]> 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
net/core: add NETDEV_BONDING_FAILOVER event
[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);
1192 }
2ab82852 1193
dd957c57 1194 if (new_active && bond->params.fail_over_mac)
3915c1e8
JV		 1195 bond_do_fail_over_mac(bond, new_active, old_active);
1196
7893b249 1197 bond->send_grat_arp = bond->params.num_grat_arp;
1053f62c
MS		 1198 if (bond->curr_active_slave &&
1199 test_bit(__LINK_STATE_LINKWATCH_PENDING,
1200 &bond->curr_active_slave->dev->state)) {
1201 dprintk("delaying gratuitous arp on %s\n",
1202 bond->curr_active_slave->dev->name);
7893b249
MS		 1203 } else {
1204 if (bond->send_grat_arp > 0) {
1205 bond_send_gratuitous_arp(bond);
1206 bond->send_grat_arp--;
1207 }
1208 }
c3ade5ca 1209 }
1da177e4
LT		 1210}
1211
1212/**
1213 * bond_select_active_slave - select a new active slave, if needed
1214 * @bond: our bonding struct
1215 *
1216 * This functions shoud be called when one of the following occurs:
1217 * - The old curr_active_slave has been released or lost its link.
1218 * - The primary_slave has got its link back.
1219 * - A slave has got its link back and there's no old curr_active_slave.
1220 *
3915c1e8 1221 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1222 */
a77b5325 1223void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1224{
1225 struct slave *best_slave;
ff59c456 1226 int rv;
1da177e4
LT		 1227
1228 best_slave = bond_find_best_slave(bond);
1229 if (best_slave != bond->curr_active_slave) {
1230 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1231 rv = bond_set_carrier(bond);
1232 if (!rv)
1233 return;
1234
1235 if (netif_carrier_ok(bond->dev)) {
1236 printk(KERN_INFO DRV_NAME
1237 ": %s: first active interface up!\n",
1238 bond->dev->name);
1239 } else {
1240 printk(KERN_INFO DRV_NAME ": %s: "
1241 "now running without any active interface !\n",
1242 bond->dev->name);
1243 }
1da177e4
LT		 1244 }
1245}
1246
1247/*--------------------------- slave list handling ---------------------------*/
1248
1249/*
1250 * This function attaches the slave to the end of list.
1251 *
1252 * bond->lock held for writing by caller.
1253 */
1254static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1255{
1256 if (bond->first_slave == NULL) { /* attaching the first slave */
1257 new_slave->next = new_slave;
1258 new_slave->prev = new_slave;
1259 bond->first_slave = new_slave;
1260 } else {
1261 new_slave->next = bond->first_slave;
1262 new_slave->prev = bond->first_slave->prev;
1263 new_slave->next->prev = new_slave;
1264 new_slave->prev->next = new_slave;
1265 }
1266
1267 bond->slave_cnt++;
1268}
1269
1270/*
1271 * This function detaches the slave from the list.
1272 * WARNING: no check is made to verify if the slave effectively
1273 * belongs to <bond>.
1274 * Nothing is freed on return, structures are just unchained.
1275 * If any slave pointer in bond was pointing to <slave>,
1276 * it should be changed by the calling function.
1277 *
1278 * bond->lock held for writing by caller.
1279 */
1280static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1281{
1282 if (slave->next) {
1283 slave->next->prev = slave->prev;
1284 }
1285
1286 if (slave->prev) {
1287 slave->prev->next = slave->next;
1288 }
1289
1290 if (bond->first_slave == slave) { /* slave is the first slave */
1291 if (bond->slave_cnt > 1) { /* there are more slave */
1292 bond->first_slave = slave->next;
1293 } else {
1294 bond->first_slave = NULL; /* slave was the last one */
1295 }
1296 }
1297
1298 slave->next = NULL;
1299 slave->prev = NULL;
1300 bond->slave_cnt--;
1301}
1302
1303/*---------------------------------- IOCTL ----------------------------------*/
1304
4ad072c9
AB		 1305static int bond_sethwaddr(struct net_device *bond_dev,
1306 struct net_device *slave_dev)
1da177e4
LT		 1307{
1308 dprintk("bond_dev=%p\n", bond_dev);
1309 dprintk("slave_dev=%p\n", slave_dev);
1310 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1311 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1312 return 0;
1313}
1314
7f353bf2
HX		 1315#define BOND_VLAN_FEATURES \
1316 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1317 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK		 1318
1319/*
8e3babcd 1320 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX		 1321 * feature bits are managed elsewhere, so preserve those feature bits
1322 * on the master device.
8531c5ff
AK		 1323 */
1324static int bond_compute_features(struct bonding *bond)
1325{
8531c5ff
AK		 1326 struct slave *slave;
1327 struct net_device *bond_dev = bond->dev;
7f353bf2 1328 unsigned long features = bond_dev->features;
3158bf7d
MS		 1329 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1330 bond_dev->hard_header_len);
8e3babcd 1331 int i;
8531c5ff 1332
7f353bf2
HX		 1333 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1334 features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1335 NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1336
54ef3137 1337 bond_for_each_slave(bond, slave, i) {
7f353bf2
HX		 1338 features = netdev_compute_features(features,
1339 slave->dev->features);
54ef3137
JV		 1340 if (slave->dev->hard_header_len > max_hard_header_len)
1341 max_hard_header_len = slave->dev->hard_header_len;
1342 }
8531c5ff 1343
7f353bf2 1344 features |= (bond_dev->features & BOND_VLAN_FEATURES);
8531c5ff 1345 bond_dev->features = features;
54ef3137 1346 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1347
1348 return 0;
1349}
1350
872254dd
MS		 1351
1352static void bond_setup_by_slave(struct net_device *bond_dev,
1353 struct net_device *slave_dev)
1354{
d90a162a
MS		 1355 struct bonding *bond = bond_dev->priv;
1356
872254dd 1357 bond_dev->neigh_setup = slave_dev->neigh_setup;
1284cd3a 1358 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1359
1360 bond_dev->type = slave_dev->type;
1361 bond_dev->hard_header_len = slave_dev->hard_header_len;
1362 bond_dev->addr_len = slave_dev->addr_len;
1363
1364 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1365 slave_dev->addr_len);
d90a162a 1366 bond->setup_by_slave = 1;
872254dd
MS		 1367}
1368
1da177e4 1369/* enslave device <slave> to bond device <master> */
a77b5325 1370int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1371{
1372 struct bonding *bond = bond_dev->priv;
1373 struct slave *new_slave = NULL;
1374 struct dev_mc_list *dmi;
1375 struct sockaddr addr;
1376 int link_reporting;
1377 int old_features = bond_dev->features;
1378 int res = 0;
1379
552709d5 1380 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1381 slave_dev->do_ioctl == NULL) {
1da177e4 1382 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1383 ": %s: Warning: no link monitoring support for %s\n",
1384 bond_dev->name, slave_dev->name);
1da177e4
LT		 1385 }
1386
1387 /* bond must be initialized by bond_open() before enslaving */
1388 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1389 printk(KERN_WARNING DRV_NAME
1390 " %s: master_dev is not up in bond_enslave\n",
1391 bond_dev->name);
1da177e4
LT		 1392 }
1393
1394 /* already enslaved */
1395 if (slave_dev->flags & IFF_SLAVE) {
1396 dprintk("Error, Device was already enslaved\n");
1397 return -EBUSY;
1398 }
1399
1400 /* vlan challenged mutual exclusion */
1401 /* no need to lock since we're protected by rtnl_lock */
1402 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1403 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1404 if (!list_empty(&bond->vlan_list)) {
1405 printk(KERN_ERR DRV_NAME
4e0952c7 1406 ": %s: Error: cannot enslave VLAN "
1da177e4 1407 "challenged slave %s on VLAN enabled "
4e0952c7 1408 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1409 bond_dev->name);
1410 return -EPERM;
1411 } else {
1412 printk(KERN_WARNING DRV_NAME
4e0952c7 1413 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1414 "slave %s. Adding VLANs will be blocked as "
1415 "long as %s is part of bond %s\n",
4e0952c7 1416 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1417 bond_dev->name);
1418 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1419 }
1420 } else {
1421 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1422 if (bond->slave_cnt == 0) {
1423 /* First slave, and it is not VLAN challenged,
1424 * so remove the block of adding VLANs over the bond.
1425 */
1426 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1427 }
1428 }
1429
217df670
JV		 1430 /*
1431 * Old ifenslave binaries are no longer supported. These can
1432 * be identified with moderate accurary by the state of the slave:
1433 * the current ifenslave will set the interface down prior to
1434 * enslaving it; the old ifenslave will not.
1435 */
1436 if ((slave_dev->flags & IFF_UP)) {
1437 printk(KERN_ERR DRV_NAME ": %s is up. "
1438 "This may be due to an out of date ifenslave.\n",
1439 slave_dev->name);
1440 res = -EPERM;
1441 goto err_undo_flags;
1442 }
1da177e4 1443
872254dd
MS		 1444 /* set bonding device ether type by slave - bonding netdevices are
1445 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1446 * there is a need to override some of the type dependent attribs/funcs.
1447 *
1448 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1449 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1450 */
1451 if (bond->slave_cnt == 0) {
1452 if (slave_dev->type != ARPHRD_ETHER)
1453 bond_setup_by_slave(bond_dev, slave_dev);
1454 } else if (bond_dev->type != slave_dev->type) {
1455 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1456 "from other slaves (%d), can not enslave it.\n",
1457 slave_dev->name,
1458 slave_dev->type, bond_dev->type);
1459 res = -EINVAL;
1460 goto err_undo_flags;
1461 }
1462
217df670 1463 if (slave_dev->set_mac_address == NULL) {
2ab82852
MS		 1464 if (bond->slave_cnt == 0) {
1465 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1466 ": %s: Warning: The first slave device "
1467 "specified does not support setting the MAC "
3915c1e8 1468 "address. Setting fail_over_mac to active.",
dd957c57 1469 bond_dev->name);
3915c1e8
JV		 1470 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1471 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1472 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1473 ": %s: Error: The slave device specified "
1474 "does not support setting the MAC address, "
3915c1e8 1475 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1476 , bond_dev->name);
1477 res = -EOPNOTSUPP;
1478 goto err_undo_flags;
1479 }
1da177e4
LT		 1480 }
1481
243cb4e5 1482 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1483 if (!new_slave) {
1484 res = -ENOMEM;
1485 goto err_undo_flags;
1486 }
1487
1da177e4
LT		 1488 /* save slave's original flags before calling
1489 * netdev_set_master and dev_open
1490 */
1491 new_slave->original_flags = slave_dev->flags;
1492
217df670
JV		 1493 /*
1494 * Save slave's original ("permanent") mac address for modes
1495 * that need it, and for restoring it upon release, and then
1496 * set it to the master's address
1497 */
1498 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1499
dd957c57 1500 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1501 /*
1502 * Set slave to master's mac address. The application already
1503 * set the master's mac address to that of the first slave
1504 */
1505 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1506 addr.sa_family = slave_dev->type;
1507 res = dev_set_mac_address(slave_dev, &addr);
1508 if (res) {
1509 dprintk("Error %d calling set_mac_address\n", res);
1510 goto err_free;
1511 }
217df670 1512 }
1da177e4 1513
c2edacf8
JV		 1514 res = netdev_set_master(slave_dev, bond_dev);
1515 if (res) {
1516 dprintk("Error %d calling netdev_set_master\n", res);
569f0c4d 1517 goto err_restore_mac;
c2edacf8 1518 }
217df670
JV		 1519 /* open the slave since the application closed it */
1520 res = dev_open(slave_dev);
1521 if (res) {
1522 dprintk("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1523 goto err_unset_master;
1da177e4
LT		 1524 }
1525
1da177e4 1526 new_slave->dev = slave_dev;
0b680e75 1527 slave_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 1528
1529 if ((bond->params.mode == BOND_MODE_TLB) ||
1530 (bond->params.mode == BOND_MODE_ALB)) {
1531 /* bond_alb_init_slave() must be called before all other stages since
1532 * it might fail and we do not want to have to undo everything
1533 */
1534 res = bond_alb_init_slave(bond, new_slave);
1535 if (res) {
569f0c4d 1536 goto err_close;
1da177e4
LT		 1537 }
1538 }
1539
1540 /* If the mode USES_PRIMARY, then the new slave gets the
1541 * master's promisc (and mc) settings only if it becomes the
1542 * curr_active_slave, and that is taken care of later when calling
1543 * bond_change_active()
1544 */
1545 if (!USES_PRIMARY(bond->params.mode)) {
1546 /* set promiscuity level to new slave */
1547 if (bond_dev->flags & IFF_PROMISC) {
1548 dev_set_promiscuity(slave_dev, 1);
1549 }
1550
1551 /* set allmulti level to new slave */
1552 if (bond_dev->flags & IFF_ALLMULTI) {
1553 dev_set_allmulti(slave_dev, 1);
1554 }
1555
80ee5ad2 1556 netif_tx_lock_bh(bond_dev);
1da177e4
LT		 1557 /* upload master's mc_list to new slave */
1558 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1559 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1560 }
80ee5ad2 1561 netif_tx_unlock_bh(bond_dev);
1da177e4
LT		 1562 }
1563
1564 if (bond->params.mode == BOND_MODE_8023AD) {
1565 /* add lacpdu mc addr to mc list */
1566 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1567
1568 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1569 }
1570
1571 bond_add_vlans_on_slave(bond, slave_dev);
1572
1573 write_lock_bh(&bond->lock);
1574
1575 bond_attach_slave(bond, new_slave);
1576
1577 new_slave->delay = 0;
1578 new_slave->link_failure_count = 0;
1579
8531c5ff
AK		 1580 bond_compute_features(bond);
1581
3915c1e8
JV		 1582 write_unlock_bh(&bond->lock);
1583
1584 read_lock(&bond->lock);
1585
f5b2b966
JV		 1586 new_slave->last_arp_rx = jiffies;
1587
1da177e4
LT		 1588 if (bond->params.miimon && !bond->params.use_carrier) {
1589 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1590
1591 if ((link_reporting == -1) && !bond->params.arp_interval) {
1592 /*
1593 * miimon is set but a bonded network driver
1594 * does not support ETHTOOL/MII and
1595 * arp_interval is not set. Note: if
1596 * use_carrier is enabled, we will never go
1597 * here (because netif_carrier is always
1598 * supported); thus, we don't need to change
1599 * the messages for netif_carrier.
1600 */
1601 printk(KERN_WARNING DRV_NAME
4e0952c7 1602 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1603 "available for interface %s, and "
1604 "arp_interval/arp_ip_target module parameters "
1605 "not specified, thus bonding will not detect "
1606 "link failures! see bonding.txt for details.\n",
4e0952c7 1607 bond_dev->name, slave_dev->name);
1da177e4
LT		 1608 } else if (link_reporting == -1) {
1609 /* unable get link status using mii/ethtool */
1610 printk(KERN_WARNING DRV_NAME
4e0952c7 1611 ": %s: Warning: can't get link status from "
1da177e4
LT		 1612 "interface %s; the network driver associated "
1613 "with this interface does not support MII or "
1614 "ETHTOOL link status reporting, thus miimon "
1615 "has no effect on this interface.\n",
4e0952c7 1616 bond_dev->name, slave_dev->name);
1da177e4
LT		 1617 }
1618 }
1619
1620 /* check for initial state */
1621 if (!bond->params.miimon ||
1622 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1623 if (bond->params.updelay) {
1624 dprintk("Initial state of slave_dev is "
1625 "BOND_LINK_BACK\n");
1626 new_slave->link = BOND_LINK_BACK;
1627 new_slave->delay = bond->params.updelay;
1628 } else {
1629 dprintk("Initial state of slave_dev is "
1630 "BOND_LINK_UP\n");
1631 new_slave->link = BOND_LINK_UP;
1632 }
1633 new_slave->jiffies = jiffies;
1634 } else {
1635 dprintk("Initial state of slave_dev is "
1636 "BOND_LINK_DOWN\n");
1637 new_slave->link = BOND_LINK_DOWN;
1638 }
1639
1640 if (bond_update_speed_duplex(new_slave) &&
1641 (new_slave->link != BOND_LINK_DOWN)) {
1642 printk(KERN_WARNING DRV_NAME
4e0952c7 1643 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1644 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1645 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1646
1647 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1648 printk(KERN_WARNING DRV_NAME
1649 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1650 "support in base driver for proper aggregator "
4e0952c7 1651 "selection.\n", bond_dev->name);
1da177e4
LT		 1652 }
1653 }
1654
1655 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1656 /* if there is a primary slave, remember it */
1657 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1658 bond->primary_slave = new_slave;
1659 }
1660 }
1661
3915c1e8
JV		 1662 write_lock_bh(&bond->curr_slave_lock);
1663
1da177e4
LT		 1664 switch (bond->params.mode) {
1665 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1666 bond_set_slave_inactive_flags(new_slave);
1667 bond_select_active_slave(bond);
1da177e4
LT		 1668 break;
1669 case BOND_MODE_8023AD:
1670 /* in 802.3ad mode, the internal mechanism
1671 * will activate the slaves in the selected
1672 * aggregator
1673 */
1674 bond_set_slave_inactive_flags(new_slave);
1675 /* if this is the first slave */
1676 if (bond->slave_cnt == 1) {
1677 SLAVE_AD_INFO(new_slave).id = 1;
1678 /* Initialize AD with the number of times that the AD timer is called in 1 second
1679 * can be called only after the mac address of the bond is set
1680 */
1681 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1682 bond->params.lacp_fast);
1683 } else {
1684 SLAVE_AD_INFO(new_slave).id =
1685 SLAVE_AD_INFO(new_slave->prev).id + 1;
1686 }
1687
1688 bond_3ad_bind_slave(new_slave);
1689 break;
1690 case BOND_MODE_TLB:
1691 case BOND_MODE_ALB:
1692 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1693 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1694 break;
1695 default:
1696 dprintk("This slave is always active in trunk mode\n");
1697
1698 /* always active in trunk mode */
1699 new_slave->state = BOND_STATE_ACTIVE;
1700
1701 /* In trunking mode there is little meaning to curr_active_slave
1702 * anyway (it holds no special properties of the bond device),
1703 * so we can change it without calling change_active_interface()
1704 */
1705 if (!bond->curr_active_slave) {
1706 bond->curr_active_slave = new_slave;
1707 }
1708 break;
1709 } /* switch(bond_mode) */
1710
3915c1e8
JV		 1711 write_unlock_bh(&bond->curr_slave_lock);
1712
ff59c456
JV		 1713 bond_set_carrier(bond);
1714
3915c1e8 1715 read_unlock(&bond->lock);
1da177e4 1716
b76cdba9
MW		 1717 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1718 if (res)
569f0c4d 1719 goto err_close;
b76cdba9 1720
1da177e4
LT		 1721 printk(KERN_INFO DRV_NAME
1722 ": %s: enslaving %s as a%s interface with a%s link.\n",
1723 bond_dev->name, slave_dev->name,
1724 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1725 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1726
1727 /* enslave is successful */
1728 return 0;
1729
1730/* Undo stages on error */
1da177e4
LT		 1731err_close:
1732 dev_close(slave_dev);
1733
569f0c4d
JV		 1734err_unset_master:
1735 netdev_set_master(slave_dev, NULL);
1736
1da177e4 1737err_restore_mac:
dd957c57 1738 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1739 /* XXX TODO - fom follow mode needs to change master's
1740 * MAC if this slave's MAC is in use by the bond, or at
1741 * least print a warning.
1742 */
2ab82852
MS		 1743 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1744 addr.sa_family = slave_dev->type;
1745 dev_set_mac_address(slave_dev, &addr);
1746 }
1da177e4
LT		 1747
1748err_free:
1749 kfree(new_slave);
1750
1751err_undo_flags:
1752 bond_dev->features = old_features;
8531c5ff 1753
1da177e4
LT		 1754 return res;
1755}
1756
1757/*
1758 * Try to release the slave device <slave> from the bond device <master>
1759 * It is legal to access curr_active_slave without a lock because all the function
1760 * is write-locked.
1761 *
1762 * The rules for slave state should be:
1763 * for Active/Backup:
1764 * Active stays on all backups go down
1765 * for Bonded connections:
1766 * The first up interface should be left on and all others downed.
1767 */
a77b5325 1768int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1769{
1770 struct bonding *bond = bond_dev->priv;
1771 struct slave *slave, *oldcurrent;
1772 struct sockaddr addr;
1773 int mac_addr_differ;
0795af57 1774 DECLARE_MAC_BUF(mac);
1da177e4
LT		 1775
1776 /* slave is not a slave or master is not master of this slave */
1777 if (!(slave_dev->flags & IFF_SLAVE) ||
1778 (slave_dev->master != bond_dev)) {
1779 printk(KERN_ERR DRV_NAME
4e0952c7 1780 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1781 bond_dev->name, slave_dev->name);
1782 return -EINVAL;
1783 }
1784
1785 write_lock_bh(&bond->lock);
1786
1787 slave = bond_get_slave_by_dev(bond, slave_dev);
1788 if (!slave) {
1789 /* not a slave of this bond */
1790 printk(KERN_INFO DRV_NAME
1791 ": %s: %s not enslaved\n",
1792 bond_dev->name, slave_dev->name);
f5e2a7b2 1793 write_unlock_bh(&bond->lock);
1da177e4
LT		 1794 return -EINVAL;
1795 }
1796
3915c1e8
JV		 1797 if (!bond->params.fail_over_mac) {
1798 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1799 ETH_ALEN);
1800 if (!mac_addr_differ && (bond->slave_cnt > 1))
1801 printk(KERN_WARNING DRV_NAME
1802 ": %s: Warning: the permanent HWaddr of %s - "
1803 "%s - is still in use by %s. "
1804 "Set the HWaddr of %s to a different address "
1805 "to avoid conflicts.\n",
1806 bond_dev->name, slave_dev->name,
1807 print_mac(mac, slave->perm_hwaddr),
1808 bond_dev->name, slave_dev->name);
1da177e4
LT		 1809 }
1810
1811 /* Inform AD package of unbinding of slave. */
1812 if (bond->params.mode == BOND_MODE_8023AD) {
1813 /* must be called before the slave is
1814 * detached from the list
1815 */
1816 bond_3ad_unbind_slave(slave);
1817 }
1818
1819 printk(KERN_INFO DRV_NAME
1820 ": %s: releasing %s interface %s\n",
1821 bond_dev->name,
1822 (slave->state == BOND_STATE_ACTIVE)
1823 ? "active" : "backup",
1824 slave_dev->name);
1825
1826 oldcurrent = bond->curr_active_slave;
1827
1828 bond->current_arp_slave = NULL;
1829
1830 /* release the slave from its bond */
1831 bond_detach_slave(bond, slave);
1832
8531c5ff
AK		 1833 bond_compute_features(bond);
1834
1da177e4
LT		 1835 if (bond->primary_slave == slave) {
1836 bond->primary_slave = NULL;
1837 }
1838
1839 if (oldcurrent == slave) {
1840 bond_change_active_slave(bond, NULL);
1841 }
1842
1843 if ((bond->params.mode == BOND_MODE_TLB) ||
1844 (bond->params.mode == BOND_MODE_ALB)) {
1845 /* Must be called only after the slave has been
1846 * detached from the list and the curr_active_slave
1847 * has been cleared (if our_slave == old_current),
1848 * but before a new active slave is selected.
1849 */
2543331d 1850 write_unlock_bh(&bond->lock);
1da177e4 1851 bond_alb_deinit_slave(bond, slave);
2543331d 1852 write_lock_bh(&bond->lock);
1da177e4
LT		 1853 }
1854
059fe7a5
JV		 1855 if (oldcurrent == slave) {
1856 /*
1857 * Note that we hold RTNL over this sequence, so there
1858 * is no concern that another slave add/remove event
1859 * will interfere.
1860 */
1861 write_unlock_bh(&bond->lock);
1862 read_lock(&bond->lock);
1863 write_lock_bh(&bond->curr_slave_lock);
1864
1da177e4
LT		 1865 bond_select_active_slave(bond);
1866
059fe7a5
JV		 1867 write_unlock_bh(&bond->curr_slave_lock);
1868 read_unlock(&bond->lock);
1869 write_lock_bh(&bond->lock);
1870 }
1871
1da177e4 1872 if (bond->slave_cnt == 0) {
ff59c456
JV		 1873 bond_set_carrier(bond);
1874
1da177e4
LT		 1875 /* if the last slave was removed, zero the mac address
1876 * of the master so it will be set by the application
1877 * to the mac address of the first slave
1878 */
1879 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1880
1881 if (list_empty(&bond->vlan_list)) {
1882 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1883 } else {
1884 printk(KERN_WARNING DRV_NAME
4e0952c7 1885 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1886 "still has VLANs.\n",
4e0952c7 1887 bond_dev->name, bond_dev->name);
1da177e4 1888 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1889 ": %s: When re-adding slaves, make sure the bond's "
1890 "HW address matches its VLANs'.\n",
1891 bond_dev->name);
1da177e4
LT		 1892 }
1893 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1894 !bond_has_challenged_slaves(bond)) {
1895 printk(KERN_INFO DRV_NAME
4e0952c7 1896 ": %s: last VLAN challenged slave %s "
1da177e4 1897 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1898 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1899 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1900 }
1901
1902 write_unlock_bh(&bond->lock);
1903
b76cdba9
MW		 1904 /* must do this from outside any spinlocks */
1905 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1906
1da177e4
LT		 1907 bond_del_vlans_from_slave(bond, slave_dev);
1908
1909 /* If the mode USES_PRIMARY, then we should only remove its
1910 * promisc and mc settings if it was the curr_active_slave, but that was
1911 * already taken care of above when we detached the slave
1912 */
1913 if (!USES_PRIMARY(bond->params.mode)) {
1914 /* unset promiscuity level from slave */
1915 if (bond_dev->flags & IFF_PROMISC) {
1916 dev_set_promiscuity(slave_dev, -1);
1917 }
1918
1919 /* unset allmulti level from slave */
1920 if (bond_dev->flags & IFF_ALLMULTI) {
1921 dev_set_allmulti(slave_dev, -1);
1922 }
1923
1924 /* flush master's mc_list from slave */
80ee5ad2 1925 netif_tx_lock_bh(bond_dev);
1da177e4 1926 bond_mc_list_flush(bond_dev, slave_dev);
80ee5ad2 1927 netif_tx_unlock_bh(bond_dev);
1da177e4
LT		 1928 }
1929
1930 netdev_set_master(slave_dev, NULL);
1931
1932 /* close slave before restoring its mac address */
1933 dev_close(slave_dev);
1934
3915c1e8 1935 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1936 /* restore original ("permanent") mac address */
1937 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1938 addr.sa_family = slave_dev->type;
1939 dev_set_mac_address(slave_dev, &addr);
1940 }
1da177e4 1941
8f903c70 1942 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1943 IFF_SLAVE_INACTIVE | IFF_BONDING |
1944 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1945
1946 kfree(slave);
1947
1948 return 0; /* deletion OK */
1949}
1950
d90a162a
MS		 1951/*
1952* Destroy a bonding device.
1953* Must be under rtnl_lock when this function is called.
1954*/
1955void bond_destroy(struct bonding *bond)
1956{
1957 bond_deinit(bond->dev);
1958 bond_destroy_sysfs_entry(bond);
8cbdeec6 1959 unregister_netdevice(bond->dev);
d90a162a
MS		 1960}
1961
1962/*
1963* First release a slave and than destroy the bond if no more slaves iare left.
1964* Must be under rtnl_lock when this function is called.
1965*/
1966int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1967{
1968 struct bonding *bond = bond_dev->priv;
1969 int ret;
1970
1971 ret = bond_release(bond_dev, slave_dev);
1972 if ((ret == 0) && (bond->slave_cnt == 0)) {
1973 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1974 bond_dev->name, bond_dev->name);
1975 bond_destroy(bond);
1976 }
1977 return ret;
1978}
1979
1da177e4
LT		 1980/*
1981 * This function releases all slaves.
1982 */
1983static int bond_release_all(struct net_device *bond_dev)
1984{
1985 struct bonding *bond = bond_dev->priv;
1986 struct slave *slave;
1987 struct net_device *slave_dev;
1988 struct sockaddr addr;
1989
1990 write_lock_bh(&bond->lock);
1991
ff59c456
JV		 1992 netif_carrier_off(bond_dev);
1993
1da177e4
LT		 1994 if (bond->slave_cnt == 0) {
1995 goto out;
1996 }
1997
1998 bond->current_arp_slave = NULL;
1999 bond->primary_slave = NULL;
2000 bond_change_active_slave(bond, NULL);
2001
2002 while ((slave = bond->first_slave) != NULL) {
2003 /* Inform AD package of unbinding of slave
2004 * before slave is detached from the list.
2005 */
2006 if (bond->params.mode == BOND_MODE_8023AD) {
2007 bond_3ad_unbind_slave(slave);
2008 }
2009
2010 slave_dev = slave->dev;
2011 bond_detach_slave(bond, slave);
2012
2543331d
JV		 2013 /* now that the slave is detached, unlock and perform
2014 * all the undo steps that should not be called from
2015 * within a lock.
2016 */
2017 write_unlock_bh(&bond->lock);
2018
1da177e4
LT		 2019 if ((bond->params.mode == BOND_MODE_TLB) ||
2020 (bond->params.mode == BOND_MODE_ALB)) {
2021 /* must be called only after the slave
2022 * has been detached from the list
2023 */
2024 bond_alb_deinit_slave(bond, slave);
2025 }
2026
8531c5ff
AK		 2027 bond_compute_features(bond);
2028
b76cdba9 2029 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2030 bond_del_vlans_from_slave(bond, slave_dev);
2031
2032 /* If the mode USES_PRIMARY, then we should only remove its
2033 * promisc and mc settings if it was the curr_active_slave, but that was
2034 * already taken care of above when we detached the slave
2035 */
2036 if (!USES_PRIMARY(bond->params.mode)) {
2037 /* unset promiscuity level from slave */
2038 if (bond_dev->flags & IFF_PROMISC) {
2039 dev_set_promiscuity(slave_dev, -1);
2040 }
2041
2042 /* unset allmulti level from slave */
2043 if (bond_dev->flags & IFF_ALLMULTI) {
2044 dev_set_allmulti(slave_dev, -1);
2045 }
2046
2047 /* flush master's mc_list from slave */
80ee5ad2 2048 netif_tx_lock_bh(bond_dev);
1da177e4 2049 bond_mc_list_flush(bond_dev, slave_dev);
80ee5ad2 2050 netif_tx_unlock_bh(bond_dev);
1da177e4
LT		 2051 }
2052
2053 netdev_set_master(slave_dev, NULL);
2054
2055 /* close slave before restoring its mac address */
2056 dev_close(slave_dev);
2057
dd957c57 2058 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2059 /* restore original ("permanent") mac address*/
2060 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2061 addr.sa_family = slave_dev->type;
2062 dev_set_mac_address(slave_dev, &addr);
2063 }
1da177e4 2064
8f903c70
JV		 2065 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2066 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2067
2068 kfree(slave);
2069
2070 /* re-acquire the lock before getting the next slave */
2071 write_lock_bh(&bond->lock);
2072 }
2073
2074 /* zero the mac address of the master so it will be
2075 * set by the application to the mac address of the
2076 * first slave
2077 */
2078 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2079
2080 if (list_empty(&bond->vlan_list)) {
2081 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2082 } else {
2083 printk(KERN_WARNING DRV_NAME
4e0952c7 2084 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2085 "still has VLANs.\n",
4e0952c7 2086 bond_dev->name, bond_dev->name);
1da177e4 2087 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2088 ": %s: When re-adding slaves, make sure the bond's "
2089 "HW address matches its VLANs'.\n",
2090 bond_dev->name);
1da177e4
LT		 2091 }
2092
2093 printk(KERN_INFO DRV_NAME
2094 ": %s: released all slaves\n",
2095 bond_dev->name);
2096
2097out:
2098 write_unlock_bh(&bond->lock);
2099
2100 return 0;
2101}
2102
2103/*
2104 * This function changes the active slave to slave <slave_dev>.
2105 * It returns -EINVAL in the following cases.
2106 * - <slave_dev> is not found in the list.
2107 * - There is not active slave now.
2108 * - <slave_dev> is already active.
2109 * - The link state of <slave_dev> is not BOND_LINK_UP.
2110 * - <slave_dev> is not running.
2111 * In these cases, this fuction does nothing.
2112 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2113 */
2114static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2115{
2116 struct bonding *bond = bond_dev->priv;
2117 struct slave *old_active = NULL;
2118 struct slave *new_active = NULL;
2119 int res = 0;
2120
2121 if (!USES_PRIMARY(bond->params.mode)) {
2122 return -EINVAL;
2123 }
2124
2125 /* Verify that master_dev is indeed the master of slave_dev */
2126 if (!(slave_dev->flags & IFF_SLAVE) ||
2127 (slave_dev->master != bond_dev)) {
2128 return -EINVAL;
2129 }
2130
059fe7a5 2131 read_lock(&bond->lock);
1da177e4 2132
059fe7a5 2133 read_lock(&bond->curr_slave_lock);
1da177e4 2134 old_active = bond->curr_active_slave;
059fe7a5
JV		 2135 read_unlock(&bond->curr_slave_lock);
2136
1da177e4
LT		 2137 new_active = bond_get_slave_by_dev(bond, slave_dev);
2138
2139 /*
2140 * Changing to the current active: do nothing; return success.
2141 */
2142 if (new_active && (new_active == old_active)) {
059fe7a5 2143 read_unlock(&bond->lock);
1da177e4
LT		 2144 return 0;
2145 }
2146
2147 if ((new_active) &&
2148 (old_active) &&
2149 (new_active->link == BOND_LINK_UP) &&
2150 IS_UP(new_active->dev)) {
059fe7a5 2151 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2152 bond_change_active_slave(bond, new_active);
059fe7a5 2153 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2154 } else {
2155 res = -EINVAL;
2156 }
2157
059fe7a5 2158 read_unlock(&bond->lock);
1da177e4
LT		 2159
2160 return res;
2161}
2162
1da177e4
LT		 2163static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2164{
2165 struct bonding *bond = bond_dev->priv;
2166
2167 info->bond_mode = bond->params.mode;
2168 info->miimon = bond->params.miimon;
2169
6603a6f2 2170 read_lock(&bond->lock);
1da177e4 2171 info->num_slaves = bond->slave_cnt;
6603a6f2 2172 read_unlock(&bond->lock);
1da177e4
LT		 2173
2174 return 0;
2175}
2176
2177static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2178{
2179 struct bonding *bond = bond_dev->priv;
2180 struct slave *slave;
2181 int i, found = 0;
2182
2183 if (info->slave_id < 0) {
2184 return -ENODEV;
2185 }
2186
6603a6f2 2187 read_lock(&bond->lock);
1da177e4
LT		 2188
2189 bond_for_each_slave(bond, slave, i) {
2190 if (i == (int)info->slave_id) {
2191 found = 1;
2192 break;
2193 }
2194 }
2195
6603a6f2 2196 read_unlock(&bond->lock);
1da177e4
LT		 2197
2198 if (found) {
2199 strcpy(info->slave_name, slave->dev->name);
2200 info->link = slave->link;
2201 info->state = slave->state;
2202 info->link_failure_count = slave->link_failure_count;
2203 } else {
2204 return -ENODEV;
2205 }
2206
2207 return 0;
2208}
2209
2210/*-------------------------------- Monitoring -------------------------------*/
2211
0b0eef66
JV		 2212/*
2213 * if !have_locks, return nonzero if a failover is necessary. if
2214 * have_locks, do whatever failover activities are needed.
2215 *
2216 * This is to separate the inspection and failover steps for locking
2217 * purposes; failover requires rtnl, but acquiring it for every
2218 * inspection is undesirable, so a wrapper first does inspection, and
2219 * the acquires the necessary locks and calls again to perform
2220 * failover if needed. Since all locks are dropped, a complete
2221 * restart is needed between calls.
2222 */
2223static int __bond_mii_monitor(struct bonding *bond, int have_locks)
1da177e4 2224{
1da177e4
LT		 2225 struct slave *slave, *oldcurrent;
2226 int do_failover = 0;
1da177e4
LT		 2227 int i;
2228
0b0eef66 2229 if (bond->slave_cnt == 0)
1da177e4 2230 goto out;
1da177e4
LT		 2231
2232 /* we will try to read the link status of each of our slaves, and
2233 * set their IFF_RUNNING flag appropriately. For each slave not
2234 * supporting MII status, we won't do anything so that a user-space
2235 * program could monitor the link itself if needed.
2236 */
2237
1053f62c
MS		 2238 if (bond->send_grat_arp) {
2239 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2240 &bond->curr_active_slave->dev->state))
2241 dprintk("Needs to send gratuitous arp but not yet\n");
2242 else {
2243 dprintk("sending delayed gratuitous arp on on %s\n",
2244 bond->curr_active_slave->dev->name);
2245 bond_send_gratuitous_arp(bond);
7893b249 2246 bond->send_grat_arp--;
1053f62c
MS		 2247 }
2248 }
1da177e4
LT		 2249 read_lock(&bond->curr_slave_lock);
2250 oldcurrent = bond->curr_active_slave;
2251 read_unlock(&bond->curr_slave_lock);
2252
2253 bond_for_each_slave(bond, slave, i) {
2254 struct net_device *slave_dev = slave->dev;
2255 int link_state;
2256 u16 old_speed = slave->speed;
2257 u8 old_duplex = slave->duplex;
2258
2259 link_state = bond_check_dev_link(bond, slave_dev, 0);
2260
2261 switch (slave->link) {
2262 case BOND_LINK_UP: /* the link was up */
2263 if (link_state == BMSR_LSTATUS) {
059fe7a5
JV		 2264 if (!oldcurrent) {
2265 if (!have_locks)
2266 return 1;
2267 do_failover = 1;
2268 }
1da177e4
LT		 2269 break;
2270 } else { /* link going down */
2271 slave->link = BOND_LINK_FAIL;
2272 slave->delay = bond->params.downdelay;
2273
2274 if (slave->link_failure_count < UINT_MAX) {
2275 slave->link_failure_count++;
2276 }
2277
2278 if (bond->params.downdelay) {
2279 printk(KERN_INFO DRV_NAME
2280 ": %s: link status down for %s "
2281 "interface %s, disabling it in "
2282 "%d ms.\n",
1b76b316 2283 bond->dev->name,
1da177e4
LT		 2284 IS_UP(slave_dev)
2285 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2286 ? ((slave == oldcurrent)
2287 ? "active " : "backup ")
2288 : "")
2289 : "idle ",
2290 slave_dev->name,
2291 bond->params.downdelay * bond->params.miimon);
2292 }
2293 }
2294 /* no break ! fall through the BOND_LINK_FAIL test to
2295 ensure proper action to be taken
2296 */
2297 case BOND_LINK_FAIL: /* the link has just gone down */
2298 if (link_state != BMSR_LSTATUS) {
2299 /* link stays down */
2300 if (slave->delay <= 0) {
0b0eef66
JV		 2301 if (!have_locks)
2302 return 1;
2303
1da177e4
LT		 2304 /* link down for too long time */
2305 slave->link = BOND_LINK_DOWN;
2306
2307 /* in active/backup mode, we must
2308 * completely disable this interface
2309 */
2310 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2311 (bond->params.mode == BOND_MODE_8023AD)) {
2312 bond_set_slave_inactive_flags(slave);
2313 }
2314
2315 printk(KERN_INFO DRV_NAME
2316 ": %s: link status definitely "
2317 "down for interface %s, "
2318 "disabling it\n",
1b76b316 2319 bond->dev->name,
1da177e4
LT		 2320 slave_dev->name);
2321
2322 /* notify ad that the link status has changed */
2323 if (bond->params.mode == BOND_MODE_8023AD) {
2324 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2325 }
2326
2327 if ((bond->params.mode == BOND_MODE_TLB) ||
2328 (bond->params.mode == BOND_MODE_ALB)) {
2329 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2330 }
2331
2332 if (slave == oldcurrent) {
2333 do_failover = 1;
2334 }
2335 } else {
2336 slave->delay--;
2337 }
2338 } else {
2339 /* link up again */
2340 slave->link = BOND_LINK_UP;
2341 slave->jiffies = jiffies;
2342 printk(KERN_INFO DRV_NAME
2343 ": %s: link status up again after %d "
2344 "ms for interface %s.\n",
1b76b316 2345 bond->dev->name,
1da177e4
LT		 2346 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2347 slave_dev->name);
2348 }
2349 break;
2350 case BOND_LINK_DOWN: /* the link was down */
2351 if (link_state != BMSR_LSTATUS) {
2352 /* the link stays down, nothing more to do */
2353 break;
2354 } else { /* link going up */
2355 slave->link = BOND_LINK_BACK;
2356 slave->delay = bond->params.updelay;
2357
2358 if (bond->params.updelay) {
2359 /* if updelay == 0, no need to
2360 advertise about a 0 ms delay */
2361 printk(KERN_INFO DRV_NAME
2362 ": %s: link status up for "
2363 "interface %s, enabling it "
2364 "in %d ms.\n",
1b76b316 2365 bond->dev->name,
1da177e4
LT		 2366 slave_dev->name,
2367 bond->params.updelay * bond->params.miimon);
2368 }
2369 }
2370 /* no break ! fall through the BOND_LINK_BACK state in
2371 case there's something to do.
2372 */
2373 case BOND_LINK_BACK: /* the link has just come back */
2374 if (link_state != BMSR_LSTATUS) {
2375 /* link down again */
2376 slave->link = BOND_LINK_DOWN;
2377
2378 printk(KERN_INFO DRV_NAME
2379 ": %s: link status down again after %d "
2380 "ms for interface %s.\n",
1b76b316 2381 bond->dev->name,
1da177e4
LT		 2382 (bond->params.updelay - slave->delay) * bond->params.miimon,
2383 slave_dev->name);
2384 } else {
2385 /* link stays up */
2386 if (slave->delay == 0) {
0b0eef66
JV		 2387 if (!have_locks)
2388 return 1;
2389
1da177e4
LT		 2390 /* now the link has been up for long time enough */
2391 slave->link = BOND_LINK_UP;
2392 slave->jiffies = jiffies;
2393
2394 if (bond->params.mode == BOND_MODE_8023AD) {
2395 /* prevent it from being the active one */
2396 slave->state = BOND_STATE_BACKUP;
2397 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2398 /* make it immediately active */
2399 slave->state = BOND_STATE_ACTIVE;
2400 } else if (slave != bond->primary_slave) {
2401 /* prevent it from being the active one */
2402 slave->state = BOND_STATE_BACKUP;
2403 }
2404
2405 printk(KERN_INFO DRV_NAME
2406 ": %s: link status definitely "
2407 "up for interface %s.\n",
1b76b316 2408 bond->dev->name,
1da177e4
LT		 2409 slave_dev->name);
2410
2411 /* notify ad that the link status has changed */
2412 if (bond->params.mode == BOND_MODE_8023AD) {
2413 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2414 }
2415
2416 if ((bond->params.mode == BOND_MODE_TLB) ||
2417 (bond->params.mode == BOND_MODE_ALB)) {
2418 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2419 }
2420
2421 if ((!oldcurrent) ||
2422 (slave == bond->primary_slave)) {
2423 do_failover = 1;
2424 }
2425 } else {
2426 slave->delay--;
2427 }
2428 }
2429 break;
2430 default:
2431 /* Should not happen */
4e0952c7
MW		 2432 printk(KERN_ERR DRV_NAME
2433 ": %s: Error: %s Illegal value (link=%d)\n",
1b76b316 2434 bond->dev->name,
4e0952c7
MW		 2435 slave->dev->name,
2436 slave->link);
1da177e4
LT		 2437 goto out;
2438 } /* end of switch (slave->link) */
2439
2440 bond_update_speed_duplex(slave);
2441
2442 if (bond->params.mode == BOND_MODE_8023AD) {
2443 if (old_speed != slave->speed) {
2444 bond_3ad_adapter_speed_changed(slave);
2445 }
2446
2447 if (old_duplex != slave->duplex) {
2448 bond_3ad_adapter_duplex_changed(slave);
2449 }
2450 }
2451
2452 } /* end of for */
2453
2454 if (do_failover) {
059fe7a5
JV		 2455 ASSERT_RTNL();
2456
2457 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2458
2459 bond_select_active_slave(bond);
2460
059fe7a5
JV		 2461 write_unlock_bh(&bond->curr_slave_lock);
2462
ff59c456
JV		 2463 } else
2464 bond_set_carrier(bond);
1da177e4 2465
1da177e4 2466out:
0b0eef66 2467 return 0;
1da177e4
LT		 2468}
2469
0b0eef66
JV		 2470/*
2471 * bond_mii_monitor
2472 *
2473 * Really a wrapper that splits the mii monitor into two phases: an
2474 * inspection, then (if inspection indicates something needs to be
2475 * done) an acquisition of appropriate locks followed by another pass
2476 * to implement whatever link state changes are indicated.
2477 */
2478void bond_mii_monitor(struct work_struct *work)
2479{
2480 struct bonding *bond = container_of(work, struct bonding,
2481 mii_work.work);
2482 unsigned long delay;
2483
2484 read_lock(&bond->lock);
2485 if (bond->kill_timers) {
2486 read_unlock(&bond->lock);
2487 return;
2488 }
2489 if (__bond_mii_monitor(bond, 0)) {
2490 read_unlock(&bond->lock);
2491 rtnl_lock();
2492 read_lock(&bond->lock);
2493 __bond_mii_monitor(bond, 1);
5655662d
JV		 2494 read_unlock(&bond->lock);
2495 rtnl_unlock(); /* might sleep, hold no other locks */
2496 read_lock(&bond->lock);
0b0eef66
JV		 2497 }
2498
5ce0da8f 2499 delay = msecs_to_jiffies(bond->params.miimon);
0b0eef66
JV		 2500 read_unlock(&bond->lock);
2501 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2502}
c3ade5ca 2503
d3bb52b0 2504static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2505{
2506 struct in_device *idev;
2507 struct in_ifaddr *ifa;
a144ea4b 2508 __be32 addr = 0;
c3ade5ca
JV		 2509
2510 if (!dev)
2511 return 0;
2512
2513 rcu_read_lock();
e5ed6399 2514 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2515 if (!idev)
2516 goto out;
2517
2518 ifa = idev->ifa_list;
2519 if (!ifa)
2520 goto out;
2521
2522 addr = ifa->ifa_local;
2523out:
2524 rcu_read_unlock();
2525 return addr;
2526}
2527
d3bb52b0 2528static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2529{
0883beca 2530 struct vlan_entry *vlan;
f5b2b966
JV		 2531
2532 if (ip == bond->master_ip)
2533 return 1;
2534
0883beca 2535 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2536 if (ip == vlan->vlan_ip)
2537 return 1;
2538 }
2539
2540 return 0;
2541}
2542
c3ade5ca
JV		 2543/*
2544 * We go to the (large) trouble of VLAN tagging ARP frames because
2545 * switches in VLAN mode (especially if ports are configured as
2546 * "native" to a VLAN) might not pass non-tagged frames.
2547 */
d3bb52b0 2548static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2549{
2550 struct sk_buff *skb;
2551
2552 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2553 slave_dev->name, dest_ip, src_ip, vlan_id);
2554
2555 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2556 NULL, slave_dev->dev_addr, NULL);
2557
2558 if (!skb) {
2559 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2560 return;
2561 }
2562 if (vlan_id) {
2563 skb = vlan_put_tag(skb, vlan_id);
2564 if (!skb) {
2565 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2566 return;
2567 }
2568 }
2569 arp_xmit(skb);
2570}
2571
2572
1da177e4
LT		 2573static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2574{
c3ade5ca 2575 int i, vlan_id, rv;
d3bb52b0 2576 __be32 *targets = bond->params.arp_targets;
0883beca 2577 struct vlan_entry *vlan;
c3ade5ca
JV		 2578 struct net_device *vlan_dev;
2579 struct flowi fl;
2580 struct rtable *rt;
1da177e4 2581
6b780567
MW		 2582 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2583 if (!targets[i])
2584 continue;
c3ade5ca
JV		 2585 dprintk("basa: target %x\n", targets[i]);
2586 if (list_empty(&bond->vlan_list)) {
2587 dprintk("basa: empty vlan: arp_send\n");
2588 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2589 bond->master_ip, 0);
2590 continue;
2591 }
2592
2593 /*
2594 * If VLANs are configured, we do a route lookup to
2595 * determine which VLAN interface would be used, so we
2596 * can tag the ARP with the proper VLAN tag.
2597 */
2598 memset(&fl, 0, sizeof(fl));
2599 fl.fl4_dst = targets[i];
2600 fl.fl4_tos = RTO_ONLINK;
2601
f206351a 2602 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2603 if (rv) {
2604 if (net_ratelimit()) {
2605 printk(KERN_WARNING DRV_NAME
2606 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2607 bond->dev->name, NIPQUAD(fl.fl4_dst));
2608 }
2609 continue;
2610 }
2611
2612 /*
2613 * This target is not on a VLAN
2614 */
2615 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2616 ip_rt_put(rt);
c3ade5ca
JV		 2617 dprintk("basa: rtdev == bond->dev: arp_send\n");
2618 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2619 bond->master_ip, 0);
2620 continue;
2621 }
2622
2623 vlan_id = 0;
0883beca 2624 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2625 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2626 if (vlan_dev == rt->u.dst.dev) {
2627 vlan_id = vlan->vlan_id;
2628 dprintk("basa: vlan match on %s %d\n",
2629 vlan_dev->name, vlan_id);
2630 break;
2631 }
2632 }
2633
2634 if (vlan_id) {
ed4b9f80 2635 ip_rt_put(rt);
c3ade5ca
JV		 2636 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2637 vlan->vlan_ip, vlan_id);
2638 continue;
2639 }
2640
2641 if (net_ratelimit()) {
2642 printk(KERN_WARNING DRV_NAME
2643 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2644 bond->dev->name, NIPQUAD(fl.fl4_dst),
2645 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2646 }
ed4b9f80 2647 ip_rt_put(rt);
c3ade5ca
JV		 2648 }
2649}
2650
2651/*
2652 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2653 * for each VLAN above us.
2654 */
2655static void bond_send_gratuitous_arp(struct bonding *bond)
2656{
2657 struct slave *slave = bond->curr_active_slave;
2658 struct vlan_entry *vlan;
2659 struct net_device *vlan_dev;
2660
2661 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2662 slave ? slave->dev->name : "NULL");
2663 if (!slave)
2664 return;
2665
2666 if (bond->master_ip) {
2667 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2668 bond->master_ip, 0);
c3ade5ca
JV		 2669 }
2670
2671 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2672 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2673 if (vlan->vlan_ip) {
2674 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2675 vlan->vlan_ip, vlan->vlan_id);
2676 }
1da177e4
LT		 2677 }
2678}
2679
d3bb52b0 2680static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2681{
2682 int i;
d3bb52b0 2683 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2684
2685 targets = bond->params.arp_targets;
2686 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2687 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2688 "%u.%u.%u.%u bhti(tip) %d\n",
2689 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2690 bond_has_this_ip(bond, tip));
2691 if (sip == targets[i]) {
2692 if (bond_has_this_ip(bond, tip))
2693 slave->last_arp_rx = jiffies;
2694 return;
2695 }
2696 }
2697}
2698
2699static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2700{
2701 struct arphdr *arp;
2702 struct slave *slave;
2703 struct bonding *bond;
2704 unsigned char *arp_ptr;
d3bb52b0 2705 __be32 sip, tip;
f5b2b966 2706
c346dca1 2707 if (dev_net(dev) != &init_net)
e730c155
EB		 2708 goto out;
2709
f5b2b966
JV		 2710 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2711 goto out;
2712
2713 bond = dev->priv;
2714 read_lock(&bond->lock);
2715
2716 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2717 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2718 orig_dev ? orig_dev->name : "NULL");
2719
2720 slave = bond_get_slave_by_dev(bond, orig_dev);
2721 if (!slave || !slave_do_arp_validate(bond, slave))
2722 goto out_unlock;
2723
988b7050 2724 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2725 goto out_unlock;
2726
d0a92be0 2727 arp = arp_hdr(skb);
f5b2b966
JV		 2728 if (arp->ar_hln != dev->addr_len ||
2729 skb->pkt_type == PACKET_OTHERHOST ||
2730 skb->pkt_type == PACKET_LOOPBACK ||
2731 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2732 arp->ar_pro != htons(ETH_P_IP) ||
2733 arp->ar_pln != 4)
2734 goto out_unlock;
2735
2736 arp_ptr = (unsigned char *)(arp + 1);
2737 arp_ptr += dev->addr_len;
2738 memcpy(&sip, arp_ptr, 4);
2739 arp_ptr += 4 + dev->addr_len;
2740 memcpy(&tip, arp_ptr, 4);
2741
2742 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2743 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2744 slave->state, bond->params.arp_validate,
2745 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2746
2747 /*
2748 * Backup slaves won't see the ARP reply, but do come through
2749 * here for each ARP probe (so we swap the sip/tip to validate
2750 * the probe). In a "redundant switch, common router" type of
2751 * configuration, the ARP probe will (hopefully) travel from
2752 * the active, through one switch, the router, then the other
2753 * switch before reaching the backup.
2754 */
2755 if (slave->state == BOND_STATE_ACTIVE)
2756 bond_validate_arp(bond, slave, sip, tip);
2757 else
2758 bond_validate_arp(bond, slave, tip, sip);
2759
2760out_unlock:
2761 read_unlock(&bond->lock);
2762out:
2763 dev_kfree_skb(skb);
2764 return NET_RX_SUCCESS;
2765}
2766
1da177e4
LT		 2767/*
2768 * this function is called regularly to monitor each slave's link
2769 * ensuring that traffic is being sent and received when arp monitoring
2770 * is used in load-balancing mode. if the adapter has been dormant, then an
2771 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2772 * arp monitoring in active backup mode.
2773 */
1b76b316 2774void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2775{
1b76b316
JV		 2776 struct bonding *bond = container_of(work, struct bonding,
2777 arp_work.work);
1da177e4
LT		 2778 struct slave *slave, *oldcurrent;
2779 int do_failover = 0;
2780 int delta_in_ticks;
2781 int i;
2782
2783 read_lock(&bond->lock);
2784
5ce0da8f 2785 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2786
2787 if (bond->kill_timers) {
2788 goto out;
2789 }
2790
2791 if (bond->slave_cnt == 0) {
2792 goto re_arm;
2793 }
2794
2795 read_lock(&bond->curr_slave_lock);
2796 oldcurrent = bond->curr_active_slave;
2797 read_unlock(&bond->curr_slave_lock);
2798
2799 /* see if any of the previous devices are up now (i.e. they have
2800 * xmt and rcv traffic). the curr_active_slave does not come into
2801 * the picture unless it is null. also, slave->jiffies is not needed
2802 * here because we send an arp on each slave and give a slave as
2803 * long as it needs to get the tx/rx within the delta.
2804 * TODO: what about up/down delay in arp mode? it wasn't here before
2805 * so it can wait
2806 */
2807 bond_for_each_slave(bond, slave, i) {
2808 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2809 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2810 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2811
2812 slave->link = BOND_LINK_UP;
2813 slave->state = BOND_STATE_ACTIVE;
2814
2815 /* primary_slave has no meaning in round-robin
2816 * mode. the window of a slave being up and
2817 * curr_active_slave being null after enslaving
2818 * is closed.
2819 */
2820 if (!oldcurrent) {
2821 printk(KERN_INFO DRV_NAME
2822 ": %s: link status definitely "
2823 "up for interface %s, ",
1b76b316 2824 bond->dev->name,
1da177e4
LT		 2825 slave->dev->name);
2826 do_failover = 1;
2827 } else {
2828 printk(KERN_INFO DRV_NAME
2829 ": %s: interface %s is now up\n",
1b76b316 2830 bond->dev->name,
1da177e4
LT		 2831 slave->dev->name);
2832 }
2833 }
2834 } else {
2835 /* slave->link == BOND_LINK_UP */
2836
2837 /* not all switches will respond to an arp request
2838 * when the source ip is 0, so don't take the link down
2839 * if we don't know our ip yet
2840 */
b63bb739 2841 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2842 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2843
2844 slave->link = BOND_LINK_DOWN;
2845 slave->state = BOND_STATE_BACKUP;
2846
2847 if (slave->link_failure_count < UINT_MAX) {
2848 slave->link_failure_count++;
2849 }
2850
2851 printk(KERN_INFO DRV_NAME
2852 ": %s: interface %s is now down.\n",
1b76b316 2853 bond->dev->name,
1da177e4
LT		 2854 slave->dev->name);
2855
2856 if (slave == oldcurrent) {
2857 do_failover = 1;
2858 }
2859 }
2860 }
2861
2862 /* note: if switch is in round-robin mode, all links
2863 * must tx arp to ensure all links rx an arp - otherwise
2864 * links may oscillate or not come up at all; if switch is
2865 * in something like xor mode, there is nothing we can
2866 * do - all replies will be rx'ed on same link causing slaves
2867 * to be unstable during low/no traffic periods
2868 */
2869 if (IS_UP(slave->dev)) {
2870 bond_arp_send_all(bond, slave);
2871 }
2872 }
2873
2874 if (do_failover) {
059fe7a5 2875 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2876
2877 bond_select_active_slave(bond);
2878
059fe7a5 2879 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2880 }
2881
2882re_arm:
1b76b316
JV		 2883 if (bond->params.arp_interval)
2884 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2885out:
2886 read_unlock(&bond->lock);
2887}
2888
2889/*
b2220cad
JV		 2890 * Called to inspect slaves for active-backup mode ARP monitor link state
2891 * changes. Sets new_link in slaves to specify what action should take
2892 * place for the slave. Returns 0 if no changes are found, >0 if changes
2893 * to link states must be committed.
2894 *
2895 * Called with bond->lock held for read.
1da177e4 2896 */
b2220cad 2897static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2898{
1da177e4 2899 struct slave *slave;
b2220cad 2900 int i, commit = 0;
1da177e4 2901
b2220cad
JV		 2902 bond_for_each_slave(bond, slave, i) {
2903 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2904
b2220cad
JV		 2905 if (slave->link != BOND_LINK_UP) {
2906 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2907 delta_in_ticks)) {
2908 slave->new_link = BOND_LINK_UP;
2909 commit++;
2910 }
1da177e4 2911
b2220cad
JV		 2912 continue;
2913 }
1da177e4 2914
b2220cad
JV		 2915 /*
2916 * Give slaves 2*delta after being enslaved or made
2917 * active. This avoids bouncing, as the last receive
2918 * times need a full ARP monitor cycle to be updated.
2919 */
2920 if (!time_after_eq(jiffies, slave->jiffies +
2921 2 * delta_in_ticks))
2922 continue;
2923
2924 /*
2925 * Backup slave is down if:
2926 * - No current_arp_slave AND
2927 * - more than 3*delta since last receive AND
2928 * - the bond has an IP address
2929 *
2930 * Note: a non-null current_arp_slave indicates
2931 * the curr_active_slave went down and we are
2932 * searching for a new one; under this condition
2933 * we only take the curr_active_slave down - this
2934 * gives each slave a chance to tx/rx traffic
2935 * before being taken out
2936 */
2937 if (slave->state == BOND_STATE_BACKUP &&
2938 !bond->current_arp_slave &&
2939 time_after(jiffies, slave_last_rx(bond, slave) +
2940 3 * delta_in_ticks)) {
2941 slave->new_link = BOND_LINK_DOWN;
2942 commit++;
2943 }
2944
2945 /*
2946 * Active slave is down if:
2947 * - more than 2*delta since transmitting OR
2948 * - (more than 2*delta since receive AND
2949 * the bond has an IP address)
2950 */
2951 if ((slave->state == BOND_STATE_ACTIVE) &&
2952 (time_after_eq(jiffies, slave->dev->trans_start +
2953 2 * delta_in_ticks) ||
2954 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2955 + 2 * delta_in_ticks)))) {
2956 slave->new_link = BOND_LINK_DOWN;
2957 commit++;
2958 }
1da177e4
LT		 2959 }
2960
b2220cad
JV		 2961 read_lock(&bond->curr_slave_lock);
2962
2963 /*
2964 * Trigger a commit if the primary option setting has changed.
1da177e4 2965 */
b2220cad
JV		 2966 if (bond->primary_slave &&
2967 (bond->primary_slave != bond->curr_active_slave) &&
2968 (bond->primary_slave->link == BOND_LINK_UP))
2969 commit++;
1da177e4 2970
b2220cad 2971 read_unlock(&bond->curr_slave_lock);
1da177e4 2972
b2220cad
JV		 2973 return commit;
2974}
1da177e4 2975
b2220cad
JV		 2976/*
2977 * Called to commit link state changes noted by inspection step of
2978 * active-backup mode ARP monitor.
2979 *
2980 * Called with RTNL and bond->lock for read.
2981 */
2982static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2983{
2984 struct slave *slave;
2985 int i;
1da177e4 2986
b2220cad
JV		 2987 bond_for_each_slave(bond, slave, i) {
2988 switch (slave->new_link) {
2989 case BOND_LINK_NOCHANGE:
2990 continue;
ff59c456 2991
b2220cad
JV		 2992 case BOND_LINK_UP:
2993 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2994
b2220cad
JV		 2995 if (!bond->curr_active_slave &&
2996 time_before_eq(jiffies, slave->dev->trans_start +
2997 delta_in_ticks)) {
2998 slave->link = BOND_LINK_UP;
2999 bond_change_active_slave(bond, slave);
3000 bond->current_arp_slave = NULL;
3001
3002 printk(KERN_INFO DRV_NAME
3003 ": %s: %s is up and now the "
3004 "active interface\n",
3005 bond->dev->name, slave->dev->name);
3006
3007 } else if (bond->curr_active_slave != slave) {
3008 /* this slave has just come up but we
3009 * already have a current slave; this can
3010 * also happen if bond_enslave adds a new
3011 * slave that is up while we are searching
3012 * for a new slave
1da177e4 3013 */
b2220cad
JV		 3014 slave->link = BOND_LINK_UP;
3015 bond_set_slave_inactive_flags(slave);
3016 bond->current_arp_slave = NULL;
3017
3018 printk(KERN_INFO DRV_NAME
3019 ": %s: backup interface %s is now up\n",
3020 bond->dev->name, slave->dev->name);
3021 }
1da177e4 3022
b2220cad 3023 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3024
b2220cad 3025 break;
1da177e4 3026
b2220cad
JV		 3027 case BOND_LINK_DOWN:
3028 if (slave->link_failure_count < UINT_MAX)
3029 slave->link_failure_count++;
3030
3031 slave->link = BOND_LINK_DOWN;
3032
3033 if (slave == bond->curr_active_slave) {
3034 printk(KERN_INFO DRV_NAME
3035 ": %s: link status down for active "
3036 "interface %s, disabling it\n",
3037 bond->dev->name, slave->dev->name);
1da177e4
LT		 3038
3039 bond_set_slave_inactive_flags(slave);
3040
b2220cad
JV		 3041 write_lock_bh(&bond->curr_slave_lock);
3042
3043 bond_select_active_slave(bond);
3044 if (bond->curr_active_slave)
3045 bond->curr_active_slave->jiffies =
3046 jiffies;
3047
3048 write_unlock_bh(&bond->curr_slave_lock);
3049
3050 bond->current_arp_slave = NULL;
3051
3052 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3053 printk(KERN_INFO DRV_NAME
3054 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3055 bond->dev->name, slave->dev->name);
3056
3057 bond_set_slave_inactive_flags(slave);
1da177e4 3058 }
b2220cad
JV		 3059 break;
3060
3061 default:
3062 printk(KERN_ERR DRV_NAME
3063 ": %s: impossible: new_link %d on slave %s\n",
3064 bond->dev->name, slave->new_link,
3065 slave->dev->name);
1da177e4
LT		 3066 }
3067 }
3068
b2220cad
JV		 3069 /*
3070 * No race with changes to primary via sysfs, as we hold rtnl.
3071 */
3072 if (bond->primary_slave &&
3073 (bond->primary_slave != bond->curr_active_slave) &&
3074 (bond->primary_slave->link == BOND_LINK_UP)) {
3075 write_lock_bh(&bond->curr_slave_lock);
3076 bond_change_active_slave(bond, bond->primary_slave);
3077 write_unlock_bh(&bond->curr_slave_lock);
3078 }
1da177e4 3079
b2220cad
JV		 3080 bond_set_carrier(bond);
3081}
1da177e4 3082
b2220cad
JV		 3083/*
3084 * Send ARP probes for active-backup mode ARP monitor.
3085 *
3086 * Called with bond->lock held for read.
3087 */
3088static void bond_ab_arp_probe(struct bonding *bond)
3089{
3090 struct slave *slave;
3091 int i;
1da177e4 3092
b2220cad 3093 read_lock(&bond->curr_slave_lock);
1da177e4 3094
b2220cad
JV		 3095 if (bond->current_arp_slave && bond->curr_active_slave)
3096 printk("PROBE: c_arp %s && cas %s BAD\n",
3097 bond->current_arp_slave->dev->name,
3098 bond->curr_active_slave->dev->name);
1da177e4 3099
b2220cad
JV		 3100 if (bond->curr_active_slave) {
3101 bond_arp_send_all(bond, bond->curr_active_slave);
3102 read_unlock(&bond->curr_slave_lock);
3103 return;
3104 }
1da177e4 3105
b2220cad 3106 read_unlock(&bond->curr_slave_lock);
059fe7a5 3107
b2220cad
JV		 3108 /* if we don't have a curr_active_slave, search for the next available
3109 * backup slave from the current_arp_slave and make it the candidate
3110 * for becoming the curr_active_slave
3111 */
1da177e4 3112
b2220cad
JV		 3113 if (!bond->current_arp_slave) {
3114 bond->current_arp_slave = bond->first_slave;
3115 if (!bond->current_arp_slave)
3116 return;
3117 }
1da177e4 3118
b2220cad 3119 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3120
b2220cad
JV		 3121 /* search for next candidate */
3122 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3123 if (IS_UP(slave->dev)) {
3124 slave->link = BOND_LINK_BACK;
3125 bond_set_slave_active_flags(slave);
3126 bond_arp_send_all(bond, slave);
1da177e4 3127 slave->jiffies = jiffies;
b2220cad
JV		 3128 bond->current_arp_slave = slave;
3129 break;
1da177e4
LT		 3130 }
3131
b2220cad
JV		 3132 /* if the link state is up at this point, we
3133 * mark it down - this can happen if we have
3134 * simultaneous link failures and
3135 * reselect_active_interface doesn't make this
3136 * one the current slave so it is still marked
3137 * up when it is actually down
1da177e4 3138 */
b2220cad
JV		 3139 if (slave->link == BOND_LINK_UP) {
3140 slave->link = BOND_LINK_DOWN;
3141 if (slave->link_failure_count < UINT_MAX)
3142 slave->link_failure_count++;
1da177e4 3143
b2220cad
JV		 3144 bond_set_slave_inactive_flags(slave);
3145
3146 printk(KERN_INFO DRV_NAME
3147 ": %s: backup interface %s is now down.\n",
3148 bond->dev->name, slave->dev->name);
1da177e4 3149 }
b2220cad
JV		 3150 }
3151}
1da177e4 3152
b2220cad
JV		 3153void bond_activebackup_arp_mon(struct work_struct *work)
3154{
3155 struct bonding *bond = container_of(work, struct bonding,
3156 arp_work.work);
3157 int delta_in_ticks;
1da177e4 3158
b2220cad 3159 read_lock(&bond->lock);
1da177e4 3160
b2220cad
JV		 3161 if (bond->kill_timers)
3162 goto out;
1da177e4 3163
b2220cad 3164 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3165
b2220cad
JV		 3166 if (bond->slave_cnt == 0)
3167 goto re_arm;
3168
3169 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3170 read_unlock(&bond->lock);
3171 rtnl_lock();
3172 read_lock(&bond->lock);
3173
3174 bond_ab_arp_commit(bond, delta_in_ticks);
3175
3176 read_unlock(&bond->lock);
3177 rtnl_unlock();
3178 read_lock(&bond->lock);
1da177e4
LT		 3179 }
3180
b2220cad
JV		 3181 bond_ab_arp_probe(bond);
3182
1da177e4
LT		 3183re_arm:
3184 if (bond->params.arp_interval) {
1b76b316 3185 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3186 }
3187out:
3188 read_unlock(&bond->lock);
3189}
3190
3191/*------------------------------ proc/seq_file-------------------------------*/
3192
3193#ifdef CONFIG_PROC_FS
3194
1da177e4
LT		 3195static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3196{
3197 struct bonding *bond = seq->private;
3198 loff_t off = 0;
3199 struct slave *slave;
3200 int i;
3201
3202 /* make sure the bond won't be taken away */
3203 read_lock(&dev_base_lock);
6603a6f2 3204 read_lock(&bond->lock);
1da177e4
LT		 3205
3206 if (*pos == 0) {
3207 return SEQ_START_TOKEN;
3208 }
3209
3210 bond_for_each_slave(bond, slave, i) {
3211 if (++off == *pos) {
3212 return slave;
3213 }
3214 }
3215
3216 return NULL;
3217}
3218
3219static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3220{
3221 struct bonding *bond = seq->private;
3222 struct slave *slave = v;
3223
3224 ++*pos;
3225 if (v == SEQ_START_TOKEN) {
3226 return bond->first_slave;
3227 }
3228
3229 slave = slave->next;
3230
3231 return (slave == bond->first_slave) ? NULL : slave;
3232}
3233
3234static void bond_info_seq_stop(struct seq_file *seq, void *v)
3235{
3236 struct bonding *bond = seq->private;
3237
6603a6f2 3238 read_unlock(&bond->lock);
1da177e4
LT		 3239 read_unlock(&dev_base_lock);
3240}
3241
3242static void bond_info_show_master(struct seq_file *seq)
3243{
3244 struct bonding *bond = seq->private;
3245 struct slave *curr;
4756b02f
MW		 3246 int i;
3247 u32 target;
1da177e4
LT		 3248
3249 read_lock(&bond->curr_slave_lock);
3250 curr = bond->curr_active_slave;
3251 read_unlock(&bond->curr_slave_lock);
3252
dd957c57 3253 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3254 bond_mode_name(bond->params.mode));
3255
dd957c57
JV		 3256 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3257 bond->params.fail_over_mac)
3915c1e8
JV		 3258 seq_printf(seq, " (fail_over_mac %s)",
3259 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3260
3261 seq_printf(seq, "\n");
3262
c61b75ad
MW		 3263 if (bond->params.mode == BOND_MODE_XOR ||
3264 bond->params.mode == BOND_MODE_8023AD) {
3265 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3266 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3267 bond->params.xmit_policy);
3268 }
3269
1da177e4
LT		 3270 if (USES_PRIMARY(bond->params.mode)) {
3271 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3272 (bond->primary_slave) ?
3273 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3274
3275 seq_printf(seq, "Currently Active Slave: %s\n",
3276 (curr) ? curr->dev->name : "None");
3277 }
3278
ff59c456
JV		 3279 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3280 "up" : "down");
1da177e4
LT		 3281 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3282 seq_printf(seq, "Up Delay (ms): %d\n",
3283 bond->params.updelay * bond->params.miimon);
3284 seq_printf(seq, "Down Delay (ms): %d\n",
3285 bond->params.downdelay * bond->params.miimon);
3286
4756b02f
MW		 3287
3288 /* ARP information */
3289 if(bond->params.arp_interval > 0) {
3290 int printed=0;
3291 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3292 bond->params.arp_interval);
3293
3294 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3295
3296 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3297 if (!bond->params.arp_targets[i])
3298 continue;
3299 if (printed)
3300 seq_printf(seq, ",");
3301 target = ntohl(bond->params.arp_targets[i]);
3302 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3303 printed = 1;
3304 }
3305 seq_printf(seq, "\n");
3306 }
3307
1da177e4
LT		 3308 if (bond->params.mode == BOND_MODE_8023AD) {
3309 struct ad_info ad_info;
0795af57 3310 DECLARE_MAC_BUF(mac);
1da177e4
LT		 3311
3312 seq_puts(seq, "\n802.3ad info\n");
3313 seq_printf(seq, "LACP rate: %s\n",
3314 (bond->params.lacp_fast) ? "fast" : "slow");
3315
3316 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3317 seq_printf(seq, "bond %s has no active aggregator\n",
3318 bond->dev->name);
3319 } else {
3320 seq_printf(seq, "Active Aggregator Info:\n");
3321
3322 seq_printf(seq, "\tAggregator ID: %d\n",
3323 ad_info.aggregator_id);
3324 seq_printf(seq, "\tNumber of ports: %d\n",
3325 ad_info.ports);
3326 seq_printf(seq, "\tActor Key: %d\n",
3327 ad_info.actor_key);
3328 seq_printf(seq, "\tPartner Key: %d\n",
3329 ad_info.partner_key);
0795af57
JP		 3330 seq_printf(seq, "\tPartner Mac Address: %s\n",
3331 print_mac(mac, ad_info.partner_system));
1da177e4
LT		 3332 }
3333 }
3334}
3335
3336static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3337{
3338 struct bonding *bond = seq->private;
0795af57 3339 DECLARE_MAC_BUF(mac);
1da177e4
LT		 3340
3341 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3342 seq_printf(seq, "MII Status: %s\n",
3343 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3344 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3345 slave->link_failure_count);
3346
217df670 3347 seq_printf(seq,
0795af57
JP		 3348 "Permanent HW addr: %s\n",
3349 print_mac(mac, slave->perm_hwaddr));
1da177e4
LT		 3350
3351 if (bond->params.mode == BOND_MODE_8023AD) {
3352 const struct aggregator *agg
3353 = SLAVE_AD_INFO(slave).port.aggregator;
3354
3355 if (agg) {
3356 seq_printf(seq, "Aggregator ID: %d\n",
3357 agg->aggregator_identifier);
3358 } else {
3359 seq_puts(seq, "Aggregator ID: N/A\n");
3360 }
3361 }
3362}
3363
3364static int bond_info_seq_show(struct seq_file *seq, void *v)
3365{
3366 if (v == SEQ_START_TOKEN) {
3367 seq_printf(seq, "%s\n", version);
3368 bond_info_show_master(seq);
3369 } else {
3370 bond_info_show_slave(seq, v);
3371 }
3372
3373 return 0;
3374}
3375
3376static struct seq_operations bond_info_seq_ops = {
3377 .start = bond_info_seq_start,
3378 .next = bond_info_seq_next,
3379 .stop = bond_info_seq_stop,
3380 .show = bond_info_seq_show,
3381};
3382
3383static int bond_info_open(struct inode *inode, struct file *file)
3384{
3385 struct seq_file *seq;
3386 struct proc_dir_entry *proc;
3387 int res;
3388
3389 res = seq_open(file, &bond_info_seq_ops);
3390 if (!res) {
3391 /* recover the pointer buried in proc_dir_entry data */
3392 seq = file->private_data;
3393 proc = PDE(inode);
3394 seq->private = proc->data;
3395 }
3396
3397 return res;
3398}
3399
d54b1fdb 3400static const struct file_operations bond_info_fops = {
1da177e4
LT		 3401 .owner = THIS_MODULE,
3402 .open = bond_info_open,
3403 .read = seq_read,
3404 .llseek = seq_lseek,
3405 .release = seq_release,
3406};
3407
3408static int bond_create_proc_entry(struct bonding *bond)
3409{
3410 struct net_device *bond_dev = bond->dev;
3411
3412 if (bond_proc_dir) {
a95609cb
DL		 3413 bond->proc_entry = proc_create_data(bond_dev->name,
3414 S_IRUGO, bond_proc_dir,
3415 &bond_info_fops, bond);
1da177e4
LT		 3416 if (bond->proc_entry == NULL) {
3417 printk(KERN_WARNING DRV_NAME
3418 ": Warning: Cannot create /proc/net/%s/%s\n",
3419 DRV_NAME, bond_dev->name);
3420 } else {
1da177e4
LT		 3421 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3422 }
3423 }
3424
3425 return 0;
3426}
3427
3428static void bond_remove_proc_entry(struct bonding *bond)
3429{
3430 if (bond_proc_dir && bond->proc_entry) {
3431 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3432 memset(bond->proc_file_name, 0, IFNAMSIZ);
3433 bond->proc_entry = NULL;
3434 }
3435}
3436
3437/* Create the bonding directory under /proc/net, if doesn't exist yet.
3438 * Caller must hold rtnl_lock.
3439 */
3440static void bond_create_proc_dir(void)
3441{
3442 int len = strlen(DRV_NAME);
3443
457c4cbc 3444 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3445 bond_proc_dir = bond_proc_dir->next) {
3446 if ((bond_proc_dir->namelen == len) &&
3447 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3448 break;
3449 }
3450 }
3451
3452 if (!bond_proc_dir) {
457c4cbc 3453 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3454 if (bond_proc_dir) {
3455 bond_proc_dir->owner = THIS_MODULE;
3456 } else {
3457 printk(KERN_WARNING DRV_NAME
3458 ": Warning: cannot create /proc/net/%s\n",
3459 DRV_NAME);
3460 }
3461 }
3462}
3463
3464/* Destroy the bonding directory under /proc/net, if empty.
3465 * Caller must hold rtnl_lock.
3466 */
3467static void bond_destroy_proc_dir(void)
3468{
3469 struct proc_dir_entry *de;
3470
3471 if (!bond_proc_dir) {
3472 return;
3473 }
3474
3475 /* verify that the /proc dir is empty */
3476 for (de = bond_proc_dir->subdir; de; de = de->next) {
3477 /* ignore . and .. */
3478 if (*(de->name) != '.') {
3479 break;
3480 }
3481 }
3482
3483 if (de) {
3484 if (bond_proc_dir->owner == THIS_MODULE) {
3485 bond_proc_dir->owner = NULL;
3486 }
3487 } else {
457c4cbc 3488 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3489 bond_proc_dir = NULL;
3490 }
3491}
3492#endif /* CONFIG_PROC_FS */
3493
3494/*-------------------------- netdev event handling --------------------------*/
3495
3496/*
3497 * Change device name
3498 */
3499static int bond_event_changename(struct bonding *bond)
3500{
3501#ifdef CONFIG_PROC_FS
3502 bond_remove_proc_entry(bond);
3503 bond_create_proc_entry(bond);
3504#endif
b76cdba9
MW		 3505 down_write(&(bonding_rwsem));
3506 bond_destroy_sysfs_entry(bond);
3507 bond_create_sysfs_entry(bond);
3508 up_write(&(bonding_rwsem));
1da177e4
LT		 3509 return NOTIFY_DONE;
3510}
3511
3512static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3513{
3514 struct bonding *event_bond = bond_dev->priv;
3515
3516 switch (event) {
3517 case NETDEV_CHANGENAME:
3518 return bond_event_changename(event_bond);
3519 case NETDEV_UNREGISTER:
3b96c858 3520 bond_release_all(event_bond->dev);
1da177e4
LT		 3521 break;
3522 default:
3523 break;
3524 }
3525
3526 return NOTIFY_DONE;
3527}
3528
3529static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3530{
3531 struct net_device *bond_dev = slave_dev->master;
8531c5ff 3532 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3533
3534 switch (event) {
3535 case NETDEV_UNREGISTER:
3536 if (bond_dev) {
1284cd3a
JV		 3537 if (bond->setup_by_slave)
3538 bond_release_and_destroy(bond_dev, slave_dev);
3539 else
3540 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3541 }
3542 break;
3543 case NETDEV_CHANGE:
3544 /*
3545 * TODO: is this what we get if somebody
3546 * sets up a hierarchical bond, then rmmod's
3547 * one of the slave bonding devices?
3548 */
3549 break;
3550 case NETDEV_DOWN:
3551 /*
3552 * ... Or is it this?
3553 */
3554 break;
3555 case NETDEV_CHANGEMTU:
3556 /*
3557 * TODO: Should slaves be allowed to
3558 * independently alter their MTU? For
3559 * an active-backup bond, slaves need
3560 * not be the same type of device, so
3561 * MTUs may vary. For other modes,
3562 * slaves arguably should have the
3563 * same MTUs. To do this, we'd need to
3564 * take over the slave's change_mtu
3565 * function for the duration of their
3566 * servitude.
3567 */
3568 break;
3569 case NETDEV_CHANGENAME:
3570 /*
3571 * TODO: handle changing the primary's name
3572 */
3573 break;
8531c5ff
AK		 3574 case NETDEV_FEAT_CHANGE:
3575 bond_compute_features(bond);
3576 break;
1da177e4
LT		 3577 default:
3578 break;
3579 }
3580
3581 return NOTIFY_DONE;
3582}
3583
3584/*
3585 * bond_netdev_event: handle netdev notifier chain events.
3586 *
3587 * This function receives events for the netdev chain. The caller (an
e041c683 3588 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3589 * locks for us to safely manipulate the slave devices (RTNL lock,
3590 * dev_probe_lock).
3591 */
3592static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3593{
3594 struct net_device *event_dev = (struct net_device *)ptr;
3595
c346dca1 3596 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3597 return NOTIFY_DONE;
3598
1da177e4
LT		 3599 dprintk("event_dev: %s, event: %lx\n",
3600 (event_dev ? event_dev->name : "None"),
3601 event);
3602
0b680e75
JV		 3603 if (!(event_dev->priv_flags & IFF_BONDING))
3604 return NOTIFY_DONE;
3605
1da177e4
LT		 3606 if (event_dev->flags & IFF_MASTER) {
3607 dprintk("IFF_MASTER\n");
3608 return bond_master_netdev_event(event, event_dev);
3609 }
3610
3611 if (event_dev->flags & IFF_SLAVE) {
3612 dprintk("IFF_SLAVE\n");
3613 return bond_slave_netdev_event(event, event_dev);
3614 }
3615
3616 return NOTIFY_DONE;
3617}
3618
c3ade5ca
JV		 3619/*
3620 * bond_inetaddr_event: handle inetaddr notifier chain events.
3621 *
3622 * We keep track of device IPs primarily to use as source addresses in
3623 * ARP monitor probes (rather than spewing out broadcasts all the time).
3624 *
3625 * We track one IP for the main device (if it has one), plus one per VLAN.
3626 */
3627static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3628{
3629 struct in_ifaddr *ifa = ptr;
3630 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3631 struct bonding *bond;
3632 struct vlan_entry *vlan;
c3ade5ca 3633
c346dca1 3634 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3635 return NOTIFY_DONE;
3636
0883beca 3637 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3638 if (bond->dev == event_dev) {
3639 switch (event) {
3640 case NETDEV_UP:
3641 bond->master_ip = ifa->ifa_local;
3642 return NOTIFY_OK;
3643 case NETDEV_DOWN:
3644 bond->master_ip = bond_glean_dev_ip(bond->dev);
3645 return NOTIFY_OK;
3646 default:
3647 return NOTIFY_DONE;
3648 }
3649 }
3650
0883beca 3651 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3652 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3653 if (vlan_dev == event_dev) {
3654 switch (event) {
3655 case NETDEV_UP:
3656 vlan->vlan_ip = ifa->ifa_local;
3657 return NOTIFY_OK;
3658 case NETDEV_DOWN:
3659 vlan->vlan_ip =
3660 bond_glean_dev_ip(vlan_dev);
3661 return NOTIFY_OK;
3662 default:
3663 return NOTIFY_DONE;
3664 }
3665 }
3666 }
3667 }
3668 return NOTIFY_DONE;
3669}
3670
1da177e4
LT		 3671static struct notifier_block bond_netdev_notifier = {
3672 .notifier_call = bond_netdev_event,
3673};
3674
c3ade5ca
JV		 3675static struct notifier_block bond_inetaddr_notifier = {
3676 .notifier_call = bond_inetaddr_event,
3677};
3678
1da177e4
LT		 3679/*-------------------------- Packet type handling ---------------------------*/
3680
3681/* register to receive lacpdus on a bond */
3682static void bond_register_lacpdu(struct bonding *bond)
3683{
3684 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3685
3686 /* initialize packet type */
3687 pk_type->type = PKT_TYPE_LACPDU;
3688 pk_type->dev = bond->dev;
3689 pk_type->func = bond_3ad_lacpdu_recv;
3690
3691 dev_add_pack(pk_type);
3692}
3693
3694/* unregister to receive lacpdus on a bond */
3695static void bond_unregister_lacpdu(struct bonding *bond)
3696{
3697 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3698}
3699
f5b2b966
JV		 3700void bond_register_arp(struct bonding *bond)
3701{
3702 struct packet_type *pt = &bond->arp_mon_pt;
3703
c4f283b1
JV		 3704 if (pt->type)
3705 return;
3706
f5b2b966 3707 pt->type = htons(ETH_P_ARP);
e245cb71 3708 pt->dev = bond->dev;
f5b2b966
JV		 3709 pt->func = bond_arp_rcv;
3710 dev_add_pack(pt);
3711}
3712
3713void bond_unregister_arp(struct bonding *bond)
3714{
c4f283b1
JV		 3715 struct packet_type *pt = &bond->arp_mon_pt;
3716
3717 dev_remove_pack(pt);
3718 pt->type = 0;
f5b2b966
JV		 3719}
3720
169a3e66
JV		 3721/*---------------------------- Hashing Policies -----------------------------*/
3722
6f6652be
JV		 3723/*
3724 * Hash for the output device based upon layer 2 and layer 3 data. If
3725 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3726 */
3727static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3728 struct net_device *bond_dev, int count)
3729{
3730 struct ethhdr *data = (struct ethhdr *)skb->data;
3731 struct iphdr *iph = ip_hdr(skb);
3732
3733 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3734 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3735 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3736 }
3737
3738 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3739}
3740
169a3e66 3741/*
59c51591 3742 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3743 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3744 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3745 */
3746static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3747 struct net_device *bond_dev, int count)
3748{
3749 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3750 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3751 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3752 int layer4_xor = 0;
3753
3754 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3755 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3756 (iph->protocol == IPPROTO_TCP ||
3757 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3758 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3759 }
3760 return (layer4_xor ^
3761 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3762
3763 }
3764
3765 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3766}
3767
3768/*
3769 * Hash for the output device based upon layer 2 data
3770 */
3771static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3772 struct net_device *bond_dev, int count)
3773{
3774 struct ethhdr *data = (struct ethhdr *)skb->data;
3775
3776 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3777}
3778
1da177e4
LT		 3779/*-------------------------- Device entry points ----------------------------*/
3780
3781static int bond_open(struct net_device *bond_dev)
3782{
3783 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3784
3785 bond->kill_timers = 0;
3786
3787 if ((bond->params.mode == BOND_MODE_TLB) ||
3788 (bond->params.mode == BOND_MODE_ALB)) {
1da177e4
LT		 3789 /* bond_alb_initialize must be called before the timer
3790 * is started.
3791 */
3792 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3793 /* something went wrong - fail the open operation */
3794 return -1;
3795 }
3796
1b76b316
JV		 3797 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3798 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3799 }
3800
3801 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3802 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3803 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3804 }
3805
3806 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3807 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3808 INIT_DELAYED_WORK(&bond->arp_work,
3809 bond_activebackup_arp_mon);
3810 else
3811 INIT_DELAYED_WORK(&bond->arp_work,
3812 bond_loadbalance_arp_mon);
3813
3814 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3815 if (bond->params.arp_validate)
3816 bond_register_arp(bond);
1da177e4
LT		 3817 }
3818
3819 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3820 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3821 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3822 /* register to receive LACPDUs */
3823 bond_register_lacpdu(bond);
3824 }
3825
3826 return 0;
3827}
3828
3829static int bond_close(struct net_device *bond_dev)
3830{
3831 struct bonding *bond = bond_dev->priv;
3832
3833 if (bond->params.mode == BOND_MODE_8023AD) {
3834 /* Unregister the receive of LACPDUs */
3835 bond_unregister_lacpdu(bond);
3836 }
3837
f5b2b966
JV		 3838 if (bond->params.arp_validate)
3839 bond_unregister_arp(bond);
3840
1da177e4
LT		 3841 write_lock_bh(&bond->lock);
3842
1da177e4
LT		 3843
3844 /* signal timers not to re-arm */
3845 bond->kill_timers = 1;
3846
3847 write_unlock_bh(&bond->lock);
3848
1da177e4 3849 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3850 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3851 }
3852
3853 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3854 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3855 }
3856
3857 switch (bond->params.mode) {
3858 case BOND_MODE_8023AD:
1b76b316 3859 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3860 break;
3861 case BOND_MODE_TLB:
3862 case BOND_MODE_ALB:
1b76b316 3863 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3864 break;
3865 default:
3866 break;
3867 }
3868
1da177e4
LT		 3869
3870 if ((bond->params.mode == BOND_MODE_TLB) ||
3871 (bond->params.mode == BOND_MODE_ALB)) {
3872 /* Must be called only after all
3873 * slaves have been released
3874 */
3875 bond_alb_deinitialize(bond);
3876 }
3877
3878 return 0;
3879}
3880
3881static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3882{
3883 struct bonding *bond = bond_dev->priv;
3884 struct net_device_stats *stats = &(bond->stats), *sstats;
2439f9eb 3885 struct net_device_stats local_stats;
1da177e4
LT		 3886 struct slave *slave;
3887 int i;
3888
2439f9eb 3889 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3890
3891 read_lock_bh(&bond->lock);
3892
3893 bond_for_each_slave(bond, slave, i) {
c45d286e 3894 sstats = slave->dev->get_stats(slave->dev);
2439f9eb
AG		 3895 local_stats.rx_packets += sstats->rx_packets;
3896 local_stats.rx_bytes += sstats->rx_bytes;
3897 local_stats.rx_errors += sstats->rx_errors;
3898 local_stats.rx_dropped += sstats->rx_dropped;
3899
3900 local_stats.tx_packets += sstats->tx_packets;
3901 local_stats.tx_bytes += sstats->tx_bytes;
3902 local_stats.tx_errors += sstats->tx_errors;
3903 local_stats.tx_dropped += sstats->tx_dropped;
3904
3905 local_stats.multicast += sstats->multicast;
3906 local_stats.collisions += sstats->collisions;
3907
3908 local_stats.rx_length_errors += sstats->rx_length_errors;
3909 local_stats.rx_over_errors += sstats->rx_over_errors;
3910 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3911 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3912 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3913 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3914
3915 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3916 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3917 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3918 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3919 local_stats.tx_window_errors += sstats->tx_window_errors;
3920 }
3921
3922 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3923
3924 read_unlock_bh(&bond->lock);
3925
3926 return stats;
3927}
3928
3929static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3930{
3931 struct net_device *slave_dev = NULL;
3932 struct ifbond k_binfo;
3933 struct ifbond __user *u_binfo = NULL;
3934 struct ifslave k_sinfo;
3935 struct ifslave __user *u_sinfo = NULL;
3936 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3937 int res = 0;
3938
3939 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3940 bond_dev->name, cmd);
3941
3942 switch (cmd) {
1da177e4
LT		 3943 case SIOCGMIIPHY:
3944 mii = if_mii(ifr);
3945 if (!mii) {
3946 return -EINVAL;
3947 }
3948 mii->phy_id = 0;
3949 /* Fall Through */
3950 case SIOCGMIIREG:
3951 /*
3952 * We do this again just in case we were called by SIOCGMIIREG
3953 * instead of SIOCGMIIPHY.
3954 */
3955 mii = if_mii(ifr);
3956 if (!mii) {
3957 return -EINVAL;
3958 }
3959
3960 if (mii->reg_num == 1) {
3961 struct bonding *bond = bond_dev->priv;
3962 mii->val_out = 0;
6603a6f2 3963 read_lock(&bond->lock);
1da177e4 3964 read_lock(&bond->curr_slave_lock);
4e140079 3965 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3966 mii->val_out = BMSR_LSTATUS;
3967 }
3968 read_unlock(&bond->curr_slave_lock);
6603a6f2 3969 read_unlock(&bond->lock);
1da177e4
LT		 3970 }
3971
3972 return 0;
3973 case BOND_INFO_QUERY_OLD:
3974 case SIOCBONDINFOQUERY:
3975 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3976
3977 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3978 return -EFAULT;
3979 }
3980
3981 res = bond_info_query(bond_dev, &k_binfo);
3982 if (res == 0) {
3983 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3984 return -EFAULT;
3985 }
3986 }
3987
3988 return res;
3989 case BOND_SLAVE_INFO_QUERY_OLD:
3990 case SIOCBONDSLAVEINFOQUERY:
3991 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3992
3993 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3994 return -EFAULT;
3995 }
3996
3997 res = bond_slave_info_query(bond_dev, &k_sinfo);
3998 if (res == 0) {
3999 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4000 return -EFAULT;
4001 }
4002 }
4003
4004 return res;
4005 default:
4006 /* Go on */
4007 break;
4008 }
4009
4010 if (!capable(CAP_NET_ADMIN)) {
4011 return -EPERM;
4012 }
4013
b76cdba9 4014 down_write(&(bonding_rwsem));
881d966b 4015 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT		 4016
4017 dprintk("slave_dev=%p: \n", slave_dev);
4018
4019 if (!slave_dev) {
4020 res = -ENODEV;
4021 } else {
4022 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4023 switch (cmd) {
4024 case BOND_ENSLAVE_OLD:
4025 case SIOCBONDENSLAVE:
4026 res = bond_enslave(bond_dev, slave_dev);
4027 break;
4028 case BOND_RELEASE_OLD:
4029 case SIOCBONDRELEASE:
4030 res = bond_release(bond_dev, slave_dev);
4031 break;
4032 case BOND_SETHWADDR_OLD:
4033 case SIOCBONDSETHWADDR:
4034 res = bond_sethwaddr(bond_dev, slave_dev);
4035 break;
4036 case BOND_CHANGE_ACTIVE_OLD:
4037 case SIOCBONDCHANGEACTIVE:
4038 res = bond_ioctl_change_active(bond_dev, slave_dev);
4039 break;
4040 default:
4041 res = -EOPNOTSUPP;
4042 }
4043
4044 dev_put(slave_dev);
4045 }
4046
b76cdba9 4047 up_write(&(bonding_rwsem));
1da177e4
LT		 4048 return res;
4049}
4050
4051static void bond_set_multicast_list(struct net_device *bond_dev)
4052{
4053 struct bonding *bond = bond_dev->priv;
4054 struct dev_mc_list *dmi;
4055
1da177e4
LT		 4056 /*
4057 * Do promisc before checking multicast_mode
4058 */
4059 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4060 bond_set_promiscuity(bond, 1);
4061 }
4062
4063 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4064 bond_set_promiscuity(bond, -1);
4065 }
4066
4067 /* set allmulti flag to slaves */
4068 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4069 bond_set_allmulti(bond, 1);
4070 }
4071
4072 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4073 bond_set_allmulti(bond, -1);
4074 }
4075
80ee5ad2
JV		 4076 read_lock(&bond->lock);
4077
1da177e4
LT		 4078 bond->flags = bond_dev->flags;
4079
4080 /* looking for addresses to add to slaves' mc list */
4081 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4082 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4083 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4084 }
4085 }
4086
4087 /* looking for addresses to delete from slaves' list */
4088 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4089 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4090 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4091 }
4092 }
4093
4094 /* save master's multicast list */
4095 bond_mc_list_destroy(bond);
4096 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4097
80ee5ad2 4098 read_unlock(&bond->lock);
1da177e4
LT		 4099}
4100
4101/*
4102 * Change the MTU of all of a master's slaves to match the master
4103 */
4104static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4105{
4106 struct bonding *bond = bond_dev->priv;
4107 struct slave *slave, *stop_at;
4108 int res = 0;
4109 int i;
4110
4111 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4112 (bond_dev ? bond_dev->name : "None"), new_mtu);
4113
4114 /* Can't hold bond->lock with bh disabled here since
4115 * some base drivers panic. On the other hand we can't
4116 * hold bond->lock without bh disabled because we'll
4117 * deadlock. The only solution is to rely on the fact
4118 * that we're under rtnl_lock here, and the slaves
4119 * list won't change. This doesn't solve the problem
4120 * of setting the slave's MTU while it is
4121 * transmitting, but the assumption is that the base
4122 * driver can handle that.
4123 *
4124 * TODO: figure out a way to safely iterate the slaves
4125 * list, but without holding a lock around the actual
4126 * call to the base driver.
4127 */
4128
4129 bond_for_each_slave(bond, slave, i) {
4130 dprintk("s %p s->p %p c_m %p\n", slave,
4131 slave->prev, slave->dev->change_mtu);
e944ef79 4132
1da177e4
LT		 4133 res = dev_set_mtu(slave->dev, new_mtu);
4134
4135 if (res) {
4136 /* If we failed to set the slave's mtu to the new value
4137 * we must abort the operation even in ACTIVE_BACKUP
4138 * mode, because if we allow the backup slaves to have
4139 * different mtu values than the active slave we'll
4140 * need to change their mtu when doing a failover. That
4141 * means changing their mtu from timer context, which
4142 * is probably not a good idea.
4143 */
4144 dprintk("err %d %s\n", res, slave->dev->name);
4145 goto unwind;
4146 }
4147 }
4148
4149 bond_dev->mtu = new_mtu;
4150
4151 return 0;
4152
4153unwind:
4154 /* unwind from head to the slave that failed */
4155 stop_at = slave;
4156 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4157 int tmp_res;
4158
4159 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4160 if (tmp_res) {
4161 dprintk("unwind err %d dev %s\n", tmp_res,
4162 slave->dev->name);
4163 }
4164 }
4165
4166 return res;
4167}
4168
4169/*
4170 * Change HW address
4171 *
4172 * Note that many devices must be down to change the HW address, and
4173 * downing the master releases all slaves. We can make bonds full of
4174 * bonding devices to test this, however.
4175 */
4176static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4177{
4178 struct bonding *bond = bond_dev->priv;
4179 struct sockaddr *sa = addr, tmp_sa;
4180 struct slave *slave, *stop_at;
4181 int res = 0;
4182 int i;
4183
4184 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4185
dd957c57 4186 /*
3915c1e8
JV		 4187 * If fail_over_mac is set to active, do nothing and return
4188 * success. Returning an error causes ifenslave to fail.
dd957c57 4189 */
3915c1e8 4190 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4191 return 0;
2ab82852 4192
1da177e4
LT		 4193 if (!is_valid_ether_addr(sa->sa_data)) {
4194 return -EADDRNOTAVAIL;
4195 }
4196
4197 /* Can't hold bond->lock with bh disabled here since
4198 * some base drivers panic. On the other hand we can't
4199 * hold bond->lock without bh disabled because we'll
4200 * deadlock. The only solution is to rely on the fact
4201 * that we're under rtnl_lock here, and the slaves
4202 * list won't change. This doesn't solve the problem
4203 * of setting the slave's hw address while it is
4204 * transmitting, but the assumption is that the base
4205 * driver can handle that.
4206 *
4207 * TODO: figure out a way to safely iterate the slaves
4208 * list, but without holding a lock around the actual
4209 * call to the base driver.
4210 */
4211
4212 bond_for_each_slave(bond, slave, i) {
4213 dprintk("slave %p %s\n", slave, slave->dev->name);
4214
4215 if (slave->dev->set_mac_address == NULL) {
4216 res = -EOPNOTSUPP;
4217 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4218 goto unwind;
4219 }
4220
4221 res = dev_set_mac_address(slave->dev, addr);
4222 if (res) {
4223 /* TODO: consider downing the slave
4224 * and retry ?
4225 * User should expect communications
4226 * breakage anyway until ARP finish
4227 * updating, so...
4228 */
4229 dprintk("err %d %s\n", res, slave->dev->name);
4230 goto unwind;
4231 }
4232 }
4233
4234 /* success */
4235 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4236 return 0;
4237
4238unwind:
4239 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4240 tmp_sa.sa_family = bond_dev->type;
4241
4242 /* unwind from head to the slave that failed */
4243 stop_at = slave;
4244 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4245 int tmp_res;
4246
4247 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4248 if (tmp_res) {
4249 dprintk("unwind err %d dev %s\n", tmp_res,
4250 slave->dev->name);
4251 }
4252 }
4253
4254 return res;
4255}
4256
4257static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4258{
4259 struct bonding *bond = bond_dev->priv;
4260 struct slave *slave, *start_at;
cf5f9044 4261 int i, slave_no, res = 1;
1da177e4
LT		 4262
4263 read_lock(&bond->lock);
4264
4265 if (!BOND_IS_OK(bond)) {
4266 goto out;
4267 }
4268
cf5f9044
JV		 4269 /*
4270 * Concurrent TX may collide on rr_tx_counter; we accept that
4271 * as being rare enough not to justify using an atomic op here
4272 */
4273 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4274
cf5f9044
JV		 4275 bond_for_each_slave(bond, slave, i) {
4276 slave_no--;
4277 if (slave_no < 0) {
4278 break;
4279 }
1da177e4
LT		 4280 }
4281
cf5f9044 4282 start_at = slave;
1da177e4
LT		 4283 bond_for_each_slave_from(bond, slave, i, start_at) {
4284 if (IS_UP(slave->dev) &&
4285 (slave->link == BOND_LINK_UP) &&
4286 (slave->state == BOND_STATE_ACTIVE)) {
4287 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4288 break;
4289 }
4290 }
4291
1da177e4
LT		 4292out:
4293 if (res) {
4294 /* no suitable interface, frame not sent */
4295 dev_kfree_skb(skb);
4296 }
4297 read_unlock(&bond->lock);
4298 return 0;
4299}
4300
075897ce 4301
1da177e4
LT		 4302/*
4303 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4304 * the bond has a usable interface.
4305 */
4306static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4307{
4308 struct bonding *bond = bond_dev->priv;
4309 int res = 1;
4310
1da177e4
LT		 4311 read_lock(&bond->lock);
4312 read_lock(&bond->curr_slave_lock);
4313
4314 if (!BOND_IS_OK(bond)) {
4315 goto out;
4316 }
4317
075897ce
JL		 4318 if (!bond->curr_active_slave)
4319 goto out;
4320
075897ce
JL		 4321 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4322
1da177e4
LT		 4323out:
4324 if (res) {
4325 /* no suitable interface, frame not sent */
4326 dev_kfree_skb(skb);
4327 }
4328 read_unlock(&bond->curr_slave_lock);
4329 read_unlock(&bond->lock);
4330 return 0;
4331}
4332
4333/*
169a3e66
JV		 4334 * In bond_xmit_xor() , we determine the output device by using a pre-
4335 * determined xmit_hash_policy(), If the selected device is not enabled,
4336 * find the next active slave.
1da177e4
LT		 4337 */
4338static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4339{
4340 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 4341 struct slave *slave, *start_at;
4342 int slave_no;
4343 int i;
4344 int res = 1;
4345
4346 read_lock(&bond->lock);
4347
4348 if (!BOND_IS_OK(bond)) {
4349 goto out;
4350 }
4351
169a3e66 4352 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4353
4354 bond_for_each_slave(bond, slave, i) {
4355 slave_no--;
4356 if (slave_no < 0) {
4357 break;
4358 }
4359 }
4360
4361 start_at = slave;
4362
4363 bond_for_each_slave_from(bond, slave, i, start_at) {
4364 if (IS_UP(slave->dev) &&
4365 (slave->link == BOND_LINK_UP) &&
4366 (slave->state == BOND_STATE_ACTIVE)) {
4367 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4368 break;
4369 }
4370 }
4371
4372out:
4373 if (res) {
4374 /* no suitable interface, frame not sent */
4375 dev_kfree_skb(skb);
4376 }
4377 read_unlock(&bond->lock);
4378 return 0;
4379}
4380
4381/*
4382 * in broadcast mode, we send everything to all usable interfaces.
4383 */
4384static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4385{
4386 struct bonding *bond = bond_dev->priv;
4387 struct slave *slave, *start_at;
4388 struct net_device *tx_dev = NULL;
4389 int i;
4390 int res = 1;
4391
4392 read_lock(&bond->lock);
4393
4394 if (!BOND_IS_OK(bond)) {
4395 goto out;
4396 }
4397
4398 read_lock(&bond->curr_slave_lock);
4399 start_at = bond->curr_active_slave;
4400 read_unlock(&bond->curr_slave_lock);
4401
4402 if (!start_at) {
4403 goto out;
4404 }
4405
4406 bond_for_each_slave_from(bond, slave, i, start_at) {
4407 if (IS_UP(slave->dev) &&
4408 (slave->link == BOND_LINK_UP) &&
4409 (slave->state == BOND_STATE_ACTIVE)) {
4410 if (tx_dev) {
4411 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4412 if (!skb2) {
4413 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4414 ": %s: Error: bond_xmit_broadcast(): "
4415 "skb_clone() failed\n",
4416 bond_dev->name);
1da177e4
LT		 4417 continue;
4418 }
4419
4420 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4421 if (res) {
4422 dev_kfree_skb(skb2);
4423 continue;
4424 }
4425 }
4426 tx_dev = slave->dev;
4427 }
4428 }
4429
4430 if (tx_dev) {
4431 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4432 }
4433
4434out:
4435 if (res) {
4436 /* no suitable interface, frame not sent */
4437 dev_kfree_skb(skb);
4438 }
4439 /* frame sent to all suitable interfaces */
4440 read_unlock(&bond->lock);
4441 return 0;
4442}
4443
4444/*------------------------- Device initialization ---------------------------*/
4445
6f6652be
JV		 4446static void bond_set_xmit_hash_policy(struct bonding *bond)
4447{
4448 switch (bond->params.xmit_policy) {
4449 case BOND_XMIT_POLICY_LAYER23:
4450 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4451 break;
4452 case BOND_XMIT_POLICY_LAYER34:
4453 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4454 break;
4455 case BOND_XMIT_POLICY_LAYER2:
4456 default:
4457 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4458 break;
4459 }
4460}
4461
1da177e4
LT		 4462/*
4463 * set bond mode specific net device operations
4464 */
a77b5325 4465void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4466{
169a3e66
JV		 4467 struct net_device *bond_dev = bond->dev;
4468
1da177e4
LT		 4469 switch (mode) {
4470 case BOND_MODE_ROUNDROBIN:
4471 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4472 break;
4473 case BOND_MODE_ACTIVEBACKUP:
4474 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4475 break;
4476 case BOND_MODE_XOR:
4477 bond_dev->hard_start_xmit = bond_xmit_xor;
6f6652be 4478 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4479 break;
4480 case BOND_MODE_BROADCAST:
4481 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4482 break;
4483 case BOND_MODE_8023AD:
8f903c70 4484 bond_set_master_3ad_flags(bond);
1da177e4 4485 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
6f6652be 4486 bond_set_xmit_hash_policy(bond);
1da177e4 4487 break;
1da177e4 4488 case BOND_MODE_ALB:
8f903c70
JV		 4489 bond_set_master_alb_flags(bond);
4490 /* FALLTHRU */
4491 case BOND_MODE_TLB:
1da177e4
LT		 4492 bond_dev->hard_start_xmit = bond_alb_xmit;
4493 bond_dev->set_mac_address = bond_alb_set_mac_address;
4494 break;
4495 default:
4496 /* Should never happen, mode already checked */
4497 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4498 ": %s: Error: Unknown bonding mode %d\n",
4499 bond_dev->name,
1da177e4
LT		 4500 mode);
4501 break;
4502 }
4503}
4504
217df670
JV		 4505static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4506 struct ethtool_drvinfo *drvinfo)
4507{
4508 strncpy(drvinfo->driver, DRV_NAME, 32);
4509 strncpy(drvinfo->version, DRV_VERSION, 32);
4510 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4511}
4512
7282d491 4513static const struct ethtool_ops bond_ethtool_ops = {
217df670 4514 .get_drvinfo = bond_ethtool_get_drvinfo,
8531c5ff
AK		 4515};
4516
1da177e4
LT		 4517/*
4518 * Does not allocate but creates a /proc entry.
4519 * Allowed to fail.
4520 */
3c535952 4521static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4
LT		 4522{
4523 struct bonding *bond = bond_dev->priv;
4524
4525 dprintk("Begin bond_init for %s\n", bond_dev->name);
4526
4527 /* initialize rwlocks */
4528 rwlock_init(&bond->lock);
4529 rwlock_init(&bond->curr_slave_lock);
4530
4531 bond->params = *params; /* copy params struct */
4532
1b76b316
JV		 4533 bond->wq = create_singlethread_workqueue(bond_dev->name);
4534 if (!bond->wq)
4535 return -ENOMEM;
4536
1da177e4
LT		 4537 /* Initialize pointers */
4538 bond->first_slave = NULL;
4539 bond->curr_active_slave = NULL;
4540 bond->current_arp_slave = NULL;
4541 bond->primary_slave = NULL;
4542 bond->dev = bond_dev;
1053f62c 4543 bond->send_grat_arp = 0;
d90a162a 4544 bond->setup_by_slave = 0;
1da177e4
LT		 4545 INIT_LIST_HEAD(&bond->vlan_list);
4546
4547 /* Initialize the device entry points */
4548 bond_dev->open = bond_open;
4549 bond_dev->stop = bond_close;
4550 bond_dev->get_stats = bond_get_stats;
4551 bond_dev->do_ioctl = bond_do_ioctl;
8531c5ff 4552 bond_dev->ethtool_ops = &bond_ethtool_ops;
1da177e4
LT		 4553 bond_dev->set_multicast_list = bond_set_multicast_list;
4554 bond_dev->change_mtu = bond_change_mtu;
4555 bond_dev->set_mac_address = bond_set_mac_address;
3a1521b7 4556 bond_dev->validate_addr = NULL;
1da177e4 4557
169a3e66 4558 bond_set_mode_ops(bond, bond->params.mode);
1da177e4
LT		 4559
4560 bond_dev->destructor = free_netdev;
4561
4562 /* Initialize the device options */
4563 bond_dev->tx_queue_len = 0;
4564 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4565 bond_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 4566
4567 /* At first, we block adding VLANs. That's the only way to
4568 * prevent problems that occur when adding VLANs over an
4569 * empty bond. The block will be removed once non-challenged
4570 * slaves are enslaved.
4571 */
4572 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4573
932ff279 4574 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4575 * transmitting */
4576 bond_dev->features |= NETIF_F_LLTX;
4577
4578 /* By default, we declare the bond to be fully
4579 * VLAN hardware accelerated capable. Special
4580 * care is taken in the various xmit functions
4581 * when there are slaves that are not hw accel
4582 * capable
4583 */
4584 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4585 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4586 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4587 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4588 NETIF_F_HW_VLAN_RX |
4589 NETIF_F_HW_VLAN_FILTER);
4590
4591#ifdef CONFIG_PROC_FS
4592 bond_create_proc_entry(bond);
4593#endif
1da177e4
LT		 4594 list_add_tail(&bond->bond_list, &bond_dev_list);
4595
4596 return 0;
4597}
4598
4599/* De-initialize device specific data.
4600 * Caller must hold rtnl_lock.
4601 */
c50b85d0 4602static void bond_deinit(struct net_device *bond_dev)
1da177e4
LT		 4603{
4604 struct bonding *bond = bond_dev->priv;
4605
4606 list_del(&bond->bond_list);
4607
4608#ifdef CONFIG_PROC_FS
4609 bond_remove_proc_entry(bond);
4610#endif
4611}
4612
fdaea7a9
JV		 4613static void bond_work_cancel_all(struct bonding *bond)
4614{
4615 write_lock_bh(&bond->lock);
4616 bond->kill_timers = 1;
4617 write_unlock_bh(&bond->lock);
4618
4619 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4620 cancel_delayed_work(&bond->mii_work);
4621
4622 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4623 cancel_delayed_work(&bond->arp_work);
4624
4625 if (bond->params.mode == BOND_MODE_ALB &&
4626 delayed_work_pending(&bond->alb_work))
4627 cancel_delayed_work(&bond->alb_work);
4628
4629 if (bond->params.mode == BOND_MODE_8023AD &&
4630 delayed_work_pending(&bond->ad_work))
4631 cancel_delayed_work(&bond->ad_work);
4632}
4633
1da177e4
LT		 4634/* Unregister and free all bond devices.
4635 * Caller must hold rtnl_lock.
4636 */
4637static void bond_free_all(void)
4638{
4639 struct bonding *bond, *nxt;
4640
4641 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4642 struct net_device *bond_dev = bond->dev;
4643
fdaea7a9 4644 bond_work_cancel_all(bond);
80ee5ad2 4645 netif_tx_lock_bh(bond_dev);
70298705 4646 bond_mc_list_destroy(bond);
80ee5ad2 4647 netif_tx_unlock_bh(bond_dev);
70298705 4648 /* Release the bonded slaves */
4649 bond_release_all(bond_dev);
92b41daa 4650 bond_destroy(bond);
1da177e4
LT		 4651 }
4652
4653#ifdef CONFIG_PROC_FS
4654 bond_destroy_proc_dir();
4655#endif
4656}
4657
4658/*------------------------- Module initialization ---------------------------*/
4659
4660/*
4661 * Convert string input module parms. Accept either the
ece95f7f
JV		 4662 * number of the mode or its string name. A bit complicated because
4663 * some mode names are substrings of other names, and calls from sysfs
4664 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4665 */
ece95f7f 4666int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4667{
ece95f7f 4668 int mode = -1, i, rv;
a42e534f 4669 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4670
a42e534f
JV		 4671 for (p = (char *)buf; *p; p++)
4672 if (!(isdigit(*p) || isspace(*p)))
4673 break;
4674
4675 if (*p)
ece95f7f 4676 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4677 else
4678 rv = sscanf(buf, "%d", &mode);
4679
4680 if (!rv)
4681 return -1;
1da177e4
LT		 4682
4683 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4684 if (mode == tbl[i].mode)
4685 return tbl[i].mode;
4686 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4687 return tbl[i].mode;
1da177e4
LT		 4688 }
4689
4690 return -1;
4691}
4692
4693static int bond_check_params(struct bond_params *params)
4694{
3915c1e8 4695 int arp_validate_value, fail_over_mac_value;
f5b2b966 4696
1da177e4
LT		 4697 /*
4698 * Convert string parameters.
4699 */
4700 if (mode) {
4701 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4702 if (bond_mode == -1) {
4703 printk(KERN_ERR DRV_NAME
4704 ": Error: Invalid bonding mode \"%s\"\n",
4705 mode == NULL ? "NULL" : mode);
4706 return -EINVAL;
4707 }
4708 }
4709
169a3e66
JV		 4710 if (xmit_hash_policy) {
4711 if ((bond_mode != BOND_MODE_XOR) &&
4712 (bond_mode != BOND_MODE_8023AD)) {
4713 printk(KERN_INFO DRV_NAME
4714 ": xor_mode param is irrelevant in mode %s\n",
4715 bond_mode_name(bond_mode));
4716 } else {
4717 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4718 xmit_hashtype_tbl);
4719 if (xmit_hashtype == -1) {
4720 printk(KERN_ERR DRV_NAME
4721 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4722 xmit_hash_policy == NULL ? "NULL" :
4723 xmit_hash_policy);
4724 return -EINVAL;
4725 }
4726 }
4727 }
4728
1da177e4
LT		 4729 if (lacp_rate) {
4730 if (bond_mode != BOND_MODE_8023AD) {
4731 printk(KERN_INFO DRV_NAME
4732 ": lacp_rate param is irrelevant in mode %s\n",
4733 bond_mode_name(bond_mode));
4734 } else {
4735 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4736 if (lacp_fast == -1) {
4737 printk(KERN_ERR DRV_NAME
4738 ": Error: Invalid lacp rate \"%s\"\n",
4739 lacp_rate == NULL ? "NULL" : lacp_rate);
4740 return -EINVAL;
4741 }
4742 }
4743 }
4744
4745 if (max_bonds < 1 || max_bonds > INT_MAX) {
4746 printk(KERN_WARNING DRV_NAME
4747 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4748 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
1da177e4
LT		 4749 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4750 max_bonds = BOND_DEFAULT_MAX_BONDS;
4751 }
4752
4753 if (miimon < 0) {
4754 printk(KERN_WARNING DRV_NAME
4755 ": Warning: miimon module parameter (%d), "
4756 "not in range 0-%d, so it was reset to %d\n",
4757 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4758 miimon = BOND_LINK_MON_INTERV;
4759 }
4760
4761 if (updelay < 0) {
4762 printk(KERN_WARNING DRV_NAME
4763 ": Warning: updelay module parameter (%d), "
4764 "not in range 0-%d, so it was reset to 0\n",
4765 updelay, INT_MAX);
4766 updelay = 0;
4767 }
4768
4769 if (downdelay < 0) {
4770 printk(KERN_WARNING DRV_NAME
4771 ": Warning: downdelay module parameter (%d), "
4772 "not in range 0-%d, so it was reset to 0\n",
4773 downdelay, INT_MAX);
4774 downdelay = 0;
4775 }
4776
4777 if ((use_carrier != 0) && (use_carrier != 1)) {
4778 printk(KERN_WARNING DRV_NAME
4779 ": Warning: use_carrier module parameter (%d), "
4780 "not of valid value (0/1), so it was set to 1\n",
4781 use_carrier);
4782 use_carrier = 1;
4783 }
4784
7893b249
MS		 4785 if (num_grat_arp < 0 || num_grat_arp > 255) {
4786 printk(KERN_WARNING DRV_NAME
4787 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4788 "was reset to 1 \n", num_grat_arp);
4789 num_grat_arp = 1;
4790 }
4791
1da177e4
LT		 4792 /* reset values for 802.3ad */
4793 if (bond_mode == BOND_MODE_8023AD) {
4794 if (!miimon) {
4795 printk(KERN_WARNING DRV_NAME
4796 ": Warning: miimon must be specified, "
4797 "otherwise bonding will not detect link "
4798 "failure, speed and duplex which are "
4799 "essential for 802.3ad operation\n");
4800 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4801 miimon = 100;
4802 }
4803 }
4804
4805 /* reset values for TLB/ALB */
4806 if ((bond_mode == BOND_MODE_TLB) ||
4807 (bond_mode == BOND_MODE_ALB)) {
4808 if (!miimon) {
4809 printk(KERN_WARNING DRV_NAME
4810 ": Warning: miimon must be specified, "
4811 "otherwise bonding will not detect link "
4812 "failure and link speed which are essential "
4813 "for TLB/ALB load balancing\n");
4814 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4815 miimon = 100;
4816 }
4817 }
4818
4819 if (bond_mode == BOND_MODE_ALB) {
4820 printk(KERN_NOTICE DRV_NAME
4821 ": In ALB mode you might experience client "
4822 "disconnections upon reconnection of a link if the "
4823 "bonding module updelay parameter (%d msec) is "
4824 "incompatible with the forwarding delay time of the "
4825 "switch\n",
4826 updelay);
4827 }
4828
4829 if (!miimon) {
4830 if (updelay || downdelay) {
4831 /* just warn the user the up/down delay will have
4832 * no effect since miimon is zero...
4833 */
4834 printk(KERN_WARNING DRV_NAME
4835 ": Warning: miimon module parameter not set "
4836 "and updelay (%d) or downdelay (%d) module "
4837 "parameter is set; updelay and downdelay have "
4838 "no effect unless miimon is set\n",
4839 updelay, downdelay);
4840 }
4841 } else {
4842 /* don't allow arp monitoring */
4843 if (arp_interval) {
4844 printk(KERN_WARNING DRV_NAME
4845 ": Warning: miimon (%d) and arp_interval (%d) "
4846 "can't be used simultaneously, disabling ARP "
4847 "monitoring\n",
4848 miimon, arp_interval);
4849 arp_interval = 0;
4850 }
4851
4852 if ((updelay % miimon) != 0) {
4853 printk(KERN_WARNING DRV_NAME
4854 ": Warning: updelay (%d) is not a multiple "
4855 "of miimon (%d), updelay rounded to %d ms\n",
4856 updelay, miimon, (updelay / miimon) * miimon);
4857 }
4858
4859 updelay /= miimon;
4860
4861 if ((downdelay % miimon) != 0) {
4862 printk(KERN_WARNING DRV_NAME
4863 ": Warning: downdelay (%d) is not a multiple "
4864 "of miimon (%d), downdelay rounded to %d ms\n",
4865 downdelay, miimon,
4866 (downdelay / miimon) * miimon);
4867 }
4868
4869 downdelay /= miimon;
4870 }
4871
4872 if (arp_interval < 0) {
4873 printk(KERN_WARNING DRV_NAME
4874 ": Warning: arp_interval module parameter (%d) "
4875 ", not in range 0-%d, so it was reset to %d\n",
4876 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4877 arp_interval = BOND_LINK_ARP_INTERV;
4878 }
4879
4880 for (arp_ip_count = 0;
4881 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4882 arp_ip_count++) {
4883 /* not complete check, but should be good enough to
4884 catch mistakes */
4885 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4886 printk(KERN_WARNING DRV_NAME
4887 ": Warning: bad arp_ip_target module parameter "
4888 "(%s), ARP monitoring will not be performed\n",
4889 arp_ip_target[arp_ip_count]);
4890 arp_interval = 0;
4891 } else {
d3bb52b0 4892 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4893 arp_target[arp_ip_count] = ip;
4894 }
4895 }
4896
4897 if (arp_interval && !arp_ip_count) {
4898 /* don't allow arping if no arp_ip_target given... */
4899 printk(KERN_WARNING DRV_NAME
4900 ": Warning: arp_interval module parameter (%d) "
4901 "specified without providing an arp_ip_target "
4902 "parameter, arp_interval was reset to 0\n",
4903 arp_interval);
4904 arp_interval = 0;
4905 }
4906
f5b2b966
JV		 4907 if (arp_validate) {
4908 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4909 printk(KERN_ERR DRV_NAME
4910 ": arp_validate only supported in active-backup mode\n");
4911 return -EINVAL;
4912 }
4913 if (!arp_interval) {
4914 printk(KERN_ERR DRV_NAME
4915 ": arp_validate requires arp_interval\n");
4916 return -EINVAL;
4917 }
4918
4919 arp_validate_value = bond_parse_parm(arp_validate,
4920 arp_validate_tbl);
4921 if (arp_validate_value == -1) {
4922 printk(KERN_ERR DRV_NAME
4923 ": Error: invalid arp_validate \"%s\"\n",
4924 arp_validate == NULL ? "NULL" : arp_validate);
4925 return -EINVAL;
4926 }
4927 } else
4928 arp_validate_value = 0;
4929
1da177e4
LT		 4930 if (miimon) {
4931 printk(KERN_INFO DRV_NAME
4932 ": MII link monitoring set to %d ms\n",
4933 miimon);
4934 } else if (arp_interval) {
4935 int i;
4936
4937 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 4938 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4939 arp_interval,
4940 arp_validate_tbl[arp_validate_value].modename,
4941 arp_ip_count);
1da177e4
LT		 4942
4943 for (i = 0; i < arp_ip_count; i++)
4944 printk (" %s", arp_ip_target[i]);
4945
4946 printk("\n");
4947
4948 } else {
4949 /* miimon and arp_interval not set, we need one so things
4950 * work as expected, see bonding.txt for details
4951 */
4952 printk(KERN_WARNING DRV_NAME
4953 ": Warning: either miimon or arp_interval and "
4954 "arp_ip_target module parameters must be specified, "
4955 "otherwise bonding will not detect link failures! see "
4956 "bonding.txt for details.\n");
4957 }
4958
4959 if (primary && !USES_PRIMARY(bond_mode)) {
4960 /* currently, using a primary only makes sense
4961 * in active backup, TLB or ALB modes
4962 */
4963 printk(KERN_WARNING DRV_NAME
4964 ": Warning: %s primary device specified but has no "
4965 "effect in %s mode\n",
4966 primary, bond_mode_name(bond_mode));
4967 primary = NULL;
4968 }
4969
3915c1e8
JV		 4970 if (fail_over_mac) {
4971 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4972 fail_over_mac_tbl);
4973 if (fail_over_mac_value == -1) {
4974 printk(KERN_ERR DRV_NAME
4975 ": Error: invalid fail_over_mac \"%s\"\n",
4976 arp_validate == NULL ? "NULL" : arp_validate);
4977 return -EINVAL;
4978 }
4979
4980 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4981 printk(KERN_WARNING DRV_NAME
4982 ": Warning: fail_over_mac only affects "
4983 "active-backup mode.\n");
4984 } else {
4985 fail_over_mac_value = BOND_FOM_NONE;
4986 }
dd957c57 4987
1da177e4
LT		 4988 /* fill params struct with the proper values */
4989 params->mode = bond_mode;
169a3e66 4990 params->xmit_policy = xmit_hashtype;
1da177e4 4991 params->miimon = miimon;
7893b249 4992 params->num_grat_arp = num_grat_arp;
1da177e4 4993 params->arp_interval = arp_interval;
f5b2b966 4994 params->arp_validate = arp_validate_value;
1da177e4
LT		 4995 params->updelay = updelay;
4996 params->downdelay = downdelay;
4997 params->use_carrier = use_carrier;
4998 params->lacp_fast = lacp_fast;
4999 params->primary[0] = 0;
3915c1e8 5000 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5001
5002 if (primary) {
5003 strncpy(params->primary, primary, IFNAMSIZ);
5004 params->primary[IFNAMSIZ - 1] = 0;
5005 }
5006
5007 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5008
5009 return 0;
5010}
5011
0daa2303
PZ		 5012static struct lock_class_key bonding_netdev_xmit_lock_key;
5013
dfe60397 5014/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5015 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5016 * Caller must NOT hold rtnl_lock; we need to release it here before we
5017 * set up our sysfs entries.
5018 */
0dd646fe 5019int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5020{
5021 struct net_device *bond_dev;
0883beca 5022 struct bonding *bond;
dfe60397
MW		 5023 int res;
5024
5025 rtnl_lock();
027ea041
JV		 5026 down_write(&bonding_rwsem);
5027
5028 /* Check to see if the bond already exists. */
4fe4763c 5029 if (name) {
0883beca 5030 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5031 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5032 printk(KERN_ERR DRV_NAME
027ea041 5033 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5034 name);
5035 res = -EPERM;
5036 goto out_rtnl;
5037 }
5038 }
027ea041 5039
e4b91c48
JV		 5040 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5041 ether_setup);
dfe60397
MW		 5042 if (!bond_dev) {
5043 printk(KERN_ERR DRV_NAME
5044 ": %s: eek! can't alloc netdev!\n",
5045 name);
5046 res = -ENOMEM;
5047 goto out_rtnl;
5048 }
5049
e4b91c48
JV		 5050 if (!name) {
5051 res = dev_alloc_name(bond_dev, "bond%d");
5052 if (res < 0)
5053 goto out_netdev;
5054 }
5055
dfe60397
MW		 5056 /* bond_init() must be called after dev_alloc_name() (for the
5057 * /proc files), but before register_netdevice(), because we
5058 * need to set function pointers.
5059 */
5060
5061 res = bond_init(bond_dev, params);
5062 if (res < 0) {
5063 goto out_netdev;
5064 }
5065
dfe60397
MW		 5066 res = register_netdevice(bond_dev);
5067 if (res < 0) {
5068 goto out_bond;
5069 }
0daa2303
PZ		 5070
5071 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
5072
ff59c456
JV		 5073 netif_carrier_off(bond_dev);
5074
027ea041 5075 up_write(&bonding_rwsem);
dfe60397 5076 rtnl_unlock(); /* allows sysfs registration of net device */
b76cdba9 5077 res = bond_create_sysfs_entry(bond_dev->priv);
09c89279
JV		 5078 if (res < 0) {
5079 rtnl_lock();
027ea041 5080 down_write(&bonding_rwsem);
822973ba
PE		 5081 bond_deinit(bond_dev);
5082 unregister_netdevice(bond_dev);
5083 goto out_rtnl;
09c89279
JV		 5084 }
5085
5086 return 0;
5087
dfe60397
MW		 5088out_bond:
5089 bond_deinit(bond_dev);
5090out_netdev:
5091 free_netdev(bond_dev);
5092out_rtnl:
027ea041 5093 up_write(&bonding_rwsem);
dfe60397 5094 rtnl_unlock();
dfe60397
MW		 5095 return res;
5096}
5097
1da177e4
LT		 5098static int __init bonding_init(void)
5099{
1da177e4
LT		 5100 int i;
5101 int res;
0883beca 5102 struct bonding *bond;
1da177e4
LT		 5103
5104 printk(KERN_INFO "%s", version);
5105
dfe60397 5106 res = bond_check_params(&bonding_defaults);
1da177e4 5107 if (res) {
dfe60397 5108 goto out;
1da177e4
LT		 5109 }
5110
1da177e4
LT		 5111#ifdef CONFIG_PROC_FS
5112 bond_create_proc_dir();
5113#endif
027ea041
JV		 5114
5115 init_rwsem(&bonding_rwsem);
5116
1da177e4 5117 for (i = 0; i < max_bonds; i++) {
0dd646fe 5118 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5119 if (res)
5120 goto err;
1da177e4
LT		 5121 }
5122
b76cdba9
MW		 5123 res = bond_create_sysfs();
5124 if (res)
5125 goto err;
5126
1da177e4 5127 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5128 register_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5129
dfe60397
MW		 5130 goto out;
5131err:
0883beca 5132 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5133 bond_work_cancel_all(bond);
5134 destroy_workqueue(bond->wq);
5135 }
5136
ae68c398
PE		 5137 bond_destroy_sysfs();
5138
40abc270 5139 rtnl_lock();
1da177e4 5140 bond_free_all();
1da177e4 5141 rtnl_unlock();
dfe60397 5142out:
1da177e4 5143 return res;
dfe60397 5144
1da177e4
LT		 5145}
5146
5147static void __exit bonding_exit(void)
5148{
5149 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5150 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5151
ae68c398
PE		 5152 bond_destroy_sysfs();
5153
1da177e4
LT		 5154 rtnl_lock();
5155 bond_free_all();
5156 rtnl_unlock();
5157}
5158
5159module_init(bonding_init);
5160module_exit(bonding_exit);
5161MODULE_LICENSE("GPL");
5162MODULE_VERSION(DRV_VERSION);
5163MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5164MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5165MODULE_SUPPORTED_DEVICE("most ethernet devices");
5166
5167/*
5168 * Local variables:
5169 * c-indent-level: 8
5170 * c-basic-offset: 8
5171 * tab-width: 8
5172 * End:
5173 */
5174
Clone of davem's net-next-2.6 tree