]> 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
macvlan: add support for ethtool get settings
[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
1981/*
1982* First release a slave and than destroy the bond if no more slaves iare left.
1983* Must be under rtnl_lock when this function is called.
1984*/
1985int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1986{
1987 struct bonding *bond = bond_dev->priv;
1988 int ret;
1989
1990 ret = bond_release(bond_dev, slave_dev);
1991 if ((ret == 0) && (bond->slave_cnt == 0)) {
1992 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1993 bond_dev->name, bond_dev->name);
1994 bond_destroy(bond);
1995 }
1996 return ret;
1997}
1998
1da177e4
LT		 1999/*
2000 * This function releases all slaves.
2001 */
2002static int bond_release_all(struct net_device *bond_dev)
2003{
2004 struct bonding *bond = bond_dev->priv;
2005 struct slave *slave;
2006 struct net_device *slave_dev;
2007 struct sockaddr addr;
2008
2009 write_lock_bh(&bond->lock);
2010
ff59c456
JV		 2011 netif_carrier_off(bond_dev);
2012
1da177e4
LT		 2013 if (bond->slave_cnt == 0) {
2014 goto out;
2015 }
2016
2017 bond->current_arp_slave = NULL;
2018 bond->primary_slave = NULL;
2019 bond_change_active_slave(bond, NULL);
2020
2021 while ((slave = bond->first_slave) != NULL) {
2022 /* Inform AD package of unbinding of slave
2023 * before slave is detached from the list.
2024 */
2025 if (bond->params.mode == BOND_MODE_8023AD) {
2026 bond_3ad_unbind_slave(slave);
2027 }
2028
2029 slave_dev = slave->dev;
2030 bond_detach_slave(bond, slave);
2031
2543331d
JV		 2032 /* now that the slave is detached, unlock and perform
2033 * all the undo steps that should not be called from
2034 * within a lock.
2035 */
2036 write_unlock_bh(&bond->lock);
2037
1da177e4
LT		 2038 if ((bond->params.mode == BOND_MODE_TLB) ||
2039 (bond->params.mode == BOND_MODE_ALB)) {
2040 /* must be called only after the slave
2041 * has been detached from the list
2042 */
2043 bond_alb_deinit_slave(bond, slave);
2044 }
2045
8531c5ff
AK		 2046 bond_compute_features(bond);
2047
b76cdba9 2048 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2049 bond_del_vlans_from_slave(bond, slave_dev);
2050
2051 /* If the mode USES_PRIMARY, then we should only remove its
2052 * promisc and mc settings if it was the curr_active_slave, but that was
2053 * already taken care of above when we detached the slave
2054 */
2055 if (!USES_PRIMARY(bond->params.mode)) {
2056 /* unset promiscuity level from slave */
2057 if (bond_dev->flags & IFF_PROMISC) {
2058 dev_set_promiscuity(slave_dev, -1);
2059 }
2060
2061 /* unset allmulti level from slave */
2062 if (bond_dev->flags & IFF_ALLMULTI) {
2063 dev_set_allmulti(slave_dev, -1);
2064 }
2065
2066 /* flush master's mc_list from slave */
b9e40857 2067 netif_addr_lock_bh(bond_dev);
1da177e4 2068 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2069 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2070 }
2071
2072 netdev_set_master(slave_dev, NULL);
2073
2074 /* close slave before restoring its mac address */
2075 dev_close(slave_dev);
2076
dd957c57 2077 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2078 /* restore original ("permanent") mac address*/
2079 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2080 addr.sa_family = slave_dev->type;
2081 dev_set_mac_address(slave_dev, &addr);
2082 }
1da177e4 2083
8f903c70
JV		 2084 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2085 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2086
2087 kfree(slave);
2088
2089 /* re-acquire the lock before getting the next slave */
2090 write_lock_bh(&bond->lock);
2091 }
2092
2093 /* zero the mac address of the master so it will be
2094 * set by the application to the mac address of the
2095 * first slave
2096 */
2097 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2098
2099 if (list_empty(&bond->vlan_list)) {
2100 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2101 } else {
2102 printk(KERN_WARNING DRV_NAME
4e0952c7 2103 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2104 "still has VLANs.\n",
4e0952c7 2105 bond_dev->name, bond_dev->name);
1da177e4 2106 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2107 ": %s: When re-adding slaves, make sure the bond's "
2108 "HW address matches its VLANs'.\n",
2109 bond_dev->name);
1da177e4
LT		 2110 }
2111
2112 printk(KERN_INFO DRV_NAME
2113 ": %s: released all slaves\n",
2114 bond_dev->name);
2115
2116out:
2117 write_unlock_bh(&bond->lock);
2118
2119 return 0;
2120}
2121
2122/*
2123 * This function changes the active slave to slave <slave_dev>.
2124 * It returns -EINVAL in the following cases.
2125 * - <slave_dev> is not found in the list.
2126 * - There is not active slave now.
2127 * - <slave_dev> is already active.
2128 * - The link state of <slave_dev> is not BOND_LINK_UP.
2129 * - <slave_dev> is not running.
2130 * In these cases, this fuction does nothing.
2131 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2132 */
2133static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2134{
2135 struct bonding *bond = bond_dev->priv;
2136 struct slave *old_active = NULL;
2137 struct slave *new_active = NULL;
2138 int res = 0;
2139
2140 if (!USES_PRIMARY(bond->params.mode)) {
2141 return -EINVAL;
2142 }
2143
2144 /* Verify that master_dev is indeed the master of slave_dev */
2145 if (!(slave_dev->flags & IFF_SLAVE) ||
2146 (slave_dev->master != bond_dev)) {
2147 return -EINVAL;
2148 }
2149
059fe7a5 2150 read_lock(&bond->lock);
1da177e4 2151
059fe7a5 2152 read_lock(&bond->curr_slave_lock);
1da177e4 2153 old_active = bond->curr_active_slave;
059fe7a5
JV		 2154 read_unlock(&bond->curr_slave_lock);
2155
1da177e4
LT		 2156 new_active = bond_get_slave_by_dev(bond, slave_dev);
2157
2158 /*
2159 * Changing to the current active: do nothing; return success.
2160 */
2161 if (new_active && (new_active == old_active)) {
059fe7a5 2162 read_unlock(&bond->lock);
1da177e4
LT		 2163 return 0;
2164 }
2165
2166 if ((new_active) &&
2167 (old_active) &&
2168 (new_active->link == BOND_LINK_UP) &&
2169 IS_UP(new_active->dev)) {
059fe7a5 2170 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2171 bond_change_active_slave(bond, new_active);
059fe7a5 2172 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2173 } else {
2174 res = -EINVAL;
2175 }
2176
059fe7a5 2177 read_unlock(&bond->lock);
1da177e4
LT		 2178
2179 return res;
2180}
2181
1da177e4
LT		 2182static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2183{
2184 struct bonding *bond = bond_dev->priv;
2185
2186 info->bond_mode = bond->params.mode;
2187 info->miimon = bond->params.miimon;
2188
6603a6f2 2189 read_lock(&bond->lock);
1da177e4 2190 info->num_slaves = bond->slave_cnt;
6603a6f2 2191 read_unlock(&bond->lock);
1da177e4
LT		 2192
2193 return 0;
2194}
2195
2196static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2197{
2198 struct bonding *bond = bond_dev->priv;
2199 struct slave *slave;
2200 int i, found = 0;
2201
2202 if (info->slave_id < 0) {
2203 return -ENODEV;
2204 }
2205
6603a6f2 2206 read_lock(&bond->lock);
1da177e4
LT		 2207
2208 bond_for_each_slave(bond, slave, i) {
2209 if (i == (int)info->slave_id) {
2210 found = 1;
2211 break;
2212 }
2213 }
2214
6603a6f2 2215 read_unlock(&bond->lock);
1da177e4
LT		 2216
2217 if (found) {
2218 strcpy(info->slave_name, slave->dev->name);
2219 info->link = slave->link;
2220 info->state = slave->state;
2221 info->link_failure_count = slave->link_failure_count;
2222 } else {
2223 return -ENODEV;
2224 }
2225
2226 return 0;
2227}
2228
2229/*-------------------------------- Monitoring -------------------------------*/
2230
1da177e4 2231
f0c76d61
JV		 2232static int bond_miimon_inspect(struct bonding *bond)
2233{
2234 struct slave *slave;
2235 int i, link_state, commit = 0;
1da177e4
LT		 2236
2237 bond_for_each_slave(bond, slave, i) {
f0c76d61 2238 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2239
f0c76d61 2240 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2241
2242 switch (slave->link) {
f0c76d61
JV		 2243 case BOND_LINK_UP:
2244 if (link_state)
2245 continue;
1da177e4 2246
f0c76d61
JV		 2247 slave->link = BOND_LINK_FAIL;
2248 slave->delay = bond->params.downdelay;
2249 if (slave->delay) {
2250 printk(KERN_INFO DRV_NAME
2251 ": %s: link status down for %s"
2252 "interface %s, disabling it in %d ms.\n",
2253 bond->dev->name,
2254 (bond->params.mode ==
2255 BOND_MODE_ACTIVEBACKUP) ?
2256 ((slave->state == BOND_STATE_ACTIVE) ?
2257 "active " : "backup ") : "",
2258 slave->dev->name,
2259 bond->params.downdelay * bond->params.miimon);
1da177e4 2260 }
f0c76d61
JV		 2261 /*FALLTHRU*/
2262 case BOND_LINK_FAIL:
2263 if (link_state) {
2264 /*
2265 * recovered before downdelay expired
2266 */
2267 slave->link = BOND_LINK_UP;
1da177e4
LT		 2268 slave->jiffies = jiffies;
2269 printk(KERN_INFO DRV_NAME
2270 ": %s: link status up again after %d "
2271 "ms for interface %s.\n",
1b76b316 2272 bond->dev->name,
f0c76d61
JV		 2273 (bond->params.downdelay - slave->delay) *
2274 bond->params.miimon,
2275 slave->dev->name);
2276 continue;
1da177e4 2277 }
f0c76d61
JV		 2278
2279 if (slave->delay <= 0) {
2280 slave->new_link = BOND_LINK_DOWN;
2281 commit++;
2282 continue;
1da177e4 2283 }
1da177e4 2284
f0c76d61
JV		 2285 slave->delay--;
2286 break;
2287
2288 case BOND_LINK_DOWN:
2289 if (!link_state)
2290 continue;
2291
2292 slave->link = BOND_LINK_BACK;
2293 slave->delay = bond->params.updelay;
2294
2295 if (slave->delay) {
2296 printk(KERN_INFO DRV_NAME
2297 ": %s: link status up for "
2298 "interface %s, enabling it in %d ms.\n",
2299 bond->dev->name, slave->dev->name,
2300 bond->params.updelay *
2301 bond->params.miimon);
2302 }
2303 /*FALLTHRU*/
2304 case BOND_LINK_BACK:
2305 if (!link_state) {
2306 slave->link = BOND_LINK_DOWN;
1da177e4
LT		 2307 printk(KERN_INFO DRV_NAME
2308 ": %s: link status down again after %d "
2309 "ms for interface %s.\n",
1b76b316 2310 bond->dev->name,
f0c76d61
JV		 2311 (bond->params.updelay - slave->delay) *
2312 bond->params.miimon,
2313 slave->dev->name);
2314
2315 continue;
2316 }
2317
2318 if (slave->delay <= 0) {
2319 slave->new_link = BOND_LINK_UP;
2320 commit++;
2321 continue;
1da177e4 2322 }
f0c76d61
JV		 2323
2324 slave->delay--;
1da177e4 2325 break;
f0c76d61
JV		 2326 }
2327 }
1da177e4 2328
f0c76d61
JV		 2329 return commit;
2330}
1da177e4 2331
f0c76d61
JV		 2332static void bond_miimon_commit(struct bonding *bond)
2333{
2334 struct slave *slave;
2335 int i;
2336
2337 bond_for_each_slave(bond, slave, i) {
2338 switch (slave->new_link) {
2339 case BOND_LINK_NOCHANGE:
2340 continue;
1da177e4 2341
f0c76d61
JV		 2342 case BOND_LINK_UP:
2343 slave->link = BOND_LINK_UP;
2344 slave->jiffies = jiffies;
2345
2346 if (bond->params.mode == BOND_MODE_8023AD) {
2347 /* prevent it from being the active one */
2348 slave->state = BOND_STATE_BACKUP;
2349 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2350 /* make it immediately active */
2351 slave->state = BOND_STATE_ACTIVE;
2352 } else if (slave != bond->primary_slave) {
2353 /* prevent it from being the active one */
2354 slave->state = BOND_STATE_BACKUP;
1da177e4 2355 }
1da177e4 2356
f0c76d61
JV		 2357 printk(KERN_INFO DRV_NAME
2358 ": %s: link status definitely "
2359 "up for interface %s.\n",
2360 bond->dev->name, slave->dev->name);
1da177e4 2361
f0c76d61
JV		 2362 /* notify ad that the link status has changed */
2363 if (bond->params.mode == BOND_MODE_8023AD)
2364 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2365
f0c76d61
JV		 2366 if ((bond->params.mode == BOND_MODE_TLB) ||
2367 (bond->params.mode == BOND_MODE_ALB))
2368 bond_alb_handle_link_change(bond, slave,
2369 BOND_LINK_UP);
1da177e4 2370
f0c76d61
JV		 2371 if (!bond->curr_active_slave ||
2372 (slave == bond->primary_slave))
2373 goto do_failover;
1da177e4 2374
f0c76d61 2375 continue;
059fe7a5 2376
f0c76d61
JV		 2377 case BOND_LINK_DOWN:
2378 slave->link = BOND_LINK_DOWN;
1da177e4 2379
f0c76d61
JV		 2380 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2381 bond->params.mode == BOND_MODE_8023AD)
2382 bond_set_slave_inactive_flags(slave);
2383
2384 printk(KERN_INFO DRV_NAME
2385 ": %s: link status definitely down for "
2386 "interface %s, disabling it\n",
2387 bond->dev->name, slave->dev->name);
2388
2389 if (bond->params.mode == BOND_MODE_8023AD)
2390 bond_3ad_handle_link_change(slave,
2391 BOND_LINK_DOWN);
2392
2393 if (bond->params.mode == BOND_MODE_TLB ||
2394 bond->params.mode == BOND_MODE_ALB)
2395 bond_alb_handle_link_change(bond, slave,
2396 BOND_LINK_DOWN);
2397
2398 if (slave == bond->curr_active_slave)
2399 goto do_failover;
2400
2401 continue;
2402
2403 default:
2404 printk(KERN_ERR DRV_NAME
2405 ": %s: invalid new link %d on slave %s\n",
2406 bond->dev->name, slave->new_link,
2407 slave->dev->name);
2408 slave->new_link = BOND_LINK_NOCHANGE;
2409
2410 continue;
2411 }
2412
2413do_failover:
2414 ASSERT_RTNL();
2415 write_lock_bh(&bond->curr_slave_lock);
2416 bond_select_active_slave(bond);
2417 write_unlock_bh(&bond->curr_slave_lock);
2418 }
2419
2420 bond_set_carrier(bond);
1da177e4
LT		 2421}
2422
0b0eef66
JV		 2423/*
2424 * bond_mii_monitor
2425 *
2426 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2427 * inspection, then (if inspection indicates something needs to be done)
2428 * an acquisition of appropriate locks followed by a commit phase to
2429 * implement whatever link state changes are indicated.
0b0eef66
JV		 2430 */
2431void bond_mii_monitor(struct work_struct *work)
2432{
2433 struct bonding *bond = container_of(work, struct bonding,
2434 mii_work.work);
0b0eef66
JV		 2435
2436 read_lock(&bond->lock);
f0c76d61
JV		 2437 if (bond->kill_timers)
2438 goto out;
2439
2440 if (bond->slave_cnt == 0)
2441 goto re_arm;
b59f9f74
JV		 2442
2443 if (bond->send_grat_arp) {
2444 read_lock(&bond->curr_slave_lock);
2445 bond_send_gratuitous_arp(bond);
2446 read_unlock(&bond->curr_slave_lock);
2447 }
2448
f0c76d61 2449 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2450 read_unlock(&bond->lock);
2451 rtnl_lock();
2452 read_lock(&bond->lock);
f0c76d61
JV		 2453
2454 bond_miimon_commit(bond);
2455
5655662d
JV		 2456 read_unlock(&bond->lock);
2457 rtnl_unlock(); /* might sleep, hold no other locks */
2458 read_lock(&bond->lock);
0b0eef66
JV		 2459 }
2460
f0c76d61
JV		 2461re_arm:
2462 if (bond->params.miimon)
2463 queue_delayed_work(bond->wq, &bond->mii_work,
2464 msecs_to_jiffies(bond->params.miimon));
2465out:
0b0eef66 2466 read_unlock(&bond->lock);
0b0eef66 2467}
c3ade5ca 2468
d3bb52b0 2469static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2470{
2471 struct in_device *idev;
2472 struct in_ifaddr *ifa;
a144ea4b 2473 __be32 addr = 0;
c3ade5ca
JV		 2474
2475 if (!dev)
2476 return 0;
2477
2478 rcu_read_lock();
e5ed6399 2479 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2480 if (!idev)
2481 goto out;
2482
2483 ifa = idev->ifa_list;
2484 if (!ifa)
2485 goto out;
2486
2487 addr = ifa->ifa_local;
2488out:
2489 rcu_read_unlock();
2490 return addr;
2491}
2492
d3bb52b0 2493static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2494{
0883beca 2495 struct vlan_entry *vlan;
f5b2b966
JV		 2496
2497 if (ip == bond->master_ip)
2498 return 1;
2499
0883beca 2500 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2501 if (ip == vlan->vlan_ip)
2502 return 1;
2503 }
2504
2505 return 0;
2506}
2507
c3ade5ca
JV		 2508/*
2509 * We go to the (large) trouble of VLAN tagging ARP frames because
2510 * switches in VLAN mode (especially if ports are configured as
2511 * "native" to a VLAN) might not pass non-tagged frames.
2512 */
d3bb52b0 2513static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2514{
2515 struct sk_buff *skb;
2516
2517 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2518 slave_dev->name, dest_ip, src_ip, vlan_id);
2519
2520 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2521 NULL, slave_dev->dev_addr, NULL);
2522
2523 if (!skb) {
2524 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2525 return;
2526 }
2527 if (vlan_id) {
2528 skb = vlan_put_tag(skb, vlan_id);
2529 if (!skb) {
2530 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2531 return;
2532 }
2533 }
2534 arp_xmit(skb);
2535}
2536
2537
1da177e4
LT		 2538static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2539{
c3ade5ca 2540 int i, vlan_id, rv;
d3bb52b0 2541 __be32 *targets = bond->params.arp_targets;
0883beca 2542 struct vlan_entry *vlan;
c3ade5ca
JV		 2543 struct net_device *vlan_dev;
2544 struct flowi fl;
2545 struct rtable *rt;
1da177e4 2546
6b780567
MW		 2547 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2548 if (!targets[i])
2549 continue;
c3ade5ca
JV		 2550 dprintk("basa: target %x\n", targets[i]);
2551 if (list_empty(&bond->vlan_list)) {
2552 dprintk("basa: empty vlan: arp_send\n");
2553 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2554 bond->master_ip, 0);
2555 continue;
2556 }
2557
2558 /*
2559 * If VLANs are configured, we do a route lookup to
2560 * determine which VLAN interface would be used, so we
2561 * can tag the ARP with the proper VLAN tag.
2562 */
2563 memset(&fl, 0, sizeof(fl));
2564 fl.fl4_dst = targets[i];
2565 fl.fl4_tos = RTO_ONLINK;
2566
f206351a 2567 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2568 if (rv) {
2569 if (net_ratelimit()) {
2570 printk(KERN_WARNING DRV_NAME
2571 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2572 bond->dev->name, NIPQUAD(fl.fl4_dst));
2573 }
2574 continue;
2575 }
2576
2577 /*
2578 * This target is not on a VLAN
2579 */
2580 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2581 ip_rt_put(rt);
c3ade5ca
JV		 2582 dprintk("basa: rtdev == bond->dev: arp_send\n");
2583 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2584 bond->master_ip, 0);
2585 continue;
2586 }
2587
2588 vlan_id = 0;
0883beca 2589 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2590 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2591 if (vlan_dev == rt->u.dst.dev) {
2592 vlan_id = vlan->vlan_id;
2593 dprintk("basa: vlan match on %s %d\n",
2594 vlan_dev->name, vlan_id);
2595 break;
2596 }
2597 }
2598
2599 if (vlan_id) {
ed4b9f80 2600 ip_rt_put(rt);
c3ade5ca
JV		 2601 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2602 vlan->vlan_ip, vlan_id);
2603 continue;
2604 }
2605
2606 if (net_ratelimit()) {
2607 printk(KERN_WARNING DRV_NAME
2608 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2609 bond->dev->name, NIPQUAD(fl.fl4_dst),
2610 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2611 }
ed4b9f80 2612 ip_rt_put(rt);
c3ade5ca
JV		 2613 }
2614}
2615
2616/*
2617 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2618 * for each VLAN above us.
b59f9f74
JV		 2619 *
2620 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2621 */
2622static void bond_send_gratuitous_arp(struct bonding *bond)
2623{
2624 struct slave *slave = bond->curr_active_slave;
2625 struct vlan_entry *vlan;
2626 struct net_device *vlan_dev;
2627
2628 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2629 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2630
2631 if (!slave || !bond->send_grat_arp ||
2632 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2633 return;
2634
b59f9f74
JV		 2635 bond->send_grat_arp--;
2636
c3ade5ca
JV		 2637 if (bond->master_ip) {
2638 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2639 bond->master_ip, 0);
c3ade5ca
JV		 2640 }
2641
2642 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2643 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2644 if (vlan->vlan_ip) {
2645 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2646 vlan->vlan_ip, vlan->vlan_id);
2647 }
1da177e4
LT		 2648 }
2649}
2650
d3bb52b0 2651static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2652{
2653 int i;
d3bb52b0 2654 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2655
2656 targets = bond->params.arp_targets;
2657 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2658 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2659 "%u.%u.%u.%u bhti(tip) %d\n",
2660 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2661 bond_has_this_ip(bond, tip));
2662 if (sip == targets[i]) {
2663 if (bond_has_this_ip(bond, tip))
2664 slave->last_arp_rx = jiffies;
2665 return;
2666 }
2667 }
2668}
2669
2670static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2671{
2672 struct arphdr *arp;
2673 struct slave *slave;
2674 struct bonding *bond;
2675 unsigned char *arp_ptr;
d3bb52b0 2676 __be32 sip, tip;
f5b2b966 2677
c346dca1 2678 if (dev_net(dev) != &init_net)
e730c155
EB		 2679 goto out;
2680
f5b2b966
JV		 2681 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2682 goto out;
2683
2684 bond = dev->priv;
2685 read_lock(&bond->lock);
2686
2687 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2688 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2689 orig_dev ? orig_dev->name : "NULL");
2690
2691 slave = bond_get_slave_by_dev(bond, orig_dev);
2692 if (!slave || !slave_do_arp_validate(bond, slave))
2693 goto out_unlock;
2694
988b7050 2695 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2696 goto out_unlock;
2697
d0a92be0 2698 arp = arp_hdr(skb);
f5b2b966
JV		 2699 if (arp->ar_hln != dev->addr_len ||
2700 skb->pkt_type == PACKET_OTHERHOST ||
2701 skb->pkt_type == PACKET_LOOPBACK ||
2702 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2703 arp->ar_pro != htons(ETH_P_IP) ||
2704 arp->ar_pln != 4)
2705 goto out_unlock;
2706
2707 arp_ptr = (unsigned char *)(arp + 1);
2708 arp_ptr += dev->addr_len;
2709 memcpy(&sip, arp_ptr, 4);
2710 arp_ptr += 4 + dev->addr_len;
2711 memcpy(&tip, arp_ptr, 4);
2712
2713 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2714 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2715 slave->state, bond->params.arp_validate,
2716 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2717
2718 /*
2719 * Backup slaves won't see the ARP reply, but do come through
2720 * here for each ARP probe (so we swap the sip/tip to validate
2721 * the probe). In a "redundant switch, common router" type of
2722 * configuration, the ARP probe will (hopefully) travel from
2723 * the active, through one switch, the router, then the other
2724 * switch before reaching the backup.
2725 */
2726 if (slave->state == BOND_STATE_ACTIVE)
2727 bond_validate_arp(bond, slave, sip, tip);
2728 else
2729 bond_validate_arp(bond, slave, tip, sip);
2730
2731out_unlock:
2732 read_unlock(&bond->lock);
2733out:
2734 dev_kfree_skb(skb);
2735 return NET_RX_SUCCESS;
2736}
2737
1da177e4
LT		 2738/*
2739 * this function is called regularly to monitor each slave's link
2740 * ensuring that traffic is being sent and received when arp monitoring
2741 * is used in load-balancing mode. if the adapter has been dormant, then an
2742 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2743 * arp monitoring in active backup mode.
2744 */
1b76b316 2745void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2746{
1b76b316
JV		 2747 struct bonding *bond = container_of(work, struct bonding,
2748 arp_work.work);
1da177e4
LT		 2749 struct slave *slave, *oldcurrent;
2750 int do_failover = 0;
2751 int delta_in_ticks;
2752 int i;
2753
2754 read_lock(&bond->lock);
2755
5ce0da8f 2756 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2757
2758 if (bond->kill_timers) {
2759 goto out;
2760 }
2761
2762 if (bond->slave_cnt == 0) {
2763 goto re_arm;
2764 }
2765
2766 read_lock(&bond->curr_slave_lock);
2767 oldcurrent = bond->curr_active_slave;
2768 read_unlock(&bond->curr_slave_lock);
2769
2770 /* see if any of the previous devices are up now (i.e. they have
2771 * xmt and rcv traffic). the curr_active_slave does not come into
2772 * the picture unless it is null. also, slave->jiffies is not needed
2773 * here because we send an arp on each slave and give a slave as
2774 * long as it needs to get the tx/rx within the delta.
2775 * TODO: what about up/down delay in arp mode? it wasn't here before
2776 * so it can wait
2777 */
2778 bond_for_each_slave(bond, slave, i) {
2779 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2780 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2781 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2782
2783 slave->link = BOND_LINK_UP;
2784 slave->state = BOND_STATE_ACTIVE;
2785
2786 /* primary_slave has no meaning in round-robin
2787 * mode. the window of a slave being up and
2788 * curr_active_slave being null after enslaving
2789 * is closed.
2790 */
2791 if (!oldcurrent) {
2792 printk(KERN_INFO DRV_NAME
2793 ": %s: link status definitely "
2794 "up for interface %s, ",
1b76b316 2795 bond->dev->name,
1da177e4
LT		 2796 slave->dev->name);
2797 do_failover = 1;
2798 } else {
2799 printk(KERN_INFO DRV_NAME
2800 ": %s: interface %s is now up\n",
1b76b316 2801 bond->dev->name,
1da177e4
LT		 2802 slave->dev->name);
2803 }
2804 }
2805 } else {
2806 /* slave->link == BOND_LINK_UP */
2807
2808 /* not all switches will respond to an arp request
2809 * when the source ip is 0, so don't take the link down
2810 * if we don't know our ip yet
2811 */
b63bb739 2812 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2813 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2814
2815 slave->link = BOND_LINK_DOWN;
2816 slave->state = BOND_STATE_BACKUP;
2817
2818 if (slave->link_failure_count < UINT_MAX) {
2819 slave->link_failure_count++;
2820 }
2821
2822 printk(KERN_INFO DRV_NAME
2823 ": %s: interface %s is now down.\n",
1b76b316 2824 bond->dev->name,
1da177e4
LT		 2825 slave->dev->name);
2826
2827 if (slave == oldcurrent) {
2828 do_failover = 1;
2829 }
2830 }
2831 }
2832
2833 /* note: if switch is in round-robin mode, all links
2834 * must tx arp to ensure all links rx an arp - otherwise
2835 * links may oscillate or not come up at all; if switch is
2836 * in something like xor mode, there is nothing we can
2837 * do - all replies will be rx'ed on same link causing slaves
2838 * to be unstable during low/no traffic periods
2839 */
2840 if (IS_UP(slave->dev)) {
2841 bond_arp_send_all(bond, slave);
2842 }
2843 }
2844
2845 if (do_failover) {
059fe7a5 2846 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2847
2848 bond_select_active_slave(bond);
2849
059fe7a5 2850 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2851 }
2852
2853re_arm:
1b76b316
JV		 2854 if (bond->params.arp_interval)
2855 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2856out:
2857 read_unlock(&bond->lock);
2858}
2859
2860/*
b2220cad
JV		 2861 * Called to inspect slaves for active-backup mode ARP monitor link state
2862 * changes. Sets new_link in slaves to specify what action should take
2863 * place for the slave. Returns 0 if no changes are found, >0 if changes
2864 * to link states must be committed.
2865 *
2866 * Called with bond->lock held for read.
1da177e4 2867 */
b2220cad 2868static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2869{
1da177e4 2870 struct slave *slave;
b2220cad 2871 int i, commit = 0;
1da177e4 2872
b2220cad
JV		 2873 bond_for_each_slave(bond, slave, i) {
2874 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2875
b2220cad
JV		 2876 if (slave->link != BOND_LINK_UP) {
2877 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2878 delta_in_ticks)) {
2879 slave->new_link = BOND_LINK_UP;
2880 commit++;
2881 }
1da177e4 2882
b2220cad
JV		 2883 continue;
2884 }
1da177e4 2885
b2220cad
JV		 2886 /*
2887 * Give slaves 2*delta after being enslaved or made
2888 * active. This avoids bouncing, as the last receive
2889 * times need a full ARP monitor cycle to be updated.
2890 */
2891 if (!time_after_eq(jiffies, slave->jiffies +
2892 2 * delta_in_ticks))
2893 continue;
2894
2895 /*
2896 * Backup slave is down if:
2897 * - No current_arp_slave AND
2898 * - more than 3*delta since last receive AND
2899 * - the bond has an IP address
2900 *
2901 * Note: a non-null current_arp_slave indicates
2902 * the curr_active_slave went down and we are
2903 * searching for a new one; under this condition
2904 * we only take the curr_active_slave down - this
2905 * gives each slave a chance to tx/rx traffic
2906 * before being taken out
2907 */
2908 if (slave->state == BOND_STATE_BACKUP &&
2909 !bond->current_arp_slave &&
2910 time_after(jiffies, slave_last_rx(bond, slave) +
2911 3 * delta_in_ticks)) {
2912 slave->new_link = BOND_LINK_DOWN;
2913 commit++;
2914 }
2915
2916 /*
2917 * Active slave is down if:
2918 * - more than 2*delta since transmitting OR
2919 * - (more than 2*delta since receive AND
2920 * the bond has an IP address)
2921 */
2922 if ((slave->state == BOND_STATE_ACTIVE) &&
2923 (time_after_eq(jiffies, slave->dev->trans_start +
2924 2 * delta_in_ticks) ||
2925 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2926 + 2 * delta_in_ticks)))) {
2927 slave->new_link = BOND_LINK_DOWN;
2928 commit++;
2929 }
1da177e4
LT		 2930 }
2931
b2220cad
JV		 2932 read_lock(&bond->curr_slave_lock);
2933
2934 /*
2935 * Trigger a commit if the primary option setting has changed.
1da177e4 2936 */
b2220cad
JV		 2937 if (bond->primary_slave &&
2938 (bond->primary_slave != bond->curr_active_slave) &&
2939 (bond->primary_slave->link == BOND_LINK_UP))
2940 commit++;
1da177e4 2941
b2220cad 2942 read_unlock(&bond->curr_slave_lock);
1da177e4 2943
b2220cad
JV		 2944 return commit;
2945}
1da177e4 2946
b2220cad
JV		 2947/*
2948 * Called to commit link state changes noted by inspection step of
2949 * active-backup mode ARP monitor.
2950 *
2951 * Called with RTNL and bond->lock for read.
2952 */
2953static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2954{
2955 struct slave *slave;
2956 int i;
1da177e4 2957
b2220cad
JV		 2958 bond_for_each_slave(bond, slave, i) {
2959 switch (slave->new_link) {
2960 case BOND_LINK_NOCHANGE:
2961 continue;
ff59c456 2962
b2220cad
JV		 2963 case BOND_LINK_UP:
2964 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2965
b2220cad
JV		 2966 if (!bond->curr_active_slave &&
2967 time_before_eq(jiffies, slave->dev->trans_start +
2968 delta_in_ticks)) {
2969 slave->link = BOND_LINK_UP;
2970 bond_change_active_slave(bond, slave);
2971 bond->current_arp_slave = NULL;
2972
2973 printk(KERN_INFO DRV_NAME
2974 ": %s: %s is up and now the "
2975 "active interface\n",
2976 bond->dev->name, slave->dev->name);
2977
2978 } else if (bond->curr_active_slave != slave) {
2979 /* this slave has just come up but we
2980 * already have a current slave; this can
2981 * also happen if bond_enslave adds a new
2982 * slave that is up while we are searching
2983 * for a new slave
1da177e4 2984 */
b2220cad
JV		 2985 slave->link = BOND_LINK_UP;
2986 bond_set_slave_inactive_flags(slave);
2987 bond->current_arp_slave = NULL;
2988
2989 printk(KERN_INFO DRV_NAME
2990 ": %s: backup interface %s is now up\n",
2991 bond->dev->name, slave->dev->name);
2992 }
1da177e4 2993
b2220cad 2994 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 2995
b2220cad 2996 break;
1da177e4 2997
b2220cad
JV		 2998 case BOND_LINK_DOWN:
2999 if (slave->link_failure_count < UINT_MAX)
3000 slave->link_failure_count++;
3001
3002 slave->link = BOND_LINK_DOWN;
3003
3004 if (slave == bond->curr_active_slave) {
3005 printk(KERN_INFO DRV_NAME
3006 ": %s: link status down for active "
3007 "interface %s, disabling it\n",
3008 bond->dev->name, slave->dev->name);
1da177e4
LT		 3009
3010 bond_set_slave_inactive_flags(slave);
3011
b2220cad
JV		 3012 write_lock_bh(&bond->curr_slave_lock);
3013
3014 bond_select_active_slave(bond);
3015 if (bond->curr_active_slave)
3016 bond->curr_active_slave->jiffies =
3017 jiffies;
3018
3019 write_unlock_bh(&bond->curr_slave_lock);
3020
3021 bond->current_arp_slave = NULL;
3022
3023 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3024 printk(KERN_INFO DRV_NAME
3025 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3026 bond->dev->name, slave->dev->name);
3027
3028 bond_set_slave_inactive_flags(slave);
1da177e4 3029 }
b2220cad
JV		 3030 break;
3031
3032 default:
3033 printk(KERN_ERR DRV_NAME
3034 ": %s: impossible: new_link %d on slave %s\n",
3035 bond->dev->name, slave->new_link,
3036 slave->dev->name);
1da177e4
LT		 3037 }
3038 }
3039
b2220cad
JV		 3040 /*
3041 * No race with changes to primary via sysfs, as we hold rtnl.
3042 */
3043 if (bond->primary_slave &&
3044 (bond->primary_slave != bond->curr_active_slave) &&
3045 (bond->primary_slave->link == BOND_LINK_UP)) {
3046 write_lock_bh(&bond->curr_slave_lock);
3047 bond_change_active_slave(bond, bond->primary_slave);
3048 write_unlock_bh(&bond->curr_slave_lock);
3049 }
1da177e4 3050
b2220cad
JV		 3051 bond_set_carrier(bond);
3052}
1da177e4 3053
b2220cad
JV		 3054/*
3055 * Send ARP probes for active-backup mode ARP monitor.
3056 *
3057 * Called with bond->lock held for read.
3058 */
3059static void bond_ab_arp_probe(struct bonding *bond)
3060{
3061 struct slave *slave;
3062 int i;
1da177e4 3063
b2220cad 3064 read_lock(&bond->curr_slave_lock);
1da177e4 3065
b2220cad
JV		 3066 if (bond->current_arp_slave && bond->curr_active_slave)
3067 printk("PROBE: c_arp %s && cas %s BAD\n",
3068 bond->current_arp_slave->dev->name,
3069 bond->curr_active_slave->dev->name);
1da177e4 3070
b2220cad
JV		 3071 if (bond->curr_active_slave) {
3072 bond_arp_send_all(bond, bond->curr_active_slave);
3073 read_unlock(&bond->curr_slave_lock);
3074 return;
3075 }
1da177e4 3076
b2220cad 3077 read_unlock(&bond->curr_slave_lock);
059fe7a5 3078
b2220cad
JV		 3079 /* if we don't have a curr_active_slave, search for the next available
3080 * backup slave from the current_arp_slave and make it the candidate
3081 * for becoming the curr_active_slave
3082 */
1da177e4 3083
b2220cad
JV		 3084 if (!bond->current_arp_slave) {
3085 bond->current_arp_slave = bond->first_slave;
3086 if (!bond->current_arp_slave)
3087 return;
3088 }
1da177e4 3089
b2220cad 3090 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3091
b2220cad
JV		 3092 /* search for next candidate */
3093 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3094 if (IS_UP(slave->dev)) {
3095 slave->link = BOND_LINK_BACK;
3096 bond_set_slave_active_flags(slave);
3097 bond_arp_send_all(bond, slave);
1da177e4 3098 slave->jiffies = jiffies;
b2220cad
JV		 3099 bond->current_arp_slave = slave;
3100 break;
1da177e4
LT		 3101 }
3102
b2220cad
JV		 3103 /* if the link state is up at this point, we
3104 * mark it down - this can happen if we have
3105 * simultaneous link failures and
3106 * reselect_active_interface doesn't make this
3107 * one the current slave so it is still marked
3108 * up when it is actually down
1da177e4 3109 */
b2220cad
JV		 3110 if (slave->link == BOND_LINK_UP) {
3111 slave->link = BOND_LINK_DOWN;
3112 if (slave->link_failure_count < UINT_MAX)
3113 slave->link_failure_count++;
1da177e4 3114
b2220cad
JV		 3115 bond_set_slave_inactive_flags(slave);
3116
3117 printk(KERN_INFO DRV_NAME
3118 ": %s: backup interface %s is now down.\n",
3119 bond->dev->name, slave->dev->name);
1da177e4 3120 }
b2220cad
JV		 3121 }
3122}
1da177e4 3123
b2220cad
JV		 3124void bond_activebackup_arp_mon(struct work_struct *work)
3125{
3126 struct bonding *bond = container_of(work, struct bonding,
3127 arp_work.work);
3128 int delta_in_ticks;
1da177e4 3129
b2220cad 3130 read_lock(&bond->lock);
1da177e4 3131
b2220cad
JV		 3132 if (bond->kill_timers)
3133 goto out;
1da177e4 3134
b2220cad 3135 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3136
b2220cad
JV		 3137 if (bond->slave_cnt == 0)
3138 goto re_arm;
3139
b59f9f74
JV		 3140 if (bond->send_grat_arp) {
3141 read_lock(&bond->curr_slave_lock);
3142 bond_send_gratuitous_arp(bond);
3143 read_unlock(&bond->curr_slave_lock);
3144 }
3145
b2220cad
JV		 3146 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3147 read_unlock(&bond->lock);
3148 rtnl_lock();
3149 read_lock(&bond->lock);
3150
3151 bond_ab_arp_commit(bond, delta_in_ticks);
3152
3153 read_unlock(&bond->lock);
3154 rtnl_unlock();
3155 read_lock(&bond->lock);
1da177e4
LT		 3156 }
3157
b2220cad
JV		 3158 bond_ab_arp_probe(bond);
3159
1da177e4
LT		 3160re_arm:
3161 if (bond->params.arp_interval) {
1b76b316 3162 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3163 }
3164out:
3165 read_unlock(&bond->lock);
3166}
3167
3168/*------------------------------ proc/seq_file-------------------------------*/
3169
3170#ifdef CONFIG_PROC_FS
3171
1da177e4
LT		 3172static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3173{
3174 struct bonding *bond = seq->private;
3175 loff_t off = 0;
3176 struct slave *slave;
3177 int i;
3178
3179 /* make sure the bond won't be taken away */
3180 read_lock(&dev_base_lock);
6603a6f2 3181 read_lock(&bond->lock);
1da177e4
LT		 3182
3183 if (*pos == 0) {
3184 return SEQ_START_TOKEN;
3185 }
3186
3187 bond_for_each_slave(bond, slave, i) {
3188 if (++off == *pos) {
3189 return slave;
3190 }
3191 }
3192
3193 return NULL;
3194}
3195
3196static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3197{
3198 struct bonding *bond = seq->private;
3199 struct slave *slave = v;
3200
3201 ++*pos;
3202 if (v == SEQ_START_TOKEN) {
3203 return bond->first_slave;
3204 }
3205
3206 slave = slave->next;
3207
3208 return (slave == bond->first_slave) ? NULL : slave;
3209}
3210
3211static void bond_info_seq_stop(struct seq_file *seq, void *v)
3212{
3213 struct bonding *bond = seq->private;
3214
6603a6f2 3215 read_unlock(&bond->lock);
1da177e4
LT		 3216 read_unlock(&dev_base_lock);
3217}
3218
3219static void bond_info_show_master(struct seq_file *seq)
3220{
3221 struct bonding *bond = seq->private;
3222 struct slave *curr;
4756b02f
MW		 3223 int i;
3224 u32 target;
1da177e4
LT		 3225
3226 read_lock(&bond->curr_slave_lock);
3227 curr = bond->curr_active_slave;
3228 read_unlock(&bond->curr_slave_lock);
3229
dd957c57 3230 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3231 bond_mode_name(bond->params.mode));
3232
dd957c57
JV		 3233 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3234 bond->params.fail_over_mac)
3915c1e8
JV		 3235 seq_printf(seq, " (fail_over_mac %s)",
3236 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3237
3238 seq_printf(seq, "\n");
3239
c61b75ad
MW		 3240 if (bond->params.mode == BOND_MODE_XOR ||
3241 bond->params.mode == BOND_MODE_8023AD) {
3242 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3243 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3244 bond->params.xmit_policy);
3245 }
3246
1da177e4
LT		 3247 if (USES_PRIMARY(bond->params.mode)) {
3248 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3249 (bond->primary_slave) ?
3250 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3251
3252 seq_printf(seq, "Currently Active Slave: %s\n",
3253 (curr) ? curr->dev->name : "None");
3254 }
3255
ff59c456
JV		 3256 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3257 "up" : "down");
1da177e4
LT		 3258 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3259 seq_printf(seq, "Up Delay (ms): %d\n",
3260 bond->params.updelay * bond->params.miimon);
3261 seq_printf(seq, "Down Delay (ms): %d\n",
3262 bond->params.downdelay * bond->params.miimon);
3263
4756b02f
MW		 3264
3265 /* ARP information */
3266 if(bond->params.arp_interval > 0) {
3267 int printed=0;
3268 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3269 bond->params.arp_interval);
3270
3271 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3272
3273 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3274 if (!bond->params.arp_targets[i])
3275 continue;
3276 if (printed)
3277 seq_printf(seq, ",");
3278 target = ntohl(bond->params.arp_targets[i]);
3279 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3280 printed = 1;
3281 }
3282 seq_printf(seq, "\n");
3283 }
3284
1da177e4
LT		 3285 if (bond->params.mode == BOND_MODE_8023AD) {
3286 struct ad_info ad_info;
3287
3288 seq_puts(seq, "\n802.3ad info\n");
3289 seq_printf(seq, "LACP rate: %s\n",
3290 (bond->params.lacp_fast) ? "fast" : "slow");
3291
3292 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3293 seq_printf(seq, "bond %s has no active aggregator\n",
3294 bond->dev->name);
3295 } else {
3296 seq_printf(seq, "Active Aggregator Info:\n");
3297
3298 seq_printf(seq, "\tAggregator ID: %d\n",
3299 ad_info.aggregator_id);
3300 seq_printf(seq, "\tNumber of ports: %d\n",
3301 ad_info.ports);
3302 seq_printf(seq, "\tActor Key: %d\n",
3303 ad_info.actor_key);
3304 seq_printf(seq, "\tPartner Key: %d\n",
3305 ad_info.partner_key);
e174961c
JB		 3306 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3307 ad_info.partner_system);
1da177e4
LT		 3308 }
3309 }
3310}
3311
3312static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3313{
3314 struct bonding *bond = seq->private;
3315
3316 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3317 seq_printf(seq, "MII Status: %s\n",
3318 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3319 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3320 slave->link_failure_count);
3321
e174961c 3322 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
1da177e4
LT		 3323
3324 if (bond->params.mode == BOND_MODE_8023AD) {
3325 const struct aggregator *agg
3326 = SLAVE_AD_INFO(slave).port.aggregator;
3327
3328 if (agg) {
3329 seq_printf(seq, "Aggregator ID: %d\n",
3330 agg->aggregator_identifier);
3331 } else {
3332 seq_puts(seq, "Aggregator ID: N/A\n");
3333 }
3334 }
3335}
3336
3337static int bond_info_seq_show(struct seq_file *seq, void *v)
3338{
3339 if (v == SEQ_START_TOKEN) {
3340 seq_printf(seq, "%s\n", version);
3341 bond_info_show_master(seq);
3342 } else {
3343 bond_info_show_slave(seq, v);
3344 }
3345
3346 return 0;
3347}
3348
3349static struct seq_operations bond_info_seq_ops = {
3350 .start = bond_info_seq_start,
3351 .next = bond_info_seq_next,
3352 .stop = bond_info_seq_stop,
3353 .show = bond_info_seq_show,
3354};
3355
3356static int bond_info_open(struct inode *inode, struct file *file)
3357{
3358 struct seq_file *seq;
3359 struct proc_dir_entry *proc;
3360 int res;
3361
3362 res = seq_open(file, &bond_info_seq_ops);
3363 if (!res) {
3364 /* recover the pointer buried in proc_dir_entry data */
3365 seq = file->private_data;
3366 proc = PDE(inode);
3367 seq->private = proc->data;
3368 }
3369
3370 return res;
3371}
3372
d54b1fdb 3373static const struct file_operations bond_info_fops = {
1da177e4
LT		 3374 .owner = THIS_MODULE,
3375 .open = bond_info_open,
3376 .read = seq_read,
3377 .llseek = seq_lseek,
3378 .release = seq_release,
3379};
3380
3381static int bond_create_proc_entry(struct bonding *bond)
3382{
3383 struct net_device *bond_dev = bond->dev;
3384
3385 if (bond_proc_dir) {
a95609cb
DL		 3386 bond->proc_entry = proc_create_data(bond_dev->name,
3387 S_IRUGO, bond_proc_dir,
3388 &bond_info_fops, bond);
1da177e4
LT		 3389 if (bond->proc_entry == NULL) {
3390 printk(KERN_WARNING DRV_NAME
3391 ": Warning: Cannot create /proc/net/%s/%s\n",
3392 DRV_NAME, bond_dev->name);
3393 } else {
1da177e4
LT		 3394 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3395 }
3396 }
3397
3398 return 0;
3399}
3400
3401static void bond_remove_proc_entry(struct bonding *bond)
3402{
3403 if (bond_proc_dir && bond->proc_entry) {
3404 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3405 memset(bond->proc_file_name, 0, IFNAMSIZ);
3406 bond->proc_entry = NULL;
3407 }
3408}
3409
3410/* Create the bonding directory under /proc/net, if doesn't exist yet.
3411 * Caller must hold rtnl_lock.
3412 */
3413static void bond_create_proc_dir(void)
3414{
3415 int len = strlen(DRV_NAME);
3416
457c4cbc 3417 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3418 bond_proc_dir = bond_proc_dir->next) {
3419 if ((bond_proc_dir->namelen == len) &&
3420 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3421 break;
3422 }
3423 }
3424
3425 if (!bond_proc_dir) {
457c4cbc 3426 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3427 if (bond_proc_dir) {
3428 bond_proc_dir->owner = THIS_MODULE;
3429 } else {
3430 printk(KERN_WARNING DRV_NAME
3431 ": Warning: cannot create /proc/net/%s\n",
3432 DRV_NAME);
3433 }
3434 }
3435}
3436
3437/* Destroy the bonding directory under /proc/net, if empty.
3438 * Caller must hold rtnl_lock.
3439 */
3440static void bond_destroy_proc_dir(void)
3441{
3442 struct proc_dir_entry *de;
3443
3444 if (!bond_proc_dir) {
3445 return;
3446 }
3447
3448 /* verify that the /proc dir is empty */
3449 for (de = bond_proc_dir->subdir; de; de = de->next) {
3450 /* ignore . and .. */
3451 if (*(de->name) != '.') {
3452 break;
3453 }
3454 }
3455
3456 if (de) {
3457 if (bond_proc_dir->owner == THIS_MODULE) {
3458 bond_proc_dir->owner = NULL;
3459 }
3460 } else {
457c4cbc 3461 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3462 bond_proc_dir = NULL;
3463 }
3464}
3465#endif /* CONFIG_PROC_FS */
3466
3467/*-------------------------- netdev event handling --------------------------*/
3468
3469/*
3470 * Change device name
3471 */
3472static int bond_event_changename(struct bonding *bond)
3473{
3474#ifdef CONFIG_PROC_FS
3475 bond_remove_proc_entry(bond);
3476 bond_create_proc_entry(bond);
3477#endif
b76cdba9
MW		 3478 down_write(&(bonding_rwsem));
3479 bond_destroy_sysfs_entry(bond);
3480 bond_create_sysfs_entry(bond);
3481 up_write(&(bonding_rwsem));
1da177e4
LT		 3482 return NOTIFY_DONE;
3483}
3484
3485static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3486{
3487 struct bonding *event_bond = bond_dev->priv;
3488
3489 switch (event) {
3490 case NETDEV_CHANGENAME:
3491 return bond_event_changename(event_bond);
3492 case NETDEV_UNREGISTER:
3b96c858 3493 bond_release_all(event_bond->dev);
1da177e4
LT		 3494 break;
3495 default:
3496 break;
3497 }
3498
3499 return NOTIFY_DONE;
3500}
3501
3502static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3503{
3504 struct net_device *bond_dev = slave_dev->master;
8531c5ff 3505 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3506
3507 switch (event) {
3508 case NETDEV_UNREGISTER:
3509 if (bond_dev) {
1284cd3a
JV		 3510 if (bond->setup_by_slave)
3511 bond_release_and_destroy(bond_dev, slave_dev);
3512 else
3513 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3514 }
3515 break;
3516 case NETDEV_CHANGE:
3517 /*
3518 * TODO: is this what we get if somebody
3519 * sets up a hierarchical bond, then rmmod's
3520 * one of the slave bonding devices?
3521 */
3522 break;
3523 case NETDEV_DOWN:
3524 /*
3525 * ... Or is it this?
3526 */
3527 break;
3528 case NETDEV_CHANGEMTU:
3529 /*
3530 * TODO: Should slaves be allowed to
3531 * independently alter their MTU? For
3532 * an active-backup bond, slaves need
3533 * not be the same type of device, so
3534 * MTUs may vary. For other modes,
3535 * slaves arguably should have the
3536 * same MTUs. To do this, we'd need to
3537 * take over the slave's change_mtu
3538 * function for the duration of their
3539 * servitude.
3540 */
3541 break;
3542 case NETDEV_CHANGENAME:
3543 /*
3544 * TODO: handle changing the primary's name
3545 */
3546 break;
8531c5ff
AK		 3547 case NETDEV_FEAT_CHANGE:
3548 bond_compute_features(bond);
3549 break;
1da177e4
LT		 3550 default:
3551 break;
3552 }
3553
3554 return NOTIFY_DONE;
3555}
3556
3557/*
3558 * bond_netdev_event: handle netdev notifier chain events.
3559 *
3560 * This function receives events for the netdev chain. The caller (an
e041c683 3561 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3562 * locks for us to safely manipulate the slave devices (RTNL lock,
3563 * dev_probe_lock).
3564 */
3565static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3566{
3567 struct net_device *event_dev = (struct net_device *)ptr;
3568
c346dca1 3569 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3570 return NOTIFY_DONE;
3571
1da177e4
LT		 3572 dprintk("event_dev: %s, event: %lx\n",
3573 (event_dev ? event_dev->name : "None"),
3574 event);
3575
0b680e75
JV		 3576 if (!(event_dev->priv_flags & IFF_BONDING))
3577 return NOTIFY_DONE;
3578
1da177e4
LT		 3579 if (event_dev->flags & IFF_MASTER) {
3580 dprintk("IFF_MASTER\n");
3581 return bond_master_netdev_event(event, event_dev);
3582 }
3583
3584 if (event_dev->flags & IFF_SLAVE) {
3585 dprintk("IFF_SLAVE\n");
3586 return bond_slave_netdev_event(event, event_dev);
3587 }
3588
3589 return NOTIFY_DONE;
3590}
3591
c3ade5ca
JV		 3592/*
3593 * bond_inetaddr_event: handle inetaddr notifier chain events.
3594 *
3595 * We keep track of device IPs primarily to use as source addresses in
3596 * ARP monitor probes (rather than spewing out broadcasts all the time).
3597 *
3598 * We track one IP for the main device (if it has one), plus one per VLAN.
3599 */
3600static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3601{
3602 struct in_ifaddr *ifa = ptr;
3603 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3604 struct bonding *bond;
3605 struct vlan_entry *vlan;
c3ade5ca 3606
c346dca1 3607 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3608 return NOTIFY_DONE;
3609
0883beca 3610 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3611 if (bond->dev == event_dev) {
3612 switch (event) {
3613 case NETDEV_UP:
3614 bond->master_ip = ifa->ifa_local;
3615 return NOTIFY_OK;
3616 case NETDEV_DOWN:
3617 bond->master_ip = bond_glean_dev_ip(bond->dev);
3618 return NOTIFY_OK;
3619 default:
3620 return NOTIFY_DONE;
3621 }
3622 }
3623
0883beca 3624 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3625 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3626 if (vlan_dev == event_dev) {
3627 switch (event) {
3628 case NETDEV_UP:
3629 vlan->vlan_ip = ifa->ifa_local;
3630 return NOTIFY_OK;
3631 case NETDEV_DOWN:
3632 vlan->vlan_ip =
3633 bond_glean_dev_ip(vlan_dev);
3634 return NOTIFY_OK;
3635 default:
3636 return NOTIFY_DONE;
3637 }
3638 }
3639 }
3640 }
3641 return NOTIFY_DONE;
3642}
3643
1da177e4
LT		 3644static struct notifier_block bond_netdev_notifier = {
3645 .notifier_call = bond_netdev_event,
3646};
3647
c3ade5ca
JV		 3648static struct notifier_block bond_inetaddr_notifier = {
3649 .notifier_call = bond_inetaddr_event,
3650};
3651
1da177e4
LT		 3652/*-------------------------- Packet type handling ---------------------------*/
3653
3654/* register to receive lacpdus on a bond */
3655static void bond_register_lacpdu(struct bonding *bond)
3656{
3657 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3658
3659 /* initialize packet type */
3660 pk_type->type = PKT_TYPE_LACPDU;
3661 pk_type->dev = bond->dev;
3662 pk_type->func = bond_3ad_lacpdu_recv;
3663
3664 dev_add_pack(pk_type);
3665}
3666
3667/* unregister to receive lacpdus on a bond */
3668static void bond_unregister_lacpdu(struct bonding *bond)
3669{
3670 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3671}
3672
f5b2b966
JV		 3673void bond_register_arp(struct bonding *bond)
3674{
3675 struct packet_type *pt = &bond->arp_mon_pt;
3676
c4f283b1
JV		 3677 if (pt->type)
3678 return;
3679
f5b2b966 3680 pt->type = htons(ETH_P_ARP);
e245cb71 3681 pt->dev = bond->dev;
f5b2b966
JV		 3682 pt->func = bond_arp_rcv;
3683 dev_add_pack(pt);
3684}
3685
3686void bond_unregister_arp(struct bonding *bond)
3687{
c4f283b1
JV		 3688 struct packet_type *pt = &bond->arp_mon_pt;
3689
3690 dev_remove_pack(pt);
3691 pt->type = 0;
f5b2b966
JV		 3692}
3693
169a3e66
JV		 3694/*---------------------------- Hashing Policies -----------------------------*/
3695
6f6652be
JV		 3696/*
3697 * Hash for the output device based upon layer 2 and layer 3 data. If
3698 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3699 */
3700static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3701 struct net_device *bond_dev, int count)
3702{
3703 struct ethhdr *data = (struct ethhdr *)skb->data;
3704 struct iphdr *iph = ip_hdr(skb);
3705
f14c4e4e 3706 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be
JV		 3707 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3708 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3709 }
3710
3711 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3712}
3713
169a3e66 3714/*
59c51591 3715 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3716 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3717 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3718 */
3719static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3720 struct net_device *bond_dev, int count)
3721{
3722 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3723 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3724 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3725 int layer4_xor = 0;
3726
f14c4e4e
BH		 3727 if (skb->protocol == htons(ETH_P_IP)) {
3728 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3729 (iph->protocol == IPPROTO_TCP ||
3730 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3731 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3732 }
3733 return (layer4_xor ^
3734 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3735
3736 }
3737
3738 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3739}
3740
3741/*
3742 * Hash for the output device based upon layer 2 data
3743 */
3744static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3745 struct net_device *bond_dev, int count)
3746{
3747 struct ethhdr *data = (struct ethhdr *)skb->data;
3748
3749 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3750}
3751
1da177e4
LT		 3752/*-------------------------- Device entry points ----------------------------*/
3753
3754static int bond_open(struct net_device *bond_dev)
3755{
3756 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 3757
3758 bond->kill_timers = 0;
3759
3760 if ((bond->params.mode == BOND_MODE_TLB) ||
3761 (bond->params.mode == BOND_MODE_ALB)) {
1da177e4
LT		 3762 /* bond_alb_initialize must be called before the timer
3763 * is started.
3764 */
3765 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3766 /* something went wrong - fail the open operation */
3767 return -1;
3768 }
3769
1b76b316
JV		 3770 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3771 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3772 }
3773
3774 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3775 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3776 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3777 }
3778
3779 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3780 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3781 INIT_DELAYED_WORK(&bond->arp_work,
3782 bond_activebackup_arp_mon);
3783 else
3784 INIT_DELAYED_WORK(&bond->arp_work,
3785 bond_loadbalance_arp_mon);
3786
3787 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3788 if (bond->params.arp_validate)
3789 bond_register_arp(bond);
1da177e4
LT		 3790 }
3791
3792 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3793 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3794 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3795 /* register to receive LACPDUs */
3796 bond_register_lacpdu(bond);
3797 }
3798
3799 return 0;
3800}
3801
3802static int bond_close(struct net_device *bond_dev)
3803{
3804 struct bonding *bond = bond_dev->priv;
3805
3806 if (bond->params.mode == BOND_MODE_8023AD) {
3807 /* Unregister the receive of LACPDUs */
3808 bond_unregister_lacpdu(bond);
3809 }
3810
f5b2b966
JV		 3811 if (bond->params.arp_validate)
3812 bond_unregister_arp(bond);
3813
1da177e4
LT		 3814 write_lock_bh(&bond->lock);
3815
b59f9f74 3816 bond->send_grat_arp = 0;
1da177e4
LT		 3817
3818 /* signal timers not to re-arm */
3819 bond->kill_timers = 1;
3820
3821 write_unlock_bh(&bond->lock);
3822
1da177e4 3823 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3824 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3825 }
3826
3827 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3828 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3829 }
3830
3831 switch (bond->params.mode) {
3832 case BOND_MODE_8023AD:
1b76b316 3833 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3834 break;
3835 case BOND_MODE_TLB:
3836 case BOND_MODE_ALB:
1b76b316 3837 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3838 break;
3839 default:
3840 break;
3841 }
3842
1da177e4
LT		 3843
3844 if ((bond->params.mode == BOND_MODE_TLB) ||
3845 (bond->params.mode == BOND_MODE_ALB)) {
3846 /* Must be called only after all
3847 * slaves have been released
3848 */
3849 bond_alb_deinitialize(bond);
3850 }
3851
3852 return 0;
3853}
3854
3855static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3856{
3857 struct bonding *bond = bond_dev->priv;
3858 struct net_device_stats *stats = &(bond->stats), *sstats;
2439f9eb 3859 struct net_device_stats local_stats;
1da177e4
LT		 3860 struct slave *slave;
3861 int i;
3862
2439f9eb 3863 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3864
3865 read_lock_bh(&bond->lock);
3866
3867 bond_for_each_slave(bond, slave, i) {
c45d286e 3868 sstats = slave->dev->get_stats(slave->dev);
2439f9eb
AG		 3869 local_stats.rx_packets += sstats->rx_packets;
3870 local_stats.rx_bytes += sstats->rx_bytes;
3871 local_stats.rx_errors += sstats->rx_errors;
3872 local_stats.rx_dropped += sstats->rx_dropped;
3873
3874 local_stats.tx_packets += sstats->tx_packets;
3875 local_stats.tx_bytes += sstats->tx_bytes;
3876 local_stats.tx_errors += sstats->tx_errors;
3877 local_stats.tx_dropped += sstats->tx_dropped;
3878
3879 local_stats.multicast += sstats->multicast;
3880 local_stats.collisions += sstats->collisions;
3881
3882 local_stats.rx_length_errors += sstats->rx_length_errors;
3883 local_stats.rx_over_errors += sstats->rx_over_errors;
3884 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3885 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3886 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3887 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3888
3889 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3890 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3891 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3892 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3893 local_stats.tx_window_errors += sstats->tx_window_errors;
3894 }
3895
3896 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3897
3898 read_unlock_bh(&bond->lock);
3899
3900 return stats;
3901}
3902
3903static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3904{
3905 struct net_device *slave_dev = NULL;
3906 struct ifbond k_binfo;
3907 struct ifbond __user *u_binfo = NULL;
3908 struct ifslave k_sinfo;
3909 struct ifslave __user *u_sinfo = NULL;
3910 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3911 int res = 0;
3912
3913 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3914 bond_dev->name, cmd);
3915
3916 switch (cmd) {
1da177e4
LT		 3917 case SIOCGMIIPHY:
3918 mii = if_mii(ifr);
3919 if (!mii) {
3920 return -EINVAL;
3921 }
3922 mii->phy_id = 0;
3923 /* Fall Through */
3924 case SIOCGMIIREG:
3925 /*
3926 * We do this again just in case we were called by SIOCGMIIREG
3927 * instead of SIOCGMIIPHY.
3928 */
3929 mii = if_mii(ifr);
3930 if (!mii) {
3931 return -EINVAL;
3932 }
3933
3934 if (mii->reg_num == 1) {
3935 struct bonding *bond = bond_dev->priv;
3936 mii->val_out = 0;
6603a6f2 3937 read_lock(&bond->lock);
1da177e4 3938 read_lock(&bond->curr_slave_lock);
4e140079 3939 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3940 mii->val_out = BMSR_LSTATUS;
3941 }
3942 read_unlock(&bond->curr_slave_lock);
6603a6f2 3943 read_unlock(&bond->lock);
1da177e4
LT		 3944 }
3945
3946 return 0;
3947 case BOND_INFO_QUERY_OLD:
3948 case SIOCBONDINFOQUERY:
3949 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3950
3951 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3952 return -EFAULT;
3953 }
3954
3955 res = bond_info_query(bond_dev, &k_binfo);
3956 if (res == 0) {
3957 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3958 return -EFAULT;
3959 }
3960 }
3961
3962 return res;
3963 case BOND_SLAVE_INFO_QUERY_OLD:
3964 case SIOCBONDSLAVEINFOQUERY:
3965 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3966
3967 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3968 return -EFAULT;
3969 }
3970
3971 res = bond_slave_info_query(bond_dev, &k_sinfo);
3972 if (res == 0) {
3973 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3974 return -EFAULT;
3975 }
3976 }
3977
3978 return res;
3979 default:
3980 /* Go on */
3981 break;
3982 }
3983
3984 if (!capable(CAP_NET_ADMIN)) {
3985 return -EPERM;
3986 }
3987
b76cdba9 3988 down_write(&(bonding_rwsem));
881d966b 3989 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT		 3990
3991 dprintk("slave_dev=%p: \n", slave_dev);
3992
3993 if (!slave_dev) {
3994 res = -ENODEV;
3995 } else {
3996 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3997 switch (cmd) {
3998 case BOND_ENSLAVE_OLD:
3999 case SIOCBONDENSLAVE:
4000 res = bond_enslave(bond_dev, slave_dev);
4001 break;
4002 case BOND_RELEASE_OLD:
4003 case SIOCBONDRELEASE:
4004 res = bond_release(bond_dev, slave_dev);
4005 break;
4006 case BOND_SETHWADDR_OLD:
4007 case SIOCBONDSETHWADDR:
4008 res = bond_sethwaddr(bond_dev, slave_dev);
4009 break;
4010 case BOND_CHANGE_ACTIVE_OLD:
4011 case SIOCBONDCHANGEACTIVE:
4012 res = bond_ioctl_change_active(bond_dev, slave_dev);
4013 break;
4014 default:
4015 res = -EOPNOTSUPP;
4016 }
4017
4018 dev_put(slave_dev);
4019 }
4020
b76cdba9 4021 up_write(&(bonding_rwsem));
1da177e4
LT		 4022 return res;
4023}
4024
4025static void bond_set_multicast_list(struct net_device *bond_dev)
4026{
4027 struct bonding *bond = bond_dev->priv;
4028 struct dev_mc_list *dmi;
4029
1da177e4
LT		 4030 /*
4031 * Do promisc before checking multicast_mode
4032 */
4033 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4034 /*
4035 * FIXME: Need to handle the error when one of the multi-slaves
4036 * encounters error.
4037 */
1da177e4
LT		 4038 bond_set_promiscuity(bond, 1);
4039 }
4040
4041 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4042 bond_set_promiscuity(bond, -1);
4043 }
4044
4045 /* set allmulti flag to slaves */
4046 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4047 /*
4048 * FIXME: Need to handle the error when one of the multi-slaves
4049 * encounters error.
4050 */
1da177e4
LT		 4051 bond_set_allmulti(bond, 1);
4052 }
4053
4054 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4055 bond_set_allmulti(bond, -1);
4056 }
4057
80ee5ad2
JV		 4058 read_lock(&bond->lock);
4059
1da177e4
LT		 4060 bond->flags = bond_dev->flags;
4061
4062 /* looking for addresses to add to slaves' mc list */
4063 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4064 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4065 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4066 }
4067 }
4068
4069 /* looking for addresses to delete from slaves' list */
4070 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4071 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4072 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4073 }
4074 }
4075
4076 /* save master's multicast list */
4077 bond_mc_list_destroy(bond);
4078 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4079
80ee5ad2 4080 read_unlock(&bond->lock);
1da177e4
LT		 4081}
4082
4083/*
4084 * Change the MTU of all of a master's slaves to match the master
4085 */
4086static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4087{
4088 struct bonding *bond = bond_dev->priv;
4089 struct slave *slave, *stop_at;
4090 int res = 0;
4091 int i;
4092
4093 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4094 (bond_dev ? bond_dev->name : "None"), new_mtu);
4095
4096 /* Can't hold bond->lock with bh disabled here since
4097 * some base drivers panic. On the other hand we can't
4098 * hold bond->lock without bh disabled because we'll
4099 * deadlock. The only solution is to rely on the fact
4100 * that we're under rtnl_lock here, and the slaves
4101 * list won't change. This doesn't solve the problem
4102 * of setting the slave's MTU while it is
4103 * transmitting, but the assumption is that the base
4104 * driver can handle that.
4105 *
4106 * TODO: figure out a way to safely iterate the slaves
4107 * list, but without holding a lock around the actual
4108 * call to the base driver.
4109 */
4110
4111 bond_for_each_slave(bond, slave, i) {
4112 dprintk("s %p s->p %p c_m %p\n", slave,
4113 slave->prev, slave->dev->change_mtu);
e944ef79 4114
1da177e4
LT		 4115 res = dev_set_mtu(slave->dev, new_mtu);
4116
4117 if (res) {
4118 /* If we failed to set the slave's mtu to the new value
4119 * we must abort the operation even in ACTIVE_BACKUP
4120 * mode, because if we allow the backup slaves to have
4121 * different mtu values than the active slave we'll
4122 * need to change their mtu when doing a failover. That
4123 * means changing their mtu from timer context, which
4124 * is probably not a good idea.
4125 */
4126 dprintk("err %d %s\n", res, slave->dev->name);
4127 goto unwind;
4128 }
4129 }
4130
4131 bond_dev->mtu = new_mtu;
4132
4133 return 0;
4134
4135unwind:
4136 /* unwind from head to the slave that failed */
4137 stop_at = slave;
4138 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4139 int tmp_res;
4140
4141 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4142 if (tmp_res) {
4143 dprintk("unwind err %d dev %s\n", tmp_res,
4144 slave->dev->name);
4145 }
4146 }
4147
4148 return res;
4149}
4150
4151/*
4152 * Change HW address
4153 *
4154 * Note that many devices must be down to change the HW address, and
4155 * downing the master releases all slaves. We can make bonds full of
4156 * bonding devices to test this, however.
4157 */
4158static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4159{
4160 struct bonding *bond = bond_dev->priv;
4161 struct sockaddr *sa = addr, tmp_sa;
4162 struct slave *slave, *stop_at;
4163 int res = 0;
4164 int i;
4165
4166 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4167
dd957c57 4168 /*
3915c1e8
JV		 4169 * If fail_over_mac is set to active, do nothing and return
4170 * success. Returning an error causes ifenslave to fail.
dd957c57 4171 */
3915c1e8 4172 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4173 return 0;
2ab82852 4174
1da177e4
LT		 4175 if (!is_valid_ether_addr(sa->sa_data)) {
4176 return -EADDRNOTAVAIL;
4177 }
4178
4179 /* Can't hold bond->lock with bh disabled here since
4180 * some base drivers panic. On the other hand we can't
4181 * hold bond->lock without bh disabled because we'll
4182 * deadlock. The only solution is to rely on the fact
4183 * that we're under rtnl_lock here, and the slaves
4184 * list won't change. This doesn't solve the problem
4185 * of setting the slave's hw address while it is
4186 * transmitting, but the assumption is that the base
4187 * driver can handle that.
4188 *
4189 * TODO: figure out a way to safely iterate the slaves
4190 * list, but without holding a lock around the actual
4191 * call to the base driver.
4192 */
4193
4194 bond_for_each_slave(bond, slave, i) {
4195 dprintk("slave %p %s\n", slave, slave->dev->name);
4196
4197 if (slave->dev->set_mac_address == NULL) {
4198 res = -EOPNOTSUPP;
4199 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4200 goto unwind;
4201 }
4202
4203 res = dev_set_mac_address(slave->dev, addr);
4204 if (res) {
4205 /* TODO: consider downing the slave
4206 * and retry ?
4207 * User should expect communications
4208 * breakage anyway until ARP finish
4209 * updating, so...
4210 */
4211 dprintk("err %d %s\n", res, slave->dev->name);
4212 goto unwind;
4213 }
4214 }
4215
4216 /* success */
4217 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4218 return 0;
4219
4220unwind:
4221 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4222 tmp_sa.sa_family = bond_dev->type;
4223
4224 /* unwind from head to the slave that failed */
4225 stop_at = slave;
4226 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4227 int tmp_res;
4228
4229 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4230 if (tmp_res) {
4231 dprintk("unwind err %d dev %s\n", tmp_res,
4232 slave->dev->name);
4233 }
4234 }
4235
4236 return res;
4237}
4238
4239static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4240{
4241 struct bonding *bond = bond_dev->priv;
4242 struct slave *slave, *start_at;
cf5f9044 4243 int i, slave_no, res = 1;
1da177e4
LT		 4244
4245 read_lock(&bond->lock);
4246
4247 if (!BOND_IS_OK(bond)) {
4248 goto out;
4249 }
4250
cf5f9044
JV		 4251 /*
4252 * Concurrent TX may collide on rr_tx_counter; we accept that
4253 * as being rare enough not to justify using an atomic op here
4254 */
4255 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4256
cf5f9044
JV		 4257 bond_for_each_slave(bond, slave, i) {
4258 slave_no--;
4259 if (slave_no < 0) {
4260 break;
4261 }
1da177e4
LT		 4262 }
4263
cf5f9044 4264 start_at = slave;
1da177e4
LT		 4265 bond_for_each_slave_from(bond, slave, i, start_at) {
4266 if (IS_UP(slave->dev) &&
4267 (slave->link == BOND_LINK_UP) &&
4268 (slave->state == BOND_STATE_ACTIVE)) {
4269 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4270 break;
4271 }
4272 }
4273
1da177e4
LT		 4274out:
4275 if (res) {
4276 /* no suitable interface, frame not sent */
4277 dev_kfree_skb(skb);
4278 }
4279 read_unlock(&bond->lock);
4280 return 0;
4281}
4282
075897ce 4283
1da177e4
LT		 4284/*
4285 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4286 * the bond has a usable interface.
4287 */
4288static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4289{
4290 struct bonding *bond = bond_dev->priv;
4291 int res = 1;
4292
1da177e4
LT		 4293 read_lock(&bond->lock);
4294 read_lock(&bond->curr_slave_lock);
4295
4296 if (!BOND_IS_OK(bond)) {
4297 goto out;
4298 }
4299
075897ce
JL		 4300 if (!bond->curr_active_slave)
4301 goto out;
4302
075897ce
JL		 4303 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4304
1da177e4
LT		 4305out:
4306 if (res) {
4307 /* no suitable interface, frame not sent */
4308 dev_kfree_skb(skb);
4309 }
4310 read_unlock(&bond->curr_slave_lock);
4311 read_unlock(&bond->lock);
4312 return 0;
4313}
4314
4315/*
169a3e66
JV		 4316 * In bond_xmit_xor() , we determine the output device by using a pre-
4317 * determined xmit_hash_policy(), If the selected device is not enabled,
4318 * find the next active slave.
1da177e4
LT		 4319 */
4320static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4321{
4322 struct bonding *bond = bond_dev->priv;
1da177e4
LT		 4323 struct slave *slave, *start_at;
4324 int slave_no;
4325 int i;
4326 int res = 1;
4327
4328 read_lock(&bond->lock);
4329
4330 if (!BOND_IS_OK(bond)) {
4331 goto out;
4332 }
4333
169a3e66 4334 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4335
4336 bond_for_each_slave(bond, slave, i) {
4337 slave_no--;
4338 if (slave_no < 0) {
4339 break;
4340 }
4341 }
4342
4343 start_at = slave;
4344
4345 bond_for_each_slave_from(bond, slave, i, start_at) {
4346 if (IS_UP(slave->dev) &&
4347 (slave->link == BOND_LINK_UP) &&
4348 (slave->state == BOND_STATE_ACTIVE)) {
4349 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4350 break;
4351 }
4352 }
4353
4354out:
4355 if (res) {
4356 /* no suitable interface, frame not sent */
4357 dev_kfree_skb(skb);
4358 }
4359 read_unlock(&bond->lock);
4360 return 0;
4361}
4362
4363/*
4364 * in broadcast mode, we send everything to all usable interfaces.
4365 */
4366static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4367{
4368 struct bonding *bond = bond_dev->priv;
4369 struct slave *slave, *start_at;
4370 struct net_device *tx_dev = NULL;
4371 int i;
4372 int res = 1;
4373
4374 read_lock(&bond->lock);
4375
4376 if (!BOND_IS_OK(bond)) {
4377 goto out;
4378 }
4379
4380 read_lock(&bond->curr_slave_lock);
4381 start_at = bond->curr_active_slave;
4382 read_unlock(&bond->curr_slave_lock);
4383
4384 if (!start_at) {
4385 goto out;
4386 }
4387
4388 bond_for_each_slave_from(bond, slave, i, start_at) {
4389 if (IS_UP(slave->dev) &&
4390 (slave->link == BOND_LINK_UP) &&
4391 (slave->state == BOND_STATE_ACTIVE)) {
4392 if (tx_dev) {
4393 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4394 if (!skb2) {
4395 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4396 ": %s: Error: bond_xmit_broadcast(): "
4397 "skb_clone() failed\n",
4398 bond_dev->name);
1da177e4
LT		 4399 continue;
4400 }
4401
4402 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4403 if (res) {
4404 dev_kfree_skb(skb2);
4405 continue;
4406 }
4407 }
4408 tx_dev = slave->dev;
4409 }
4410 }
4411
4412 if (tx_dev) {
4413 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4414 }
4415
4416out:
4417 if (res) {
4418 /* no suitable interface, frame not sent */
4419 dev_kfree_skb(skb);
4420 }
4421 /* frame sent to all suitable interfaces */
4422 read_unlock(&bond->lock);
4423 return 0;
4424}
4425
4426/*------------------------- Device initialization ---------------------------*/
4427
6f6652be
JV		 4428static void bond_set_xmit_hash_policy(struct bonding *bond)
4429{
4430 switch (bond->params.xmit_policy) {
4431 case BOND_XMIT_POLICY_LAYER23:
4432 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4433 break;
4434 case BOND_XMIT_POLICY_LAYER34:
4435 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4436 break;
4437 case BOND_XMIT_POLICY_LAYER2:
4438 default:
4439 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4440 break;
4441 }
4442}
4443
1da177e4
LT		 4444/*
4445 * set bond mode specific net device operations
4446 */
a77b5325 4447void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4448{
169a3e66
JV		 4449 struct net_device *bond_dev = bond->dev;
4450
1da177e4
LT		 4451 switch (mode) {
4452 case BOND_MODE_ROUNDROBIN:
4453 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4454 break;
4455 case BOND_MODE_ACTIVEBACKUP:
4456 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4457 break;
4458 case BOND_MODE_XOR:
4459 bond_dev->hard_start_xmit = bond_xmit_xor;
6f6652be 4460 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4461 break;
4462 case BOND_MODE_BROADCAST:
4463 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4464 break;
4465 case BOND_MODE_8023AD:
8f903c70 4466 bond_set_master_3ad_flags(bond);
1da177e4 4467 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
6f6652be 4468 bond_set_xmit_hash_policy(bond);
1da177e4 4469 break;
1da177e4 4470 case BOND_MODE_ALB:
8f903c70
JV		 4471 bond_set_master_alb_flags(bond);
4472 /* FALLTHRU */
4473 case BOND_MODE_TLB:
1da177e4
LT		 4474 bond_dev->hard_start_xmit = bond_alb_xmit;
4475 bond_dev->set_mac_address = bond_alb_set_mac_address;
4476 break;
4477 default:
4478 /* Should never happen, mode already checked */
4479 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4480 ": %s: Error: Unknown bonding mode %d\n",
4481 bond_dev->name,
1da177e4
LT		 4482 mode);
4483 break;
4484 }
4485}
4486
217df670
JV		 4487static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4488 struct ethtool_drvinfo *drvinfo)
4489{
4490 strncpy(drvinfo->driver, DRV_NAME, 32);
4491 strncpy(drvinfo->version, DRV_VERSION, 32);
4492 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4493}
4494
7282d491 4495static const struct ethtool_ops bond_ethtool_ops = {
217df670 4496 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac
SH		 4497 .get_link = ethtool_op_get_link,
4498 .get_tx_csum = ethtool_op_get_tx_csum,
4499 .get_sg = ethtool_op_get_sg,
4500 .get_tso = ethtool_op_get_tso,
4501 .get_ufo = ethtool_op_get_ufo,
4502 .get_flags = ethtool_op_get_flags,
8531c5ff
AK		 4503};
4504
1da177e4
LT		 4505/*
4506 * Does not allocate but creates a /proc entry.
4507 * Allowed to fail.
4508 */
3c535952 4509static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4
LT		 4510{
4511 struct bonding *bond = bond_dev->priv;
4512
4513 dprintk("Begin bond_init for %s\n", bond_dev->name);
4514
4515 /* initialize rwlocks */
4516 rwlock_init(&bond->lock);
4517 rwlock_init(&bond->curr_slave_lock);
4518
4519 bond->params = *params; /* copy params struct */
4520
1b76b316
JV		 4521 bond->wq = create_singlethread_workqueue(bond_dev->name);
4522 if (!bond->wq)
4523 return -ENOMEM;
4524
1da177e4
LT		 4525 /* Initialize pointers */
4526 bond->first_slave = NULL;
4527 bond->curr_active_slave = NULL;
4528 bond->current_arp_slave = NULL;
4529 bond->primary_slave = NULL;
4530 bond->dev = bond_dev;
1053f62c 4531 bond->send_grat_arp = 0;
d90a162a 4532 bond->setup_by_slave = 0;
1da177e4
LT		 4533 INIT_LIST_HEAD(&bond->vlan_list);
4534
4535 /* Initialize the device entry points */
4536 bond_dev->open = bond_open;
4537 bond_dev->stop = bond_close;
4538 bond_dev->get_stats = bond_get_stats;
4539 bond_dev->do_ioctl = bond_do_ioctl;
8531c5ff 4540 bond_dev->ethtool_ops = &bond_ethtool_ops;
1da177e4
LT		 4541 bond_dev->set_multicast_list = bond_set_multicast_list;
4542 bond_dev->change_mtu = bond_change_mtu;
4543 bond_dev->set_mac_address = bond_set_mac_address;
3a1521b7 4544 bond_dev->validate_addr = NULL;
1da177e4 4545
169a3e66 4546 bond_set_mode_ops(bond, bond->params.mode);
1da177e4
LT		 4547
4548 bond_dev->destructor = free_netdev;
4549
4550 /* Initialize the device options */
4551 bond_dev->tx_queue_len = 0;
4552 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4553 bond_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 4554
4555 /* At first, we block adding VLANs. That's the only way to
4556 * prevent problems that occur when adding VLANs over an
4557 * empty bond. The block will be removed once non-challenged
4558 * slaves are enslaved.
4559 */
4560 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4561
932ff279 4562 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4563 * transmitting */
4564 bond_dev->features |= NETIF_F_LLTX;
4565
4566 /* By default, we declare the bond to be fully
4567 * VLAN hardware accelerated capable. Special
4568 * care is taken in the various xmit functions
4569 * when there are slaves that are not hw accel
4570 * capable
4571 */
4572 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4573 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4574 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4575 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4576 NETIF_F_HW_VLAN_RX |
4577 NETIF_F_HW_VLAN_FILTER);
4578
4579#ifdef CONFIG_PROC_FS
4580 bond_create_proc_entry(bond);
4581#endif
1da177e4
LT		 4582 list_add_tail(&bond->bond_list, &bond_dev_list);
4583
4584 return 0;
4585}
4586
4587/* De-initialize device specific data.
4588 * Caller must hold rtnl_lock.
4589 */
c50b85d0 4590static void bond_deinit(struct net_device *bond_dev)
1da177e4
LT		 4591{
4592 struct bonding *bond = bond_dev->priv;
4593
4594 list_del(&bond->bond_list);
4595
4596#ifdef CONFIG_PROC_FS
4597 bond_remove_proc_entry(bond);
4598#endif
4599}
4600
fdaea7a9
JV		 4601static void bond_work_cancel_all(struct bonding *bond)
4602{
4603 write_lock_bh(&bond->lock);
4604 bond->kill_timers = 1;
4605 write_unlock_bh(&bond->lock);
4606
4607 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4608 cancel_delayed_work(&bond->mii_work);
4609
4610 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4611 cancel_delayed_work(&bond->arp_work);
4612
4613 if (bond->params.mode == BOND_MODE_ALB &&
4614 delayed_work_pending(&bond->alb_work))
4615 cancel_delayed_work(&bond->alb_work);
4616
4617 if (bond->params.mode == BOND_MODE_8023AD &&
4618 delayed_work_pending(&bond->ad_work))
4619 cancel_delayed_work(&bond->ad_work);
4620}
4621
1da177e4
LT		 4622/* Unregister and free all bond devices.
4623 * Caller must hold rtnl_lock.
4624 */
4625static void bond_free_all(void)
4626{
4627 struct bonding *bond, *nxt;
4628
4629 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4630 struct net_device *bond_dev = bond->dev;
4631
fdaea7a9 4632 bond_work_cancel_all(bond);
b9e40857 4633 netif_addr_lock_bh(bond_dev);
70298705 4634 bond_mc_list_destroy(bond);
b9e40857 4635 netif_addr_unlock_bh(bond_dev);
70298705 4636 /* Release the bonded slaves */
4637 bond_release_all(bond_dev);
92b41daa 4638 bond_destroy(bond);
1da177e4
LT		 4639 }
4640
4641#ifdef CONFIG_PROC_FS
4642 bond_destroy_proc_dir();
4643#endif
4644}
4645
4646/*------------------------- Module initialization ---------------------------*/
4647
4648/*
4649 * Convert string input module parms. Accept either the
ece95f7f
JV		 4650 * number of the mode or its string name. A bit complicated because
4651 * some mode names are substrings of other names, and calls from sysfs
4652 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4653 */
ece95f7f 4654int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4655{
ece95f7f 4656 int mode = -1, i, rv;
a42e534f 4657 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4658
a42e534f
JV		 4659 for (p = (char *)buf; *p; p++)
4660 if (!(isdigit(*p) || isspace(*p)))
4661 break;
4662
4663 if (*p)
ece95f7f 4664 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4665 else
4666 rv = sscanf(buf, "%d", &mode);
4667
4668 if (!rv)
4669 return -1;
1da177e4
LT		 4670
4671 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4672 if (mode == tbl[i].mode)
4673 return tbl[i].mode;
4674 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4675 return tbl[i].mode;
1da177e4
LT		 4676 }
4677
4678 return -1;
4679}
4680
4681static int bond_check_params(struct bond_params *params)
4682{
3915c1e8 4683 int arp_validate_value, fail_over_mac_value;
f5b2b966 4684
1da177e4
LT		 4685 /*
4686 * Convert string parameters.
4687 */
4688 if (mode) {
4689 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4690 if (bond_mode == -1) {
4691 printk(KERN_ERR DRV_NAME
4692 ": Error: Invalid bonding mode \"%s\"\n",
4693 mode == NULL ? "NULL" : mode);
4694 return -EINVAL;
4695 }
4696 }
4697
169a3e66
JV		 4698 if (xmit_hash_policy) {
4699 if ((bond_mode != BOND_MODE_XOR) &&
4700 (bond_mode != BOND_MODE_8023AD)) {
4701 printk(KERN_INFO DRV_NAME
4702 ": xor_mode param is irrelevant in mode %s\n",
4703 bond_mode_name(bond_mode));
4704 } else {
4705 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4706 xmit_hashtype_tbl);
4707 if (xmit_hashtype == -1) {
4708 printk(KERN_ERR DRV_NAME
4709 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4710 xmit_hash_policy == NULL ? "NULL" :
4711 xmit_hash_policy);
4712 return -EINVAL;
4713 }
4714 }
4715 }
4716
1da177e4
LT		 4717 if (lacp_rate) {
4718 if (bond_mode != BOND_MODE_8023AD) {
4719 printk(KERN_INFO DRV_NAME
4720 ": lacp_rate param is irrelevant in mode %s\n",
4721 bond_mode_name(bond_mode));
4722 } else {
4723 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4724 if (lacp_fast == -1) {
4725 printk(KERN_ERR DRV_NAME
4726 ": Error: Invalid lacp rate \"%s\"\n",
4727 lacp_rate == NULL ? "NULL" : lacp_rate);
4728 return -EINVAL;
4729 }
4730 }
4731 }
4732
b8a9787e 4733 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4734 printk(KERN_WARNING DRV_NAME
4735 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4736 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4737 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4738 max_bonds = BOND_DEFAULT_MAX_BONDS;
4739 }
4740
4741 if (miimon < 0) {
4742 printk(KERN_WARNING DRV_NAME
4743 ": Warning: miimon module parameter (%d), "
4744 "not in range 0-%d, so it was reset to %d\n",
4745 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4746 miimon = BOND_LINK_MON_INTERV;
4747 }
4748
4749 if (updelay < 0) {
4750 printk(KERN_WARNING DRV_NAME
4751 ": Warning: updelay module parameter (%d), "
4752 "not in range 0-%d, so it was reset to 0\n",
4753 updelay, INT_MAX);
4754 updelay = 0;
4755 }
4756
4757 if (downdelay < 0) {
4758 printk(KERN_WARNING DRV_NAME
4759 ": Warning: downdelay module parameter (%d), "
4760 "not in range 0-%d, so it was reset to 0\n",
4761 downdelay, INT_MAX);
4762 downdelay = 0;
4763 }
4764
4765 if ((use_carrier != 0) && (use_carrier != 1)) {
4766 printk(KERN_WARNING DRV_NAME
4767 ": Warning: use_carrier module parameter (%d), "
4768 "not of valid value (0/1), so it was set to 1\n",
4769 use_carrier);
4770 use_carrier = 1;
4771 }
4772
7893b249
MS		 4773 if (num_grat_arp < 0 || num_grat_arp > 255) {
4774 printk(KERN_WARNING DRV_NAME
4775 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4776 "was reset to 1 \n", num_grat_arp);
4777 num_grat_arp = 1;
4778 }
4779
1da177e4
LT		 4780 /* reset values for 802.3ad */
4781 if (bond_mode == BOND_MODE_8023AD) {
4782 if (!miimon) {
4783 printk(KERN_WARNING DRV_NAME
4784 ": Warning: miimon must be specified, "
4785 "otherwise bonding will not detect link "
4786 "failure, speed and duplex which are "
4787 "essential for 802.3ad operation\n");
4788 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4789 miimon = 100;
4790 }
4791 }
4792
4793 /* reset values for TLB/ALB */
4794 if ((bond_mode == BOND_MODE_TLB) ||
4795 (bond_mode == BOND_MODE_ALB)) {
4796 if (!miimon) {
4797 printk(KERN_WARNING DRV_NAME
4798 ": Warning: miimon must be specified, "
4799 "otherwise bonding will not detect link "
4800 "failure and link speed which are essential "
4801 "for TLB/ALB load balancing\n");
4802 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4803 miimon = 100;
4804 }
4805 }
4806
4807 if (bond_mode == BOND_MODE_ALB) {
4808 printk(KERN_NOTICE DRV_NAME
4809 ": In ALB mode you might experience client "
4810 "disconnections upon reconnection of a link if the "
4811 "bonding module updelay parameter (%d msec) is "
4812 "incompatible with the forwarding delay time of the "
4813 "switch\n",
4814 updelay);
4815 }
4816
4817 if (!miimon) {
4818 if (updelay || downdelay) {
4819 /* just warn the user the up/down delay will have
4820 * no effect since miimon is zero...
4821 */
4822 printk(KERN_WARNING DRV_NAME
4823 ": Warning: miimon module parameter not set "
4824 "and updelay (%d) or downdelay (%d) module "
4825 "parameter is set; updelay and downdelay have "
4826 "no effect unless miimon is set\n",
4827 updelay, downdelay);
4828 }
4829 } else {
4830 /* don't allow arp monitoring */
4831 if (arp_interval) {
4832 printk(KERN_WARNING DRV_NAME
4833 ": Warning: miimon (%d) and arp_interval (%d) "
4834 "can't be used simultaneously, disabling ARP "
4835 "monitoring\n",
4836 miimon, arp_interval);
4837 arp_interval = 0;
4838 }
4839
4840 if ((updelay % miimon) != 0) {
4841 printk(KERN_WARNING DRV_NAME
4842 ": Warning: updelay (%d) is not a multiple "
4843 "of miimon (%d), updelay rounded to %d ms\n",
4844 updelay, miimon, (updelay / miimon) * miimon);
4845 }
4846
4847 updelay /= miimon;
4848
4849 if ((downdelay % miimon) != 0) {
4850 printk(KERN_WARNING DRV_NAME
4851 ": Warning: downdelay (%d) is not a multiple "
4852 "of miimon (%d), downdelay rounded to %d ms\n",
4853 downdelay, miimon,
4854 (downdelay / miimon) * miimon);
4855 }
4856
4857 downdelay /= miimon;
4858 }
4859
4860 if (arp_interval < 0) {
4861 printk(KERN_WARNING DRV_NAME
4862 ": Warning: arp_interval module parameter (%d) "
4863 ", not in range 0-%d, so it was reset to %d\n",
4864 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4865 arp_interval = BOND_LINK_ARP_INTERV;
4866 }
4867
4868 for (arp_ip_count = 0;
4869 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4870 arp_ip_count++) {
4871 /* not complete check, but should be good enough to
4872 catch mistakes */
4873 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4874 printk(KERN_WARNING DRV_NAME
4875 ": Warning: bad arp_ip_target module parameter "
4876 "(%s), ARP monitoring will not be performed\n",
4877 arp_ip_target[arp_ip_count]);
4878 arp_interval = 0;
4879 } else {
d3bb52b0 4880 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4881 arp_target[arp_ip_count] = ip;
4882 }
4883 }
4884
4885 if (arp_interval && !arp_ip_count) {
4886 /* don't allow arping if no arp_ip_target given... */
4887 printk(KERN_WARNING DRV_NAME
4888 ": Warning: arp_interval module parameter (%d) "
4889 "specified without providing an arp_ip_target "
4890 "parameter, arp_interval was reset to 0\n",
4891 arp_interval);
4892 arp_interval = 0;
4893 }
4894
f5b2b966
JV		 4895 if (arp_validate) {
4896 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4897 printk(KERN_ERR DRV_NAME
4898 ": arp_validate only supported in active-backup mode\n");
4899 return -EINVAL;
4900 }
4901 if (!arp_interval) {
4902 printk(KERN_ERR DRV_NAME
4903 ": arp_validate requires arp_interval\n");
4904 return -EINVAL;
4905 }
4906
4907 arp_validate_value = bond_parse_parm(arp_validate,
4908 arp_validate_tbl);
4909 if (arp_validate_value == -1) {
4910 printk(KERN_ERR DRV_NAME
4911 ": Error: invalid arp_validate \"%s\"\n",
4912 arp_validate == NULL ? "NULL" : arp_validate);
4913 return -EINVAL;
4914 }
4915 } else
4916 arp_validate_value = 0;
4917
1da177e4
LT		 4918 if (miimon) {
4919 printk(KERN_INFO DRV_NAME
4920 ": MII link monitoring set to %d ms\n",
4921 miimon);
4922 } else if (arp_interval) {
4923 int i;
4924
4925 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 4926 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4927 arp_interval,
4928 arp_validate_tbl[arp_validate_value].modename,
4929 arp_ip_count);
1da177e4
LT		 4930
4931 for (i = 0; i < arp_ip_count; i++)
4932 printk (" %s", arp_ip_target[i]);
4933
4934 printk("\n");
4935
b8a9787e 4936 } else if (max_bonds) {
1da177e4
LT		 4937 /* miimon and arp_interval not set, we need one so things
4938 * work as expected, see bonding.txt for details
4939 */
4940 printk(KERN_WARNING DRV_NAME
4941 ": Warning: either miimon or arp_interval and "
4942 "arp_ip_target module parameters must be specified, "
4943 "otherwise bonding will not detect link failures! see "
4944 "bonding.txt for details.\n");
4945 }
4946
4947 if (primary && !USES_PRIMARY(bond_mode)) {
4948 /* currently, using a primary only makes sense
4949 * in active backup, TLB or ALB modes
4950 */
4951 printk(KERN_WARNING DRV_NAME
4952 ": Warning: %s primary device specified but has no "
4953 "effect in %s mode\n",
4954 primary, bond_mode_name(bond_mode));
4955 primary = NULL;
4956 }
4957
3915c1e8
JV		 4958 if (fail_over_mac) {
4959 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4960 fail_over_mac_tbl);
4961 if (fail_over_mac_value == -1) {
4962 printk(KERN_ERR DRV_NAME
4963 ": Error: invalid fail_over_mac \"%s\"\n",
4964 arp_validate == NULL ? "NULL" : arp_validate);
4965 return -EINVAL;
4966 }
4967
4968 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4969 printk(KERN_WARNING DRV_NAME
4970 ": Warning: fail_over_mac only affects "
4971 "active-backup mode.\n");
4972 } else {
4973 fail_over_mac_value = BOND_FOM_NONE;
4974 }
dd957c57 4975
1da177e4
LT		 4976 /* fill params struct with the proper values */
4977 params->mode = bond_mode;
169a3e66 4978 params->xmit_policy = xmit_hashtype;
1da177e4 4979 params->miimon = miimon;
7893b249 4980 params->num_grat_arp = num_grat_arp;
1da177e4 4981 params->arp_interval = arp_interval;
f5b2b966 4982 params->arp_validate = arp_validate_value;
1da177e4
LT		 4983 params->updelay = updelay;
4984 params->downdelay = downdelay;
4985 params->use_carrier = use_carrier;
4986 params->lacp_fast = lacp_fast;
4987 params->primary[0] = 0;
3915c1e8 4988 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 4989
4990 if (primary) {
4991 strncpy(params->primary, primary, IFNAMSIZ);
4992 params->primary[IFNAMSIZ - 1] = 0;
4993 }
4994
4995 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4996
4997 return 0;
4998}
4999
0daa2303 5000static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 5001static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 5002
e8a0464c
DM		 5003static void bond_set_lockdep_class_one(struct net_device *dev,
5004 struct netdev_queue *txq,
5005 void *_unused)
c773e847
DM		 5006{
5007 lockdep_set_class(&txq->_xmit_lock,
5008 &bonding_netdev_xmit_lock_key);
5009}
5010
5011static void bond_set_lockdep_class(struct net_device *dev)
5012{
cf508b12
DM		 5013 lockdep_set_class(&dev->addr_list_lock,
5014 &bonding_netdev_addr_lock_key);
e8a0464c 5015 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 5016}
5017
dfe60397 5018/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5019 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5020 * Caller must NOT hold rtnl_lock; we need to release it here before we
5021 * set up our sysfs entries.
5022 */
0dd646fe 5023int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5024{
5025 struct net_device *bond_dev;
0883beca 5026 struct bonding *bond;
dfe60397
MW		 5027 int res;
5028
5029 rtnl_lock();
027ea041
JV		 5030 down_write(&bonding_rwsem);
5031
5032 /* Check to see if the bond already exists. */
4fe4763c 5033 if (name) {
0883beca 5034 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5035 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5036 printk(KERN_ERR DRV_NAME
027ea041 5037 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5038 name);
5039 res = -EPERM;
5040 goto out_rtnl;
5041 }
5042 }
027ea041 5043
e4b91c48
JV		 5044 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5045 ether_setup);
dfe60397
MW		 5046 if (!bond_dev) {
5047 printk(KERN_ERR DRV_NAME
5048 ": %s: eek! can't alloc netdev!\n",
5049 name);
5050 res = -ENOMEM;
5051 goto out_rtnl;
5052 }
5053
e4b91c48
JV		 5054 if (!name) {
5055 res = dev_alloc_name(bond_dev, "bond%d");
5056 if (res < 0)
5057 goto out_netdev;
5058 }
5059
dfe60397
MW		 5060 /* bond_init() must be called after dev_alloc_name() (for the
5061 * /proc files), but before register_netdevice(), because we
5062 * need to set function pointers.
5063 */
5064
5065 res = bond_init(bond_dev, params);
5066 if (res < 0) {
5067 goto out_netdev;
5068 }
5069
dfe60397
MW		 5070 res = register_netdevice(bond_dev);
5071 if (res < 0) {
5072 goto out_bond;
5073 }
0daa2303 5074
c773e847 5075 bond_set_lockdep_class(bond_dev);
0daa2303 5076
ff59c456
JV		 5077 netif_carrier_off(bond_dev);
5078
027ea041 5079 up_write(&bonding_rwsem);
dfe60397 5080 rtnl_unlock(); /* allows sysfs registration of net device */
b76cdba9 5081 res = bond_create_sysfs_entry(bond_dev->priv);
09c89279
JV		 5082 if (res < 0) {
5083 rtnl_lock();
027ea041 5084 down_write(&bonding_rwsem);
822973ba
PE		 5085 bond_deinit(bond_dev);
5086 unregister_netdevice(bond_dev);
5087 goto out_rtnl;
09c89279
JV		 5088 }
5089
5090 return 0;
5091
dfe60397
MW		 5092out_bond:
5093 bond_deinit(bond_dev);
5094out_netdev:
5095 free_netdev(bond_dev);
5096out_rtnl:
027ea041 5097 up_write(&bonding_rwsem);
dfe60397 5098 rtnl_unlock();
dfe60397
MW		 5099 return res;
5100}
5101
1da177e4
LT		 5102static int __init bonding_init(void)
5103{
1da177e4
LT		 5104 int i;
5105 int res;
0883beca 5106 struct bonding *bond;
1da177e4
LT		 5107
5108 printk(KERN_INFO "%s", version);
5109
dfe60397 5110 res = bond_check_params(&bonding_defaults);
1da177e4 5111 if (res) {
dfe60397 5112 goto out;
1da177e4
LT		 5113 }
5114
1da177e4
LT		 5115#ifdef CONFIG_PROC_FS
5116 bond_create_proc_dir();
5117#endif
027ea041
JV		 5118
5119 init_rwsem(&bonding_rwsem);
5120
1da177e4 5121 for (i = 0; i < max_bonds; i++) {
0dd646fe 5122 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5123 if (res)
5124 goto err;
1da177e4
LT		 5125 }
5126
b76cdba9
MW		 5127 res = bond_create_sysfs();
5128 if (res)
5129 goto err;
5130
1da177e4 5131 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5132 register_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5133
dfe60397
MW		 5134 goto out;
5135err:
0883beca 5136 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5137 bond_work_cancel_all(bond);
5138 destroy_workqueue(bond->wq);
5139 }
5140
ae68c398
PE		 5141 bond_destroy_sysfs();
5142
40abc270 5143 rtnl_lock();
1da177e4 5144 bond_free_all();
1da177e4 5145 rtnl_unlock();
dfe60397 5146out:
1da177e4 5147 return res;
dfe60397 5148
1da177e4
LT		 5149}
5150
5151static void __exit bonding_exit(void)
5152{
5153 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5154 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 5155
ae68c398
PE		 5156 bond_destroy_sysfs();
5157
1da177e4
LT		 5158 rtnl_lock();
5159 bond_free_all();
5160 rtnl_unlock();
5161}
5162
5163module_init(bonding_init);
5164module_exit(bonding_exit);
5165MODULE_LICENSE("GPL");
5166MODULE_VERSION(DRV_VERSION);
5167MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5168MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5169MODULE_SUPPORTED_DEVICE("most ethernet devices");
5170
5171/*
5172 * Local variables:
5173 * c-indent-level: 8
5174 * c-basic-offset: 8
5175 * tab-width: 8
5176 * End:
5177 */
5178
Clone of davem's net-next-2.6 tree