]> 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
netdev: Add netdev->addr_list_lock protection.
[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 */
7e1a1ac1 775static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 776{
7e1a1ac1 777 int err = 0;
1da177e4
LT		 778 if (USES_PRIMARY(bond->params.mode)) {
779 /* write lock already acquired */
780 if (bond->curr_active_slave) {
7e1a1ac1
WC		 781 err = dev_set_promiscuity(bond->curr_active_slave->dev,
782 inc);
1da177e4
LT		 783 }
784 } else {
785 struct slave *slave;
786 int i;
787 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 788 err = dev_set_promiscuity(slave->dev, inc);
789 if (err)
790 return err;
1da177e4
LT		 791 }
792 }
7e1a1ac1 793 return err;
1da177e4
LT		 794}
795
796/*
797 * Push the allmulti flag down to all slaves
798 */
7e1a1ac1 799static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 800{
7e1a1ac1 801 int err = 0;
1da177e4
LT		 802 if (USES_PRIMARY(bond->params.mode)) {
803 /* write lock already acquired */
804 if (bond->curr_active_slave) {
7e1a1ac1
WC		 805 err = dev_set_allmulti(bond->curr_active_slave->dev,
806 inc);
1da177e4
LT		 807 }
808 } else {
809 struct slave *slave;
810 int i;
811 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 812 err = dev_set_allmulti(slave->dev, inc);
813 if (err)
814 return err;
1da177e4
LT		 815 }
816 }
7e1a1ac1 817 return err;
1da177e4
LT		 818}
819
820/*
821 * Add a Multicast address to slaves
822 * according to mode
823 */
824static void bond_mc_add(struct bonding *bond, void *addr, int alen)
825{
826 if (USES_PRIMARY(bond->params.mode)) {
827 /* write lock already acquired */
828 if (bond->curr_active_slave) {
829 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
830 }
831 } else {
832 struct slave *slave;
833 int i;
834 bond_for_each_slave(bond, slave, i) {
835 dev_mc_add(slave->dev, addr, alen, 0);
836 }
837 }
838}
839
840/*
841 * Remove a multicast address from slave
842 * according to mode
843 */
844static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
845{
846 if (USES_PRIMARY(bond->params.mode)) {
847 /* write lock already acquired */
848 if (bond->curr_active_slave) {
849 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
850 }
851 } else {
852 struct slave *slave;
853 int i;
854 bond_for_each_slave(bond, slave, i) {
855 dev_mc_delete(slave->dev, addr, alen, 0);
856 }
857 }
858}
859
a816c7c7
JV		 860
861/*
862 * Retrieve the list of registered multicast addresses for the bonding
863 * device and retransmit an IGMP JOIN request to the current active
864 * slave.
865 */
866static void bond_resend_igmp_join_requests(struct bonding *bond)
867{
868 struct in_device *in_dev;
869 struct ip_mc_list *im;
870
871 rcu_read_lock();
872 in_dev = __in_dev_get_rcu(bond->dev);
873 if (in_dev) {
874 for (im = in_dev->mc_list; im; im = im->next) {
875 ip_mc_rejoin_group(im);
876 }
877 }
878
879 rcu_read_unlock();
880}
881
1da177e4
LT		 882/*
883 * Totally destroys the mc_list in bond
884 */
885static void bond_mc_list_destroy(struct bonding *bond)
886{
887 struct dev_mc_list *dmi;
888
889 dmi = bond->mc_list;
890 while (dmi) {
891 bond->mc_list = dmi->next;
892 kfree(dmi);
893 dmi = bond->mc_list;
894 }
a816c7c7 895 bond->mc_list = NULL;
1da177e4
LT		 896}
897
898/*
899 * Copy all the Multicast addresses from src to the bonding device dst
900 */
de54f390 901static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 902 gfp_t gfp_flag)
1da177e4
LT		 903{
904 struct dev_mc_list *dmi, *new_dmi;
905
906 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 907 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT		 908
909 if (!new_dmi) {
910 /* FIXME: Potential memory leak !!! */
911 return -ENOMEM;
912 }
913
914 new_dmi->next = bond->mc_list;
915 bond->mc_list = new_dmi;
916 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
917 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
918 new_dmi->dmi_users = dmi->dmi_users;
919 new_dmi->dmi_gusers = dmi->dmi_gusers;
920 }
921
922 return 0;
923}
924
925/*
926 * flush all members of flush->mc_list from device dev->mc_list
927 */
928static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
929{
930 struct bonding *bond = bond_dev->priv;
931 struct dev_mc_list *dmi;
932
933 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
934 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
935 }
936
937 if (bond->params.mode == BOND_MODE_8023AD) {
938 /* del lacpdu mc addr from mc list */
939 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
940
941 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
942 }
943}
944
945/*--------------------------- Active slave change ---------------------------*/
946
947/*
948 * Update the mc list and multicast-related flags for the new and
949 * old active slaves (if any) according to the multicast mode, and
950 * promiscuous flags unconditionally.
951 */
952static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
953{
954 struct dev_mc_list *dmi;
955
956 if (!USES_PRIMARY(bond->params.mode)) {
957 /* nothing to do - mc list is already up-to-date on
958 * all slaves
959 */
960 return;
961 }
962
963 if (old_active) {
964 if (bond->dev->flags & IFF_PROMISC) {
965 dev_set_promiscuity(old_active->dev, -1);
966 }
967
968 if (bond->dev->flags & IFF_ALLMULTI) {
969 dev_set_allmulti(old_active->dev, -1);
970 }
971
972 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
973 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
974 }
975 }
976
977 if (new_active) {
7e1a1ac1 978 /* FIXME: Signal errors upstream. */
1da177e4
LT		 979 if (bond->dev->flags & IFF_PROMISC) {
980 dev_set_promiscuity(new_active->dev, 1);
981 }
982
983 if (bond->dev->flags & IFF_ALLMULTI) {
984 dev_set_allmulti(new_active->dev, 1);
985 }
986
987 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
988 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
989 }
a816c7c7 990 bond_resend_igmp_join_requests(bond);
1da177e4
LT		 991 }
992}
993
3915c1e8
JV		 994/*
995 * bond_do_fail_over_mac
996 *
997 * Perform special MAC address swapping for fail_over_mac settings
998 *
999 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1000 */
1001static void bond_do_fail_over_mac(struct bonding *bond,
1002 struct slave *new_active,
1003 struct slave *old_active)
1004{
1005 u8 tmp_mac[ETH_ALEN];
1006 struct sockaddr saddr;
1007 int rv;
1008
1009 switch (bond->params.fail_over_mac) {
1010 case BOND_FOM_ACTIVE:
1011 if (new_active)
1012 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1013 new_active->dev->addr_len);
1014 break;
1015 case BOND_FOM_FOLLOW:
1016 /*
1017 * if new_active && old_active, swap them
1018 * if just old_active, do nothing (going to no active slave)
1019 * if just new_active, set new_active to bond's MAC
1020 */
1021 if (!new_active)
1022 return;
1023
1024 write_unlock_bh(&bond->curr_slave_lock);
1025 read_unlock(&bond->lock);
1026
1027 if (old_active) {
1028 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1029 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1030 ETH_ALEN);
1031 saddr.sa_family = new_active->dev->type;
1032 } else {
1033 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1034 saddr.sa_family = bond->dev->type;
1035 }
1036
1037 rv = dev_set_mac_address(new_active->dev, &saddr);
1038 if (rv) {
1039 printk(KERN_ERR DRV_NAME
1040 ": %s: Error %d setting MAC of slave %s\n",
1041 bond->dev->name, -rv, new_active->dev->name);
1042 goto out;
1043 }
1044
1045 if (!old_active)
1046 goto out;
1047
1048 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1049 saddr.sa_family = old_active->dev->type;
1050
1051 rv = dev_set_mac_address(old_active->dev, &saddr);
1052 if (rv)
1053 printk(KERN_ERR DRV_NAME
1054 ": %s: Error %d setting MAC of slave %s\n",
1055 bond->dev->name, -rv, new_active->dev->name);
1056out:
1057 read_lock(&bond->lock);
1058 write_lock_bh(&bond->curr_slave_lock);
1059 break;
1060 default:
1061 printk(KERN_ERR DRV_NAME
1062 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1063 bond->dev->name, bond->params.fail_over_mac);
1064 break;
1065 }
1066
1067}
1068
1069
1da177e4
LT		 1070/**
1071 * find_best_interface - select the best available slave to be the active one
1072 * @bond: our bonding struct
1073 *
1074 * Warning: Caller must hold curr_slave_lock for writing.
1075 */
1076static struct slave *bond_find_best_slave(struct bonding *bond)
1077{
1078 struct slave *new_active, *old_active;
1079 struct slave *bestslave = NULL;
1080 int mintime = bond->params.updelay;
1081 int i;
1082
1083 new_active = old_active = bond->curr_active_slave;
1084
1085 if (!new_active) { /* there were no active slaves left */
1086 if (bond->slave_cnt > 0) { /* found one slave */
1087 new_active = bond->first_slave;
1088 } else {
1089 return NULL; /* still no slave, return NULL */
1090 }
1091 }
1092
1093 /* first try the primary link; if arping, a link must tx/rx traffic
1094 * before it can be considered the curr_active_slave - also, we would skip
1095 * slaves between the curr_active_slave and primary_slave that may be up
1096 * and able to arp
1097 */
1098 if ((bond->primary_slave) &&
1099 (!bond->params.arp_interval) &&
1100 (IS_UP(bond->primary_slave->dev))) {
1101 new_active = bond->primary_slave;
1102 }
1103
1104 /* remember where to stop iterating over the slaves */
1105 old_active = new_active;
1106
1107 bond_for_each_slave_from(bond, new_active, i, old_active) {
1108 if (IS_UP(new_active->dev)) {
1109 if (new_active->link == BOND_LINK_UP) {
1110 return new_active;
1111 } else if (new_active->link == BOND_LINK_BACK) {
1112 /* link up, but waiting for stabilization */
1113 if (new_active->delay < mintime) {
1114 mintime = new_active->delay;
1115 bestslave = new_active;
1116 }
1117 }
1118 }
1119 }
1120
1121 return bestslave;
1122}
1123
1124/**
1125 * change_active_interface - change the active slave into the specified one
1126 * @bond: our bonding struct
1127 * @new: the new slave to make the active one
1128 *
1129 * Set the new slave to the bond's settings and unset them on the old
1130 * curr_active_slave.
1131 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1132 *
1133 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1134 * because it is apparently the best available slave we have, even though its
1135 * updelay hasn't timed out yet.
1136 *
3915c1e8
JV		 1137 * If new_active is not NULL, caller must hold bond->lock for read and
1138 * curr_slave_lock for write_bh.
1da177e4 1139 */
a77b5325 1140void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1141{
1142 struct slave *old_active = bond->curr_active_slave;
1143
1144 if (old_active == new_active) {
1145 return;
1146 }
1147
1148 if (new_active) {
b2220cad
JV		 1149 new_active->jiffies = jiffies;
1150
1da177e4
LT		 1151 if (new_active->link == BOND_LINK_BACK) {
1152 if (USES_PRIMARY(bond->params.mode)) {
1153 printk(KERN_INFO DRV_NAME
1154 ": %s: making interface %s the new "
1155 "active one %d ms earlier.\n",
1156 bond->dev->name, new_active->dev->name,
1157 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1158 }
1159
1160 new_active->delay = 0;
1161 new_active->link = BOND_LINK_UP;
1da177e4
LT		 1162
1163 if (bond->params.mode == BOND_MODE_8023AD) {
1164 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1165 }
1166
1167 if ((bond->params.mode == BOND_MODE_TLB) ||
1168 (bond->params.mode == BOND_MODE_ALB)) {
1169 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1170 }
1171 } else {
1172 if (USES_PRIMARY(bond->params.mode)) {
1173 printk(KERN_INFO DRV_NAME
1174 ": %s: making interface %s the new "
1175 "active one.\n",
1176 bond->dev->name, new_active->dev->name);
1177 }
1178 }
1179 }
1180
1da177e4
LT		 1181 if (USES_PRIMARY(bond->params.mode)) {
1182 bond_mc_swap(bond, new_active, old_active);
1183 }
1184
1185 if ((bond->params.mode == BOND_MODE_TLB) ||
1186 (bond->params.mode == BOND_MODE_ALB)) {
1187 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1188 if (old_active)
1189 bond_set_slave_inactive_flags(old_active);
1190 if (new_active)
1191 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1192 } else {
1193 bond->curr_active_slave = new_active;
1194 }
c3ade5ca
JV		 1195
1196 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1197 if (old_active) {
1198 bond_set_slave_inactive_flags(old_active);
1199 }
1200
1201 if (new_active) {
1202 bond_set_slave_active_flags(new_active);
2ab82852 1203
709f8a45
OG		 1204 if (bond->params.fail_over_mac)
1205 bond_do_fail_over_mac(bond, new_active,
1206 old_active);
3915c1e8 1207
709f8a45 1208 bond->send_grat_arp = bond->params.num_grat_arp;
b59f9f74 1209 bond_send_gratuitous_arp(bond);
01f3109d
OG		 1210
1211 write_unlock_bh(&bond->curr_slave_lock);
1212 read_unlock(&bond->lock);
1213
1214 netdev_bonding_change(bond->dev);
1215
1216 read_lock(&bond->lock);
1217 write_lock_bh(&bond->curr_slave_lock);
7893b249 1218 }
c3ade5ca 1219 }
1da177e4
LT		 1220}
1221
1222/**
1223 * bond_select_active_slave - select a new active slave, if needed
1224 * @bond: our bonding struct
1225 *
1226 * This functions shoud be called when one of the following occurs:
1227 * - The old curr_active_slave has been released or lost its link.
1228 * - The primary_slave has got its link back.
1229 * - A slave has got its link back and there's no old curr_active_slave.
1230 *
3915c1e8 1231 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1232 */
a77b5325 1233void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1234{
1235 struct slave *best_slave;
ff59c456 1236 int rv;
1da177e4
LT		 1237
1238 best_slave = bond_find_best_slave(bond);
1239 if (best_slave != bond->curr_active_slave) {
1240 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1241 rv = bond_set_carrier(bond);
1242 if (!rv)
1243 return;
1244
1245 if (netif_carrier_ok(bond->dev)) {
1246 printk(KERN_INFO DRV_NAME
1247 ": %s: first active interface up!\n",
1248 bond->dev->name);
1249 } else {
1250 printk(KERN_INFO DRV_NAME ": %s: "
1251 "now running without any active interface !\n",
1252 bond->dev->name);
1253 }
1da177e4
LT		 1254 }
1255}
1256
1257/*--------------------------- slave list handling ---------------------------*/
1258
1259/*
1260 * This function attaches the slave to the end of list.
1261 *
1262 * bond->lock held for writing by caller.
1263 */
1264static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1265{
1266 if (bond->first_slave == NULL) { /* attaching the first slave */
1267 new_slave->next = new_slave;
1268 new_slave->prev = new_slave;
1269 bond->first_slave = new_slave;
1270 } else {
1271 new_slave->next = bond->first_slave;
1272 new_slave->prev = bond->first_slave->prev;
1273 new_slave->next->prev = new_slave;
1274 new_slave->prev->next = new_slave;
1275 }
1276
1277 bond->slave_cnt++;
1278}
1279
1280/*
1281 * This function detaches the slave from the list.
1282 * WARNING: no check is made to verify if the slave effectively
1283 * belongs to <bond>.
1284 * Nothing is freed on return, structures are just unchained.
1285 * If any slave pointer in bond was pointing to <slave>,
1286 * it should be changed by the calling function.
1287 *
1288 * bond->lock held for writing by caller.
1289 */
1290static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1291{
1292 if (slave->next) {
1293 slave->next->prev = slave->prev;
1294 }
1295
1296 if (slave->prev) {
1297 slave->prev->next = slave->next;
1298 }
1299
1300 if (bond->first_slave == slave) { /* slave is the first slave */
1301 if (bond->slave_cnt > 1) { /* there are more slave */
1302 bond->first_slave = slave->next;
1303 } else {
1304 bond->first_slave = NULL; /* slave was the last one */
1305 }
1306 }
1307
1308 slave->next = NULL;
1309 slave->prev = NULL;
1310 bond->slave_cnt--;
1311}
1312
1313/*---------------------------------- IOCTL ----------------------------------*/
1314
4ad072c9
AB		 1315static int bond_sethwaddr(struct net_device *bond_dev,
1316 struct net_device *slave_dev)
1da177e4
LT		 1317{
1318 dprintk("bond_dev=%p\n", bond_dev);
1319 dprintk("slave_dev=%p\n", slave_dev);
1320 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1321 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1322 return 0;
1323}
1324
7f353bf2
HX		 1325#define BOND_VLAN_FEATURES \
1326 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1327 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK		 1328
1329/*
8e3babcd 1330 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX		 1331 * feature bits are managed elsewhere, so preserve those feature bits
1332 * on the master device.
8531c5ff
AK		 1333 */
1334static int bond_compute_features(struct bonding *bond)
1335{
8531c5ff
AK		 1336 struct slave *slave;
1337 struct net_device *bond_dev = bond->dev;
7f353bf2 1338 unsigned long features = bond_dev->features;
3158bf7d
MS		 1339 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1340 bond_dev->hard_header_len);
8e3babcd 1341 int i;
8531c5ff 1342
7f353bf2
HX		 1343 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1344 features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1345 NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1346
54ef3137 1347 bond_for_each_slave(bond, slave, i) {
7f353bf2
HX		 1348 features = netdev_compute_features(features,
1349 slave->dev->features);
54ef3137
JV		 1350 if (slave->dev->hard_header_len > max_hard_header_len)
1351 max_hard_header_len = slave->dev->hard_header_len;
1352 }
8531c5ff 1353
7f353bf2 1354 features |= (bond_dev->features & BOND_VLAN_FEATURES);
8531c5ff 1355 bond_dev->features = features;
54ef3137 1356 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1357
1358 return 0;
1359}
1360
872254dd
MS		 1361
1362static void bond_setup_by_slave(struct net_device *bond_dev,
1363 struct net_device *slave_dev)
1364{
d90a162a
MS		 1365 struct bonding *bond = bond_dev->priv;
1366
872254dd 1367 bond_dev->neigh_setup = slave_dev->neigh_setup;
1284cd3a 1368 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1369
1370 bond_dev->type = slave_dev->type;
1371 bond_dev->hard_header_len = slave_dev->hard_header_len;
1372 bond_dev->addr_len = slave_dev->addr_len;
1373
1374 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1375 slave_dev->addr_len);
d90a162a 1376 bond->setup_by_slave = 1;
872254dd
MS		 1377}
1378
1da177e4 1379/* enslave device <slave> to bond device <master> */
a77b5325 1380int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1381{
1382 struct bonding *bond = bond_dev->priv;
1383 struct slave *new_slave = NULL;
1384 struct dev_mc_list *dmi;
1385 struct sockaddr addr;
1386 int link_reporting;
1387 int old_features = bond_dev->features;
1388 int res = 0;
1389
552709d5 1390 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1391 slave_dev->do_ioctl == NULL) {
1da177e4 1392 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1393 ": %s: Warning: no link monitoring support for %s\n",
1394 bond_dev->name, slave_dev->name);
1da177e4
LT		 1395 }
1396
1397 /* bond must be initialized by bond_open() before enslaving */
1398 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1399 printk(KERN_WARNING DRV_NAME
1400 " %s: master_dev is not up in bond_enslave\n",
1401 bond_dev->name);
1da177e4
LT		 1402 }
1403
1404 /* already enslaved */
1405 if (slave_dev->flags & IFF_SLAVE) {
1406 dprintk("Error, Device was already enslaved\n");
1407 return -EBUSY;
1408 }
1409
1410 /* vlan challenged mutual exclusion */
1411 /* no need to lock since we're protected by rtnl_lock */
1412 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1413 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1414 if (!list_empty(&bond->vlan_list)) {
1415 printk(KERN_ERR DRV_NAME
4e0952c7 1416 ": %s: Error: cannot enslave VLAN "
1da177e4 1417 "challenged slave %s on VLAN enabled "
4e0952c7 1418 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1419 bond_dev->name);
1420 return -EPERM;
1421 } else {
1422 printk(KERN_WARNING DRV_NAME
4e0952c7 1423 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1424 "slave %s. Adding VLANs will be blocked as "
1425 "long as %s is part of bond %s\n",
4e0952c7 1426 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1427 bond_dev->name);
1428 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1429 }
1430 } else {
1431 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1432 if (bond->slave_cnt == 0) {
1433 /* First slave, and it is not VLAN challenged,
1434 * so remove the block of adding VLANs over the bond.
1435 */
1436 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1437 }
1438 }
1439
217df670
JV		 1440 /*
1441 * Old ifenslave binaries are no longer supported. These can
1442 * be identified with moderate accurary by the state of the slave:
1443 * the current ifenslave will set the interface down prior to
1444 * enslaving it; the old ifenslave will not.
1445 */
1446 if ((slave_dev->flags & IFF_UP)) {
1447 printk(KERN_ERR DRV_NAME ": %s is up. "
1448 "This may be due to an out of date ifenslave.\n",
1449 slave_dev->name);
1450 res = -EPERM;
1451 goto err_undo_flags;
1452 }
1da177e4 1453
872254dd
MS		 1454 /* set bonding device ether type by slave - bonding netdevices are
1455 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1456 * there is a need to override some of the type dependent attribs/funcs.
1457 *
1458 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1459 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1460 */
1461 if (bond->slave_cnt == 0) {
1462 if (slave_dev->type != ARPHRD_ETHER)
1463 bond_setup_by_slave(bond_dev, slave_dev);
1464 } else if (bond_dev->type != slave_dev->type) {
1465 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1466 "from other slaves (%d), can not enslave it.\n",
1467 slave_dev->name,
1468 slave_dev->type, bond_dev->type);
1469 res = -EINVAL;
1470 goto err_undo_flags;
1471 }
1472
217df670 1473 if (slave_dev->set_mac_address == NULL) {
2ab82852
MS		 1474 if (bond->slave_cnt == 0) {
1475 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1476 ": %s: Warning: The first slave device "
1477 "specified does not support setting the MAC "
3915c1e8 1478 "address. Setting fail_over_mac to active.",
dd957c57 1479 bond_dev->name);
3915c1e8
JV		 1480 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1481 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1482 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1483 ": %s: Error: The slave device specified "
1484 "does not support setting the MAC address, "
3915c1e8 1485 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1486 , bond_dev->name);
1487 res = -EOPNOTSUPP;
1488 goto err_undo_flags;
1489 }
1da177e4
LT		 1490 }
1491
243cb4e5 1492 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1493 if (!new_slave) {
1494 res = -ENOMEM;
1495 goto err_undo_flags;
1496 }
1497
1da177e4
LT		 1498 /* save slave's original flags before calling
1499 * netdev_set_master and dev_open
1500 */
1501 new_slave->original_flags = slave_dev->flags;
1502
217df670
JV		 1503 /*
1504 * Save slave's original ("permanent") mac address for modes
1505 * that need it, and for restoring it upon release, and then
1506 * set it to the master's address
1507 */
1508 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1509
dd957c57 1510 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1511 /*
1512 * Set slave to master's mac address. The application already
1513 * set the master's mac address to that of the first slave
1514 */
1515 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1516 addr.sa_family = slave_dev->type;
1517 res = dev_set_mac_address(slave_dev, &addr);
1518 if (res) {
1519 dprintk("Error %d calling set_mac_address\n", res);
1520 goto err_free;
1521 }
217df670 1522 }
1da177e4 1523
c2edacf8
JV		 1524 res = netdev_set_master(slave_dev, bond_dev);
1525 if (res) {
1526 dprintk("Error %d calling netdev_set_master\n", res);
569f0c4d 1527 goto err_restore_mac;
c2edacf8 1528 }
217df670
JV		 1529 /* open the slave since the application closed it */
1530 res = dev_open(slave_dev);
1531 if (res) {
1532 dprintk("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1533 goto err_unset_master;
1da177e4
LT		 1534 }
1535
1da177e4 1536 new_slave->dev = slave_dev;
0b680e75 1537 slave_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 1538
1539 if ((bond->params.mode == BOND_MODE_TLB) ||
1540 (bond->params.mode == BOND_MODE_ALB)) {
1541 /* bond_alb_init_slave() must be called before all other stages since
1542 * it might fail and we do not want to have to undo everything
1543 */
1544 res = bond_alb_init_slave(bond, new_slave);
1545 if (res) {
569f0c4d 1546 goto err_close;
1da177e4
LT		 1547 }
1548 }
1549
1550 /* If the mode USES_PRIMARY, then the new slave gets the
1551 * master's promisc (and mc) settings only if it becomes the
1552 * curr_active_slave, and that is taken care of later when calling
1553 * bond_change_active()
1554 */
1555 if (!USES_PRIMARY(bond->params.mode)) {
1556 /* set promiscuity level to new slave */
1557 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1558 res = dev_set_promiscuity(slave_dev, 1);
1559 if (res)
1560 goto err_close;
1da177e4
LT		 1561 }
1562
1563 /* set allmulti level to new slave */
1564 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1565 res = dev_set_allmulti(slave_dev, 1);
1566 if (res)
1567 goto err_close;
1da177e4
LT		 1568 }
1569
80ee5ad2 1570 netif_tx_lock_bh(bond_dev);
e308a5d8 1571 netif_addr_lock(bond_dev);
1da177e4
LT		 1572 /* upload master's mc_list to new slave */
1573 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1574 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1575 }
e308a5d8 1576 netif_addr_unlock(bond_dev);
80ee5ad2 1577 netif_tx_unlock_bh(bond_dev);
1da177e4
LT		 1578 }
1579
1580 if (bond->params.mode == BOND_MODE_8023AD) {
1581 /* add lacpdu mc addr to mc list */
1582 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1583
1584 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1585 }
1586
1587 bond_add_vlans_on_slave(bond, slave_dev);
1588
1589 write_lock_bh(&bond->lock);
1590
1591 bond_attach_slave(bond, new_slave);
1592
1593 new_slave->delay = 0;
1594 new_slave->link_failure_count = 0;
1595
8531c5ff
AK		 1596 bond_compute_features(bond);
1597
3915c1e8
JV		 1598 write_unlock_bh(&bond->lock);
1599
1600 read_lock(&bond->lock);
1601
f5b2b966
JV		 1602 new_slave->last_arp_rx = jiffies;
1603
1da177e4
LT		 1604 if (bond->params.miimon && !bond->params.use_carrier) {
1605 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1606
1607 if ((link_reporting == -1) && !bond->params.arp_interval) {
1608 /*
1609 * miimon is set but a bonded network driver
1610 * does not support ETHTOOL/MII and
1611 * arp_interval is not set. Note: if
1612 * use_carrier is enabled, we will never go
1613 * here (because netif_carrier is always
1614 * supported); thus, we don't need to change
1615 * the messages for netif_carrier.
1616 */
1617 printk(KERN_WARNING DRV_NAME
4e0952c7 1618 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1619 "available for interface %s, and "
1620 "arp_interval/arp_ip_target module parameters "
1621 "not specified, thus bonding will not detect "
1622 "link failures! see bonding.txt for details.\n",
4e0952c7 1623 bond_dev->name, slave_dev->name);
1da177e4
LT		 1624 } else if (link_reporting == -1) {
1625 /* unable get link status using mii/ethtool */
1626 printk(KERN_WARNING DRV_NAME
4e0952c7 1627 ": %s: Warning: can't get link status from "
1da177e4
LT		 1628 "interface %s; the network driver associated "
1629 "with this interface does not support MII or "
1630 "ETHTOOL link status reporting, thus miimon "
1631 "has no effect on this interface.\n",
4e0952c7 1632 bond_dev->name, slave_dev->name);
1da177e4
LT		 1633 }
1634 }
1635
1636 /* check for initial state */
1637 if (!bond->params.miimon ||
1638 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1639 if (bond->params.updelay) {
1640 dprintk("Initial state of slave_dev is "
1641 "BOND_LINK_BACK\n");
1642 new_slave->link = BOND_LINK_BACK;
1643 new_slave->delay = bond->params.updelay;
1644 } else {
1645 dprintk("Initial state of slave_dev is "
1646 "BOND_LINK_UP\n");
1647 new_slave->link = BOND_LINK_UP;
1648 }
1649 new_slave->jiffies = jiffies;
1650 } else {
1651 dprintk("Initial state of slave_dev is "
1652 "BOND_LINK_DOWN\n");
1653 new_slave->link = BOND_LINK_DOWN;
1654 }
1655
1656 if (bond_update_speed_duplex(new_slave) &&
1657 (new_slave->link != BOND_LINK_DOWN)) {
1658 printk(KERN_WARNING DRV_NAME
4e0952c7 1659 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1660 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1661 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1662
1663 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1664 printk(KERN_WARNING DRV_NAME
1665 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1666 "support in base driver for proper aggregator "
4e0952c7 1667 "selection.\n", bond_dev->name);
1da177e4
LT		 1668 }
1669 }
1670
1671 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1672 /* if there is a primary slave, remember it */
1673 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1674 bond->primary_slave = new_slave;
1675 }
1676 }
1677
3915c1e8
JV		 1678 write_lock_bh(&bond->curr_slave_lock);
1679
1da177e4
LT		 1680 switch (bond->params.mode) {
1681 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1682 bond_set_slave_inactive_flags(new_slave);
1683 bond_select_active_slave(bond);
1da177e4
LT		 1684 break;
1685 case BOND_MODE_8023AD:
1686 /* in 802.3ad mode, the internal mechanism
1687 * will activate the slaves in the selected
1688 * aggregator
1689 */
1690 bond_set_slave_inactive_flags(new_slave);
1691 /* if this is the first slave */
1692 if (bond->slave_cnt == 1) {
1693 SLAVE_AD_INFO(new_slave).id = 1;
1694 /* Initialize AD with the number of times that the AD timer is called in 1 second
1695 * can be called only after the mac address of the bond is set
1696 */
1697 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1698 bond->params.lacp_fast);
1699 } else {
1700 SLAVE_AD_INFO(new_slave).id =
1701 SLAVE_AD_INFO(new_slave->prev).id + 1;
1702 }
1703
1704 bond_3ad_bind_slave(new_slave);
1705 break;
1706 case BOND_MODE_TLB:
1707 case BOND_MODE_ALB:
1708 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1709 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1710 break;
1711 default:
1712 dprintk("This slave is always active in trunk mode\n");
1713
1714 /* always active in trunk mode */
1715 new_slave->state = BOND_STATE_ACTIVE;
1716
1717 /* In trunking mode there is little meaning to curr_active_slave
1718 * anyway (it holds no special properties of the bond device),
1719 * so we can change it without calling change_active_interface()
1720 */
1721 if (!bond->curr_active_slave) {
1722 bond->curr_active_slave = new_slave;
1723 }
1724 break;
1725 } /* switch(bond_mode) */
1726
3915c1e8
JV		 1727 write_unlock_bh(&bond->curr_slave_lock);
1728
ff59c456
JV		 1729 bond_set_carrier(bond);
1730
3915c1e8 1731 read_unlock(&bond->lock);
1da177e4 1732
b76cdba9
MW		 1733 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1734 if (res)
569f0c4d 1735 goto err_close;
b76cdba9 1736
1da177e4
LT		 1737 printk(KERN_INFO DRV_NAME
1738 ": %s: enslaving %s as a%s interface with a%s link.\n",
1739 bond_dev->name, slave_dev->name,
1740 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1741 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1742
1743 /* enslave is successful */
1744 return 0;
1745
1746/* Undo stages on error */
1da177e4
LT		 1747err_close:
1748 dev_close(slave_dev);
1749
569f0c4d
JV		 1750err_unset_master:
1751 netdev_set_master(slave_dev, NULL);
1752
1da177e4 1753err_restore_mac:
dd957c57 1754 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1755 /* XXX TODO - fom follow mode needs to change master's
1756 * MAC if this slave's MAC is in use by the bond, or at
1757 * least print a warning.
1758 */
2ab82852
MS		 1759 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1760 addr.sa_family = slave_dev->type;
1761 dev_set_mac_address(slave_dev, &addr);
1762 }
1da177e4
LT		 1763
1764err_free:
1765 kfree(new_slave);
1766
1767err_undo_flags:
1768 bond_dev->features = old_features;
8531c5ff 1769
1da177e4
LT		 1770 return res;
1771}
1772
1773/*
1774 * Try to release the slave device <slave> from the bond device <master>
1775 * It is legal to access curr_active_slave without a lock because all the function
1776 * is write-locked.
1777 *
1778 * The rules for slave state should be:
1779 * for Active/Backup:
1780 * Active stays on all backups go down
1781 * for Bonded connections:
1782 * The first up interface should be left on and all others downed.
1783 */
a77b5325 1784int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1785{
1786 struct bonding *bond = bond_dev->priv;
1787 struct slave *slave, *oldcurrent;
1788 struct sockaddr addr;
1789 int mac_addr_differ;
0795af57 1790 DECLARE_MAC_BUF(mac);
1da177e4
LT		 1791
1792 /* slave is not a slave or master is not master of this slave */
1793 if (!(slave_dev->flags & IFF_SLAVE) ||
1794 (slave_dev->master != bond_dev)) {
1795 printk(KERN_ERR DRV_NAME
4e0952c7 1796 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1797 bond_dev->name, slave_dev->name);
1798 return -EINVAL;
1799 }
1800
1801 write_lock_bh(&bond->lock);
1802
1803 slave = bond_get_slave_by_dev(bond, slave_dev);
1804 if (!slave) {
1805 /* not a slave of this bond */
1806 printk(KERN_INFO DRV_NAME
1807 ": %s: %s not enslaved\n",
1808 bond_dev->name, slave_dev->name);
f5e2a7b2 1809 write_unlock_bh(&bond->lock);
1da177e4
LT		 1810 return -EINVAL;
1811 }
1812
3915c1e8
JV		 1813 if (!bond->params.fail_over_mac) {
1814 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1815 ETH_ALEN);
1816 if (!mac_addr_differ && (bond->slave_cnt > 1))
1817 printk(KERN_WARNING DRV_NAME
1818 ": %s: Warning: the permanent HWaddr of %s - "
1819 "%s - is still in use by %s. "
1820 "Set the HWaddr of %s to a different address "
1821 "to avoid conflicts.\n",
1822 bond_dev->name, slave_dev->name,
1823 print_mac(mac, slave->perm_hwaddr),
1824 bond_dev->name, slave_dev->name);
1da177e4
LT		 1825 }
1826
1827 /* Inform AD package of unbinding of slave. */
1828 if (bond->params.mode == BOND_MODE_8023AD) {
1829 /* must be called before the slave is
1830 * detached from the list
1831 */
1832 bond_3ad_unbind_slave(slave);
1833 }
1834
1835 printk(KERN_INFO DRV_NAME
1836 ": %s: releasing %s interface %s\n",
1837 bond_dev->name,
1838 (slave->state == BOND_STATE_ACTIVE)
1839 ? "active" : "backup",
1840 slave_dev->name);
1841
1842 oldcurrent = bond->curr_active_slave;
1843
1844 bond->current_arp_slave = NULL;
1845
1846 /* release the slave from its bond */
1847 bond_detach_slave(bond, slave);
1848
8531c5ff
AK		 1849 bond_compute_features(bond);
1850
1da177e4
LT		 1851 if (bond->primary_slave == slave) {
1852 bond->primary_slave = NULL;
1853 }
1854
1855 if (oldcurrent == slave) {
1856 bond_change_active_slave(bond, NULL);
1857 }
1858
1859 if ((bond->params.mode == BOND_MODE_TLB) ||
1860 (bond->params.mode == BOND_MODE_ALB)) {
1861 /* Must be called only after the slave has been
1862 * detached from the list and the curr_active_slave
1863 * has been cleared (if our_slave == old_current),
1864 * but before a new active slave is selected.
1865 */
2543331d 1866 write_unlock_bh(&bond->lock);
1da177e4 1867 bond_alb_deinit_slave(bond, slave);
2543331d 1868 write_lock_bh(&bond->lock);
1da177e4
LT		 1869 }
1870
059fe7a5
JV		 1871 if (oldcurrent == slave) {
1872 /*
1873 * Note that we hold RTNL over this sequence, so there
1874 * is no concern that another slave add/remove event
1875 * will interfere.
1876 */
1877 write_unlock_bh(&bond->lock);
1878 read_lock(&bond->lock);
1879 write_lock_bh(&bond->curr_slave_lock);
1880
1da177e4
LT		 1881 bond_select_active_slave(bond);
1882
059fe7a5
JV		 1883 write_unlock_bh(&bond->curr_slave_lock);
1884 read_unlock(&bond->lock);
1885 write_lock_bh(&bond->lock);
1886 }
1887
1da177e4 1888 if (bond->slave_cnt == 0) {
ff59c456
JV		 1889 bond_set_carrier(bond);
1890
1da177e4
LT		 1891 /* if the last slave was removed, zero the mac address
1892 * of the master so it will be set by the application
1893 * to the mac address of the first slave
1894 */
1895 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1896
1897 if (list_empty(&bond->vlan_list)) {
1898 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1899 } else {
1900 printk(KERN_WARNING DRV_NAME
4e0952c7 1901 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1902 "still has VLANs.\n",
4e0952c7 1903 bond_dev->name, bond_dev->name);
1da177e4 1904 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1905 ": %s: When re-adding slaves, make sure the bond's "
1906 "HW address matches its VLANs'.\n",
1907 bond_dev->name);
1da177e4
LT		 1908 }
1909 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1910 !bond_has_challenged_slaves(bond)) {
1911 printk(KERN_INFO DRV_NAME
4e0952c7 1912 ": %s: last VLAN challenged slave %s "
1da177e4 1913 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1914 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1915 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1916 }
1917
1918 write_unlock_bh(&bond->lock);
1919
b76cdba9
MW		 1920 /* must do this from outside any spinlocks */
1921 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1922
1da177e4
LT		 1923 bond_del_vlans_from_slave(bond, slave_dev);
1924
1925 /* If the mode USES_PRIMARY, then we should only remove its
1926 * promisc and mc settings if it was the curr_active_slave, but that was
1927 * already taken care of above when we detached the slave
1928 */
1929 if (!USES_PRIMARY(bond->params.mode)) {
1930 /* unset promiscuity level from slave */
1931 if (bond_dev->flags & IFF_PROMISC) {
1932 dev_set_promiscuity(slave_dev, -1);
1933 }
1934
1935 /* unset allmulti level from slave */
1936 if (bond_dev->flags & IFF_ALLMULTI) {
1937 dev_set_allmulti(slave_dev, -1);
1938 }
1939
1940 /* flush master's mc_list from slave */
80ee5ad2 1941 netif_tx_lock_bh(bond_dev);
e308a5d8 1942 netif_addr_lock(bond_dev);
1da177e4 1943 bond_mc_list_flush(bond_dev, slave_dev);
e308a5d8 1944 netif_addr_unlock(bond_dev);
80ee5ad2 1945 netif_tx_unlock_bh(bond_dev);
1da177e4
LT		 1946 }
1947
1948 netdev_set_master(slave_dev, NULL);
1949
1950 /* close slave before restoring its mac address */
1951 dev_close(slave_dev);
1952
3915c1e8 1953 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1954 /* restore original ("permanent") mac address */
1955 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1956 addr.sa_family = slave_dev->type;
1957 dev_set_mac_address(slave_dev, &addr);
1958 }
1da177e4 1959
8f903c70 1960 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1961 IFF_SLAVE_INACTIVE | IFF_BONDING |
1962 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1963
1964 kfree(slave);
1965
1966 return 0; /* deletion OK */
1967}
1968
d90a162a
MS		 1969/*
1970* Destroy a bonding device.
1971* Must be under rtnl_lock when this function is called.
1972*/
1973void bond_destroy(struct bonding *bond)
1974{
1975 bond_deinit(bond->dev);
1976 bond_destroy_sysfs_entry(bond);
8cbdeec6 1977 unregister_netdevice(bond->dev);
d90a162a
MS		 1978}
1979
1980/*
1981* First release a slave and than destroy the bond if no more slaves iare left.
1982* Must be under rtnl_lock when this function is called.
1983*/
1984int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1985{
1986 struct bonding *bond = bond_dev->priv;
1987 int ret;
1988
1989 ret = bond_release(bond_dev, slave_dev);
1990 if ((ret == 0) && (bond->slave_cnt == 0)) {
1991 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1992 bond_dev->name, bond_dev->name);
1993 bond_destroy(bond);
1994 }
1995 return ret;
1996}
1997
1da177e4
LT		 1998/*
1999 * This function releases all slaves.
2000 */
2001static int bond_release_all(struct net_device *bond_dev)
2002{
2003 struct bonding *bond = bond_dev->priv;
2004 struct slave *slave;
2005 struct net_device *slave_dev;
2006 struct sockaddr addr;
2007
2008 write_lock_bh(&bond->lock);
2009
ff59c456
JV		 2010 netif_carrier_off(bond_dev);
2011
1da177e4
LT		 2012 if (bond->slave_cnt == 0) {
2013 goto out;
2014 }
2015
2016 bond->current_arp_slave = NULL;
2017 bond->primary_slave = NULL;
2018 bond_change_active_slave(bond, NULL);
2019
2020 while ((slave = bond->first_slave) != NULL) {
2021 /* Inform AD package of unbinding of slave
2022 * before slave is detached from the list.
2023 */
2024 if (bond->params.mode == BOND_MODE_8023AD) {
2025 bond_3ad_unbind_slave(slave);
2026 }
2027
2028 slave_dev = slave->dev;
2029 bond_detach_slave(bond, slave);
2030
2543331d
JV		 2031 /* now that the slave is detached, unlock and perform
2032 * all the undo steps that should not be called from
2033 * within a lock.
2034 */
2035 write_unlock_bh(&bond->lock);
2036
1da177e4
LT		 2037 if ((bond->params.mode == BOND_MODE_TLB) ||
2038 (bond->params.mode == BOND_MODE_ALB)) {
2039 /* must be called only after the slave
2040 * has been detached from the list
2041 */
2042 bond_alb_deinit_slave(bond, slave);
2043 }
2044
8531c5ff
AK		 2045 bond_compute_features(bond);
2046
b76cdba9 2047 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2048 bond_del_vlans_from_slave(bond, slave_dev);
2049
2050 /* If the mode USES_PRIMARY, then we should only remove its
2051 * promisc and mc settings if it was the curr_active_slave, but that was
2052 * already taken care of above when we detached the slave
2053 */
2054 if (!USES_PRIMARY(bond->params.mode)) {
2055 /* unset promiscuity level from slave */
2056 if (bond_dev->flags & IFF_PROMISC) {
2057 dev_set_promiscuity(slave_dev, -1);
2058 }
2059
2060 /* unset allmulti level from slave */
2061 if (bond_dev->flags & IFF_ALLMULTI) {
2062 dev_set_allmulti(slave_dev, -1);
2063 }
2064
2065 /* flush master's mc_list from slave */
80ee5ad2 2066 netif_tx_lock_bh(bond_dev);
e308a5d8 2067 netif_addr_lock(bond_dev);
1da177e4 2068 bond_mc_list_flush(bond_dev, slave_dev);
e308a5d8 2069 netif_addr_unlock(bond_dev);
80ee5ad2 2070 netif_tx_unlock_bh(bond_dev);
1da177e4
LT		 2071 }
2072
2073 netdev_set_master(slave_dev, NULL);
2074
2075 /* close slave before restoring its mac address */
2076 dev_close(slave_dev);
2077
dd957c57 2078 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2079 /* restore original ("permanent") mac address*/
2080 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2081 addr.sa_family = slave_dev->type;
2082 dev_set_mac_address(slave_dev, &addr);
2083 }
1da177e4 2084
8f903c70
JV		 2085 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2086 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2087
2088 kfree(slave);
2089
2090 /* re-acquire the lock before getting the next slave */
2091 write_lock_bh(&bond->lock);
2092 }
2093
2094 /* zero the mac address of the master so it will be
2095 * set by the application to the mac address of the
2096 * first slave
2097 */
2098 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2099
2100 if (list_empty(&bond->vlan_list)) {
2101 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2102 } else {
2103 printk(KERN_WARNING DRV_NAME
4e0952c7 2104 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2105 "still has VLANs.\n",
4e0952c7 2106 bond_dev->name, bond_dev->name);
1da177e4 2107 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2108 ": %s: When re-adding slaves, make sure the bond's "
2109 "HW address matches its VLANs'.\n",
2110 bond_dev->name);
1da177e4
LT		 2111 }
2112
2113 printk(KERN_INFO DRV_NAME
2114 ": %s: released all slaves\n",
2115 bond_dev->name);
2116
2117out:
2118 write_unlock_bh(&bond->lock);
2119
2120 return 0;
2121}
2122
2123/*
2124 * This function changes the active slave to slave <slave_dev>.
2125 * It returns -EINVAL in the following cases.
2126 * - <slave_dev> is not found in the list.
2127 * - There is not active slave now.
2128 * - <slave_dev> is already active.
2129 * - The link state of <slave_dev> is not BOND_LINK_UP.
2130 * - <slave_dev> is not running.
2131 * In these cases, this fuction does nothing.
2132 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2133 */
2134static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2135{
2136 struct bonding *bond = bond_dev->priv;
2137 struct slave *old_active = NULL;
2138 struct slave *new_active = NULL;
2139 int res = 0;
2140
2141 if (!USES_PRIMARY(bond->params.mode)) {
2142 return -EINVAL;
2143 }
2144
2145 /* Verify that master_dev is indeed the master of slave_dev */
2146 if (!(slave_dev->flags & IFF_SLAVE) ||
2147 (slave_dev->master != bond_dev)) {
2148 return -EINVAL;
2149 }
2150
059fe7a5 2151 read_lock(&bond->lock);
1da177e4 2152
059fe7a5 2153 read_lock(&bond->curr_slave_lock);
1da177e4 2154 old_active = bond->curr_active_slave;
059fe7a5
JV		 2155 read_unlock(&bond->curr_slave_lock);
2156
1da177e4
LT		 2157 new_active = bond_get_slave_by_dev(bond, slave_dev);
2158
2159 /*
2160 * Changing to the current active: do nothing; return success.
2161 */
2162 if (new_active && (new_active == old_active)) {
059fe7a5 2163 read_unlock(&bond->lock);
1da177e4
LT		 2164 return 0;
2165 }
2166
2167 if ((new_active) &&
2168 (old_active) &&
2169 (new_active->link == BOND_LINK_UP) &&
2170 IS_UP(new_active->dev)) {
059fe7a5 2171 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2172 bond_change_active_slave(bond, new_active);
059fe7a5 2173 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2174 } else {
2175 res = -EINVAL;
2176 }
2177
059fe7a5 2178 read_unlock(&bond->lock);
1da177e4
LT		 2179
2180 return res;
2181}
2182
1da177e4
LT		 2183static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2184{
2185 struct bonding *bond = bond_dev->priv;
2186
2187 info->bond_mode = bond->params.mode;
2188 info->miimon = bond->params.miimon;
2189
6603a6f2 2190 read_lock(&bond->lock);
1da177e4 2191 info->num_slaves = bond->slave_cnt;
6603a6f2 2192 read_unlock(&bond->lock);
1da177e4
LT		 2193
2194 return 0;
2195}
2196
2197static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2198{
2199 struct bonding *bond = bond_dev->priv;
2200 struct slave *slave;
2201 int i, found = 0;
2202
2203 if (info->slave_id < 0) {
2204 return -ENODEV;
2205 }
2206
6603a6f2 2207 read_lock(&bond->lock);
1da177e4
LT		 2208
2209 bond_for_each_slave(bond, slave, i) {
2210 if (i == (int)info->slave_id) {
2211 found = 1;
2212 break;
2213 }
2214 }
2215
6603a6f2 2216 read_unlock(&bond->lock);
1da177e4
LT		 2217
2218 if (found) {
2219 strcpy(info->slave_name, slave->dev->name);
2220 info->link = slave->link;
2221 info->state = slave->state;
2222 info->link_failure_count = slave->link_failure_count;
2223 } else {
2224 return -ENODEV;
2225 }
2226
2227 return 0;
2228}
2229
2230/*-------------------------------- Monitoring -------------------------------*/
2231
0b0eef66
JV		 2232/*
2233 * if !have_locks, return nonzero if a failover is necessary. if
2234 * have_locks, do whatever failover activities are needed.
2235 *
2236 * This is to separate the inspection and failover steps for locking
2237 * purposes; failover requires rtnl, but acquiring it for every
2238 * inspection is undesirable, so a wrapper first does inspection, and
2239 * the acquires the necessary locks and calls again to perform
2240 * failover if needed. Since all locks are dropped, a complete
2241 * restart is needed between calls.
2242 */
2243static int __bond_mii_monitor(struct bonding *bond, int have_locks)
1da177e4 2244{
1da177e4
LT		 2245 struct slave *slave, *oldcurrent;
2246 int do_failover = 0;
1da177e4
LT		 2247 int i;
2248
0b0eef66 2249 if (bond->slave_cnt == 0)
1da177e4 2250 goto out;
1da177e4
LT		 2251
2252 /* we will try to read the link status of each of our slaves, and
2253 * set their IFF_RUNNING flag appropriately. For each slave not
2254 * supporting MII status, we won't do anything so that a user-space
2255 * program could monitor the link itself if needed.
2256 */
2257
2258 read_lock(&bond->curr_slave_lock);
2259 oldcurrent = bond->curr_active_slave;
2260 read_unlock(&bond->curr_slave_lock);
2261
2262 bond_for_each_slave(bond, slave, i) {
2263 struct net_device *slave_dev = slave->dev;
2264 int link_state;
2265 u16 old_speed = slave->speed;
2266 u8 old_duplex = slave->duplex;
2267
2268 link_state = bond_check_dev_link(bond, slave_dev, 0);
2269
2270 switch (slave->link) {
2271 case BOND_LINK_UP: /* the link was up */
2272 if (link_state == BMSR_LSTATUS) {
059fe7a5
JV		 2273 if (!oldcurrent) {
2274 if (!have_locks)
2275 return 1;
2276 do_failover = 1;
2277 }
1da177e4
LT		 2278 break;
2279 } else { /* link going down */
2280 slave->link = BOND_LINK_FAIL;
2281 slave->delay = bond->params.downdelay;
2282
2283 if (slave->link_failure_count < UINT_MAX) {
2284 slave->link_failure_count++;
2285 }
2286
2287 if (bond->params.downdelay) {
2288 printk(KERN_INFO DRV_NAME
2289 ": %s: link status down for %s "
2290 "interface %s, disabling it in "
2291 "%d ms.\n",
1b76b316 2292 bond->dev->name,
1da177e4
LT		 2293 IS_UP(slave_dev)
2294 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2295 ? ((slave == oldcurrent)
2296 ? "active " : "backup ")
2297 : "")
2298 : "idle ",
2299 slave_dev->name,
2300 bond->params.downdelay * bond->params.miimon);
2301 }
2302 }
2303 /* no break ! fall through the BOND_LINK_FAIL test to
2304 ensure proper action to be taken
2305 */
2306 case BOND_LINK_FAIL: /* the link has just gone down */
2307 if (link_state != BMSR_LSTATUS) {
2308 /* link stays down */
2309 if (slave->delay <= 0) {
0b0eef66
JV		 2310 if (!have_locks)
2311 return 1;
2312
1da177e4
LT		 2313 /* link down for too long time */
2314 slave->link = BOND_LINK_DOWN;
2315
2316 /* in active/backup mode, we must
2317 * completely disable this interface
2318 */
2319 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2320 (bond->params.mode == BOND_MODE_8023AD)) {
2321 bond_set_slave_inactive_flags(slave);
2322 }
2323
2324 printk(KERN_INFO DRV_NAME
2325 ": %s: link status definitely "
2326 "down for interface %s, "
2327 "disabling it\n",
1b76b316 2328 bond->dev->name,
1da177e4
LT		 2329 slave_dev->name);
2330
2331 /* notify ad that the link status has changed */
2332 if (bond->params.mode == BOND_MODE_8023AD) {
2333 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2334 }
2335
2336 if ((bond->params.mode == BOND_MODE_TLB) ||
2337 (bond->params.mode == BOND_MODE_ALB)) {
2338 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2339 }
2340
2341 if (slave == oldcurrent) {
2342 do_failover = 1;
2343 }
2344 } else {
2345 slave->delay--;
2346 }
2347 } else {
2348 /* link up again */
2349 slave->link = BOND_LINK_UP;
2350 slave->jiffies = jiffies;
2351 printk(KERN_INFO DRV_NAME
2352 ": %s: link status up again after %d "
2353 "ms for interface %s.\n",
1b76b316 2354 bond->dev->name,
1da177e4
LT		 2355 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2356 slave_dev->name);
2357 }
2358 break;
2359 case BOND_LINK_DOWN: /* the link was down */
2360 if (link_state != BMSR_LSTATUS) {
2361 /* the link stays down, nothing more to do */
2362 break;
2363 } else { /* link going up */
2364 slave->link = BOND_LINK_BACK;
2365 slave->delay = bond->params.updelay;
2366
2367 if (bond->params.updelay) {
2368 /* if updelay == 0, no need to
2369 advertise about a 0 ms delay */
2370 printk(KERN_INFO DRV_NAME
2371 ": %s: link status up for "
2372 "interface %s, enabling it "
2373 "in %d ms.\n",
1b76b316 2374 bond->dev->name,
1da177e4
LT		 2375 slave_dev->name,
2376 bond->params.updelay * bond->params.miimon);
2377 }
2378 }
2379 /* no break ! fall through the BOND_LINK_BACK state in
2380 case there's something to do.
2381 */
2382 case BOND_LINK_BACK: /* the link has just come back */
2383 if (link_state != BMSR_LSTATUS) {
2384 /* link down again */
2385 slave->link = BOND_LINK_DOWN;
2386
2387 printk(KERN_INFO DRV_NAME
2388 ": %s: link status down again after %d "
2389 "ms for interface %s.\n",
1b76b316 2390 bond->dev->name,
1da177e4
LT		 2391 (bond->params.updelay - slave->delay) * bond->params.miimon,
2392 slave_dev->name);
2393 } else {
2394 /* link stays up */
2395 if (slave->delay == 0) {
0b0eef66
JV		 2396 if (!have_locks)
2397 return 1;
2398
1da177e4
LT		 2399 /* now the link has been up for long time enough */
2400 slave->link = BOND_LINK_UP;
2401 slave->jiffies = jiffies;
2402
2403 if (bond->params.mode == BOND_MODE_8023AD) {
2404 /* prevent it from being the active one */
2405 slave->state = BOND_STATE_BACKUP;
2406 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2407 /* make it immediately active */
2408 slave->state = BOND_STATE_ACTIVE;
2409 } else if (slave != bond->primary_slave) {
2410 /* prevent it from being the active one */
2411 slave->state = BOND_STATE_BACKUP;
2412 }
2413
2414 printk(KERN_INFO DRV_NAME
2415 ": %s: link status definitely "
2416 "up for interface %s.\n",
1b76b316 2417 bond->dev->name,
1da177e4
LT		 2418 slave_dev->name);
2419
2420 /* notify ad that the link status has changed */
2421 if (bond->params.mode == BOND_MODE_8023AD) {
2422 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2423 }
2424
2425 if ((bond->params.mode == BOND_MODE_TLB) ||
2426 (bond->params.mode == BOND_MODE_ALB)) {
2427 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2428 }
2429
2430 if ((!oldcurrent) ||
2431 (slave == bond->primary_slave)) {
2432 do_failover = 1;
2433 }
2434 } else {
2435 slave->delay--;
2436 }
2437 }
2438 break;
2439 default:
2440 /* Should not happen */
4e0952c7
MW		 2441 printk(KERN_ERR DRV_NAME
2442 ": %s: Error: %s Illegal value (link=%d)\n",
1b76b316 2443 bond->dev->name,
4e0952c7
MW		 2444 slave->dev->name,
2445 slave->link);
1da177e4
LT		 2446 goto out;
2447 } /* end of switch (slave->link) */
2448
2449 bond_update_speed_duplex(slave);
2450
2451 if (bond->params.mode == BOND_MODE_8023AD) {
2452 if (old_speed != slave->speed) {
2453 bond_3ad_adapter_speed_changed(slave);
2454 }
2455
2456 if (old_duplex != slave->duplex) {
2457 bond_3ad_adapter_duplex_changed(slave);
2458 }
2459 }
2460
2461 } /* end of for */
2462
2463 if (do_failover) {
059fe7a5
JV		 2464 ASSERT_RTNL();
2465
2466 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2467
2468 bond_select_active_slave(bond);
2469
059fe7a5
JV		 2470 write_unlock_bh(&bond->curr_slave_lock);
2471
ff59c456
JV		 2472 } else
2473 bond_set_carrier(bond);
1da177e4 2474
1da177e4 2475out:
0b0eef66 2476 return 0;
1da177e4
LT		 2477}
2478
0b0eef66
JV		 2479/*
2480 * bond_mii_monitor
2481 *
2482 * Really a wrapper that splits the mii monitor into two phases: an
2483 * inspection, then (if inspection indicates something needs to be
2484 * done) an acquisition of appropriate locks followed by another pass
2485 * to implement whatever link state changes are indicated.
2486 */
2487void bond_mii_monitor(struct work_struct *work)
2488{
2489 struct bonding *bond = container_of(work, struct bonding,
2490 mii_work.work);
2491 unsigned long delay;
2492
2493 read_lock(&bond->lock);
2494 if (bond->kill_timers) {
2495 read_unlock(&bond->lock);
2496 return;
2497 }
b59f9f74
JV		 2498
2499 if (bond->send_grat_arp) {
2500 read_lock(&bond->curr_slave_lock);
2501 bond_send_gratuitous_arp(bond);
2502 read_unlock(&bond->curr_slave_lock);
2503 }
2504
0b0eef66
JV		 2505 if (__bond_mii_monitor(bond, 0)) {
2506 read_unlock(&bond->lock);
2507 rtnl_lock();
2508 read_lock(&bond->lock);
2509 __bond_mii_monitor(bond, 1);
5655662d
JV		 2510 read_unlock(&bond->lock);
2511 rtnl_unlock(); /* might sleep, hold no other locks */
2512 read_lock(&bond->lock);
0b0eef66
JV		 2513 }
2514
5ce0da8f 2515 delay = msecs_to_jiffies(bond->params.miimon);
0b0eef66
JV		 2516 read_unlock(&bond->lock);
2517 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2518}
c3ade5ca 2519
d3bb52b0 2520static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2521{
2522 struct in_device *idev;
2523 struct in_ifaddr *ifa;
a144ea4b 2524 __be32 addr = 0;
c3ade5ca
JV		 2525
2526 if (!dev)
2527 return 0;
2528
2529 rcu_read_lock();
e5ed6399 2530 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2531 if (!idev)
2532 goto out;
2533
2534 ifa = idev->ifa_list;
2535 if (!ifa)
2536 goto out;
2537
2538 addr = ifa->ifa_local;
2539out:
2540 rcu_read_unlock();
2541 return addr;
2542}
2543
d3bb52b0 2544static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2545{
0883beca 2546 struct vlan_entry *vlan;
f5b2b966
JV		 2547
2548 if (ip == bond->master_ip)
2549 return 1;
2550
0883beca 2551 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2552 if (ip == vlan->vlan_ip)
2553 return 1;
2554 }
2555
2556 return 0;
2557}
2558
c3ade5ca
JV		 2559/*
2560 * We go to the (large) trouble of VLAN tagging ARP frames because
2561 * switches in VLAN mode (especially if ports are configured as
2562 * "native" to a VLAN) might not pass non-tagged frames.
2563 */
d3bb52b0 2564static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2565{
2566 struct sk_buff *skb;
2567
2568 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2569 slave_dev->name, dest_ip, src_ip, vlan_id);
2570
2571 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2572 NULL, slave_dev->dev_addr, NULL);
2573
2574 if (!skb) {
2575 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2576 return;
2577 }
2578 if (vlan_id) {
2579 skb = vlan_put_tag(skb, vlan_id);
2580 if (!skb) {
2581 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2582 return;
2583 }
2584 }
2585 arp_xmit(skb);
2586}
2587
2588
1da177e4
LT		 2589static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2590{
c3ade5ca 2591 int i, vlan_id, rv;
d3bb52b0 2592 __be32 *targets = bond->params.arp_targets;
0883beca 2593 struct vlan_entry *vlan;
c3ade5ca
JV		 2594 struct net_device *vlan_dev;
2595 struct flowi fl;
2596 struct rtable *rt;
1da177e4 2597
6b780567
MW		 2598 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2599 if (!targets[i])
2600 continue;
c3ade5ca
JV		 2601 dprintk("basa: target %x\n", targets[i]);
2602 if (list_empty(&bond->vlan_list)) {
2603 dprintk("basa: empty vlan: arp_send\n");
2604 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2605 bond->master_ip, 0);
2606 continue;
2607 }
2608
2609 /*
2610 * If VLANs are configured, we do a route lookup to
2611 * determine which VLAN interface would be used, so we
2612 * can tag the ARP with the proper VLAN tag.
2613 */
2614 memset(&fl, 0, sizeof(fl));
2615 fl.fl4_dst = targets[i];
2616 fl.fl4_tos = RTO_ONLINK;
2617
f206351a 2618 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2619 if (rv) {
2620 if (net_ratelimit()) {
2621 printk(KERN_WARNING DRV_NAME
2622 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2623 bond->dev->name, NIPQUAD(fl.fl4_dst));
2624 }
2625 continue;
2626 }
2627
2628 /*
2629 * This target is not on a VLAN
2630 */
2631 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2632 ip_rt_put(rt);
c3ade5ca
JV		 2633 dprintk("basa: rtdev == bond->dev: arp_send\n");
2634 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2635 bond->master_ip, 0);
2636 continue;
2637 }
2638
2639 vlan_id = 0;
0883beca 2640 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2641 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2642 if (vlan_dev == rt->u.dst.dev) {
2643 vlan_id = vlan->vlan_id;
2644 dprintk("basa: vlan match on %s %d\n",
2645 vlan_dev->name, vlan_id);
2646 break;
2647 }
2648 }
2649
2650 if (vlan_id) {
ed4b9f80 2651 ip_rt_put(rt);
c3ade5ca
JV		 2652 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2653 vlan->vlan_ip, vlan_id);
2654 continue;
2655 }
2656
2657 if (net_ratelimit()) {
2658 printk(KERN_WARNING DRV_NAME
2659 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2660 bond->dev->name, NIPQUAD(fl.fl4_dst),
2661 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2662 }
ed4b9f80 2663 ip_rt_put(rt);
c3ade5ca
JV		 2664 }
2665}
2666
2667/*
2668 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2669 * for each VLAN above us.
b59f9f74
JV		 2670 *
2671 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2672 */
2673static void bond_send_gratuitous_arp(struct bonding *bond)
2674{
2675 struct slave *slave = bond->curr_active_slave;
2676 struct vlan_entry *vlan;
2677 struct net_device *vlan_dev;
2678
2679 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2680 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2681
2682 if (!slave || !bond->send_grat_arp ||
2683 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2684 return;
2685
b59f9f74
JV		 2686 bond->send_grat_arp--;
2687
c3ade5ca
JV		 2688 if (bond->master_ip) {
2689 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2690 bond->master_ip, 0);
c3ade5ca
JV		 2691 }
2692
2693 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2694 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2695 if (vlan->vlan_ip) {
2696 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2697 vlan->vlan_ip, vlan->vlan_id);
2698 }
1da177e4
LT		 2699 }
2700}
2701
d3bb52b0 2702static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2703{
2704 int i;
d3bb52b0 2705 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2706
2707 targets = bond->params.arp_targets;
2708 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2709 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2710 "%u.%u.%u.%u bhti(tip) %d\n",
2711 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2712 bond_has_this_ip(bond, tip));
2713 if (sip == targets[i]) {
2714 if (bond_has_this_ip(bond, tip))
2715 slave->last_arp_rx = jiffies;
2716 return;
2717 }
2718 }
2719}
2720
2721static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2722{
2723 struct arphdr *arp;
2724 struct slave *slave;
2725 struct bonding *bond;
2726 unsigned char *arp_ptr;
d3bb52b0 2727 __be32 sip, tip;
f5b2b966 2728
c346dca1 2729 if (dev_net(dev) != &init_net)
e730c155
EB		 2730 goto out;
2731
f5b2b966
JV		 2732 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2733 goto out;
2734
2735 bond = dev->priv;
2736 read_lock(&bond->lock);
2737
2738 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2739 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2740 orig_dev ? orig_dev->name : "NULL");
2741
2742 slave = bond_get_slave_by_dev(bond, orig_dev);
2743 if (!slave || !slave_do_arp_validate(bond, slave))
2744 goto out_unlock;
2745
988b7050 2746 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2747 goto out_unlock;
2748
d0a92be0 2749 arp = arp_hdr(skb);
f5b2b966
JV		 2750 if (arp->ar_hln != dev->addr_len ||
2751 skb->pkt_type == PACKET_OTHERHOST ||
2752 skb->pkt_type == PACKET_LOOPBACK ||
2753 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2754 arp->ar_pro != htons(ETH_P_IP) ||
2755 arp->ar_pln != 4)
2756 goto out_unlock;
2757
2758 arp_ptr = (unsigned char *)(arp + 1);
2759 arp_ptr += dev->addr_len;
2760 memcpy(&sip, arp_ptr, 4);
2761 arp_ptr += 4 + dev->addr_len;
2762 memcpy(&tip, arp_ptr, 4);
2763
2764 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2765 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2766 slave->state, bond->params.arp_validate,
2767 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2768
2769 /*
2770 * Backup slaves won't see the ARP reply, but do come through
2771 * here for each ARP probe (so we swap the sip/tip to validate
2772 * the probe). In a "redundant switch, common router" type of
2773 * configuration, the ARP probe will (hopefully) travel from
2774 * the active, through one switch, the router, then the other
2775 * switch before reaching the backup.
2776 */
2777 if (slave->state == BOND_STATE_ACTIVE)
2778 bond_validate_arp(bond, slave, sip, tip);
2779 else
2780 bond_validate_arp(bond, slave, tip, sip);
2781
2782out_unlock:
2783 read_unlock(&bond->lock);
2784out:
2785 dev_kfree_skb(skb);
2786 return NET_RX_SUCCESS;
2787}
2788
1da177e4
LT		 2789/*
2790 * this function is called regularly to monitor each slave's link
2791 * ensuring that traffic is being sent and received when arp monitoring
2792 * is used in load-balancing mode. if the adapter has been dormant, then an
2793 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2794 * arp monitoring in active backup mode.
2795 */
1b76b316 2796void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2797{
1b76b316
JV		 2798 struct bonding *bond = container_of(work, struct bonding,
2799 arp_work.work);
1da177e4
LT		 2800 struct slave *slave, *oldcurrent;
2801 int do_failover = 0;
2802 int delta_in_ticks;
2803 int i;
2804
2805 read_lock(&bond->lock);
2806
5ce0da8f 2807 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2808
2809 if (bond->kill_timers) {
2810 goto out;
2811 }
2812
2813 if (bond->slave_cnt == 0) {
2814 goto re_arm;
2815 }
2816
2817 read_lock(&bond->curr_slave_lock);
2818 oldcurrent = bond->curr_active_slave;
2819 read_unlock(&bond->curr_slave_lock);
2820
2821 /* see if any of the previous devices are up now (i.e. they have
2822 * xmt and rcv traffic). the curr_active_slave does not come into
2823 * the picture unless it is null. also, slave->jiffies is not needed
2824 * here because we send an arp on each slave and give a slave as
2825 * long as it needs to get the tx/rx within the delta.
2826 * TODO: what about up/down delay in arp mode? it wasn't here before
2827 * so it can wait
2828 */
2829 bond_for_each_slave(bond, slave, i) {
2830 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2831 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2832 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2833
2834 slave->link = BOND_LINK_UP;
2835 slave->state = BOND_STATE_ACTIVE;
2836
2837 /* primary_slave has no meaning in round-robin
2838 * mode. the window of a slave being up and
2839 * curr_active_slave being null after enslaving
2840 * is closed.
2841 */
2842 if (!oldcurrent) {
2843 printk(KERN_INFO DRV_NAME
2844 ": %s: link status definitely "
2845 "up for interface %s, ",
1b76b316 2846 bond->dev->name,
1da177e4
LT		 2847 slave->dev->name);
2848 do_failover = 1;
2849 } else {
2850 printk(KERN_INFO DRV_NAME
2851 ": %s: interface %s is now up\n",
1b76b316 2852 bond->dev->name,
1da177e4
LT		 2853 slave->dev->name);
2854 }
2855 }
2856 } else {
2857 /* slave->link == BOND_LINK_UP */
2858
2859 /* not all switches will respond to an arp request
2860 * when the source ip is 0, so don't take the link down
2861 * if we don't know our ip yet
2862 */
b63bb739 2863 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2864 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2865
2866 slave->link = BOND_LINK_DOWN;
2867 slave->state = BOND_STATE_BACKUP;
2868
2869 if (slave->link_failure_count < UINT_MAX) {
2870 slave->link_failure_count++;
2871 }
2872
2873 printk(KERN_INFO DRV_NAME
2874 ": %s: interface %s is now down.\n",
1b76b316 2875 bond->dev->name,
1da177e4
LT		 2876 slave->dev->name);
2877
2878 if (slave == oldcurrent) {
2879 do_failover = 1;
2880 }
2881 }
2882 }
2883
2884 /* note: if switch is in round-robin mode, all links
2885 * must tx arp to ensure all links rx an arp - otherwise
2886 * links may oscillate or not come up at all; if switch is
2887 * in something like xor mode, there is nothing we can
2888 * do - all replies will be rx'ed on same link causing slaves
2889 * to be unstable during low/no traffic periods
2890 */
2891 if (IS_UP(slave->dev)) {
2892 bond_arp_send_all(bond, slave);
2893 }
2894 }
2895
2896 if (do_failover) {
059fe7a5 2897 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2898
2899 bond_select_active_slave(bond);
2900
059fe7a5 2901 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2902 }
2903
2904re_arm:
1b76b316
JV		 2905 if (bond->params.arp_interval)
2906 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2907out:
2908 read_unlock(&bond->lock);
2909}
2910
2911/*
b2220cad
JV		 2912 * Called to inspect slaves for active-backup mode ARP monitor link state
2913 * changes. Sets new_link in slaves to specify what action should take
2914 * place for the slave. Returns 0 if no changes are found, >0 if changes
2915 * to link states must be committed.
2916 *
2917 * Called with bond->lock held for read.
1da177e4 2918 */
b2220cad 2919static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2920{
1da177e4 2921 struct slave *slave;
b2220cad 2922 int i, commit = 0;
1da177e4 2923
b2220cad
JV		 2924 bond_for_each_slave(bond, slave, i) {
2925 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2926
b2220cad
JV		 2927 if (slave->link != BOND_LINK_UP) {
2928 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2929 delta_in_ticks)) {
2930 slave->new_link = BOND_LINK_UP;
2931 commit++;
2932 }
1da177e4 2933
b2220cad
JV		 2934 continue;
2935 }
1da177e4 2936
b2220cad
JV		 2937 /*
2938 * Give slaves 2*delta after being enslaved or made
2939 * active. This avoids bouncing, as the last receive
2940 * times need a full ARP monitor cycle to be updated.
2941 */
2942 if (!time_after_eq(jiffies, slave->jiffies +
2943 2 * delta_in_ticks))
2944 continue;
2945
2946 /*
2947 * Backup slave is down if:
2948 * - No current_arp_slave AND
2949 * - more than 3*delta since last receive AND
2950 * - the bond has an IP address
2951 *
2952 * Note: a non-null current_arp_slave indicates
2953 * the curr_active_slave went down and we are
2954 * searching for a new one; under this condition
2955 * we only take the curr_active_slave down - this
2956 * gives each slave a chance to tx/rx traffic
2957 * before being taken out
2958 */
2959 if (slave->state == BOND_STATE_BACKUP &&
2960 !bond->current_arp_slave &&
2961 time_after(jiffies, slave_last_rx(bond, slave) +
2962 3 * delta_in_ticks)) {
2963 slave->new_link = BOND_LINK_DOWN;
2964 commit++;
2965 }
2966
2967 /*
2968 * Active slave is down if:
2969 * - more than 2*delta since transmitting OR
2970 * - (more than 2*delta since receive AND
2971 * the bond has an IP address)
2972 */
2973 if ((slave->state == BOND_STATE_ACTIVE) &&
2974 (time_after_eq(jiffies, slave->dev->trans_start +
2975 2 * delta_in_ticks) ||
2976 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2977 + 2 * delta_in_ticks)))) {
2978 slave->new_link = BOND_LINK_DOWN;
2979 commit++;
2980 }
1da177e4
LT		 2981 }
2982
b2220cad
JV		 2983 read_lock(&bond->curr_slave_lock);
2984
2985 /*
2986 * Trigger a commit if the primary option setting has changed.
1da177e4 2987 */
b2220cad
JV		 2988 if (bond->primary_slave &&
2989 (bond->primary_slave != bond->curr_active_slave) &&
2990 (bond->primary_slave->link == BOND_LINK_UP))
2991 commit++;
1da177e4 2992
b2220cad 2993 read_unlock(&bond->curr_slave_lock);
1da177e4 2994
b2220cad
JV		 2995 return commit;
2996}
1da177e4 2997
b2220cad
JV		 2998/*
2999 * Called to commit link state changes noted by inspection step of
3000 * active-backup mode ARP monitor.
3001 *
3002 * Called with RTNL and bond->lock for read.
3003 */
3004static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3005{
3006 struct slave *slave;
3007 int i;
1da177e4 3008
b2220cad
JV		 3009 bond_for_each_slave(bond, slave, i) {
3010 switch (slave->new_link) {
3011 case BOND_LINK_NOCHANGE:
3012 continue;
ff59c456 3013
b2220cad
JV		 3014 case BOND_LINK_UP:
3015 write_lock_bh(&bond->curr_slave_lock);
1da177e4 3016
b2220cad
JV		 3017 if (!bond->curr_active_slave &&
3018 time_before_eq(jiffies, slave->dev->trans_start +
3019 delta_in_ticks)) {
3020 slave->link = BOND_LINK_UP;
3021 bond_change_active_slave(bond, slave);
3022 bond->current_arp_slave = NULL;
3023
3024 printk(KERN_INFO DRV_NAME
3025 ": %s: %s is up and now the "
3026 "active interface\n",
3027 bond->dev->name, slave->dev->name);
3028
3029 } else if (bond->curr_active_slave != slave) {
3030 /* this slave has just come up but we
3031 * already have a current slave; this can
3032 * also happen if bond_enslave adds a new
3033 * slave that is up while we are searching
3034 * for a new slave
1da177e4 3035 */
b2220cad
JV		 3036 slave->link = BOND_LINK_UP;
3037 bond_set_slave_inactive_flags(slave);
3038 bond->current_arp_slave = NULL;
3039
3040 printk(KERN_INFO DRV_NAME
3041 ": %s: backup interface %s is now up\n",
3042 bond->dev->name, slave->dev->name);
3043 }
1da177e4 3044
b2220cad 3045 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3046
b2220cad 3047 break;
1da177e4 3048
b2220cad
JV		 3049 case BOND_LINK_DOWN:
3050 if (slave->link_failure_count < UINT_MAX)
3051 slave->link_failure_count++;
3052
3053 slave->link = BOND_LINK_DOWN;
3054
3055 if (slave == bond->curr_active_slave) {
3056 printk(KERN_INFO DRV_NAME
3057 ": %s: link status down for active "
3058 "interface %s, disabling it\n",
3059 bond->dev->name, slave->dev->name);
1da177e4
LT		 3060
3061 bond_set_slave_inactive_flags(slave);
3062
b2220cad
JV		 3063 write_lock_bh(&bond->curr_slave_lock);
3064
3065 bond_select_active_slave(bond);
3066 if (bond->curr_active_slave)
3067 bond->curr_active_slave->jiffies =
3068 jiffies;
3069
3070 write_unlock_bh(&bond->curr_slave_lock);
3071
3072 bond->current_arp_slave = NULL;
3073
3074 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3075 printk(KERN_INFO DRV_NAME
3076 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3077 bond->dev->name, slave->dev->name);
3078
3079 bond_set_slave_inactive_flags(slave);
1da177e4 3080 }
b2220cad
JV		 3081 break;
3082
3083 default:
3084 printk(KERN_ERR DRV_NAME
3085 ": %s: impossible: new_link %d on slave %s\n",
3086 bond->dev->name, slave->new_link,
3087 slave->dev->name);
1da177e4
LT		 3088 }
3089 }
3090
b2220cad
JV		 3091 /*
3092 * No race with changes to primary via sysfs, as we hold rtnl.
3093 */
3094 if (bond->primary_slave &&
3095 (bond->primary_slave != bond->curr_active_slave) &&
3096 (bond->primary_slave->link == BOND_LINK_UP)) {
3097 write_lock_bh(&bond->curr_slave_lock);
3098 bond_change_active_slave(bond, bond->primary_slave);
3099 write_unlock_bh(&bond->curr_slave_lock);
3100 }
1da177e4 3101
b2220cad
JV		 3102 bond_set_carrier(bond);
3103}
1da177e4 3104
b2220cad
JV		 3105/*
3106 * Send ARP probes for active-backup mode ARP monitor.
3107 *
3108 * Called with bond->lock held for read.
3109 */
3110static void bond_ab_arp_probe(struct bonding *bond)
3111{
3112 struct slave *slave;
3113 int i;
1da177e4 3114
b2220cad 3115 read_lock(&bond->curr_slave_lock);
1da177e4 3116
b2220cad
JV		 3117 if (bond->current_arp_slave && bond->curr_active_slave)
3118 printk("PROBE: c_arp %s && cas %s BAD\n",
3119 bond->current_arp_slave->dev->name,
3120 bond->curr_active_slave->dev->name);
1da177e4 3121
b2220cad
JV		 3122 if (bond->curr_active_slave) {
3123 bond_arp_send_all(bond, bond->curr_active_slave);
3124 read_unlock(&bond->curr_slave_lock);
3125 return;
3126 }
1da177e4 3127
b2220cad 3128 read_unlock(&bond->curr_slave_lock);
059fe7a5 3129
b2220cad
JV		 3130 /* if we don't have a curr_active_slave, search for the next available
3131 * backup slave from the current_arp_slave and make it the candidate
3132 * for becoming the curr_active_slave
3133 */
1da177e4 3134
b2220cad
JV		 3135 if (!bond->current_arp_slave) {
3136 bond->current_arp_slave = bond->first_slave;
3137 if (!bond->current_arp_slave)
3138 return;
3139 }
1da177e4 3140
b2220cad 3141 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3142
b2220cad
JV		 3143 /* search for next candidate */
3144 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3145 if (IS_UP(slave->dev)) {
3146 slave->link = BOND_LINK_BACK;
3147 bond_set_slave_active_flags(slave);
3148 bond_arp_send_all(bond, slave);
1da177e4 3149 slave->jiffies = jiffies;
b2220cad
JV		 3150 bond->current_arp_slave = slave;
3151 break;
1da177e4
LT		 3152 }
3153
b2220cad
JV		 3154 /* if the link state is up at this point, we
3155 * mark it down - this can happen if we have
3156 * simultaneous link failures and
3157 * reselect_active_interface doesn't make this
3158 * one the current slave so it is still marked
3159 * up when it is actually down
1da177e4 3160 */
b2220cad
JV		 3161 if (slave->link == BOND_LINK_UP) {
3162 slave->link = BOND_LINK_DOWN;
3163 if (slave->link_failure_count < UINT_MAX)
3164 slave->link_failure_count++;
1da177e4 3165
b2220cad
JV		 3166 bond_set_slave_inactive_flags(slave);
3167
3168 printk(KERN_INFO DRV_NAME
3169 ": %s: backup interface %s is now down.\n",
3170 bond->dev->name, slave->dev->name);
1da177e4 3171 }
b2220cad
JV		 3172 }
3173}
1da177e4 3174
b2220cad
JV		 3175void bond_activebackup_arp_mon(struct work_struct *work)
3176{
3177 struct bonding *bond = container_of(work, struct bonding,
3178 arp_work.work);
3179 int delta_in_ticks;
1da177e4 3180
b2220cad 3181 read_lock(&bond->lock);
1da177e4 3182
b2220cad
JV		 3183 if (bond->kill_timers)
3184 goto out;
1da177e4 3185
b2220cad 3186 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3187
b2220cad
JV		 3188 if (bond->slave_cnt == 0)
3189 goto re_arm;
3190
b59f9f74
JV		 3191 if (bond->send_grat_arp) {
3192 read_lock(&bond->curr_slave_lock);
3193 bond_send_gratuitous_arp(bond);
3194 read_unlock(&bond->curr_slave_lock);
3195 }
3196
b2220cad
JV		 3197 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3198 read_unlock(&bond->lock);
3199 rtnl_lock();
3200 read_lock(&bond->lock);
3201
3202 bond_ab_arp_commit(bond, delta_in_ticks);
3203
3204 read_unlock(&bond->lock);
3205 rtnl_unlock();
3206 read_lock(&bond->lock);
1da177e4
LT		 3207 }
3208
b2220cad
JV		 3209 bond_ab_arp_probe(bond);
3210
1da177e4
LT		 3211re_arm:
3212 if (bond->params.arp_interval) {
1b76b316 3213 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3214 }
3215out:
3216 read_unlock(&bond->lock);
3217}
3218
3219/*------------------------------ proc/seq_file-------------------------------*/
3220
3221#ifdef CONFIG_PROC_FS
3222
1da177e4
LT		 3223static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3224{
3225 struct bonding *bond = seq->private;
3226 loff_t off = 0;
3227 struct slave *slave;
3228 int i;
3229
3230 /* make sure the bond won't be taken away */
3231 read_lock(&dev_base_lock);
6603a6f2 3232 read_lock(&bond->lock);
1da177e4
LT		 3233
3234 if (*pos == 0) {
3235 return SEQ_START_TOKEN;
3236 }
3237
3238 bond_for_each_slave(bond, slave, i) {
3239 if (++off == *pos) {
3240 return slave;
3241 }
3242 }
3243
3244 return NULL;
3245}
3246
3247static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3248{
3249 struct bonding *bond = seq->private;
3250 struct slave *slave = v;
3251
3252 ++*pos;
3253 if (v == SEQ_START_TOKEN) {
3254 return bond->first_slave;
3255 }
3256
3257 slave = slave->next;
3258
3259 return (slave == bond->first_slave) ? NULL : slave;
3260}
3261
3262static void bond_info_seq_stop(struct seq_file *seq, void *v)
3263{
3264 struct bonding *bond = seq->private;
3265
6603a6f2 3266 read_unlock(&bond->lock);
1da177e4
LT		 3267 read_unlock(&dev_base_lock);
3268}
3269
3270static void bond_info_show_master(struct seq_file *seq)
3271{
3272 struct bonding *bond = seq->private;
3273 struct slave *curr;
4756b02f
MW		 3274 int i;
3275 u32 target;
1da177e4
LT		 3276
3277 read_lock(&bond->curr_slave_lock);
3278 curr = bond->curr_active_slave;
3279 read_unlock(&bond->curr_slave_lock);
3280
dd957c57 3281 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3282 bond_mode_name(bond->params.mode));
3283
dd957c57
JV		 3284 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3285 bond->params.fail_over_mac)
3915c1e8
JV		 3286 seq_printf(seq, " (fail_over_mac %s)",
3287 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3288
3289 seq_printf(seq, "\n");
3290
c61b75ad
MW		 3291 if (bond->params.mode == BOND_MODE_XOR ||
3292 bond->params.mode == BOND_MODE_8023AD) {
3293 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3294 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3295 bond->params.xmit_policy);
3296 }
3297
1da177e4
LT		 3298 if (USES_PRIMARY(bond->params.mode)) {
3299 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3300 (bond->primary_slave) ?
3301 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3302
3303 seq_printf(seq, "Currently Active Slave: %s\n",
3304 (curr) ? curr->dev->name : "None");
3305 }
3306
ff59c456
JV		 3307 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3308 "up" : "down");
1da177e4
LT		 3309 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3310 seq_printf(seq, "Up Delay (ms): %d\n",
3311 bond->params.updelay * bond->params.miimon);
3312 seq_printf(seq, "Down Delay (ms): %d\n",
3313 bond->params.downdelay * bond->params.miimon);
3314
4756b02f
MW		 3315
3316 /* ARP information */
3317 if(bond->params.arp_interval > 0) {
3318 int printed=0;
3319 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3320 bond->params.arp_interval);
3321
3322 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3323
3324 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3325 if (!bond->params.arp_targets[i])
3326 continue;
3327 if (printed)
3328 seq_printf(seq, ",");
3329 target = ntohl(bond->params.arp_targets[i]);
3330 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3331 printed = 1;
3332 }
3333 seq_printf(seq, "\n");
3334 }
3335
1da177e4
LT		 3336 if (bond->params.mode == BOND_MODE_8023AD) {
3337 struct ad_info ad_info;
0795af57 3338 DECLARE_MAC_BUF(mac);
1da177e4
LT		 3339
3340 seq_puts(seq, "\n802.3ad info\n");
3341 seq_printf(seq, "LACP rate: %s\n",
3342 (bond->params.lacp_fast) ? "fast" : "slow");
3343
3344 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3345 seq_printf(seq, "bond %s has no active aggregator\n",
3346 bond->dev->name);
3347 } else {
3348 seq_printf(seq, "Active Aggregator Info:\n");
3349
3350 seq_printf(seq, "\tAggregator ID: %d\n",
3351 ad_info.aggregator_id);
3352 seq_printf(seq, "\tNumber of ports: %d\n",
3353 ad_info.ports);
3354 seq_printf(seq, "\tActor Key: %d\n",
3355 ad_info.actor_key);
3356 seq_printf(seq, "\tPartner Key: %d\n",
3357 ad_info.partner_key);
0795af57
JP		 3358 seq_printf(seq, "\tPartner Mac Address: %s\n",
3359 print_mac(mac, ad_info.partner_system));
1da177e4
LT		 3360 }
3361 }
3362}
3363
3364static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3365{
3366 struct bonding *bond = seq->private;
0795af57 3367 DECLARE_MAC_BUF(mac);
1da177e4
LT		 3368
3369 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3370 seq_printf(seq, "MII Status: %s\n",
3371 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3372 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3373 slave->link_failure_count);
3374
217df670 3375 seq_printf(seq,
0795af57
JP		 3376 "Permanent HW addr: %s\n",
3377 print_mac(mac, slave->perm_hwaddr));
1da177e4
LT		 3378
3379 if (bond->params.mode == BOND_MODE_8023AD) {
3380 const struct aggregator *agg
3381 = SLAVE_AD_INFO(slave).port.aggregator;
3382
3383 if (agg) {
3384 seq_printf(seq, "Aggregator ID: %d\n",
3385 agg->aggregator_identifier);
3386 } else {
3387 seq_puts(seq, "Aggregator ID: N/A\n");
3388 }
3389 }
3390}
3391
3392static int bond_info_seq_show(struct seq_file *seq, void *v)
3393{
3394 if (v == SEQ_START_TOKEN) {
3395 seq_printf(seq, "%s\n", version);
3396 bond_info_show_master(seq);
3397 } else {
3398 bond_info_show_slave(seq, v);
3399 }
3400
3401 return 0;
3402}
3403
3404static struct seq_operations bond_info_seq_ops = {
3405 .start = bond_info_seq_start,
3406 .next = bond_info_seq_next,
3407 .stop = bond_info_seq_stop,
3408 .show = bond_info_seq_show,
3409};
3410
3411static int bond_info_open(struct inode *inode, struct file *file)
3412{
3413 struct seq_file *seq;
3414 struct proc_dir_entry *proc;
3415 int res;
3416
3417 res = seq_open(file, &bond_info_seq_ops);
3418 if (!res) {
3419 /* recover the pointer buried in proc_dir_entry data */
3420 seq = file->private_data;
3421 proc = PDE(inode);
3422 seq->private = proc->data;
3423 }
3424
3425 return res;
3426}
3427
d54b1fdb 3428static const struct file_operations bond_info_fops = {
1da177e4
LT		 3429 .owner = THIS_MODULE,
3430 .open = bond_info_open,
3431 .read = seq_read,
3432 .llseek = seq_lseek,
3433 .release = seq_release,
3434};
3435
3436static int bond_create_proc_entry(struct bonding *bond)
3437{
3438 struct net_device *bond_dev = bond->dev;
3439
3440 if (bond_proc_dir) {
a95609cb
DL		 3441 bond->proc_entry = proc_create_data(bond_dev->name,
3442 S_IRUGO, bond_proc_dir,
3443 &bond_info_fops, bond);
1da177e4
LT		 3444 if (bond->proc_entry == NULL) {
3445 printk(KERN_WARNING DRV_NAME
3446 ": Warning: Cannot create /proc/net/%s/%s\n",
3447 DRV_NAME, bond_dev->name);
3448 } else {
1da177e4
LT		 3449 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3450 }
3451 }
3452
3453 return 0;
3454}
3455
3456static void bond_remove_proc_entry(struct bonding *bond)
3457{
3458 if (bond_proc_dir && bond->proc_entry) {
3459 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3460 memset(bond->proc_file_name, 0, IFNAMSIZ);
3461 bond->proc_entry = NULL;
3462 }
3463}
3464
3465/* Create the bonding directory under /proc/net, if doesn't exist yet.
3466 * Caller must hold rtnl_lock.
3467 */
3468static void bond_create_proc_dir(void)
3469{
3470 int len = strlen(DRV_NAME);
3471
457c4cbc 3472 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3473 bond_proc_dir = bond_proc_dir->next) {
3474 if ((bond_proc_dir->namelen == len) &&
3475 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3476 break;
3477 }
3478 }
3479
3480 if (!bond_proc_dir) {
457c4cbc 3481 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3482 if (bond_proc_dir) {
3483 bond_proc_dir->owner = THIS_MODULE;
3484 } else {
3485 printk(KERN_WARNING DRV_NAME
3486 ": Warning: cannot create /proc/net/%s\n",
3487 DRV_NAME);
3488 }
3489 }
3490}
3491
3492/* Destroy the bonding directory under /proc/net, if empty.
3493 * Caller must hold rtnl_lock.
3494 */
3495static void bond_destroy_proc_dir(void)
3496{
3497 struct proc_dir_entry *de;
3498
3499 if (!bond_proc_dir) {
3500 return;
3501 }
3502
3503 /* verify that the /proc dir is empty */
3504 for (de = bond_proc_dir->subdir; de; de = de->next) {
3505 /* ignore . and .. */
3506 if (*(de->name) != '.') {
3507 break;
3508 }
3509 }
3510
3511 if (de) {
3512 if (bond_proc_dir->owner == THIS_MODULE) {
3513 bond_proc_dir->owner = NULL;
3514 }
3515 } else {
457c4cbc 3516 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3517 bond_proc_dir = NULL;
3518 }
3519}
3520#endif /* CONFIG_PROC_FS */
3521
3522/*-------------------------- netdev event handling --------------------------*/
3523
3524/*
3525 * Change device name
3526 */
3527static int bond_event_changename(struct bonding *bond)
3528{
3529#ifdef CONFIG_PROC_FS
3530 bond_remove_proc_entry(bond);
3531 bond_create_proc_entry(bond);
3532#endif
b76cdba9
MW		 3533 down_write(&(bonding_rwsem));
3534 bond_destroy_sysfs_entry(bond);
3535 bond_create_sysfs_entry(bond);
3536 up_write(&(bonding_rwsem));
1da177e4
LT		 3537 return NOTIFY_DONE;
3538}
3539
3540static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3541{
3542 struct bonding *event_bond = bond_dev->priv;
3543
3544 switch (event) {
3545 case NETDEV_CHANGENAME:
3546 return bond_event_changename(event_bond);
3547 case NETDEV_UNREGISTER:
3b96c858 3548 bond_release_all(event_bond->dev);
1da177e4
LT		 3549 break;
3550 default:
3551 break;
3552 }
3553
3554 return NOTIFY_DONE;
3555}
3556
3557static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3558{
3559 struct net_device *bond_dev = slave_dev->master;
8531c5ff 3560 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3561
3562 switch (event) {
3563 case NETDEV_UNREGISTER:
3564 if (bond_dev) {
1284cd3a
JV		 3565 if (bond->setup_by_slave)
3566 bond_release_and_destroy(bond_dev, slave_dev);
3567 else
3568 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3569 }
3570 break;
3571 case NETDEV_CHANGE:
3572 /*
3573 * TODO: is this what we get if somebody
3574 * sets up a hierarchical bond, then rmmod's
3575 * one of the slave bonding devices?
3576 */
3577 break;
3578 case NETDEV_DOWN:
3579 /*
3580 * ... Or is it this?
3581 */
3582 break;
3583 case NETDEV_CHANGEMTU:
3584 /*
3585 * TODO: Should slaves be allowed to
3586 * independently alter their MTU? For
3587 * an active-backup bond, slaves need
3588 * not be the same type of device, so
3589 * MTUs may vary. For other modes,
3590 * slaves arguably should have the
3591 * same MTUs. To do this, we'd need to
3592 * take over the slave's change_mtu
3593 * function for the duration of their
3594 * servitude.
3595 */
3596 break;
3597 case NETDEV_CHANGENAME:
3598 /*
3599 * TODO: handle changing the primary's name
3600 */
3601 break;
8531c5ff
AK		 3602 case NETDEV_FEAT_CHANGE:
3603 bond_compute_features(bond);
3604 break;
1da177e4
LT		 3605 default:
3606 break;
3607 }
3608
3609 return NOTIFY_DONE;
3610}
3611
3612/*
3613 * bond_netdev_event: handle netdev notifier chain events.
3614 *
3615 * This function receives events for the netdev chain. The caller (an
e041c683 3616 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3617 * locks for us to safely manipulate the slave devices (RTNL lock,
3618 * dev_probe_lock).
3619 */
3620static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3621{
3622 struct net_device *event_dev = (struct net_device *)ptr;
3623
c346dca1 3624 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3625 return NOTIFY_DONE;
3626
1da177e4
LT		 3627 dprintk("event_dev: %s, event: %lx\n",
3628 (event_dev ? event_dev->name : "None"),
3629 event);
3630
0b680e75
JV		 3631 if (!(event_dev->priv_flags & IFF_BONDING))
3632 return NOTIFY_DONE;
3633
1da177e4
LT		 3634 if (event_dev->flags & IFF_MASTER) {
3635 dprintk("IFF_MASTER\n");
3636 return bond_master_netdev_event(event, event_dev);
3637 }
3638
3639 if (event_dev->flags & IFF_SLAVE) {
3640 dprintk("IFF_SLAVE\n");
3641 return bond_slave_netdev_event(event, event_dev);
3642 }
3643
3644 return NOTIFY_DONE;
3645}
3646
c3ade5ca
JV		 3647/*
3648 * bond_inetaddr_event: handle inetaddr notifier chain events.
3649 *
3650 * We keep track of device IPs primarily to use as source addresses in
3651 * ARP monitor probes (rather than spewing out broadcasts all the time).
3652 *
3653 * We track one IP for the main device (if it has one), plus one per VLAN.
3654 */
3655static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3656{
3657 struct in_ifaddr *ifa = ptr;
3658 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3659 struct bonding *bond;
3660 struct vlan_entry *vlan;
c3ade5ca 3661
c346dca1 3662 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3663 return NOTIFY_DONE;
3664
0883beca 3665 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3666 if (bond->dev == event_dev) {
3667 switch (event) {
3668 case NETDEV_UP:
3669 bond->master_ip = ifa->ifa_local;
3670 return NOTIFY_OK;
3671 case NETDEV_DOWN:
3672 bond->master_ip = bond_glean_dev_ip(bond->dev);
3673 return NOTIFY_OK;
3674 default:
3675 return NOTIFY_DONE;
3676 }
3677 }
3678
0883beca 3679 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3680 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3681 if (vlan_dev == event_dev) {
3682 switch (event) {
3683 case NETDEV_UP:
3684 vlan->vlan_ip = ifa->ifa_local;
3685 return NOTIFY_OK;
3686 case NETDEV_DOWN:
3687 vlan->vlan_ip =
3688 bond_glean_dev_ip(vlan_dev);
3689 return NOTIFY_OK;
3690 default:
3691 return NOTIFY_DONE;
3692 }
3693 }
3694 }
3695 }
3696 return NOTIFY_DONE;
3697}
3698
1da177e4
LT		 3699static struct notifier_block bond_netdev_notifier = {
3700 .notifier_call = bond_netdev_event,
3701};
3702
c3ade5ca
JV		 3703static struct notifier_block bond_inetaddr_notifier = {
3704 .notifier_call = bond_inetaddr_event,
3705};
3706
1da177e4
LT		 3707/*-------------------------- Packet type handling ---------------------------*/
3708
3709/* register to receive lacpdus on a bond */
3710static void bond_register_lacpdu(struct bonding *bond)
3711{
3712 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3713
3714 /* initialize packet type */
3715 pk_type->type = PKT_TYPE_LACPDU;
3716 pk_type->dev = bond->dev;
3717 pk_type->func = bond_3ad_lacpdu_recv;
3718
3719 dev_add_pack(pk_type);
3720}
3721
3722/* unregister to receive lacpdus on a bond */
3723static void bond_unregister_lacpdu(struct bonding *bond)
3724{
3725 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3726}
3727
f5b2b966
JV		 3728void bond_register_arp(struct bonding *bond)
3729{
3730 struct packet_type *pt = &bond->arp_mon_pt;
3731
c4f283b1
JV		 3732 if (pt->type)
3733 return;
3734
f5b2b966 3735 pt->type = htons(ETH_P_ARP);
e245cb71 3736 pt->dev = bond->dev;
f5b2b966
JV		 3737 pt->func = bond_arp_rcv;
3738 dev_add_pack(pt);
3739}
3740
3741void bond_unregister_arp(struct bonding *bond)
3742{
c4f283b1
JV		 3743 struct packet_type *pt = &bond->arp_mon_pt;
3744
3745 dev_remove_pack(pt);
3746 pt->type = 0;
f5b2b966
JV		 3747}
3748
169a3e66
JV		 3749/*---------------------------- Hashing Policies -----------------------------*/
3750
6f6652be
JV		 3751/*
3752 * Hash for the output device based upon layer 2 and layer 3 data. If
3753 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3754 */
3755static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3756 struct net_device *bond_dev, int count)
3757{
3758 struct ethhdr *data = (struct ethhdr *)skb->data;
3759 struct iphdr *iph = ip_hdr(skb);
3760
3761 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3762 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3763 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3764 }
3765
3766 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3767}
3768
169a3e66 3769/*
59c51591 3770 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3771 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3772 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3773 */
3774static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3775 struct net_device *bond_dev, int count)
3776{
3777 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3778 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3779 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3780 int layer4_xor = 0;
3781
3782 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3783 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3784 (iph->protocol == IPPROTO_TCP ||
3785 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3786 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3787 }
3788 return (layer4_xor ^
3789 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3790
3791 }
3792
3793 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3794}
3795
3796/*
3797 * Hash for the output device based upon layer 2 data
3798 */
3799static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3800 struct net_device *bond_dev, int count)
3801{
3802 struct ethhdr *data = (struct ethhdr *)skb->data;
3803
3804 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3805}
3806
1da177e4
LT		 3807/*-------------------------- Device entry points ----------------------------*/
3808
3809static int bond_open(struct net_device *bond_dev)
3810{
3811 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3812
3813 bond->kill_timers = 0;
3814
3815 if ((bond->params.mode == BOND_MODE_TLB) ||
3816 (bond->params.mode == BOND_MODE_ALB)) {
1da177e4
LT		 3817 /* bond_alb_initialize must be called before the timer
3818 * is started.
3819 */
3820 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3821 /* something went wrong - fail the open operation */
3822 return -1;
3823 }
3824
1b76b316
JV		 3825 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3826 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3827 }
3828
3829 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3830 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3831 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3832 }
3833
3834 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3835 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3836 INIT_DELAYED_WORK(&bond->arp_work,
3837 bond_activebackup_arp_mon);
3838 else
3839 INIT_DELAYED_WORK(&bond->arp_work,
3840 bond_loadbalance_arp_mon);
3841
3842 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3843 if (bond->params.arp_validate)
3844 bond_register_arp(bond);
1da177e4
LT		 3845 }
3846
3847 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3848 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3849 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3850 /* register to receive LACPDUs */
3851 bond_register_lacpdu(bond);
3852 }
3853
3854 return 0;
3855}
3856
3857static int bond_close(struct net_device *bond_dev)
3858{
3859 struct bonding *bond = bond_dev->priv;
3860
3861 if (bond->params.mode == BOND_MODE_8023AD) {
3862 /* Unregister the receive of LACPDUs */
3863 bond_unregister_lacpdu(bond);
3864 }
3865
f5b2b966
JV		 3866 if (bond->params.arp_validate)
3867 bond_unregister_arp(bond);
3868
1da177e4
LT		 3869 write_lock_bh(&bond->lock);
3870
b59f9f74 3871 bond->send_grat_arp = 0;
1da177e4
LT		 3872
3873 /* signal timers not to re-arm */
3874 bond->kill_timers = 1;
3875
3876 write_unlock_bh(&bond->lock);
3877
1da177e4 3878 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3879 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3880 }
3881
3882 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3883 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3884 }
3885
3886 switch (bond->params.mode) {
3887 case BOND_MODE_8023AD:
1b76b316 3888 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3889 break;
3890 case BOND_MODE_TLB:
3891 case BOND_MODE_ALB:
1b76b316 3892 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3893 break;
3894 default:
3895 break;
3896 }
3897
1da177e4
LT		 3898
3899 if ((bond->params.mode == BOND_MODE_TLB) ||
3900 (bond->params.mode == BOND_MODE_ALB)) {
3901 /* Must be called only after all
3902 * slaves have been released
3903 */
3904 bond_alb_deinitialize(bond);
3905 }
3906
3907 return 0;
3908}
3909
3910static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3911{
3912 struct bonding *bond = bond_dev->priv;
3913 struct net_device_stats *stats = &(bond->stats), *sstats;
2439f9eb 3914 struct net_device_stats local_stats;
1da177e4
LT		 3915 struct slave *slave;
3916 int i;
3917
2439f9eb 3918 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3919
3920 read_lock_bh(&bond->lock);
3921
3922 bond_for_each_slave(bond, slave, i) {
c45d286e 3923 sstats = slave->dev->get_stats(slave->dev);
2439f9eb
AG		 3924 local_stats.rx_packets += sstats->rx_packets;
3925 local_stats.rx_bytes += sstats->rx_bytes;
3926 local_stats.rx_errors += sstats->rx_errors;
3927 local_stats.rx_dropped += sstats->rx_dropped;
3928
3929 local_stats.tx_packets += sstats->tx_packets;
3930 local_stats.tx_bytes += sstats->tx_bytes;
3931 local_stats.tx_errors += sstats->tx_errors;
3932 local_stats.tx_dropped += sstats->tx_dropped;
3933
3934 local_stats.multicast += sstats->multicast;
3935 local_stats.collisions += sstats->collisions;
3936
3937 local_stats.rx_length_errors += sstats->rx_length_errors;
3938 local_stats.rx_over_errors += sstats->rx_over_errors;
3939 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3940 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3941 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3942 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3943
3944 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3945 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3946 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3947 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3948 local_stats.tx_window_errors += sstats->tx_window_errors;
3949 }
3950
3951 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3952
3953 read_unlock_bh(&bond->lock);
3954
3955 return stats;
3956}
3957
3958static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3959{
3960 struct net_device *slave_dev = NULL;
3961 struct ifbond k_binfo;
3962 struct ifbond __user *u_binfo = NULL;
3963 struct ifslave k_sinfo;
3964 struct ifslave __user *u_sinfo = NULL;
3965 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3966 int res = 0;
3967
3968 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3969 bond_dev->name, cmd);
3970
3971 switch (cmd) {
1da177e4
LT		 3972 case SIOCGMIIPHY:
3973 mii = if_mii(ifr);
3974 if (!mii) {
3975 return -EINVAL;
3976 }
3977 mii->phy_id = 0;
3978 /* Fall Through */
3979 case SIOCGMIIREG:
3980 /*
3981 * We do this again just in case we were called by SIOCGMIIREG
3982 * instead of SIOCGMIIPHY.
3983 */
3984 mii = if_mii(ifr);
3985 if (!mii) {
3986 return -EINVAL;
3987 }
3988
3989 if (mii->reg_num == 1) {
3990 struct bonding *bond = bond_dev->priv;
3991 mii->val_out = 0;
6603a6f2 3992 read_lock(&bond->lock);
1da177e4 3993 read_lock(&bond->curr_slave_lock);
4e140079 3994 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3995 mii->val_out = BMSR_LSTATUS;
3996 }
3997 read_unlock(&bond->curr_slave_lock);
6603a6f2 3998 read_unlock(&bond->lock);
1da177e4
LT		 3999 }
4000
4001 return 0;
4002 case BOND_INFO_QUERY_OLD:
4003 case SIOCBONDINFOQUERY:
4004 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4005
4006 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
4007 return -EFAULT;
4008 }
4009
4010 res = bond_info_query(bond_dev, &k_binfo);
4011 if (res == 0) {
4012 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
4013 return -EFAULT;
4014 }
4015 }
4016
4017 return res;
4018 case BOND_SLAVE_INFO_QUERY_OLD:
4019 case SIOCBONDSLAVEINFOQUERY:
4020 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4021
4022 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4023 return -EFAULT;
4024 }
4025
4026 res = bond_slave_info_query(bond_dev, &k_sinfo);
4027 if (res == 0) {
4028 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4029 return -EFAULT;
4030 }
4031 }
4032
4033 return res;
4034 default:
4035 /* Go on */
4036 break;
4037 }
4038
4039 if (!capable(CAP_NET_ADMIN)) {
4040 return -EPERM;
4041 }
4042
b76cdba9 4043 down_write(&(bonding_rwsem));
881d966b 4044 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT		 4045
4046 dprintk("slave_dev=%p: \n", slave_dev);
4047
4048 if (!slave_dev) {
4049 res = -ENODEV;
4050 } else {
4051 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4052 switch (cmd) {
4053 case BOND_ENSLAVE_OLD:
4054 case SIOCBONDENSLAVE:
4055 res = bond_enslave(bond_dev, slave_dev);
4056 break;
4057 case BOND_RELEASE_OLD:
4058 case SIOCBONDRELEASE:
4059 res = bond_release(bond_dev, slave_dev);
4060 break;
4061 case BOND_SETHWADDR_OLD:
4062 case SIOCBONDSETHWADDR:
4063 res = bond_sethwaddr(bond_dev, slave_dev);
4064 break;
4065 case BOND_CHANGE_ACTIVE_OLD:
4066 case SIOCBONDCHANGEACTIVE:
4067 res = bond_ioctl_change_active(bond_dev, slave_dev);
4068 break;
4069 default:
4070 res = -EOPNOTSUPP;
4071 }
4072
4073 dev_put(slave_dev);
4074 }
4075
b76cdba9 4076 up_write(&(bonding_rwsem));
1da177e4
LT		 4077 return res;
4078}
4079
4080static void bond_set_multicast_list(struct net_device *bond_dev)
4081{
4082 struct bonding *bond = bond_dev->priv;
4083 struct dev_mc_list *dmi;
4084
1da177e4
LT		 4085 /*
4086 * Do promisc before checking multicast_mode
4087 */
4088 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4089 /*
4090 * FIXME: Need to handle the error when one of the multi-slaves
4091 * encounters error.
4092 */
1da177e4
LT		 4093 bond_set_promiscuity(bond, 1);
4094 }
4095
4096 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4097 bond_set_promiscuity(bond, -1);
4098 }
4099
4100 /* set allmulti flag to slaves */
4101 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4102 /*
4103 * FIXME: Need to handle the error when one of the multi-slaves
4104 * encounters error.
4105 */
1da177e4
LT		 4106 bond_set_allmulti(bond, 1);
4107 }
4108
4109 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4110 bond_set_allmulti(bond, -1);
4111 }
4112
80ee5ad2
JV		 4113 read_lock(&bond->lock);
4114
1da177e4
LT		 4115 bond->flags = bond_dev->flags;
4116
4117 /* looking for addresses to add to slaves' mc list */
4118 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4119 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4120 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4121 }
4122 }
4123
4124 /* looking for addresses to delete from slaves' list */
4125 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4126 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4127 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4128 }
4129 }
4130
4131 /* save master's multicast list */
4132 bond_mc_list_destroy(bond);
4133 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4134
80ee5ad2 4135 read_unlock(&bond->lock);
1da177e4
LT		 4136}
4137
4138/*
4139 * Change the MTU of all of a master's slaves to match the master
4140 */
4141static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4142{
4143 struct bonding *bond = bond_dev->priv;
4144 struct slave *slave, *stop_at;
4145 int res = 0;
4146 int i;
4147
4148 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4149 (bond_dev ? bond_dev->name : "None"), new_mtu);
4150
4151 /* Can't hold bond->lock with bh disabled here since
4152 * some base drivers panic. On the other hand we can't
4153 * hold bond->lock without bh disabled because we'll
4154 * deadlock. The only solution is to rely on the fact
4155 * that we're under rtnl_lock here, and the slaves
4156 * list won't change. This doesn't solve the problem
4157 * of setting the slave's MTU while it is
4158 * transmitting, but the assumption is that the base
4159 * driver can handle that.
4160 *
4161 * TODO: figure out a way to safely iterate the slaves
4162 * list, but without holding a lock around the actual
4163 * call to the base driver.
4164 */
4165
4166 bond_for_each_slave(bond, slave, i) {
4167 dprintk("s %p s->p %p c_m %p\n", slave,
4168 slave->prev, slave->dev->change_mtu);
e944ef79 4169
1da177e4
LT		 4170 res = dev_set_mtu(slave->dev, new_mtu);
4171
4172 if (res) {
4173 /* If we failed to set the slave's mtu to the new value
4174 * we must abort the operation even in ACTIVE_BACKUP
4175 * mode, because if we allow the backup slaves to have
4176 * different mtu values than the active slave we'll
4177 * need to change their mtu when doing a failover. That
4178 * means changing their mtu from timer context, which
4179 * is probably not a good idea.
4180 */
4181 dprintk("err %d %s\n", res, slave->dev->name);
4182 goto unwind;
4183 }
4184 }
4185
4186 bond_dev->mtu = new_mtu;
4187
4188 return 0;
4189
4190unwind:
4191 /* unwind from head to the slave that failed */
4192 stop_at = slave;
4193 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4194 int tmp_res;
4195
4196 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4197 if (tmp_res) {
4198 dprintk("unwind err %d dev %s\n", tmp_res,
4199 slave->dev->name);
4200 }
4201 }
4202
4203 return res;
4204}
4205
4206/*
4207 * Change HW address
4208 *
4209 * Note that many devices must be down to change the HW address, and
4210 * downing the master releases all slaves. We can make bonds full of
4211 * bonding devices to test this, however.
4212 */
4213static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4214{
4215 struct bonding *bond = bond_dev->priv;
4216 struct sockaddr *sa = addr, tmp_sa;
4217 struct slave *slave, *stop_at;
4218 int res = 0;
4219 int i;
4220
4221 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4222
dd957c57 4223 /*
3915c1e8
JV		 4224 * If fail_over_mac is set to active, do nothing and return
4225 * success. Returning an error causes ifenslave to fail.
dd957c57 4226 */
3915c1e8 4227 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4228 return 0;
2ab82852 4229
1da177e4
LT		 4230 if (!is_valid_ether_addr(sa->sa_data)) {
4231 return -EADDRNOTAVAIL;
4232 }
4233
4234 /* Can't hold bond->lock with bh disabled here since
4235 * some base drivers panic. On the other hand we can't
4236 * hold bond->lock without bh disabled because we'll
4237 * deadlock. The only solution is to rely on the fact
4238 * that we're under rtnl_lock here, and the slaves
4239 * list won't change. This doesn't solve the problem
4240 * of setting the slave's hw address while it is
4241 * transmitting, but the assumption is that the base
4242 * driver can handle that.
4243 *
4244 * TODO: figure out a way to safely iterate the slaves
4245 * list, but without holding a lock around the actual
4246 * call to the base driver.
4247 */
4248
4249 bond_for_each_slave(bond, slave, i) {
4250 dprintk("slave %p %s\n", slave, slave->dev->name);
4251
4252 if (slave->dev->set_mac_address == NULL) {
4253 res = -EOPNOTSUPP;
4254 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4255 goto unwind;
4256 }
4257
4258 res = dev_set_mac_address(slave->dev, addr);
4259 if (res) {
4260 /* TODO: consider downing the slave
4261 * and retry ?
4262 * User should expect communications
4263 * breakage anyway until ARP finish
4264 * updating, so...
4265 */
4266 dprintk("err %d %s\n", res, slave->dev->name);
4267 goto unwind;
4268 }
4269 }
4270
4271 /* success */
4272 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4273 return 0;
4274
4275unwind:
4276 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4277 tmp_sa.sa_family = bond_dev->type;
4278
4279 /* unwind from head to the slave that failed */
4280 stop_at = slave;
4281 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4282 int tmp_res;
4283
4284 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4285 if (tmp_res) {
4286 dprintk("unwind err %d dev %s\n", tmp_res,
4287 slave->dev->name);
4288 }
4289 }
4290
4291 return res;
4292}
4293
4294static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4295{
4296 struct bonding *bond = bond_dev->priv;
4297 struct slave *slave, *start_at;
cf5f9044 4298 int i, slave_no, res = 1;
1da177e4
LT		 4299
4300 read_lock(&bond->lock);
4301
4302 if (!BOND_IS_OK(bond)) {
4303 goto out;
4304 }
4305
cf5f9044
JV		 4306 /*
4307 * Concurrent TX may collide on rr_tx_counter; we accept that
4308 * as being rare enough not to justify using an atomic op here
4309 */
4310 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4311
cf5f9044
JV		 4312 bond_for_each_slave(bond, slave, i) {
4313 slave_no--;
4314 if (slave_no < 0) {
4315 break;
4316 }
1da177e4
LT		 4317 }
4318
cf5f9044 4319 start_at = slave;
1da177e4
LT		 4320 bond_for_each_slave_from(bond, slave, i, start_at) {
4321 if (IS_UP(slave->dev) &&
4322 (slave->link == BOND_LINK_UP) &&
4323 (slave->state == BOND_STATE_ACTIVE)) {
4324 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4325 break;
4326 }
4327 }
4328
1da177e4
LT		 4329out:
4330 if (res) {
4331 /* no suitable interface, frame not sent */
4332 dev_kfree_skb(skb);
4333 }
4334 read_unlock(&bond->lock);
4335 return 0;
4336}
4337
075897ce 4338
1da177e4
LT		 4339/*
4340 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4341 * the bond has a usable interface.
4342 */
4343static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4344{
4345 struct bonding *bond = bond_dev->priv;
4346 int res = 1;
4347
1da177e4
LT		 4348 read_lock(&bond->lock);
4349 read_lock(&bond->curr_slave_lock);
4350
4351 if (!BOND_IS_OK(bond)) {
4352 goto out;
4353 }
4354
075897ce
JL		 4355 if (!bond->curr_active_slave)
4356 goto out;
4357
075897ce
JL		 4358 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4359
1da177e4
LT		 4360out:
4361 if (res) {
4362 /* no suitable interface, frame not sent */
4363 dev_kfree_skb(skb);
4364 }
4365 read_unlock(&bond->curr_slave_lock);
4366 read_unlock(&bond->lock);
4367 return 0;
4368}
4369
4370/*
169a3e66
JV		 4371 * In bond_xmit_xor() , we determine the output device by using a pre-
4372 * determined xmit_hash_policy(), If the selected device is not enabled,
4373 * find the next active slave.
1da177e4
LT		 4374 */
4375static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4376{
4377 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 4378 struct slave *slave, *start_at;
4379 int slave_no;
4380 int i;
4381 int res = 1;
4382
4383 read_lock(&bond->lock);
4384
4385 if (!BOND_IS_OK(bond)) {
4386 goto out;
4387 }
4388
169a3e66 4389 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4390
4391 bond_for_each_slave(bond, slave, i) {
4392 slave_no--;
4393 if (slave_no < 0) {
4394 break;
4395 }
4396 }
4397
4398 start_at = slave;
4399
4400 bond_for_each_slave_from(bond, slave, i, start_at) {
4401 if (IS_UP(slave->dev) &&
4402 (slave->link == BOND_LINK_UP) &&
4403 (slave->state == BOND_STATE_ACTIVE)) {
4404 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4405 break;
4406 }
4407 }
4408
4409out:
4410 if (res) {
4411 /* no suitable interface, frame not sent */
4412 dev_kfree_skb(skb);
4413 }
4414 read_unlock(&bond->lock);
4415 return 0;
4416}
4417
4418/*
4419 * in broadcast mode, we send everything to all usable interfaces.
4420 */
4421static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4422{
4423 struct bonding *bond = bond_dev->priv;
4424 struct slave *slave, *start_at;
4425 struct net_device *tx_dev = NULL;
4426 int i;
4427 int res = 1;
4428
4429 read_lock(&bond->lock);
4430
4431 if (!BOND_IS_OK(bond)) {
4432 goto out;
4433 }
4434
4435 read_lock(&bond->curr_slave_lock);
4436 start_at = bond->curr_active_slave;
4437 read_unlock(&bond->curr_slave_lock);
4438
4439 if (!start_at) {
4440 goto out;
4441 }
4442
4443 bond_for_each_slave_from(bond, slave, i, start_at) {
4444 if (IS_UP(slave->dev) &&
4445 (slave->link == BOND_LINK_UP) &&
4446 (slave->state == BOND_STATE_ACTIVE)) {
4447 if (tx_dev) {
4448 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4449 if (!skb2) {
4450 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4451 ": %s: Error: bond_xmit_broadcast(): "
4452 "skb_clone() failed\n",
4453 bond_dev->name);
1da177e4
LT		 4454 continue;
4455 }
4456
4457 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4458 if (res) {
4459 dev_kfree_skb(skb2);
4460 continue;
4461 }
4462 }
4463 tx_dev = slave->dev;
4464 }
4465 }
4466
4467 if (tx_dev) {
4468 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4469 }
4470
4471out:
4472 if (res) {
4473 /* no suitable interface, frame not sent */
4474 dev_kfree_skb(skb);
4475 }
4476 /* frame sent to all suitable interfaces */
4477 read_unlock(&bond->lock);
4478 return 0;
4479}
4480
4481/*------------------------- Device initialization ---------------------------*/
4482
6f6652be
JV		 4483static void bond_set_xmit_hash_policy(struct bonding *bond)
4484{
4485 switch (bond->params.xmit_policy) {
4486 case BOND_XMIT_POLICY_LAYER23:
4487 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4488 break;
4489 case BOND_XMIT_POLICY_LAYER34:
4490 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4491 break;
4492 case BOND_XMIT_POLICY_LAYER2:
4493 default:
4494 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4495 break;
4496 }
4497}
4498
1da177e4
LT		 4499/*
4500 * set bond mode specific net device operations
4501 */
a77b5325 4502void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4503{
169a3e66
JV		 4504 struct net_device *bond_dev = bond->dev;
4505
1da177e4
LT		 4506 switch (mode) {
4507 case BOND_MODE_ROUNDROBIN:
4508 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4509 break;
4510 case BOND_MODE_ACTIVEBACKUP:
4511 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4512 break;
4513 case BOND_MODE_XOR:
4514 bond_dev->hard_start_xmit = bond_xmit_xor;
6f6652be 4515 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4516 break;
4517 case BOND_MODE_BROADCAST:
4518 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4519 break;
4520 case BOND_MODE_8023AD:
8f903c70 4521 bond_set_master_3ad_flags(bond);
1da177e4 4522 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
6f6652be 4523 bond_set_xmit_hash_policy(bond);
1da177e4 4524 break;
1da177e4 4525 case BOND_MODE_ALB:
8f903c70
JV		 4526 bond_set_master_alb_flags(bond);
4527 /* FALLTHRU */
4528 case BOND_MODE_TLB:
1da177e4
LT		 4529 bond_dev->hard_start_xmit = bond_alb_xmit;
4530 bond_dev->set_mac_address = bond_alb_set_mac_address;
4531 break;
4532 default:
4533 /* Should never happen, mode already checked */
4534 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4535 ": %s: Error: Unknown bonding mode %d\n",
4536 bond_dev->name,
1da177e4
LT		 4537 mode);
4538 break;
4539 }
4540}
4541
217df670
JV		 4542static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4543 struct ethtool_drvinfo *drvinfo)
4544{
4545 strncpy(drvinfo->driver, DRV_NAME, 32);
4546 strncpy(drvinfo->version, DRV_VERSION, 32);
4547 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4548}
4549
7282d491 4550static const struct ethtool_ops bond_ethtool_ops = {
217df670 4551 .get_drvinfo = bond_ethtool_get_drvinfo,
8531c5ff
AK		 4552};
4553
1da177e4
LT		 4554/*
4555 * Does not allocate but creates a /proc entry.
4556 * Allowed to fail.
4557 */
3c535952 4558static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4
LT		 4559{
4560 struct bonding *bond = bond_dev->priv;
4561
4562 dprintk("Begin bond_init for %s\n", bond_dev->name);
4563
4564 /* initialize rwlocks */
4565 rwlock_init(&bond->lock);
4566 rwlock_init(&bond->curr_slave_lock);
4567
4568 bond->params = *params; /* copy params struct */
4569
1b76b316
JV		 4570 bond->wq = create_singlethread_workqueue(bond_dev->name);
4571 if (!bond->wq)
4572 return -ENOMEM;
4573
1da177e4
LT		 4574 /* Initialize pointers */
4575 bond->first_slave = NULL;
4576 bond->curr_active_slave = NULL;
4577 bond->current_arp_slave = NULL;
4578 bond->primary_slave = NULL;
4579 bond->dev = bond_dev;
1053f62c 4580 bond->send_grat_arp = 0;
d90a162a 4581 bond->setup_by_slave = 0;
1da177e4
LT		 4582 INIT_LIST_HEAD(&bond->vlan_list);
4583
4584 /* Initialize the device entry points */
4585 bond_dev->open = bond_open;
4586 bond_dev->stop = bond_close;
4587 bond_dev->get_stats = bond_get_stats;
4588 bond_dev->do_ioctl = bond_do_ioctl;
8531c5ff 4589 bond_dev->ethtool_ops = &bond_ethtool_ops;
1da177e4
LT		 4590 bond_dev->set_multicast_list = bond_set_multicast_list;
4591 bond_dev->change_mtu = bond_change_mtu;
4592 bond_dev->set_mac_address = bond_set_mac_address;
3a1521b7 4593 bond_dev->validate_addr = NULL;
1da177e4 4594
169a3e66 4595 bond_set_mode_ops(bond, bond->params.mode);
1da177e4
LT		 4596
4597 bond_dev->destructor = free_netdev;
4598
4599 /* Initialize the device options */
4600 bond_dev->tx_queue_len = 0;
4601 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4602 bond_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 4603
4604 /* At first, we block adding VLANs. That's the only way to
4605 * prevent problems that occur when adding VLANs over an
4606 * empty bond. The block will be removed once non-challenged
4607 * slaves are enslaved.
4608 */
4609 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4610
932ff279 4611 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4612 * transmitting */
4613 bond_dev->features |= NETIF_F_LLTX;
4614
4615 /* By default, we declare the bond to be fully
4616 * VLAN hardware accelerated capable. Special
4617 * care is taken in the various xmit functions
4618 * when there are slaves that are not hw accel
4619 * capable
4620 */
4621 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4622 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4623 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4624 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4625 NETIF_F_HW_VLAN_RX |
4626 NETIF_F_HW_VLAN_FILTER);
4627
4628#ifdef CONFIG_PROC_FS
4629 bond_create_proc_entry(bond);
4630#endif
1da177e4
LT		 4631 list_add_tail(&bond->bond_list, &bond_dev_list);
4632
4633 return 0;
4634}
4635
4636/* De-initialize device specific data.
4637 * Caller must hold rtnl_lock.
4638 */
c50b85d0 4639static void bond_deinit(struct net_device *bond_dev)
1da177e4
LT		 4640{
4641 struct bonding *bond = bond_dev->priv;
4642
4643 list_del(&bond->bond_list);
4644
4645#ifdef CONFIG_PROC_FS
4646 bond_remove_proc_entry(bond);
4647#endif
4648}
4649
fdaea7a9
JV		 4650static void bond_work_cancel_all(struct bonding *bond)
4651{
4652 write_lock_bh(&bond->lock);
4653 bond->kill_timers = 1;
4654 write_unlock_bh(&bond->lock);
4655
4656 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4657 cancel_delayed_work(&bond->mii_work);
4658
4659 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4660 cancel_delayed_work(&bond->arp_work);
4661
4662 if (bond->params.mode == BOND_MODE_ALB &&
4663 delayed_work_pending(&bond->alb_work))
4664 cancel_delayed_work(&bond->alb_work);
4665
4666 if (bond->params.mode == BOND_MODE_8023AD &&
4667 delayed_work_pending(&bond->ad_work))
4668 cancel_delayed_work(&bond->ad_work);
4669}
4670
1da177e4
LT		 4671/* Unregister and free all bond devices.
4672 * Caller must hold rtnl_lock.
4673 */
4674static void bond_free_all(void)
4675{
4676 struct bonding *bond, *nxt;
4677
4678 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4679 struct net_device *bond_dev = bond->dev;
4680
fdaea7a9 4681 bond_work_cancel_all(bond);
80ee5ad2 4682 netif_tx_lock_bh(bond_dev);
e308a5d8 4683 netif_addr_lock(bond_dev);
70298705 4684 bond_mc_list_destroy(bond);
e308a5d8 4685 netif_addr_unlock(bond_dev);
80ee5ad2 4686 netif_tx_unlock_bh(bond_dev);
70298705 4687 /* Release the bonded slaves */
4688 bond_release_all(bond_dev);
92b41daa 4689 bond_destroy(bond);
1da177e4
LT		 4690 }
4691
4692#ifdef CONFIG_PROC_FS
4693 bond_destroy_proc_dir();
4694#endif
4695}
4696
4697/*------------------------- Module initialization ---------------------------*/
4698
4699/*
4700 * Convert string input module parms. Accept either the
ece95f7f
JV		 4701 * number of the mode or its string name. A bit complicated because
4702 * some mode names are substrings of other names, and calls from sysfs
4703 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4704 */
ece95f7f 4705int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4706{
ece95f7f 4707 int mode = -1, i, rv;
a42e534f 4708 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4709
a42e534f
JV		 4710 for (p = (char *)buf; *p; p++)
4711 if (!(isdigit(*p) || isspace(*p)))
4712 break;
4713
4714 if (*p)
ece95f7f 4715 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4716 else
4717 rv = sscanf(buf, "%d", &mode);
4718
4719 if (!rv)
4720 return -1;
1da177e4
LT		 4721
4722 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4723 if (mode == tbl[i].mode)
4724 return tbl[i].mode;
4725 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4726 return tbl[i].mode;
1da177e4
LT		 4727 }
4728
4729 return -1;
4730}
4731
4732static int bond_check_params(struct bond_params *params)
4733{
3915c1e8 4734 int arp_validate_value, fail_over_mac_value;
f5b2b966 4735
1da177e4
LT		 4736 /*
4737 * Convert string parameters.
4738 */
4739 if (mode) {
4740 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4741 if (bond_mode == -1) {
4742 printk(KERN_ERR DRV_NAME
4743 ": Error: Invalid bonding mode \"%s\"\n",
4744 mode == NULL ? "NULL" : mode);
4745 return -EINVAL;
4746 }
4747 }
4748
169a3e66
JV		 4749 if (xmit_hash_policy) {
4750 if ((bond_mode != BOND_MODE_XOR) &&
4751 (bond_mode != BOND_MODE_8023AD)) {
4752 printk(KERN_INFO DRV_NAME
4753 ": xor_mode param is irrelevant in mode %s\n",
4754 bond_mode_name(bond_mode));
4755 } else {
4756 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4757 xmit_hashtype_tbl);
4758 if (xmit_hashtype == -1) {
4759 printk(KERN_ERR DRV_NAME
4760 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4761 xmit_hash_policy == NULL ? "NULL" :
4762 xmit_hash_policy);
4763 return -EINVAL;
4764 }
4765 }
4766 }
4767
1da177e4
LT		 4768 if (lacp_rate) {
4769 if (bond_mode != BOND_MODE_8023AD) {
4770 printk(KERN_INFO DRV_NAME
4771 ": lacp_rate param is irrelevant in mode %s\n",
4772 bond_mode_name(bond_mode));
4773 } else {
4774 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4775 if (lacp_fast == -1) {
4776 printk(KERN_ERR DRV_NAME
4777 ": Error: Invalid lacp rate \"%s\"\n",
4778 lacp_rate == NULL ? "NULL" : lacp_rate);
4779 return -EINVAL;
4780 }
4781 }
4782 }
4783
b8a9787e 4784 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4785 printk(KERN_WARNING DRV_NAME
4786 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4787 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4788 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4789 max_bonds = BOND_DEFAULT_MAX_BONDS;
4790 }
4791
4792 if (miimon < 0) {
4793 printk(KERN_WARNING DRV_NAME
4794 ": Warning: miimon module parameter (%d), "
4795 "not in range 0-%d, so it was reset to %d\n",
4796 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4797 miimon = BOND_LINK_MON_INTERV;
4798 }
4799
4800 if (updelay < 0) {
4801 printk(KERN_WARNING DRV_NAME
4802 ": Warning: updelay module parameter (%d), "
4803 "not in range 0-%d, so it was reset to 0\n",
4804 updelay, INT_MAX);
4805 updelay = 0;
4806 }
4807
4808 if (downdelay < 0) {
4809 printk(KERN_WARNING DRV_NAME
4810 ": Warning: downdelay module parameter (%d), "
4811 "not in range 0-%d, so it was reset to 0\n",
4812 downdelay, INT_MAX);
4813 downdelay = 0;
4814 }
4815
4816 if ((use_carrier != 0) && (use_carrier != 1)) {
4817 printk(KERN_WARNING DRV_NAME
4818 ": Warning: use_carrier module parameter (%d), "
4819 "not of valid value (0/1), so it was set to 1\n",
4820 use_carrier);
4821 use_carrier = 1;
4822 }
4823
7893b249
MS		 4824 if (num_grat_arp < 0 || num_grat_arp > 255) {
4825 printk(KERN_WARNING DRV_NAME
4826 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4827 "was reset to 1 \n", num_grat_arp);
4828 num_grat_arp = 1;
4829 }
4830
1da177e4
LT		 4831 /* reset values for 802.3ad */
4832 if (bond_mode == BOND_MODE_8023AD) {
4833 if (!miimon) {
4834 printk(KERN_WARNING DRV_NAME
4835 ": Warning: miimon must be specified, "
4836 "otherwise bonding will not detect link "
4837 "failure, speed and duplex which are "
4838 "essential for 802.3ad operation\n");
4839 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4840 miimon = 100;
4841 }
4842 }
4843
4844 /* reset values for TLB/ALB */
4845 if ((bond_mode == BOND_MODE_TLB) ||
4846 (bond_mode == BOND_MODE_ALB)) {
4847 if (!miimon) {
4848 printk(KERN_WARNING DRV_NAME
4849 ": Warning: miimon must be specified, "
4850 "otherwise bonding will not detect link "
4851 "failure and link speed which are essential "
4852 "for TLB/ALB load balancing\n");
4853 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4854 miimon = 100;
4855 }
4856 }
4857
4858 if (bond_mode == BOND_MODE_ALB) {
4859 printk(KERN_NOTICE DRV_NAME
4860 ": In ALB mode you might experience client "
4861 "disconnections upon reconnection of a link if the "
4862 "bonding module updelay parameter (%d msec) is "
4863 "incompatible with the forwarding delay time of the "
4864 "switch\n",
4865 updelay);
4866 }
4867
4868 if (!miimon) {
4869 if (updelay || downdelay) {
4870 /* just warn the user the up/down delay will have
4871 * no effect since miimon is zero...
4872 */
4873 printk(KERN_WARNING DRV_NAME
4874 ": Warning: miimon module parameter not set "
4875 "and updelay (%d) or downdelay (%d) module "
4876 "parameter is set; updelay and downdelay have "
4877 "no effect unless miimon is set\n",
4878 updelay, downdelay);
4879 }
4880 } else {
4881 /* don't allow arp monitoring */
4882 if (arp_interval) {
4883 printk(KERN_WARNING DRV_NAME
4884 ": Warning: miimon (%d) and arp_interval (%d) "
4885 "can't be used simultaneously, disabling ARP "
4886 "monitoring\n",
4887 miimon, arp_interval);
4888 arp_interval = 0;
4889 }
4890
4891 if ((updelay % miimon) != 0) {
4892 printk(KERN_WARNING DRV_NAME
4893 ": Warning: updelay (%d) is not a multiple "
4894 "of miimon (%d), updelay rounded to %d ms\n",
4895 updelay, miimon, (updelay / miimon) * miimon);
4896 }
4897
4898 updelay /= miimon;
4899
4900 if ((downdelay % miimon) != 0) {
4901 printk(KERN_WARNING DRV_NAME
4902 ": Warning: downdelay (%d) is not a multiple "
4903 "of miimon (%d), downdelay rounded to %d ms\n",
4904 downdelay, miimon,
4905 (downdelay / miimon) * miimon);
4906 }
4907
4908 downdelay /= miimon;
4909 }
4910
4911 if (arp_interval < 0) {
4912 printk(KERN_WARNING DRV_NAME
4913 ": Warning: arp_interval module parameter (%d) "
4914 ", not in range 0-%d, so it was reset to %d\n",
4915 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4916 arp_interval = BOND_LINK_ARP_INTERV;
4917 }
4918
4919 for (arp_ip_count = 0;
4920 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4921 arp_ip_count++) {
4922 /* not complete check, but should be good enough to
4923 catch mistakes */
4924 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4925 printk(KERN_WARNING DRV_NAME
4926 ": Warning: bad arp_ip_target module parameter "
4927 "(%s), ARP monitoring will not be performed\n",
4928 arp_ip_target[arp_ip_count]);
4929 arp_interval = 0;
4930 } else {
d3bb52b0 4931 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4932 arp_target[arp_ip_count] = ip;
4933 }
4934 }
4935
4936 if (arp_interval && !arp_ip_count) {
4937 /* don't allow arping if no arp_ip_target given... */
4938 printk(KERN_WARNING DRV_NAME
4939 ": Warning: arp_interval module parameter (%d) "
4940 "specified without providing an arp_ip_target "
4941 "parameter, arp_interval was reset to 0\n",
4942 arp_interval);
4943 arp_interval = 0;
4944 }
4945
f5b2b966
JV		 4946 if (arp_validate) {
4947 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4948 printk(KERN_ERR DRV_NAME
4949 ": arp_validate only supported in active-backup mode\n");
4950 return -EINVAL;
4951 }
4952 if (!arp_interval) {
4953 printk(KERN_ERR DRV_NAME
4954 ": arp_validate requires arp_interval\n");
4955 return -EINVAL;
4956 }
4957
4958 arp_validate_value = bond_parse_parm(arp_validate,
4959 arp_validate_tbl);
4960 if (arp_validate_value == -1) {
4961 printk(KERN_ERR DRV_NAME
4962 ": Error: invalid arp_validate \"%s\"\n",
4963 arp_validate == NULL ? "NULL" : arp_validate);
4964 return -EINVAL;
4965 }
4966 } else
4967 arp_validate_value = 0;
4968
1da177e4
LT		 4969 if (miimon) {
4970 printk(KERN_INFO DRV_NAME
4971 ": MII link monitoring set to %d ms\n",
4972 miimon);
4973 } else if (arp_interval) {
4974 int i;
4975
4976 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 4977 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4978 arp_interval,
4979 arp_validate_tbl[arp_validate_value].modename,
4980 arp_ip_count);
1da177e4
LT		 4981
4982 for (i = 0; i < arp_ip_count; i++)
4983 printk (" %s", arp_ip_target[i]);
4984
4985 printk("\n");
4986
b8a9787e 4987 } else if (max_bonds) {
1da177e4
LT		 4988 /* miimon and arp_interval not set, we need one so things
4989 * work as expected, see bonding.txt for details
4990 */
4991 printk(KERN_WARNING DRV_NAME
4992 ": Warning: either miimon or arp_interval and "
4993 "arp_ip_target module parameters must be specified, "
4994 "otherwise bonding will not detect link failures! see "
4995 "bonding.txt for details.\n");
4996 }
4997
4998 if (primary && !USES_PRIMARY(bond_mode)) {
4999 /* currently, using a primary only makes sense
5000 * in active backup, TLB or ALB modes
5001 */
5002 printk(KERN_WARNING DRV_NAME
5003 ": Warning: %s primary device specified but has no "
5004 "effect in %s mode\n",
5005 primary, bond_mode_name(bond_mode));
5006 primary = NULL;
5007 }
5008
3915c1e8
JV		 5009 if (fail_over_mac) {
5010 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5011 fail_over_mac_tbl);
5012 if (fail_over_mac_value == -1) {
5013 printk(KERN_ERR DRV_NAME
5014 ": Error: invalid fail_over_mac \"%s\"\n",
5015 arp_validate == NULL ? "NULL" : arp_validate);
5016 return -EINVAL;
5017 }
5018
5019 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5020 printk(KERN_WARNING DRV_NAME
5021 ": Warning: fail_over_mac only affects "
5022 "active-backup mode.\n");
5023 } else {
5024 fail_over_mac_value = BOND_FOM_NONE;
5025 }
dd957c57 5026
1da177e4
LT		 5027 /* fill params struct with the proper values */
5028 params->mode = bond_mode;
169a3e66 5029 params->xmit_policy = xmit_hashtype;
1da177e4 5030 params->miimon = miimon;
7893b249 5031 params->num_grat_arp = num_grat_arp;
1da177e4 5032 params->arp_interval = arp_interval;
f5b2b966 5033 params->arp_validate = arp_validate_value;
1da177e4
LT		 5034 params->updelay = updelay;
5035 params->downdelay = downdelay;
5036 params->use_carrier = use_carrier;
5037 params->lacp_fast = lacp_fast;
5038 params->primary[0] = 0;
3915c1e8 5039 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5040
5041 if (primary) {
5042 strncpy(params->primary, primary, IFNAMSIZ);
5043 params->primary[IFNAMSIZ - 1] = 0;
5044 }
5045
5046 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5047
5048 return 0;
5049}
5050
0daa2303
PZ		 5051static struct lock_class_key bonding_netdev_xmit_lock_key;
5052
c773e847
DM		 5053static void bond_set_lockdep_class_one(struct netdev_queue *txq)
5054{
5055 lockdep_set_class(&txq->_xmit_lock,
5056 &bonding_netdev_xmit_lock_key);
5057}
5058
5059static void bond_set_lockdep_class(struct net_device *dev)
5060{
5061 bond_set_lockdep_class_one(&dev->tx_queue);
5062}
5063
dfe60397 5064/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5065 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5066 * Caller must NOT hold rtnl_lock; we need to release it here before we
5067 * set up our sysfs entries.
5068 */
0dd646fe 5069int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5070{
5071 struct net_device *bond_dev;
0883beca 5072 struct bonding *bond;
dfe60397
MW		 5073 int res;
5074
5075 rtnl_lock();
027ea041
JV		 5076 down_write(&bonding_rwsem);
5077
5078 /* Check to see if the bond already exists. */
4fe4763c 5079 if (name) {
0883beca 5080 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5081 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5082 printk(KERN_ERR DRV_NAME
027ea041 5083 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5084 name);
5085 res = -EPERM;
5086 goto out_rtnl;
5087 }
5088 }
027ea041 5089
e4b91c48
JV		 5090 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5091 ether_setup);
dfe60397
MW		 5092 if (!bond_dev) {
5093 printk(KERN_ERR DRV_NAME
5094 ": %s: eek! can't alloc netdev!\n",
5095 name);
5096 res = -ENOMEM;
5097 goto out_rtnl;
5098 }
5099
e4b91c48
JV		 5100 if (!name) {
5101 res = dev_alloc_name(bond_dev, "bond%d");
5102 if (res < 0)
5103 goto out_netdev;
5104 }
5105
dfe60397
MW		 5106 /* bond_init() must be called after dev_alloc_name() (for the
5107 * /proc files), but before register_netdevice(), because we
5108 * need to set function pointers.
5109 */
5110
5111 res = bond_init(bond_dev, params);
5112 if (res < 0) {
5113 goto out_netdev;
5114 }
5115
dfe60397
MW		 5116 res = register_netdevice(bond_dev);
5117 if (res < 0) {
5118 goto out_bond;
5119 }
0daa2303 5120
c773e847 5121 bond_set_lockdep_class(bond_dev);
0daa2303 5122
ff59c456
JV		 5123 netif_carrier_off(bond_dev);
5124
027ea041 5125 up_write(&bonding_rwsem);
dfe60397 5126 rtnl_unlock(); /* allows sysfs registration of net device */
b76cdba9 5127 res = bond_create_sysfs_entry(bond_dev->priv);
09c89279
JV		 5128 if (res < 0) {
5129 rtnl_lock();
027ea041 5130 down_write(&bonding_rwsem);
822973ba
PE		 5131 bond_deinit(bond_dev);
5132 unregister_netdevice(bond_dev);
5133 goto out_rtnl;
09c89279
JV		 5134 }
5135
5136 return 0;
5137
dfe60397
MW		 5138out_bond:
5139 bond_deinit(bond_dev);
5140out_netdev:
5141 free_netdev(bond_dev);
5142out_rtnl:
027ea041 5143 up_write(&bonding_rwsem);
dfe60397 5144 rtnl_unlock();
dfe60397
MW		 5145 return res;
5146}
5147
1da177e4
LT		 5148static int __init bonding_init(void)
5149{
1da177e4
LT		 5150 int i;
5151 int res;
0883beca 5152 struct bonding *bond;
1da177e4
LT		 5153
5154 printk(KERN_INFO "%s", version);
5155
dfe60397 5156 res = bond_check_params(&bonding_defaults);
1da177e4 5157 if (res) {
dfe60397 5158 goto out;
1da177e4
LT		 5159 }
5160
1da177e4
LT		 5161#ifdef CONFIG_PROC_FS
5162 bond_create_proc_dir();
5163#endif
027ea041
JV		 5164
5165 init_rwsem(&bonding_rwsem);
5166
1da177e4 5167 for (i = 0; i < max_bonds; i++) {
0dd646fe 5168 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5169 if (res)
5170 goto err;
1da177e4
LT		 5171 }
5172
b76cdba9
MW		 5173 res = bond_create_sysfs();
5174 if (res)
5175 goto err;
5176
1da177e4 5177 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5178 register_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5179
dfe60397
MW		 5180 goto out;
5181err:
0883beca 5182 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5183 bond_work_cancel_all(bond);
5184 destroy_workqueue(bond->wq);
5185 }
5186
ae68c398
PE		 5187 bond_destroy_sysfs();
5188
40abc270 5189 rtnl_lock();
1da177e4 5190 bond_free_all();
1da177e4 5191 rtnl_unlock();
dfe60397 5192out:
1da177e4 5193 return res;
dfe60397 5194
1da177e4
LT		 5195}
5196
5197static void __exit bonding_exit(void)
5198{
5199 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5200 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5201
ae68c398
PE		 5202 bond_destroy_sysfs();
5203
1da177e4
LT		 5204 rtnl_lock();
5205 bond_free_all();
5206 rtnl_unlock();
5207}
5208
5209module_init(bonding_init);
5210module_exit(bonding_exit);
5211MODULE_LICENSE("GPL");
5212MODULE_VERSION(DRV_VERSION);
5213MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5214MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5215MODULE_SUPPORTED_DEVICE("most ethernet devices");
5216
5217/*
5218 * Local variables:
5219 * c-indent-level: 8
5220 * c-basic-offset: 8
5221 * tab-width: 8
5222 * End:
5223 */
5224
Clone of davem's net-next-2.6 tree