4129 }
4130
4131 read_lock(&bond->curr_slave_lock);
4132 start_at = bond->curr_active_slave;
4133 read_unlock(&bond->curr_slave_lock);
4134
4135 if (!start_at) {
4136 goto out;
4137 }
4138
4139 bond_for_each_slave_from(bond, slave, i, start_at) {
4140 if (IS_UP(slave->dev) &&
4141 (slave->link == BOND_LINK_UP) &&
4142 (slave->state == BOND_STATE_ACTIVE)) {
4143 if (tx_dev) {
4144 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4145 if (!skb2) {
4146 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4147 ": %s: Error: bond_xmit_broadcast(): "
4148 "skb_clone() failed\n",
4149 bond_dev->name);
1da177e4
LT		 4150 continue;
4151 }
4152
4153 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4154 if (res) {
4155 dev_kfree_skb(skb2);
4156 continue;
4157 }
4158 }
4159 tx_dev = slave->dev;
4160 }
4161 }
4162
4163 if (tx_dev) {
4164 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4165 }
4166
4167out:
4168 if (res) {
4169 /* no suitable interface, frame not sent */
4170 dev_kfree_skb(skb);
4171 }
4172 /* frame sent to all suitable interfaces */
4173 read_unlock(&bond->lock);
4174 return 0;
4175}
4176
4177/*------------------------- Device initialization ---------------------------*/
4178
4179/*
4180 * set bond mode specific net device operations
4181 */
a77b5325 4182void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4183{
169a3e66
JV		 4184 struct net_device *bond_dev = bond->dev;
4185
1da177e4
LT		 4186 switch (mode) {
4187 case BOND_MODE_ROUNDROBIN:
4188 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4189 break;
4190 case BOND_MODE_ACTIVEBACKUP:
4191 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4192 break;
4193 case BOND_MODE_XOR:
4194 bond_dev->hard_start_xmit = bond_xmit_xor;
169a3e66
JV		 4195 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4196 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4197 else
4198 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
1da177e4
LT		 4199 break;
4200 case BOND_MODE_BROADCAST:
4201 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4202 break;
4203 case BOND_MODE_8023AD:
8f903c70 4204 bond_set_master_3ad_flags(bond);
1da177e4 4205 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
169a3e66
JV		 4206 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4207 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4208 else
4209 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
1da177e4 4210 break;
1da177e4 4211 case BOND_MODE_ALB:
8f903c70
JV		 4212 bond_set_master_alb_flags(bond);
4213 /* FALLTHRU */
4214 case BOND_MODE_TLB:
1da177e4
LT		 4215 bond_dev->hard_start_xmit = bond_alb_xmit;
4216 bond_dev->set_mac_address = bond_alb_set_mac_address;
4217 break;
4218 default:
4219 /* Should never happen, mode already checked */
4220 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4221 ": %s: Error: Unknown bonding mode %d\n",
4222 bond_dev->name,
1da177e4
LT		 4223 mode);
4224 break;
4225 }
4226}
4227
217df670
JV		 4228static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4229 struct ethtool_drvinfo *drvinfo)
4230{
4231 strncpy(drvinfo->driver, DRV_NAME, 32);
4232 strncpy(drvinfo->version, DRV_VERSION, 32);
4233 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4234}
4235
7282d491 4236static const struct ethtool_ops bond_ethtool_ops = {
8531c5ff 4237 .get_tx_csum = ethtool_op_get_tx_csum,
a0de3adf
JV		 4238 .get_tso = ethtool_op_get_tso,
4239 .get_ufo = ethtool_op_get_ufo,
8531c5ff 4240 .get_sg = ethtool_op_get_sg,
217df670 4241 .get_drvinfo = bond_ethtool_get_drvinfo,
8531c5ff
AK		 4242};
4243
1da177e4
LT		 4244/*
4245 * Does not allocate but creates a /proc entry.
4246 * Allowed to fail.
4247 */
3c535952 4248static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4
LT		 4249{
4250 struct bonding *bond = bond_dev->priv;
4251
4252 dprintk("Begin bond_init for %s\n", bond_dev->name);
4253
4254 /* initialize rwlocks */
4255 rwlock_init(&bond->lock);
4256 rwlock_init(&bond->curr_slave_lock);
4257
4258 bond->params = *params; /* copy params struct */
4259
4260 /* Initialize pointers */
4261 bond->first_slave = NULL;
4262 bond->curr_active_slave = NULL;
4263 bond->current_arp_slave = NULL;
4264 bond->primary_slave = NULL;
4265 bond->dev = bond_dev;
4266 INIT_LIST_HEAD(&bond->vlan_list);
4267
4268 /* Initialize the device entry points */
4269 bond_dev->open = bond_open;
4270 bond_dev->stop = bond_close;
4271 bond_dev->get_stats = bond_get_stats;
4272 bond_dev->do_ioctl = bond_do_ioctl;
8531c5ff 4273 bond_dev->ethtool_ops = &bond_ethtool_ops;
1da177e4
LT		 4274 bond_dev->set_multicast_list = bond_set_multicast_list;
4275 bond_dev->change_mtu = bond_change_mtu;
4276 bond_dev->set_mac_address = bond_set_mac_address;
4277
169a3e66 4278 bond_set_mode_ops(bond, bond->params.mode);
1da177e4
LT		 4279
4280 bond_dev->destructor = free_netdev;
4281
4282 /* Initialize the device options */
4283 bond_dev->tx_queue_len = 0;
4284 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4285 bond_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 4286
4287 /* At first, we block adding VLANs. That's the only way to
4288 * prevent problems that occur when adding VLANs over an
4289 * empty bond. The block will be removed once non-challenged
4290 * slaves are enslaved.
4291 */
4292 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4293
932ff279 4294 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4295 * transmitting */
4296 bond_dev->features |= NETIF_F_LLTX;
4297
4298 /* By default, we declare the bond to be fully
4299 * VLAN hardware accelerated capable. Special
4300 * care is taken in the various xmit functions
4301 * when there are slaves that are not hw accel
4302 * capable
4303 */
4304 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4305 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4306 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4307 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4308 NETIF_F_HW_VLAN_RX |
4309 NETIF_F_HW_VLAN_FILTER);
4310
4311#ifdef CONFIG_PROC_FS
4312 bond_create_proc_entry(bond);
4313#endif
4314
4315 list_add_tail(&bond->bond_list, &bond_dev_list);
4316
4317 return 0;
4318}
4319
4320/* De-initialize device specific data.
4321 * Caller must hold rtnl_lock.
4322 */
a77b5325 4323void bond_deinit(struct net_device *bond_dev)
1da177e4
LT		 4324{
4325 struct bonding *bond = bond_dev->priv;
4326
4327 list_del(&bond->bond_list);
4328
4329#ifdef CONFIG_PROC_FS
4330 bond_remove_proc_entry(bond);
4331#endif
4332}
4333
4334/* Unregister and free all bond devices.
4335 * Caller must hold rtnl_lock.
4336 */
4337static void bond_free_all(void)
4338{
4339 struct bonding *bond, *nxt;
4340
4341 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4342 struct net_device *bond_dev = bond->dev;
4343
70298705 4344 bond_mc_list_destroy(bond);
4345 /* Release the bonded slaves */
4346 bond_release_all(bond_dev);
1da177e4 4347 bond_deinit(bond_dev);
3201e656 4348 unregister_netdevice(bond_dev);
1da177e4
LT		 4349 }
4350
4351#ifdef CONFIG_PROC_FS
4352 bond_destroy_proc_dir();
4353#endif
4354}
4355
4356/*------------------------- Module initialization ---------------------------*/
4357
4358/*
4359 * Convert string input module parms. Accept either the
4360 * number of the mode or its string name.
4361 */
a77b5325 4362int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
1da177e4
LT		 4363{
4364 int i;
4365
4366 for (i = 0; tbl[i].modename; i++) {
4367 if ((isdigit(*mode_arg) &&
4368 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4369 (strncmp(mode_arg, tbl[i].modename,
4370 strlen(tbl[i].modename)) == 0)) {
4371 return tbl[i].mode;
4372 }
4373 }
4374
4375 return -1;
4376}
4377
4378static int bond_check_params(struct bond_params *params)
4379{
f5b2b966
JV		 4380 int arp_validate_value;
4381
1da177e4
LT		 4382 /*
4383 * Convert string parameters.
4384 */
4385 if (mode) {
4386 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4387 if (bond_mode == -1) {
4388 printk(KERN_ERR DRV_NAME
4389 ": Error: Invalid bonding mode \"%s\"\n",
4390 mode == NULL ? "NULL" : mode);
4391 return -EINVAL;
4392 }
4393 }
4394
169a3e66
JV		 4395 if (xmit_hash_policy) {
4396 if ((bond_mode != BOND_MODE_XOR) &&
4397 (bond_mode != BOND_MODE_8023AD)) {
4398 printk(KERN_INFO DRV_NAME
4399 ": xor_mode param is irrelevant in mode %s\n",
4400 bond_mode_name(bond_mode));
4401 } else {
4402 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4403 xmit_hashtype_tbl);
4404 if (xmit_hashtype == -1) {
4405 printk(KERN_ERR DRV_NAME
4406 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4407 xmit_hash_policy == NULL ? "NULL" :
4408 xmit_hash_policy);
4409 return -EINVAL;
4410 }
4411 }
4412 }
4413
1da177e4
LT		 4414 if (lacp_rate) {
4415 if (bond_mode != BOND_MODE_8023AD) {
4416 printk(KERN_INFO DRV_NAME
4417 ": lacp_rate param is irrelevant in mode %s\n",
4418 bond_mode_name(bond_mode));
4419 } else {
4420 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4421 if (lacp_fast == -1) {
4422 printk(KERN_ERR DRV_NAME
4423 ": Error: Invalid lacp rate \"%s\"\n",
4424 lacp_rate == NULL ? "NULL" : lacp_rate);
4425 return -EINVAL;
4426 }
4427 }
4428 }
4429
4430 if (max_bonds < 1 || max_bonds > INT_MAX) {
4431 printk(KERN_WARNING DRV_NAME
4432 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4433 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
1da177e4
LT		 4434 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4435 max_bonds = BOND_DEFAULT_MAX_BONDS;
4436 }
4437
4438 if (miimon < 0) {
4439 printk(KERN_WARNING DRV_NAME
4440 ": Warning: miimon module parameter (%d), "
4441 "not in range 0-%d, so it was reset to %d\n",
4442 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4443 miimon = BOND_LINK_MON_INTERV;
4444 }
4445
4446 if (updelay < 0) {
4447 printk(KERN_WARNING DRV_NAME
4448 ": Warning: updelay module parameter (%d), "
4449 "not in range 0-%d, so it was reset to 0\n",
4450 updelay, INT_MAX);
4451 updelay = 0;
4452 }
4453
4454 if (downdelay < 0) {
4455 printk(KERN_WARNING DRV_NAME
4456 ": Warning: downdelay module parameter (%d), "
4457 "not in range 0-%d, so it was reset to 0\n",
4458 downdelay, INT_MAX);
4459 downdelay = 0;
4460 }
4461
4462 if ((use_carrier != 0) && (use_carrier != 1)) {
4463 printk(KERN_WARNING DRV_NAME
4464 ": Warning: use_carrier module parameter (%d), "
4465 "not of valid value (0/1), so it was set to 1\n",
4466 use_carrier);
4467 use_carrier = 1;
4468 }
4469
4470 /* reset values for 802.3ad */
4471 if (bond_mode == BOND_MODE_8023AD) {
4472 if (!miimon) {
4473 printk(KERN_WARNING DRV_NAME
4474 ": Warning: miimon must be specified, "
4475 "otherwise bonding will not detect link "
4476 "failure, speed and duplex which are "
4477 "essential for 802.3ad operation\n");
4478 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4479 miimon = 100;
4480 }
4481 }
4482
4483 /* reset values for TLB/ALB */
4484 if ((bond_mode == BOND_MODE_TLB) ||
4485 (bond_mode == BOND_MODE_ALB)) {
4486 if (!miimon) {
4487 printk(KERN_WARNING DRV_NAME
4488 ": Warning: miimon must be specified, "
4489 "otherwise bonding will not detect link "
4490 "failure and link speed which are essential "
4491 "for TLB/ALB load balancing\n");
4492 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4493 miimon = 100;
4494 }
4495 }
4496
4497 if (bond_mode == BOND_MODE_ALB) {
4498 printk(KERN_NOTICE DRV_NAME
4499 ": In ALB mode you might experience client "
4500 "disconnections upon reconnection of a link if the "
4501 "bonding module updelay parameter (%d msec) is "
4502 "incompatible with the forwarding delay time of the "
4503 "switch\n",
4504 updelay);
4505 }
4506
4507 if (!miimon) {
4508 if (updelay || downdelay) {
4509 /* just warn the user the up/down delay will have
4510 * no effect since miimon is zero...
4511 */
4512 printk(KERN_WARNING DRV_NAME
4513 ": Warning: miimon module parameter not set "
4514 "and updelay (%d) or downdelay (%d) module "
4515 "parameter is set; updelay and downdelay have "
4516 "no effect unless miimon is set\n",
4517 updelay, downdelay);
4518 }
4519 } else {
4520 /* don't allow arp monitoring */
4521 if (arp_interval) {
4522 printk(KERN_WARNING DRV_NAME
4523 ": Warning: miimon (%d) and arp_interval (%d) "
4524 "can't be used simultaneously, disabling ARP "
4525 "monitoring\n",
4526 miimon, arp_interval);
4527 arp_interval = 0;
4528 }
4529
4530 if ((updelay % miimon) != 0) {
4531 printk(KERN_WARNING DRV_NAME
4532 ": Warning: updelay (%d) is not a multiple "
4533 "of miimon (%d), updelay rounded to %d ms\n",
4534 updelay, miimon, (updelay / miimon) * miimon);
4535 }
4536
4537 updelay /= miimon;
4538
4539 if ((downdelay % miimon) != 0) {
4540 printk(KERN_WARNING DRV_NAME
4541 ": Warning: downdelay (%d) is not a multiple "
4542 "of miimon (%d), downdelay rounded to %d ms\n",
4543 downdelay, miimon,
4544 (downdelay / miimon) * miimon);
4545 }
4546
4547 downdelay /= miimon;
4548 }
4549
4550 if (arp_interval < 0) {
4551 printk(KERN_WARNING DRV_NAME
4552 ": Warning: arp_interval module parameter (%d) "
4553 ", not in range 0-%d, so it was reset to %d\n",
4554 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4555 arp_interval = BOND_LINK_ARP_INTERV;
4556 }
4557
4558 for (arp_ip_count = 0;
4559 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4560 arp_ip_count++) {
4561 /* not complete check, but should be good enough to
4562 catch mistakes */
4563 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4564 printk(KERN_WARNING DRV_NAME
4565 ": Warning: bad arp_ip_target module parameter "
4566 "(%s), ARP monitoring will not be performed\n",
4567 arp_ip_target[arp_ip_count]);
4568 arp_interval = 0;
4569 } else {
4570 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4571 arp_target[arp_ip_count] = ip;
4572 }
4573 }
4574
4575 if (arp_interval && !arp_ip_count) {
4576 /* don't allow arping if no arp_ip_target given... */
4577 printk(KERN_WARNING DRV_NAME
4578 ": Warning: arp_interval module parameter (%d) "
4579 "specified without providing an arp_ip_target "
4580 "parameter, arp_interval was reset to 0\n",
4581 arp_interval);
4582 arp_interval = 0;
4583 }
4584
f5b2b966
JV		 4585 if (arp_validate) {
4586 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4587 printk(KERN_ERR DRV_NAME
4588 ": arp_validate only supported in active-backup mode\n");
4589 return -EINVAL;
4590 }
4591 if (!arp_interval) {
4592 printk(KERN_ERR DRV_NAME
4593 ": arp_validate requires arp_interval\n");
4594 return -EINVAL;
4595 }
4596
4597 arp_validate_value = bond_parse_parm(arp_validate,
4598 arp_validate_tbl);
4599 if (arp_validate_value == -1) {
4600 printk(KERN_ERR DRV_NAME
4601 ": Error: invalid arp_validate \"%s\"\n",
4602 arp_validate == NULL ? "NULL" : arp_validate);
4603 return -EINVAL;
4604 }
4605 } else
4606 arp_validate_value = 0;
4607
1da177e4
LT		 4608 if (miimon) {
4609 printk(KERN_INFO DRV_NAME
4610 ": MII link monitoring set to %d ms\n",
4611 miimon);
4612 } else if (arp_interval) {
4613 int i;
4614
4615 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 4616 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4617 arp_interval,
4618 arp_validate_tbl[arp_validate_value].modename,
4619 arp_ip_count);
1da177e4
LT		 4620
4621 for (i = 0; i < arp_ip_count; i++)
4622 printk (" %s", arp_ip_target[i]);
4623
4624 printk("\n");
4625
4626 } else {
4627 /* miimon and arp_interval not set, we need one so things
4628 * work as expected, see bonding.txt for details
4629 */
4630 printk(KERN_WARNING DRV_NAME
4631 ": Warning: either miimon or arp_interval and "
4632 "arp_ip_target module parameters must be specified, "
4633 "otherwise bonding will not detect link failures! see "
4634 "bonding.txt for details.\n");
4635 }
4636
4637 if (primary && !USES_PRIMARY(bond_mode)) {
4638 /* currently, using a primary only makes sense
4639 * in active backup, TLB or ALB modes
4640 */
4641 printk(KERN_WARNING DRV_NAME
4642 ": Warning: %s primary device specified but has no "
4643 "effect in %s mode\n",
4644 primary, bond_mode_name(bond_mode));
4645 primary = NULL;
4646 }
4647
4648 /* fill params struct with the proper values */
4649 params->mode = bond_mode;
169a3e66 4650 params->xmit_policy = xmit_hashtype;
1da177e4
LT		 4651 params->miimon = miimon;
4652 params->arp_interval = arp_interval;
f5b2b966 4653 params->arp_validate = arp_validate_value;
1da177e4
LT		 4654 params->updelay = updelay;
4655 params->downdelay = downdelay;
4656 params->use_carrier = use_carrier;
4657 params->lacp_fast = lacp_fast;
4658 params->primary[0] = 0;
4659
4660 if (primary) {
4661 strncpy(params->primary, primary, IFNAMSIZ);
4662 params->primary[IFNAMSIZ - 1] = 0;
4663 }
4664
4665 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4666
4667 return 0;
4668}
4669
0daa2303
PZ		 4670static struct lock_class_key bonding_netdev_xmit_lock_key;
4671
dfe60397 4672/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 4673 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 4674 * Caller must NOT hold rtnl_lock; we need to release it here before we
4675 * set up our sysfs entries.
4676 */
4677int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4678{
4679 struct net_device *bond_dev;
4680 int res;
4681
4682 rtnl_lock();
e4b91c48
JV		 4683 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4684 ether_setup);
dfe60397
MW		 4685 if (!bond_dev) {
4686 printk(KERN_ERR DRV_NAME
4687 ": %s: eek! can't alloc netdev!\n",
4688 name);
4689 res = -ENOMEM;
4690 goto out_rtnl;
4691 }
4692
e4b91c48
JV		 4693 if (!name) {
4694 res = dev_alloc_name(bond_dev, "bond%d");
4695 if (res < 0)
4696 goto out_netdev;
4697 }
4698
dfe60397
MW		 4699 /* bond_init() must be called after dev_alloc_name() (for the
4700 * /proc files), but before register_netdevice(), because we
4701 * need to set function pointers.
4702 */
4703
4704 res = bond_init(bond_dev, params);
4705 if (res < 0) {
4706 goto out_netdev;
4707 }
4708
4709 SET_MODULE_OWNER(bond_dev);
4710
4711 res = register_netdevice(bond_dev);
4712 if (res < 0) {
4713 goto out_bond;
4714 }
0daa2303
PZ		 4715
4716 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4717
dfe60397
MW		 4718 if (newbond)
4719 *newbond = bond_dev->priv;
4720
ff59c456
JV		 4721 netif_carrier_off(bond_dev);
4722
dfe60397 4723 rtnl_unlock(); /* allows sysfs registration of net device */
b76cdba9 4724 res = bond_create_sysfs_entry(bond_dev->priv);
09c89279
JV		 4725 if (res < 0) {
4726 rtnl_lock();
4727 goto out_bond;
4728 }
4729
4730 return 0;
4731
dfe60397
MW		 4732out_bond:
4733 bond_deinit(bond_dev);
4734out_netdev:
4735 free_netdev(bond_dev);
4736out_rtnl:
4737 rtnl_unlock();
dfe60397
MW		 4738 return res;
4739}
4740
1da177e4
LT		 4741static int __init bonding_init(void)
4742{
1da177e4
LT		 4743 int i;
4744 int res;
4745
4746 printk(KERN_INFO "%s", version);
4747
dfe60397 4748 res = bond_check_params(&bonding_defaults);
1da177e4 4749 if (res) {
dfe60397 4750 goto out;
1da177e4
LT		 4751 }
4752
1da177e4
LT		 4753#ifdef CONFIG_PROC_FS
4754 bond_create_proc_dir();
4755#endif
1da177e4 4756 for (i = 0; i < max_bonds; i++) {
e4b91c48 4757 res = bond_create(NULL, &bonding_defaults, NULL);
dfe60397
MW		 4758 if (res)
4759 goto err;
1da177e4
LT		 4760 }
4761
b76cdba9
MW		 4762 res = bond_create_sysfs();
4763 if (res)
4764 goto err;
4765
1da177e4 4766 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 4767 register_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 4768
dfe60397
MW		 4769 goto out;
4770err:
40abc270 4771 rtnl_lock();
1da177e4 4772 bond_free_all();
b76cdba9 4773 bond_destroy_sysfs();
1da177e4 4774 rtnl_unlock();
dfe60397 4775out:
1da177e4 4776 return res;
dfe60397 4777
1da177e4
LT		 4778}
4779
4780static void __exit bonding_exit(void)
4781{
4782 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 4783 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4
LT		 4784
4785 rtnl_lock();
4786 bond_free_all();
b76cdba9 4787 bond_destroy_sysfs();
1da177e4
LT		 4788 rtnl_unlock();
4789}
4790
4791module_init(bonding_init);
4792module_exit(bonding_exit);
4793MODULE_LICENSE("GPL");
4794MODULE_VERSION(DRV_VERSION);
4795MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4796MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4797MODULE_SUPPORTED_DEVICE("most ethernet devices");
4798
4799/*
4800 * Local variables:
4801 * c-indent-level: 8
4802 * c-basic-offset: 8
4803 * tab-width: 8
4804 * End:
4805 */
4806
Clone of davem's net-next-2.6 tree