]> 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
bonding: bond_change_active_slave() cleanup under active-backup
[net-next-2.6.git] / drivers / net / bonding / bond_main.c
CommitLineData
1da177e4
LT		 1/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
1da177e4
LT		 32 */
33
34//#define BONDING_DEBUG 1
35
1da177e4
LT		 36#include <linux/kernel.h>
37#include <linux/module.h>
1da177e4
LT		 38#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
169a3e66 44#include <net/ip.h>
1da177e4 45#include <linux/ip.h>
169a3e66
JV		 46#include <linux/tcp.h>
47#include <linux/udp.h>
1da177e4
LT		 48#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
56#include <asm/system.h>
57#include <asm/io.h>
58#include <asm/dma.h>
59#include <asm/uaccess.h>
60#include <linux/errno.h>
61#include <linux/netdevice.h>
62#include <linux/inetdevice.h>
a816c7c7 63#include <linux/igmp.h>
1da177e4
LT		 64#include <linux/etherdevice.h>
65#include <linux/skbuff.h>
66#include <net/sock.h>
67#include <linux/rtnetlink.h>
68#include <linux/proc_fs.h>
69#include <linux/seq_file.h>
70#include <linux/smp.h>
71#include <linux/if_ether.h>
72#include <net/arp.h>
73#include <linux/mii.h>
74#include <linux/ethtool.h>
75#include <linux/if_vlan.h>
76#include <linux/if_bonding.h>
b63bb739 77#include <linux/jiffies.h>
c3ade5ca 78#include <net/route.h>
457c4cbc 79#include <net/net_namespace.h>
1da177e4
LT		 80#include "bonding.h"
81#include "bond_3ad.h"
82#include "bond_alb.h"
83
84/*---------------------------- Module parameters ----------------------------*/
85
86/* monitor all links that often (in milliseconds). <=0 disables monitoring */
87#define BOND_LINK_MON_INTERV 0
88#define BOND_LINK_ARP_INTERV 0
89
90static int max_bonds = BOND_DEFAULT_MAX_BONDS;
7893b249 91static int num_grat_arp = 1;
1da177e4
LT		 92static int miimon = BOND_LINK_MON_INTERV;
93static int updelay = 0;
94static int downdelay = 0;
95static int use_carrier = 1;
96static char *mode = NULL;
97static char *primary = NULL;
98static char *lacp_rate = NULL;
169a3e66 99static char *xmit_hash_policy = NULL;
1da177e4
LT		 100static int arp_interval = BOND_LINK_ARP_INTERV;
101static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
f5b2b966 102static char *arp_validate = NULL;
3915c1e8 103static char *fail_over_mac = NULL;
12479f9a 104struct bond_params bonding_defaults;
1da177e4
LT		 105
106module_param(max_bonds, int, 0);
107MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
7893b249
MS		 108module_param(num_grat_arp, int, 0644);
109MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
1da177e4
LT		 110module_param(miimon, int, 0);
111MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
112module_param(updelay, int, 0);
113MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
114module_param(downdelay, int, 0);
2ac47660
MW		 115MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
116 "in milliseconds");
1da177e4 117module_param(use_carrier, int, 0);
2ac47660
MW		 118MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
119 "0 for off, 1 for on (default)");
1da177e4 120module_param(mode, charp, 0);
2ac47660
MW		 121MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
122 "1 for active-backup, 2 for balance-xor, "
123 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
124 "6 for balance-alb");
1da177e4
LT		 125module_param(primary, charp, 0);
126MODULE_PARM_DESC(primary, "Primary network device to use");
127module_param(lacp_rate, charp, 0);
2ac47660
MW		 128MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
129 "(slow/fast)");
169a3e66 130module_param(xmit_hash_policy, charp, 0);
2ac47660
MW		 131MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
132 ", 1 for layer 3+4");
1da177e4
LT		 133module_param(arp_interval, int, 0);
134MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
135module_param_array(arp_ip_target, charp, NULL, 0);
136MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
f5b2b966
JV		 137module_param(arp_validate, charp, 0);
138MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
3915c1e8
JV		 139module_param(fail_over_mac, charp, 0);
140MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
1da177e4
LT		 141
142/*----------------------------- Global variables ----------------------------*/
143
f71e1309 144static const char * const version =
1da177e4
LT		 145 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
146
12479f9a 147LIST_HEAD(bond_dev_list);
1da177e4
LT		 148
149#ifdef CONFIG_PROC_FS
150static struct proc_dir_entry *bond_proc_dir = NULL;
151#endif
152
b76cdba9 153extern struct rw_semaphore bonding_rwsem;
d3bb52b0 154static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
1da177e4 155static int arp_ip_count = 0;
1da177e4 156static int bond_mode = BOND_MODE_ROUNDROBIN;
169a3e66 157static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
1da177e4 158static int lacp_fast = 0;
217df670 159
1da177e4 160
12479f9a 161struct bond_parm_tbl bond_lacp_tbl[] = {
1da177e4
LT		 162{ "slow", AD_LACP_SLOW},
163{ "fast", AD_LACP_FAST},
164{ NULL, -1},
165};
166
12479f9a 167struct bond_parm_tbl bond_mode_tbl[] = {
1da177e4
LT		 168{ "balance-rr", BOND_MODE_ROUNDROBIN},
169{ "active-backup", BOND_MODE_ACTIVEBACKUP},
170{ "balance-xor", BOND_MODE_XOR},
171{ "broadcast", BOND_MODE_BROADCAST},
172{ "802.3ad", BOND_MODE_8023AD},
173{ "balance-tlb", BOND_MODE_TLB},
174{ "balance-alb", BOND_MODE_ALB},
175{ NULL, -1},
176};
177
12479f9a 178struct bond_parm_tbl xmit_hashtype_tbl[] = {
169a3e66
JV		 179{ "layer2", BOND_XMIT_POLICY_LAYER2},
180{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
6f6652be 181{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
169a3e66
JV		 182{ NULL, -1},
183};
184
f5b2b966
JV		 185struct bond_parm_tbl arp_validate_tbl[] = {
186{ "none", BOND_ARP_VALIDATE_NONE},
187{ "active", BOND_ARP_VALIDATE_ACTIVE},
188{ "backup", BOND_ARP_VALIDATE_BACKUP},
189{ "all", BOND_ARP_VALIDATE_ALL},
190{ NULL, -1},
191};
192
3915c1e8
JV		 193struct bond_parm_tbl fail_over_mac_tbl[] = {
194{ "none", BOND_FOM_NONE},
195{ "active", BOND_FOM_ACTIVE},
196{ "follow", BOND_FOM_FOLLOW},
197{ NULL, -1},
198};
199
1da177e4
LT		 200/*-------------------------- Forward declarations ---------------------------*/
201
c3ade5ca 202static void bond_send_gratuitous_arp(struct bonding *bond);
c50b85d0 203static void bond_deinit(struct net_device *bond_dev);
1da177e4
LT		 204
205/*---------------------------- General routines -----------------------------*/
206
4ad072c9 207static const char *bond_mode_name(int mode)
1da177e4
LT		 208{
209 switch (mode) {
210 case BOND_MODE_ROUNDROBIN :
211 return "load balancing (round-robin)";
212 case BOND_MODE_ACTIVEBACKUP :
213 return "fault-tolerance (active-backup)";
214 case BOND_MODE_XOR :
215 return "load balancing (xor)";
216 case BOND_MODE_BROADCAST :
217 return "fault-tolerance (broadcast)";
218 case BOND_MODE_8023AD:
219 return "IEEE 802.3ad Dynamic link aggregation";
220 case BOND_MODE_TLB:
221 return "transmit load balancing";
222 case BOND_MODE_ALB:
223 return "adaptive load balancing";
224 default:
225 return "unknown";
226 }
227}
228
229/*---------------------------------- VLAN -----------------------------------*/
230
231/**
232 * bond_add_vlan - add a new vlan id on bond
233 * @bond: bond that got the notification
234 * @vlan_id: the vlan id to add
235 *
236 * Returns -ENOMEM if allocation failed.
237 */
238static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
239{
240 struct vlan_entry *vlan;
241
242 dprintk("bond: %s, vlan id %d\n",
243 (bond ? bond->dev->name: "None"), vlan_id);
244
245 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
246 if (!vlan) {
247 return -ENOMEM;
248 }
249
250 INIT_LIST_HEAD(&vlan->vlan_list);
251 vlan->vlan_id = vlan_id;
c3ade5ca 252 vlan->vlan_ip = 0;
1da177e4
LT		 253
254 write_lock_bh(&bond->lock);
255
256 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
257
258 write_unlock_bh(&bond->lock);
259
260 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
261
262 return 0;
263}
264
265/**
266 * bond_del_vlan - delete a vlan id from bond
267 * @bond: bond that got the notification
268 * @vlan_id: the vlan id to delete
269 *
270 * returns -ENODEV if @vlan_id was not found in @bond.
271 */
272static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
273{
0883beca 274 struct vlan_entry *vlan;
1da177e4
LT		 275 int res = -ENODEV;
276
277 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
278
279 write_lock_bh(&bond->lock);
280
0883beca 281 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
1da177e4
LT		 282 if (vlan->vlan_id == vlan_id) {
283 list_del(&vlan->vlan_list);
284
285 if ((bond->params.mode == BOND_MODE_TLB) ||
286 (bond->params.mode == BOND_MODE_ALB)) {
287 bond_alb_clear_vlan(bond, vlan_id);
288 }
289
290 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
291 bond->dev->name);
292
293 kfree(vlan);
294
295 if (list_empty(&bond->vlan_list) &&
296 (bond->slave_cnt == 0)) {
297 /* Last VLAN removed and no slaves, so
298 * restore block on adding VLANs. This will
299 * be removed once new slaves that are not
300 * VLAN challenged will be added.
301 */
302 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
303 }
304
305 res = 0;
306 goto out;
307 }
308 }
309
310 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
311 bond->dev->name);
312
313out:
314 write_unlock_bh(&bond->lock);
315 return res;
316}
317
318/**
319 * bond_has_challenged_slaves
320 * @bond: the bond we're working on
321 *
322 * Searches the slave list. Returns 1 if a vlan challenged slave
323 * was found, 0 otherwise.
324 *
325 * Assumes bond->lock is held.
326 */
327static int bond_has_challenged_slaves(struct bonding *bond)
328{
329 struct slave *slave;
330 int i;
331
332 bond_for_each_slave(bond, slave, i) {
333 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
334 dprintk("found VLAN challenged slave - %s\n",
335 slave->dev->name);
336 return 1;
337 }
338 }
339
340 dprintk("no VLAN challenged slaves found\n");
341 return 0;
342}
343
344/**
345 * bond_next_vlan - safely skip to the next item in the vlans list.
346 * @bond: the bond we're working on
347 * @curr: item we're advancing from
348 *
349 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
350 * or @curr->next otherwise (even if it is @curr itself again).
351 *
352 * Caller must hold bond->lock
353 */
354struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
355{
356 struct vlan_entry *next, *last;
357
358 if (list_empty(&bond->vlan_list)) {
359 return NULL;
360 }
361
362 if (!curr) {
363 next = list_entry(bond->vlan_list.next,
364 struct vlan_entry, vlan_list);
365 } else {
366 last = list_entry(bond->vlan_list.prev,
367 struct vlan_entry, vlan_list);
368 if (last == curr) {
369 next = list_entry(bond->vlan_list.next,
370 struct vlan_entry, vlan_list);
371 } else {
372 next = list_entry(curr->vlan_list.next,
373 struct vlan_entry, vlan_list);
374 }
375 }
376
377 return next;
378}
379
380/**
381 * bond_dev_queue_xmit - Prepare skb for xmit.
382 *
383 * @bond: bond device that got this skb for tx.
384 * @skb: hw accel VLAN tagged skb to transmit
385 * @slave_dev: slave that is supposed to xmit this skbuff
386 *
387 * When the bond gets an skb to transmit that is
388 * already hardware accelerated VLAN tagged, and it
389 * needs to relay this skb to a slave that is not
390 * hw accel capable, the skb needs to be "unaccelerated",
391 * i.e. strip the hwaccel tag and re-insert it as part
392 * of the payload.
393 */
394int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
395{
966bc6f4 396 unsigned short uninitialized_var(vlan_id);
1da177e4
LT		 397
398 if (!list_empty(&bond->vlan_list) &&
399 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
400 vlan_get_tag(skb, &vlan_id) == 0) {
401 skb->dev = slave_dev;
402 skb = vlan_put_tag(skb, vlan_id);
403 if (!skb) {
404 /* vlan_put_tag() frees the skb in case of error,
405 * so return success here so the calling functions
406 * won't attempt to free is again.
407 */
408 return 0;
409 }
410 } else {
411 skb->dev = slave_dev;
412 }
413
414 skb->priority = 1;
415 dev_queue_xmit(skb);
416
417 return 0;
418}
419
420/*
421 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
422 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
423 * lock because:
424 * a. This operation is performed in IOCTL context,
425 * b. The operation is protected by the RTNL semaphore in the 8021q code,
426 * c. Holding a lock with BH disabled while directly calling a base driver
427 * entry point is generally a BAD idea.
428 *
429 * The design of synchronization/protection for this operation in the 8021q
430 * module is good for one or more VLAN devices over a single physical device
431 * and cannot be extended for a teaming solution like bonding, so there is a
432 * potential race condition here where a net device from the vlan group might
433 * be referenced (either by a base driver or the 8021q code) while it is being
434 * removed from the system. However, it turns out we're not making matters
435 * worse, and if it works for regular VLAN usage it will work here too.
436*/
437
438/**
439 * bond_vlan_rx_register - Propagates registration to slaves
440 * @bond_dev: bonding net device that got called
441 * @grp: vlan group being registered
442 */
443static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
444{
445 struct bonding *bond = bond_dev->priv;
446 struct slave *slave;
447 int i;
448
449 bond->vlgrp = grp;
450
451 bond_for_each_slave(bond, slave, i) {
452 struct net_device *slave_dev = slave->dev;
453
454 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
455 slave_dev->vlan_rx_register) {
456 slave_dev->vlan_rx_register(slave_dev, grp);
457 }
458 }
459}
460
461/**
462 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
463 * @bond_dev: bonding net device that got called
464 * @vid: vlan id being added
465 */
466static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
467{
468 struct bonding *bond = bond_dev->priv;
469 struct slave *slave;
470 int i, res;
471
472 bond_for_each_slave(bond, slave, i) {
473 struct net_device *slave_dev = slave->dev;
474
475 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
476 slave_dev->vlan_rx_add_vid) {
477 slave_dev->vlan_rx_add_vid(slave_dev, vid);
478 }
479 }
480
481 res = bond_add_vlan(bond, vid);
482 if (res) {
483 printk(KERN_ERR DRV_NAME
4e0952c7 484 ": %s: Error: Failed to add vlan id %d\n",
1da177e4
LT		 485 bond_dev->name, vid);
486 }
487}
488
489/**
490 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
491 * @bond_dev: bonding net device that got called
492 * @vid: vlan id being removed
493 */
494static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
495{
496 struct bonding *bond = bond_dev->priv;
497 struct slave *slave;
498 struct net_device *vlan_dev;
499 int i, res;
500
501 bond_for_each_slave(bond, slave, i) {
502 struct net_device *slave_dev = slave->dev;
503
504 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
505 slave_dev->vlan_rx_kill_vid) {
506 /* Save and then restore vlan_dev in the grp array,
507 * since the slave's driver might clear it.
508 */
5c15bdec 509 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
1da177e4 510 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
5c15bdec 511 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
1da177e4
LT		 512 }
513 }
514
515 res = bond_del_vlan(bond, vid);
516 if (res) {
517 printk(KERN_ERR DRV_NAME
4e0952c7 518 ": %s: Error: Failed to remove vlan id %d\n",
1da177e4
LT		 519 bond_dev->name, vid);
520 }
521}
522
523static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
524{
525 struct vlan_entry *vlan;
526
527 write_lock_bh(&bond->lock);
528
529 if (list_empty(&bond->vlan_list)) {
530 goto out;
531 }
532
533 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
534 slave_dev->vlan_rx_register) {
535 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
536 }
537
538 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
539 !(slave_dev->vlan_rx_add_vid)) {
540 goto out;
541 }
542
543 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
544 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
545 }
546
547out:
548 write_unlock_bh(&bond->lock);
549}
550
551static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
552{
553 struct vlan_entry *vlan;
554 struct net_device *vlan_dev;
555
556 write_lock_bh(&bond->lock);
557
558 if (list_empty(&bond->vlan_list)) {
559 goto out;
560 }
561
562 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
563 !(slave_dev->vlan_rx_kill_vid)) {
564 goto unreg;
565 }
566
567 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
568 /* Save and then restore vlan_dev in the grp array,
569 * since the slave's driver might clear it.
570 */
5c15bdec 571 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
1da177e4 572 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
5c15bdec 573 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
1da177e4
LT		 574 }
575
576unreg:
577 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
578 slave_dev->vlan_rx_register) {
579 slave_dev->vlan_rx_register(slave_dev, NULL);
580 }
581
582out:
583 write_unlock_bh(&bond->lock);
584}
585
586/*------------------------------- Link status -------------------------------*/
587
ff59c456
JV		 588/*
589 * Set the carrier state for the master according to the state of its
590 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
591 * do special 802.3ad magic.
592 *
593 * Returns zero if carrier state does not change, nonzero if it does.
594 */
595static int bond_set_carrier(struct bonding *bond)
596{
597 struct slave *slave;
598 int i;
599
600 if (bond->slave_cnt == 0)
601 goto down;
602
603 if (bond->params.mode == BOND_MODE_8023AD)
604 return bond_3ad_set_carrier(bond);
605
606 bond_for_each_slave(bond, slave, i) {
607 if (slave->link == BOND_LINK_UP) {
608 if (!netif_carrier_ok(bond->dev)) {
609 netif_carrier_on(bond->dev);
610 return 1;
611 }
612 return 0;
613 }
614 }
615
616down:
617 if (netif_carrier_ok(bond->dev)) {
618 netif_carrier_off(bond->dev);
619 return 1;
620 }
621 return 0;
622}
623
1da177e4
LT		 624/*
625 * Get link speed and duplex from the slave's base driver
626 * using ethtool. If for some reason the call fails or the
627 * values are invalid, fake speed and duplex to 100/Full
628 * and return error.
629 */
630static int bond_update_speed_duplex(struct slave *slave)
631{
632 struct net_device *slave_dev = slave->dev;
1da177e4 633 struct ethtool_cmd etool;
61a44b9c 634 int res;
1da177e4
LT		 635
636 /* Fake speed and duplex */
637 slave->speed = SPEED_100;
638 slave->duplex = DUPLEX_FULL;
639
61a44b9c
MW		 640 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
641 return -1;
1da177e4 642
61a44b9c
MW		 643 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
644 if (res < 0)
1da177e4 645 return -1;
1da177e4 646
1da177e4
LT		 647 switch (etool.speed) {
648 case SPEED_10:
649 case SPEED_100:
650 case SPEED_1000:
94dbffd5 651 case SPEED_10000:
1da177e4
LT		 652 break;
653 default:
654 return -1;
655 }
656
657 switch (etool.duplex) {
658 case DUPLEX_FULL:
659 case DUPLEX_HALF:
660 break;
661 default:
662 return -1;
663 }
664
665 slave->speed = etool.speed;
666 slave->duplex = etool.duplex;
667
668 return 0;
669}
670
671/*
672 * if <dev> supports MII link status reporting, check its link status.
673 *
674 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
675 * depening upon the setting of the use_carrier parameter.
676 *
677 * Return either BMSR_LSTATUS, meaning that the link is up (or we
678 * can't tell and just pretend it is), or 0, meaning that the link is
679 * down.
680 *
681 * If reporting is non-zero, instead of faking link up, return -1 if
682 * both ETHTOOL and MII ioctls fail (meaning the device does not
683 * support them). If use_carrier is set, return whatever it says.
684 * It'd be nice if there was a good way to tell if a driver supports
685 * netif_carrier, but there really isn't.
686 */
687static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
688{
689 static int (* ioctl)(struct net_device *, struct ifreq *, int);
690 struct ifreq ifr;
691 struct mii_ioctl_data *mii;
1da177e4
LT		 692
693 if (bond->params.use_carrier) {
694 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
695 }
696
697 ioctl = slave_dev->do_ioctl;
698 if (ioctl) {
699 /* TODO: set pointer to correct ioctl on a per team member */
700 /* bases to make this more efficient. that is, once */
701 /* we determine the correct ioctl, we will always */
702 /* call it and not the others for that team */
703 /* member. */
704
705 /*
706 * We cannot assume that SIOCGMIIPHY will also read a
707 * register; not all network drivers (e.g., e100)
708 * support that.
709 */
710
711 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
712 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
713 mii = if_mii(&ifr);
714 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
715 mii->reg_num = MII_BMSR;
716 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
717 return (mii->val_out & BMSR_LSTATUS);
718 }
719 }
720 }
721
61a44b9c
MW		 722 /*
723 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
724 * attempt to get link status from it if the above MII ioctls fail.
725 */
1da177e4
LT		 726 if (slave_dev->ethtool_ops) {
727 if (slave_dev->ethtool_ops->get_link) {
728 u32 link;
729
730 link = slave_dev->ethtool_ops->get_link(slave_dev);
731
732 return link ? BMSR_LSTATUS : 0;
733 }
734 }
735
1da177e4
LT		 736 /*
737 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 738 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 739 * cannot report link status). If not reporting, pretend
740 * we're ok.
741 */
742 return (reporting ? -1 : BMSR_LSTATUS);
743}
744
745/*----------------------------- Multicast list ------------------------------*/
746
747/*
748 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
749 */
750static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
751{
752 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
753 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
754}
755
756/*
757 * returns dmi entry if found, NULL otherwise
758 */
759static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
760{
761 struct dev_mc_list *idmi;
762
763 for (idmi = mc_list; idmi; idmi = idmi->next) {
764 if (bond_is_dmi_same(dmi, idmi)) {
765 return idmi;
766 }
767 }
768
769 return NULL;
770}
771
772/*
773 * Push the promiscuity flag down to appropriate slaves
774 */
775static void bond_set_promiscuity(struct bonding *bond, int inc)
776{
777 if (USES_PRIMARY(bond->params.mode)) {
778 /* write lock already acquired */
779 if (bond->curr_active_slave) {
780 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
781 }
782 } else {
783 struct slave *slave;
784 int i;
785 bond_for_each_slave(bond, slave, i) {
786 dev_set_promiscuity(slave->dev, inc);
787 }
788 }
789}
790
791/*
792 * Push the allmulti flag down to all slaves
793 */
794static void bond_set_allmulti(struct bonding *bond, int inc)
795{
796 if (USES_PRIMARY(bond->params.mode)) {
797 /* write lock already acquired */
798 if (bond->curr_active_slave) {
799 dev_set_allmulti(bond->curr_active_slave->dev, inc);
800 }
801 } else {
802 struct slave *slave;
803 int i;
804 bond_for_each_slave(bond, slave, i) {
805 dev_set_allmulti(slave->dev, inc);
806 }
807 }
808}
809
810/*
811 * Add a Multicast address to slaves
812 * according to mode
813 */
814static void bond_mc_add(struct bonding *bond, void *addr, int alen)
815{
816 if (USES_PRIMARY(bond->params.mode)) {
817 /* write lock already acquired */
818 if (bond->curr_active_slave) {
819 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
820 }
821 } else {
822 struct slave *slave;
823 int i;
824 bond_for_each_slave(bond, slave, i) {
825 dev_mc_add(slave->dev, addr, alen, 0);
826 }
827 }
828}
829
830/*
831 * Remove a multicast address from slave
832 * according to mode
833 */
834static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
835{
836 if (USES_PRIMARY(bond->params.mode)) {
837 /* write lock already acquired */
838 if (bond->curr_active_slave) {
839 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
840 }
841 } else {
842 struct slave *slave;
843 int i;
844 bond_for_each_slave(bond, slave, i) {
845 dev_mc_delete(slave->dev, addr, alen, 0);
846 }
847 }
848}
849
a816c7c7
JV		 850
851/*
852 * Retrieve the list of registered multicast addresses for the bonding
853 * device and retransmit an IGMP JOIN request to the current active
854 * slave.
855 */
856static void bond_resend_igmp_join_requests(struct bonding *bond)
857{
858 struct in_device *in_dev;
859 struct ip_mc_list *im;
860
861 rcu_read_lock();
862 in_dev = __in_dev_get_rcu(bond->dev);
863 if (in_dev) {
864 for (im = in_dev->mc_list; im; im = im->next) {
865 ip_mc_rejoin_group(im);
866 }
867 }
868
869 rcu_read_unlock();
870}
871
1da177e4
LT		 872/*
873 * Totally destroys the mc_list in bond
874 */
875static void bond_mc_list_destroy(struct bonding *bond)
876{
877 struct dev_mc_list *dmi;
878
879 dmi = bond->mc_list;
880 while (dmi) {
881 bond->mc_list = dmi->next;
882 kfree(dmi);
883 dmi = bond->mc_list;
884 }
a816c7c7 885 bond->mc_list = NULL;
1da177e4
LT		 886}
887
888/*
889 * Copy all the Multicast addresses from src to the bonding device dst
890 */
de54f390 891static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 892 gfp_t gfp_flag)
1da177e4
LT		 893{
894 struct dev_mc_list *dmi, *new_dmi;
895
896 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 897 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT		 898
899 if (!new_dmi) {
900 /* FIXME: Potential memory leak !!! */
901 return -ENOMEM;
902 }
903
904 new_dmi->next = bond->mc_list;
905 bond->mc_list = new_dmi;
906 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
907 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
908 new_dmi->dmi_users = dmi->dmi_users;
909 new_dmi->dmi_gusers = dmi->dmi_gusers;
910 }
911
912 return 0;
913}
914
915/*
916 * flush all members of flush->mc_list from device dev->mc_list
917 */
918static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
919{
920 struct bonding *bond = bond_dev->priv;
921 struct dev_mc_list *dmi;
922
923 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
924 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
925 }
926
927 if (bond->params.mode == BOND_MODE_8023AD) {
928 /* del lacpdu mc addr from mc list */
929 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
930
931 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
932 }
933}
934
935/*--------------------------- Active slave change ---------------------------*/
936
937/*
938 * Update the mc list and multicast-related flags for the new and
939 * old active slaves (if any) according to the multicast mode, and
940 * promiscuous flags unconditionally.
941 */
942static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
943{
944 struct dev_mc_list *dmi;
945
946 if (!USES_PRIMARY(bond->params.mode)) {
947 /* nothing to do - mc list is already up-to-date on
948 * all slaves
949 */
950 return;
951 }
952
953 if (old_active) {
954 if (bond->dev->flags & IFF_PROMISC) {
955 dev_set_promiscuity(old_active->dev, -1);
956 }
957
958 if (bond->dev->flags & IFF_ALLMULTI) {
959 dev_set_allmulti(old_active->dev, -1);
960 }
961
962 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
963 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
964 }
965 }
966
967 if (new_active) {
968 if (bond->dev->flags & IFF_PROMISC) {
969 dev_set_promiscuity(new_active->dev, 1);
970 }
971
972 if (bond->dev->flags & IFF_ALLMULTI) {
973 dev_set_allmulti(new_active->dev, 1);
974 }
975
976 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
977 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
978 }
a816c7c7 979 bond_resend_igmp_join_requests(bond);
1da177e4
LT		 980 }
981}
982
3915c1e8
JV		 983/*
984 * bond_do_fail_over_mac
985 *
986 * Perform special MAC address swapping for fail_over_mac settings
987 *
988 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
989 */
990static void bond_do_fail_over_mac(struct bonding *bond,
991 struct slave *new_active,
992 struct slave *old_active)
993{
994 u8 tmp_mac[ETH_ALEN];
995 struct sockaddr saddr;
996 int rv;
997
998 switch (bond->params.fail_over_mac) {
999 case BOND_FOM_ACTIVE:
1000 if (new_active)
1001 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1002 new_active->dev->addr_len);
1003 break;
1004 case BOND_FOM_FOLLOW:
1005 /*
1006 * if new_active && old_active, swap them
1007 * if just old_active, do nothing (going to no active slave)
1008 * if just new_active, set new_active to bond's MAC
1009 */
1010 if (!new_active)
1011 return;
1012
1013 write_unlock_bh(&bond->curr_slave_lock);
1014 read_unlock(&bond->lock);
1015
1016 if (old_active) {
1017 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1018 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1019 ETH_ALEN);
1020 saddr.sa_family = new_active->dev->type;
1021 } else {
1022 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1023 saddr.sa_family = bond->dev->type;
1024 }
1025
1026 rv = dev_set_mac_address(new_active->dev, &saddr);
1027 if (rv) {
1028 printk(KERN_ERR DRV_NAME
1029 ": %s: Error %d setting MAC of slave %s\n",
1030 bond->dev->name, -rv, new_active->dev->name);
1031 goto out;
1032 }
1033
1034 if (!old_active)
1035 goto out;
1036
1037 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1038 saddr.sa_family = old_active->dev->type;
1039
1040 rv = dev_set_mac_address(old_active->dev, &saddr);
1041 if (rv)
1042 printk(KERN_ERR DRV_NAME
1043 ": %s: Error %d setting MAC of slave %s\n",
1044 bond->dev->name, -rv, new_active->dev->name);
1045out:
1046 read_lock(&bond->lock);
1047 write_lock_bh(&bond->curr_slave_lock);
1048 break;
1049 default:
1050 printk(KERN_ERR DRV_NAME
1051 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1052 bond->dev->name, bond->params.fail_over_mac);
1053 break;
1054 }
1055
1056}
1057
1058
1da177e4
LT		 1059/**
1060 * find_best_interface - select the best available slave to be the active one
1061 * @bond: our bonding struct
1062 *
1063 * Warning: Caller must hold curr_slave_lock for writing.
1064 */
1065static struct slave *bond_find_best_slave(struct bonding *bond)
1066{
1067 struct slave *new_active, *old_active;
1068 struct slave *bestslave = NULL;
1069 int mintime = bond->params.updelay;
1070 int i;
1071
1072 new_active = old_active = bond->curr_active_slave;
1073
1074 if (!new_active) { /* there were no active slaves left */
1075 if (bond->slave_cnt > 0) { /* found one slave */
1076 new_active = bond->first_slave;
1077 } else {
1078 return NULL; /* still no slave, return NULL */
1079 }
1080 }
1081
1082 /* first try the primary link; if arping, a link must tx/rx traffic
1083 * before it can be considered the curr_active_slave - also, we would skip
1084 * slaves between the curr_active_slave and primary_slave that may be up
1085 * and able to arp
1086 */
1087 if ((bond->primary_slave) &&
1088 (!bond->params.arp_interval) &&
1089 (IS_UP(bond->primary_slave->dev))) {
1090 new_active = bond->primary_slave;
1091 }
1092
1093 /* remember where to stop iterating over the slaves */
1094 old_active = new_active;
1095
1096 bond_for_each_slave_from(bond, new_active, i, old_active) {
1097 if (IS_UP(new_active->dev)) {
1098 if (new_active->link == BOND_LINK_UP) {
1099 return new_active;
1100 } else if (new_active->link == BOND_LINK_BACK) {
1101 /* link up, but waiting for stabilization */
1102 if (new_active->delay < mintime) {
1103 mintime = new_active->delay;
1104 bestslave = new_active;
1105 }
1106 }
1107 }
1108 }
1109
1110 return bestslave;
1111}
1112
1113/**
1114 * change_active_interface - change the active slave into the specified one
1115 * @bond: our bonding struct
1116 * @new: the new slave to make the active one
1117 *
1118 * Set the new slave to the bond's settings and unset them on the old
1119 * curr_active_slave.
1120 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1121 *
1122 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1123 * because it is apparently the best available slave we have, even though its
1124 * updelay hasn't timed out yet.
1125 *
3915c1e8
JV		 1126 * If new_active is not NULL, caller must hold bond->lock for read and
1127 * curr_slave_lock for write_bh.
1da177e4 1128 */
a77b5325 1129void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1130{
1131 struct slave *old_active = bond->curr_active_slave;
1132
1133 if (old_active == new_active) {
1134 return;
1135 }
1136
1137 if (new_active) {
b2220cad
JV		 1138 new_active->jiffies = jiffies;
1139
1da177e4
LT		 1140 if (new_active->link == BOND_LINK_BACK) {
1141 if (USES_PRIMARY(bond->params.mode)) {
1142 printk(KERN_INFO DRV_NAME
1143 ": %s: making interface %s the new "
1144 "active one %d ms earlier.\n",
1145 bond->dev->name, new_active->dev->name,
1146 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1147 }
1148
1149 new_active->delay = 0;
1150 new_active->link = BOND_LINK_UP;
1da177e4
LT		 1151
1152 if (bond->params.mode == BOND_MODE_8023AD) {
1153 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1154 }
1155
1156 if ((bond->params.mode == BOND_MODE_TLB) ||
1157 (bond->params.mode == BOND_MODE_ALB)) {
1158 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1159 }
1160 } else {
1161 if (USES_PRIMARY(bond->params.mode)) {
1162 printk(KERN_INFO DRV_NAME
1163 ": %s: making interface %s the new "
1164 "active one.\n",
1165 bond->dev->name, new_active->dev->name);
1166 }
1167 }
1168 }
1169
1da177e4
LT		 1170 if (USES_PRIMARY(bond->params.mode)) {
1171 bond_mc_swap(bond, new_active, old_active);
1172 }
1173
1174 if ((bond->params.mode == BOND_MODE_TLB) ||
1175 (bond->params.mode == BOND_MODE_ALB)) {
1176 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1177 if (old_active)
1178 bond_set_slave_inactive_flags(old_active);
1179 if (new_active)
1180 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1181 } else {
1182 bond->curr_active_slave = new_active;
1183 }
c3ade5ca
JV		 1184
1185 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1186 if (old_active) {
1187 bond_set_slave_inactive_flags(old_active);
1188 }
1189
1190 if (new_active) {
1191 bond_set_slave_active_flags(new_active);
2ab82852 1192
709f8a45
OG		 1193 if (bond->params.fail_over_mac)
1194 bond_do_fail_over_mac(bond, new_active,
1195 old_active);
3915c1e8 1196
709f8a45
OG		 1197 bond->send_grat_arp = bond->params.num_grat_arp;
1198 if (!test_bit(__LINK_STATE_LINKWATCH_PENDING,
1053f62c 1199 &bond->curr_active_slave->dev->state)) {
7893b249
MS		 1200 bond_send_gratuitous_arp(bond);
1201 bond->send_grat_arp--;
709f8a45
OG		 1202 } else {
1203 dprintk("delaying gratuitous arp on %s\n",
1204 bond->curr_active_slave->dev->name);
7893b249
MS		 1205 }
1206 }
c3ade5ca 1207 }
1da177e4
LT		 1208}
1209
1210/**
1211 * bond_select_active_slave - select a new active slave, if needed
1212 * @bond: our bonding struct
1213 *
1214 * This functions shoud be called when one of the following occurs:
1215 * - The old curr_active_slave has been released or lost its link.
1216 * - The primary_slave has got its link back.
1217 * - A slave has got its link back and there's no old curr_active_slave.
1218 *
3915c1e8 1219 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1220 */
a77b5325 1221void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1222{
1223 struct slave *best_slave;
ff59c456 1224 int rv;
1da177e4
LT		 1225
1226 best_slave = bond_find_best_slave(bond);
1227 if (best_slave != bond->curr_active_slave) {
1228 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1229 rv = bond_set_carrier(bond);
1230 if (!rv)
1231 return;
1232
1233 if (netif_carrier_ok(bond->dev)) {
1234 printk(KERN_INFO DRV_NAME
1235 ": %s: first active interface up!\n",
1236 bond->dev->name);
1237 } else {
1238 printk(KERN_INFO DRV_NAME ": %s: "
1239 "now running without any active interface !\n",
1240 bond->dev->name);
1241 }
1da177e4
LT		 1242 }
1243}
1244
1245/*--------------------------- slave list handling ---------------------------*/
1246
1247/*
1248 * This function attaches the slave to the end of list.
1249 *
1250 * bond->lock held for writing by caller.
1251 */
1252static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1253{
1254 if (bond->first_slave == NULL) { /* attaching the first slave */
1255 new_slave->next = new_slave;
1256 new_slave->prev = new_slave;
1257 bond->first_slave = new_slave;
1258 } else {
1259 new_slave->next = bond->first_slave;
1260 new_slave->prev = bond->first_slave->prev;
1261 new_slave->next->prev = new_slave;
1262 new_slave->prev->next = new_slave;
1263 }
1264
1265 bond->slave_cnt++;
1266}
1267
1268/*
1269 * This function detaches the slave from the list.
1270 * WARNING: no check is made to verify if the slave effectively
1271 * belongs to <bond>.
1272 * Nothing is freed on return, structures are just unchained.
1273 * If any slave pointer in bond was pointing to <slave>,
1274 * it should be changed by the calling function.
1275 *
1276 * bond->lock held for writing by caller.
1277 */
1278static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1279{
1280 if (slave->next) {
1281 slave->next->prev = slave->prev;
1282 }
1283
1284 if (slave->prev) {
1285 slave->prev->next = slave->next;
1286 }
1287
1288 if (bond->first_slave == slave) { /* slave is the first slave */
1289 if (bond->slave_cnt > 1) { /* there are more slave */
1290 bond->first_slave = slave->next;
1291 } else {
1292 bond->first_slave = NULL; /* slave was the last one */
1293 }
1294 }
1295
1296 slave->next = NULL;
1297 slave->prev = NULL;
1298 bond->slave_cnt--;
1299}
1300
1301/*---------------------------------- IOCTL ----------------------------------*/
1302
4ad072c9
AB		 1303static int bond_sethwaddr(struct net_device *bond_dev,
1304 struct net_device *slave_dev)
1da177e4
LT		 1305{
1306 dprintk("bond_dev=%p\n", bond_dev);
1307 dprintk("slave_dev=%p\n", slave_dev);
1308 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1309 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1310 return 0;
1311}
1312
7f353bf2
HX		 1313#define BOND_VLAN_FEATURES \
1314 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1315 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK		 1316
1317/*
8e3babcd 1318 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX		 1319 * feature bits are managed elsewhere, so preserve those feature bits
1320 * on the master device.
8531c5ff
AK		 1321 */
1322static int bond_compute_features(struct bonding *bond)
1323{
8531c5ff
AK		 1324 struct slave *slave;
1325 struct net_device *bond_dev = bond->dev;
7f353bf2 1326 unsigned long features = bond_dev->features;
3158bf7d
MS		 1327 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1328 bond_dev->hard_header_len);
8e3babcd 1329 int i;
8531c5ff 1330
7f353bf2
HX		 1331 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1332 features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1333 NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1334
54ef3137 1335 bond_for_each_slave(bond, slave, i) {
7f353bf2
HX		 1336 features = netdev_compute_features(features,
1337 slave->dev->features);
54ef3137
JV		 1338 if (slave->dev->hard_header_len > max_hard_header_len)
1339 max_hard_header_len = slave->dev->hard_header_len;
1340 }
8531c5ff 1341
7f353bf2 1342 features |= (bond_dev->features & BOND_VLAN_FEATURES);
8531c5ff 1343 bond_dev->features = features;
54ef3137 1344 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1345
1346 return 0;
1347}
1348
872254dd
MS		 1349
1350static void bond_setup_by_slave(struct net_device *bond_dev,
1351 struct net_device *slave_dev)
1352{
d90a162a
MS		 1353 struct bonding *bond = bond_dev->priv;
1354
872254dd 1355 bond_dev->neigh_setup = slave_dev->neigh_setup;
1284cd3a 1356 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1357
1358 bond_dev->type = slave_dev->type;
1359 bond_dev->hard_header_len = slave_dev->hard_header_len;
1360 bond_dev->addr_len = slave_dev->addr_len;
1361
1362 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1363 slave_dev->addr_len);
d90a162a 1364 bond->setup_by_slave = 1;
872254dd
MS		 1365}
1366
1da177e4 1367/* enslave device <slave> to bond device <master> */
a77b5325 1368int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1369{
1370 struct bonding *bond = bond_dev->priv;
1371 struct slave *new_slave = NULL;
1372 struct dev_mc_list *dmi;
1373 struct sockaddr addr;
1374 int link_reporting;
1375 int old_features = bond_dev->features;
1376 int res = 0;
1377
552709d5 1378 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1379 slave_dev->do_ioctl == NULL) {
1da177e4 1380 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1381 ": %s: Warning: no link monitoring support for %s\n",
1382 bond_dev->name, slave_dev->name);
1da177e4
LT		 1383 }
1384
1385 /* bond must be initialized by bond_open() before enslaving */
1386 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1387 printk(KERN_WARNING DRV_NAME
1388 " %s: master_dev is not up in bond_enslave\n",
1389 bond_dev->name);
1da177e4
LT		 1390 }
1391
1392 /* already enslaved */
1393 if (slave_dev->flags & IFF_SLAVE) {
1394 dprintk("Error, Device was already enslaved\n");
1395 return -EBUSY;
1396 }
1397
1398 /* vlan challenged mutual exclusion */
1399 /* no need to lock since we're protected by rtnl_lock */
1400 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1401 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1402 if (!list_empty(&bond->vlan_list)) {
1403 printk(KERN_ERR DRV_NAME
4e0952c7 1404 ": %s: Error: cannot enslave VLAN "
1da177e4 1405 "challenged slave %s on VLAN enabled "
4e0952c7 1406 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1407 bond_dev->name);
1408 return -EPERM;
1409 } else {
1410 printk(KERN_WARNING DRV_NAME
4e0952c7 1411 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1412 "slave %s. Adding VLANs will be blocked as "
1413 "long as %s is part of bond %s\n",
4e0952c7 1414 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1415 bond_dev->name);
1416 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1417 }
1418 } else {
1419 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1420 if (bond->slave_cnt == 0) {
1421 /* First slave, and it is not VLAN challenged,
1422 * so remove the block of adding VLANs over the bond.
1423 */
1424 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1425 }
1426 }
1427
217df670
JV		 1428 /*
1429 * Old ifenslave binaries are no longer supported. These can
1430 * be identified with moderate accurary by the state of the slave:
1431 * the current ifenslave will set the interface down prior to
1432 * enslaving it; the old ifenslave will not.
1433 */
1434 if ((slave_dev->flags & IFF_UP)) {
1435 printk(KERN_ERR DRV_NAME ": %s is up. "
1436 "This may be due to an out of date ifenslave.\n",
1437 slave_dev->name);
1438 res = -EPERM;
1439 goto err_undo_flags;
1440 }
1da177e4 1441
872254dd
MS		 1442 /* set bonding device ether type by slave - bonding netdevices are
1443 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1444 * there is a need to override some of the type dependent attribs/funcs.
1445 *
1446 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1447 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1448 */
1449 if (bond->slave_cnt == 0) {
1450 if (slave_dev->type != ARPHRD_ETHER)
1451 bond_setup_by_slave(bond_dev, slave_dev);
1452 } else if (bond_dev->type != slave_dev->type) {
1453 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1454 "from other slaves (%d), can not enslave it.\n",
1455 slave_dev->name,
1456 slave_dev->type, bond_dev->type);
1457 res = -EINVAL;
1458 goto err_undo_flags;
1459 }
1460
217df670 1461 if (slave_dev->set_mac_address == NULL) {
2ab82852
MS		 1462 if (bond->slave_cnt == 0) {
1463 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1464 ": %s: Warning: The first slave device "
1465 "specified does not support setting the MAC "
3915c1e8 1466 "address. Setting fail_over_mac to active.",
dd957c57 1467 bond_dev->name);
3915c1e8
JV		 1468 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1469 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1470 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1471 ": %s: Error: The slave device specified "
1472 "does not support setting the MAC address, "
3915c1e8 1473 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1474 , bond_dev->name);
1475 res = -EOPNOTSUPP;
1476 goto err_undo_flags;
1477 }
1da177e4
LT		 1478 }
1479
243cb4e5 1480 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1481 if (!new_slave) {
1482 res = -ENOMEM;
1483 goto err_undo_flags;
1484 }
1485
1da177e4
LT		 1486 /* save slave's original flags before calling
1487 * netdev_set_master and dev_open
1488 */
1489 new_slave->original_flags = slave_dev->flags;
1490
217df670
JV		 1491 /*
1492 * Save slave's original ("permanent") mac address for modes
1493 * that need it, and for restoring it upon release, and then
1494 * set it to the master's address
1495 */
1496 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1497
dd957c57 1498 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1499 /*
1500 * Set slave to master's mac address. The application already
1501 * set the master's mac address to that of the first slave
1502 */
1503 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1504 addr.sa_family = slave_dev->type;
1505 res = dev_set_mac_address(slave_dev, &addr);
1506 if (res) {
1507 dprintk("Error %d calling set_mac_address\n", res);
1508 goto err_free;
1509 }
217df670 1510 }
1da177e4 1511
c2edacf8
JV		 1512 res = netdev_set_master(slave_dev, bond_dev);
1513 if (res) {
1514 dprintk("Error %d calling netdev_set_master\n", res);
569f0c4d 1515 goto err_restore_mac;
c2edacf8 1516 }
217df670
JV		 1517 /* open the slave since the application closed it */
1518 res = dev_open(slave_dev);
1519 if (res) {
1520 dprintk("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1521 goto err_unset_master;
1da177e4
LT		 1522 }
1523
1da177e4 1524 new_slave->dev = slave_dev;
0b680e75 1525 slave_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 1526
1527 if ((bond->params.mode == BOND_MODE_TLB) ||
1528 (bond->params.mode == BOND_MODE_ALB)) {
1529 /* bond_alb_init_slave() must be called before all other stages since
1530 * it might fail and we do not want to have to undo everything
1531 */
1532 res = bond_alb_init_slave(bond, new_slave);
1533 if (res) {
569f0c4d 1534 goto err_close;
1da177e4
LT		 1535 }
1536 }
1537
1538 /* If the mode USES_PRIMARY, then the new slave gets the
1539 * master's promisc (and mc) settings only if it becomes the
1540 * curr_active_slave, and that is taken care of later when calling
1541 * bond_change_active()
1542 */
1543 if (!USES_PRIMARY(bond->params.mode)) {
1544 /* set promiscuity level to new slave */
1545 if (bond_dev->flags & IFF_PROMISC) {
1546 dev_set_promiscuity(slave_dev, 1);
1547 }
1548
1549 /* set allmulti level to new slave */
1550 if (bond_dev->flags & IFF_ALLMULTI) {
1551 dev_set_allmulti(slave_dev, 1);
1552 }
1553
80ee5ad2 1554 netif_tx_lock_bh(bond_dev);
1da177e4
LT		 1555 /* upload master's mc_list to new slave */
1556 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1557 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1558 }
80ee5ad2 1559 netif_tx_unlock_bh(bond_dev);
1da177e4
LT		 1560 }
1561
1562 if (bond->params.mode == BOND_MODE_8023AD) {
1563 /* add lacpdu mc addr to mc list */
1564 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1565
1566 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1567 }
1568
1569 bond_add_vlans_on_slave(bond, slave_dev);
1570
1571 write_lock_bh(&bond->lock);
1572
1573 bond_attach_slave(bond, new_slave);
1574
1575 new_slave->delay = 0;
1576 new_slave->link_failure_count = 0;
1577
8531c5ff
AK		 1578 bond_compute_features(bond);
1579
3915c1e8
JV		 1580 write_unlock_bh(&bond->lock);
1581
1582 read_lock(&bond->lock);
1583
f5b2b966
JV		 1584 new_slave->last_arp_rx = jiffies;
1585
1da177e4
LT		 1586 if (bond->params.miimon && !bond->params.use_carrier) {
1587 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1588
1589 if ((link_reporting == -1) && !bond->params.arp_interval) {
1590 /*
1591 * miimon is set but a bonded network driver
1592 * does not support ETHTOOL/MII and
1593 * arp_interval is not set. Note: if
1594 * use_carrier is enabled, we will never go
1595 * here (because netif_carrier is always
1596 * supported); thus, we don't need to change
1597 * the messages for netif_carrier.
1598 */
1599 printk(KERN_WARNING DRV_NAME
4e0952c7 1600 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1601 "available for interface %s, and "
1602 "arp_interval/arp_ip_target module parameters "
1603 "not specified, thus bonding will not detect "
1604 "link failures! see bonding.txt for details.\n",
4e0952c7 1605 bond_dev->name, slave_dev->name);
1da177e4
LT		 1606 } else if (link_reporting == -1) {
1607 /* unable get link status using mii/ethtool */
1608 printk(KERN_WARNING DRV_NAME
4e0952c7 1609 ": %s: Warning: can't get link status from "
1da177e4
LT		 1610 "interface %s; the network driver associated "
1611 "with this interface does not support MII or "
1612 "ETHTOOL link status reporting, thus miimon "
1613 "has no effect on this interface.\n",
4e0952c7 1614 bond_dev->name, slave_dev->name);
1da177e4
LT		 1615 }
1616 }
1617
1618 /* check for initial state */
1619 if (!bond->params.miimon ||
1620 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1621 if (bond->params.updelay) {
1622 dprintk("Initial state of slave_dev is "
1623 "BOND_LINK_BACK\n");
1624 new_slave->link = BOND_LINK_BACK;
1625 new_slave->delay = bond->params.updelay;
1626 } else {
1627 dprintk("Initial state of slave_dev is "
1628 "BOND_LINK_UP\n");
1629 new_slave->link = BOND_LINK_UP;
1630 }
1631 new_slave->jiffies = jiffies;
1632 } else {
1633 dprintk("Initial state of slave_dev is "
1634 "BOND_LINK_DOWN\n");
1635 new_slave->link = BOND_LINK_DOWN;
1636 }
1637
1638 if (bond_update_speed_duplex(new_slave) &&
1639 (new_slave->link != BOND_LINK_DOWN)) {
1640 printk(KERN_WARNING DRV_NAME
4e0952c7 1641 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1642 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1643 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1644
1645 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1646 printk(KERN_WARNING DRV_NAME
1647 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1648 "support in base driver for proper aggregator "
4e0952c7 1649 "selection.\n", bond_dev->name);
1da177e4
LT		 1650 }
1651 }
1652
1653 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1654 /* if there is a primary slave, remember it */
1655 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1656 bond->primary_slave = new_slave;
1657 }
1658 }
1659
3915c1e8
JV		 1660 write_lock_bh(&bond->curr_slave_lock);
1661
1da177e4
LT		 1662 switch (bond->params.mode) {
1663 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1664 bond_set_slave_inactive_flags(new_slave);
1665 bond_select_active_slave(bond);
1da177e4
LT		 1666 break;
1667 case BOND_MODE_8023AD:
1668 /* in 802.3ad mode, the internal mechanism
1669 * will activate the slaves in the selected
1670 * aggregator
1671 */
1672 bond_set_slave_inactive_flags(new_slave);
1673 /* if this is the first slave */
1674 if (bond->slave_cnt == 1) {
1675 SLAVE_AD_INFO(new_slave).id = 1;
1676 /* Initialize AD with the number of times that the AD timer is called in 1 second
1677 * can be called only after the mac address of the bond is set
1678 */
1679 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1680 bond->params.lacp_fast);
1681 } else {
1682 SLAVE_AD_INFO(new_slave).id =
1683 SLAVE_AD_INFO(new_slave->prev).id + 1;
1684 }
1685
1686 bond_3ad_bind_slave(new_slave);
1687 break;
1688 case BOND_MODE_TLB:
1689 case BOND_MODE_ALB:
1690 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1691 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1692 break;
1693 default:
1694 dprintk("This slave is always active in trunk mode\n");
1695
1696 /* always active in trunk mode */
1697 new_slave->state = BOND_STATE_ACTIVE;
1698
1699 /* In trunking mode there is little meaning to curr_active_slave
1700 * anyway (it holds no special properties of the bond device),
1701 * so we can change it without calling change_active_interface()
1702 */
1703 if (!bond->curr_active_slave) {
1704 bond->curr_active_slave = new_slave;
1705 }
1706 break;
1707 } /* switch(bond_mode) */
1708
3915c1e8
JV		 1709 write_unlock_bh(&bond->curr_slave_lock);
1710
ff59c456
JV		 1711 bond_set_carrier(bond);
1712
3915c1e8 1713 read_unlock(&bond->lock);
1da177e4 1714
b76cdba9
MW		 1715 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1716 if (res)
569f0c4d 1717 goto err_close;
b76cdba9 1718
1da177e4
LT		 1719 printk(KERN_INFO DRV_NAME
1720 ": %s: enslaving %s as a%s interface with a%s link.\n",
1721 bond_dev->name, slave_dev->name,
1722 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1723 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1724
1725 /* enslave is successful */
1726 return 0;
1727
1728/* Undo stages on error */
1da177e4
LT		 1729err_close:
1730 dev_close(slave_dev);
1731
569f0c4d
JV		 1732err_unset_master:
1733 netdev_set_master(slave_dev, NULL);
1734
1da177e4 1735err_restore_mac:
dd957c57 1736 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1737 /* XXX TODO - fom follow mode needs to change master's
1738 * MAC if this slave's MAC is in use by the bond, or at
1739 * least print a warning.
1740 */
2ab82852
MS		 1741 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1742 addr.sa_family = slave_dev->type;
1743 dev_set_mac_address(slave_dev, &addr);
1744 }
1da177e4
LT		 1745
1746err_free:
1747 kfree(new_slave);
1748
1749err_undo_flags:
1750 bond_dev->features = old_features;
8531c5ff 1751
1da177e4
LT		 1752 return res;
1753}
1754
1755/*
1756 * Try to release the slave device <slave> from the bond device <master>
1757 * It is legal to access curr_active_slave without a lock because all the function
1758 * is write-locked.
1759 *
1760 * The rules for slave state should be:
1761 * for Active/Backup:
1762 * Active stays on all backups go down
1763 * for Bonded connections:
1764 * The first up interface should be left on and all others downed.
1765 */
a77b5325 1766int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1767{
1768 struct bonding *bond = bond_dev->priv;
1769 struct slave *slave, *oldcurrent;
1770 struct sockaddr addr;
1771 int mac_addr_differ;
0795af57 1772 DECLARE_MAC_BUF(mac);
1da177e4
LT		 1773
1774 /* slave is not a slave or master is not master of this slave */
1775 if (!(slave_dev->flags & IFF_SLAVE) ||
1776 (slave_dev->master != bond_dev)) {
1777 printk(KERN_ERR DRV_NAME
4e0952c7 1778 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1779 bond_dev->name, slave_dev->name);
1780 return -EINVAL;
1781 }
1782
1783 write_lock_bh(&bond->lock);
1784
1785 slave = bond_get_slave_by_dev(bond, slave_dev);
1786 if (!slave) {
1787 /* not a slave of this bond */
1788 printk(KERN_INFO DRV_NAME
1789 ": %s: %s not enslaved\n",
1790 bond_dev->name, slave_dev->name);
f5e2a7b2 1791 write_unlock_bh(&bond->lock);
1da177e4
LT		 1792 return -EINVAL;
1793 }
1794
3915c1e8
JV		 1795 if (!bond->params.fail_over_mac) {
1796 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1797 ETH_ALEN);
1798 if (!mac_addr_differ && (bond->slave_cnt > 1))
1799 printk(KERN_WARNING DRV_NAME
1800 ": %s: Warning: the permanent HWaddr of %s - "
1801 "%s - is still in use by %s. "
1802 "Set the HWaddr of %s to a different address "
1803 "to avoid conflicts.\n",
1804 bond_dev->name, slave_dev->name,
1805 print_mac(mac, slave->perm_hwaddr),
1806 bond_dev->name, slave_dev->name);
1da177e4
LT		 1807 }
1808
1809 /* Inform AD package of unbinding of slave. */
1810 if (bond->params.mode == BOND_MODE_8023AD) {
1811 /* must be called before the slave is
1812 * detached from the list
1813 */
1814 bond_3ad_unbind_slave(slave);
1815 }
1816
1817 printk(KERN_INFO DRV_NAME
1818 ": %s: releasing %s interface %s\n",
1819 bond_dev->name,
1820 (slave->state == BOND_STATE_ACTIVE)
1821 ? "active" : "backup",
1822 slave_dev->name);
1823
1824 oldcurrent = bond->curr_active_slave;
1825
1826 bond->current_arp_slave = NULL;
1827
1828 /* release the slave from its bond */
1829 bond_detach_slave(bond, slave);
1830
8531c5ff
AK		 1831 bond_compute_features(bond);
1832
1da177e4
LT		 1833 if (bond->primary_slave == slave) {
1834 bond->primary_slave = NULL;
1835 }
1836
1837 if (oldcurrent == slave) {
1838 bond_change_active_slave(bond, NULL);
1839 }
1840
1841 if ((bond->params.mode == BOND_MODE_TLB) ||
1842 (bond->params.mode == BOND_MODE_ALB)) {
1843 /* Must be called only after the slave has been
1844 * detached from the list and the curr_active_slave
1845 * has been cleared (if our_slave == old_current),
1846 * but before a new active slave is selected.
1847 */
2543331d 1848 write_unlock_bh(&bond->lock);
1da177e4 1849 bond_alb_deinit_slave(bond, slave);
2543331d 1850 write_lock_bh(&bond->lock);
1da177e4
LT		 1851 }
1852
059fe7a5
JV		 1853 if (oldcurrent == slave) {
1854 /*
1855 * Note that we hold RTNL over this sequence, so there
1856 * is no concern that another slave add/remove event
1857 * will interfere.
1858 */
1859 write_unlock_bh(&bond->lock);
1860 read_lock(&bond->lock);
1861 write_lock_bh(&bond->curr_slave_lock);
1862
1da177e4
LT		 1863 bond_select_active_slave(bond);
1864
059fe7a5
JV		 1865 write_unlock_bh(&bond->curr_slave_lock);
1866 read_unlock(&bond->lock);
1867 write_lock_bh(&bond->lock);
1868 }
1869
1da177e4 1870 if (bond->slave_cnt == 0) {
ff59c456
JV		 1871 bond_set_carrier(bond);
1872
1da177e4
LT		 1873 /* if the last slave was removed, zero the mac address
1874 * of the master so it will be set by the application
1875 * to the mac address of the first slave
1876 */
1877 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1878
1879 if (list_empty(&bond->vlan_list)) {
1880 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1881 } else {
1882 printk(KERN_WARNING DRV_NAME
4e0952c7 1883 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1884 "still has VLANs.\n",
4e0952c7 1885 bond_dev->name, bond_dev->name);
1da177e4 1886 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1887 ": %s: When re-adding slaves, make sure the bond's "
1888 "HW address matches its VLANs'.\n",
1889 bond_dev->name);
1da177e4
LT		 1890 }
1891 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1892 !bond_has_challenged_slaves(bond)) {
1893 printk(KERN_INFO DRV_NAME
4e0952c7 1894 ": %s: last VLAN challenged slave %s "
1da177e4 1895 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1896 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1897 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1898 }
1899
1900 write_unlock_bh(&bond->lock);
1901
b76cdba9
MW		 1902 /* must do this from outside any spinlocks */
1903 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1904
1da177e4
LT		 1905 bond_del_vlans_from_slave(bond, slave_dev);
1906
1907 /* If the mode USES_PRIMARY, then we should only remove its
1908 * promisc and mc settings if it was the curr_active_slave, but that was
1909 * already taken care of above when we detached the slave
1910 */
1911 if (!USES_PRIMARY(bond->params.mode)) {
1912 /* unset promiscuity level from slave */
1913 if (bond_dev->flags & IFF_PROMISC) {
1914 dev_set_promiscuity(slave_dev, -1);
1915 }
1916
1917 /* unset allmulti level from slave */
1918 if (bond_dev->flags & IFF_ALLMULTI) {
1919 dev_set_allmulti(slave_dev, -1);
1920 }
1921
1922 /* flush master's mc_list from slave */
80ee5ad2 1923 netif_tx_lock_bh(bond_dev);
1da177e4 1924 bond_mc_list_flush(bond_dev, slave_dev);
80ee5ad2 1925 netif_tx_unlock_bh(bond_dev);
1da177e4
LT		 1926 }
1927
1928 netdev_set_master(slave_dev, NULL);
1929
1930 /* close slave before restoring its mac address */
1931 dev_close(slave_dev);
1932
3915c1e8 1933 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1934 /* restore original ("permanent") mac address */
1935 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1936 addr.sa_family = slave_dev->type;
1937 dev_set_mac_address(slave_dev, &addr);
1938 }
1da177e4 1939
8f903c70 1940 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1941 IFF_SLAVE_INACTIVE | IFF_BONDING |
1942 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1943
1944 kfree(slave);
1945
1946 return 0; /* deletion OK */
1947}
1948
d90a162a
MS		 1949/*
1950* Destroy a bonding device.
1951* Must be under rtnl_lock when this function is called.
1952*/
1953void bond_destroy(struct bonding *bond)
1954{
1955 bond_deinit(bond->dev);
1956 bond_destroy_sysfs_entry(bond);
8cbdeec6 1957 unregister_netdevice(bond->dev);
d90a162a
MS		 1958}
1959
1960/*
1961* First release a slave and than destroy the bond if no more slaves iare left.
1962* Must be under rtnl_lock when this function is called.
1963*/
1964int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1965{
1966 struct bonding *bond = bond_dev->priv;
1967 int ret;
1968
1969 ret = bond_release(bond_dev, slave_dev);
1970 if ((ret == 0) && (bond->slave_cnt == 0)) {
1971 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1972 bond_dev->name, bond_dev->name);
1973 bond_destroy(bond);
1974 }
1975 return ret;
1976}
1977
1da177e4
LT		 1978/*
1979 * This function releases all slaves.
1980 */
1981static int bond_release_all(struct net_device *bond_dev)
1982{
1983 struct bonding *bond = bond_dev->priv;
1984 struct slave *slave;
1985 struct net_device *slave_dev;
1986 struct sockaddr addr;
1987
1988 write_lock_bh(&bond->lock);
1989
ff59c456
JV		 1990 netif_carrier_off(bond_dev);
1991
1da177e4
LT		 1992 if (bond->slave_cnt == 0) {
1993 goto out;
1994 }
1995
1996 bond->current_arp_slave = NULL;
1997 bond->primary_slave = NULL;
1998 bond_change_active_slave(bond, NULL);
1999
2000 while ((slave = bond->first_slave) != NULL) {
2001 /* Inform AD package of unbinding of slave
2002 * before slave is detached from the list.
2003 */
2004 if (bond->params.mode == BOND_MODE_8023AD) {
2005 bond_3ad_unbind_slave(slave);
2006 }
2007
2008 slave_dev = slave->dev;
2009 bond_detach_slave(bond, slave);
2010
2543331d
JV		 2011 /* now that the slave is detached, unlock and perform
2012 * all the undo steps that should not be called from
2013 * within a lock.
2014 */
2015 write_unlock_bh(&bond->lock);
2016
1da177e4
LT		 2017 if ((bond->params.mode == BOND_MODE_TLB) ||
2018 (bond->params.mode == BOND_MODE_ALB)) {
2019 /* must be called only after the slave
2020 * has been detached from the list
2021 */
2022 bond_alb_deinit_slave(bond, slave);
2023 }
2024
8531c5ff
AK		 2025 bond_compute_features(bond);
2026
b76cdba9 2027 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2028 bond_del_vlans_from_slave(bond, slave_dev);
2029
2030 /* If the mode USES_PRIMARY, then we should only remove its
2031 * promisc and mc settings if it was the curr_active_slave, but that was
2032 * already taken care of above when we detached the slave
2033 */
2034 if (!USES_PRIMARY(bond->params.mode)) {
2035 /* unset promiscuity level from slave */
2036 if (bond_dev->flags & IFF_PROMISC) {
2037 dev_set_promiscuity(slave_dev, -1);
2038 }
2039
2040 /* unset allmulti level from slave */
2041 if (bond_dev->flags & IFF_ALLMULTI) {
2042 dev_set_allmulti(slave_dev, -1);
2043 }
2044
2045 /* flush master's mc_list from slave */
80ee5ad2 2046 netif_tx_lock_bh(bond_dev);
1da177e4 2047 bond_mc_list_flush(bond_dev, slave_dev);
80ee5ad2 2048 netif_tx_unlock_bh(bond_dev);
1da177e4
LT		 2049 }
2050
2051 netdev_set_master(slave_dev, NULL);
2052
2053 /* close slave before restoring its mac address */
2054 dev_close(slave_dev);
2055
dd957c57 2056 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2057 /* restore original ("permanent") mac address*/
2058 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2059 addr.sa_family = slave_dev->type;
2060 dev_set_mac_address(slave_dev, &addr);
2061 }
1da177e4 2062
8f903c70
JV		 2063 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2064 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2065
2066 kfree(slave);
2067
2068 /* re-acquire the lock before getting the next slave */
2069 write_lock_bh(&bond->lock);
2070 }
2071
2072 /* zero the mac address of the master so it will be
2073 * set by the application to the mac address of the
2074 * first slave
2075 */
2076 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2077
2078 if (list_empty(&bond->vlan_list)) {
2079 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2080 } else {
2081 printk(KERN_WARNING DRV_NAME
4e0952c7 2082 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2083 "still has VLANs.\n",
4e0952c7 2084 bond_dev->name, bond_dev->name);
1da177e4 2085 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2086 ": %s: When re-adding slaves, make sure the bond's "
2087 "HW address matches its VLANs'.\n",
2088 bond_dev->name);
1da177e4
LT		 2089 }
2090
2091 printk(KERN_INFO DRV_NAME
2092 ": %s: released all slaves\n",
2093 bond_dev->name);
2094
2095out:
2096 write_unlock_bh(&bond->lock);
2097
2098 return 0;
2099}
2100
2101/*
2102 * This function changes the active slave to slave <slave_dev>.
2103 * It returns -EINVAL in the following cases.
2104 * - <slave_dev> is not found in the list.
2105 * - There is not active slave now.
2106 * - <slave_dev> is already active.
2107 * - The link state of <slave_dev> is not BOND_LINK_UP.
2108 * - <slave_dev> is not running.
2109 * In these cases, this fuction does nothing.
2110 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2111 */
2112static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2113{
2114 struct bonding *bond = bond_dev->priv;
2115 struct slave *old_active = NULL;
2116 struct slave *new_active = NULL;
2117 int res = 0;
2118
2119 if (!USES_PRIMARY(bond->params.mode)) {
2120 return -EINVAL;
2121 }
2122
2123 /* Verify that master_dev is indeed the master of slave_dev */
2124 if (!(slave_dev->flags & IFF_SLAVE) ||
2125 (slave_dev->master != bond_dev)) {
2126 return -EINVAL;
2127 }
2128
059fe7a5 2129 read_lock(&bond->lock);
1da177e4 2130
059fe7a5 2131 read_lock(&bond->curr_slave_lock);
1da177e4 2132 old_active = bond->curr_active_slave;
059fe7a5
JV		 2133 read_unlock(&bond->curr_slave_lock);
2134
1da177e4
LT		 2135 new_active = bond_get_slave_by_dev(bond, slave_dev);
2136
2137 /*
2138 * Changing to the current active: do nothing; return success.
2139 */
2140 if (new_active && (new_active == old_active)) {
059fe7a5 2141 read_unlock(&bond->lock);
1da177e4
LT		 2142 return 0;
2143 }
2144
2145 if ((new_active) &&
2146 (old_active) &&
2147 (new_active->link == BOND_LINK_UP) &&
2148 IS_UP(new_active->dev)) {
059fe7a5 2149 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2150 bond_change_active_slave(bond, new_active);
059fe7a5 2151 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2152 } else {
2153 res = -EINVAL;
2154 }
2155
059fe7a5 2156 read_unlock(&bond->lock);
1da177e4
LT		 2157
2158 return res;
2159}
2160
1da177e4
LT		 2161static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2162{
2163 struct bonding *bond = bond_dev->priv;
2164
2165 info->bond_mode = bond->params.mode;
2166 info->miimon = bond->params.miimon;
2167
6603a6f2 2168 read_lock(&bond->lock);
1da177e4 2169 info->num_slaves = bond->slave_cnt;
6603a6f2 2170 read_unlock(&bond->lock);
1da177e4
LT		 2171
2172 return 0;
2173}
2174
2175static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2176{
2177 struct bonding *bond = bond_dev->priv;
2178 struct slave *slave;
2179 int i, found = 0;
2180
2181 if (info->slave_id < 0) {
2182 return -ENODEV;
2183 }
2184
6603a6f2 2185 read_lock(&bond->lock);
1da177e4
LT		 2186
2187 bond_for_each_slave(bond, slave, i) {
2188 if (i == (int)info->slave_id) {
2189 found = 1;
2190 break;
2191 }
2192 }
2193
6603a6f2 2194 read_unlock(&bond->lock);
1da177e4
LT		 2195
2196 if (found) {
2197 strcpy(info->slave_name, slave->dev->name);
2198 info->link = slave->link;
2199 info->state = slave->state;
2200 info->link_failure_count = slave->link_failure_count;
2201 } else {
2202 return -ENODEV;
2203 }
2204
2205 return 0;
2206}
2207
2208/*-------------------------------- Monitoring -------------------------------*/
2209
0b0eef66
JV		 2210/*
2211 * if !have_locks, return nonzero if a failover is necessary. if
2212 * have_locks, do whatever failover activities are needed.
2213 *
2214 * This is to separate the inspection and failover steps for locking
2215 * purposes; failover requires rtnl, but acquiring it for every
2216 * inspection is undesirable, so a wrapper first does inspection, and
2217 * the acquires the necessary locks and calls again to perform
2218 * failover if needed. Since all locks are dropped, a complete
2219 * restart is needed between calls.
2220 */
2221static int __bond_mii_monitor(struct bonding *bond, int have_locks)
1da177e4 2222{
1da177e4
LT		 2223 struct slave *slave, *oldcurrent;
2224 int do_failover = 0;
1da177e4
LT		 2225 int i;
2226
0b0eef66 2227 if (bond->slave_cnt == 0)
1da177e4 2228 goto out;
1da177e4
LT		 2229
2230 /* we will try to read the link status of each of our slaves, and
2231 * set their IFF_RUNNING flag appropriately. For each slave not
2232 * supporting MII status, we won't do anything so that a user-space
2233 * program could monitor the link itself if needed.
2234 */
2235
1053f62c
MS		 2236 if (bond->send_grat_arp) {
2237 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2238 &bond->curr_active_slave->dev->state))
2239 dprintk("Needs to send gratuitous arp but not yet\n");
2240 else {
2241 dprintk("sending delayed gratuitous arp on on %s\n",
2242 bond->curr_active_slave->dev->name);
2243 bond_send_gratuitous_arp(bond);
7893b249 2244 bond->send_grat_arp--;
1053f62c
MS		 2245 }
2246 }
1da177e4
LT		 2247 read_lock(&bond->curr_slave_lock);
2248 oldcurrent = bond->curr_active_slave;
2249 read_unlock(&bond->curr_slave_lock);
2250
2251 bond_for_each_slave(bond, slave, i) {
2252 struct net_device *slave_dev = slave->dev;
2253 int link_state;
2254 u16 old_speed = slave->speed;
2255 u8 old_duplex = slave->duplex;
2256
2257 link_state = bond_check_dev_link(bond, slave_dev, 0);
2258
2259 switch (slave->link) {
2260 case BOND_LINK_UP: /* the link was up */
2261 if (link_state == BMSR_LSTATUS) {
059fe7a5
JV		 2262 if (!oldcurrent) {
2263 if (!have_locks)
2264 return 1;
2265 do_failover = 1;
2266 }
1da177e4
LT		 2267 break;
2268 } else { /* link going down */
2269 slave->link = BOND_LINK_FAIL;
2270 slave->delay = bond->params.downdelay;
2271
2272 if (slave->link_failure_count < UINT_MAX) {
2273 slave->link_failure_count++;
2274 }
2275
2276 if (bond->params.downdelay) {
2277 printk(KERN_INFO DRV_NAME
2278 ": %s: link status down for %s "
2279 "interface %s, disabling it in "
2280 "%d ms.\n",
1b76b316 2281 bond->dev->name,
1da177e4
LT		 2282 IS_UP(slave_dev)
2283 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2284 ? ((slave == oldcurrent)
2285 ? "active " : "backup ")
2286 : "")
2287 : "idle ",
2288 slave_dev->name,
2289 bond->params.downdelay * bond->params.miimon);
2290 }
2291 }
2292 /* no break ! fall through the BOND_LINK_FAIL test to
2293 ensure proper action to be taken
2294 */
2295 case BOND_LINK_FAIL: /* the link has just gone down */
2296 if (link_state != BMSR_LSTATUS) {
2297 /* link stays down */
2298 if (slave->delay <= 0) {
0b0eef66
JV		 2299 if (!have_locks)
2300 return 1;
2301
1da177e4
LT		 2302 /* link down for too long time */
2303 slave->link = BOND_LINK_DOWN;
2304
2305 /* in active/backup mode, we must
2306 * completely disable this interface
2307 */
2308 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2309 (bond->params.mode == BOND_MODE_8023AD)) {
2310 bond_set_slave_inactive_flags(slave);
2311 }
2312
2313 printk(KERN_INFO DRV_NAME
2314 ": %s: link status definitely "
2315 "down for interface %s, "
2316 "disabling it\n",
1b76b316 2317 bond->dev->name,
1da177e4
LT		 2318 slave_dev->name);
2319
2320 /* notify ad that the link status has changed */
2321 if (bond->params.mode == BOND_MODE_8023AD) {
2322 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2323 }
2324
2325 if ((bond->params.mode == BOND_MODE_TLB) ||
2326 (bond->params.mode == BOND_MODE_ALB)) {
2327 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2328 }
2329
2330 if (slave == oldcurrent) {
2331 do_failover = 1;
2332 }
2333 } else {
2334 slave->delay--;
2335 }
2336 } else {
2337 /* link up again */
2338 slave->link = BOND_LINK_UP;
2339 slave->jiffies = jiffies;
2340 printk(KERN_INFO DRV_NAME
2341 ": %s: link status up again after %d "
2342 "ms for interface %s.\n",
1b76b316 2343 bond->dev->name,
1da177e4
LT		 2344 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2345 slave_dev->name);
2346 }
2347 break;
2348 case BOND_LINK_DOWN: /* the link was down */
2349 if (link_state != BMSR_LSTATUS) {
2350 /* the link stays down, nothing more to do */
2351 break;
2352 } else { /* link going up */
2353 slave->link = BOND_LINK_BACK;
2354 slave->delay = bond->params.updelay;
2355
2356 if (bond->params.updelay) {
2357 /* if updelay == 0, no need to
2358 advertise about a 0 ms delay */
2359 printk(KERN_INFO DRV_NAME
2360 ": %s: link status up for "
2361 "interface %s, enabling it "
2362 "in %d ms.\n",
1b76b316 2363 bond->dev->name,
1da177e4
LT		 2364 slave_dev->name,
2365 bond->params.updelay * bond->params.miimon);
2366 }
2367 }
2368 /* no break ! fall through the BOND_LINK_BACK state in
2369 case there's something to do.
2370 */
2371 case BOND_LINK_BACK: /* the link has just come back */
2372 if (link_state != BMSR_LSTATUS) {
2373 /* link down again */
2374 slave->link = BOND_LINK_DOWN;
2375
2376 printk(KERN_INFO DRV_NAME
2377 ": %s: link status down again after %d "
2378 "ms for interface %s.\n",
1b76b316 2379 bond->dev->name,
1da177e4
LT		 2380 (bond->params.updelay - slave->delay) * bond->params.miimon,
2381 slave_dev->name);
2382 } else {
2383 /* link stays up */
2384 if (slave->delay == 0) {
0b0eef66
JV		 2385 if (!have_locks)
2386 return 1;
2387
1da177e4
LT		 2388 /* now the link has been up for long time enough */
2389 slave->link = BOND_LINK_UP;
2390 slave->jiffies = jiffies;
2391
2392 if (bond->params.mode == BOND_MODE_8023AD) {
2393 /* prevent it from being the active one */
2394 slave->state = BOND_STATE_BACKUP;
2395 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2396 /* make it immediately active */
2397 slave->state = BOND_STATE_ACTIVE;
2398 } else if (slave != bond->primary_slave) {
2399 /* prevent it from being the active one */
2400 slave->state = BOND_STATE_BACKUP;
2401 }
2402
2403 printk(KERN_INFO DRV_NAME
2404 ": %s: link status definitely "
2405 "up for interface %s.\n",
1b76b316 2406 bond->dev->name,
1da177e4
LT		 2407 slave_dev->name);
2408
2409 /* notify ad that the link status has changed */
2410 if (bond->params.mode == BOND_MODE_8023AD) {
2411 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2412 }
2413
2414 if ((bond->params.mode == BOND_MODE_TLB) ||
2415 (bond->params.mode == BOND_MODE_ALB)) {
2416 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2417 }
2418
2419 if ((!oldcurrent) ||
2420 (slave == bond->primary_slave)) {
2421 do_failover = 1;
2422 }
2423 } else {
2424 slave->delay--;
2425 }
2426 }
2427 break;
2428 default:
2429 /* Should not happen */
4e0952c7
MW		 2430 printk(KERN_ERR DRV_NAME
2431 ": %s: Error: %s Illegal value (link=%d)\n",
1b76b316 2432 bond->dev->name,
4e0952c7
MW		 2433 slave->dev->name,
2434 slave->link);
1da177e4
LT		 2435 goto out;
2436 } /* end of switch (slave->link) */
2437
2438 bond_update_speed_duplex(slave);
2439
2440 if (bond->params.mode == BOND_MODE_8023AD) {
2441 if (old_speed != slave->speed) {
2442 bond_3ad_adapter_speed_changed(slave);
2443 }
2444
2445 if (old_duplex != slave->duplex) {
2446 bond_3ad_adapter_duplex_changed(slave);
2447 }
2448 }
2449
2450 } /* end of for */
2451
2452 if (do_failover) {
059fe7a5
JV		 2453 ASSERT_RTNL();
2454
2455 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2456
2457 bond_select_active_slave(bond);
2458
059fe7a5
JV		 2459 write_unlock_bh(&bond->curr_slave_lock);
2460
ff59c456
JV		 2461 } else
2462 bond_set_carrier(bond);
1da177e4 2463
1da177e4 2464out:
0b0eef66 2465 return 0;
1da177e4
LT		 2466}
2467
0b0eef66
JV		 2468/*
2469 * bond_mii_monitor
2470 *
2471 * Really a wrapper that splits the mii monitor into two phases: an
2472 * inspection, then (if inspection indicates something needs to be
2473 * done) an acquisition of appropriate locks followed by another pass
2474 * to implement whatever link state changes are indicated.
2475 */
2476void bond_mii_monitor(struct work_struct *work)
2477{
2478 struct bonding *bond = container_of(work, struct bonding,
2479 mii_work.work);
2480 unsigned long delay;
2481
2482 read_lock(&bond->lock);
2483 if (bond->kill_timers) {
2484 read_unlock(&bond->lock);
2485 return;
2486 }
2487 if (__bond_mii_monitor(bond, 0)) {
2488 read_unlock(&bond->lock);
2489 rtnl_lock();
2490 read_lock(&bond->lock);
2491 __bond_mii_monitor(bond, 1);
5655662d
JV		 2492 read_unlock(&bond->lock);
2493 rtnl_unlock(); /* might sleep, hold no other locks */
2494 read_lock(&bond->lock);
0b0eef66
JV		 2495 }
2496
5ce0da8f 2497 delay = msecs_to_jiffies(bond->params.miimon);
0b0eef66
JV		 2498 read_unlock(&bond->lock);
2499 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2500}
c3ade5ca 2501
d3bb52b0 2502static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2503{
2504 struct in_device *idev;
2505 struct in_ifaddr *ifa;
a144ea4b 2506 __be32 addr = 0;
c3ade5ca
JV		 2507
2508 if (!dev)
2509 return 0;
2510
2511 rcu_read_lock();
e5ed6399 2512 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2513 if (!idev)
2514 goto out;
2515
2516 ifa = idev->ifa_list;
2517 if (!ifa)
2518 goto out;
2519
2520 addr = ifa->ifa_local;
2521out:
2522 rcu_read_unlock();
2523 return addr;
2524}
2525
d3bb52b0 2526static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2527{
0883beca 2528 struct vlan_entry *vlan;
f5b2b966
JV		 2529
2530 if (ip == bond->master_ip)
2531 return 1;
2532
0883beca 2533 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2534 if (ip == vlan->vlan_ip)
2535 return 1;
2536 }
2537
2538 return 0;
2539}
2540
c3ade5ca
JV		 2541/*
2542 * We go to the (large) trouble of VLAN tagging ARP frames because
2543 * switches in VLAN mode (especially if ports are configured as
2544 * "native" to a VLAN) might not pass non-tagged frames.
2545 */
d3bb52b0 2546static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2547{
2548 struct sk_buff *skb;
2549
2550 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2551 slave_dev->name, dest_ip, src_ip, vlan_id);
2552
2553 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2554 NULL, slave_dev->dev_addr, NULL);
2555
2556 if (!skb) {
2557 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2558 return;
2559 }
2560 if (vlan_id) {
2561 skb = vlan_put_tag(skb, vlan_id);
2562 if (!skb) {
2563 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2564 return;
2565 }
2566 }
2567 arp_xmit(skb);
2568}
2569
2570
1da177e4
LT		 2571static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2572{
c3ade5ca 2573 int i, vlan_id, rv;
d3bb52b0 2574 __be32 *targets = bond->params.arp_targets;
0883beca 2575 struct vlan_entry *vlan;
c3ade5ca
JV		 2576 struct net_device *vlan_dev;
2577 struct flowi fl;
2578 struct rtable *rt;
1da177e4 2579
6b780567
MW		 2580 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2581 if (!targets[i])
2582 continue;
c3ade5ca
JV		 2583 dprintk("basa: target %x\n", targets[i]);
2584 if (list_empty(&bond->vlan_list)) {
2585 dprintk("basa: empty vlan: arp_send\n");
2586 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2587 bond->master_ip, 0);
2588 continue;
2589 }
2590
2591 /*
2592 * If VLANs are configured, we do a route lookup to
2593 * determine which VLAN interface would be used, so we
2594 * can tag the ARP with the proper VLAN tag.
2595 */
2596 memset(&fl, 0, sizeof(fl));
2597 fl.fl4_dst = targets[i];
2598 fl.fl4_tos = RTO_ONLINK;
2599
f206351a 2600 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2601 if (rv) {
2602 if (net_ratelimit()) {
2603 printk(KERN_WARNING DRV_NAME
2604 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2605 bond->dev->name, NIPQUAD(fl.fl4_dst));
2606 }
2607 continue;
2608 }
2609
2610 /*
2611 * This target is not on a VLAN
2612 */
2613 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2614 ip_rt_put(rt);
c3ade5ca
JV		 2615 dprintk("basa: rtdev == bond->dev: arp_send\n");
2616 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2617 bond->master_ip, 0);
2618 continue;
2619 }
2620
2621 vlan_id = 0;
0883beca 2622 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2623 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2624 if (vlan_dev == rt->u.dst.dev) {
2625 vlan_id = vlan->vlan_id;
2626 dprintk("basa: vlan match on %s %d\n",
2627 vlan_dev->name, vlan_id);
2628 break;
2629 }
2630 }
2631
2632 if (vlan_id) {
ed4b9f80 2633 ip_rt_put(rt);
c3ade5ca
JV		 2634 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2635 vlan->vlan_ip, vlan_id);
2636 continue;
2637 }
2638
2639 if (net_ratelimit()) {
2640 printk(KERN_WARNING DRV_NAME
2641 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2642 bond->dev->name, NIPQUAD(fl.fl4_dst),
2643 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2644 }
ed4b9f80 2645 ip_rt_put(rt);
c3ade5ca
JV		 2646 }
2647}
2648
2649/*
2650 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2651 * for each VLAN above us.
2652 */
2653static void bond_send_gratuitous_arp(struct bonding *bond)
2654{
2655 struct slave *slave = bond->curr_active_slave;
2656 struct vlan_entry *vlan;
2657 struct net_device *vlan_dev;
2658
2659 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2660 slave ? slave->dev->name : "NULL");
2661 if (!slave)
2662 return;
2663
2664 if (bond->master_ip) {
2665 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2666 bond->master_ip, 0);
c3ade5ca
JV		 2667 }
2668
2669 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2670 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2671 if (vlan->vlan_ip) {
2672 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2673 vlan->vlan_ip, vlan->vlan_id);
2674 }
1da177e4
LT		 2675 }
2676}
2677
d3bb52b0 2678static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2679{
2680 int i;
d3bb52b0 2681 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2682
2683 targets = bond->params.arp_targets;
2684 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2685 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2686 "%u.%u.%u.%u bhti(tip) %d\n",
2687 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2688 bond_has_this_ip(bond, tip));
2689 if (sip == targets[i]) {
2690 if (bond_has_this_ip(bond, tip))
2691 slave->last_arp_rx = jiffies;
2692 return;
2693 }
2694 }
2695}
2696
2697static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2698{
2699 struct arphdr *arp;
2700 struct slave *slave;
2701 struct bonding *bond;
2702 unsigned char *arp_ptr;
d3bb52b0 2703 __be32 sip, tip;
f5b2b966 2704
c346dca1 2705 if (dev_net(dev) != &init_net)
e730c155
EB		 2706 goto out;
2707
f5b2b966
JV		 2708 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2709 goto out;
2710
2711 bond = dev->priv;
2712 read_lock(&bond->lock);
2713
2714 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2715 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2716 orig_dev ? orig_dev->name : "NULL");
2717
2718 slave = bond_get_slave_by_dev(bond, orig_dev);
2719 if (!slave || !slave_do_arp_validate(bond, slave))
2720 goto out_unlock;
2721
988b7050 2722 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2723 goto out_unlock;
2724
d0a92be0 2725 arp = arp_hdr(skb);
f5b2b966
JV		 2726 if (arp->ar_hln != dev->addr_len ||
2727 skb->pkt_type == PACKET_OTHERHOST ||
2728 skb->pkt_type == PACKET_LOOPBACK ||
2729 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2730 arp->ar_pro != htons(ETH_P_IP) ||
2731 arp->ar_pln != 4)
2732 goto out_unlock;
2733
2734 arp_ptr = (unsigned char *)(arp + 1);
2735 arp_ptr += dev->addr_len;
2736 memcpy(&sip, arp_ptr, 4);
2737 arp_ptr += 4 + dev->addr_len;
2738 memcpy(&tip, arp_ptr, 4);
2739
2740 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2741 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2742 slave->state, bond->params.arp_validate,
2743 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2744
2745 /*
2746 * Backup slaves won't see the ARP reply, but do come through
2747 * here for each ARP probe (so we swap the sip/tip to validate
2748 * the probe). In a "redundant switch, common router" type of
2749 * configuration, the ARP probe will (hopefully) travel from
2750 * the active, through one switch, the router, then the other
2751 * switch before reaching the backup.
2752 */
2753 if (slave->state == BOND_STATE_ACTIVE)
2754 bond_validate_arp(bond, slave, sip, tip);
2755 else
2756 bond_validate_arp(bond, slave, tip, sip);
2757
2758out_unlock:
2759 read_unlock(&bond->lock);
2760out:
2761 dev_kfree_skb(skb);
2762 return NET_RX_SUCCESS;
2763}
2764
1da177e4
LT		 2765/*
2766 * this function is called regularly to monitor each slave's link
2767 * ensuring that traffic is being sent and received when arp monitoring
2768 * is used in load-balancing mode. if the adapter has been dormant, then an
2769 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2770 * arp monitoring in active backup mode.
2771 */
1b76b316 2772void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2773{
1b76b316
JV		 2774 struct bonding *bond = container_of(work, struct bonding,
2775 arp_work.work);
1da177e4
LT		 2776 struct slave *slave, *oldcurrent;
2777 int do_failover = 0;
2778 int delta_in_ticks;
2779 int i;
2780
2781 read_lock(&bond->lock);
2782
5ce0da8f 2783 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2784
2785 if (bond->kill_timers) {
2786 goto out;
2787 }
2788
2789 if (bond->slave_cnt == 0) {
2790 goto re_arm;
2791 }
2792
2793 read_lock(&bond->curr_slave_lock);
2794 oldcurrent = bond->curr_active_slave;
2795 read_unlock(&bond->curr_slave_lock);
2796
2797 /* see if any of the previous devices are up now (i.e. they have
2798 * xmt and rcv traffic). the curr_active_slave does not come into
2799 * the picture unless it is null. also, slave->jiffies is not needed
2800 * here because we send an arp on each slave and give a slave as
2801 * long as it needs to get the tx/rx within the delta.
2802 * TODO: what about up/down delay in arp mode? it wasn't here before
2803 * so it can wait
2804 */
2805 bond_for_each_slave(bond, slave, i) {
2806 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2807 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2808 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2809
2810 slave->link = BOND_LINK_UP;
2811 slave->state = BOND_STATE_ACTIVE;
2812
2813 /* primary_slave has no meaning in round-robin
2814 * mode. the window of a slave being up and
2815 * curr_active_slave being null after enslaving
2816 * is closed.
2817 */
2818 if (!oldcurrent) {
2819 printk(KERN_INFO DRV_NAME
2820 ": %s: link status definitely "
2821 "up for interface %s, ",
1b76b316 2822 bond->dev->name,
1da177e4
LT		 2823 slave->dev->name);
2824 do_failover = 1;
2825 } else {
2826 printk(KERN_INFO DRV_NAME
2827 ": %s: interface %s is now up\n",
1b76b316 2828 bond->dev->name,
1da177e4
LT		 2829 slave->dev->name);
2830 }
2831 }
2832 } else {
2833 /* slave->link == BOND_LINK_UP */
2834
2835 /* not all switches will respond to an arp request
2836 * when the source ip is 0, so don't take the link down
2837 * if we don't know our ip yet
2838 */
b63bb739 2839 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2840 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2841
2842 slave->link = BOND_LINK_DOWN;
2843 slave->state = BOND_STATE_BACKUP;
2844
2845 if (slave->link_failure_count < UINT_MAX) {
2846 slave->link_failure_count++;
2847 }
2848
2849 printk(KERN_INFO DRV_NAME
2850 ": %s: interface %s is now down.\n",
1b76b316 2851 bond->dev->name,
1da177e4
LT		 2852 slave->dev->name);
2853
2854 if (slave == oldcurrent) {
2855 do_failover = 1;
2856 }
2857 }
2858 }
2859
2860 /* note: if switch is in round-robin mode, all links
2861 * must tx arp to ensure all links rx an arp - otherwise
2862 * links may oscillate or not come up at all; if switch is
2863 * in something like xor mode, there is nothing we can
2864 * do - all replies will be rx'ed on same link causing slaves
2865 * to be unstable during low/no traffic periods
2866 */
2867 if (IS_UP(slave->dev)) {
2868 bond_arp_send_all(bond, slave);
2869 }
2870 }
2871
2872 if (do_failover) {
059fe7a5 2873 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2874
2875 bond_select_active_slave(bond);
2876
059fe7a5 2877 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2878 }
2879
2880re_arm:
1b76b316
JV		 2881 if (bond->params.arp_interval)
2882 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2883out:
2884 read_unlock(&bond->lock);
2885}
2886
2887/*
b2220cad
JV		 2888 * Called to inspect slaves for active-backup mode ARP monitor link state
2889 * changes. Sets new_link in slaves to specify what action should take
2890 * place for the slave. Returns 0 if no changes are found, >0 if changes
2891 * to link states must be committed.
2892 *
2893 * Called with bond->lock held for read.
1da177e4 2894 */
b2220cad 2895static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2896{
1da177e4 2897 struct slave *slave;
b2220cad 2898 int i, commit = 0;
1da177e4 2899
b2220cad
JV		 2900 bond_for_each_slave(bond, slave, i) {
2901 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2902
b2220cad
JV		 2903 if (slave->link != BOND_LINK_UP) {
2904 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2905 delta_in_ticks)) {
2906 slave->new_link = BOND_LINK_UP;
2907 commit++;
2908 }
1da177e4 2909
b2220cad
JV		 2910 continue;
2911 }
1da177e4 2912
b2220cad
JV		 2913 /*
2914 * Give slaves 2*delta after being enslaved or made
2915 * active. This avoids bouncing, as the last receive
2916 * times need a full ARP monitor cycle to be updated.
2917 */
2918 if (!time_after_eq(jiffies, slave->jiffies +
2919 2 * delta_in_ticks))
2920 continue;
2921
2922 /*
2923 * Backup slave is down if:
2924 * - No current_arp_slave AND
2925 * - more than 3*delta since last receive AND
2926 * - the bond has an IP address
2927 *
2928 * Note: a non-null current_arp_slave indicates
2929 * the curr_active_slave went down and we are
2930 * searching for a new one; under this condition
2931 * we only take the curr_active_slave down - this
2932 * gives each slave a chance to tx/rx traffic
2933 * before being taken out
2934 */
2935 if (slave->state == BOND_STATE_BACKUP &&
2936 !bond->current_arp_slave &&
2937 time_after(jiffies, slave_last_rx(bond, slave) +
2938 3 * delta_in_ticks)) {
2939 slave->new_link = BOND_LINK_DOWN;
2940 commit++;
2941 }
2942
2943 /*
2944 * Active slave is down if:
2945 * - more than 2*delta since transmitting OR
2946 * - (more than 2*delta since receive AND
2947 * the bond has an IP address)
2948 */
2949 if ((slave->state == BOND_STATE_ACTIVE) &&
2950 (time_after_eq(jiffies, slave->dev->trans_start +
2951 2 * delta_in_ticks) ||
2952 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2953 + 2 * delta_in_ticks)))) {
2954 slave->new_link = BOND_LINK_DOWN;
2955 commit++;
2956 }
1da177e4
LT		 2957 }
2958
b2220cad
JV		 2959 read_lock(&bond->curr_slave_lock);
2960
2961 /*
2962 * Trigger a commit if the primary option setting has changed.
1da177e4 2963 */
b2220cad
JV		 2964 if (bond->primary_slave &&
2965 (bond->primary_slave != bond->curr_active_slave) &&
2966 (bond->primary_slave->link == BOND_LINK_UP))
2967 commit++;
1da177e4 2968
b2220cad 2969 read_unlock(&bond->curr_slave_lock);
1da177e4 2970
b2220cad
JV		 2971 return commit;
2972}
1da177e4 2973
b2220cad
JV		 2974/*
2975 * Called to commit link state changes noted by inspection step of
2976 * active-backup mode ARP monitor.
2977 *
2978 * Called with RTNL and bond->lock for read.
2979 */
2980static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2981{
2982 struct slave *slave;
2983 int i;
1da177e4 2984
b2220cad
JV		 2985 bond_for_each_slave(bond, slave, i) {
2986 switch (slave->new_link) {
2987 case BOND_LINK_NOCHANGE:
2988 continue;
ff59c456 2989
b2220cad
JV		 2990 case BOND_LINK_UP:
2991 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2992
b2220cad
JV		 2993 if (!bond->curr_active_slave &&
2994 time_before_eq(jiffies, slave->dev->trans_start +
2995 delta_in_ticks)) {
2996 slave->link = BOND_LINK_UP;
2997 bond_change_active_slave(bond, slave);
2998 bond->current_arp_slave = NULL;
2999
3000 printk(KERN_INFO DRV_NAME
3001 ": %s: %s is up and now the "
3002 "active interface\n",
3003 bond->dev->name, slave->dev->name);
3004
3005 } else if (bond->curr_active_slave != slave) {
3006 /* this slave has just come up but we
3007 * already have a current slave; this can
3008 * also happen if bond_enslave adds a new
3009 * slave that is up while we are searching
3010 * for a new slave
1da177e4 3011 */
b2220cad
JV		 3012 slave->link = BOND_LINK_UP;
3013 bond_set_slave_inactive_flags(slave);
3014 bond->current_arp_slave = NULL;
3015
3016 printk(KERN_INFO DRV_NAME
3017 ": %s: backup interface %s is now up\n",
3018 bond->dev->name, slave->dev->name);
3019 }
1da177e4 3020
b2220cad 3021 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3022
b2220cad 3023 break;
1da177e4 3024
b2220cad
JV		 3025 case BOND_LINK_DOWN:
3026 if (slave->link_failure_count < UINT_MAX)
3027 slave->link_failure_count++;
3028
3029 slave->link = BOND_LINK_DOWN;
3030
3031 if (slave == bond->curr_active_slave) {
3032 printk(KERN_INFO DRV_NAME
3033 ": %s: link status down for active "
3034 "interface %s, disabling it\n",
3035 bond->dev->name, slave->dev->name);
1da177e4
LT		 3036
3037 bond_set_slave_inactive_flags(slave);
3038
b2220cad
JV		 3039 write_lock_bh(&bond->curr_slave_lock);
3040
3041 bond_select_active_slave(bond);
3042 if (bond->curr_active_slave)
3043 bond->curr_active_slave->jiffies =
3044 jiffies;
3045
3046 write_unlock_bh(&bond->curr_slave_lock);
3047
3048 bond->current_arp_slave = NULL;
3049
3050 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3051 printk(KERN_INFO DRV_NAME
3052 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3053 bond->dev->name, slave->dev->name);
3054
3055 bond_set_slave_inactive_flags(slave);
1da177e4 3056 }
b2220cad
JV		 3057 break;
3058
3059 default:
3060 printk(KERN_ERR DRV_NAME
3061 ": %s: impossible: new_link %d on slave %s\n",
3062 bond->dev->name, slave->new_link,
3063 slave->dev->name);
1da177e4
LT		 3064 }
3065 }
3066
b2220cad
JV		 3067 /*
3068 * No race with changes to primary via sysfs, as we hold rtnl.
3069 */
3070 if (bond->primary_slave &&
3071 (bond->primary_slave != bond->curr_active_slave) &&
3072 (bond->primary_slave->link == BOND_LINK_UP)) {
3073 write_lock_bh(&bond->curr_slave_lock);
3074 bond_change_active_slave(bond, bond->primary_slave);
3075 write_unlock_bh(&bond->curr_slave_lock);
3076 }
1da177e4 3077
b2220cad
JV		 3078 bond_set_carrier(bond);
3079}
1da177e4 3080
b2220cad
JV		 3081/*
3082 * Send ARP probes for active-backup mode ARP monitor.
3083 *
3084 * Called with bond->lock held for read.
3085 */
3086static void bond_ab_arp_probe(struct bonding *bond)
3087{
3088 struct slave *slave;
3089 int i;
1da177e4 3090
b2220cad 3091 read_lock(&bond->curr_slave_lock);
1da177e4 3092
b2220cad
JV		 3093 if (bond->current_arp_slave && bond->curr_active_slave)
3094 printk("PROBE: c_arp %s && cas %s BAD\n",
3095 bond->current_arp_slave->dev->name,
3096 bond->curr_active_slave->dev->name);
1da177e4 3097
b2220cad
JV		 3098 if (bond->curr_active_slave) {
3099 bond_arp_send_all(bond, bond->curr_active_slave);
3100 read_unlock(&bond->curr_slave_lock);
3101 return;
3102 }
1da177e4 3103
b2220cad 3104 read_unlock(&bond->curr_slave_lock);
059fe7a5 3105
b2220cad
JV		 3106 /* if we don't have a curr_active_slave, search for the next available
3107 * backup slave from the current_arp_slave and make it the candidate
3108 * for becoming the curr_active_slave
3109 */
1da177e4 3110
b2220cad
JV		 3111 if (!bond->current_arp_slave) {
3112 bond->current_arp_slave = bond->first_slave;
3113 if (!bond->current_arp_slave)
3114 return;
3115 }
1da177e4 3116
b2220cad 3117 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3118
b2220cad
JV		 3119 /* search for next candidate */
3120 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3121 if (IS_UP(slave->dev)) {
3122 slave->link = BOND_LINK_BACK;
3123 bond_set_slave_active_flags(slave);
3124 bond_arp_send_all(bond, slave);
1da177e4 3125 slave->jiffies = jiffies;
b2220cad
JV		 3126 bond->current_arp_slave = slave;
3127 break;
1da177e4
LT		 3128 }
3129
b2220cad
JV		 3130 /* if the link state is up at this point, we
3131 * mark it down - this can happen if we have
3132 * simultaneous link failures and
3133 * reselect_active_interface doesn't make this
3134 * one the current slave so it is still marked
3135 * up when it is actually down
1da177e4 3136 */
b2220cad
JV		 3137 if (slave->link == BOND_LINK_UP) {
3138 slave->link = BOND_LINK_DOWN;
3139 if (slave->link_failure_count < UINT_MAX)
3140 slave->link_failure_count++;
1da177e4 3141
b2220cad
JV		 3142 bond_set_slave_inactive_flags(slave);
3143
3144 printk(KERN_INFO DRV_NAME
3145 ": %s: backup interface %s is now down.\n",
3146 bond->dev->name, slave->dev->name);
1da177e4 3147 }
b2220cad
JV		 3148 }
3149}
1da177e4 3150
b2220cad
JV		 3151void bond_activebackup_arp_mon(struct work_struct *work)
3152{
3153 struct bonding *bond = container_of(work, struct bonding,
3154 arp_work.work);
3155 int delta_in_ticks;
1da177e4 3156
b2220cad 3157 read_lock(&bond->lock);
1da177e4 3158
b2220cad
JV		 3159 if (bond->kill_timers)
3160 goto out;
1da177e4 3161
b2220cad 3162 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3163
b2220cad
JV		 3164 if (bond->slave_cnt == 0)
3165 goto re_arm;
3166
3167 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3168 read_unlock(&bond->lock);
3169 rtnl_lock();
3170 read_lock(&bond->lock);
3171
3172 bond_ab_arp_commit(bond, delta_in_ticks);
3173
3174 read_unlock(&bond->lock);
3175 rtnl_unlock();
3176 read_lock(&bond->lock);
1da177e4
LT		 3177 }
3178
b2220cad
JV		 3179 bond_ab_arp_probe(bond);
3180
1da177e4
LT		 3181re_arm:
3182 if (bond->params.arp_interval) {
1b76b316 3183 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3184 }
3185out:
3186 read_unlock(&bond->lock);
3187}
3188
3189/*------------------------------ proc/seq_file-------------------------------*/
3190
3191#ifdef CONFIG_PROC_FS
3192
1da177e4
LT		 3193static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3194{
3195 struct bonding *bond = seq->private;
3196 loff_t off = 0;
3197 struct slave *slave;
3198 int i;
3199
3200 /* make sure the bond won't be taken away */
3201 read_lock(&dev_base_lock);
6603a6f2 3202 read_lock(&bond->lock);
1da177e4
LT		 3203
3204 if (*pos == 0) {
3205 return SEQ_START_TOKEN;
3206 }
3207
3208 bond_for_each_slave(bond, slave, i) {
3209 if (++off == *pos) {
3210 return slave;
3211 }
3212 }
3213
3214 return NULL;
3215}
3216
3217static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3218{
3219 struct bonding *bond = seq->private;
3220 struct slave *slave = v;
3221
3222 ++*pos;
3223 if (v == SEQ_START_TOKEN) {
3224 return bond->first_slave;
3225 }
3226
3227 slave = slave->next;
3228
3229 return (slave == bond->first_slave) ? NULL : slave;
3230}
3231
3232static void bond_info_seq_stop(struct seq_file *seq, void *v)
3233{
3234 struct bonding *bond = seq->private;
3235
6603a6f2 3236 read_unlock(&bond->lock);
1da177e4
LT		 3237 read_unlock(&dev_base_lock);
3238}
3239
3240static void bond_info_show_master(struct seq_file *seq)
3241{
3242 struct bonding *bond = seq->private;
3243 struct slave *curr;
4756b02f
MW		 3244 int i;
3245 u32 target;
1da177e4
LT		 3246
3247 read_lock(&bond->curr_slave_lock);
3248 curr = bond->curr_active_slave;
3249 read_unlock(&bond->curr_slave_lock);
3250
dd957c57 3251 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3252 bond_mode_name(bond->params.mode));
3253
dd957c57
JV		 3254 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3255 bond->params.fail_over_mac)
3915c1e8
JV		 3256 seq_printf(seq, " (fail_over_mac %s)",
3257 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3258
3259 seq_printf(seq, "\n");
3260
c61b75ad
MW		 3261 if (bond->params.mode == BOND_MODE_XOR ||
3262 bond->params.mode == BOND_MODE_8023AD) {
3263 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3264 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3265 bond->params.xmit_policy);
3266 }
3267
1da177e4
LT		 3268 if (USES_PRIMARY(bond->params.mode)) {
3269 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3270 (bond->primary_slave) ?
3271 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3272
3273 seq_printf(seq, "Currently Active Slave: %s\n",
3274 (curr) ? curr->dev->name : "None");
3275 }
3276
ff59c456
JV		 3277 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3278 "up" : "down");
1da177e4
LT		 3279 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3280 seq_printf(seq, "Up Delay (ms): %d\n",
3281 bond->params.updelay * bond->params.miimon);
3282 seq_printf(seq, "Down Delay (ms): %d\n",
3283 bond->params.downdelay * bond->params.miimon);
3284
4756b02f
MW		 3285
3286 /* ARP information */
3287 if(bond->params.arp_interval > 0) {
3288 int printed=0;
3289 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3290 bond->params.arp_interval);
3291
3292 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3293
3294 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3295 if (!bond->params.arp_targets[i])
3296 continue;
3297 if (printed)
3298 seq_printf(seq, ",");
3299 target = ntohl(bond->params.arp_targets[i]);
3300 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3301 printed = 1;
3302 }
3303 seq_printf(seq, "\n");
3304 }
3305
1da177e4
LT		 3306 if (bond->params.mode == BOND_MODE_8023AD) {
3307 struct ad_info ad_info;
0795af57 3308 DECLARE_MAC_BUF(mac);
1da177e4
LT		 3309
3310 seq_puts(seq, "\n802.3ad info\n");
3311 seq_printf(seq, "LACP rate: %s\n",
3312 (bond->params.lacp_fast) ? "fast" : "slow");
3313
3314 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3315 seq_printf(seq, "bond %s has no active aggregator\n",
3316 bond->dev->name);
3317 } else {
3318 seq_printf(seq, "Active Aggregator Info:\n");
3319
3320 seq_printf(seq, "\tAggregator ID: %d\n",
3321 ad_info.aggregator_id);
3322 seq_printf(seq, "\tNumber of ports: %d\n",
3323 ad_info.ports);
3324 seq_printf(seq, "\tActor Key: %d\n",
3325 ad_info.actor_key);
3326 seq_printf(seq, "\tPartner Key: %d\n",
3327 ad_info.partner_key);
0795af57
JP		 3328 seq_printf(seq, "\tPartner Mac Address: %s\n",
3329 print_mac(mac, ad_info.partner_system));
1da177e4
LT		 3330 }
3331 }
3332}
3333
3334static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3335{
3336 struct bonding *bond = seq->private;
0795af57 3337 DECLARE_MAC_BUF(mac);
1da177e4
LT		 3338
3339 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3340 seq_printf(seq, "MII Status: %s\n",
3341 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3342 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3343 slave->link_failure_count);
3344
217df670 3345 seq_printf(seq,
0795af57
JP		 3346 "Permanent HW addr: %s\n",
3347 print_mac(mac, slave->perm_hwaddr));
1da177e4
LT		 3348
3349 if (bond->params.mode == BOND_MODE_8023AD) {
3350 const struct aggregator *agg
3351 = SLAVE_AD_INFO(slave).port.aggregator;
3352
3353 if (agg) {
3354 seq_printf(seq, "Aggregator ID: %d\n",
3355 agg->aggregator_identifier);
3356 } else {
3357 seq_puts(seq, "Aggregator ID: N/A\n");
3358 }
3359 }
3360}
3361
3362static int bond_info_seq_show(struct seq_file *seq, void *v)
3363{
3364 if (v == SEQ_START_TOKEN) {
3365 seq_printf(seq, "%s\n", version);
3366 bond_info_show_master(seq);
3367 } else {
3368 bond_info_show_slave(seq, v);
3369 }
3370
3371 return 0;
3372}
3373
3374static struct seq_operations bond_info_seq_ops = {
3375 .start = bond_info_seq_start,
3376 .next = bond_info_seq_next,
3377 .stop = bond_info_seq_stop,
3378 .show = bond_info_seq_show,
3379};
3380
3381static int bond_info_open(struct inode *inode, struct file *file)
3382{
3383 struct seq_file *seq;
3384 struct proc_dir_entry *proc;
3385 int res;
3386
3387 res = seq_open(file, &bond_info_seq_ops);
3388 if (!res) {
3389 /* recover the pointer buried in proc_dir_entry data */
3390 seq = file->private_data;
3391 proc = PDE(inode);
3392 seq->private = proc->data;
3393 }
3394
3395 return res;
3396}
3397
d54b1fdb 3398static const struct file_operations bond_info_fops = {
1da177e4
LT		 3399 .owner = THIS_MODULE,
3400 .open = bond_info_open,
3401 .read = seq_read,
3402 .llseek = seq_lseek,
3403 .release = seq_release,
3404};
3405
3406static int bond_create_proc_entry(struct bonding *bond)
3407{
3408 struct net_device *bond_dev = bond->dev;
3409
3410 if (bond_proc_dir) {
a95609cb
DL		 3411 bond->proc_entry = proc_create_data(bond_dev->name,
3412 S_IRUGO, bond_proc_dir,
3413 &bond_info_fops, bond);
1da177e4
LT		 3414 if (bond->proc_entry == NULL) {
3415 printk(KERN_WARNING DRV_NAME
3416 ": Warning: Cannot create /proc/net/%s/%s\n",
3417 DRV_NAME, bond_dev->name);
3418 } else {
1da177e4
LT		 3419 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3420 }
3421 }
3422
3423 return 0;
3424}
3425
3426static void bond_remove_proc_entry(struct bonding *bond)
3427{
3428 if (bond_proc_dir && bond->proc_entry) {
3429 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3430 memset(bond->proc_file_name, 0, IFNAMSIZ);
3431 bond->proc_entry = NULL;
3432 }
3433}
3434
3435/* Create the bonding directory under /proc/net, if doesn't exist yet.
3436 * Caller must hold rtnl_lock.
3437 */
3438static void bond_create_proc_dir(void)
3439{
3440 int len = strlen(DRV_NAME);
3441
457c4cbc 3442 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3443 bond_proc_dir = bond_proc_dir->next) {
3444 if ((bond_proc_dir->namelen == len) &&
3445 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3446 break;
3447 }
3448 }
3449
3450 if (!bond_proc_dir) {
457c4cbc 3451 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3452 if (bond_proc_dir) {
3453 bond_proc_dir->owner = THIS_MODULE;
3454 } else {
3455 printk(KERN_WARNING DRV_NAME
3456 ": Warning: cannot create /proc/net/%s\n",
3457 DRV_NAME);
3458 }
3459 }
3460}
3461
3462/* Destroy the bonding directory under /proc/net, if empty.
3463 * Caller must hold rtnl_lock.
3464 */
3465static void bond_destroy_proc_dir(void)
3466{
3467 struct proc_dir_entry *de;
3468
3469 if (!bond_proc_dir) {
3470 return;
3471 }
3472
3473 /* verify that the /proc dir is empty */
3474 for (de = bond_proc_dir->subdir; de; de = de->next) {
3475 /* ignore . and .. */
3476 if (*(de->name) != '.') {
3477 break;
3478 }
3479 }
3480
3481 if (de) {
3482 if (bond_proc_dir->owner == THIS_MODULE) {
3483 bond_proc_dir->owner = NULL;
3484 }
3485 } else {
457c4cbc 3486 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3487 bond_proc_dir = NULL;
3488 }
3489}
3490#endif /* CONFIG_PROC_FS */
3491
3492/*-------------------------- netdev event handling --------------------------*/
3493
3494/*
3495 * Change device name
3496 */
3497static int bond_event_changename(struct bonding *bond)
3498{
3499#ifdef CONFIG_PROC_FS
3500 bond_remove_proc_entry(bond);
3501 bond_create_proc_entry(bond);
3502#endif
b76cdba9
MW		 3503 down_write(&(bonding_rwsem));
3504 bond_destroy_sysfs_entry(bond);
3505 bond_create_sysfs_entry(bond);
3506 up_write(&(bonding_rwsem));
1da177e4
LT		 3507 return NOTIFY_DONE;
3508}
3509
3510static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3511{
3512 struct bonding *event_bond = bond_dev->priv;
3513
3514 switch (event) {
3515 case NETDEV_CHANGENAME:
3516 return bond_event_changename(event_bond);
3517 case NETDEV_UNREGISTER:
3b96c858 3518 bond_release_all(event_bond->dev);
1da177e4
LT		 3519 break;
3520 default:
3521 break;
3522 }
3523
3524 return NOTIFY_DONE;
3525}
3526
3527static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3528{
3529 struct net_device *bond_dev = slave_dev->master;
8531c5ff 3530 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3531
3532 switch (event) {
3533 case NETDEV_UNREGISTER:
3534 if (bond_dev) {
1284cd3a
JV		 3535 if (bond->setup_by_slave)
3536 bond_release_and_destroy(bond_dev, slave_dev);
3537 else
3538 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3539 }
3540 break;
3541 case NETDEV_CHANGE:
3542 /*
3543 * TODO: is this what we get if somebody
3544 * sets up a hierarchical bond, then rmmod's
3545 * one of the slave bonding devices?
3546 */
3547 break;
3548 case NETDEV_DOWN:
3549 /*
3550 * ... Or is it this?
3551 */
3552 break;
3553 case NETDEV_CHANGEMTU:
3554 /*
3555 * TODO: Should slaves be allowed to
3556 * independently alter their MTU? For
3557 * an active-backup bond, slaves need
3558 * not be the same type of device, so
3559 * MTUs may vary. For other modes,
3560 * slaves arguably should have the
3561 * same MTUs. To do this, we'd need to
3562 * take over the slave's change_mtu
3563 * function for the duration of their
3564 * servitude.
3565 */
3566 break;
3567 case NETDEV_CHANGENAME:
3568 /*
3569 * TODO: handle changing the primary's name
3570 */
3571 break;
8531c5ff
AK		 3572 case NETDEV_FEAT_CHANGE:
3573 bond_compute_features(bond);
3574 break;
1da177e4
LT		 3575 default:
3576 break;
3577 }
3578
3579 return NOTIFY_DONE;
3580}
3581
3582/*
3583 * bond_netdev_event: handle netdev notifier chain events.
3584 *
3585 * This function receives events for the netdev chain. The caller (an
e041c683 3586 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3587 * locks for us to safely manipulate the slave devices (RTNL lock,
3588 * dev_probe_lock).
3589 */
3590static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3591{
3592 struct net_device *event_dev = (struct net_device *)ptr;
3593
c346dca1 3594 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3595 return NOTIFY_DONE;
3596
1da177e4
LT		 3597 dprintk("event_dev: %s, event: %lx\n",
3598 (event_dev ? event_dev->name : "None"),
3599 event);
3600
0b680e75
JV		 3601 if (!(event_dev->priv_flags & IFF_BONDING))
3602 return NOTIFY_DONE;
3603
1da177e4
LT		 3604 if (event_dev->flags & IFF_MASTER) {
3605 dprintk("IFF_MASTER\n");
3606 return bond_master_netdev_event(event, event_dev);
3607 }
3608
3609 if (event_dev->flags & IFF_SLAVE) {
3610 dprintk("IFF_SLAVE\n");
3611 return bond_slave_netdev_event(event, event_dev);
3612 }
3613
3614 return NOTIFY_DONE;
3615}
3616
c3ade5ca
JV		 3617/*
3618 * bond_inetaddr_event: handle inetaddr notifier chain events.
3619 *
3620 * We keep track of device IPs primarily to use as source addresses in
3621 * ARP monitor probes (rather than spewing out broadcasts all the time).
3622 *
3623 * We track one IP for the main device (if it has one), plus one per VLAN.
3624 */
3625static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3626{
3627 struct in_ifaddr *ifa = ptr;
3628 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3629 struct bonding *bond;
3630 struct vlan_entry *vlan;
c3ade5ca 3631
c346dca1 3632 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3633 return NOTIFY_DONE;
3634
0883beca 3635 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3636 if (bond->dev == event_dev) {
3637 switch (event) {
3638 case NETDEV_UP:
3639 bond->master_ip = ifa->ifa_local;
3640 return NOTIFY_OK;
3641 case NETDEV_DOWN:
3642 bond->master_ip = bond_glean_dev_ip(bond->dev);
3643 return NOTIFY_OK;
3644 default:
3645 return NOTIFY_DONE;
3646 }
3647 }
3648
0883beca 3649 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3650 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3651 if (vlan_dev == event_dev) {
3652 switch (event) {
3653 case NETDEV_UP:
3654 vlan->vlan_ip = ifa->ifa_local;
3655 return NOTIFY_OK;
3656 case NETDEV_DOWN:
3657 vlan->vlan_ip =
3658 bond_glean_dev_ip(vlan_dev);
3659 return NOTIFY_OK;
3660 default:
3661 return NOTIFY_DONE;
3662 }
3663 }
3664 }
3665 }
3666 return NOTIFY_DONE;
3667}
3668
1da177e4
LT		 3669static struct notifier_block bond_netdev_notifier = {
3670 .notifier_call = bond_netdev_event,
3671};
3672
c3ade5ca
JV		 3673static struct notifier_block bond_inetaddr_notifier = {
3674 .notifier_call = bond_inetaddr_event,
3675};
3676
1da177e4
LT		 3677/*-------------------------- Packet type handling ---------------------------*/
3678
3679/* register to receive lacpdus on a bond */
3680static void bond_register_lacpdu(struct bonding *bond)
3681{
3682 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3683
3684 /* initialize packet type */
3685 pk_type->type = PKT_TYPE_LACPDU;
3686 pk_type->dev = bond->dev;
3687 pk_type->func = bond_3ad_lacpdu_recv;
3688
3689 dev_add_pack(pk_type);
3690}
3691
3692/* unregister to receive lacpdus on a bond */
3693static void bond_unregister_lacpdu(struct bonding *bond)
3694{
3695 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3696}
3697
f5b2b966
JV		 3698void bond_register_arp(struct bonding *bond)
3699{
3700 struct packet_type *pt = &bond->arp_mon_pt;
3701
c4f283b1
JV		 3702 if (pt->type)
3703 return;
3704
f5b2b966 3705 pt->type = htons(ETH_P_ARP);
e245cb71 3706 pt->dev = bond->dev;
f5b2b966
JV		 3707 pt->func = bond_arp_rcv;
3708 dev_add_pack(pt);
3709}
3710
3711void bond_unregister_arp(struct bonding *bond)
3712{
c4f283b1
JV		 3713 struct packet_type *pt = &bond->arp_mon_pt;
3714
3715 dev_remove_pack(pt);
3716 pt->type = 0;
f5b2b966
JV		 3717}
3718
169a3e66
JV		 3719/*---------------------------- Hashing Policies -----------------------------*/
3720
6f6652be
JV		 3721/*
3722 * Hash for the output device based upon layer 2 and layer 3 data. If
3723 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3724 */
3725static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3726 struct net_device *bond_dev, int count)
3727{
3728 struct ethhdr *data = (struct ethhdr *)skb->data;
3729 struct iphdr *iph = ip_hdr(skb);
3730
3731 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3732 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3733 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3734 }
3735
3736 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3737}
3738
169a3e66 3739/*
59c51591 3740 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3741 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3742 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3743 */
3744static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3745 struct net_device *bond_dev, int count)
3746{
3747 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3748 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3749 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3750 int layer4_xor = 0;
3751
3752 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3753 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3754 (iph->protocol == IPPROTO_TCP ||
3755 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3756 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3757 }
3758 return (layer4_xor ^
3759 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3760
3761 }
3762
3763 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3764}
3765
3766/*
3767 * Hash for the output device based upon layer 2 data
3768 */
3769static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3770 struct net_device *bond_dev, int count)
3771{
3772 struct ethhdr *data = (struct ethhdr *)skb->data;
3773
3774 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3775}
3776
1da177e4
LT		 3777/*-------------------------- Device entry points ----------------------------*/
3778
3779static int bond_open(struct net_device *bond_dev)
3780{
3781 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3782
3783 bond->kill_timers = 0;
3784
3785 if ((bond->params.mode == BOND_MODE_TLB) ||
3786 (bond->params.mode == BOND_MODE_ALB)) {
1da177e4
LT		 3787 /* bond_alb_initialize must be called before the timer
3788 * is started.
3789 */
3790 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3791 /* something went wrong - fail the open operation */
3792 return -1;
3793 }
3794
1b76b316
JV		 3795 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3796 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3797 }
3798
3799 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3800 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3801 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3802 }
3803
3804 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3805 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3806 INIT_DELAYED_WORK(&bond->arp_work,
3807 bond_activebackup_arp_mon);
3808 else
3809 INIT_DELAYED_WORK(&bond->arp_work,
3810 bond_loadbalance_arp_mon);
3811
3812 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3813 if (bond->params.arp_validate)
3814 bond_register_arp(bond);
1da177e4
LT		 3815 }
3816
3817 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3818 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3819 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3820 /* register to receive LACPDUs */
3821 bond_register_lacpdu(bond);
3822 }
3823
3824 return 0;
3825}
3826
3827static int bond_close(struct net_device *bond_dev)
3828{
3829 struct bonding *bond = bond_dev->priv;
3830
3831 if (bond->params.mode == BOND_MODE_8023AD) {
3832 /* Unregister the receive of LACPDUs */
3833 bond_unregister_lacpdu(bond);
3834 }
3835
f5b2b966
JV		 3836 if (bond->params.arp_validate)
3837 bond_unregister_arp(bond);
3838
1da177e4
LT		 3839 write_lock_bh(&bond->lock);
3840
1da177e4
LT		 3841
3842 /* signal timers not to re-arm */
3843 bond->kill_timers = 1;
3844
3845 write_unlock_bh(&bond->lock);
3846
1da177e4 3847 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3848 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3849 }
3850
3851 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3852 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3853 }
3854
3855 switch (bond->params.mode) {
3856 case BOND_MODE_8023AD:
1b76b316 3857 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3858 break;
3859 case BOND_MODE_TLB:
3860 case BOND_MODE_ALB:
1b76b316 3861 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3862 break;
3863 default:
3864 break;
3865 }
3866
1da177e4
LT		 3867
3868 if ((bond->params.mode == BOND_MODE_TLB) ||
3869 (bond->params.mode == BOND_MODE_ALB)) {
3870 /* Must be called only after all
3871 * slaves have been released
3872 */
3873 bond_alb_deinitialize(bond);
3874 }
3875
3876 return 0;
3877}
3878
3879static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3880{
3881 struct bonding *bond = bond_dev->priv;
3882 struct net_device_stats *stats = &(bond->stats), *sstats;
2439f9eb 3883 struct net_device_stats local_stats;
1da177e4
LT		 3884 struct slave *slave;
3885 int i;
3886
2439f9eb 3887 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3888
3889 read_lock_bh(&bond->lock);
3890
3891 bond_for_each_slave(bond, slave, i) {
c45d286e 3892 sstats = slave->dev->get_stats(slave->dev);
2439f9eb
AG		 3893 local_stats.rx_packets += sstats->rx_packets;
3894 local_stats.rx_bytes += sstats->rx_bytes;
3895 local_stats.rx_errors += sstats->rx_errors;
3896 local_stats.rx_dropped += sstats->rx_dropped;
3897
3898 local_stats.tx_packets += sstats->tx_packets;
3899 local_stats.tx_bytes += sstats->tx_bytes;
3900 local_stats.tx_errors += sstats->tx_errors;
3901 local_stats.tx_dropped += sstats->tx_dropped;
3902
3903 local_stats.multicast += sstats->multicast;
3904 local_stats.collisions += sstats->collisions;
3905
3906 local_stats.rx_length_errors += sstats->rx_length_errors;
3907 local_stats.rx_over_errors += sstats->rx_over_errors;
3908 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3909 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3910 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3911 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3912
3913 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3914 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3915 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3916 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3917 local_stats.tx_window_errors += sstats->tx_window_errors;
3918 }
3919
3920 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3921
3922 read_unlock_bh(&bond->lock);
3923
3924 return stats;
3925}
3926
3927static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3928{
3929 struct net_device *slave_dev = NULL;
3930 struct ifbond k_binfo;
3931 struct ifbond __user *u_binfo = NULL;
3932 struct ifslave k_sinfo;
3933 struct ifslave __user *u_sinfo = NULL;
3934 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3935 int res = 0;
3936
3937 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3938 bond_dev->name, cmd);
3939
3940 switch (cmd) {
1da177e4
LT		 3941 case SIOCGMIIPHY:
3942 mii = if_mii(ifr);
3943 if (!mii) {
3944 return -EINVAL;
3945 }
3946 mii->phy_id = 0;
3947 /* Fall Through */
3948 case SIOCGMIIREG:
3949 /*
3950 * We do this again just in case we were called by SIOCGMIIREG
3951 * instead of SIOCGMIIPHY.
3952 */
3953 mii = if_mii(ifr);
3954 if (!mii) {
3955 return -EINVAL;
3956 }
3957
3958 if (mii->reg_num == 1) {
3959 struct bonding *bond = bond_dev->priv;
3960 mii->val_out = 0;
6603a6f2 3961 read_lock(&bond->lock);
1da177e4 3962 read_lock(&bond->curr_slave_lock);
4e140079 3963 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3964 mii->val_out = BMSR_LSTATUS;
3965 }
3966 read_unlock(&bond->curr_slave_lock);
6603a6f2 3967 read_unlock(&bond->lock);
1da177e4
LT		 3968 }
3969
3970 return 0;
3971 case BOND_INFO_QUERY_OLD:
3972 case SIOCBONDINFOQUERY:
3973 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3974
3975 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3976 return -EFAULT;
3977 }
3978
3979 res = bond_info_query(bond_dev, &k_binfo);
3980 if (res == 0) {
3981 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3982 return -EFAULT;
3983 }
3984 }
3985
3986 return res;
3987 case BOND_SLAVE_INFO_QUERY_OLD:
3988 case SIOCBONDSLAVEINFOQUERY:
3989 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3990
3991 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3992 return -EFAULT;
3993 }
3994
3995 res = bond_slave_info_query(bond_dev, &k_sinfo);
3996 if (res == 0) {
3997 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3998 return -EFAULT;
3999 }
4000 }
4001
4002 return res;
4003 default:
4004 /* Go on */
4005 break;
4006 }
4007
4008 if (!capable(CAP_NET_ADMIN)) {
4009 return -EPERM;
4010 }
4011
b76cdba9 4012 down_write(&(bonding_rwsem));
881d966b 4013 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT		 4014
4015 dprintk("slave_dev=%p: \n", slave_dev);
4016
4017 if (!slave_dev) {
4018 res = -ENODEV;
4019 } else {
4020 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4021 switch (cmd) {
4022 case BOND_ENSLAVE_OLD:
4023 case SIOCBONDENSLAVE:
4024 res = bond_enslave(bond_dev, slave_dev);
4025 break;
4026 case BOND_RELEASE_OLD:
4027 case SIOCBONDRELEASE:
4028 res = bond_release(bond_dev, slave_dev);
4029 break;
4030 case BOND_SETHWADDR_OLD:
4031 case SIOCBONDSETHWADDR:
4032 res = bond_sethwaddr(bond_dev, slave_dev);
4033 break;
4034 case BOND_CHANGE_ACTIVE_OLD:
4035 case SIOCBONDCHANGEACTIVE:
4036 res = bond_ioctl_change_active(bond_dev, slave_dev);
4037 break;
4038 default:
4039 res = -EOPNOTSUPP;
4040 }
4041
4042 dev_put(slave_dev);
4043 }
4044
b76cdba9 4045 up_write(&(bonding_rwsem));
1da177e4
LT		 4046 return res;
4047}
4048
4049static void bond_set_multicast_list(struct net_device *bond_dev)
4050{
4051 struct bonding *bond = bond_dev->priv;
4052 struct dev_mc_list *dmi;
4053
1da177e4
LT		 4054 /*
4055 * Do promisc before checking multicast_mode
4056 */
4057 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4058 bond_set_promiscuity(bond, 1);
4059 }
4060
4061 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4062 bond_set_promiscuity(bond, -1);
4063 }
4064
4065 /* set allmulti flag to slaves */
4066 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4067 bond_set_allmulti(bond, 1);
4068 }
4069
4070 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4071 bond_set_allmulti(bond, -1);
4072 }
4073
80ee5ad2
JV		 4074 read_lock(&bond->lock);
4075
1da177e4
LT		 4076 bond->flags = bond_dev->flags;
4077
4078 /* looking for addresses to add to slaves' mc list */
4079 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4080 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4081 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4082 }
4083 }
4084
4085 /* looking for addresses to delete from slaves' list */
4086 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4087 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4088 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4089 }
4090 }
4091
4092 /* save master's multicast list */
4093 bond_mc_list_destroy(bond);
4094 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4095
80ee5ad2 4096 read_unlock(&bond->lock);
1da177e4
LT		 4097}
4098
4099/*
4100 * Change the MTU of all of a master's slaves to match the master
4101 */
4102static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4103{
4104 struct bonding *bond = bond_dev->priv;
4105 struct slave *slave, *stop_at;
4106 int res = 0;
4107 int i;
4108
4109 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4110 (bond_dev ? bond_dev->name : "None"), new_mtu);
4111
4112 /* Can't hold bond->lock with bh disabled here since
4113 * some base drivers panic. On the other hand we can't
4114 * hold bond->lock without bh disabled because we'll
4115 * deadlock. The only solution is to rely on the fact
4116 * that we're under rtnl_lock here, and the slaves
4117 * list won't change. This doesn't solve the problem
4118 * of setting the slave's MTU while it is
4119 * transmitting, but the assumption is that the base
4120 * driver can handle that.
4121 *
4122 * TODO: figure out a way to safely iterate the slaves
4123 * list, but without holding a lock around the actual
4124 * call to the base driver.
4125 */
4126
4127 bond_for_each_slave(bond, slave, i) {
4128 dprintk("s %p s->p %p c_m %p\n", slave,
4129 slave->prev, slave->dev->change_mtu);
e944ef79 4130
1da177e4
LT		 4131 res = dev_set_mtu(slave->dev, new_mtu);
4132
4133 if (res) {
4134 /* If we failed to set the slave's mtu to the new value
4135 * we must abort the operation even in ACTIVE_BACKUP
4136 * mode, because if we allow the backup slaves to have
4137 * different mtu values than the active slave we'll
4138 * need to change their mtu when doing a failover. That
4139 * means changing their mtu from timer context, which
4140 * is probably not a good idea.
4141 */
4142 dprintk("err %d %s\n", res, slave->dev->name);
4143 goto unwind;
4144 }
4145 }
4146
4147 bond_dev->mtu = new_mtu;
4148
4149 return 0;
4150
4151unwind:
4152 /* unwind from head to the slave that failed */
4153 stop_at = slave;
4154 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4155 int tmp_res;
4156
4157 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4158 if (tmp_res) {
4159 dprintk("unwind err %d dev %s\n", tmp_res,
4160 slave->dev->name);
4161 }
4162 }
4163
4164 return res;
4165}
4166
4167/*
4168 * Change HW address
4169 *
4170 * Note that many devices must be down to change the HW address, and
4171 * downing the master releases all slaves. We can make bonds full of
4172 * bonding devices to test this, however.
4173 */
4174static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4175{
4176 struct bonding *bond = bond_dev->priv;
4177 struct sockaddr *sa = addr, tmp_sa;
4178 struct slave *slave, *stop_at;
4179 int res = 0;
4180 int i;
4181
4182 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4183
dd957c57 4184 /*
3915c1e8
JV		 4185 * If fail_over_mac is set to active, do nothing and return
4186 * success. Returning an error causes ifenslave to fail.
dd957c57 4187 */
3915c1e8 4188 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4189 return 0;
2ab82852 4190
1da177e4
LT		 4191 if (!is_valid_ether_addr(sa->sa_data)) {
4192 return -EADDRNOTAVAIL;
4193 }
4194
4195 /* Can't hold bond->lock with bh disabled here since
4196 * some base drivers panic. On the other hand we can't
4197 * hold bond->lock without bh disabled because we'll
4198 * deadlock. The only solution is to rely on the fact
4199 * that we're under rtnl_lock here, and the slaves
4200 * list won't change. This doesn't solve the problem
4201 * of setting the slave's hw address while it is
4202 * transmitting, but the assumption is that the base
4203 * driver can handle that.
4204 *
4205 * TODO: figure out a way to safely iterate the slaves
4206 * list, but without holding a lock around the actual
4207 * call to the base driver.
4208 */
4209
4210 bond_for_each_slave(bond, slave, i) {
4211 dprintk("slave %p %s\n", slave, slave->dev->name);
4212
4213 if (slave->dev->set_mac_address == NULL) {
4214 res = -EOPNOTSUPP;
4215 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4216 goto unwind;
4217 }
4218
4219 res = dev_set_mac_address(slave->dev, addr);
4220 if (res) {
4221 /* TODO: consider downing the slave
4222 * and retry ?
4223 * User should expect communications
4224 * breakage anyway until ARP finish
4225 * updating, so...
4226 */
4227 dprintk("err %d %s\n", res, slave->dev->name);
4228 goto unwind;
4229 }
4230 }
4231
4232 /* success */
4233 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4234 return 0;
4235
4236unwind:
4237 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4238 tmp_sa.sa_family = bond_dev->type;
4239
4240 /* unwind from head to the slave that failed */
4241 stop_at = slave;
4242 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4243 int tmp_res;
4244
4245 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4246 if (tmp_res) {
4247 dprintk("unwind err %d dev %s\n", tmp_res,
4248 slave->dev->name);
4249 }
4250 }
4251
4252 return res;
4253}
4254
4255static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4256{
4257 struct bonding *bond = bond_dev->priv;
4258 struct slave *slave, *start_at;
cf5f9044 4259 int i, slave_no, res = 1;
1da177e4
LT		 4260
4261 read_lock(&bond->lock);
4262
4263 if (!BOND_IS_OK(bond)) {
4264 goto out;
4265 }
4266
cf5f9044
JV		 4267 /*
4268 * Concurrent TX may collide on rr_tx_counter; we accept that
4269 * as being rare enough not to justify using an atomic op here
4270 */
4271 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4272
cf5f9044
JV		 4273 bond_for_each_slave(bond, slave, i) {
4274 slave_no--;
4275 if (slave_no < 0) {
4276 break;
4277 }
1da177e4
LT		 4278 }
4279
cf5f9044 4280 start_at = slave;
1da177e4
LT		 4281 bond_for_each_slave_from(bond, slave, i, start_at) {
4282 if (IS_UP(slave->dev) &&
4283 (slave->link == BOND_LINK_UP) &&
4284 (slave->state == BOND_STATE_ACTIVE)) {
4285 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4286 break;
4287 }
4288 }
4289
1da177e4
LT		 4290out:
4291 if (res) {
4292 /* no suitable interface, frame not sent */
4293 dev_kfree_skb(skb);
4294 }
4295 read_unlock(&bond->lock);
4296 return 0;
4297}
4298
075897ce 4299
1da177e4
LT		 4300/*
4301 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4302 * the bond has a usable interface.
4303 */
4304static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4305{
4306 struct bonding *bond = bond_dev->priv;
4307 int res = 1;
4308
1da177e4
LT		 4309 read_lock(&bond->lock);
4310 read_lock(&bond->curr_slave_lock);
4311
4312 if (!BOND_IS_OK(bond)) {
4313 goto out;
4314 }
4315
075897ce
JL		 4316 if (!bond->curr_active_slave)
4317 goto out;
4318
075897ce
JL		 4319 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4320
1da177e4
LT		 4321out:
4322 if (res) {
4323 /* no suitable interface, frame not sent */
4324 dev_kfree_skb(skb);
4325 }
4326 read_unlock(&bond->curr_slave_lock);
4327 read_unlock(&bond->lock);
4328 return 0;
4329}
4330
4331/*
169a3e66
JV		 4332 * In bond_xmit_xor() , we determine the output device by using a pre-
4333 * determined xmit_hash_policy(), If the selected device is not enabled,
4334 * find the next active slave.
1da177e4
LT		 4335 */
4336static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4337{
4338 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 4339 struct slave *slave, *start_at;
4340 int slave_no;
4341 int i;
4342 int res = 1;
4343
4344 read_lock(&bond->lock);
4345
4346 if (!BOND_IS_OK(bond)) {
4347 goto out;
4348 }
4349
169a3e66 4350 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4351
4352 bond_for_each_slave(bond, slave, i) {
4353 slave_no--;
4354 if (slave_no < 0) {
4355 break;
4356 }
4357 }
4358
4359 start_at = slave;
4360
4361 bond_for_each_slave_from(bond, slave, i, start_at) {
4362 if (IS_UP(slave->dev) &&
4363 (slave->link == BOND_LINK_UP) &&
4364 (slave->state == BOND_STATE_ACTIVE)) {
4365 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4366 break;
4367 }
4368 }
4369
4370out:
4371 if (res) {
4372 /* no suitable interface, frame not sent */
4373 dev_kfree_skb(skb);
4374 }
4375 read_unlock(&bond->lock);
4376 return 0;
4377}
4378
4379/*
4380 * in broadcast mode, we send everything to all usable interfaces.
4381 */
4382static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4383{
4384 struct bonding *bond = bond_dev->priv;
4385 struct slave *slave, *start_at;
4386 struct net_device *tx_dev = NULL;
4387 int i;
4388 int res = 1;
4389
4390 read_lock(&bond->lock);
4391
4392 if (!BOND_IS_OK(bond)) {
4393 goto out;
4394 }
4395
4396 read_lock(&bond->curr_slave_lock);
4397 start_at = bond->curr_active_slave;
4398 read_unlock(&bond->curr_slave_lock);
4399
4400 if (!start_at) {
4401 goto out;
4402 }
4403
4404 bond_for_each_slave_from(bond, slave, i, start_at) {
4405 if (IS_UP(slave->dev) &&
4406 (slave->link == BOND_LINK_UP) &&
4407 (slave->state == BOND_STATE_ACTIVE)) {
4408 if (tx_dev) {
4409 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4410 if (!skb2) {
4411 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4412 ": %s: Error: bond_xmit_broadcast(): "
4413 "skb_clone() failed\n",
4414 bond_dev->name);
1da177e4
LT		 4415 continue;
4416 }
4417
4418 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4419 if (res) {
4420 dev_kfree_skb(skb2);
4421 continue;
4422 }
4423 }
4424 tx_dev = slave->dev;
4425 }
4426 }
4427
4428 if (tx_dev) {
4429 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4430 }
4431
4432out:
4433 if (res) {
4434 /* no suitable interface, frame not sent */
4435 dev_kfree_skb(skb);
4436 }
4437 /* frame sent to all suitable interfaces */
4438 read_unlock(&bond->lock);
4439 return 0;
4440}
4441
4442/*------------------------- Device initialization ---------------------------*/
4443
6f6652be
JV		 4444static void bond_set_xmit_hash_policy(struct bonding *bond)
4445{
4446 switch (bond->params.xmit_policy) {
4447 case BOND_XMIT_POLICY_LAYER23:
4448 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4449 break;
4450 case BOND_XMIT_POLICY_LAYER34:
4451 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4452 break;
4453 case BOND_XMIT_POLICY_LAYER2:
4454 default:
4455 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4456 break;
4457 }
4458}
4459
1da177e4
LT		 4460/*
4461 * set bond mode specific net device operations
4462 */
a77b5325 4463void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4464{
169a3e66
JV		 4465 struct net_device *bond_dev = bond->dev;
4466
1da177e4
LT		 4467 switch (mode) {
4468 case BOND_MODE_ROUNDROBIN:
4469 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4470 break;
4471 case BOND_MODE_ACTIVEBACKUP:
4472 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4473 break;
4474 case BOND_MODE_XOR:
4475 bond_dev->hard_start_xmit = bond_xmit_xor;
6f6652be 4476 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4477 break;
4478 case BOND_MODE_BROADCAST:
4479 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4480 break;
4481 case BOND_MODE_8023AD:
8f903c70 4482 bond_set_master_3ad_flags(bond);
1da177e4 4483 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
6f6652be 4484 bond_set_xmit_hash_policy(bond);
1da177e4 4485 break;
1da177e4 4486 case BOND_MODE_ALB:
8f903c70
JV		 4487 bond_set_master_alb_flags(bond);
4488 /* FALLTHRU */
4489 case BOND_MODE_TLB:
1da177e4
LT		 4490 bond_dev->hard_start_xmit = bond_alb_xmit;
4491 bond_dev->set_mac_address = bond_alb_set_mac_address;
4492 break;
4493 default:
4494 /* Should never happen, mode already checked */
4495 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4496 ": %s: Error: Unknown bonding mode %d\n",
4497 bond_dev->name,
1da177e4
LT		 4498 mode);
4499 break;
4500 }
4501}
4502
217df670
JV		 4503static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4504 struct ethtool_drvinfo *drvinfo)
4505{
4506 strncpy(drvinfo->driver, DRV_NAME, 32);
4507 strncpy(drvinfo->version, DRV_VERSION, 32);
4508 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4509}
4510
7282d491 4511static const struct ethtool_ops bond_ethtool_ops = {
217df670 4512 .get_drvinfo = bond_ethtool_get_drvinfo,
8531c5ff
AK		 4513};
4514
1da177e4
LT		 4515/*
4516 * Does not allocate but creates a /proc entry.
4517 * Allowed to fail.
4518 */
3c535952 4519static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4
LT		 4520{
4521 struct bonding *bond = bond_dev->priv;
4522
4523 dprintk("Begin bond_init for %s\n", bond_dev->name);
4524
4525 /* initialize rwlocks */
4526 rwlock_init(&bond->lock);
4527 rwlock_init(&bond->curr_slave_lock);
4528
4529 bond->params = *params; /* copy params struct */
4530
1b76b316
JV		 4531 bond->wq = create_singlethread_workqueue(bond_dev->name);
4532 if (!bond->wq)
4533 return -ENOMEM;
4534
1da177e4
LT		 4535 /* Initialize pointers */
4536 bond->first_slave = NULL;
4537 bond->curr_active_slave = NULL;
4538 bond->current_arp_slave = NULL;
4539 bond->primary_slave = NULL;
4540 bond->dev = bond_dev;
1053f62c 4541 bond->send_grat_arp = 0;
d90a162a 4542 bond->setup_by_slave = 0;
1da177e4
LT		 4543 INIT_LIST_HEAD(&bond->vlan_list);
4544
4545 /* Initialize the device entry points */
4546 bond_dev->open = bond_open;
4547 bond_dev->stop = bond_close;
4548 bond_dev->get_stats = bond_get_stats;
4549 bond_dev->do_ioctl = bond_do_ioctl;
8531c5ff 4550 bond_dev->ethtool_ops = &bond_ethtool_ops;
1da177e4
LT		 4551 bond_dev->set_multicast_list = bond_set_multicast_list;
4552 bond_dev->change_mtu = bond_change_mtu;
4553 bond_dev->set_mac_address = bond_set_mac_address;
3a1521b7 4554 bond_dev->validate_addr = NULL;
1da177e4 4555
169a3e66 4556 bond_set_mode_ops(bond, bond->params.mode);
1da177e4
LT		 4557
4558 bond_dev->destructor = free_netdev;
4559
4560 /* Initialize the device options */
4561 bond_dev->tx_queue_len = 0;
4562 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4563 bond_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 4564
4565 /* At first, we block adding VLANs. That's the only way to
4566 * prevent problems that occur when adding VLANs over an
4567 * empty bond. The block will be removed once non-challenged
4568 * slaves are enslaved.
4569 */
4570 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4571
932ff279 4572 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4573 * transmitting */
4574 bond_dev->features |= NETIF_F_LLTX;
4575
4576 /* By default, we declare the bond to be fully
4577 * VLAN hardware accelerated capable. Special
4578 * care is taken in the various xmit functions
4579 * when there are slaves that are not hw accel
4580 * capable
4581 */
4582 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4583 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4584 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4585 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4586 NETIF_F_HW_VLAN_RX |
4587 NETIF_F_HW_VLAN_FILTER);
4588
4589#ifdef CONFIG_PROC_FS
4590 bond_create_proc_entry(bond);
4591#endif
1da177e4
LT		 4592 list_add_tail(&bond->bond_list, &bond_dev_list);
4593
4594 return 0;
4595}
4596
4597/* De-initialize device specific data.
4598 * Caller must hold rtnl_lock.
4599 */
c50b85d0 4600static void bond_deinit(struct net_device *bond_dev)
1da177e4
LT		 4601{
4602 struct bonding *bond = bond_dev->priv;
4603
4604 list_del(&bond->bond_list);
4605
4606#ifdef CONFIG_PROC_FS
4607 bond_remove_proc_entry(bond);
4608#endif
4609}
4610
fdaea7a9
JV		 4611static void bond_work_cancel_all(struct bonding *bond)
4612{
4613 write_lock_bh(&bond->lock);
4614 bond->kill_timers = 1;
4615 write_unlock_bh(&bond->lock);
4616
4617 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4618 cancel_delayed_work(&bond->mii_work);
4619
4620 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4621 cancel_delayed_work(&bond->arp_work);
4622
4623 if (bond->params.mode == BOND_MODE_ALB &&
4624 delayed_work_pending(&bond->alb_work))
4625 cancel_delayed_work(&bond->alb_work);
4626
4627 if (bond->params.mode == BOND_MODE_8023AD &&
4628 delayed_work_pending(&bond->ad_work))
4629 cancel_delayed_work(&bond->ad_work);
4630}
4631
1da177e4
LT		 4632/* Unregister and free all bond devices.
4633 * Caller must hold rtnl_lock.
4634 */
4635static void bond_free_all(void)
4636{
4637 struct bonding *bond, *nxt;
4638
4639 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4640 struct net_device *bond_dev = bond->dev;
4641
fdaea7a9 4642 bond_work_cancel_all(bond);
80ee5ad2 4643 netif_tx_lock_bh(bond_dev);
70298705 4644 bond_mc_list_destroy(bond);
80ee5ad2 4645 netif_tx_unlock_bh(bond_dev);
70298705 4646 /* Release the bonded slaves */
4647 bond_release_all(bond_dev);
92b41daa 4648 bond_destroy(bond);
1da177e4
LT		 4649 }
4650
4651#ifdef CONFIG_PROC_FS
4652 bond_destroy_proc_dir();
4653#endif
4654}
4655
4656/*------------------------- Module initialization ---------------------------*/
4657
4658/*
4659 * Convert string input module parms. Accept either the
ece95f7f
JV		 4660 * number of the mode or its string name. A bit complicated because
4661 * some mode names are substrings of other names, and calls from sysfs
4662 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4663 */
ece95f7f 4664int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4665{
ece95f7f 4666 int mode = -1, i, rv;
a42e534f 4667 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4668
a42e534f
JV		 4669 for (p = (char *)buf; *p; p++)
4670 if (!(isdigit(*p) || isspace(*p)))
4671 break;
4672
4673 if (*p)
ece95f7f 4674 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4675 else
4676 rv = sscanf(buf, "%d", &mode);
4677
4678 if (!rv)
4679 return -1;
1da177e4
LT		 4680
4681 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4682 if (mode == tbl[i].mode)
4683 return tbl[i].mode;
4684 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4685 return tbl[i].mode;
1da177e4
LT		 4686 }
4687
4688 return -1;
4689}
4690
4691static int bond_check_params(struct bond_params *params)
4692{
3915c1e8 4693 int arp_validate_value, fail_over_mac_value;
f5b2b966 4694
1da177e4
LT		 4695 /*
4696 * Convert string parameters.
4697 */
4698 if (mode) {
4699 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4700 if (bond_mode == -1) {
4701 printk(KERN_ERR DRV_NAME
4702 ": Error: Invalid bonding mode \"%s\"\n",
4703 mode == NULL ? "NULL" : mode);
4704 return -EINVAL;
4705 }
4706 }
4707
169a3e66
JV		 4708 if (xmit_hash_policy) {
4709 if ((bond_mode != BOND_MODE_XOR) &&
4710 (bond_mode != BOND_MODE_8023AD)) {
4711 printk(KERN_INFO DRV_NAME
4712 ": xor_mode param is irrelevant in mode %s\n",
4713 bond_mode_name(bond_mode));
4714 } else {
4715 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4716 xmit_hashtype_tbl);
4717 if (xmit_hashtype == -1) {
4718 printk(KERN_ERR DRV_NAME
4719 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4720 xmit_hash_policy == NULL ? "NULL" :
4721 xmit_hash_policy);
4722 return -EINVAL;
4723 }
4724 }
4725 }
4726
1da177e4
LT		 4727 if (lacp_rate) {
4728 if (bond_mode != BOND_MODE_8023AD) {
4729 printk(KERN_INFO DRV_NAME
4730 ": lacp_rate param is irrelevant in mode %s\n",
4731 bond_mode_name(bond_mode));
4732 } else {
4733 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4734 if (lacp_fast == -1) {
4735 printk(KERN_ERR DRV_NAME
4736 ": Error: Invalid lacp rate \"%s\"\n",
4737 lacp_rate == NULL ? "NULL" : lacp_rate);
4738 return -EINVAL;
4739 }
4740 }
4741 }
4742
4743 if (max_bonds < 1 || max_bonds > INT_MAX) {
4744 printk(KERN_WARNING DRV_NAME
4745 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4746 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
1da177e4
LT		 4747 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4748 max_bonds = BOND_DEFAULT_MAX_BONDS;
4749 }
4750
4751 if (miimon < 0) {
4752 printk(KERN_WARNING DRV_NAME
4753 ": Warning: miimon module parameter (%d), "
4754 "not in range 0-%d, so it was reset to %d\n",
4755 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4756 miimon = BOND_LINK_MON_INTERV;
4757 }
4758
4759 if (updelay < 0) {
4760 printk(KERN_WARNING DRV_NAME
4761 ": Warning: updelay module parameter (%d), "
4762 "not in range 0-%d, so it was reset to 0\n",
4763 updelay, INT_MAX);
4764 updelay = 0;
4765 }
4766
4767 if (downdelay < 0) {
4768 printk(KERN_WARNING DRV_NAME
4769 ": Warning: downdelay module parameter (%d), "
4770 "not in range 0-%d, so it was reset to 0\n",
4771 downdelay, INT_MAX);
4772 downdelay = 0;
4773 }
4774
4775 if ((use_carrier != 0) && (use_carrier != 1)) {
4776 printk(KERN_WARNING DRV_NAME
4777 ": Warning: use_carrier module parameter (%d), "
4778 "not of valid value (0/1), so it was set to 1\n",
4779 use_carrier);
4780 use_carrier = 1;
4781 }
4782
7893b249
MS		 4783 if (num_grat_arp < 0 || num_grat_arp > 255) {
4784 printk(KERN_WARNING DRV_NAME
4785 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4786 "was reset to 1 \n", num_grat_arp);
4787 num_grat_arp = 1;
4788 }
4789
1da177e4
LT		 4790 /* reset values for 802.3ad */
4791 if (bond_mode == BOND_MODE_8023AD) {
4792 if (!miimon) {
4793 printk(KERN_WARNING DRV_NAME
4794 ": Warning: miimon must be specified, "
4795 "otherwise bonding will not detect link "
4796 "failure, speed and duplex which are "
4797 "essential for 802.3ad operation\n");
4798 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4799 miimon = 100;
4800 }
4801 }
4802
4803 /* reset values for TLB/ALB */
4804 if ((bond_mode == BOND_MODE_TLB) ||
4805 (bond_mode == BOND_MODE_ALB)) {
4806 if (!miimon) {
4807 printk(KERN_WARNING DRV_NAME
4808 ": Warning: miimon must be specified, "
4809 "otherwise bonding will not detect link "
4810 "failure and link speed which are essential "
4811 "for TLB/ALB load balancing\n");
4812 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4813 miimon = 100;
4814 }
4815 }
4816
4817 if (bond_mode == BOND_MODE_ALB) {
4818 printk(KERN_NOTICE DRV_NAME
4819 ": In ALB mode you might experience client "
4820 "disconnections upon reconnection of a link if the "
4821 "bonding module updelay parameter (%d msec) is "
4822 "incompatible with the forwarding delay time of the "
4823 "switch\n",
4824 updelay);
4825 }
4826
4827 if (!miimon) {
4828 if (updelay || downdelay) {
4829 /* just warn the user the up/down delay will have
4830 * no effect since miimon is zero...
4831 */
4832 printk(KERN_WARNING DRV_NAME
4833 ": Warning: miimon module parameter not set "
4834 "and updelay (%d) or downdelay (%d) module "
4835 "parameter is set; updelay and downdelay have "
4836 "no effect unless miimon is set\n",
4837 updelay, downdelay);
4838 }
4839 } else {
4840 /* don't allow arp monitoring */
4841 if (arp_interval) {
4842 printk(KERN_WARNING DRV_NAME
4843 ": Warning: miimon (%d) and arp_interval (%d) "
4844 "can't be used simultaneously, disabling ARP "
4845 "monitoring\n",
4846 miimon, arp_interval);
4847 arp_interval = 0;
4848 }
4849
4850 if ((updelay % miimon) != 0) {
4851 printk(KERN_WARNING DRV_NAME
4852 ": Warning: updelay (%d) is not a multiple "
4853 "of miimon (%d), updelay rounded to %d ms\n",
4854 updelay, miimon, (updelay / miimon) * miimon);
4855 }
4856
4857 updelay /= miimon;
4858
4859 if ((downdelay % miimon) != 0) {
4860 printk(KERN_WARNING DRV_NAME
4861 ": Warning: downdelay (%d) is not a multiple "
4862 "of miimon (%d), downdelay rounded to %d ms\n",
4863 downdelay, miimon,
4864 (downdelay / miimon) * miimon);
4865 }
4866
4867 downdelay /= miimon;
4868 }
4869
4870 if (arp_interval < 0) {
4871 printk(KERN_WARNING DRV_NAME
4872 ": Warning: arp_interval module parameter (%d) "
4873 ", not in range 0-%d, so it was reset to %d\n",
4874 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4875 arp_interval = BOND_LINK_ARP_INTERV;
4876 }
4877
4878 for (arp_ip_count = 0;
4879 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4880 arp_ip_count++) {
4881 /* not complete check, but should be good enough to
4882 catch mistakes */
4883 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4884 printk(KERN_WARNING DRV_NAME
4885 ": Warning: bad arp_ip_target module parameter "
4886 "(%s), ARP monitoring will not be performed\n",
4887 arp_ip_target[arp_ip_count]);
4888 arp_interval = 0;
4889 } else {
d3bb52b0 4890 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4891 arp_target[arp_ip_count] = ip;
4892 }
4893 }
4894
4895 if (arp_interval && !arp_ip_count) {
4896 /* don't allow arping if no arp_ip_target given... */
4897 printk(KERN_WARNING DRV_NAME
4898 ": Warning: arp_interval module parameter (%d) "
4899 "specified without providing an arp_ip_target "
4900 "parameter, arp_interval was reset to 0\n",
4901 arp_interval);
4902 arp_interval = 0;
4903 }
4904
f5b2b966
JV		 4905 if (arp_validate) {
4906 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4907 printk(KERN_ERR DRV_NAME
4908 ": arp_validate only supported in active-backup mode\n");
4909 return -EINVAL;
4910 }
4911 if (!arp_interval) {
4912 printk(KERN_ERR DRV_NAME
4913 ": arp_validate requires arp_interval\n");
4914 return -EINVAL;
4915 }
4916
4917 arp_validate_value = bond_parse_parm(arp_validate,
4918 arp_validate_tbl);
4919 if (arp_validate_value == -1) {
4920 printk(KERN_ERR DRV_NAME
4921 ": Error: invalid arp_validate \"%s\"\n",
4922 arp_validate == NULL ? "NULL" : arp_validate);
4923 return -EINVAL;
4924 }
4925 } else
4926 arp_validate_value = 0;
4927
1da177e4
LT		 4928 if (miimon) {
4929 printk(KERN_INFO DRV_NAME
4930 ": MII link monitoring set to %d ms\n",
4931 miimon);
4932 } else if (arp_interval) {
4933 int i;
4934
4935 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 4936 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4937 arp_interval,
4938 arp_validate_tbl[arp_validate_value].modename,
4939 arp_ip_count);
1da177e4
LT		 4940
4941 for (i = 0; i < arp_ip_count; i++)
4942 printk (" %s", arp_ip_target[i]);
4943
4944 printk("\n");
4945
4946 } else {
4947 /* miimon and arp_interval not set, we need one so things
4948 * work as expected, see bonding.txt for details
4949 */
4950 printk(KERN_WARNING DRV_NAME
4951 ": Warning: either miimon or arp_interval and "
4952 "arp_ip_target module parameters must be specified, "
4953 "otherwise bonding will not detect link failures! see "
4954 "bonding.txt for details.\n");
4955 }
4956
4957 if (primary && !USES_PRIMARY(bond_mode)) {
4958 /* currently, using a primary only makes sense
4959 * in active backup, TLB or ALB modes
4960 */
4961 printk(KERN_WARNING DRV_NAME
4962 ": Warning: %s primary device specified but has no "
4963 "effect in %s mode\n",
4964 primary, bond_mode_name(bond_mode));
4965 primary = NULL;
4966 }
4967
3915c1e8
JV		 4968 if (fail_over_mac) {
4969 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4970 fail_over_mac_tbl);
4971 if (fail_over_mac_value == -1) {
4972 printk(KERN_ERR DRV_NAME
4973 ": Error: invalid fail_over_mac \"%s\"\n",
4974 arp_validate == NULL ? "NULL" : arp_validate);
4975 return -EINVAL;
4976 }
4977
4978 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4979 printk(KERN_WARNING DRV_NAME
4980 ": Warning: fail_over_mac only affects "
4981 "active-backup mode.\n");
4982 } else {
4983 fail_over_mac_value = BOND_FOM_NONE;
4984 }
dd957c57 4985
1da177e4
LT		 4986 /* fill params struct with the proper values */
4987 params->mode = bond_mode;
169a3e66 4988 params->xmit_policy = xmit_hashtype;
1da177e4 4989 params->miimon = miimon;
7893b249 4990 params->num_grat_arp = num_grat_arp;
1da177e4 4991 params->arp_interval = arp_interval;
f5b2b966 4992 params->arp_validate = arp_validate_value;
1da177e4
LT		 4993 params->updelay = updelay;
4994 params->downdelay = downdelay;
4995 params->use_carrier = use_carrier;
4996 params->lacp_fast = lacp_fast;
4997 params->primary[0] = 0;
3915c1e8 4998 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 4999
5000 if (primary) {
5001 strncpy(params->primary, primary, IFNAMSIZ);
5002 params->primary[IFNAMSIZ - 1] = 0;
5003 }
5004
5005 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5006
5007 return 0;
5008}
5009
0daa2303
PZ		 5010static struct lock_class_key bonding_netdev_xmit_lock_key;
5011
dfe60397 5012/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5013 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5014 * Caller must NOT hold rtnl_lock; we need to release it here before we
5015 * set up our sysfs entries.
5016 */
0dd646fe 5017int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5018{
5019 struct net_device *bond_dev;
0883beca 5020 struct bonding *bond;
dfe60397
MW		 5021 int res;
5022
5023 rtnl_lock();
027ea041
JV		 5024 down_write(&bonding_rwsem);
5025
5026 /* Check to see if the bond already exists. */
4fe4763c 5027 if (name) {
0883beca 5028 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5029 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5030 printk(KERN_ERR DRV_NAME
027ea041 5031 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5032 name);
5033 res = -EPERM;
5034 goto out_rtnl;
5035 }
5036 }
027ea041 5037
e4b91c48
JV		 5038 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5039 ether_setup);
dfe60397
MW		 5040 if (!bond_dev) {
5041 printk(KERN_ERR DRV_NAME
5042 ": %s: eek! can't alloc netdev!\n",
5043 name);
5044 res = -ENOMEM;
5045 goto out_rtnl;
5046 }
5047
e4b91c48
JV		 5048 if (!name) {
5049 res = dev_alloc_name(bond_dev, "bond%d");
5050 if (res < 0)
5051 goto out_netdev;
5052 }
5053
dfe60397
MW		 5054 /* bond_init() must be called after dev_alloc_name() (for the
5055 * /proc files), but before register_netdevice(), because we
5056 * need to set function pointers.
5057 */
5058
5059 res = bond_init(bond_dev, params);
5060 if (res < 0) {
5061 goto out_netdev;
5062 }
5063
dfe60397
MW		 5064 res = register_netdevice(bond_dev);
5065 if (res < 0) {
5066 goto out_bond;
5067 }
0daa2303
PZ		 5068
5069 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
5070
ff59c456
JV		 5071 netif_carrier_off(bond_dev);
5072
027ea041 5073 up_write(&bonding_rwsem);
dfe60397 5074 rtnl_unlock(); /* allows sysfs registration of net device */
b76cdba9 5075 res = bond_create_sysfs_entry(bond_dev->priv);
09c89279
JV		 5076 if (res < 0) {
5077 rtnl_lock();
027ea041 5078 down_write(&bonding_rwsem);
822973ba
PE		 5079 bond_deinit(bond_dev);
5080 unregister_netdevice(bond_dev);
5081 goto out_rtnl;
09c89279
JV		 5082 }
5083
5084 return 0;
5085
dfe60397
MW		 5086out_bond:
5087 bond_deinit(bond_dev);
5088out_netdev:
5089 free_netdev(bond_dev);
5090out_rtnl:
027ea041 5091 up_write(&bonding_rwsem);
dfe60397 5092 rtnl_unlock();
dfe60397
MW		 5093 return res;
5094}
5095
1da177e4
LT		 5096static int __init bonding_init(void)
5097{
1da177e4
LT		 5098 int i;
5099 int res;
0883beca 5100 struct bonding *bond;
1da177e4
LT		 5101
5102 printk(KERN_INFO "%s", version);
5103
dfe60397 5104 res = bond_check_params(&bonding_defaults);
1da177e4 5105 if (res) {
dfe60397 5106 goto out;
1da177e4
LT		 5107 }
5108
1da177e4
LT		 5109#ifdef CONFIG_PROC_FS
5110 bond_create_proc_dir();
5111#endif
027ea041
JV		 5112
5113 init_rwsem(&bonding_rwsem);
5114
1da177e4 5115 for (i = 0; i < max_bonds; i++) {
0dd646fe 5116 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5117 if (res)
5118 goto err;
1da177e4
LT		 5119 }
5120
b76cdba9
MW		 5121 res = bond_create_sysfs();
5122 if (res)
5123 goto err;
5124
1da177e4 5125 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5126 register_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5127
dfe60397
MW		 5128 goto out;
5129err:
0883beca 5130 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5131 bond_work_cancel_all(bond);
5132 destroy_workqueue(bond->wq);
5133 }
5134
ae68c398
PE		 5135 bond_destroy_sysfs();
5136
40abc270 5137 rtnl_lock();
1da177e4 5138 bond_free_all();
1da177e4 5139 rtnl_unlock();
dfe60397 5140out:
1da177e4 5141 return res;
dfe60397 5142
1da177e4
LT		 5143}
5144
5145static void __exit bonding_exit(void)
5146{
5147 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5148 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5149
ae68c398
PE		 5150 bond_destroy_sysfs();
5151
1da177e4
LT		 5152 rtnl_lock();
5153 bond_free_all();
5154 rtnl_unlock();
5155}
5156
5157module_init(bonding_init);
5158module_exit(bonding_exit);
5159MODULE_LICENSE("GPL");
5160MODULE_VERSION(DRV_VERSION);
5161MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5162MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5163MODULE_SUPPORTED_DEVICE("most ethernet devices");
5164
5165/*
5166 * Local variables:
5167 * c-indent-level: 8
5168 * c-basic-offset: 8
5169 * tab-width: 8
5170 * End:
5171 */
5172
Clone of davem's net-next-2.6 tree