]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/bonding/bond_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net: replace NIPQUAD() in net/*/
[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 1343 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
b63365a2
HX		 1344 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1345
1346 if (!bond->first_slave)
1347 goto done;
1348
1349 features &= ~NETIF_F_ONE_FOR_ALL;
7f353bf2 1350
54ef3137 1351 bond_for_each_slave(bond, slave, i) {
b63365a2
HX		 1352 features = netdev_increment_features(features,
1353 slave->dev->features,
1354 NETIF_F_ONE_FOR_ALL);
54ef3137
JV		 1355 if (slave->dev->hard_header_len > max_hard_header_len)
1356 max_hard_header_len = slave->dev->hard_header_len;
1357 }
8531c5ff 1358
b63365a2 1359done:
7f353bf2 1360 features |= (bond_dev->features & BOND_VLAN_FEATURES);
b63365a2 1361 bond_dev->features = netdev_fix_features(features, NULL);
54ef3137 1362 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1363
1364 return 0;
1365}
1366
872254dd
MS		 1367
1368static void bond_setup_by_slave(struct net_device *bond_dev,
1369 struct net_device *slave_dev)
1370{
d90a162a
MS		 1371 struct bonding *bond = bond_dev->priv;
1372
872254dd 1373 bond_dev->neigh_setup = slave_dev->neigh_setup;
1284cd3a 1374 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1375
1376 bond_dev->type = slave_dev->type;
1377 bond_dev->hard_header_len = slave_dev->hard_header_len;
1378 bond_dev->addr_len = slave_dev->addr_len;
1379
1380 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1381 slave_dev->addr_len);
d90a162a 1382 bond->setup_by_slave = 1;
872254dd
MS		 1383}
1384
1da177e4 1385/* enslave device <slave> to bond device <master> */
a77b5325 1386int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1387{
1388 struct bonding *bond = bond_dev->priv;
1389 struct slave *new_slave = NULL;
1390 struct dev_mc_list *dmi;
1391 struct sockaddr addr;
1392 int link_reporting;
1393 int old_features = bond_dev->features;
1394 int res = 0;
1395
552709d5 1396 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1397 slave_dev->do_ioctl == NULL) {
1da177e4 1398 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1399 ": %s: Warning: no link monitoring support for %s\n",
1400 bond_dev->name, slave_dev->name);
1da177e4
LT		 1401 }
1402
1403 /* bond must be initialized by bond_open() before enslaving */
1404 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1405 printk(KERN_WARNING DRV_NAME
1406 " %s: master_dev is not up in bond_enslave\n",
1407 bond_dev->name);
1da177e4
LT		 1408 }
1409
1410 /* already enslaved */
1411 if (slave_dev->flags & IFF_SLAVE) {
1412 dprintk("Error, Device was already enslaved\n");
1413 return -EBUSY;
1414 }
1415
1416 /* vlan challenged mutual exclusion */
1417 /* no need to lock since we're protected by rtnl_lock */
1418 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1419 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1420 if (!list_empty(&bond->vlan_list)) {
1421 printk(KERN_ERR DRV_NAME
4e0952c7 1422 ": %s: Error: cannot enslave VLAN "
1da177e4 1423 "challenged slave %s on VLAN enabled "
4e0952c7 1424 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1425 bond_dev->name);
1426 return -EPERM;
1427 } else {
1428 printk(KERN_WARNING DRV_NAME
4e0952c7 1429 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1430 "slave %s. Adding VLANs will be blocked as "
1431 "long as %s is part of bond %s\n",
4e0952c7 1432 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1433 bond_dev->name);
1434 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1435 }
1436 } else {
1437 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1438 if (bond->slave_cnt == 0) {
1439 /* First slave, and it is not VLAN challenged,
1440 * so remove the block of adding VLANs over the bond.
1441 */
1442 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1443 }
1444 }
1445
217df670
JV		 1446 /*
1447 * Old ifenslave binaries are no longer supported. These can
1448 * be identified with moderate accurary by the state of the slave:
1449 * the current ifenslave will set the interface down prior to
1450 * enslaving it; the old ifenslave will not.
1451 */
1452 if ((slave_dev->flags & IFF_UP)) {
1453 printk(KERN_ERR DRV_NAME ": %s is up. "
1454 "This may be due to an out of date ifenslave.\n",
1455 slave_dev->name);
1456 res = -EPERM;
1457 goto err_undo_flags;
1458 }
1da177e4 1459
872254dd
MS		 1460 /* set bonding device ether type by slave - bonding netdevices are
1461 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1462 * there is a need to override some of the type dependent attribs/funcs.
1463 *
1464 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1465 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1466 */
1467 if (bond->slave_cnt == 0) {
1468 if (slave_dev->type != ARPHRD_ETHER)
1469 bond_setup_by_slave(bond_dev, slave_dev);
1470 } else if (bond_dev->type != slave_dev->type) {
1471 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1472 "from other slaves (%d), can not enslave it.\n",
1473 slave_dev->name,
1474 slave_dev->type, bond_dev->type);
1475 res = -EINVAL;
1476 goto err_undo_flags;
1477 }
1478
217df670 1479 if (slave_dev->set_mac_address == NULL) {
2ab82852
MS		 1480 if (bond->slave_cnt == 0) {
1481 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1482 ": %s: Warning: The first slave device "
1483 "specified does not support setting the MAC "
3915c1e8 1484 "address. Setting fail_over_mac to active.",
dd957c57 1485 bond_dev->name);
3915c1e8
JV		 1486 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1487 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1488 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1489 ": %s: Error: The slave device specified "
1490 "does not support setting the MAC address, "
3915c1e8 1491 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1492 , bond_dev->name);
1493 res = -EOPNOTSUPP;
1494 goto err_undo_flags;
1495 }
1da177e4
LT		 1496 }
1497
243cb4e5 1498 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1499 if (!new_slave) {
1500 res = -ENOMEM;
1501 goto err_undo_flags;
1502 }
1503
1da177e4
LT		 1504 /* save slave's original flags before calling
1505 * netdev_set_master and dev_open
1506 */
1507 new_slave->original_flags = slave_dev->flags;
1508
217df670
JV		 1509 /*
1510 * Save slave's original ("permanent") mac address for modes
1511 * that need it, and for restoring it upon release, and then
1512 * set it to the master's address
1513 */
1514 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1515
dd957c57 1516 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1517 /*
1518 * Set slave to master's mac address. The application already
1519 * set the master's mac address to that of the first slave
1520 */
1521 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1522 addr.sa_family = slave_dev->type;
1523 res = dev_set_mac_address(slave_dev, &addr);
1524 if (res) {
1525 dprintk("Error %d calling set_mac_address\n", res);
1526 goto err_free;
1527 }
217df670 1528 }
1da177e4 1529
c2edacf8
JV		 1530 res = netdev_set_master(slave_dev, bond_dev);
1531 if (res) {
1532 dprintk("Error %d calling netdev_set_master\n", res);
569f0c4d 1533 goto err_restore_mac;
c2edacf8 1534 }
217df670
JV		 1535 /* open the slave since the application closed it */
1536 res = dev_open(slave_dev);
1537 if (res) {
1538 dprintk("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1539 goto err_unset_master;
1da177e4
LT		 1540 }
1541
1da177e4 1542 new_slave->dev = slave_dev;
0b680e75 1543 slave_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 1544
1545 if ((bond->params.mode == BOND_MODE_TLB) ||
1546 (bond->params.mode == BOND_MODE_ALB)) {
1547 /* bond_alb_init_slave() must be called before all other stages since
1548 * it might fail and we do not want to have to undo everything
1549 */
1550 res = bond_alb_init_slave(bond, new_slave);
1551 if (res) {
569f0c4d 1552 goto err_close;
1da177e4
LT		 1553 }
1554 }
1555
1556 /* If the mode USES_PRIMARY, then the new slave gets the
1557 * master's promisc (and mc) settings only if it becomes the
1558 * curr_active_slave, and that is taken care of later when calling
1559 * bond_change_active()
1560 */
1561 if (!USES_PRIMARY(bond->params.mode)) {
1562 /* set promiscuity level to new slave */
1563 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1564 res = dev_set_promiscuity(slave_dev, 1);
1565 if (res)
1566 goto err_close;
1da177e4
LT		 1567 }
1568
1569 /* set allmulti level to new slave */
1570 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1571 res = dev_set_allmulti(slave_dev, 1);
1572 if (res)
1573 goto err_close;
1da177e4
LT		 1574 }
1575
b9e40857 1576 netif_addr_lock_bh(bond_dev);
1da177e4
LT		 1577 /* upload master's mc_list to new slave */
1578 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1579 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1580 }
b9e40857 1581 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1582 }
1583
1584 if (bond->params.mode == BOND_MODE_8023AD) {
1585 /* add lacpdu mc addr to mc list */
1586 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1587
1588 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1589 }
1590
1591 bond_add_vlans_on_slave(bond, slave_dev);
1592
1593 write_lock_bh(&bond->lock);
1594
1595 bond_attach_slave(bond, new_slave);
1596
1597 new_slave->delay = 0;
1598 new_slave->link_failure_count = 0;
1599
8531c5ff
AK		 1600 bond_compute_features(bond);
1601
3915c1e8
JV		 1602 write_unlock_bh(&bond->lock);
1603
1604 read_lock(&bond->lock);
1605
f5b2b966
JV		 1606 new_slave->last_arp_rx = jiffies;
1607
1da177e4
LT		 1608 if (bond->params.miimon && !bond->params.use_carrier) {
1609 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1610
1611 if ((link_reporting == -1) && !bond->params.arp_interval) {
1612 /*
1613 * miimon is set but a bonded network driver
1614 * does not support ETHTOOL/MII and
1615 * arp_interval is not set. Note: if
1616 * use_carrier is enabled, we will never go
1617 * here (because netif_carrier is always
1618 * supported); thus, we don't need to change
1619 * the messages for netif_carrier.
1620 */
1621 printk(KERN_WARNING DRV_NAME
4e0952c7 1622 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1623 "available for interface %s, and "
1624 "arp_interval/arp_ip_target module parameters "
1625 "not specified, thus bonding will not detect "
1626 "link failures! see bonding.txt for details.\n",
4e0952c7 1627 bond_dev->name, slave_dev->name);
1da177e4
LT		 1628 } else if (link_reporting == -1) {
1629 /* unable get link status using mii/ethtool */
1630 printk(KERN_WARNING DRV_NAME
4e0952c7 1631 ": %s: Warning: can't get link status from "
1da177e4
LT		 1632 "interface %s; the network driver associated "
1633 "with this interface does not support MII or "
1634 "ETHTOOL link status reporting, thus miimon "
1635 "has no effect on this interface.\n",
4e0952c7 1636 bond_dev->name, slave_dev->name);
1da177e4
LT		 1637 }
1638 }
1639
1640 /* check for initial state */
1641 if (!bond->params.miimon ||
1642 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1643 if (bond->params.updelay) {
1644 dprintk("Initial state of slave_dev is "
1645 "BOND_LINK_BACK\n");
1646 new_slave->link = BOND_LINK_BACK;
1647 new_slave->delay = bond->params.updelay;
1648 } else {
1649 dprintk("Initial state of slave_dev is "
1650 "BOND_LINK_UP\n");
1651 new_slave->link = BOND_LINK_UP;
1652 }
1653 new_slave->jiffies = jiffies;
1654 } else {
1655 dprintk("Initial state of slave_dev is "
1656 "BOND_LINK_DOWN\n");
1657 new_slave->link = BOND_LINK_DOWN;
1658 }
1659
1660 if (bond_update_speed_duplex(new_slave) &&
1661 (new_slave->link != BOND_LINK_DOWN)) {
1662 printk(KERN_WARNING DRV_NAME
4e0952c7 1663 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1664 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1665 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1666
1667 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1668 printk(KERN_WARNING DRV_NAME
1669 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1670 "support in base driver for proper aggregator "
4e0952c7 1671 "selection.\n", bond_dev->name);
1da177e4
LT		 1672 }
1673 }
1674
1675 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1676 /* if there is a primary slave, remember it */
1677 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1678 bond->primary_slave = new_slave;
1679 }
1680 }
1681
3915c1e8
JV		 1682 write_lock_bh(&bond->curr_slave_lock);
1683
1da177e4
LT		 1684 switch (bond->params.mode) {
1685 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1686 bond_set_slave_inactive_flags(new_slave);
1687 bond_select_active_slave(bond);
1da177e4
LT		 1688 break;
1689 case BOND_MODE_8023AD:
1690 /* in 802.3ad mode, the internal mechanism
1691 * will activate the slaves in the selected
1692 * aggregator
1693 */
1694 bond_set_slave_inactive_flags(new_slave);
1695 /* if this is the first slave */
1696 if (bond->slave_cnt == 1) {
1697 SLAVE_AD_INFO(new_slave).id = 1;
1698 /* Initialize AD with the number of times that the AD timer is called in 1 second
1699 * can be called only after the mac address of the bond is set
1700 */
1701 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1702 bond->params.lacp_fast);
1703 } else {
1704 SLAVE_AD_INFO(new_slave).id =
1705 SLAVE_AD_INFO(new_slave->prev).id + 1;
1706 }
1707
1708 bond_3ad_bind_slave(new_slave);
1709 break;
1710 case BOND_MODE_TLB:
1711 case BOND_MODE_ALB:
1712 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1713 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1714 break;
1715 default:
1716 dprintk("This slave is always active in trunk mode\n");
1717
1718 /* always active in trunk mode */
1719 new_slave->state = BOND_STATE_ACTIVE;
1720
1721 /* In trunking mode there is little meaning to curr_active_slave
1722 * anyway (it holds no special properties of the bond device),
1723 * so we can change it without calling change_active_interface()
1724 */
1725 if (!bond->curr_active_slave) {
1726 bond->curr_active_slave = new_slave;
1727 }
1728 break;
1729 } /* switch(bond_mode) */
1730
3915c1e8
JV		 1731 write_unlock_bh(&bond->curr_slave_lock);
1732
ff59c456
JV		 1733 bond_set_carrier(bond);
1734
3915c1e8 1735 read_unlock(&bond->lock);
1da177e4 1736
b76cdba9
MW		 1737 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1738 if (res)
569f0c4d 1739 goto err_close;
b76cdba9 1740
1da177e4
LT		 1741 printk(KERN_INFO DRV_NAME
1742 ": %s: enslaving %s as a%s interface with a%s link.\n",
1743 bond_dev->name, slave_dev->name,
1744 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1745 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1746
1747 /* enslave is successful */
1748 return 0;
1749
1750/* Undo stages on error */
1da177e4
LT		 1751err_close:
1752 dev_close(slave_dev);
1753
569f0c4d
JV		 1754err_unset_master:
1755 netdev_set_master(slave_dev, NULL);
1756
1da177e4 1757err_restore_mac:
dd957c57 1758 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1759 /* XXX TODO - fom follow mode needs to change master's
1760 * MAC if this slave's MAC is in use by the bond, or at
1761 * least print a warning.
1762 */
2ab82852
MS		 1763 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1764 addr.sa_family = slave_dev->type;
1765 dev_set_mac_address(slave_dev, &addr);
1766 }
1da177e4
LT		 1767
1768err_free:
1769 kfree(new_slave);
1770
1771err_undo_flags:
1772 bond_dev->features = old_features;
8531c5ff 1773
1da177e4
LT		 1774 return res;
1775}
1776
1777/*
1778 * Try to release the slave device <slave> from the bond device <master>
1779 * It is legal to access curr_active_slave without a lock because all the function
1780 * is write-locked.
1781 *
1782 * The rules for slave state should be:
1783 * for Active/Backup:
1784 * Active stays on all backups go down
1785 * for Bonded connections:
1786 * The first up interface should be left on and all others downed.
1787 */
a77b5325 1788int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4
LT		 1789{
1790 struct bonding *bond = bond_dev->priv;
1791 struct slave *slave, *oldcurrent;
1792 struct sockaddr addr;
1793 int mac_addr_differ;
1794
1795 /* slave is not a slave or master is not master of this slave */
1796 if (!(slave_dev->flags & IFF_SLAVE) ||
1797 (slave_dev->master != bond_dev)) {
1798 printk(KERN_ERR DRV_NAME
4e0952c7 1799 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1800 bond_dev->name, slave_dev->name);
1801 return -EINVAL;
1802 }
1803
1804 write_lock_bh(&bond->lock);
1805
1806 slave = bond_get_slave_by_dev(bond, slave_dev);
1807 if (!slave) {
1808 /* not a slave of this bond */
1809 printk(KERN_INFO DRV_NAME
1810 ": %s: %s not enslaved\n",
1811 bond_dev->name, slave_dev->name);
f5e2a7b2 1812 write_unlock_bh(&bond->lock);
1da177e4
LT		 1813 return -EINVAL;
1814 }
1815
3915c1e8
JV		 1816 if (!bond->params.fail_over_mac) {
1817 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1818 ETH_ALEN);
1819 if (!mac_addr_differ && (bond->slave_cnt > 1))
1820 printk(KERN_WARNING DRV_NAME
1821 ": %s: Warning: the permanent HWaddr of %s - "
e174961c 1822 "%pM - is still in use by %s. "
3915c1e8
JV		 1823 "Set the HWaddr of %s to a different address "
1824 "to avoid conflicts.\n",
1825 bond_dev->name, slave_dev->name,
e174961c 1826 slave->perm_hwaddr,
3915c1e8 1827 bond_dev->name, slave_dev->name);
1da177e4
LT		 1828 }
1829
1830 /* Inform AD package of unbinding of slave. */
1831 if (bond->params.mode == BOND_MODE_8023AD) {
1832 /* must be called before the slave is
1833 * detached from the list
1834 */
1835 bond_3ad_unbind_slave(slave);
1836 }
1837
1838 printk(KERN_INFO DRV_NAME
1839 ": %s: releasing %s interface %s\n",
1840 bond_dev->name,
1841 (slave->state == BOND_STATE_ACTIVE)
1842 ? "active" : "backup",
1843 slave_dev->name);
1844
1845 oldcurrent = bond->curr_active_slave;
1846
1847 bond->current_arp_slave = NULL;
1848
1849 /* release the slave from its bond */
1850 bond_detach_slave(bond, slave);
1851
8531c5ff
AK		 1852 bond_compute_features(bond);
1853
1da177e4
LT		 1854 if (bond->primary_slave == slave) {
1855 bond->primary_slave = NULL;
1856 }
1857
1858 if (oldcurrent == slave) {
1859 bond_change_active_slave(bond, NULL);
1860 }
1861
1862 if ((bond->params.mode == BOND_MODE_TLB) ||
1863 (bond->params.mode == BOND_MODE_ALB)) {
1864 /* Must be called only after the slave has been
1865 * detached from the list and the curr_active_slave
1866 * has been cleared (if our_slave == old_current),
1867 * but before a new active slave is selected.
1868 */
2543331d 1869 write_unlock_bh(&bond->lock);
1da177e4 1870 bond_alb_deinit_slave(bond, slave);
2543331d 1871 write_lock_bh(&bond->lock);
1da177e4
LT		 1872 }
1873
059fe7a5
JV		 1874 if (oldcurrent == slave) {
1875 /*
1876 * Note that we hold RTNL over this sequence, so there
1877 * is no concern that another slave add/remove event
1878 * will interfere.
1879 */
1880 write_unlock_bh(&bond->lock);
1881 read_lock(&bond->lock);
1882 write_lock_bh(&bond->curr_slave_lock);
1883
1da177e4
LT		 1884 bond_select_active_slave(bond);
1885
059fe7a5
JV		 1886 write_unlock_bh(&bond->curr_slave_lock);
1887 read_unlock(&bond->lock);
1888 write_lock_bh(&bond->lock);
1889 }
1890
1da177e4 1891 if (bond->slave_cnt == 0) {
ff59c456
JV		 1892 bond_set_carrier(bond);
1893
1da177e4
LT		 1894 /* if the last slave was removed, zero the mac address
1895 * of the master so it will be set by the application
1896 * to the mac address of the first slave
1897 */
1898 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1899
1900 if (list_empty(&bond->vlan_list)) {
1901 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1902 } else {
1903 printk(KERN_WARNING DRV_NAME
4e0952c7 1904 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1905 "still has VLANs.\n",
4e0952c7 1906 bond_dev->name, bond_dev->name);
1da177e4 1907 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1908 ": %s: When re-adding slaves, make sure the bond's "
1909 "HW address matches its VLANs'.\n",
1910 bond_dev->name);
1da177e4
LT		 1911 }
1912 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1913 !bond_has_challenged_slaves(bond)) {
1914 printk(KERN_INFO DRV_NAME
4e0952c7 1915 ": %s: last VLAN challenged slave %s "
1da177e4 1916 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1917 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1918 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1919 }
1920
1921 write_unlock_bh(&bond->lock);
1922
b76cdba9
MW		 1923 /* must do this from outside any spinlocks */
1924 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1925
1da177e4
LT		 1926 bond_del_vlans_from_slave(bond, slave_dev);
1927
1928 /* If the mode USES_PRIMARY, then we should only remove its
1929 * promisc and mc settings if it was the curr_active_slave, but that was
1930 * already taken care of above when we detached the slave
1931 */
1932 if (!USES_PRIMARY(bond->params.mode)) {
1933 /* unset promiscuity level from slave */
1934 if (bond_dev->flags & IFF_PROMISC) {
1935 dev_set_promiscuity(slave_dev, -1);
1936 }
1937
1938 /* unset allmulti level from slave */
1939 if (bond_dev->flags & IFF_ALLMULTI) {
1940 dev_set_allmulti(slave_dev, -1);
1941 }
1942
1943 /* flush master's mc_list from slave */
b9e40857 1944 netif_addr_lock_bh(bond_dev);
1da177e4 1945 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 1946 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1947 }
1948
1949 netdev_set_master(slave_dev, NULL);
1950
1951 /* close slave before restoring its mac address */
1952 dev_close(slave_dev);
1953
3915c1e8 1954 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1955 /* restore original ("permanent") mac address */
1956 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1957 addr.sa_family = slave_dev->type;
1958 dev_set_mac_address(slave_dev, &addr);
1959 }
1da177e4 1960
8f903c70 1961 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1962 IFF_SLAVE_INACTIVE | IFF_BONDING |
1963 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1964
1965 kfree(slave);
1966
1967 return 0; /* deletion OK */
1968}
1969
d90a162a
MS		 1970/*
1971* Destroy a bonding device.
1972* Must be under rtnl_lock when this function is called.
1973*/
1974void bond_destroy(struct bonding *bond)
1975{
1976 bond_deinit(bond->dev);
1977 bond_destroy_sysfs_entry(bond);
8cbdeec6 1978 unregister_netdevice(bond->dev);
d90a162a
MS		 1979}
1980
a434e43f
JV		 1981static void bond_destructor(struct net_device *bond_dev)
1982{
1983 struct bonding *bond = bond_dev->priv;
1984
1985 if (bond->wq)
1986 destroy_workqueue(bond->wq);
1987
1988 netif_addr_lock_bh(bond_dev);
1989 bond_mc_list_destroy(bond);
1990 netif_addr_unlock_bh(bond_dev);
1991
1992 free_netdev(bond_dev);
1993}
1994
d90a162a
MS		 1995/*
1996* First release a slave and than destroy the bond if no more slaves iare left.
1997* Must be under rtnl_lock when this function is called.
1998*/
1999int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2000{
2001 struct bonding *bond = bond_dev->priv;
2002 int ret;
2003
2004 ret = bond_release(bond_dev, slave_dev);
2005 if ((ret == 0) && (bond->slave_cnt == 0)) {
2006 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2007 bond_dev->name, bond_dev->name);
2008 bond_destroy(bond);
2009 }
2010 return ret;
2011}
2012
1da177e4
LT		 2013/*
2014 * This function releases all slaves.
2015 */
2016static int bond_release_all(struct net_device *bond_dev)
2017{
2018 struct bonding *bond = bond_dev->priv;
2019 struct slave *slave;
2020 struct net_device *slave_dev;
2021 struct sockaddr addr;
2022
2023 write_lock_bh(&bond->lock);
2024
ff59c456
JV		 2025 netif_carrier_off(bond_dev);
2026
1da177e4
LT		 2027 if (bond->slave_cnt == 0) {
2028 goto out;
2029 }
2030
2031 bond->current_arp_slave = NULL;
2032 bond->primary_slave = NULL;
2033 bond_change_active_slave(bond, NULL);
2034
2035 while ((slave = bond->first_slave) != NULL) {
2036 /* Inform AD package of unbinding of slave
2037 * before slave is detached from the list.
2038 */
2039 if (bond->params.mode == BOND_MODE_8023AD) {
2040 bond_3ad_unbind_slave(slave);
2041 }
2042
2043 slave_dev = slave->dev;
2044 bond_detach_slave(bond, slave);
2045
2543331d
JV		 2046 /* now that the slave is detached, unlock and perform
2047 * all the undo steps that should not be called from
2048 * within a lock.
2049 */
2050 write_unlock_bh(&bond->lock);
2051
1da177e4
LT		 2052 if ((bond->params.mode == BOND_MODE_TLB) ||
2053 (bond->params.mode == BOND_MODE_ALB)) {
2054 /* must be called only after the slave
2055 * has been detached from the list
2056 */
2057 bond_alb_deinit_slave(bond, slave);
2058 }
2059
8531c5ff
AK		 2060 bond_compute_features(bond);
2061
b76cdba9 2062 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2063 bond_del_vlans_from_slave(bond, slave_dev);
2064
2065 /* If the mode USES_PRIMARY, then we should only remove its
2066 * promisc and mc settings if it was the curr_active_slave, but that was
2067 * already taken care of above when we detached the slave
2068 */
2069 if (!USES_PRIMARY(bond->params.mode)) {
2070 /* unset promiscuity level from slave */
2071 if (bond_dev->flags & IFF_PROMISC) {
2072 dev_set_promiscuity(slave_dev, -1);
2073 }
2074
2075 /* unset allmulti level from slave */
2076 if (bond_dev->flags & IFF_ALLMULTI) {
2077 dev_set_allmulti(slave_dev, -1);
2078 }
2079
2080 /* flush master's mc_list from slave */
b9e40857 2081 netif_addr_lock_bh(bond_dev);
1da177e4 2082 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2083 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2084 }
2085
2086 netdev_set_master(slave_dev, NULL);
2087
2088 /* close slave before restoring its mac address */
2089 dev_close(slave_dev);
2090
dd957c57 2091 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2092 /* restore original ("permanent") mac address*/
2093 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2094 addr.sa_family = slave_dev->type;
2095 dev_set_mac_address(slave_dev, &addr);
2096 }
1da177e4 2097
8f903c70
JV		 2098 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2099 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2100
2101 kfree(slave);
2102
2103 /* re-acquire the lock before getting the next slave */
2104 write_lock_bh(&bond->lock);
2105 }
2106
2107 /* zero the mac address of the master so it will be
2108 * set by the application to the mac address of the
2109 * first slave
2110 */
2111 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2112
2113 if (list_empty(&bond->vlan_list)) {
2114 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2115 } else {
2116 printk(KERN_WARNING DRV_NAME
4e0952c7 2117 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2118 "still has VLANs.\n",
4e0952c7 2119 bond_dev->name, bond_dev->name);
1da177e4 2120 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2121 ": %s: When re-adding slaves, make sure the bond's "
2122 "HW address matches its VLANs'.\n",
2123 bond_dev->name);
1da177e4
LT		 2124 }
2125
2126 printk(KERN_INFO DRV_NAME
2127 ": %s: released all slaves\n",
2128 bond_dev->name);
2129
2130out:
2131 write_unlock_bh(&bond->lock);
2132
2133 return 0;
2134}
2135
2136/*
2137 * This function changes the active slave to slave <slave_dev>.
2138 * It returns -EINVAL in the following cases.
2139 * - <slave_dev> is not found in the list.
2140 * - There is not active slave now.
2141 * - <slave_dev> is already active.
2142 * - The link state of <slave_dev> is not BOND_LINK_UP.
2143 * - <slave_dev> is not running.
2144 * In these cases, this fuction does nothing.
2145 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2146 */
2147static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2148{
2149 struct bonding *bond = bond_dev->priv;
2150 struct slave *old_active = NULL;
2151 struct slave *new_active = NULL;
2152 int res = 0;
2153
2154 if (!USES_PRIMARY(bond->params.mode)) {
2155 return -EINVAL;
2156 }
2157
2158 /* Verify that master_dev is indeed the master of slave_dev */
2159 if (!(slave_dev->flags & IFF_SLAVE) ||
2160 (slave_dev->master != bond_dev)) {
2161 return -EINVAL;
2162 }
2163
059fe7a5 2164 read_lock(&bond->lock);
1da177e4 2165
059fe7a5 2166 read_lock(&bond->curr_slave_lock);
1da177e4 2167 old_active = bond->curr_active_slave;
059fe7a5
JV		 2168 read_unlock(&bond->curr_slave_lock);
2169
1da177e4
LT		 2170 new_active = bond_get_slave_by_dev(bond, slave_dev);
2171
2172 /*
2173 * Changing to the current active: do nothing; return success.
2174 */
2175 if (new_active && (new_active == old_active)) {
059fe7a5 2176 read_unlock(&bond->lock);
1da177e4
LT		 2177 return 0;
2178 }
2179
2180 if ((new_active) &&
2181 (old_active) &&
2182 (new_active->link == BOND_LINK_UP) &&
2183 IS_UP(new_active->dev)) {
059fe7a5 2184 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2185 bond_change_active_slave(bond, new_active);
059fe7a5 2186 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2187 } else {
2188 res = -EINVAL;
2189 }
2190
059fe7a5 2191 read_unlock(&bond->lock);
1da177e4
LT		 2192
2193 return res;
2194}
2195
1da177e4
LT		 2196static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2197{
2198 struct bonding *bond = bond_dev->priv;
2199
2200 info->bond_mode = bond->params.mode;
2201 info->miimon = bond->params.miimon;
2202
6603a6f2 2203 read_lock(&bond->lock);
1da177e4 2204 info->num_slaves = bond->slave_cnt;
6603a6f2 2205 read_unlock(&bond->lock);
1da177e4
LT		 2206
2207 return 0;
2208}
2209
2210static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2211{
2212 struct bonding *bond = bond_dev->priv;
2213 struct slave *slave;
2214 int i, found = 0;
2215
2216 if (info->slave_id < 0) {
2217 return -ENODEV;
2218 }
2219
6603a6f2 2220 read_lock(&bond->lock);
1da177e4
LT		 2221
2222 bond_for_each_slave(bond, slave, i) {
2223 if (i == (int)info->slave_id) {
2224 found = 1;
2225 break;
2226 }
2227 }
2228
6603a6f2 2229 read_unlock(&bond->lock);
1da177e4
LT		 2230
2231 if (found) {
2232 strcpy(info->slave_name, slave->dev->name);
2233 info->link = slave->link;
2234 info->state = slave->state;
2235 info->link_failure_count = slave->link_failure_count;
2236 } else {
2237 return -ENODEV;
2238 }
2239
2240 return 0;
2241}
2242
2243/*-------------------------------- Monitoring -------------------------------*/
2244
1da177e4 2245
f0c76d61
JV		 2246static int bond_miimon_inspect(struct bonding *bond)
2247{
2248 struct slave *slave;
2249 int i, link_state, commit = 0;
1da177e4
LT		 2250
2251 bond_for_each_slave(bond, slave, i) {
f0c76d61 2252 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2253
f0c76d61 2254 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2255
2256 switch (slave->link) {
f0c76d61
JV		 2257 case BOND_LINK_UP:
2258 if (link_state)
2259 continue;
1da177e4 2260
f0c76d61
JV		 2261 slave->link = BOND_LINK_FAIL;
2262 slave->delay = bond->params.downdelay;
2263 if (slave->delay) {
2264 printk(KERN_INFO DRV_NAME
2265 ": %s: link status down for %s"
2266 "interface %s, disabling it in %d ms.\n",
2267 bond->dev->name,
2268 (bond->params.mode ==
2269 BOND_MODE_ACTIVEBACKUP) ?
2270 ((slave->state == BOND_STATE_ACTIVE) ?
2271 "active " : "backup ") : "",
2272 slave->dev->name,
2273 bond->params.downdelay * bond->params.miimon);
1da177e4 2274 }
f0c76d61
JV		 2275 /*FALLTHRU*/
2276 case BOND_LINK_FAIL:
2277 if (link_state) {
2278 /*
2279 * recovered before downdelay expired
2280 */
2281 slave->link = BOND_LINK_UP;
1da177e4
LT		 2282 slave->jiffies = jiffies;
2283 printk(KERN_INFO DRV_NAME
2284 ": %s: link status up again after %d "
2285 "ms for interface %s.\n",
1b76b316 2286 bond->dev->name,
f0c76d61
JV		 2287 (bond->params.downdelay - slave->delay) *
2288 bond->params.miimon,
2289 slave->dev->name);
2290 continue;
1da177e4 2291 }
f0c76d61
JV		 2292
2293 if (slave->delay <= 0) {
2294 slave->new_link = BOND_LINK_DOWN;
2295 commit++;
2296 continue;
1da177e4 2297 }
1da177e4 2298
f0c76d61
JV		 2299 slave->delay--;
2300 break;
2301
2302 case BOND_LINK_DOWN:
2303 if (!link_state)
2304 continue;
2305
2306 slave->link = BOND_LINK_BACK;
2307 slave->delay = bond->params.updelay;
2308
2309 if (slave->delay) {
2310 printk(KERN_INFO DRV_NAME
2311 ": %s: link status up for "
2312 "interface %s, enabling it in %d ms.\n",
2313 bond->dev->name, slave->dev->name,
2314 bond->params.updelay *
2315 bond->params.miimon);
2316 }
2317 /*FALLTHRU*/
2318 case BOND_LINK_BACK:
2319 if (!link_state) {
2320 slave->link = BOND_LINK_DOWN;
1da177e4
LT		 2321 printk(KERN_INFO DRV_NAME
2322 ": %s: link status down again after %d "
2323 "ms for interface %s.\n",
1b76b316 2324 bond->dev->name,
f0c76d61
JV		 2325 (bond->params.updelay - slave->delay) *
2326 bond->params.miimon,
2327 slave->dev->name);
2328
2329 continue;
2330 }
2331
2332 if (slave->delay <= 0) {
2333 slave->new_link = BOND_LINK_UP;
2334 commit++;
2335 continue;
1da177e4 2336 }
f0c76d61
JV		 2337
2338 slave->delay--;
1da177e4 2339 break;
f0c76d61
JV		 2340 }
2341 }
1da177e4 2342
f0c76d61
JV		 2343 return commit;
2344}
1da177e4 2345
f0c76d61
JV		 2346static void bond_miimon_commit(struct bonding *bond)
2347{
2348 struct slave *slave;
2349 int i;
2350
2351 bond_for_each_slave(bond, slave, i) {
2352 switch (slave->new_link) {
2353 case BOND_LINK_NOCHANGE:
2354 continue;
1da177e4 2355
f0c76d61
JV		 2356 case BOND_LINK_UP:
2357 slave->link = BOND_LINK_UP;
2358 slave->jiffies = jiffies;
2359
2360 if (bond->params.mode == BOND_MODE_8023AD) {
2361 /* prevent it from being the active one */
2362 slave->state = BOND_STATE_BACKUP;
2363 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2364 /* make it immediately active */
2365 slave->state = BOND_STATE_ACTIVE;
2366 } else if (slave != bond->primary_slave) {
2367 /* prevent it from being the active one */
2368 slave->state = BOND_STATE_BACKUP;
1da177e4 2369 }
1da177e4 2370
f0c76d61
JV		 2371 printk(KERN_INFO DRV_NAME
2372 ": %s: link status definitely "
2373 "up for interface %s.\n",
2374 bond->dev->name, slave->dev->name);
1da177e4 2375
f0c76d61
JV		 2376 /* notify ad that the link status has changed */
2377 if (bond->params.mode == BOND_MODE_8023AD)
2378 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2379
f0c76d61
JV		 2380 if ((bond->params.mode == BOND_MODE_TLB) ||
2381 (bond->params.mode == BOND_MODE_ALB))
2382 bond_alb_handle_link_change(bond, slave,
2383 BOND_LINK_UP);
1da177e4 2384
f0c76d61
JV		 2385 if (!bond->curr_active_slave ||
2386 (slave == bond->primary_slave))
2387 goto do_failover;
1da177e4 2388
f0c76d61 2389 continue;
059fe7a5 2390
f0c76d61 2391 case BOND_LINK_DOWN:
fba4acda
JV		 2392 if (slave->link_failure_count < UINT_MAX)
2393 slave->link_failure_count++;
2394
f0c76d61 2395 slave->link = BOND_LINK_DOWN;
1da177e4 2396
f0c76d61
JV		 2397 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2398 bond->params.mode == BOND_MODE_8023AD)
2399 bond_set_slave_inactive_flags(slave);
2400
2401 printk(KERN_INFO DRV_NAME
2402 ": %s: link status definitely down for "
2403 "interface %s, disabling it\n",
2404 bond->dev->name, slave->dev->name);
2405
2406 if (bond->params.mode == BOND_MODE_8023AD)
2407 bond_3ad_handle_link_change(slave,
2408 BOND_LINK_DOWN);
2409
2410 if (bond->params.mode == BOND_MODE_TLB ||
2411 bond->params.mode == BOND_MODE_ALB)
2412 bond_alb_handle_link_change(bond, slave,
2413 BOND_LINK_DOWN);
2414
2415 if (slave == bond->curr_active_slave)
2416 goto do_failover;
2417
2418 continue;
2419
2420 default:
2421 printk(KERN_ERR DRV_NAME
2422 ": %s: invalid new link %d on slave %s\n",
2423 bond->dev->name, slave->new_link,
2424 slave->dev->name);
2425 slave->new_link = BOND_LINK_NOCHANGE;
2426
2427 continue;
2428 }
2429
2430do_failover:
2431 ASSERT_RTNL();
2432 write_lock_bh(&bond->curr_slave_lock);
2433 bond_select_active_slave(bond);
2434 write_unlock_bh(&bond->curr_slave_lock);
2435 }
2436
2437 bond_set_carrier(bond);
1da177e4
LT		 2438}
2439
0b0eef66
JV		 2440/*
2441 * bond_mii_monitor
2442 *
2443 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2444 * inspection, then (if inspection indicates something needs to be done)
2445 * an acquisition of appropriate locks followed by a commit phase to
2446 * implement whatever link state changes are indicated.
0b0eef66
JV		 2447 */
2448void bond_mii_monitor(struct work_struct *work)
2449{
2450 struct bonding *bond = container_of(work, struct bonding,
2451 mii_work.work);
0b0eef66
JV		 2452
2453 read_lock(&bond->lock);
f0c76d61
JV		 2454 if (bond->kill_timers)
2455 goto out;
2456
2457 if (bond->slave_cnt == 0)
2458 goto re_arm;
b59f9f74
JV		 2459
2460 if (bond->send_grat_arp) {
2461 read_lock(&bond->curr_slave_lock);
2462 bond_send_gratuitous_arp(bond);
2463 read_unlock(&bond->curr_slave_lock);
2464 }
2465
f0c76d61 2466 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2467 read_unlock(&bond->lock);
2468 rtnl_lock();
2469 read_lock(&bond->lock);
f0c76d61
JV		 2470
2471 bond_miimon_commit(bond);
2472
5655662d
JV		 2473 read_unlock(&bond->lock);
2474 rtnl_unlock(); /* might sleep, hold no other locks */
2475 read_lock(&bond->lock);
0b0eef66
JV		 2476 }
2477
f0c76d61
JV		 2478re_arm:
2479 if (bond->params.miimon)
2480 queue_delayed_work(bond->wq, &bond->mii_work,
2481 msecs_to_jiffies(bond->params.miimon));
2482out:
0b0eef66 2483 read_unlock(&bond->lock);
0b0eef66 2484}
c3ade5ca 2485
d3bb52b0 2486static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2487{
2488 struct in_device *idev;
2489 struct in_ifaddr *ifa;
a144ea4b 2490 __be32 addr = 0;
c3ade5ca
JV		 2491
2492 if (!dev)
2493 return 0;
2494
2495 rcu_read_lock();
e5ed6399 2496 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2497 if (!idev)
2498 goto out;
2499
2500 ifa = idev->ifa_list;
2501 if (!ifa)
2502 goto out;
2503
2504 addr = ifa->ifa_local;
2505out:
2506 rcu_read_unlock();
2507 return addr;
2508}
2509
d3bb52b0 2510static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2511{
0883beca 2512 struct vlan_entry *vlan;
f5b2b966
JV		 2513
2514 if (ip == bond->master_ip)
2515 return 1;
2516
0883beca 2517 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2518 if (ip == vlan->vlan_ip)
2519 return 1;
2520 }
2521
2522 return 0;
2523}
2524
c3ade5ca
JV		 2525/*
2526 * We go to the (large) trouble of VLAN tagging ARP frames because
2527 * switches in VLAN mode (especially if ports are configured as
2528 * "native" to a VLAN) might not pass non-tagged frames.
2529 */
d3bb52b0 2530static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2531{
2532 struct sk_buff *skb;
2533
2534 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2535 slave_dev->name, dest_ip, src_ip, vlan_id);
2536
2537 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2538 NULL, slave_dev->dev_addr, NULL);
2539
2540 if (!skb) {
2541 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2542 return;
2543 }
2544 if (vlan_id) {
2545 skb = vlan_put_tag(skb, vlan_id);
2546 if (!skb) {
2547 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2548 return;
2549 }
2550 }
2551 arp_xmit(skb);
2552}
2553
2554
1da177e4
LT		 2555static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2556{
c3ade5ca 2557 int i, vlan_id, rv;
d3bb52b0 2558 __be32 *targets = bond->params.arp_targets;
0883beca 2559 struct vlan_entry *vlan;
c3ade5ca
JV		 2560 struct net_device *vlan_dev;
2561 struct flowi fl;
2562 struct rtable *rt;
1da177e4 2563
6b780567
MW		 2564 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2565 if (!targets[i])
2566 continue;
c3ade5ca
JV		 2567 dprintk("basa: target %x\n", targets[i]);
2568 if (list_empty(&bond->vlan_list)) {
2569 dprintk("basa: empty vlan: arp_send\n");
2570 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2571 bond->master_ip, 0);
2572 continue;
2573 }
2574
2575 /*
2576 * If VLANs are configured, we do a route lookup to
2577 * determine which VLAN interface would be used, so we
2578 * can tag the ARP with the proper VLAN tag.
2579 */
2580 memset(&fl, 0, sizeof(fl));
2581 fl.fl4_dst = targets[i];
2582 fl.fl4_tos = RTO_ONLINK;
2583
f206351a 2584 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2585 if (rv) {
2586 if (net_ratelimit()) {
2587 printk(KERN_WARNING DRV_NAME
2588 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2589 bond->dev->name, NIPQUAD(fl.fl4_dst));
2590 }
2591 continue;
2592 }
2593
2594 /*
2595 * This target is not on a VLAN
2596 */
2597 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2598 ip_rt_put(rt);
c3ade5ca
JV		 2599 dprintk("basa: rtdev == bond->dev: arp_send\n");
2600 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2601 bond->master_ip, 0);
2602 continue;
2603 }
2604
2605 vlan_id = 0;
0883beca 2606 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2607 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2608 if (vlan_dev == rt->u.dst.dev) {
2609 vlan_id = vlan->vlan_id;
2610 dprintk("basa: vlan match on %s %d\n",
2611 vlan_dev->name, vlan_id);
2612 break;
2613 }
2614 }
2615
2616 if (vlan_id) {
ed4b9f80 2617 ip_rt_put(rt);
c3ade5ca
JV		 2618 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2619 vlan->vlan_ip, vlan_id);
2620 continue;
2621 }
2622
2623 if (net_ratelimit()) {
2624 printk(KERN_WARNING DRV_NAME
2625 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2626 bond->dev->name, NIPQUAD(fl.fl4_dst),
2627 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2628 }
ed4b9f80 2629 ip_rt_put(rt);
c3ade5ca
JV		 2630 }
2631}
2632
2633/*
2634 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2635 * for each VLAN above us.
b59f9f74
JV		 2636 *
2637 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2638 */
2639static void bond_send_gratuitous_arp(struct bonding *bond)
2640{
2641 struct slave *slave = bond->curr_active_slave;
2642 struct vlan_entry *vlan;
2643 struct net_device *vlan_dev;
2644
2645 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2646 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2647
2648 if (!slave || !bond->send_grat_arp ||
2649 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2650 return;
2651
b59f9f74
JV		 2652 bond->send_grat_arp--;
2653
c3ade5ca
JV		 2654 if (bond->master_ip) {
2655 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2656 bond->master_ip, 0);
c3ade5ca
JV		 2657 }
2658
2659 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2660 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2661 if (vlan->vlan_ip) {
2662 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2663 vlan->vlan_ip, vlan->vlan_id);
2664 }
1da177e4
LT		 2665 }
2666}
2667
d3bb52b0 2668static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2669{
2670 int i;
d3bb52b0 2671 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2672
2673 targets = bond->params.arp_targets;
2674 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2675 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2676 "%u.%u.%u.%u bhti(tip) %d\n",
2677 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2678 bond_has_this_ip(bond, tip));
2679 if (sip == targets[i]) {
2680 if (bond_has_this_ip(bond, tip))
2681 slave->last_arp_rx = jiffies;
2682 return;
2683 }
2684 }
2685}
2686
2687static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2688{
2689 struct arphdr *arp;
2690 struct slave *slave;
2691 struct bonding *bond;
2692 unsigned char *arp_ptr;
d3bb52b0 2693 __be32 sip, tip;
f5b2b966 2694
c346dca1 2695 if (dev_net(dev) != &init_net)
e730c155
EB		 2696 goto out;
2697
f5b2b966
JV		 2698 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2699 goto out;
2700
2701 bond = dev->priv;
2702 read_lock(&bond->lock);
2703
2704 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2705 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2706 orig_dev ? orig_dev->name : "NULL");
2707
2708 slave = bond_get_slave_by_dev(bond, orig_dev);
2709 if (!slave || !slave_do_arp_validate(bond, slave))
2710 goto out_unlock;
2711
988b7050 2712 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2713 goto out_unlock;
2714
d0a92be0 2715 arp = arp_hdr(skb);
f5b2b966
JV		 2716 if (arp->ar_hln != dev->addr_len ||
2717 skb->pkt_type == PACKET_OTHERHOST ||
2718 skb->pkt_type == PACKET_LOOPBACK ||
2719 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2720 arp->ar_pro != htons(ETH_P_IP) ||
2721 arp->ar_pln != 4)
2722 goto out_unlock;
2723
2724 arp_ptr = (unsigned char *)(arp + 1);
2725 arp_ptr += dev->addr_len;
2726 memcpy(&sip, arp_ptr, 4);
2727 arp_ptr += 4 + dev->addr_len;
2728 memcpy(&tip, arp_ptr, 4);
2729
2730 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2731 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2732 slave->state, bond->params.arp_validate,
2733 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2734
2735 /*
2736 * Backup slaves won't see the ARP reply, but do come through
2737 * here for each ARP probe (so we swap the sip/tip to validate
2738 * the probe). In a "redundant switch, common router" type of
2739 * configuration, the ARP probe will (hopefully) travel from
2740 * the active, through one switch, the router, then the other
2741 * switch before reaching the backup.
2742 */
2743 if (slave->state == BOND_STATE_ACTIVE)
2744 bond_validate_arp(bond, slave, sip, tip);
2745 else
2746 bond_validate_arp(bond, slave, tip, sip);
2747
2748out_unlock:
2749 read_unlock(&bond->lock);
2750out:
2751 dev_kfree_skb(skb);
2752 return NET_RX_SUCCESS;
2753}
2754
1da177e4
LT		 2755/*
2756 * this function is called regularly to monitor each slave's link
2757 * ensuring that traffic is being sent and received when arp monitoring
2758 * is used in load-balancing mode. if the adapter has been dormant, then an
2759 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2760 * arp monitoring in active backup mode.
2761 */
1b76b316 2762void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2763{
1b76b316
JV		 2764 struct bonding *bond = container_of(work, struct bonding,
2765 arp_work.work);
1da177e4
LT		 2766 struct slave *slave, *oldcurrent;
2767 int do_failover = 0;
2768 int delta_in_ticks;
2769 int i;
2770
2771 read_lock(&bond->lock);
2772
5ce0da8f 2773 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2774
2775 if (bond->kill_timers) {
2776 goto out;
2777 }
2778
2779 if (bond->slave_cnt == 0) {
2780 goto re_arm;
2781 }
2782
2783 read_lock(&bond->curr_slave_lock);
2784 oldcurrent = bond->curr_active_slave;
2785 read_unlock(&bond->curr_slave_lock);
2786
2787 /* see if any of the previous devices are up now (i.e. they have
2788 * xmt and rcv traffic). the curr_active_slave does not come into
2789 * the picture unless it is null. also, slave->jiffies is not needed
2790 * here because we send an arp on each slave and give a slave as
2791 * long as it needs to get the tx/rx within the delta.
2792 * TODO: what about up/down delay in arp mode? it wasn't here before
2793 * so it can wait
2794 */
2795 bond_for_each_slave(bond, slave, i) {
2796 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2797 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2798 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2799
2800 slave->link = BOND_LINK_UP;
2801 slave->state = BOND_STATE_ACTIVE;
2802
2803 /* primary_slave has no meaning in round-robin
2804 * mode. the window of a slave being up and
2805 * curr_active_slave being null after enslaving
2806 * is closed.
2807 */
2808 if (!oldcurrent) {
2809 printk(KERN_INFO DRV_NAME
2810 ": %s: link status definitely "
2811 "up for interface %s, ",
1b76b316 2812 bond->dev->name,
1da177e4
LT		 2813 slave->dev->name);
2814 do_failover = 1;
2815 } else {
2816 printk(KERN_INFO DRV_NAME
2817 ": %s: interface %s is now up\n",
1b76b316 2818 bond->dev->name,
1da177e4
LT		 2819 slave->dev->name);
2820 }
2821 }
2822 } else {
2823 /* slave->link == BOND_LINK_UP */
2824
2825 /* not all switches will respond to an arp request
2826 * when the source ip is 0, so don't take the link down
2827 * if we don't know our ip yet
2828 */
b63bb739 2829 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2830 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2831
2832 slave->link = BOND_LINK_DOWN;
2833 slave->state = BOND_STATE_BACKUP;
2834
2835 if (slave->link_failure_count < UINT_MAX) {
2836 slave->link_failure_count++;
2837 }
2838
2839 printk(KERN_INFO DRV_NAME
2840 ": %s: interface %s is now down.\n",
1b76b316 2841 bond->dev->name,
1da177e4
LT		 2842 slave->dev->name);
2843
2844 if (slave == oldcurrent) {
2845 do_failover = 1;
2846 }
2847 }
2848 }
2849
2850 /* note: if switch is in round-robin mode, all links
2851 * must tx arp to ensure all links rx an arp - otherwise
2852 * links may oscillate or not come up at all; if switch is
2853 * in something like xor mode, there is nothing we can
2854 * do - all replies will be rx'ed on same link causing slaves
2855 * to be unstable during low/no traffic periods
2856 */
2857 if (IS_UP(slave->dev)) {
2858 bond_arp_send_all(bond, slave);
2859 }
2860 }
2861
2862 if (do_failover) {
059fe7a5 2863 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2864
2865 bond_select_active_slave(bond);
2866
059fe7a5 2867 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2868 }
2869
2870re_arm:
1b76b316
JV		 2871 if (bond->params.arp_interval)
2872 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2873out:
2874 read_unlock(&bond->lock);
2875}
2876
2877/*
b2220cad
JV		 2878 * Called to inspect slaves for active-backup mode ARP monitor link state
2879 * changes. Sets new_link in slaves to specify what action should take
2880 * place for the slave. Returns 0 if no changes are found, >0 if changes
2881 * to link states must be committed.
2882 *
2883 * Called with bond->lock held for read.
1da177e4 2884 */
b2220cad 2885static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2886{
1da177e4 2887 struct slave *slave;
b2220cad 2888 int i, commit = 0;
1da177e4 2889
b2220cad
JV		 2890 bond_for_each_slave(bond, slave, i) {
2891 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2892
b2220cad
JV		 2893 if (slave->link != BOND_LINK_UP) {
2894 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2895 delta_in_ticks)) {
2896 slave->new_link = BOND_LINK_UP;
2897 commit++;
2898 }
1da177e4 2899
b2220cad
JV		 2900 continue;
2901 }
1da177e4 2902
b2220cad
JV		 2903 /*
2904 * Give slaves 2*delta after being enslaved or made
2905 * active. This avoids bouncing, as the last receive
2906 * times need a full ARP monitor cycle to be updated.
2907 */
2908 if (!time_after_eq(jiffies, slave->jiffies +
2909 2 * delta_in_ticks))
2910 continue;
2911
2912 /*
2913 * Backup slave is down if:
2914 * - No current_arp_slave AND
2915 * - more than 3*delta since last receive AND
2916 * - the bond has an IP address
2917 *
2918 * Note: a non-null current_arp_slave indicates
2919 * the curr_active_slave went down and we are
2920 * searching for a new one; under this condition
2921 * we only take the curr_active_slave down - this
2922 * gives each slave a chance to tx/rx traffic
2923 * before being taken out
2924 */
2925 if (slave->state == BOND_STATE_BACKUP &&
2926 !bond->current_arp_slave &&
2927 time_after(jiffies, slave_last_rx(bond, slave) +
2928 3 * delta_in_ticks)) {
2929 slave->new_link = BOND_LINK_DOWN;
2930 commit++;
2931 }
2932
2933 /*
2934 * Active slave is down if:
2935 * - more than 2*delta since transmitting OR
2936 * - (more than 2*delta since receive AND
2937 * the bond has an IP address)
2938 */
2939 if ((slave->state == BOND_STATE_ACTIVE) &&
2940 (time_after_eq(jiffies, slave->dev->trans_start +
2941 2 * delta_in_ticks) ||
2942 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2943 + 2 * delta_in_ticks)))) {
2944 slave->new_link = BOND_LINK_DOWN;
2945 commit++;
2946 }
1da177e4
LT		 2947 }
2948
b2220cad
JV		 2949 read_lock(&bond->curr_slave_lock);
2950
2951 /*
2952 * Trigger a commit if the primary option setting has changed.
1da177e4 2953 */
b2220cad
JV		 2954 if (bond->primary_slave &&
2955 (bond->primary_slave != bond->curr_active_slave) &&
2956 (bond->primary_slave->link == BOND_LINK_UP))
2957 commit++;
1da177e4 2958
b2220cad 2959 read_unlock(&bond->curr_slave_lock);
1da177e4 2960
b2220cad
JV		 2961 return commit;
2962}
1da177e4 2963
b2220cad
JV		 2964/*
2965 * Called to commit link state changes noted by inspection step of
2966 * active-backup mode ARP monitor.
2967 *
2968 * Called with RTNL and bond->lock for read.
2969 */
2970static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2971{
2972 struct slave *slave;
2973 int i;
1da177e4 2974
b2220cad
JV		 2975 bond_for_each_slave(bond, slave, i) {
2976 switch (slave->new_link) {
2977 case BOND_LINK_NOCHANGE:
2978 continue;
ff59c456 2979
b2220cad
JV		 2980 case BOND_LINK_UP:
2981 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2982
b2220cad
JV		 2983 if (!bond->curr_active_slave &&
2984 time_before_eq(jiffies, slave->dev->trans_start +
2985 delta_in_ticks)) {
2986 slave->link = BOND_LINK_UP;
2987 bond_change_active_slave(bond, slave);
2988 bond->current_arp_slave = NULL;
2989
2990 printk(KERN_INFO DRV_NAME
2991 ": %s: %s is up and now the "
2992 "active interface\n",
2993 bond->dev->name, slave->dev->name);
2994
2995 } else if (bond->curr_active_slave != slave) {
2996 /* this slave has just come up but we
2997 * already have a current slave; this can
2998 * also happen if bond_enslave adds a new
2999 * slave that is up while we are searching
3000 * for a new slave
1da177e4 3001 */
b2220cad
JV		 3002 slave->link = BOND_LINK_UP;
3003 bond_set_slave_inactive_flags(slave);
3004 bond->current_arp_slave = NULL;
3005
3006 printk(KERN_INFO DRV_NAME
3007 ": %s: backup interface %s is now up\n",
3008 bond->dev->name, slave->dev->name);
3009 }
1da177e4 3010
b2220cad 3011 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3012
b2220cad 3013 break;
1da177e4 3014
b2220cad
JV		 3015 case BOND_LINK_DOWN:
3016 if (slave->link_failure_count < UINT_MAX)
3017 slave->link_failure_count++;
3018
3019 slave->link = BOND_LINK_DOWN;
3020
3021 if (slave == bond->curr_active_slave) {
3022 printk(KERN_INFO DRV_NAME
3023 ": %s: link status down for active "
3024 "interface %s, disabling it\n",
3025 bond->dev->name, slave->dev->name);
1da177e4
LT		 3026
3027 bond_set_slave_inactive_flags(slave);
3028
b2220cad
JV		 3029 write_lock_bh(&bond->curr_slave_lock);
3030
3031 bond_select_active_slave(bond);
3032 if (bond->curr_active_slave)
3033 bond->curr_active_slave->jiffies =
3034 jiffies;
3035
3036 write_unlock_bh(&bond->curr_slave_lock);
3037
3038 bond->current_arp_slave = NULL;
3039
3040 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3041 printk(KERN_INFO DRV_NAME
3042 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3043 bond->dev->name, slave->dev->name);
3044
3045 bond_set_slave_inactive_flags(slave);
1da177e4 3046 }
b2220cad
JV		 3047 break;
3048
3049 default:
3050 printk(KERN_ERR DRV_NAME
3051 ": %s: impossible: new_link %d on slave %s\n",
3052 bond->dev->name, slave->new_link,
3053 slave->dev->name);
1da177e4
LT		 3054 }
3055 }
3056
b2220cad
JV		 3057 /*
3058 * No race with changes to primary via sysfs, as we hold rtnl.
3059 */
3060 if (bond->primary_slave &&
3061 (bond->primary_slave != bond->curr_active_slave) &&
3062 (bond->primary_slave->link == BOND_LINK_UP)) {
3063 write_lock_bh(&bond->curr_slave_lock);
3064 bond_change_active_slave(bond, bond->primary_slave);
3065 write_unlock_bh(&bond->curr_slave_lock);
3066 }
1da177e4 3067
b2220cad
JV		 3068 bond_set_carrier(bond);
3069}
1da177e4 3070
b2220cad
JV		 3071/*
3072 * Send ARP probes for active-backup mode ARP monitor.
3073 *
3074 * Called with bond->lock held for read.
3075 */
3076static void bond_ab_arp_probe(struct bonding *bond)
3077{
3078 struct slave *slave;
3079 int i;
1da177e4 3080
b2220cad 3081 read_lock(&bond->curr_slave_lock);
1da177e4 3082
b2220cad
JV		 3083 if (bond->current_arp_slave && bond->curr_active_slave)
3084 printk("PROBE: c_arp %s && cas %s BAD\n",
3085 bond->current_arp_slave->dev->name,
3086 bond->curr_active_slave->dev->name);
1da177e4 3087
b2220cad
JV		 3088 if (bond->curr_active_slave) {
3089 bond_arp_send_all(bond, bond->curr_active_slave);
3090 read_unlock(&bond->curr_slave_lock);
3091 return;
3092 }
1da177e4 3093
b2220cad 3094 read_unlock(&bond->curr_slave_lock);
059fe7a5 3095
b2220cad
JV		 3096 /* if we don't have a curr_active_slave, search for the next available
3097 * backup slave from the current_arp_slave and make it the candidate
3098 * for becoming the curr_active_slave
3099 */
1da177e4 3100
b2220cad
JV		 3101 if (!bond->current_arp_slave) {
3102 bond->current_arp_slave = bond->first_slave;
3103 if (!bond->current_arp_slave)
3104 return;
3105 }
1da177e4 3106
b2220cad 3107 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3108
b2220cad
JV		 3109 /* search for next candidate */
3110 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3111 if (IS_UP(slave->dev)) {
3112 slave->link = BOND_LINK_BACK;
3113 bond_set_slave_active_flags(slave);
3114 bond_arp_send_all(bond, slave);
1da177e4 3115 slave->jiffies = jiffies;
b2220cad
JV		 3116 bond->current_arp_slave = slave;
3117 break;
1da177e4
LT		 3118 }
3119
b2220cad
JV		 3120 /* if the link state is up at this point, we
3121 * mark it down - this can happen if we have
3122 * simultaneous link failures and
3123 * reselect_active_interface doesn't make this
3124 * one the current slave so it is still marked
3125 * up when it is actually down
1da177e4 3126 */
b2220cad
JV		 3127 if (slave->link == BOND_LINK_UP) {
3128 slave->link = BOND_LINK_DOWN;
3129 if (slave->link_failure_count < UINT_MAX)
3130 slave->link_failure_count++;
1da177e4 3131
b2220cad
JV		 3132 bond_set_slave_inactive_flags(slave);
3133
3134 printk(KERN_INFO DRV_NAME
3135 ": %s: backup interface %s is now down.\n",
3136 bond->dev->name, slave->dev->name);
1da177e4 3137 }
b2220cad
JV		 3138 }
3139}
1da177e4 3140
b2220cad
JV		 3141void bond_activebackup_arp_mon(struct work_struct *work)
3142{
3143 struct bonding *bond = container_of(work, struct bonding,
3144 arp_work.work);
3145 int delta_in_ticks;
1da177e4 3146
b2220cad 3147 read_lock(&bond->lock);
1da177e4 3148
b2220cad
JV		 3149 if (bond->kill_timers)
3150 goto out;
1da177e4 3151
b2220cad 3152 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3153
b2220cad
JV		 3154 if (bond->slave_cnt == 0)
3155 goto re_arm;
3156
b59f9f74
JV		 3157 if (bond->send_grat_arp) {
3158 read_lock(&bond->curr_slave_lock);
3159 bond_send_gratuitous_arp(bond);
3160 read_unlock(&bond->curr_slave_lock);
3161 }
3162
b2220cad
JV		 3163 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3164 read_unlock(&bond->lock);
3165 rtnl_lock();
3166 read_lock(&bond->lock);
3167
3168 bond_ab_arp_commit(bond, delta_in_ticks);
3169
3170 read_unlock(&bond->lock);
3171 rtnl_unlock();
3172 read_lock(&bond->lock);
1da177e4
LT		 3173 }
3174
b2220cad
JV		 3175 bond_ab_arp_probe(bond);
3176
1da177e4
LT		 3177re_arm:
3178 if (bond->params.arp_interval) {
1b76b316 3179 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3180 }
3181out:
3182 read_unlock(&bond->lock);
3183}
3184
3185/*------------------------------ proc/seq_file-------------------------------*/
3186
3187#ifdef CONFIG_PROC_FS
3188
1da177e4
LT		 3189static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3190{
3191 struct bonding *bond = seq->private;
3192 loff_t off = 0;
3193 struct slave *slave;
3194 int i;
3195
3196 /* make sure the bond won't be taken away */
3197 read_lock(&dev_base_lock);
6603a6f2 3198 read_lock(&bond->lock);
1da177e4
LT		 3199
3200 if (*pos == 0) {
3201 return SEQ_START_TOKEN;
3202 }
3203
3204 bond_for_each_slave(bond, slave, i) {
3205 if (++off == *pos) {
3206 return slave;
3207 }
3208 }
3209
3210 return NULL;
3211}
3212
3213static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3214{
3215 struct bonding *bond = seq->private;
3216 struct slave *slave = v;
3217
3218 ++*pos;
3219 if (v == SEQ_START_TOKEN) {
3220 return bond->first_slave;
3221 }
3222
3223 slave = slave->next;
3224
3225 return (slave == bond->first_slave) ? NULL : slave;
3226}
3227
3228static void bond_info_seq_stop(struct seq_file *seq, void *v)
3229{
3230 struct bonding *bond = seq->private;
3231
6603a6f2 3232 read_unlock(&bond->lock);
1da177e4
LT		 3233 read_unlock(&dev_base_lock);
3234}
3235
3236static void bond_info_show_master(struct seq_file *seq)
3237{
3238 struct bonding *bond = seq->private;
3239 struct slave *curr;
4756b02f 3240 int i;
1da177e4
LT		 3241
3242 read_lock(&bond->curr_slave_lock);
3243 curr = bond->curr_active_slave;
3244 read_unlock(&bond->curr_slave_lock);
3245
dd957c57 3246 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3247 bond_mode_name(bond->params.mode));
3248
dd957c57
JV		 3249 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3250 bond->params.fail_over_mac)
3915c1e8
JV		 3251 seq_printf(seq, " (fail_over_mac %s)",
3252 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3253
3254 seq_printf(seq, "\n");
3255
c61b75ad
MW		 3256 if (bond->params.mode == BOND_MODE_XOR ||
3257 bond->params.mode == BOND_MODE_8023AD) {
3258 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3259 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3260 bond->params.xmit_policy);
3261 }
3262
1da177e4
LT		 3263 if (USES_PRIMARY(bond->params.mode)) {
3264 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3265 (bond->primary_slave) ?
3266 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3267
3268 seq_printf(seq, "Currently Active Slave: %s\n",
3269 (curr) ? curr->dev->name : "None");
3270 }
3271
ff59c456
JV		 3272 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3273 "up" : "down");
1da177e4
LT		 3274 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3275 seq_printf(seq, "Up Delay (ms): %d\n",
3276 bond->params.updelay * bond->params.miimon);
3277 seq_printf(seq, "Down Delay (ms): %d\n",
3278 bond->params.downdelay * bond->params.miimon);
3279
4756b02f
MW		 3280
3281 /* ARP information */
3282 if(bond->params.arp_interval > 0) {
3283 int printed=0;
3284 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3285 bond->params.arp_interval);
3286
3287 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3288
3289 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3290 if (!bond->params.arp_targets[i])
3291 continue;
3292 if (printed)
3293 seq_printf(seq, ",");
8cf14e38 3294 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
4756b02f
MW		 3295 printed = 1;
3296 }
3297 seq_printf(seq, "\n");
3298 }
3299
1da177e4
LT		 3300 if (bond->params.mode == BOND_MODE_8023AD) {
3301 struct ad_info ad_info;
3302
3303 seq_puts(seq, "\n802.3ad info\n");
3304 seq_printf(seq, "LACP rate: %s\n",
3305 (bond->params.lacp_fast) ? "fast" : "slow");
3306
3307 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3308 seq_printf(seq, "bond %s has no active aggregator\n",
3309 bond->dev->name);
3310 } else {
3311 seq_printf(seq, "Active Aggregator Info:\n");
3312
3313 seq_printf(seq, "\tAggregator ID: %d\n",
3314 ad_info.aggregator_id);
3315 seq_printf(seq, "\tNumber of ports: %d\n",
3316 ad_info.ports);
3317 seq_printf(seq, "\tActor Key: %d\n",
3318 ad_info.actor_key);
3319 seq_printf(seq, "\tPartner Key: %d\n",
3320 ad_info.partner_key);
e174961c
JB		 3321 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3322 ad_info.partner_system);
1da177e4
LT		 3323 }
3324 }
3325}
3326
3327static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3328{
3329 struct bonding *bond = seq->private;
3330
3331 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3332 seq_printf(seq, "MII Status: %s\n",
3333 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3334 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3335 slave->link_failure_count);
3336
e174961c 3337 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
1da177e4
LT		 3338
3339 if (bond->params.mode == BOND_MODE_8023AD) {
3340 const struct aggregator *agg
3341 = SLAVE_AD_INFO(slave).port.aggregator;
3342
3343 if (agg) {
3344 seq_printf(seq, "Aggregator ID: %d\n",
3345 agg->aggregator_identifier);
3346 } else {
3347 seq_puts(seq, "Aggregator ID: N/A\n");
3348 }
3349 }
3350}
3351
3352static int bond_info_seq_show(struct seq_file *seq, void *v)
3353{
3354 if (v == SEQ_START_TOKEN) {
3355 seq_printf(seq, "%s\n", version);
3356 bond_info_show_master(seq);
3357 } else {
3358 bond_info_show_slave(seq, v);
3359 }
3360
3361 return 0;
3362}
3363
3364static struct seq_operations bond_info_seq_ops = {
3365 .start = bond_info_seq_start,
3366 .next = bond_info_seq_next,
3367 .stop = bond_info_seq_stop,
3368 .show = bond_info_seq_show,
3369};
3370
3371static int bond_info_open(struct inode *inode, struct file *file)
3372{
3373 struct seq_file *seq;
3374 struct proc_dir_entry *proc;
3375 int res;
3376
3377 res = seq_open(file, &bond_info_seq_ops);
3378 if (!res) {
3379 /* recover the pointer buried in proc_dir_entry data */
3380 seq = file->private_data;
3381 proc = PDE(inode);
3382 seq->private = proc->data;
3383 }
3384
3385 return res;
3386}
3387
d54b1fdb 3388static const struct file_operations bond_info_fops = {
1da177e4
LT		 3389 .owner = THIS_MODULE,
3390 .open = bond_info_open,
3391 .read = seq_read,
3392 .llseek = seq_lseek,
3393 .release = seq_release,
3394};
3395
3396static int bond_create_proc_entry(struct bonding *bond)
3397{
3398 struct net_device *bond_dev = bond->dev;
3399
3400 if (bond_proc_dir) {
a95609cb
DL		 3401 bond->proc_entry = proc_create_data(bond_dev->name,
3402 S_IRUGO, bond_proc_dir,
3403 &bond_info_fops, bond);
1da177e4
LT		 3404 if (bond->proc_entry == NULL) {
3405 printk(KERN_WARNING DRV_NAME
3406 ": Warning: Cannot create /proc/net/%s/%s\n",
3407 DRV_NAME, bond_dev->name);
3408 } else {
1da177e4
LT		 3409 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3410 }
3411 }
3412
3413 return 0;
3414}
3415
3416static void bond_remove_proc_entry(struct bonding *bond)
3417{
3418 if (bond_proc_dir && bond->proc_entry) {
3419 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3420 memset(bond->proc_file_name, 0, IFNAMSIZ);
3421 bond->proc_entry = NULL;
3422 }
3423}
3424
3425/* Create the bonding directory under /proc/net, if doesn't exist yet.
3426 * Caller must hold rtnl_lock.
3427 */
3428static void bond_create_proc_dir(void)
3429{
3430 int len = strlen(DRV_NAME);
3431
457c4cbc 3432 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3433 bond_proc_dir = bond_proc_dir->next) {
3434 if ((bond_proc_dir->namelen == len) &&
3435 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3436 break;
3437 }
3438 }
3439
3440 if (!bond_proc_dir) {
457c4cbc 3441 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3442 if (bond_proc_dir) {
3443 bond_proc_dir->owner = THIS_MODULE;
3444 } else {
3445 printk(KERN_WARNING DRV_NAME
3446 ": Warning: cannot create /proc/net/%s\n",
3447 DRV_NAME);
3448 }
3449 }
3450}
3451
3452/* Destroy the bonding directory under /proc/net, if empty.
3453 * Caller must hold rtnl_lock.
3454 */
3455static void bond_destroy_proc_dir(void)
3456{
3457 struct proc_dir_entry *de;
3458
3459 if (!bond_proc_dir) {
3460 return;
3461 }
3462
3463 /* verify that the /proc dir is empty */
3464 for (de = bond_proc_dir->subdir; de; de = de->next) {
3465 /* ignore . and .. */
3466 if (*(de->name) != '.') {
3467 break;
3468 }
3469 }
3470
3471 if (de) {
3472 if (bond_proc_dir->owner == THIS_MODULE) {
3473 bond_proc_dir->owner = NULL;
3474 }
3475 } else {
457c4cbc 3476 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3477 bond_proc_dir = NULL;
3478 }
3479}
3480#endif /* CONFIG_PROC_FS */
3481
3482/*-------------------------- netdev event handling --------------------------*/
3483
3484/*
3485 * Change device name
3486 */
3487static int bond_event_changename(struct bonding *bond)
3488{
3489#ifdef CONFIG_PROC_FS
3490 bond_remove_proc_entry(bond);
3491 bond_create_proc_entry(bond);
3492#endif
b76cdba9
MW		 3493 down_write(&(bonding_rwsem));
3494 bond_destroy_sysfs_entry(bond);
3495 bond_create_sysfs_entry(bond);
3496 up_write(&(bonding_rwsem));
1da177e4
LT		 3497 return NOTIFY_DONE;
3498}
3499
3500static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3501{
3502 struct bonding *event_bond = bond_dev->priv;
3503
3504 switch (event) {
3505 case NETDEV_CHANGENAME:
3506 return bond_event_changename(event_bond);
3507 case NETDEV_UNREGISTER:
3b96c858 3508 bond_release_all(event_bond->dev);
1da177e4
LT		 3509 break;
3510 default:
3511 break;
3512 }
3513
3514 return NOTIFY_DONE;
3515}
3516
3517static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3518{
3519 struct net_device *bond_dev = slave_dev->master;
8531c5ff 3520 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3521
3522 switch (event) {
3523 case NETDEV_UNREGISTER:
3524 if (bond_dev) {
1284cd3a
JV		 3525 if (bond->setup_by_slave)
3526 bond_release_and_destroy(bond_dev, slave_dev);
3527 else
3528 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3529 }
3530 break;
3531 case NETDEV_CHANGE:
3532 /*
3533 * TODO: is this what we get if somebody
3534 * sets up a hierarchical bond, then rmmod's
3535 * one of the slave bonding devices?
3536 */
3537 break;
3538 case NETDEV_DOWN:
3539 /*
3540 * ... Or is it this?
3541 */
3542 break;
3543 case NETDEV_CHANGEMTU:
3544 /*
3545 * TODO: Should slaves be allowed to
3546 * independently alter their MTU? For
3547 * an active-backup bond, slaves need
3548 * not be the same type of device, so
3549 * MTUs may vary. For other modes,
3550 * slaves arguably should have the
3551 * same MTUs. To do this, we'd need to
3552 * take over the slave's change_mtu
3553 * function for the duration of their
3554 * servitude.
3555 */
3556 break;
3557 case NETDEV_CHANGENAME:
3558 /*
3559 * TODO: handle changing the primary's name
3560 */
3561 break;
8531c5ff
AK		 3562 case NETDEV_FEAT_CHANGE:
3563 bond_compute_features(bond);
3564 break;
1da177e4
LT		 3565 default:
3566 break;
3567 }
3568
3569 return NOTIFY_DONE;
3570}
3571
3572/*
3573 * bond_netdev_event: handle netdev notifier chain events.
3574 *
3575 * This function receives events for the netdev chain. The caller (an
e041c683 3576 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3577 * locks for us to safely manipulate the slave devices (RTNL lock,
3578 * dev_probe_lock).
3579 */
3580static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3581{
3582 struct net_device *event_dev = (struct net_device *)ptr;
3583
c346dca1 3584 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3585 return NOTIFY_DONE;
3586
1da177e4
LT		 3587 dprintk("event_dev: %s, event: %lx\n",
3588 (event_dev ? event_dev->name : "None"),
3589 event);
3590
0b680e75
JV		 3591 if (!(event_dev->priv_flags & IFF_BONDING))
3592 return NOTIFY_DONE;
3593
1da177e4
LT		 3594 if (event_dev->flags & IFF_MASTER) {
3595 dprintk("IFF_MASTER\n");
3596 return bond_master_netdev_event(event, event_dev);
3597 }
3598
3599 if (event_dev->flags & IFF_SLAVE) {
3600 dprintk("IFF_SLAVE\n");
3601 return bond_slave_netdev_event(event, event_dev);
3602 }
3603
3604 return NOTIFY_DONE;
3605}
3606
c3ade5ca
JV		 3607/*
3608 * bond_inetaddr_event: handle inetaddr notifier chain events.
3609 *
3610 * We keep track of device IPs primarily to use as source addresses in
3611 * ARP monitor probes (rather than spewing out broadcasts all the time).
3612 *
3613 * We track one IP for the main device (if it has one), plus one per VLAN.
3614 */
3615static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3616{
3617 struct in_ifaddr *ifa = ptr;
3618 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3619 struct bonding *bond;
3620 struct vlan_entry *vlan;
c3ade5ca 3621
c346dca1 3622 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3623 return NOTIFY_DONE;
3624
0883beca 3625 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3626 if (bond->dev == event_dev) {
3627 switch (event) {
3628 case NETDEV_UP:
3629 bond->master_ip = ifa->ifa_local;
3630 return NOTIFY_OK;
3631 case NETDEV_DOWN:
3632 bond->master_ip = bond_glean_dev_ip(bond->dev);
3633 return NOTIFY_OK;
3634 default:
3635 return NOTIFY_DONE;
3636 }
3637 }
3638
0883beca 3639 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3640 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3641 if (vlan_dev == event_dev) {
3642 switch (event) {
3643 case NETDEV_UP:
3644 vlan->vlan_ip = ifa->ifa_local;
3645 return NOTIFY_OK;
3646 case NETDEV_DOWN:
3647 vlan->vlan_ip =
3648 bond_glean_dev_ip(vlan_dev);
3649 return NOTIFY_OK;
3650 default:
3651 return NOTIFY_DONE;
3652 }
3653 }
3654 }
3655 }
3656 return NOTIFY_DONE;
3657}
3658
1da177e4
LT		 3659static struct notifier_block bond_netdev_notifier = {
3660 .notifier_call = bond_netdev_event,
3661};
3662
c3ade5ca
JV		 3663static struct notifier_block bond_inetaddr_notifier = {
3664 .notifier_call = bond_inetaddr_event,
3665};
3666
1da177e4
LT		 3667/*-------------------------- Packet type handling ---------------------------*/
3668
3669/* register to receive lacpdus on a bond */
3670static void bond_register_lacpdu(struct bonding *bond)
3671{
3672 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3673
3674 /* initialize packet type */
3675 pk_type->type = PKT_TYPE_LACPDU;
3676 pk_type->dev = bond->dev;
3677 pk_type->func = bond_3ad_lacpdu_recv;
3678
3679 dev_add_pack(pk_type);
3680}
3681
3682/* unregister to receive lacpdus on a bond */
3683static void bond_unregister_lacpdu(struct bonding *bond)
3684{
3685 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3686}
3687
f5b2b966
JV		 3688void bond_register_arp(struct bonding *bond)
3689{
3690 struct packet_type *pt = &bond->arp_mon_pt;
3691
c4f283b1
JV		 3692 if (pt->type)
3693 return;
3694
f5b2b966 3695 pt->type = htons(ETH_P_ARP);
e245cb71 3696 pt->dev = bond->dev;
f5b2b966
JV		 3697 pt->func = bond_arp_rcv;
3698 dev_add_pack(pt);
3699}
3700
3701void bond_unregister_arp(struct bonding *bond)
3702{
c4f283b1
JV		 3703 struct packet_type *pt = &bond->arp_mon_pt;
3704
3705 dev_remove_pack(pt);
3706 pt->type = 0;
f5b2b966
JV		 3707}
3708
169a3e66
JV		 3709/*---------------------------- Hashing Policies -----------------------------*/
3710
6f6652be
JV		 3711/*
3712 * Hash for the output device based upon layer 2 and layer 3 data. If
3713 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3714 */
3715static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3716 struct net_device *bond_dev, int count)
3717{
3718 struct ethhdr *data = (struct ethhdr *)skb->data;
3719 struct iphdr *iph = ip_hdr(skb);
3720
f14c4e4e 3721 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be
JV		 3722 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3723 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3724 }
3725
3726 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3727}
3728
169a3e66 3729/*
59c51591 3730 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3731 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3732 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3733 */
3734static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3735 struct net_device *bond_dev, int count)
3736{
3737 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3738 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3739 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3740 int layer4_xor = 0;
3741
f14c4e4e
BH		 3742 if (skb->protocol == htons(ETH_P_IP)) {
3743 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3744 (iph->protocol == IPPROTO_TCP ||
3745 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3746 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3747 }
3748 return (layer4_xor ^
3749 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3750
3751 }
3752
3753 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3754}
3755
3756/*
3757 * Hash for the output device based upon layer 2 data
3758 */
3759static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3760 struct net_device *bond_dev, int count)
3761{
3762 struct ethhdr *data = (struct ethhdr *)skb->data;
3763
3764 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3765}
3766
1da177e4
LT		 3767/*-------------------------- Device entry points ----------------------------*/
3768
3769static int bond_open(struct net_device *bond_dev)
3770{
3771 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3772
3773 bond->kill_timers = 0;
3774
3775 if ((bond->params.mode == BOND_MODE_TLB) ||
3776 (bond->params.mode == BOND_MODE_ALB)) {
1da177e4
LT		 3777 /* bond_alb_initialize must be called before the timer
3778 * is started.
3779 */
3780 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3781 /* something went wrong - fail the open operation */
3782 return -1;
3783 }
3784
1b76b316
JV		 3785 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3786 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3787 }
3788
3789 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3790 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3791 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3792 }
3793
3794 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3795 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3796 INIT_DELAYED_WORK(&bond->arp_work,
3797 bond_activebackup_arp_mon);
3798 else
3799 INIT_DELAYED_WORK(&bond->arp_work,
3800 bond_loadbalance_arp_mon);
3801
3802 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3803 if (bond->params.arp_validate)
3804 bond_register_arp(bond);
1da177e4
LT		 3805 }
3806
3807 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3808 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3809 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3810 /* register to receive LACPDUs */
3811 bond_register_lacpdu(bond);
3812 }
3813
3814 return 0;
3815}
3816
3817static int bond_close(struct net_device *bond_dev)
3818{
3819 struct bonding *bond = bond_dev->priv;
3820
3821 if (bond->params.mode == BOND_MODE_8023AD) {
3822 /* Unregister the receive of LACPDUs */
3823 bond_unregister_lacpdu(bond);
3824 }
3825
f5b2b966
JV		 3826 if (bond->params.arp_validate)
3827 bond_unregister_arp(bond);
3828
1da177e4
LT		 3829 write_lock_bh(&bond->lock);
3830
b59f9f74 3831 bond->send_grat_arp = 0;
1da177e4
LT		 3832
3833 /* signal timers not to re-arm */
3834 bond->kill_timers = 1;
3835
3836 write_unlock_bh(&bond->lock);
3837
1da177e4 3838 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3839 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3840 }
3841
3842 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3843 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3844 }
3845
3846 switch (bond->params.mode) {
3847 case BOND_MODE_8023AD:
1b76b316 3848 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3849 break;
3850 case BOND_MODE_TLB:
3851 case BOND_MODE_ALB:
1b76b316 3852 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3853 break;
3854 default:
3855 break;
3856 }
3857
1da177e4
LT		 3858
3859 if ((bond->params.mode == BOND_MODE_TLB) ||
3860 (bond->params.mode == BOND_MODE_ALB)) {
3861 /* Must be called only after all
3862 * slaves have been released
3863 */
3864 bond_alb_deinitialize(bond);
3865 }
3866
3867 return 0;
3868}
3869
3870static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3871{
3872 struct bonding *bond = bond_dev->priv;
3873 struct net_device_stats *stats = &(bond->stats), *sstats;
2439f9eb 3874 struct net_device_stats local_stats;
1da177e4
LT		 3875 struct slave *slave;
3876 int i;
3877
2439f9eb 3878 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3879
3880 read_lock_bh(&bond->lock);
3881
3882 bond_for_each_slave(bond, slave, i) {
c45d286e 3883 sstats = slave->dev->get_stats(slave->dev);
2439f9eb
AG		 3884 local_stats.rx_packets += sstats->rx_packets;
3885 local_stats.rx_bytes += sstats->rx_bytes;
3886 local_stats.rx_errors += sstats->rx_errors;
3887 local_stats.rx_dropped += sstats->rx_dropped;
3888
3889 local_stats.tx_packets += sstats->tx_packets;
3890 local_stats.tx_bytes += sstats->tx_bytes;
3891 local_stats.tx_errors += sstats->tx_errors;
3892 local_stats.tx_dropped += sstats->tx_dropped;
3893
3894 local_stats.multicast += sstats->multicast;
3895 local_stats.collisions += sstats->collisions;
3896
3897 local_stats.rx_length_errors += sstats->rx_length_errors;
3898 local_stats.rx_over_errors += sstats->rx_over_errors;
3899 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3900 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3901 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3902 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3903
3904 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3905 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3906 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3907 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3908 local_stats.tx_window_errors += sstats->tx_window_errors;
3909 }
3910
3911 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3912
3913 read_unlock_bh(&bond->lock);
3914
3915 return stats;
3916}
3917
3918static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3919{
3920 struct net_device *slave_dev = NULL;
3921 struct ifbond k_binfo;
3922 struct ifbond __user *u_binfo = NULL;
3923 struct ifslave k_sinfo;
3924 struct ifslave __user *u_sinfo = NULL;
3925 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3926 int res = 0;
3927
3928 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3929 bond_dev->name, cmd);
3930
3931 switch (cmd) {
1da177e4
LT		 3932 case SIOCGMIIPHY:
3933 mii = if_mii(ifr);
3934 if (!mii) {
3935 return -EINVAL;
3936 }
3937 mii->phy_id = 0;
3938 /* Fall Through */
3939 case SIOCGMIIREG:
3940 /*
3941 * We do this again just in case we were called by SIOCGMIIREG
3942 * instead of SIOCGMIIPHY.
3943 */
3944 mii = if_mii(ifr);
3945 if (!mii) {
3946 return -EINVAL;
3947 }
3948
3949 if (mii->reg_num == 1) {
3950 struct bonding *bond = bond_dev->priv;
3951 mii->val_out = 0;
6603a6f2 3952 read_lock(&bond->lock);
1da177e4 3953 read_lock(&bond->curr_slave_lock);
4e140079 3954 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3955 mii->val_out = BMSR_LSTATUS;
3956 }
3957 read_unlock(&bond->curr_slave_lock);
6603a6f2 3958 read_unlock(&bond->lock);
1da177e4
LT		 3959 }
3960
3961 return 0;
3962 case BOND_INFO_QUERY_OLD:
3963 case SIOCBONDINFOQUERY:
3964 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3965
3966 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3967 return -EFAULT;
3968 }
3969
3970 res = bond_info_query(bond_dev, &k_binfo);
3971 if (res == 0) {
3972 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3973 return -EFAULT;
3974 }
3975 }
3976
3977 return res;
3978 case BOND_SLAVE_INFO_QUERY_OLD:
3979 case SIOCBONDSLAVEINFOQUERY:
3980 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3981
3982 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3983 return -EFAULT;
3984 }
3985
3986 res = bond_slave_info_query(bond_dev, &k_sinfo);
3987 if (res == 0) {
3988 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3989 return -EFAULT;
3990 }
3991 }
3992
3993 return res;
3994 default:
3995 /* Go on */
3996 break;
3997 }
3998
3999 if (!capable(CAP_NET_ADMIN)) {
4000 return -EPERM;
4001 }
4002
b76cdba9 4003 down_write(&(bonding_rwsem));
881d966b 4004 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT		 4005
4006 dprintk("slave_dev=%p: \n", slave_dev);
4007
4008 if (!slave_dev) {
4009 res = -ENODEV;
4010 } else {
4011 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4012 switch (cmd) {
4013 case BOND_ENSLAVE_OLD:
4014 case SIOCBONDENSLAVE:
4015 res = bond_enslave(bond_dev, slave_dev);
4016 break;
4017 case BOND_RELEASE_OLD:
4018 case SIOCBONDRELEASE:
4019 res = bond_release(bond_dev, slave_dev);
4020 break;
4021 case BOND_SETHWADDR_OLD:
4022 case SIOCBONDSETHWADDR:
4023 res = bond_sethwaddr(bond_dev, slave_dev);
4024 break;
4025 case BOND_CHANGE_ACTIVE_OLD:
4026 case SIOCBONDCHANGEACTIVE:
4027 res = bond_ioctl_change_active(bond_dev, slave_dev);
4028 break;
4029 default:
4030 res = -EOPNOTSUPP;
4031 }
4032
4033 dev_put(slave_dev);
4034 }
4035
b76cdba9 4036 up_write(&(bonding_rwsem));
1da177e4
LT		 4037 return res;
4038}
4039
4040static void bond_set_multicast_list(struct net_device *bond_dev)
4041{
4042 struct bonding *bond = bond_dev->priv;
4043 struct dev_mc_list *dmi;
4044
1da177e4
LT		 4045 /*
4046 * Do promisc before checking multicast_mode
4047 */
4048 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4049 /*
4050 * FIXME: Need to handle the error when one of the multi-slaves
4051 * encounters error.
4052 */
1da177e4
LT		 4053 bond_set_promiscuity(bond, 1);
4054 }
4055
4056 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4057 bond_set_promiscuity(bond, -1);
4058 }
4059
4060 /* set allmulti flag to slaves */
4061 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4062 /*
4063 * FIXME: Need to handle the error when one of the multi-slaves
4064 * encounters error.
4065 */
1da177e4
LT		 4066 bond_set_allmulti(bond, 1);
4067 }
4068
4069 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4070 bond_set_allmulti(bond, -1);
4071 }
4072
80ee5ad2
JV		 4073 read_lock(&bond->lock);
4074
1da177e4
LT		 4075 bond->flags = bond_dev->flags;
4076
4077 /* looking for addresses to add to slaves' mc list */
4078 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4079 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4080 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4081 }
4082 }
4083
4084 /* looking for addresses to delete from slaves' list */
4085 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4086 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4087 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4088 }
4089 }
4090
4091 /* save master's multicast list */
4092 bond_mc_list_destroy(bond);
4093 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4094
80ee5ad2 4095 read_unlock(&bond->lock);
1da177e4
LT		 4096}
4097
4098/*
4099 * Change the MTU of all of a master's slaves to match the master
4100 */
4101static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4102{
4103 struct bonding *bond = bond_dev->priv;
4104 struct slave *slave, *stop_at;
4105 int res = 0;
4106 int i;
4107
4108 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4109 (bond_dev ? bond_dev->name : "None"), new_mtu);
4110
4111 /* Can't hold bond->lock with bh disabled here since
4112 * some base drivers panic. On the other hand we can't
4113 * hold bond->lock without bh disabled because we'll
4114 * deadlock. The only solution is to rely on the fact
4115 * that we're under rtnl_lock here, and the slaves
4116 * list won't change. This doesn't solve the problem
4117 * of setting the slave's MTU while it is
4118 * transmitting, but the assumption is that the base
4119 * driver can handle that.
4120 *
4121 * TODO: figure out a way to safely iterate the slaves
4122 * list, but without holding a lock around the actual
4123 * call to the base driver.
4124 */
4125
4126 bond_for_each_slave(bond, slave, i) {
4127 dprintk("s %p s->p %p c_m %p\n", slave,
4128 slave->prev, slave->dev->change_mtu);
e944ef79 4129
1da177e4
LT		 4130 res = dev_set_mtu(slave->dev, new_mtu);
4131
4132 if (res) {
4133 /* If we failed to set the slave's mtu to the new value
4134 * we must abort the operation even in ACTIVE_BACKUP
4135 * mode, because if we allow the backup slaves to have
4136 * different mtu values than the active slave we'll
4137 * need to change their mtu when doing a failover. That
4138 * means changing their mtu from timer context, which
4139 * is probably not a good idea.
4140 */
4141 dprintk("err %d %s\n", res, slave->dev->name);
4142 goto unwind;
4143 }
4144 }
4145
4146 bond_dev->mtu = new_mtu;
4147
4148 return 0;
4149
4150unwind:
4151 /* unwind from head to the slave that failed */
4152 stop_at = slave;
4153 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4154 int tmp_res;
4155
4156 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4157 if (tmp_res) {
4158 dprintk("unwind err %d dev %s\n", tmp_res,
4159 slave->dev->name);
4160 }
4161 }
4162
4163 return res;
4164}
4165
4166/*
4167 * Change HW address
4168 *
4169 * Note that many devices must be down to change the HW address, and
4170 * downing the master releases all slaves. We can make bonds full of
4171 * bonding devices to test this, however.
4172 */
4173static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4174{
4175 struct bonding *bond = bond_dev->priv;
4176 struct sockaddr *sa = addr, tmp_sa;
4177 struct slave *slave, *stop_at;
4178 int res = 0;
4179 int i;
4180
4181 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4182
dd957c57 4183 /*
3915c1e8
JV		 4184 * If fail_over_mac is set to active, do nothing and return
4185 * success. Returning an error causes ifenslave to fail.
dd957c57 4186 */
3915c1e8 4187 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4188 return 0;
2ab82852 4189
1da177e4
LT		 4190 if (!is_valid_ether_addr(sa->sa_data)) {
4191 return -EADDRNOTAVAIL;
4192 }
4193
4194 /* Can't hold bond->lock with bh disabled here since
4195 * some base drivers panic. On the other hand we can't
4196 * hold bond->lock without bh disabled because we'll
4197 * deadlock. The only solution is to rely on the fact
4198 * that we're under rtnl_lock here, and the slaves
4199 * list won't change. This doesn't solve the problem
4200 * of setting the slave's hw address while it is
4201 * transmitting, but the assumption is that the base
4202 * driver can handle that.
4203 *
4204 * TODO: figure out a way to safely iterate the slaves
4205 * list, but without holding a lock around the actual
4206 * call to the base driver.
4207 */
4208
4209 bond_for_each_slave(bond, slave, i) {
4210 dprintk("slave %p %s\n", slave, slave->dev->name);
4211
4212 if (slave->dev->set_mac_address == NULL) {
4213 res = -EOPNOTSUPP;
4214 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4215 goto unwind;
4216 }
4217
4218 res = dev_set_mac_address(slave->dev, addr);
4219 if (res) {
4220 /* TODO: consider downing the slave
4221 * and retry ?
4222 * User should expect communications
4223 * breakage anyway until ARP finish
4224 * updating, so...
4225 */
4226 dprintk("err %d %s\n", res, slave->dev->name);
4227 goto unwind;
4228 }
4229 }
4230
4231 /* success */
4232 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4233 return 0;
4234
4235unwind:
4236 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4237 tmp_sa.sa_family = bond_dev->type;
4238
4239 /* unwind from head to the slave that failed */
4240 stop_at = slave;
4241 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4242 int tmp_res;
4243
4244 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4245 if (tmp_res) {
4246 dprintk("unwind err %d dev %s\n", tmp_res,
4247 slave->dev->name);
4248 }
4249 }
4250
4251 return res;
4252}
4253
4254static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4255{
4256 struct bonding *bond = bond_dev->priv;
4257 struct slave *slave, *start_at;
cf5f9044 4258 int i, slave_no, res = 1;
1da177e4
LT		 4259
4260 read_lock(&bond->lock);
4261
4262 if (!BOND_IS_OK(bond)) {
4263 goto out;
4264 }
4265
cf5f9044
JV		 4266 /*
4267 * Concurrent TX may collide on rr_tx_counter; we accept that
4268 * as being rare enough not to justify using an atomic op here
4269 */
4270 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4271
cf5f9044
JV		 4272 bond_for_each_slave(bond, slave, i) {
4273 slave_no--;
4274 if (slave_no < 0) {
4275 break;
4276 }
1da177e4
LT		 4277 }
4278
cf5f9044 4279 start_at = slave;
1da177e4
LT		 4280 bond_for_each_slave_from(bond, slave, i, start_at) {
4281 if (IS_UP(slave->dev) &&
4282 (slave->link == BOND_LINK_UP) &&
4283 (slave->state == BOND_STATE_ACTIVE)) {
4284 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4285 break;
4286 }
4287 }
4288
1da177e4
LT		 4289out:
4290 if (res) {
4291 /* no suitable interface, frame not sent */
4292 dev_kfree_skb(skb);
4293 }
4294 read_unlock(&bond->lock);
4295 return 0;
4296}
4297
075897ce 4298
1da177e4
LT		 4299/*
4300 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4301 * the bond has a usable interface.
4302 */
4303static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4304{
4305 struct bonding *bond = bond_dev->priv;
4306 int res = 1;
4307
1da177e4
LT		 4308 read_lock(&bond->lock);
4309 read_lock(&bond->curr_slave_lock);
4310
4311 if (!BOND_IS_OK(bond)) {
4312 goto out;
4313 }
4314
075897ce
JL		 4315 if (!bond->curr_active_slave)
4316 goto out;
4317
075897ce
JL		 4318 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4319
1da177e4
LT		 4320out:
4321 if (res) {
4322 /* no suitable interface, frame not sent */
4323 dev_kfree_skb(skb);
4324 }
4325 read_unlock(&bond->curr_slave_lock);
4326 read_unlock(&bond->lock);
4327 return 0;
4328}
4329
4330/*
169a3e66
JV		 4331 * In bond_xmit_xor() , we determine the output device by using a pre-
4332 * determined xmit_hash_policy(), If the selected device is not enabled,
4333 * find the next active slave.
1da177e4
LT		 4334 */
4335static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4336{
4337 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 4338 struct slave *slave, *start_at;
4339 int slave_no;
4340 int i;
4341 int res = 1;
4342
4343 read_lock(&bond->lock);
4344
4345 if (!BOND_IS_OK(bond)) {
4346 goto out;
4347 }
4348
169a3e66 4349 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4350
4351 bond_for_each_slave(bond, slave, i) {
4352 slave_no--;
4353 if (slave_no < 0) {
4354 break;
4355 }
4356 }
4357
4358 start_at = slave;
4359
4360 bond_for_each_slave_from(bond, slave, i, start_at) {
4361 if (IS_UP(slave->dev) &&
4362 (slave->link == BOND_LINK_UP) &&
4363 (slave->state == BOND_STATE_ACTIVE)) {
4364 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4365 break;
4366 }
4367 }
4368
4369out:
4370 if (res) {
4371 /* no suitable interface, frame not sent */
4372 dev_kfree_skb(skb);
4373 }
4374 read_unlock(&bond->lock);
4375 return 0;
4376}
4377
4378/*
4379 * in broadcast mode, we send everything to all usable interfaces.
4380 */
4381static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4382{
4383 struct bonding *bond = bond_dev->priv;
4384 struct slave *slave, *start_at;
4385 struct net_device *tx_dev = NULL;
4386 int i;
4387 int res = 1;
4388
4389 read_lock(&bond->lock);
4390
4391 if (!BOND_IS_OK(bond)) {
4392 goto out;
4393 }
4394
4395 read_lock(&bond->curr_slave_lock);
4396 start_at = bond->curr_active_slave;
4397 read_unlock(&bond->curr_slave_lock);
4398
4399 if (!start_at) {
4400 goto out;
4401 }
4402
4403 bond_for_each_slave_from(bond, slave, i, start_at) {
4404 if (IS_UP(slave->dev) &&
4405 (slave->link == BOND_LINK_UP) &&
4406 (slave->state == BOND_STATE_ACTIVE)) {
4407 if (tx_dev) {
4408 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4409 if (!skb2) {
4410 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4411 ": %s: Error: bond_xmit_broadcast(): "
4412 "skb_clone() failed\n",
4413 bond_dev->name);
1da177e4
LT		 4414 continue;
4415 }
4416
4417 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4418 if (res) {
4419 dev_kfree_skb(skb2);
4420 continue;
4421 }
4422 }
4423 tx_dev = slave->dev;
4424 }
4425 }
4426
4427 if (tx_dev) {
4428 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4429 }
4430
4431out:
4432 if (res) {
4433 /* no suitable interface, frame not sent */
4434 dev_kfree_skb(skb);
4435 }
4436 /* frame sent to all suitable interfaces */
4437 read_unlock(&bond->lock);
4438 return 0;
4439}
4440
4441/*------------------------- Device initialization ---------------------------*/
4442
6f6652be
JV		 4443static void bond_set_xmit_hash_policy(struct bonding *bond)
4444{
4445 switch (bond->params.xmit_policy) {
4446 case BOND_XMIT_POLICY_LAYER23:
4447 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4448 break;
4449 case BOND_XMIT_POLICY_LAYER34:
4450 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4451 break;
4452 case BOND_XMIT_POLICY_LAYER2:
4453 default:
4454 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4455 break;
4456 }
4457}
4458
1da177e4
LT		 4459/*
4460 * set bond mode specific net device operations
4461 */
a77b5325 4462void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4463{
169a3e66
JV		 4464 struct net_device *bond_dev = bond->dev;
4465
1da177e4
LT		 4466 switch (mode) {
4467 case BOND_MODE_ROUNDROBIN:
4468 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4469 break;
4470 case BOND_MODE_ACTIVEBACKUP:
4471 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4472 break;
4473 case BOND_MODE_XOR:
4474 bond_dev->hard_start_xmit = bond_xmit_xor;
6f6652be 4475 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4476 break;
4477 case BOND_MODE_BROADCAST:
4478 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4479 break;
4480 case BOND_MODE_8023AD:
8f903c70 4481 bond_set_master_3ad_flags(bond);
1da177e4 4482 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
6f6652be 4483 bond_set_xmit_hash_policy(bond);
1da177e4 4484 break;
1da177e4 4485 case BOND_MODE_ALB:
8f903c70
JV		 4486 bond_set_master_alb_flags(bond);
4487 /* FALLTHRU */
4488 case BOND_MODE_TLB:
1da177e4
LT		 4489 bond_dev->hard_start_xmit = bond_alb_xmit;
4490 bond_dev->set_mac_address = bond_alb_set_mac_address;
4491 break;
4492 default:
4493 /* Should never happen, mode already checked */
4494 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4495 ": %s: Error: Unknown bonding mode %d\n",
4496 bond_dev->name,
1da177e4
LT		 4497 mode);
4498 break;
4499 }
4500}
4501
217df670
JV		 4502static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4503 struct ethtool_drvinfo *drvinfo)
4504{
4505 strncpy(drvinfo->driver, DRV_NAME, 32);
4506 strncpy(drvinfo->version, DRV_VERSION, 32);
4507 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4508}
4509
7282d491 4510static const struct ethtool_ops bond_ethtool_ops = {
217df670 4511 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac
SH		 4512 .get_link = ethtool_op_get_link,
4513 .get_tx_csum = ethtool_op_get_tx_csum,
4514 .get_sg = ethtool_op_get_sg,
4515 .get_tso = ethtool_op_get_tso,
4516 .get_ufo = ethtool_op_get_ufo,
4517 .get_flags = ethtool_op_get_flags,
8531c5ff
AK		 4518};
4519
1da177e4
LT		 4520/*
4521 * Does not allocate but creates a /proc entry.
4522 * Allowed to fail.
4523 */
3c535952 4524static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4
LT		 4525{
4526 struct bonding *bond = bond_dev->priv;
4527
4528 dprintk("Begin bond_init for %s\n", bond_dev->name);
4529
4530 /* initialize rwlocks */
4531 rwlock_init(&bond->lock);
4532 rwlock_init(&bond->curr_slave_lock);
4533
4534 bond->params = *params; /* copy params struct */
4535
1b76b316
JV		 4536 bond->wq = create_singlethread_workqueue(bond_dev->name);
4537 if (!bond->wq)
4538 return -ENOMEM;
4539
1da177e4
LT		 4540 /* Initialize pointers */
4541 bond->first_slave = NULL;
4542 bond->curr_active_slave = NULL;
4543 bond->current_arp_slave = NULL;
4544 bond->primary_slave = NULL;
4545 bond->dev = bond_dev;
1053f62c 4546 bond->send_grat_arp = 0;
d90a162a 4547 bond->setup_by_slave = 0;
1da177e4
LT		 4548 INIT_LIST_HEAD(&bond->vlan_list);
4549
4550 /* Initialize the device entry points */
4551 bond_dev->open = bond_open;
4552 bond_dev->stop = bond_close;
4553 bond_dev->get_stats = bond_get_stats;
4554 bond_dev->do_ioctl = bond_do_ioctl;
8531c5ff 4555 bond_dev->ethtool_ops = &bond_ethtool_ops;
1da177e4
LT		 4556 bond_dev->set_multicast_list = bond_set_multicast_list;
4557 bond_dev->change_mtu = bond_change_mtu;
4558 bond_dev->set_mac_address = bond_set_mac_address;
3a1521b7 4559 bond_dev->validate_addr = NULL;
1da177e4 4560
169a3e66 4561 bond_set_mode_ops(bond, bond->params.mode);
1da177e4 4562
a434e43f 4563 bond_dev->destructor = bond_destructor;
1da177e4
LT		 4564
4565 /* Initialize the device options */
4566 bond_dev->tx_queue_len = 0;
4567 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4568 bond_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 4569
4570 /* At first, we block adding VLANs. That's the only way to
4571 * prevent problems that occur when adding VLANs over an
4572 * empty bond. The block will be removed once non-challenged
4573 * slaves are enslaved.
4574 */
4575 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4576
932ff279 4577 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4578 * transmitting */
4579 bond_dev->features |= NETIF_F_LLTX;
4580
4581 /* By default, we declare the bond to be fully
4582 * VLAN hardware accelerated capable. Special
4583 * care is taken in the various xmit functions
4584 * when there are slaves that are not hw accel
4585 * capable
4586 */
4587 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4588 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4589 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4590 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4591 NETIF_F_HW_VLAN_RX |
4592 NETIF_F_HW_VLAN_FILTER);
4593
4594#ifdef CONFIG_PROC_FS
4595 bond_create_proc_entry(bond);
4596#endif
1da177e4
LT		 4597 list_add_tail(&bond->bond_list, &bond_dev_list);
4598
4599 return 0;
4600}
4601
fdaea7a9
JV		 4602static void bond_work_cancel_all(struct bonding *bond)
4603{
4604 write_lock_bh(&bond->lock);
4605 bond->kill_timers = 1;
4606 write_unlock_bh(&bond->lock);
4607
4608 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4609 cancel_delayed_work(&bond->mii_work);
4610
4611 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4612 cancel_delayed_work(&bond->arp_work);
4613
4614 if (bond->params.mode == BOND_MODE_ALB &&
4615 delayed_work_pending(&bond->alb_work))
4616 cancel_delayed_work(&bond->alb_work);
4617
4618 if (bond->params.mode == BOND_MODE_8023AD &&
4619 delayed_work_pending(&bond->ad_work))
4620 cancel_delayed_work(&bond->ad_work);
4621}
4622
a434e43f
JV		 4623/* De-initialize device specific data.
4624 * Caller must hold rtnl_lock.
4625 */
4626static void bond_deinit(struct net_device *bond_dev)
4627{
4628 struct bonding *bond = bond_dev->priv;
4629
4630 list_del(&bond->bond_list);
4631
4632 bond_work_cancel_all(bond);
4633
4634#ifdef CONFIG_PROC_FS
4635 bond_remove_proc_entry(bond);
4636#endif
4637}
4638
1da177e4
LT		 4639/* Unregister and free all bond devices.
4640 * Caller must hold rtnl_lock.
4641 */
4642static void bond_free_all(void)
4643{
4644 struct bonding *bond, *nxt;
4645
4646 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4647 struct net_device *bond_dev = bond->dev;
4648
fdaea7a9 4649 bond_work_cancel_all(bond);
70298705 4650 /* Release the bonded slaves */
4651 bond_release_all(bond_dev);
92b41daa 4652 bond_destroy(bond);
1da177e4
LT		 4653 }
4654
4655#ifdef CONFIG_PROC_FS
4656 bond_destroy_proc_dir();
4657#endif
4658}
4659
4660/*------------------------- Module initialization ---------------------------*/
4661
4662/*
4663 * Convert string input module parms. Accept either the
ece95f7f
JV		 4664 * number of the mode or its string name. A bit complicated because
4665 * some mode names are substrings of other names, and calls from sysfs
4666 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4667 */
ece95f7f 4668int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4669{
ece95f7f 4670 int mode = -1, i, rv;
a42e534f 4671 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4672
a42e534f
JV		 4673 for (p = (char *)buf; *p; p++)
4674 if (!(isdigit(*p) || isspace(*p)))
4675 break;
4676
4677 if (*p)
ece95f7f 4678 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4679 else
4680 rv = sscanf(buf, "%d", &mode);
4681
4682 if (!rv)
4683 return -1;
1da177e4
LT		 4684
4685 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4686 if (mode == tbl[i].mode)
4687 return tbl[i].mode;
4688 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4689 return tbl[i].mode;
1da177e4
LT		 4690 }
4691
4692 return -1;
4693}
4694
4695static int bond_check_params(struct bond_params *params)
4696{
3915c1e8 4697 int arp_validate_value, fail_over_mac_value;
f5b2b966 4698
1da177e4
LT		 4699 /*
4700 * Convert string parameters.
4701 */
4702 if (mode) {
4703 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4704 if (bond_mode == -1) {
4705 printk(KERN_ERR DRV_NAME
4706 ": Error: Invalid bonding mode \"%s\"\n",
4707 mode == NULL ? "NULL" : mode);
4708 return -EINVAL;
4709 }
4710 }
4711
169a3e66
JV		 4712 if (xmit_hash_policy) {
4713 if ((bond_mode != BOND_MODE_XOR) &&
4714 (bond_mode != BOND_MODE_8023AD)) {
4715 printk(KERN_INFO DRV_NAME
4716 ": xor_mode param is irrelevant in mode %s\n",
4717 bond_mode_name(bond_mode));
4718 } else {
4719 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4720 xmit_hashtype_tbl);
4721 if (xmit_hashtype == -1) {
4722 printk(KERN_ERR DRV_NAME
4723 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4724 xmit_hash_policy == NULL ? "NULL" :
4725 xmit_hash_policy);
4726 return -EINVAL;
4727 }
4728 }
4729 }
4730
1da177e4
LT		 4731 if (lacp_rate) {
4732 if (bond_mode != BOND_MODE_8023AD) {
4733 printk(KERN_INFO DRV_NAME
4734 ": lacp_rate param is irrelevant in mode %s\n",
4735 bond_mode_name(bond_mode));
4736 } else {
4737 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4738 if (lacp_fast == -1) {
4739 printk(KERN_ERR DRV_NAME
4740 ": Error: Invalid lacp rate \"%s\"\n",
4741 lacp_rate == NULL ? "NULL" : lacp_rate);
4742 return -EINVAL;
4743 }
4744 }
4745 }
4746
b8a9787e 4747 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4748 printk(KERN_WARNING DRV_NAME
4749 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4750 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4751 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4752 max_bonds = BOND_DEFAULT_MAX_BONDS;
4753 }
4754
4755 if (miimon < 0) {
4756 printk(KERN_WARNING DRV_NAME
4757 ": Warning: miimon module parameter (%d), "
4758 "not in range 0-%d, so it was reset to %d\n",
4759 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4760 miimon = BOND_LINK_MON_INTERV;
4761 }
4762
4763 if (updelay < 0) {
4764 printk(KERN_WARNING DRV_NAME
4765 ": Warning: updelay module parameter (%d), "
4766 "not in range 0-%d, so it was reset to 0\n",
4767 updelay, INT_MAX);
4768 updelay = 0;
4769 }
4770
4771 if (downdelay < 0) {
4772 printk(KERN_WARNING DRV_NAME
4773 ": Warning: downdelay module parameter (%d), "
4774 "not in range 0-%d, so it was reset to 0\n",
4775 downdelay, INT_MAX);
4776 downdelay = 0;
4777 }
4778
4779 if ((use_carrier != 0) && (use_carrier != 1)) {
4780 printk(KERN_WARNING DRV_NAME
4781 ": Warning: use_carrier module parameter (%d), "
4782 "not of valid value (0/1), so it was set to 1\n",
4783 use_carrier);
4784 use_carrier = 1;
4785 }
4786
7893b249
MS		 4787 if (num_grat_arp < 0 || num_grat_arp > 255) {
4788 printk(KERN_WARNING DRV_NAME
4789 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4790 "was reset to 1 \n", num_grat_arp);
4791 num_grat_arp = 1;
4792 }
4793
1da177e4
LT		 4794 /* reset values for 802.3ad */
4795 if (bond_mode == BOND_MODE_8023AD) {
4796 if (!miimon) {
4797 printk(KERN_WARNING DRV_NAME
4798 ": Warning: miimon must be specified, "
4799 "otherwise bonding will not detect link "
4800 "failure, speed and duplex which are "
4801 "essential for 802.3ad operation\n");
4802 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4803 miimon = 100;
4804 }
4805 }
4806
4807 /* reset values for TLB/ALB */
4808 if ((bond_mode == BOND_MODE_TLB) ||
4809 (bond_mode == BOND_MODE_ALB)) {
4810 if (!miimon) {
4811 printk(KERN_WARNING DRV_NAME
4812 ": Warning: miimon must be specified, "
4813 "otherwise bonding will not detect link "
4814 "failure and link speed which are essential "
4815 "for TLB/ALB load balancing\n");
4816 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4817 miimon = 100;
4818 }
4819 }
4820
4821 if (bond_mode == BOND_MODE_ALB) {
4822 printk(KERN_NOTICE DRV_NAME
4823 ": In ALB mode you might experience client "
4824 "disconnections upon reconnection of a link if the "
4825 "bonding module updelay parameter (%d msec) is "
4826 "incompatible with the forwarding delay time of the "
4827 "switch\n",
4828 updelay);
4829 }
4830
4831 if (!miimon) {
4832 if (updelay || downdelay) {
4833 /* just warn the user the up/down delay will have
4834 * no effect since miimon is zero...
4835 */
4836 printk(KERN_WARNING DRV_NAME
4837 ": Warning: miimon module parameter not set "
4838 "and updelay (%d) or downdelay (%d) module "
4839 "parameter is set; updelay and downdelay have "
4840 "no effect unless miimon is set\n",
4841 updelay, downdelay);
4842 }
4843 } else {
4844 /* don't allow arp monitoring */
4845 if (arp_interval) {
4846 printk(KERN_WARNING DRV_NAME
4847 ": Warning: miimon (%d) and arp_interval (%d) "
4848 "can't be used simultaneously, disabling ARP "
4849 "monitoring\n",
4850 miimon, arp_interval);
4851 arp_interval = 0;
4852 }
4853
4854 if ((updelay % miimon) != 0) {
4855 printk(KERN_WARNING DRV_NAME
4856 ": Warning: updelay (%d) is not a multiple "
4857 "of miimon (%d), updelay rounded to %d ms\n",
4858 updelay, miimon, (updelay / miimon) * miimon);
4859 }
4860
4861 updelay /= miimon;
4862
4863 if ((downdelay % miimon) != 0) {
4864 printk(KERN_WARNING DRV_NAME
4865 ": Warning: downdelay (%d) is not a multiple "
4866 "of miimon (%d), downdelay rounded to %d ms\n",
4867 downdelay, miimon,
4868 (downdelay / miimon) * miimon);
4869 }
4870
4871 downdelay /= miimon;
4872 }
4873
4874 if (arp_interval < 0) {
4875 printk(KERN_WARNING DRV_NAME
4876 ": Warning: arp_interval module parameter (%d) "
4877 ", not in range 0-%d, so it was reset to %d\n",
4878 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4879 arp_interval = BOND_LINK_ARP_INTERV;
4880 }
4881
4882 for (arp_ip_count = 0;
4883 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4884 arp_ip_count++) {
4885 /* not complete check, but should be good enough to
4886 catch mistakes */
4887 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4888 printk(KERN_WARNING DRV_NAME
4889 ": Warning: bad arp_ip_target module parameter "
4890 "(%s), ARP monitoring will not be performed\n",
4891 arp_ip_target[arp_ip_count]);
4892 arp_interval = 0;
4893 } else {
d3bb52b0 4894 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4895 arp_target[arp_ip_count] = ip;
4896 }
4897 }
4898
4899 if (arp_interval && !arp_ip_count) {
4900 /* don't allow arping if no arp_ip_target given... */
4901 printk(KERN_WARNING DRV_NAME
4902 ": Warning: arp_interval module parameter (%d) "
4903 "specified without providing an arp_ip_target "
4904 "parameter, arp_interval was reset to 0\n",
4905 arp_interval);
4906 arp_interval = 0;
4907 }
4908
f5b2b966
JV		 4909 if (arp_validate) {
4910 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4911 printk(KERN_ERR DRV_NAME
4912 ": arp_validate only supported in active-backup mode\n");
4913 return -EINVAL;
4914 }
4915 if (!arp_interval) {
4916 printk(KERN_ERR DRV_NAME
4917 ": arp_validate requires arp_interval\n");
4918 return -EINVAL;
4919 }
4920
4921 arp_validate_value = bond_parse_parm(arp_validate,
4922 arp_validate_tbl);
4923 if (arp_validate_value == -1) {
4924 printk(KERN_ERR DRV_NAME
4925 ": Error: invalid arp_validate \"%s\"\n",
4926 arp_validate == NULL ? "NULL" : arp_validate);
4927 return -EINVAL;
4928 }
4929 } else
4930 arp_validate_value = 0;
4931
1da177e4
LT		 4932 if (miimon) {
4933 printk(KERN_INFO DRV_NAME
4934 ": MII link monitoring set to %d ms\n",
4935 miimon);
4936 } else if (arp_interval) {
4937 int i;
4938
4939 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 4940 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4941 arp_interval,
4942 arp_validate_tbl[arp_validate_value].modename,
4943 arp_ip_count);
1da177e4
LT		 4944
4945 for (i = 0; i < arp_ip_count; i++)
4946 printk (" %s", arp_ip_target[i]);
4947
4948 printk("\n");
4949
b8a9787e 4950 } else if (max_bonds) {
1da177e4
LT		 4951 /* miimon and arp_interval not set, we need one so things
4952 * work as expected, see bonding.txt for details
4953 */
4954 printk(KERN_WARNING DRV_NAME
4955 ": Warning: either miimon or arp_interval and "
4956 "arp_ip_target module parameters must be specified, "
4957 "otherwise bonding will not detect link failures! see "
4958 "bonding.txt for details.\n");
4959 }
4960
4961 if (primary && !USES_PRIMARY(bond_mode)) {
4962 /* currently, using a primary only makes sense
4963 * in active backup, TLB or ALB modes
4964 */
4965 printk(KERN_WARNING DRV_NAME
4966 ": Warning: %s primary device specified but has no "
4967 "effect in %s mode\n",
4968 primary, bond_mode_name(bond_mode));
4969 primary = NULL;
4970 }
4971
3915c1e8
JV		 4972 if (fail_over_mac) {
4973 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4974 fail_over_mac_tbl);
4975 if (fail_over_mac_value == -1) {
4976 printk(KERN_ERR DRV_NAME
4977 ": Error: invalid fail_over_mac \"%s\"\n",
4978 arp_validate == NULL ? "NULL" : arp_validate);
4979 return -EINVAL;
4980 }
4981
4982 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4983 printk(KERN_WARNING DRV_NAME
4984 ": Warning: fail_over_mac only affects "
4985 "active-backup mode.\n");
4986 } else {
4987 fail_over_mac_value = BOND_FOM_NONE;
4988 }
dd957c57 4989
1da177e4
LT		 4990 /* fill params struct with the proper values */
4991 params->mode = bond_mode;
169a3e66 4992 params->xmit_policy = xmit_hashtype;
1da177e4 4993 params->miimon = miimon;
7893b249 4994 params->num_grat_arp = num_grat_arp;
1da177e4 4995 params->arp_interval = arp_interval;
f5b2b966 4996 params->arp_validate = arp_validate_value;
1da177e4
LT		 4997 params->updelay = updelay;
4998 params->downdelay = downdelay;
4999 params->use_carrier = use_carrier;
5000 params->lacp_fast = lacp_fast;
5001 params->primary[0] = 0;
3915c1e8 5002 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5003
5004 if (primary) {
5005 strncpy(params->primary, primary, IFNAMSIZ);
5006 params->primary[IFNAMSIZ - 1] = 0;
5007 }
5008
5009 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5010
5011 return 0;
5012}
5013
0daa2303 5014static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 5015static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 5016
e8a0464c
DM		 5017static void bond_set_lockdep_class_one(struct net_device *dev,
5018 struct netdev_queue *txq,
5019 void *_unused)
c773e847
DM		 5020{
5021 lockdep_set_class(&txq->_xmit_lock,
5022 &bonding_netdev_xmit_lock_key);
5023}
5024
5025static void bond_set_lockdep_class(struct net_device *dev)
5026{
cf508b12
DM		 5027 lockdep_set_class(&dev->addr_list_lock,
5028 &bonding_netdev_addr_lock_key);
e8a0464c 5029 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 5030}
5031
dfe60397 5032/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5033 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5034 * Caller must NOT hold rtnl_lock; we need to release it here before we
5035 * set up our sysfs entries.
5036 */
0dd646fe 5037int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5038{
5039 struct net_device *bond_dev;
0883beca 5040 struct bonding *bond;
dfe60397
MW		 5041 int res;
5042
5043 rtnl_lock();
027ea041
JV		 5044 down_write(&bonding_rwsem);
5045
5046 /* Check to see if the bond already exists. */
4fe4763c 5047 if (name) {
0883beca 5048 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5049 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5050 printk(KERN_ERR DRV_NAME
027ea041 5051 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5052 name);
5053 res = -EPERM;
5054 goto out_rtnl;
5055 }
5056 }
027ea041 5057
e4b91c48
JV		 5058 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5059 ether_setup);
dfe60397
MW		 5060 if (!bond_dev) {
5061 printk(KERN_ERR DRV_NAME
5062 ": %s: eek! can't alloc netdev!\n",
5063 name);
5064 res = -ENOMEM;
5065 goto out_rtnl;
5066 }
5067
e4b91c48
JV		 5068 if (!name) {
5069 res = dev_alloc_name(bond_dev, "bond%d");
5070 if (res < 0)
5071 goto out_netdev;
5072 }
5073
dfe60397
MW		 5074 /* bond_init() must be called after dev_alloc_name() (for the
5075 * /proc files), but before register_netdevice(), because we
5076 * need to set function pointers.
5077 */
5078
5079 res = bond_init(bond_dev, params);
5080 if (res < 0) {
5081 goto out_netdev;
5082 }
5083
dfe60397
MW		 5084 res = register_netdevice(bond_dev);
5085 if (res < 0) {
5086 goto out_bond;
5087 }
0daa2303 5088
c773e847 5089 bond_set_lockdep_class(bond_dev);
0daa2303 5090
ff59c456
JV		 5091 netif_carrier_off(bond_dev);
5092
027ea041 5093 up_write(&bonding_rwsem);
dfe60397 5094 rtnl_unlock(); /* allows sysfs registration of net device */
b76cdba9 5095 res = bond_create_sysfs_entry(bond_dev->priv);
09c89279
JV		 5096 if (res < 0) {
5097 rtnl_lock();
027ea041 5098 down_write(&bonding_rwsem);
822973ba
PE		 5099 bond_deinit(bond_dev);
5100 unregister_netdevice(bond_dev);
5101 goto out_rtnl;
09c89279
JV		 5102 }
5103
5104 return 0;
5105
dfe60397
MW		 5106out_bond:
5107 bond_deinit(bond_dev);
5108out_netdev:
5109 free_netdev(bond_dev);
5110out_rtnl:
027ea041 5111 up_write(&bonding_rwsem);
dfe60397 5112 rtnl_unlock();
dfe60397
MW		 5113 return res;
5114}
5115
1da177e4
LT		 5116static int __init bonding_init(void)
5117{
1da177e4
LT		 5118 int i;
5119 int res;
0883beca 5120 struct bonding *bond;
1da177e4
LT		 5121
5122 printk(KERN_INFO "%s", version);
5123
dfe60397 5124 res = bond_check_params(&bonding_defaults);
1da177e4 5125 if (res) {
dfe60397 5126 goto out;
1da177e4
LT		 5127 }
5128
1da177e4
LT		 5129#ifdef CONFIG_PROC_FS
5130 bond_create_proc_dir();
5131#endif
027ea041
JV		 5132
5133 init_rwsem(&bonding_rwsem);
5134
1da177e4 5135 for (i = 0; i < max_bonds; i++) {
0dd646fe 5136 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5137 if (res)
5138 goto err;
1da177e4
LT		 5139 }
5140
b76cdba9
MW		 5141 res = bond_create_sysfs();
5142 if (res)
5143 goto err;
5144
1da177e4 5145 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5146 register_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5147
dfe60397
MW		 5148 goto out;
5149err:
0883beca 5150 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5151 bond_work_cancel_all(bond);
5152 destroy_workqueue(bond->wq);
5153 }
5154
ae68c398
PE		 5155 bond_destroy_sysfs();
5156
40abc270 5157 rtnl_lock();
1da177e4 5158 bond_free_all();
1da177e4 5159 rtnl_unlock();
dfe60397 5160out:
1da177e4 5161 return res;
dfe60397 5162
1da177e4
LT		 5163}
5164
5165static void __exit bonding_exit(void)
5166{
5167 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5168 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5169
ae68c398
PE		 5170 bond_destroy_sysfs();
5171
1da177e4
LT		 5172 rtnl_lock();
5173 bond_free_all();
5174 rtnl_unlock();
5175}
5176
5177module_init(bonding_init);
5178module_exit(bonding_exit);
5179MODULE_LICENSE("GPL");
5180MODULE_VERSION(DRV_VERSION);
5181MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5182MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5183MODULE_SUPPORTED_DEVICE("most ethernet devices");
5184
5185/*
5186 * Local variables:
5187 * c-indent-level: 8
5188 * c-basic-offset: 8
5189 * tab-width: 8
5190 * End:
5191 */
5192
Clone of davem's net-next-2.6 tree