]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/bonding/bond_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bonding: convert to net_device_ops
[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;
305d552a 92static int num_unsol_na = 1;
1da177e4
LT		 93static int miimon = BOND_LINK_MON_INTERV;
94static int updelay = 0;
95static int downdelay = 0;
96static int use_carrier = 1;
97static char *mode = NULL;
98static char *primary = NULL;
99static char *lacp_rate = NULL;
fd989c83 100static char *ad_select = NULL;
169a3e66 101static char *xmit_hash_policy = NULL;
1da177e4
LT		 102static int arp_interval = BOND_LINK_ARP_INTERV;
103static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
f5b2b966 104static char *arp_validate = NULL;
3915c1e8 105static char *fail_over_mac = NULL;
12479f9a 106struct bond_params bonding_defaults;
1da177e4
LT		 107
108module_param(max_bonds, int, 0);
109MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
7893b249
MS		 110module_param(num_grat_arp, int, 0644);
111MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
305d552a
BH		 112module_param(num_unsol_na, int, 0644);
113MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
1da177e4
LT		 114module_param(miimon, int, 0);
115MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
116module_param(updelay, int, 0);
117MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
118module_param(downdelay, int, 0);
2ac47660
MW		 119MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
120 "in milliseconds");
1da177e4 121module_param(use_carrier, int, 0);
2ac47660
MW		 122MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
123 "0 for off, 1 for on (default)");
1da177e4 124module_param(mode, charp, 0);
2ac47660
MW		 125MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
126 "1 for active-backup, 2 for balance-xor, "
127 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
128 "6 for balance-alb");
1da177e4
LT		 129module_param(primary, charp, 0);
130MODULE_PARM_DESC(primary, "Primary network device to use");
131module_param(lacp_rate, charp, 0);
2ac47660
MW		 132MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
133 "(slow/fast)");
fd989c83
JV		 134module_param(ad_select, charp, 0);
135MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
169a3e66 136module_param(xmit_hash_policy, charp, 0);
2ac47660
MW		 137MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
138 ", 1 for layer 3+4");
1da177e4
LT		 139module_param(arp_interval, int, 0);
140MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
141module_param_array(arp_ip_target, charp, NULL, 0);
142MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
f5b2b966
JV		 143module_param(arp_validate, charp, 0);
144MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
3915c1e8
JV		 145module_param(fail_over_mac, charp, 0);
146MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
1da177e4
LT		 147
148/*----------------------------- Global variables ----------------------------*/
149
f71e1309 150static const char * const version =
1da177e4
LT		 151 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
152
12479f9a 153LIST_HEAD(bond_dev_list);
1da177e4
LT		 154
155#ifdef CONFIG_PROC_FS
156static struct proc_dir_entry *bond_proc_dir = NULL;
157#endif
158
b76cdba9 159extern struct rw_semaphore bonding_rwsem;
d3bb52b0 160static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
1da177e4 161static int arp_ip_count = 0;
1da177e4 162static int bond_mode = BOND_MODE_ROUNDROBIN;
169a3e66 163static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
1da177e4 164static int lacp_fast = 0;
217df670 165
1da177e4 166
12479f9a 167struct bond_parm_tbl bond_lacp_tbl[] = {
1da177e4
LT		 168{ "slow", AD_LACP_SLOW},
169{ "fast", AD_LACP_FAST},
170{ NULL, -1},
171};
172
12479f9a 173struct bond_parm_tbl bond_mode_tbl[] = {
1da177e4
LT		 174{ "balance-rr", BOND_MODE_ROUNDROBIN},
175{ "active-backup", BOND_MODE_ACTIVEBACKUP},
176{ "balance-xor", BOND_MODE_XOR},
177{ "broadcast", BOND_MODE_BROADCAST},
178{ "802.3ad", BOND_MODE_8023AD},
179{ "balance-tlb", BOND_MODE_TLB},
180{ "balance-alb", BOND_MODE_ALB},
181{ NULL, -1},
182};
183
12479f9a 184struct bond_parm_tbl xmit_hashtype_tbl[] = {
169a3e66
JV		 185{ "layer2", BOND_XMIT_POLICY_LAYER2},
186{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
6f6652be 187{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
169a3e66
JV		 188{ NULL, -1},
189};
190
f5b2b966
JV		 191struct bond_parm_tbl arp_validate_tbl[] = {
192{ "none", BOND_ARP_VALIDATE_NONE},
193{ "active", BOND_ARP_VALIDATE_ACTIVE},
194{ "backup", BOND_ARP_VALIDATE_BACKUP},
195{ "all", BOND_ARP_VALIDATE_ALL},
196{ NULL, -1},
197};
198
3915c1e8
JV		 199struct bond_parm_tbl fail_over_mac_tbl[] = {
200{ "none", BOND_FOM_NONE},
201{ "active", BOND_FOM_ACTIVE},
202{ "follow", BOND_FOM_FOLLOW},
203{ NULL, -1},
204};
205
fd989c83
JV		 206struct bond_parm_tbl ad_select_tbl[] = {
207{ "stable", BOND_AD_STABLE},
208{ "bandwidth", BOND_AD_BANDWIDTH},
209{ "count", BOND_AD_COUNT},
210{ NULL, -1},
211};
212
1da177e4
LT		 213/*-------------------------- Forward declarations ---------------------------*/
214
c3ade5ca 215static void bond_send_gratuitous_arp(struct bonding *bond);
c50b85d0 216static void bond_deinit(struct net_device *bond_dev);
1da177e4
LT		 217
218/*---------------------------- General routines -----------------------------*/
219
4ad072c9 220static const char *bond_mode_name(int mode)
1da177e4
LT		 221{
222 switch (mode) {
223 case BOND_MODE_ROUNDROBIN :
224 return "load balancing (round-robin)";
225 case BOND_MODE_ACTIVEBACKUP :
226 return "fault-tolerance (active-backup)";
227 case BOND_MODE_XOR :
228 return "load balancing (xor)";
229 case BOND_MODE_BROADCAST :
230 return "fault-tolerance (broadcast)";
231 case BOND_MODE_8023AD:
232 return "IEEE 802.3ad Dynamic link aggregation";
233 case BOND_MODE_TLB:
234 return "transmit load balancing";
235 case BOND_MODE_ALB:
236 return "adaptive load balancing";
237 default:
238 return "unknown";
239 }
240}
241
242/*---------------------------------- VLAN -----------------------------------*/
243
244/**
245 * bond_add_vlan - add a new vlan id on bond
246 * @bond: bond that got the notification
247 * @vlan_id: the vlan id to add
248 *
249 * Returns -ENOMEM if allocation failed.
250 */
251static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
252{
253 struct vlan_entry *vlan;
254
255 dprintk("bond: %s, vlan id %d\n",
256 (bond ? bond->dev->name: "None"), vlan_id);
257
305d552a 258 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
1da177e4
LT		 259 if (!vlan) {
260 return -ENOMEM;
261 }
262
263 INIT_LIST_HEAD(&vlan->vlan_list);
264 vlan->vlan_id = vlan_id;
265
266 write_lock_bh(&bond->lock);
267
268 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
269
270 write_unlock_bh(&bond->lock);
271
272 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
273
274 return 0;
275}
276
277/**
278 * bond_del_vlan - delete a vlan id from bond
279 * @bond: bond that got the notification
280 * @vlan_id: the vlan id to delete
281 *
282 * returns -ENODEV if @vlan_id was not found in @bond.
283 */
284static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
285{
0883beca 286 struct vlan_entry *vlan;
1da177e4
LT		 287 int res = -ENODEV;
288
289 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
290
291 write_lock_bh(&bond->lock);
292
0883beca 293 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
1da177e4
LT		 294 if (vlan->vlan_id == vlan_id) {
295 list_del(&vlan->vlan_list);
296
297 if ((bond->params.mode == BOND_MODE_TLB) ||
298 (bond->params.mode == BOND_MODE_ALB)) {
299 bond_alb_clear_vlan(bond, vlan_id);
300 }
301
302 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
303 bond->dev->name);
304
305 kfree(vlan);
306
307 if (list_empty(&bond->vlan_list) &&
308 (bond->slave_cnt == 0)) {
309 /* Last VLAN removed and no slaves, so
310 * restore block on adding VLANs. This will
311 * be removed once new slaves that are not
312 * VLAN challenged will be added.
313 */
314 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
315 }
316
317 res = 0;
318 goto out;
319 }
320 }
321
322 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
323 bond->dev->name);
324
325out:
326 write_unlock_bh(&bond->lock);
327 return res;
328}
329
330/**
331 * bond_has_challenged_slaves
332 * @bond: the bond we're working on
333 *
334 * Searches the slave list. Returns 1 if a vlan challenged slave
335 * was found, 0 otherwise.
336 *
337 * Assumes bond->lock is held.
338 */
339static int bond_has_challenged_slaves(struct bonding *bond)
340{
341 struct slave *slave;
342 int i;
343
344 bond_for_each_slave(bond, slave, i) {
345 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
346 dprintk("found VLAN challenged slave - %s\n",
347 slave->dev->name);
348 return 1;
349 }
350 }
351
352 dprintk("no VLAN challenged slaves found\n");
353 return 0;
354}
355
356/**
357 * bond_next_vlan - safely skip to the next item in the vlans list.
358 * @bond: the bond we're working on
359 * @curr: item we're advancing from
360 *
361 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
362 * or @curr->next otherwise (even if it is @curr itself again).
363 *
364 * Caller must hold bond->lock
365 */
366struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
367{
368 struct vlan_entry *next, *last;
369
370 if (list_empty(&bond->vlan_list)) {
371 return NULL;
372 }
373
374 if (!curr) {
375 next = list_entry(bond->vlan_list.next,
376 struct vlan_entry, vlan_list);
377 } else {
378 last = list_entry(bond->vlan_list.prev,
379 struct vlan_entry, vlan_list);
380 if (last == curr) {
381 next = list_entry(bond->vlan_list.next,
382 struct vlan_entry, vlan_list);
383 } else {
384 next = list_entry(curr->vlan_list.next,
385 struct vlan_entry, vlan_list);
386 }
387 }
388
389 return next;
390}
391
392/**
393 * bond_dev_queue_xmit - Prepare skb for xmit.
394 *
395 * @bond: bond device that got this skb for tx.
396 * @skb: hw accel VLAN tagged skb to transmit
397 * @slave_dev: slave that is supposed to xmit this skbuff
398 *
399 * When the bond gets an skb to transmit that is
400 * already hardware accelerated VLAN tagged, and it
401 * needs to relay this skb to a slave that is not
402 * hw accel capable, the skb needs to be "unaccelerated",
403 * i.e. strip the hwaccel tag and re-insert it as part
404 * of the payload.
405 */
406int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
407{
966bc6f4 408 unsigned short uninitialized_var(vlan_id);
1da177e4
LT		 409
410 if (!list_empty(&bond->vlan_list) &&
411 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
412 vlan_get_tag(skb, &vlan_id) == 0) {
413 skb->dev = slave_dev;
414 skb = vlan_put_tag(skb, vlan_id);
415 if (!skb) {
416 /* vlan_put_tag() frees the skb in case of error,
417 * so return success here so the calling functions
418 * won't attempt to free is again.
419 */
420 return 0;
421 }
422 } else {
423 skb->dev = slave_dev;
424 }
425
426 skb->priority = 1;
427 dev_queue_xmit(skb);
428
429 return 0;
430}
431
432/*
433 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
434 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
435 * lock because:
436 * a. This operation is performed in IOCTL context,
437 * b. The operation is protected by the RTNL semaphore in the 8021q code,
438 * c. Holding a lock with BH disabled while directly calling a base driver
439 * entry point is generally a BAD idea.
440 *
441 * The design of synchronization/protection for this operation in the 8021q
442 * module is good for one or more VLAN devices over a single physical device
443 * and cannot be extended for a teaming solution like bonding, so there is a
444 * potential race condition here where a net device from the vlan group might
445 * be referenced (either by a base driver or the 8021q code) while it is being
446 * removed from the system. However, it turns out we're not making matters
447 * worse, and if it works for regular VLAN usage it will work here too.
448*/
449
450/**
451 * bond_vlan_rx_register - Propagates registration to slaves
452 * @bond_dev: bonding net device that got called
453 * @grp: vlan group being registered
454 */
455static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
456{
454d7c9b 457 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 458 struct slave *slave;
459 int i;
460
461 bond->vlgrp = grp;
462
463 bond_for_each_slave(bond, slave, i) {
464 struct net_device *slave_dev = slave->dev;
eb7cc59a 465 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 466
467 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 468 slave_ops->ndo_vlan_rx_register) {
469 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
1da177e4
LT		 470 }
471 }
472}
473
474/**
475 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
476 * @bond_dev: bonding net device that got called
477 * @vid: vlan id being added
478 */
479static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
480{
454d7c9b 481 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 482 struct slave *slave;
483 int i, res;
484
485 bond_for_each_slave(bond, slave, i) {
486 struct net_device *slave_dev = slave->dev;
eb7cc59a 487 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 488
489 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a
SH		 490 slave_ops->ndo_vlan_rx_add_vid) {
491 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
1da177e4
LT		 492 }
493 }
494
495 res = bond_add_vlan(bond, vid);
496 if (res) {
497 printk(KERN_ERR DRV_NAME
4e0952c7 498 ": %s: Error: Failed to add vlan id %d\n",
1da177e4
LT		 499 bond_dev->name, vid);
500 }
501}
502
503/**
504 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
505 * @bond_dev: bonding net device that got called
506 * @vid: vlan id being removed
507 */
508static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
509{
454d7c9b 510 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 511 struct slave *slave;
512 struct net_device *vlan_dev;
513 int i, res;
514
515 bond_for_each_slave(bond, slave, i) {
516 struct net_device *slave_dev = slave->dev;
eb7cc59a 517 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 518
519 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a 520 slave_ops->ndo_vlan_rx_kill_vid) {
1da177e4
LT		 521 /* Save and then restore vlan_dev in the grp array,
522 * since the slave's driver might clear it.
523 */
5c15bdec 524 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
eb7cc59a 525 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
5c15bdec 526 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
1da177e4
LT		 527 }
528 }
529
530 res = bond_del_vlan(bond, vid);
531 if (res) {
532 printk(KERN_ERR DRV_NAME
4e0952c7 533 ": %s: Error: Failed to remove vlan id %d\n",
1da177e4
LT		 534 bond_dev->name, vid);
535 }
536}
537
538static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
539{
540 struct vlan_entry *vlan;
eb7cc59a 541 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 542
543 write_lock_bh(&bond->lock);
544
eb7cc59a 545 if (list_empty(&bond->vlan_list))
1da177e4 546 goto out;
1da177e4
LT		 547
548 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 549 slave_ops->ndo_vlan_rx_register)
550 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
1da177e4
LT		 551
552 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 553 !(slave_ops->ndo_vlan_rx_add_vid))
1da177e4 554 goto out;
1da177e4 555
eb7cc59a
SH		 556 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
557 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
1da177e4
LT		 558
559out:
560 write_unlock_bh(&bond->lock);
561}
562
563static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
564{
eb7cc59a 565 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 566 struct vlan_entry *vlan;
567 struct net_device *vlan_dev;
568
569 write_lock_bh(&bond->lock);
570
eb7cc59a 571 if (list_empty(&bond->vlan_list))
1da177e4 572 goto out;
1da177e4
LT		 573
574 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 575 !(slave_ops->ndo_vlan_rx_kill_vid))
1da177e4 576 goto unreg;
1da177e4
LT		 577
578 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
579 /* Save and then restore vlan_dev in the grp array,
580 * since the slave's driver might clear it.
581 */
5c15bdec 582 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
eb7cc59a 583 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
5c15bdec 584 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
1da177e4
LT		 585 }
586
587unreg:
588 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 589 slave_ops->ndo_vlan_rx_register)
590 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
1da177e4
LT		 591
592out:
593 write_unlock_bh(&bond->lock);
594}
595
596/*------------------------------- Link status -------------------------------*/
597
ff59c456
JV		 598/*
599 * Set the carrier state for the master according to the state of its
600 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
601 * do special 802.3ad magic.
602 *
603 * Returns zero if carrier state does not change, nonzero if it does.
604 */
605static int bond_set_carrier(struct bonding *bond)
606{
607 struct slave *slave;
608 int i;
609
610 if (bond->slave_cnt == 0)
611 goto down;
612
613 if (bond->params.mode == BOND_MODE_8023AD)
614 return bond_3ad_set_carrier(bond);
615
616 bond_for_each_slave(bond, slave, i) {
617 if (slave->link == BOND_LINK_UP) {
618 if (!netif_carrier_ok(bond->dev)) {
619 netif_carrier_on(bond->dev);
620 return 1;
621 }
622 return 0;
623 }
624 }
625
626down:
627 if (netif_carrier_ok(bond->dev)) {
628 netif_carrier_off(bond->dev);
629 return 1;
630 }
631 return 0;
632}
633
1da177e4
LT		 634/*
635 * Get link speed and duplex from the slave's base driver
636 * using ethtool. If for some reason the call fails or the
637 * values are invalid, fake speed and duplex to 100/Full
638 * and return error.
639 */
640static int bond_update_speed_duplex(struct slave *slave)
641{
642 struct net_device *slave_dev = slave->dev;
1da177e4 643 struct ethtool_cmd etool;
61a44b9c 644 int res;
1da177e4
LT		 645
646 /* Fake speed and duplex */
647 slave->speed = SPEED_100;
648 slave->duplex = DUPLEX_FULL;
649
61a44b9c
MW		 650 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
651 return -1;
1da177e4 652
61a44b9c
MW		 653 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
654 if (res < 0)
1da177e4 655 return -1;
1da177e4 656
1da177e4
LT		 657 switch (etool.speed) {
658 case SPEED_10:
659 case SPEED_100:
660 case SPEED_1000:
94dbffd5 661 case SPEED_10000:
1da177e4
LT		 662 break;
663 default:
664 return -1;
665 }
666
667 switch (etool.duplex) {
668 case DUPLEX_FULL:
669 case DUPLEX_HALF:
670 break;
671 default:
672 return -1;
673 }
674
675 slave->speed = etool.speed;
676 slave->duplex = etool.duplex;
677
678 return 0;
679}
680
681/*
682 * if <dev> supports MII link status reporting, check its link status.
683 *
684 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
685 * depening upon the setting of the use_carrier parameter.
686 *
687 * Return either BMSR_LSTATUS, meaning that the link is up (or we
688 * can't tell and just pretend it is), or 0, meaning that the link is
689 * down.
690 *
691 * If reporting is non-zero, instead of faking link up, return -1 if
692 * both ETHTOOL and MII ioctls fail (meaning the device does not
693 * support them). If use_carrier is set, return whatever it says.
694 * It'd be nice if there was a good way to tell if a driver supports
695 * netif_carrier, but there really isn't.
696 */
697static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
698{
eb7cc59a 699 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 700 static int (* ioctl)(struct net_device *, struct ifreq *, int);
701 struct ifreq ifr;
702 struct mii_ioctl_data *mii;
1da177e4 703
eb7cc59a 704 if (bond->params.use_carrier)
1da177e4 705 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1da177e4 706
eb7cc59a 707 ioctl = slave_ops->ndo_do_ioctl;
1da177e4
LT		 708 if (ioctl) {
709 /* TODO: set pointer to correct ioctl on a per team member */
710 /* bases to make this more efficient. that is, once */
711 /* we determine the correct ioctl, we will always */
712 /* call it and not the others for that team */
713 /* member. */
714
715 /*
716 * We cannot assume that SIOCGMIIPHY will also read a
717 * register; not all network drivers (e.g., e100)
718 * support that.
719 */
720
721 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
722 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
723 mii = if_mii(&ifr);
724 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
725 mii->reg_num = MII_BMSR;
726 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
727 return (mii->val_out & BMSR_LSTATUS);
728 }
729 }
730 }
731
61a44b9c
MW		 732 /*
733 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
734 * attempt to get link status from it if the above MII ioctls fail.
735 */
1da177e4
LT		 736 if (slave_dev->ethtool_ops) {
737 if (slave_dev->ethtool_ops->get_link) {
738 u32 link;
739
740 link = slave_dev->ethtool_ops->get_link(slave_dev);
741
742 return link ? BMSR_LSTATUS : 0;
743 }
744 }
745
1da177e4
LT		 746 /*
747 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 748 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 749 * cannot report link status). If not reporting, pretend
750 * we're ok.
751 */
752 return (reporting ? -1 : BMSR_LSTATUS);
753}
754
755/*----------------------------- Multicast list ------------------------------*/
756
757/*
758 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
759 */
760static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
761{
762 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
763 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
764}
765
766/*
767 * returns dmi entry if found, NULL otherwise
768 */
769static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
770{
771 struct dev_mc_list *idmi;
772
773 for (idmi = mc_list; idmi; idmi = idmi->next) {
774 if (bond_is_dmi_same(dmi, idmi)) {
775 return idmi;
776 }
777 }
778
779 return NULL;
780}
781
782/*
783 * Push the promiscuity flag down to appropriate slaves
784 */
7e1a1ac1 785static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 786{
7e1a1ac1 787 int err = 0;
1da177e4
LT		 788 if (USES_PRIMARY(bond->params.mode)) {
789 /* write lock already acquired */
790 if (bond->curr_active_slave) {
7e1a1ac1
WC		 791 err = dev_set_promiscuity(bond->curr_active_slave->dev,
792 inc);
1da177e4
LT		 793 }
794 } else {
795 struct slave *slave;
796 int i;
797 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 798 err = dev_set_promiscuity(slave->dev, inc);
799 if (err)
800 return err;
1da177e4
LT		 801 }
802 }
7e1a1ac1 803 return err;
1da177e4
LT		 804}
805
806/*
807 * Push the allmulti flag down to all slaves
808 */
7e1a1ac1 809static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 810{
7e1a1ac1 811 int err = 0;
1da177e4
LT		 812 if (USES_PRIMARY(bond->params.mode)) {
813 /* write lock already acquired */
814 if (bond->curr_active_slave) {
7e1a1ac1
WC		 815 err = dev_set_allmulti(bond->curr_active_slave->dev,
816 inc);
1da177e4
LT		 817 }
818 } else {
819 struct slave *slave;
820 int i;
821 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 822 err = dev_set_allmulti(slave->dev, inc);
823 if (err)
824 return err;
1da177e4
LT		 825 }
826 }
7e1a1ac1 827 return err;
1da177e4
LT		 828}
829
830/*
831 * Add a Multicast address to slaves
832 * according to mode
833 */
834static void bond_mc_add(struct bonding *bond, void *addr, int alen)
835{
836 if (USES_PRIMARY(bond->params.mode)) {
837 /* write lock already acquired */
838 if (bond->curr_active_slave) {
839 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
840 }
841 } else {
842 struct slave *slave;
843 int i;
844 bond_for_each_slave(bond, slave, i) {
845 dev_mc_add(slave->dev, addr, alen, 0);
846 }
847 }
848}
849
850/*
851 * Remove a multicast address from slave
852 * according to mode
853 */
854static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
855{
856 if (USES_PRIMARY(bond->params.mode)) {
857 /* write lock already acquired */
858 if (bond->curr_active_slave) {
859 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
860 }
861 } else {
862 struct slave *slave;
863 int i;
864 bond_for_each_slave(bond, slave, i) {
865 dev_mc_delete(slave->dev, addr, alen, 0);
866 }
867 }
868}
869
a816c7c7
JV		 870
871/*
872 * Retrieve the list of registered multicast addresses for the bonding
873 * device and retransmit an IGMP JOIN request to the current active
874 * slave.
875 */
876static void bond_resend_igmp_join_requests(struct bonding *bond)
877{
878 struct in_device *in_dev;
879 struct ip_mc_list *im;
880
881 rcu_read_lock();
882 in_dev = __in_dev_get_rcu(bond->dev);
883 if (in_dev) {
884 for (im = in_dev->mc_list; im; im = im->next) {
885 ip_mc_rejoin_group(im);
886 }
887 }
888
889 rcu_read_unlock();
890}
891
1da177e4
LT		 892/*
893 * Totally destroys the mc_list in bond
894 */
895static void bond_mc_list_destroy(struct bonding *bond)
896{
897 struct dev_mc_list *dmi;
898
899 dmi = bond->mc_list;
900 while (dmi) {
901 bond->mc_list = dmi->next;
902 kfree(dmi);
903 dmi = bond->mc_list;
904 }
a816c7c7 905 bond->mc_list = NULL;
1da177e4
LT		 906}
907
908/*
909 * Copy all the Multicast addresses from src to the bonding device dst
910 */
de54f390 911static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 912 gfp_t gfp_flag)
1da177e4
LT		 913{
914 struct dev_mc_list *dmi, *new_dmi;
915
916 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 917 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT		 918
919 if (!new_dmi) {
920 /* FIXME: Potential memory leak !!! */
921 return -ENOMEM;
922 }
923
924 new_dmi->next = bond->mc_list;
925 bond->mc_list = new_dmi;
926 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
927 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
928 new_dmi->dmi_users = dmi->dmi_users;
929 new_dmi->dmi_gusers = dmi->dmi_gusers;
930 }
931
932 return 0;
933}
934
935/*
936 * flush all members of flush->mc_list from device dev->mc_list
937 */
938static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
939{
454d7c9b 940 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 941 struct dev_mc_list *dmi;
942
943 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
944 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
945 }
946
947 if (bond->params.mode == BOND_MODE_8023AD) {
948 /* del lacpdu mc addr from mc list */
949 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
950
951 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
952 }
953}
954
955/*--------------------------- Active slave change ---------------------------*/
956
957/*
958 * Update the mc list and multicast-related flags for the new and
959 * old active slaves (if any) according to the multicast mode, and
960 * promiscuous flags unconditionally.
961 */
962static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
963{
964 struct dev_mc_list *dmi;
965
966 if (!USES_PRIMARY(bond->params.mode)) {
967 /* nothing to do - mc list is already up-to-date on
968 * all slaves
969 */
970 return;
971 }
972
973 if (old_active) {
974 if (bond->dev->flags & IFF_PROMISC) {
975 dev_set_promiscuity(old_active->dev, -1);
976 }
977
978 if (bond->dev->flags & IFF_ALLMULTI) {
979 dev_set_allmulti(old_active->dev, -1);
980 }
981
982 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
983 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
984 }
985 }
986
987 if (new_active) {
7e1a1ac1 988 /* FIXME: Signal errors upstream. */
1da177e4
LT		 989 if (bond->dev->flags & IFF_PROMISC) {
990 dev_set_promiscuity(new_active->dev, 1);
991 }
992
993 if (bond->dev->flags & IFF_ALLMULTI) {
994 dev_set_allmulti(new_active->dev, 1);
995 }
996
997 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
998 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
999 }
a816c7c7 1000 bond_resend_igmp_join_requests(bond);
1da177e4
LT		 1001 }
1002}
1003
3915c1e8
JV		 1004/*
1005 * bond_do_fail_over_mac
1006 *
1007 * Perform special MAC address swapping for fail_over_mac settings
1008 *
1009 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1010 */
1011static void bond_do_fail_over_mac(struct bonding *bond,
1012 struct slave *new_active,
1013 struct slave *old_active)
1014{
1015 u8 tmp_mac[ETH_ALEN];
1016 struct sockaddr saddr;
1017 int rv;
1018
1019 switch (bond->params.fail_over_mac) {
1020 case BOND_FOM_ACTIVE:
1021 if (new_active)
1022 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1023 new_active->dev->addr_len);
1024 break;
1025 case BOND_FOM_FOLLOW:
1026 /*
1027 * if new_active && old_active, swap them
1028 * if just old_active, do nothing (going to no active slave)
1029 * if just new_active, set new_active to bond's MAC
1030 */
1031 if (!new_active)
1032 return;
1033
1034 write_unlock_bh(&bond->curr_slave_lock);
1035 read_unlock(&bond->lock);
1036
1037 if (old_active) {
1038 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1039 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1040 ETH_ALEN);
1041 saddr.sa_family = new_active->dev->type;
1042 } else {
1043 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1044 saddr.sa_family = bond->dev->type;
1045 }
1046
1047 rv = dev_set_mac_address(new_active->dev, &saddr);
1048 if (rv) {
1049 printk(KERN_ERR DRV_NAME
1050 ": %s: Error %d setting MAC of slave %s\n",
1051 bond->dev->name, -rv, new_active->dev->name);
1052 goto out;
1053 }
1054
1055 if (!old_active)
1056 goto out;
1057
1058 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1059 saddr.sa_family = old_active->dev->type;
1060
1061 rv = dev_set_mac_address(old_active->dev, &saddr);
1062 if (rv)
1063 printk(KERN_ERR DRV_NAME
1064 ": %s: Error %d setting MAC of slave %s\n",
1065 bond->dev->name, -rv, new_active->dev->name);
1066out:
1067 read_lock(&bond->lock);
1068 write_lock_bh(&bond->curr_slave_lock);
1069 break;
1070 default:
1071 printk(KERN_ERR DRV_NAME
1072 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1073 bond->dev->name, bond->params.fail_over_mac);
1074 break;
1075 }
1076
1077}
1078
1079
1da177e4
LT		 1080/**
1081 * find_best_interface - select the best available slave to be the active one
1082 * @bond: our bonding struct
1083 *
1084 * Warning: Caller must hold curr_slave_lock for writing.
1085 */
1086static struct slave *bond_find_best_slave(struct bonding *bond)
1087{
1088 struct slave *new_active, *old_active;
1089 struct slave *bestslave = NULL;
1090 int mintime = bond->params.updelay;
1091 int i;
1092
1093 new_active = old_active = bond->curr_active_slave;
1094
1095 if (!new_active) { /* there were no active slaves left */
1096 if (bond->slave_cnt > 0) { /* found one slave */
1097 new_active = bond->first_slave;
1098 } else {
1099 return NULL; /* still no slave, return NULL */
1100 }
1101 }
1102
1103 /* first try the primary link; if arping, a link must tx/rx traffic
1104 * before it can be considered the curr_active_slave - also, we would skip
1105 * slaves between the curr_active_slave and primary_slave that may be up
1106 * and able to arp
1107 */
1108 if ((bond->primary_slave) &&
1109 (!bond->params.arp_interval) &&
1110 (IS_UP(bond->primary_slave->dev))) {
1111 new_active = bond->primary_slave;
1112 }
1113
1114 /* remember where to stop iterating over the slaves */
1115 old_active = new_active;
1116
1117 bond_for_each_slave_from(bond, new_active, i, old_active) {
1118 if (IS_UP(new_active->dev)) {
1119 if (new_active->link == BOND_LINK_UP) {
1120 return new_active;
1121 } else if (new_active->link == BOND_LINK_BACK) {
1122 /* link up, but waiting for stabilization */
1123 if (new_active->delay < mintime) {
1124 mintime = new_active->delay;
1125 bestslave = new_active;
1126 }
1127 }
1128 }
1129 }
1130
1131 return bestslave;
1132}
1133
1134/**
1135 * change_active_interface - change the active slave into the specified one
1136 * @bond: our bonding struct
1137 * @new: the new slave to make the active one
1138 *
1139 * Set the new slave to the bond's settings and unset them on the old
1140 * curr_active_slave.
1141 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1142 *
1143 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1144 * because it is apparently the best available slave we have, even though its
1145 * updelay hasn't timed out yet.
1146 *
3915c1e8
JV		 1147 * If new_active is not NULL, caller must hold bond->lock for read and
1148 * curr_slave_lock for write_bh.
1da177e4 1149 */
a77b5325 1150void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1151{
1152 struct slave *old_active = bond->curr_active_slave;
1153
1154 if (old_active == new_active) {
1155 return;
1156 }
1157
1158 if (new_active) {
b2220cad
JV		 1159 new_active->jiffies = jiffies;
1160
1da177e4
LT		 1161 if (new_active->link == BOND_LINK_BACK) {
1162 if (USES_PRIMARY(bond->params.mode)) {
1163 printk(KERN_INFO DRV_NAME
1164 ": %s: making interface %s the new "
1165 "active one %d ms earlier.\n",
1166 bond->dev->name, new_active->dev->name,
1167 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1168 }
1169
1170 new_active->delay = 0;
1171 new_active->link = BOND_LINK_UP;
1da177e4
LT		 1172
1173 if (bond->params.mode == BOND_MODE_8023AD) {
1174 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1175 }
1176
1177 if ((bond->params.mode == BOND_MODE_TLB) ||
1178 (bond->params.mode == BOND_MODE_ALB)) {
1179 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1180 }
1181 } else {
1182 if (USES_PRIMARY(bond->params.mode)) {
1183 printk(KERN_INFO DRV_NAME
1184 ": %s: making interface %s the new "
1185 "active one.\n",
1186 bond->dev->name, new_active->dev->name);
1187 }
1188 }
1189 }
1190
1da177e4
LT		 1191 if (USES_PRIMARY(bond->params.mode)) {
1192 bond_mc_swap(bond, new_active, old_active);
1193 }
1194
1195 if ((bond->params.mode == BOND_MODE_TLB) ||
1196 (bond->params.mode == BOND_MODE_ALB)) {
1197 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1198 if (old_active)
1199 bond_set_slave_inactive_flags(old_active);
1200 if (new_active)
1201 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1202 } else {
1203 bond->curr_active_slave = new_active;
1204 }
c3ade5ca
JV		 1205
1206 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1207 if (old_active) {
1208 bond_set_slave_inactive_flags(old_active);
1209 }
1210
1211 if (new_active) {
1212 bond_set_slave_active_flags(new_active);
2ab82852 1213
709f8a45
OG		 1214 if (bond->params.fail_over_mac)
1215 bond_do_fail_over_mac(bond, new_active,
1216 old_active);
3915c1e8 1217
709f8a45 1218 bond->send_grat_arp = bond->params.num_grat_arp;
b59f9f74 1219 bond_send_gratuitous_arp(bond);
01f3109d 1220
305d552a
BH		 1221 bond->send_unsol_na = bond->params.num_unsol_na;
1222 bond_send_unsolicited_na(bond);
1223
01f3109d
OG		 1224 write_unlock_bh(&bond->curr_slave_lock);
1225 read_unlock(&bond->lock);
1226
1227 netdev_bonding_change(bond->dev);
1228
1229 read_lock(&bond->lock);
1230 write_lock_bh(&bond->curr_slave_lock);
7893b249 1231 }
c3ade5ca 1232 }
1da177e4
LT		 1233}
1234
1235/**
1236 * bond_select_active_slave - select a new active slave, if needed
1237 * @bond: our bonding struct
1238 *
1239 * This functions shoud be called when one of the following occurs:
1240 * - The old curr_active_slave has been released or lost its link.
1241 * - The primary_slave has got its link back.
1242 * - A slave has got its link back and there's no old curr_active_slave.
1243 *
3915c1e8 1244 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1245 */
a77b5325 1246void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1247{
1248 struct slave *best_slave;
ff59c456 1249 int rv;
1da177e4
LT		 1250
1251 best_slave = bond_find_best_slave(bond);
1252 if (best_slave != bond->curr_active_slave) {
1253 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1254 rv = bond_set_carrier(bond);
1255 if (!rv)
1256 return;
1257
1258 if (netif_carrier_ok(bond->dev)) {
1259 printk(KERN_INFO DRV_NAME
1260 ": %s: first active interface up!\n",
1261 bond->dev->name);
1262 } else {
1263 printk(KERN_INFO DRV_NAME ": %s: "
1264 "now running without any active interface !\n",
1265 bond->dev->name);
1266 }
1da177e4
LT		 1267 }
1268}
1269
1270/*--------------------------- slave list handling ---------------------------*/
1271
1272/*
1273 * This function attaches the slave to the end of list.
1274 *
1275 * bond->lock held for writing by caller.
1276 */
1277static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1278{
1279 if (bond->first_slave == NULL) { /* attaching the first slave */
1280 new_slave->next = new_slave;
1281 new_slave->prev = new_slave;
1282 bond->first_slave = new_slave;
1283 } else {
1284 new_slave->next = bond->first_slave;
1285 new_slave->prev = bond->first_slave->prev;
1286 new_slave->next->prev = new_slave;
1287 new_slave->prev->next = new_slave;
1288 }
1289
1290 bond->slave_cnt++;
1291}
1292
1293/*
1294 * This function detaches the slave from the list.
1295 * WARNING: no check is made to verify if the slave effectively
1296 * belongs to <bond>.
1297 * Nothing is freed on return, structures are just unchained.
1298 * If any slave pointer in bond was pointing to <slave>,
1299 * it should be changed by the calling function.
1300 *
1301 * bond->lock held for writing by caller.
1302 */
1303static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1304{
1305 if (slave->next) {
1306 slave->next->prev = slave->prev;
1307 }
1308
1309 if (slave->prev) {
1310 slave->prev->next = slave->next;
1311 }
1312
1313 if (bond->first_slave == slave) { /* slave is the first slave */
1314 if (bond->slave_cnt > 1) { /* there are more slave */
1315 bond->first_slave = slave->next;
1316 } else {
1317 bond->first_slave = NULL; /* slave was the last one */
1318 }
1319 }
1320
1321 slave->next = NULL;
1322 slave->prev = NULL;
1323 bond->slave_cnt--;
1324}
1325
1326/*---------------------------------- IOCTL ----------------------------------*/
1327
4ad072c9
AB		 1328static int bond_sethwaddr(struct net_device *bond_dev,
1329 struct net_device *slave_dev)
1da177e4
LT		 1330{
1331 dprintk("bond_dev=%p\n", bond_dev);
1332 dprintk("slave_dev=%p\n", slave_dev);
1333 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1334 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1335 return 0;
1336}
1337
7f353bf2
HX		 1338#define BOND_VLAN_FEATURES \
1339 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1340 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK		 1341
1342/*
8e3babcd 1343 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX		 1344 * feature bits are managed elsewhere, so preserve those feature bits
1345 * on the master device.
8531c5ff
AK		 1346 */
1347static int bond_compute_features(struct bonding *bond)
1348{
8531c5ff
AK		 1349 struct slave *slave;
1350 struct net_device *bond_dev = bond->dev;
7f353bf2 1351 unsigned long features = bond_dev->features;
3158bf7d
MS		 1352 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1353 bond_dev->hard_header_len);
8e3babcd 1354 int i;
8531c5ff 1355
7f353bf2 1356 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
b63365a2
HX		 1357 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1358
1359 if (!bond->first_slave)
1360 goto done;
1361
1362 features &= ~NETIF_F_ONE_FOR_ALL;
7f353bf2 1363
54ef3137 1364 bond_for_each_slave(bond, slave, i) {
b63365a2
HX		 1365 features = netdev_increment_features(features,
1366 slave->dev->features,
1367 NETIF_F_ONE_FOR_ALL);
54ef3137
JV		 1368 if (slave->dev->hard_header_len > max_hard_header_len)
1369 max_hard_header_len = slave->dev->hard_header_len;
1370 }
8531c5ff 1371
b63365a2 1372done:
7f353bf2 1373 features |= (bond_dev->features & BOND_VLAN_FEATURES);
b63365a2 1374 bond_dev->features = netdev_fix_features(features, NULL);
54ef3137 1375 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1376
1377 return 0;
1378}
1379
872254dd
MS		 1380
1381static void bond_setup_by_slave(struct net_device *bond_dev,
1382 struct net_device *slave_dev)
1383{
454d7c9b 1384 struct bonding *bond = netdev_priv(bond_dev);
d90a162a 1385
872254dd 1386 bond_dev->neigh_setup = slave_dev->neigh_setup;
1284cd3a 1387 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1388
1389 bond_dev->type = slave_dev->type;
1390 bond_dev->hard_header_len = slave_dev->hard_header_len;
1391 bond_dev->addr_len = slave_dev->addr_len;
1392
1393 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1394 slave_dev->addr_len);
d90a162a 1395 bond->setup_by_slave = 1;
872254dd
MS		 1396}
1397
1da177e4 1398/* enslave device <slave> to bond device <master> */
a77b5325 1399int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1400{
454d7c9b 1401 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1402 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 1403 struct slave *new_slave = NULL;
1404 struct dev_mc_list *dmi;
1405 struct sockaddr addr;
1406 int link_reporting;
1407 int old_features = bond_dev->features;
1408 int res = 0;
1409
552709d5 1410 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
eb7cc59a 1411 slave_ops->ndo_do_ioctl == NULL) {
1da177e4 1412 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1413 ": %s: Warning: no link monitoring support for %s\n",
1414 bond_dev->name, slave_dev->name);
1da177e4
LT		 1415 }
1416
1417 /* bond must be initialized by bond_open() before enslaving */
1418 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1419 printk(KERN_WARNING DRV_NAME
1420 " %s: master_dev is not up in bond_enslave\n",
1421 bond_dev->name);
1da177e4
LT		 1422 }
1423
1424 /* already enslaved */
1425 if (slave_dev->flags & IFF_SLAVE) {
1426 dprintk("Error, Device was already enslaved\n");
1427 return -EBUSY;
1428 }
1429
1430 /* vlan challenged mutual exclusion */
1431 /* no need to lock since we're protected by rtnl_lock */
1432 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1433 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1434 if (!list_empty(&bond->vlan_list)) {
1435 printk(KERN_ERR DRV_NAME
4e0952c7 1436 ": %s: Error: cannot enslave VLAN "
1da177e4 1437 "challenged slave %s on VLAN enabled "
4e0952c7 1438 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1439 bond_dev->name);
1440 return -EPERM;
1441 } else {
1442 printk(KERN_WARNING DRV_NAME
4e0952c7 1443 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1444 "slave %s. Adding VLANs will be blocked as "
1445 "long as %s is part of bond %s\n",
4e0952c7 1446 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1447 bond_dev->name);
1448 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1449 }
1450 } else {
1451 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1452 if (bond->slave_cnt == 0) {
1453 /* First slave, and it is not VLAN challenged,
1454 * so remove the block of adding VLANs over the bond.
1455 */
1456 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1457 }
1458 }
1459
217df670
JV		 1460 /*
1461 * Old ifenslave binaries are no longer supported. These can
1462 * be identified with moderate accurary by the state of the slave:
1463 * the current ifenslave will set the interface down prior to
1464 * enslaving it; the old ifenslave will not.
1465 */
1466 if ((slave_dev->flags & IFF_UP)) {
1467 printk(KERN_ERR DRV_NAME ": %s is up. "
1468 "This may be due to an out of date ifenslave.\n",
1469 slave_dev->name);
1470 res = -EPERM;
1471 goto err_undo_flags;
1472 }
1da177e4 1473
872254dd
MS		 1474 /* set bonding device ether type by slave - bonding netdevices are
1475 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1476 * there is a need to override some of the type dependent attribs/funcs.
1477 *
1478 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1479 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1480 */
1481 if (bond->slave_cnt == 0) {
1482 if (slave_dev->type != ARPHRD_ETHER)
1483 bond_setup_by_slave(bond_dev, slave_dev);
1484 } else if (bond_dev->type != slave_dev->type) {
1485 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1486 "from other slaves (%d), can not enslave it.\n",
1487 slave_dev->name,
1488 slave_dev->type, bond_dev->type);
1489 res = -EINVAL;
1490 goto err_undo_flags;
1491 }
1492
eb7cc59a 1493 if (slave_ops->ndo_set_mac_address == NULL) {
2ab82852
MS		 1494 if (bond->slave_cnt == 0) {
1495 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1496 ": %s: Warning: The first slave device "
1497 "specified does not support setting the MAC "
3915c1e8 1498 "address. Setting fail_over_mac to active.",
dd957c57 1499 bond_dev->name);
3915c1e8
JV		 1500 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1501 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1502 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1503 ": %s: Error: The slave device specified "
1504 "does not support setting the MAC address, "
3915c1e8 1505 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1506 , bond_dev->name);
1507 res = -EOPNOTSUPP;
1508 goto err_undo_flags;
1509 }
1da177e4
LT		 1510 }
1511
243cb4e5 1512 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1513 if (!new_slave) {
1514 res = -ENOMEM;
1515 goto err_undo_flags;
1516 }
1517
1da177e4
LT		 1518 /* save slave's original flags before calling
1519 * netdev_set_master and dev_open
1520 */
1521 new_slave->original_flags = slave_dev->flags;
1522
217df670
JV		 1523 /*
1524 * Save slave's original ("permanent") mac address for modes
1525 * that need it, and for restoring it upon release, and then
1526 * set it to the master's address
1527 */
1528 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1529
dd957c57 1530 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1531 /*
1532 * Set slave to master's mac address. The application already
1533 * set the master's mac address to that of the first slave
1534 */
1535 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1536 addr.sa_family = slave_dev->type;
1537 res = dev_set_mac_address(slave_dev, &addr);
1538 if (res) {
1539 dprintk("Error %d calling set_mac_address\n", res);
1540 goto err_free;
1541 }
217df670 1542 }
1da177e4 1543
c2edacf8
JV		 1544 res = netdev_set_master(slave_dev, bond_dev);
1545 if (res) {
1546 dprintk("Error %d calling netdev_set_master\n", res);
569f0c4d 1547 goto err_restore_mac;
c2edacf8 1548 }
217df670
JV		 1549 /* open the slave since the application closed it */
1550 res = dev_open(slave_dev);
1551 if (res) {
1552 dprintk("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1553 goto err_unset_master;
1da177e4
LT		 1554 }
1555
1da177e4 1556 new_slave->dev = slave_dev;
0b680e75 1557 slave_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 1558
1559 if ((bond->params.mode == BOND_MODE_TLB) ||
1560 (bond->params.mode == BOND_MODE_ALB)) {
1561 /* bond_alb_init_slave() must be called before all other stages since
1562 * it might fail and we do not want to have to undo everything
1563 */
1564 res = bond_alb_init_slave(bond, new_slave);
1565 if (res) {
569f0c4d 1566 goto err_close;
1da177e4
LT		 1567 }
1568 }
1569
1570 /* If the mode USES_PRIMARY, then the new slave gets the
1571 * master's promisc (and mc) settings only if it becomes the
1572 * curr_active_slave, and that is taken care of later when calling
1573 * bond_change_active()
1574 */
1575 if (!USES_PRIMARY(bond->params.mode)) {
1576 /* set promiscuity level to new slave */
1577 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1578 res = dev_set_promiscuity(slave_dev, 1);
1579 if (res)
1580 goto err_close;
1da177e4
LT		 1581 }
1582
1583 /* set allmulti level to new slave */
1584 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1585 res = dev_set_allmulti(slave_dev, 1);
1586 if (res)
1587 goto err_close;
1da177e4
LT		 1588 }
1589
b9e40857 1590 netif_addr_lock_bh(bond_dev);
1da177e4
LT		 1591 /* upload master's mc_list to new slave */
1592 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1593 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1594 }
b9e40857 1595 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1596 }
1597
1598 if (bond->params.mode == BOND_MODE_8023AD) {
1599 /* add lacpdu mc addr to mc list */
1600 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1601
1602 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1603 }
1604
1605 bond_add_vlans_on_slave(bond, slave_dev);
1606
1607 write_lock_bh(&bond->lock);
1608
1609 bond_attach_slave(bond, new_slave);
1610
1611 new_slave->delay = 0;
1612 new_slave->link_failure_count = 0;
1613
8531c5ff
AK		 1614 bond_compute_features(bond);
1615
3915c1e8
JV		 1616 write_unlock_bh(&bond->lock);
1617
1618 read_lock(&bond->lock);
1619
f5b2b966
JV		 1620 new_slave->last_arp_rx = jiffies;
1621
1da177e4
LT		 1622 if (bond->params.miimon && !bond->params.use_carrier) {
1623 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1624
1625 if ((link_reporting == -1) && !bond->params.arp_interval) {
1626 /*
1627 * miimon is set but a bonded network driver
1628 * does not support ETHTOOL/MII and
1629 * arp_interval is not set. Note: if
1630 * use_carrier is enabled, we will never go
1631 * here (because netif_carrier is always
1632 * supported); thus, we don't need to change
1633 * the messages for netif_carrier.
1634 */
1635 printk(KERN_WARNING DRV_NAME
4e0952c7 1636 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1637 "available for interface %s, and "
1638 "arp_interval/arp_ip_target module parameters "
1639 "not specified, thus bonding will not detect "
1640 "link failures! see bonding.txt for details.\n",
4e0952c7 1641 bond_dev->name, slave_dev->name);
1da177e4
LT		 1642 } else if (link_reporting == -1) {
1643 /* unable get link status using mii/ethtool */
1644 printk(KERN_WARNING DRV_NAME
4e0952c7 1645 ": %s: Warning: can't get link status from "
1da177e4
LT		 1646 "interface %s; the network driver associated "
1647 "with this interface does not support MII or "
1648 "ETHTOOL link status reporting, thus miimon "
1649 "has no effect on this interface.\n",
4e0952c7 1650 bond_dev->name, slave_dev->name);
1da177e4
LT		 1651 }
1652 }
1653
1654 /* check for initial state */
1655 if (!bond->params.miimon ||
1656 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1657 if (bond->params.updelay) {
1658 dprintk("Initial state of slave_dev is "
1659 "BOND_LINK_BACK\n");
1660 new_slave->link = BOND_LINK_BACK;
1661 new_slave->delay = bond->params.updelay;
1662 } else {
1663 dprintk("Initial state of slave_dev is "
1664 "BOND_LINK_UP\n");
1665 new_slave->link = BOND_LINK_UP;
1666 }
1667 new_slave->jiffies = jiffies;
1668 } else {
1669 dprintk("Initial state of slave_dev is "
1670 "BOND_LINK_DOWN\n");
1671 new_slave->link = BOND_LINK_DOWN;
1672 }
1673
1674 if (bond_update_speed_duplex(new_slave) &&
1675 (new_slave->link != BOND_LINK_DOWN)) {
1676 printk(KERN_WARNING DRV_NAME
4e0952c7 1677 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1678 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1679 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1680
1681 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1682 printk(KERN_WARNING DRV_NAME
1683 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1684 "support in base driver for proper aggregator "
4e0952c7 1685 "selection.\n", bond_dev->name);
1da177e4
LT		 1686 }
1687 }
1688
1689 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1690 /* if there is a primary slave, remember it */
1691 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1692 bond->primary_slave = new_slave;
1693 }
1694 }
1695
3915c1e8
JV		 1696 write_lock_bh(&bond->curr_slave_lock);
1697
1da177e4
LT		 1698 switch (bond->params.mode) {
1699 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1700 bond_set_slave_inactive_flags(new_slave);
1701 bond_select_active_slave(bond);
1da177e4
LT		 1702 break;
1703 case BOND_MODE_8023AD:
1704 /* in 802.3ad mode, the internal mechanism
1705 * will activate the slaves in the selected
1706 * aggregator
1707 */
1708 bond_set_slave_inactive_flags(new_slave);
1709 /* if this is the first slave */
1710 if (bond->slave_cnt == 1) {
1711 SLAVE_AD_INFO(new_slave).id = 1;
1712 /* Initialize AD with the number of times that the AD timer is called in 1 second
1713 * can be called only after the mac address of the bond is set
1714 */
1715 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1716 bond->params.lacp_fast);
1717 } else {
1718 SLAVE_AD_INFO(new_slave).id =
1719 SLAVE_AD_INFO(new_slave->prev).id + 1;
1720 }
1721
1722 bond_3ad_bind_slave(new_slave);
1723 break;
1724 case BOND_MODE_TLB:
1725 case BOND_MODE_ALB:
1726 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1727 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1728 break;
1729 default:
1730 dprintk("This slave is always active in trunk mode\n");
1731
1732 /* always active in trunk mode */
1733 new_slave->state = BOND_STATE_ACTIVE;
1734
1735 /* In trunking mode there is little meaning to curr_active_slave
1736 * anyway (it holds no special properties of the bond device),
1737 * so we can change it without calling change_active_interface()
1738 */
1739 if (!bond->curr_active_slave) {
1740 bond->curr_active_slave = new_slave;
1741 }
1742 break;
1743 } /* switch(bond_mode) */
1744
3915c1e8
JV		 1745 write_unlock_bh(&bond->curr_slave_lock);
1746
ff59c456
JV		 1747 bond_set_carrier(bond);
1748
3915c1e8 1749 read_unlock(&bond->lock);
1da177e4 1750
b76cdba9
MW		 1751 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1752 if (res)
569f0c4d 1753 goto err_close;
b76cdba9 1754
1da177e4
LT		 1755 printk(KERN_INFO DRV_NAME
1756 ": %s: enslaving %s as a%s interface with a%s link.\n",
1757 bond_dev->name, slave_dev->name,
1758 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1759 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1760
1761 /* enslave is successful */
1762 return 0;
1763
1764/* Undo stages on error */
1da177e4
LT		 1765err_close:
1766 dev_close(slave_dev);
1767
569f0c4d
JV		 1768err_unset_master:
1769 netdev_set_master(slave_dev, NULL);
1770
1da177e4 1771err_restore_mac:
dd957c57 1772 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1773 /* XXX TODO - fom follow mode needs to change master's
1774 * MAC if this slave's MAC is in use by the bond, or at
1775 * least print a warning.
1776 */
2ab82852
MS		 1777 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1778 addr.sa_family = slave_dev->type;
1779 dev_set_mac_address(slave_dev, &addr);
1780 }
1da177e4
LT		 1781
1782err_free:
1783 kfree(new_slave);
1784
1785err_undo_flags:
1786 bond_dev->features = old_features;
8531c5ff 1787
1da177e4
LT		 1788 return res;
1789}
1790
1791/*
1792 * Try to release the slave device <slave> from the bond device <master>
1793 * It is legal to access curr_active_slave without a lock because all the function
1794 * is write-locked.
1795 *
1796 * The rules for slave state should be:
1797 * for Active/Backup:
1798 * Active stays on all backups go down
1799 * for Bonded connections:
1800 * The first up interface should be left on and all others downed.
1801 */
a77b5325 1802int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1803{
454d7c9b 1804 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1805 struct slave *slave, *oldcurrent;
1806 struct sockaddr addr;
1807 int mac_addr_differ;
1808
1809 /* slave is not a slave or master is not master of this slave */
1810 if (!(slave_dev->flags & IFF_SLAVE) ||
1811 (slave_dev->master != bond_dev)) {
1812 printk(KERN_ERR DRV_NAME
4e0952c7 1813 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1814 bond_dev->name, slave_dev->name);
1815 return -EINVAL;
1816 }
1817
1818 write_lock_bh(&bond->lock);
1819
1820 slave = bond_get_slave_by_dev(bond, slave_dev);
1821 if (!slave) {
1822 /* not a slave of this bond */
1823 printk(KERN_INFO DRV_NAME
1824 ": %s: %s not enslaved\n",
1825 bond_dev->name, slave_dev->name);
f5e2a7b2 1826 write_unlock_bh(&bond->lock);
1da177e4
LT		 1827 return -EINVAL;
1828 }
1829
3915c1e8
JV		 1830 if (!bond->params.fail_over_mac) {
1831 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1832 ETH_ALEN);
1833 if (!mac_addr_differ && (bond->slave_cnt > 1))
1834 printk(KERN_WARNING DRV_NAME
1835 ": %s: Warning: the permanent HWaddr of %s - "
e174961c 1836 "%pM - is still in use by %s. "
3915c1e8
JV		 1837 "Set the HWaddr of %s to a different address "
1838 "to avoid conflicts.\n",
1839 bond_dev->name, slave_dev->name,
e174961c 1840 slave->perm_hwaddr,
3915c1e8 1841 bond_dev->name, slave_dev->name);
1da177e4
LT		 1842 }
1843
1844 /* Inform AD package of unbinding of slave. */
1845 if (bond->params.mode == BOND_MODE_8023AD) {
1846 /* must be called before the slave is
1847 * detached from the list
1848 */
1849 bond_3ad_unbind_slave(slave);
1850 }
1851
1852 printk(KERN_INFO DRV_NAME
1853 ": %s: releasing %s interface %s\n",
1854 bond_dev->name,
1855 (slave->state == BOND_STATE_ACTIVE)
1856 ? "active" : "backup",
1857 slave_dev->name);
1858
1859 oldcurrent = bond->curr_active_slave;
1860
1861 bond->current_arp_slave = NULL;
1862
1863 /* release the slave from its bond */
1864 bond_detach_slave(bond, slave);
1865
8531c5ff
AK		 1866 bond_compute_features(bond);
1867
1da177e4
LT		 1868 if (bond->primary_slave == slave) {
1869 bond->primary_slave = NULL;
1870 }
1871
1872 if (oldcurrent == slave) {
1873 bond_change_active_slave(bond, NULL);
1874 }
1875
1876 if ((bond->params.mode == BOND_MODE_TLB) ||
1877 (bond->params.mode == BOND_MODE_ALB)) {
1878 /* Must be called only after the slave has been
1879 * detached from the list and the curr_active_slave
1880 * has been cleared (if our_slave == old_current),
1881 * but before a new active slave is selected.
1882 */
2543331d 1883 write_unlock_bh(&bond->lock);
1da177e4 1884 bond_alb_deinit_slave(bond, slave);
2543331d 1885 write_lock_bh(&bond->lock);
1da177e4
LT		 1886 }
1887
059fe7a5
JV		 1888 if (oldcurrent == slave) {
1889 /*
1890 * Note that we hold RTNL over this sequence, so there
1891 * is no concern that another slave add/remove event
1892 * will interfere.
1893 */
1894 write_unlock_bh(&bond->lock);
1895 read_lock(&bond->lock);
1896 write_lock_bh(&bond->curr_slave_lock);
1897
1da177e4
LT		 1898 bond_select_active_slave(bond);
1899
059fe7a5
JV		 1900 write_unlock_bh(&bond->curr_slave_lock);
1901 read_unlock(&bond->lock);
1902 write_lock_bh(&bond->lock);
1903 }
1904
1da177e4 1905 if (bond->slave_cnt == 0) {
ff59c456
JV		 1906 bond_set_carrier(bond);
1907
1da177e4
LT		 1908 /* if the last slave was removed, zero the mac address
1909 * of the master so it will be set by the application
1910 * to the mac address of the first slave
1911 */
1912 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1913
1914 if (list_empty(&bond->vlan_list)) {
1915 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1916 } else {
1917 printk(KERN_WARNING DRV_NAME
4e0952c7 1918 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1919 "still has VLANs.\n",
4e0952c7 1920 bond_dev->name, bond_dev->name);
1da177e4 1921 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1922 ": %s: When re-adding slaves, make sure the bond's "
1923 "HW address matches its VLANs'.\n",
1924 bond_dev->name);
1da177e4
LT		 1925 }
1926 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1927 !bond_has_challenged_slaves(bond)) {
1928 printk(KERN_INFO DRV_NAME
4e0952c7 1929 ": %s: last VLAN challenged slave %s "
1da177e4 1930 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1931 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1932 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1933 }
1934
1935 write_unlock_bh(&bond->lock);
1936
b76cdba9
MW		 1937 /* must do this from outside any spinlocks */
1938 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1939
1da177e4
LT		 1940 bond_del_vlans_from_slave(bond, slave_dev);
1941
1942 /* If the mode USES_PRIMARY, then we should only remove its
1943 * promisc and mc settings if it was the curr_active_slave, but that was
1944 * already taken care of above when we detached the slave
1945 */
1946 if (!USES_PRIMARY(bond->params.mode)) {
1947 /* unset promiscuity level from slave */
1948 if (bond_dev->flags & IFF_PROMISC) {
1949 dev_set_promiscuity(slave_dev, -1);
1950 }
1951
1952 /* unset allmulti level from slave */
1953 if (bond_dev->flags & IFF_ALLMULTI) {
1954 dev_set_allmulti(slave_dev, -1);
1955 }
1956
1957 /* flush master's mc_list from slave */
b9e40857 1958 netif_addr_lock_bh(bond_dev);
1da177e4 1959 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 1960 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1961 }
1962
1963 netdev_set_master(slave_dev, NULL);
1964
1965 /* close slave before restoring its mac address */
1966 dev_close(slave_dev);
1967
3915c1e8 1968 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1969 /* restore original ("permanent") mac address */
1970 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1971 addr.sa_family = slave_dev->type;
1972 dev_set_mac_address(slave_dev, &addr);
1973 }
1da177e4 1974
8f903c70 1975 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1976 IFF_SLAVE_INACTIVE | IFF_BONDING |
1977 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1978
1979 kfree(slave);
1980
1981 return 0; /* deletion OK */
1982}
1983
d90a162a
MS		 1984/*
1985* Destroy a bonding device.
1986* Must be under rtnl_lock when this function is called.
1987*/
1988void bond_destroy(struct bonding *bond)
1989{
1990 bond_deinit(bond->dev);
1991 bond_destroy_sysfs_entry(bond);
8cbdeec6 1992 unregister_netdevice(bond->dev);
d90a162a
MS		 1993}
1994
a434e43f
JV		 1995static void bond_destructor(struct net_device *bond_dev)
1996{
1997 struct bonding *bond = bond_dev->priv;
1998
1999 if (bond->wq)
2000 destroy_workqueue(bond->wq);
2001
2002 netif_addr_lock_bh(bond_dev);
2003 bond_mc_list_destroy(bond);
2004 netif_addr_unlock_bh(bond_dev);
2005
2006 free_netdev(bond_dev);
2007}
2008
d90a162a
MS		 2009/*
2010* First release a slave and than destroy the bond if no more slaves iare left.
2011* Must be under rtnl_lock when this function is called.
2012*/
2013int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2014{
454d7c9b 2015 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 2016 int ret;
2017
2018 ret = bond_release(bond_dev, slave_dev);
2019 if ((ret == 0) && (bond->slave_cnt == 0)) {
2020 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2021 bond_dev->name, bond_dev->name);
2022 bond_destroy(bond);
2023 }
2024 return ret;
2025}
2026
1da177e4
LT		 2027/*
2028 * This function releases all slaves.
2029 */
2030static int bond_release_all(struct net_device *bond_dev)
2031{
454d7c9b 2032 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2033 struct slave *slave;
2034 struct net_device *slave_dev;
2035 struct sockaddr addr;
2036
2037 write_lock_bh(&bond->lock);
2038
ff59c456
JV		 2039 netif_carrier_off(bond_dev);
2040
1da177e4
LT		 2041 if (bond->slave_cnt == 0) {
2042 goto out;
2043 }
2044
2045 bond->current_arp_slave = NULL;
2046 bond->primary_slave = NULL;
2047 bond_change_active_slave(bond, NULL);
2048
2049 while ((slave = bond->first_slave) != NULL) {
2050 /* Inform AD package of unbinding of slave
2051 * before slave is detached from the list.
2052 */
2053 if (bond->params.mode == BOND_MODE_8023AD) {
2054 bond_3ad_unbind_slave(slave);
2055 }
2056
2057 slave_dev = slave->dev;
2058 bond_detach_slave(bond, slave);
2059
2543331d
JV		 2060 /* now that the slave is detached, unlock and perform
2061 * all the undo steps that should not be called from
2062 * within a lock.
2063 */
2064 write_unlock_bh(&bond->lock);
2065
1da177e4
LT		 2066 if ((bond->params.mode == BOND_MODE_TLB) ||
2067 (bond->params.mode == BOND_MODE_ALB)) {
2068 /* must be called only after the slave
2069 * has been detached from the list
2070 */
2071 bond_alb_deinit_slave(bond, slave);
2072 }
2073
8531c5ff
AK		 2074 bond_compute_features(bond);
2075
b76cdba9 2076 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2077 bond_del_vlans_from_slave(bond, slave_dev);
2078
2079 /* If the mode USES_PRIMARY, then we should only remove its
2080 * promisc and mc settings if it was the curr_active_slave, but that was
2081 * already taken care of above when we detached the slave
2082 */
2083 if (!USES_PRIMARY(bond->params.mode)) {
2084 /* unset promiscuity level from slave */
2085 if (bond_dev->flags & IFF_PROMISC) {
2086 dev_set_promiscuity(slave_dev, -1);
2087 }
2088
2089 /* unset allmulti level from slave */
2090 if (bond_dev->flags & IFF_ALLMULTI) {
2091 dev_set_allmulti(slave_dev, -1);
2092 }
2093
2094 /* flush master's mc_list from slave */
b9e40857 2095 netif_addr_lock_bh(bond_dev);
1da177e4 2096 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2097 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2098 }
2099
2100 netdev_set_master(slave_dev, NULL);
2101
2102 /* close slave before restoring its mac address */
2103 dev_close(slave_dev);
2104
dd957c57 2105 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2106 /* restore original ("permanent") mac address*/
2107 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2108 addr.sa_family = slave_dev->type;
2109 dev_set_mac_address(slave_dev, &addr);
2110 }
1da177e4 2111
8f903c70
JV		 2112 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2113 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2114
2115 kfree(slave);
2116
2117 /* re-acquire the lock before getting the next slave */
2118 write_lock_bh(&bond->lock);
2119 }
2120
2121 /* zero the mac address of the master so it will be
2122 * set by the application to the mac address of the
2123 * first slave
2124 */
2125 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2126
2127 if (list_empty(&bond->vlan_list)) {
2128 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2129 } else {
2130 printk(KERN_WARNING DRV_NAME
4e0952c7 2131 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2132 "still has VLANs.\n",
4e0952c7 2133 bond_dev->name, bond_dev->name);
1da177e4 2134 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2135 ": %s: When re-adding slaves, make sure the bond's "
2136 "HW address matches its VLANs'.\n",
2137 bond_dev->name);
1da177e4
LT		 2138 }
2139
2140 printk(KERN_INFO DRV_NAME
2141 ": %s: released all slaves\n",
2142 bond_dev->name);
2143
2144out:
2145 write_unlock_bh(&bond->lock);
2146
2147 return 0;
2148}
2149
2150/*
2151 * This function changes the active slave to slave <slave_dev>.
2152 * It returns -EINVAL in the following cases.
2153 * - <slave_dev> is not found in the list.
2154 * - There is not active slave now.
2155 * - <slave_dev> is already active.
2156 * - The link state of <slave_dev> is not BOND_LINK_UP.
2157 * - <slave_dev> is not running.
2158 * In these cases, this fuction does nothing.
2159 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2160 */
2161static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2162{
454d7c9b 2163 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2164 struct slave *old_active = NULL;
2165 struct slave *new_active = NULL;
2166 int res = 0;
2167
2168 if (!USES_PRIMARY(bond->params.mode)) {
2169 return -EINVAL;
2170 }
2171
2172 /* Verify that master_dev is indeed the master of slave_dev */
2173 if (!(slave_dev->flags & IFF_SLAVE) ||
2174 (slave_dev->master != bond_dev)) {
2175 return -EINVAL;
2176 }
2177
059fe7a5 2178 read_lock(&bond->lock);
1da177e4 2179
059fe7a5 2180 read_lock(&bond->curr_slave_lock);
1da177e4 2181 old_active = bond->curr_active_slave;
059fe7a5
JV		 2182 read_unlock(&bond->curr_slave_lock);
2183
1da177e4
LT		 2184 new_active = bond_get_slave_by_dev(bond, slave_dev);
2185
2186 /*
2187 * Changing to the current active: do nothing; return success.
2188 */
2189 if (new_active && (new_active == old_active)) {
059fe7a5 2190 read_unlock(&bond->lock);
1da177e4
LT		 2191 return 0;
2192 }
2193
2194 if ((new_active) &&
2195 (old_active) &&
2196 (new_active->link == BOND_LINK_UP) &&
2197 IS_UP(new_active->dev)) {
059fe7a5 2198 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2199 bond_change_active_slave(bond, new_active);
059fe7a5 2200 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2201 } else {
2202 res = -EINVAL;
2203 }
2204
059fe7a5 2205 read_unlock(&bond->lock);
1da177e4
LT		 2206
2207 return res;
2208}
2209
1da177e4
LT		 2210static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2211{
454d7c9b 2212 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2213
2214 info->bond_mode = bond->params.mode;
2215 info->miimon = bond->params.miimon;
2216
6603a6f2 2217 read_lock(&bond->lock);
1da177e4 2218 info->num_slaves = bond->slave_cnt;
6603a6f2 2219 read_unlock(&bond->lock);
1da177e4
LT		 2220
2221 return 0;
2222}
2223
2224static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2225{
454d7c9b 2226 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2227 struct slave *slave;
2228 int i, found = 0;
2229
2230 if (info->slave_id < 0) {
2231 return -ENODEV;
2232 }
2233
6603a6f2 2234 read_lock(&bond->lock);
1da177e4
LT		 2235
2236 bond_for_each_slave(bond, slave, i) {
2237 if (i == (int)info->slave_id) {
2238 found = 1;
2239 break;
2240 }
2241 }
2242
6603a6f2 2243 read_unlock(&bond->lock);
1da177e4
LT		 2244
2245 if (found) {
2246 strcpy(info->slave_name, slave->dev->name);
2247 info->link = slave->link;
2248 info->state = slave->state;
2249 info->link_failure_count = slave->link_failure_count;
2250 } else {
2251 return -ENODEV;
2252 }
2253
2254 return 0;
2255}
2256
2257/*-------------------------------- Monitoring -------------------------------*/
2258
1da177e4 2259
f0c76d61
JV		 2260static int bond_miimon_inspect(struct bonding *bond)
2261{
2262 struct slave *slave;
2263 int i, link_state, commit = 0;
1da177e4
LT		 2264
2265 bond_for_each_slave(bond, slave, i) {
f0c76d61 2266 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2267
f0c76d61 2268 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2269
2270 switch (slave->link) {
f0c76d61
JV		 2271 case BOND_LINK_UP:
2272 if (link_state)
2273 continue;
1da177e4 2274
f0c76d61
JV		 2275 slave->link = BOND_LINK_FAIL;
2276 slave->delay = bond->params.downdelay;
2277 if (slave->delay) {
2278 printk(KERN_INFO DRV_NAME
2279 ": %s: link status down for %s"
2280 "interface %s, disabling it in %d ms.\n",
2281 bond->dev->name,
2282 (bond->params.mode ==
2283 BOND_MODE_ACTIVEBACKUP) ?
2284 ((slave->state == BOND_STATE_ACTIVE) ?
2285 "active " : "backup ") : "",
2286 slave->dev->name,
2287 bond->params.downdelay * bond->params.miimon);
1da177e4 2288 }
f0c76d61
JV		 2289 /*FALLTHRU*/
2290 case BOND_LINK_FAIL:
2291 if (link_state) {
2292 /*
2293 * recovered before downdelay expired
2294 */
2295 slave->link = BOND_LINK_UP;
1da177e4
LT		 2296 slave->jiffies = jiffies;
2297 printk(KERN_INFO DRV_NAME
2298 ": %s: link status up again after %d "
2299 "ms for interface %s.\n",
1b76b316 2300 bond->dev->name,
f0c76d61
JV		 2301 (bond->params.downdelay - slave->delay) *
2302 bond->params.miimon,
2303 slave->dev->name);
2304 continue;
1da177e4 2305 }
f0c76d61
JV		 2306
2307 if (slave->delay <= 0) {
2308 slave->new_link = BOND_LINK_DOWN;
2309 commit++;
2310 continue;
1da177e4 2311 }
1da177e4 2312
f0c76d61
JV		 2313 slave->delay--;
2314 break;
2315
2316 case BOND_LINK_DOWN:
2317 if (!link_state)
2318 continue;
2319
2320 slave->link = BOND_LINK_BACK;
2321 slave->delay = bond->params.updelay;
2322
2323 if (slave->delay) {
2324 printk(KERN_INFO DRV_NAME
2325 ": %s: link status up for "
2326 "interface %s, enabling it in %d ms.\n",
2327 bond->dev->name, slave->dev->name,
2328 bond->params.updelay *
2329 bond->params.miimon);
2330 }
2331 /*FALLTHRU*/
2332 case BOND_LINK_BACK:
2333 if (!link_state) {
2334 slave->link = BOND_LINK_DOWN;
1da177e4
LT		 2335 printk(KERN_INFO DRV_NAME
2336 ": %s: link status down again after %d "
2337 "ms for interface %s.\n",
1b76b316 2338 bond->dev->name,
f0c76d61
JV		 2339 (bond->params.updelay - slave->delay) *
2340 bond->params.miimon,
2341 slave->dev->name);
2342
2343 continue;
2344 }
2345
2346 if (slave->delay <= 0) {
2347 slave->new_link = BOND_LINK_UP;
2348 commit++;
2349 continue;
1da177e4 2350 }
f0c76d61
JV		 2351
2352 slave->delay--;
1da177e4 2353 break;
f0c76d61
JV		 2354 }
2355 }
1da177e4 2356
f0c76d61
JV		 2357 return commit;
2358}
1da177e4 2359
f0c76d61
JV		 2360static void bond_miimon_commit(struct bonding *bond)
2361{
2362 struct slave *slave;
2363 int i;
2364
2365 bond_for_each_slave(bond, slave, i) {
2366 switch (slave->new_link) {
2367 case BOND_LINK_NOCHANGE:
2368 continue;
1da177e4 2369
f0c76d61
JV		 2370 case BOND_LINK_UP:
2371 slave->link = BOND_LINK_UP;
2372 slave->jiffies = jiffies;
2373
2374 if (bond->params.mode == BOND_MODE_8023AD) {
2375 /* prevent it from being the active one */
2376 slave->state = BOND_STATE_BACKUP;
2377 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2378 /* make it immediately active */
2379 slave->state = BOND_STATE_ACTIVE;
2380 } else if (slave != bond->primary_slave) {
2381 /* prevent it from being the active one */
2382 slave->state = BOND_STATE_BACKUP;
1da177e4 2383 }
1da177e4 2384
f0c76d61
JV		 2385 printk(KERN_INFO DRV_NAME
2386 ": %s: link status definitely "
2387 "up for interface %s.\n",
2388 bond->dev->name, slave->dev->name);
1da177e4 2389
f0c76d61
JV		 2390 /* notify ad that the link status has changed */
2391 if (bond->params.mode == BOND_MODE_8023AD)
2392 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2393
f0c76d61
JV		 2394 if ((bond->params.mode == BOND_MODE_TLB) ||
2395 (bond->params.mode == BOND_MODE_ALB))
2396 bond_alb_handle_link_change(bond, slave,
2397 BOND_LINK_UP);
1da177e4 2398
f0c76d61
JV		 2399 if (!bond->curr_active_slave ||
2400 (slave == bond->primary_slave))
2401 goto do_failover;
1da177e4 2402
f0c76d61 2403 continue;
059fe7a5 2404
f0c76d61 2405 case BOND_LINK_DOWN:
fba4acda
JV		 2406 if (slave->link_failure_count < UINT_MAX)
2407 slave->link_failure_count++;
2408
f0c76d61 2409 slave->link = BOND_LINK_DOWN;
1da177e4 2410
f0c76d61
JV		 2411 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2412 bond->params.mode == BOND_MODE_8023AD)
2413 bond_set_slave_inactive_flags(slave);
2414
2415 printk(KERN_INFO DRV_NAME
2416 ": %s: link status definitely down for "
2417 "interface %s, disabling it\n",
2418 bond->dev->name, slave->dev->name);
2419
2420 if (bond->params.mode == BOND_MODE_8023AD)
2421 bond_3ad_handle_link_change(slave,
2422 BOND_LINK_DOWN);
2423
2424 if (bond->params.mode == BOND_MODE_TLB ||
2425 bond->params.mode == BOND_MODE_ALB)
2426 bond_alb_handle_link_change(bond, slave,
2427 BOND_LINK_DOWN);
2428
2429 if (slave == bond->curr_active_slave)
2430 goto do_failover;
2431
2432 continue;
2433
2434 default:
2435 printk(KERN_ERR DRV_NAME
2436 ": %s: invalid new link %d on slave %s\n",
2437 bond->dev->name, slave->new_link,
2438 slave->dev->name);
2439 slave->new_link = BOND_LINK_NOCHANGE;
2440
2441 continue;
2442 }
2443
2444do_failover:
2445 ASSERT_RTNL();
2446 write_lock_bh(&bond->curr_slave_lock);
2447 bond_select_active_slave(bond);
2448 write_unlock_bh(&bond->curr_slave_lock);
2449 }
2450
2451 bond_set_carrier(bond);
1da177e4
LT		 2452}
2453
0b0eef66
JV		 2454/*
2455 * bond_mii_monitor
2456 *
2457 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2458 * inspection, then (if inspection indicates something needs to be done)
2459 * an acquisition of appropriate locks followed by a commit phase to
2460 * implement whatever link state changes are indicated.
0b0eef66
JV		 2461 */
2462void bond_mii_monitor(struct work_struct *work)
2463{
2464 struct bonding *bond = container_of(work, struct bonding,
2465 mii_work.work);
0b0eef66
JV		 2466
2467 read_lock(&bond->lock);
f0c76d61
JV		 2468 if (bond->kill_timers)
2469 goto out;
2470
2471 if (bond->slave_cnt == 0)
2472 goto re_arm;
b59f9f74
JV		 2473
2474 if (bond->send_grat_arp) {
2475 read_lock(&bond->curr_slave_lock);
2476 bond_send_gratuitous_arp(bond);
2477 read_unlock(&bond->curr_slave_lock);
2478 }
2479
305d552a
BH		 2480 if (bond->send_unsol_na) {
2481 read_lock(&bond->curr_slave_lock);
2482 bond_send_unsolicited_na(bond);
2483 read_unlock(&bond->curr_slave_lock);
2484 }
2485
f0c76d61 2486 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2487 read_unlock(&bond->lock);
2488 rtnl_lock();
2489 read_lock(&bond->lock);
f0c76d61
JV		 2490
2491 bond_miimon_commit(bond);
2492
5655662d
JV		 2493 read_unlock(&bond->lock);
2494 rtnl_unlock(); /* might sleep, hold no other locks */
2495 read_lock(&bond->lock);
0b0eef66
JV		 2496 }
2497
f0c76d61
JV		 2498re_arm:
2499 if (bond->params.miimon)
2500 queue_delayed_work(bond->wq, &bond->mii_work,
2501 msecs_to_jiffies(bond->params.miimon));
2502out:
0b0eef66 2503 read_unlock(&bond->lock);
0b0eef66 2504}
c3ade5ca 2505
d3bb52b0 2506static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2507{
2508 struct in_device *idev;
2509 struct in_ifaddr *ifa;
a144ea4b 2510 __be32 addr = 0;
c3ade5ca
JV		 2511
2512 if (!dev)
2513 return 0;
2514
2515 rcu_read_lock();
e5ed6399 2516 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2517 if (!idev)
2518 goto out;
2519
2520 ifa = idev->ifa_list;
2521 if (!ifa)
2522 goto out;
2523
2524 addr = ifa->ifa_local;
2525out:
2526 rcu_read_unlock();
2527 return addr;
2528}
2529
d3bb52b0 2530static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2531{
0883beca 2532 struct vlan_entry *vlan;
f5b2b966
JV		 2533
2534 if (ip == bond->master_ip)
2535 return 1;
2536
0883beca 2537 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2538 if (ip == vlan->vlan_ip)
2539 return 1;
2540 }
2541
2542 return 0;
2543}
2544
c3ade5ca
JV		 2545/*
2546 * We go to the (large) trouble of VLAN tagging ARP frames because
2547 * switches in VLAN mode (especially if ports are configured as
2548 * "native" to a VLAN) might not pass non-tagged frames.
2549 */
d3bb52b0 2550static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2551{
2552 struct sk_buff *skb;
2553
2554 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2555 slave_dev->name, dest_ip, src_ip, vlan_id);
2556
2557 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2558 NULL, slave_dev->dev_addr, NULL);
2559
2560 if (!skb) {
2561 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2562 return;
2563 }
2564 if (vlan_id) {
2565 skb = vlan_put_tag(skb, vlan_id);
2566 if (!skb) {
2567 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2568 return;
2569 }
2570 }
2571 arp_xmit(skb);
2572}
2573
2574
1da177e4
LT		 2575static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2576{
c3ade5ca 2577 int i, vlan_id, rv;
d3bb52b0 2578 __be32 *targets = bond->params.arp_targets;
0883beca 2579 struct vlan_entry *vlan;
c3ade5ca
JV		 2580 struct net_device *vlan_dev;
2581 struct flowi fl;
2582 struct rtable *rt;
1da177e4 2583
6b780567
MW		 2584 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2585 if (!targets[i])
2586 continue;
c3ade5ca
JV		 2587 dprintk("basa: target %x\n", targets[i]);
2588 if (list_empty(&bond->vlan_list)) {
2589 dprintk("basa: empty vlan: arp_send\n");
2590 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2591 bond->master_ip, 0);
2592 continue;
2593 }
2594
2595 /*
2596 * If VLANs are configured, we do a route lookup to
2597 * determine which VLAN interface would be used, so we
2598 * can tag the ARP with the proper VLAN tag.
2599 */
2600 memset(&fl, 0, sizeof(fl));
2601 fl.fl4_dst = targets[i];
2602 fl.fl4_tos = RTO_ONLINK;
2603
f206351a 2604 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2605 if (rv) {
2606 if (net_ratelimit()) {
2607 printk(KERN_WARNING DRV_NAME
63779436
HH		 2608 ": %s: no route to arp_ip_target %pI4\n",
2609 bond->dev->name, &fl.fl4_dst);
c3ade5ca
JV		 2610 }
2611 continue;
2612 }
2613
2614 /*
2615 * This target is not on a VLAN
2616 */
2617 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2618 ip_rt_put(rt);
c3ade5ca
JV		 2619 dprintk("basa: rtdev == bond->dev: arp_send\n");
2620 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2621 bond->master_ip, 0);
2622 continue;
2623 }
2624
2625 vlan_id = 0;
0883beca 2626 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2627 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2628 if (vlan_dev == rt->u.dst.dev) {
2629 vlan_id = vlan->vlan_id;
2630 dprintk("basa: vlan match on %s %d\n",
2631 vlan_dev->name, vlan_id);
2632 break;
2633 }
2634 }
2635
2636 if (vlan_id) {
ed4b9f80 2637 ip_rt_put(rt);
c3ade5ca
JV		 2638 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2639 vlan->vlan_ip, vlan_id);
2640 continue;
2641 }
2642
2643 if (net_ratelimit()) {
2644 printk(KERN_WARNING DRV_NAME
63779436
HH		 2645 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2646 bond->dev->name, &fl.fl4_dst,
c3ade5ca
JV		 2647 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2648 }
ed4b9f80 2649 ip_rt_put(rt);
c3ade5ca
JV		 2650 }
2651}
2652
2653/*
2654 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2655 * for each VLAN above us.
b59f9f74
JV		 2656 *
2657 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2658 */
2659static void bond_send_gratuitous_arp(struct bonding *bond)
2660{
2661 struct slave *slave = bond->curr_active_slave;
2662 struct vlan_entry *vlan;
2663 struct net_device *vlan_dev;
2664
2665 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2666 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2667
2668 if (!slave || !bond->send_grat_arp ||
2669 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2670 return;
2671
b59f9f74
JV		 2672 bond->send_grat_arp--;
2673
c3ade5ca
JV		 2674 if (bond->master_ip) {
2675 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2676 bond->master_ip, 0);
c3ade5ca
JV		 2677 }
2678
2679 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2680 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2681 if (vlan->vlan_ip) {
2682 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2683 vlan->vlan_ip, vlan->vlan_id);
2684 }
1da177e4
LT		 2685 }
2686}
2687
d3bb52b0 2688static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2689{
2690 int i;
d3bb52b0 2691 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2692
2693 targets = bond->params.arp_targets;
2694 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
63779436
HH		 2695 dprintk("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2696 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
f5b2b966
JV		 2697 if (sip == targets[i]) {
2698 if (bond_has_this_ip(bond, tip))
2699 slave->last_arp_rx = jiffies;
2700 return;
2701 }
2702 }
2703}
2704
2705static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2706{
2707 struct arphdr *arp;
2708 struct slave *slave;
2709 struct bonding *bond;
2710 unsigned char *arp_ptr;
d3bb52b0 2711 __be32 sip, tip;
f5b2b966 2712
c346dca1 2713 if (dev_net(dev) != &init_net)
e730c155
EB		 2714 goto out;
2715
f5b2b966
JV		 2716 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2717 goto out;
2718
454d7c9b 2719 bond = netdev_priv(dev);
f5b2b966
JV		 2720 read_lock(&bond->lock);
2721
2722 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2723 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2724 orig_dev ? orig_dev->name : "NULL");
2725
2726 slave = bond_get_slave_by_dev(bond, orig_dev);
2727 if (!slave || !slave_do_arp_validate(bond, slave))
2728 goto out_unlock;
2729
988b7050 2730 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2731 goto out_unlock;
2732
d0a92be0 2733 arp = arp_hdr(skb);
f5b2b966
JV		 2734 if (arp->ar_hln != dev->addr_len ||
2735 skb->pkt_type == PACKET_OTHERHOST ||
2736 skb->pkt_type == PACKET_LOOPBACK ||
2737 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2738 arp->ar_pro != htons(ETH_P_IP) ||
2739 arp->ar_pln != 4)
2740 goto out_unlock;
2741
2742 arp_ptr = (unsigned char *)(arp + 1);
2743 arp_ptr += dev->addr_len;
2744 memcpy(&sip, arp_ptr, 4);
2745 arp_ptr += 4 + dev->addr_len;
2746 memcpy(&tip, arp_ptr, 4);
2747
63779436
HH		 2748 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2749 bond->dev->name, slave->dev->name, slave->state,
2750 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2751 &sip, &tip);
f5b2b966
JV		 2752
2753 /*
2754 * Backup slaves won't see the ARP reply, but do come through
2755 * here for each ARP probe (so we swap the sip/tip to validate
2756 * the probe). In a "redundant switch, common router" type of
2757 * configuration, the ARP probe will (hopefully) travel from
2758 * the active, through one switch, the router, then the other
2759 * switch before reaching the backup.
2760 */
2761 if (slave->state == BOND_STATE_ACTIVE)
2762 bond_validate_arp(bond, slave, sip, tip);
2763 else
2764 bond_validate_arp(bond, slave, tip, sip);
2765
2766out_unlock:
2767 read_unlock(&bond->lock);
2768out:
2769 dev_kfree_skb(skb);
2770 return NET_RX_SUCCESS;
2771}
2772
1da177e4
LT		 2773/*
2774 * this function is called regularly to monitor each slave's link
2775 * ensuring that traffic is being sent and received when arp monitoring
2776 * is used in load-balancing mode. if the adapter has been dormant, then an
2777 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2778 * arp monitoring in active backup mode.
2779 */
1b76b316 2780void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2781{
1b76b316
JV		 2782 struct bonding *bond = container_of(work, struct bonding,
2783 arp_work.work);
1da177e4
LT		 2784 struct slave *slave, *oldcurrent;
2785 int do_failover = 0;
2786 int delta_in_ticks;
2787 int i;
2788
2789 read_lock(&bond->lock);
2790
5ce0da8f 2791 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2792
2793 if (bond->kill_timers) {
2794 goto out;
2795 }
2796
2797 if (bond->slave_cnt == 0) {
2798 goto re_arm;
2799 }
2800
2801 read_lock(&bond->curr_slave_lock);
2802 oldcurrent = bond->curr_active_slave;
2803 read_unlock(&bond->curr_slave_lock);
2804
2805 /* see if any of the previous devices are up now (i.e. they have
2806 * xmt and rcv traffic). the curr_active_slave does not come into
2807 * the picture unless it is null. also, slave->jiffies is not needed
2808 * here because we send an arp on each slave and give a slave as
2809 * long as it needs to get the tx/rx within the delta.
2810 * TODO: what about up/down delay in arp mode? it wasn't here before
2811 * so it can wait
2812 */
2813 bond_for_each_slave(bond, slave, i) {
2814 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2815 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2816 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2817
2818 slave->link = BOND_LINK_UP;
2819 slave->state = BOND_STATE_ACTIVE;
2820
2821 /* primary_slave has no meaning in round-robin
2822 * mode. the window of a slave being up and
2823 * curr_active_slave being null after enslaving
2824 * is closed.
2825 */
2826 if (!oldcurrent) {
2827 printk(KERN_INFO DRV_NAME
2828 ": %s: link status definitely "
2829 "up for interface %s, ",
1b76b316 2830 bond->dev->name,
1da177e4
LT		 2831 slave->dev->name);
2832 do_failover = 1;
2833 } else {
2834 printk(KERN_INFO DRV_NAME
2835 ": %s: interface %s is now up\n",
1b76b316 2836 bond->dev->name,
1da177e4
LT		 2837 slave->dev->name);
2838 }
2839 }
2840 } else {
2841 /* slave->link == BOND_LINK_UP */
2842
2843 /* not all switches will respond to an arp request
2844 * when the source ip is 0, so don't take the link down
2845 * if we don't know our ip yet
2846 */
b63bb739 2847 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2848 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2849
2850 slave->link = BOND_LINK_DOWN;
2851 slave->state = BOND_STATE_BACKUP;
2852
2853 if (slave->link_failure_count < UINT_MAX) {
2854 slave->link_failure_count++;
2855 }
2856
2857 printk(KERN_INFO DRV_NAME
2858 ": %s: interface %s is now down.\n",
1b76b316 2859 bond->dev->name,
1da177e4
LT		 2860 slave->dev->name);
2861
2862 if (slave == oldcurrent) {
2863 do_failover = 1;
2864 }
2865 }
2866 }
2867
2868 /* note: if switch is in round-robin mode, all links
2869 * must tx arp to ensure all links rx an arp - otherwise
2870 * links may oscillate or not come up at all; if switch is
2871 * in something like xor mode, there is nothing we can
2872 * do - all replies will be rx'ed on same link causing slaves
2873 * to be unstable during low/no traffic periods
2874 */
2875 if (IS_UP(slave->dev)) {
2876 bond_arp_send_all(bond, slave);
2877 }
2878 }
2879
2880 if (do_failover) {
059fe7a5 2881 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2882
2883 bond_select_active_slave(bond);
2884
059fe7a5 2885 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2886 }
2887
2888re_arm:
1b76b316
JV		 2889 if (bond->params.arp_interval)
2890 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2891out:
2892 read_unlock(&bond->lock);
2893}
2894
2895/*
b2220cad
JV		 2896 * Called to inspect slaves for active-backup mode ARP monitor link state
2897 * changes. Sets new_link in slaves to specify what action should take
2898 * place for the slave. Returns 0 if no changes are found, >0 if changes
2899 * to link states must be committed.
2900 *
2901 * Called with bond->lock held for read.
1da177e4 2902 */
b2220cad 2903static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2904{
1da177e4 2905 struct slave *slave;
b2220cad 2906 int i, commit = 0;
1da177e4 2907
b2220cad
JV		 2908 bond_for_each_slave(bond, slave, i) {
2909 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2910
b2220cad
JV		 2911 if (slave->link != BOND_LINK_UP) {
2912 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2913 delta_in_ticks)) {
2914 slave->new_link = BOND_LINK_UP;
2915 commit++;
2916 }
1da177e4 2917
b2220cad
JV		 2918 continue;
2919 }
1da177e4 2920
b2220cad
JV		 2921 /*
2922 * Give slaves 2*delta after being enslaved or made
2923 * active. This avoids bouncing, as the last receive
2924 * times need a full ARP monitor cycle to be updated.
2925 */
2926 if (!time_after_eq(jiffies, slave->jiffies +
2927 2 * delta_in_ticks))
2928 continue;
2929
2930 /*
2931 * Backup slave is down if:
2932 * - No current_arp_slave AND
2933 * - more than 3*delta since last receive AND
2934 * - the bond has an IP address
2935 *
2936 * Note: a non-null current_arp_slave indicates
2937 * the curr_active_slave went down and we are
2938 * searching for a new one; under this condition
2939 * we only take the curr_active_slave down - this
2940 * gives each slave a chance to tx/rx traffic
2941 * before being taken out
2942 */
2943 if (slave->state == BOND_STATE_BACKUP &&
2944 !bond->current_arp_slave &&
2945 time_after(jiffies, slave_last_rx(bond, slave) +
2946 3 * delta_in_ticks)) {
2947 slave->new_link = BOND_LINK_DOWN;
2948 commit++;
2949 }
2950
2951 /*
2952 * Active slave is down if:
2953 * - more than 2*delta since transmitting OR
2954 * - (more than 2*delta since receive AND
2955 * the bond has an IP address)
2956 */
2957 if ((slave->state == BOND_STATE_ACTIVE) &&
2958 (time_after_eq(jiffies, slave->dev->trans_start +
2959 2 * delta_in_ticks) ||
2960 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2961 + 2 * delta_in_ticks)))) {
2962 slave->new_link = BOND_LINK_DOWN;
2963 commit++;
2964 }
1da177e4
LT		 2965 }
2966
b2220cad
JV		 2967 read_lock(&bond->curr_slave_lock);
2968
2969 /*
2970 * Trigger a commit if the primary option setting has changed.
1da177e4 2971 */
b2220cad
JV		 2972 if (bond->primary_slave &&
2973 (bond->primary_slave != bond->curr_active_slave) &&
2974 (bond->primary_slave->link == BOND_LINK_UP))
2975 commit++;
1da177e4 2976
b2220cad 2977 read_unlock(&bond->curr_slave_lock);
1da177e4 2978
b2220cad
JV		 2979 return commit;
2980}
1da177e4 2981
b2220cad
JV		 2982/*
2983 * Called to commit link state changes noted by inspection step of
2984 * active-backup mode ARP monitor.
2985 *
2986 * Called with RTNL and bond->lock for read.
2987 */
2988static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2989{
2990 struct slave *slave;
2991 int i;
1da177e4 2992
b2220cad
JV		 2993 bond_for_each_slave(bond, slave, i) {
2994 switch (slave->new_link) {
2995 case BOND_LINK_NOCHANGE:
2996 continue;
ff59c456 2997
b2220cad
JV		 2998 case BOND_LINK_UP:
2999 write_lock_bh(&bond->curr_slave_lock);
1da177e4 3000
b2220cad
JV		 3001 if (!bond->curr_active_slave &&
3002 time_before_eq(jiffies, slave->dev->trans_start +
3003 delta_in_ticks)) {
3004 slave->link = BOND_LINK_UP;
3005 bond_change_active_slave(bond, slave);
3006 bond->current_arp_slave = NULL;
3007
3008 printk(KERN_INFO DRV_NAME
3009 ": %s: %s is up and now the "
3010 "active interface\n",
3011 bond->dev->name, slave->dev->name);
3012
3013 } else if (bond->curr_active_slave != slave) {
3014 /* this slave has just come up but we
3015 * already have a current slave; this can
3016 * also happen if bond_enslave adds a new
3017 * slave that is up while we are searching
3018 * for a new slave
1da177e4 3019 */
b2220cad
JV		 3020 slave->link = BOND_LINK_UP;
3021 bond_set_slave_inactive_flags(slave);
3022 bond->current_arp_slave = NULL;
3023
3024 printk(KERN_INFO DRV_NAME
3025 ": %s: backup interface %s is now up\n",
3026 bond->dev->name, slave->dev->name);
3027 }
1da177e4 3028
b2220cad 3029 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3030
b2220cad 3031 break;
1da177e4 3032
b2220cad
JV		 3033 case BOND_LINK_DOWN:
3034 if (slave->link_failure_count < UINT_MAX)
3035 slave->link_failure_count++;
3036
3037 slave->link = BOND_LINK_DOWN;
3038
3039 if (slave == bond->curr_active_slave) {
3040 printk(KERN_INFO DRV_NAME
3041 ": %s: link status down for active "
3042 "interface %s, disabling it\n",
3043 bond->dev->name, slave->dev->name);
1da177e4
LT		 3044
3045 bond_set_slave_inactive_flags(slave);
3046
b2220cad
JV		 3047 write_lock_bh(&bond->curr_slave_lock);
3048
3049 bond_select_active_slave(bond);
3050 if (bond->curr_active_slave)
3051 bond->curr_active_slave->jiffies =
3052 jiffies;
3053
3054 write_unlock_bh(&bond->curr_slave_lock);
3055
3056 bond->current_arp_slave = NULL;
3057
3058 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3059 printk(KERN_INFO DRV_NAME
3060 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3061 bond->dev->name, slave->dev->name);
3062
3063 bond_set_slave_inactive_flags(slave);
1da177e4 3064 }
b2220cad
JV		 3065 break;
3066
3067 default:
3068 printk(KERN_ERR DRV_NAME
3069 ": %s: impossible: new_link %d on slave %s\n",
3070 bond->dev->name, slave->new_link,
3071 slave->dev->name);
1da177e4
LT		 3072 }
3073 }
3074
b2220cad
JV		 3075 /*
3076 * No race with changes to primary via sysfs, as we hold rtnl.
3077 */
3078 if (bond->primary_slave &&
3079 (bond->primary_slave != bond->curr_active_slave) &&
3080 (bond->primary_slave->link == BOND_LINK_UP)) {
3081 write_lock_bh(&bond->curr_slave_lock);
3082 bond_change_active_slave(bond, bond->primary_slave);
3083 write_unlock_bh(&bond->curr_slave_lock);
3084 }
1da177e4 3085
b2220cad
JV		 3086 bond_set_carrier(bond);
3087}
1da177e4 3088
b2220cad
JV		 3089/*
3090 * Send ARP probes for active-backup mode ARP monitor.
3091 *
3092 * Called with bond->lock held for read.
3093 */
3094static void bond_ab_arp_probe(struct bonding *bond)
3095{
3096 struct slave *slave;
3097 int i;
1da177e4 3098
b2220cad 3099 read_lock(&bond->curr_slave_lock);
1da177e4 3100
b2220cad
JV		 3101 if (bond->current_arp_slave && bond->curr_active_slave)
3102 printk("PROBE: c_arp %s && cas %s BAD\n",
3103 bond->current_arp_slave->dev->name,
3104 bond->curr_active_slave->dev->name);
1da177e4 3105
b2220cad
JV		 3106 if (bond->curr_active_slave) {
3107 bond_arp_send_all(bond, bond->curr_active_slave);
3108 read_unlock(&bond->curr_slave_lock);
3109 return;
3110 }
1da177e4 3111
b2220cad 3112 read_unlock(&bond->curr_slave_lock);
059fe7a5 3113
b2220cad
JV		 3114 /* if we don't have a curr_active_slave, search for the next available
3115 * backup slave from the current_arp_slave and make it the candidate
3116 * for becoming the curr_active_slave
3117 */
1da177e4 3118
b2220cad
JV		 3119 if (!bond->current_arp_slave) {
3120 bond->current_arp_slave = bond->first_slave;
3121 if (!bond->current_arp_slave)
3122 return;
3123 }
1da177e4 3124
b2220cad 3125 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3126
b2220cad
JV		 3127 /* search for next candidate */
3128 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3129 if (IS_UP(slave->dev)) {
3130 slave->link = BOND_LINK_BACK;
3131 bond_set_slave_active_flags(slave);
3132 bond_arp_send_all(bond, slave);
1da177e4 3133 slave->jiffies = jiffies;
b2220cad
JV		 3134 bond->current_arp_slave = slave;
3135 break;
1da177e4
LT		 3136 }
3137
b2220cad
JV		 3138 /* if the link state is up at this point, we
3139 * mark it down - this can happen if we have
3140 * simultaneous link failures and
3141 * reselect_active_interface doesn't make this
3142 * one the current slave so it is still marked
3143 * up when it is actually down
1da177e4 3144 */
b2220cad
JV		 3145 if (slave->link == BOND_LINK_UP) {
3146 slave->link = BOND_LINK_DOWN;
3147 if (slave->link_failure_count < UINT_MAX)
3148 slave->link_failure_count++;
1da177e4 3149
b2220cad
JV		 3150 bond_set_slave_inactive_flags(slave);
3151
3152 printk(KERN_INFO DRV_NAME
3153 ": %s: backup interface %s is now down.\n",
3154 bond->dev->name, slave->dev->name);
1da177e4 3155 }
b2220cad
JV		 3156 }
3157}
1da177e4 3158
b2220cad
JV		 3159void bond_activebackup_arp_mon(struct work_struct *work)
3160{
3161 struct bonding *bond = container_of(work, struct bonding,
3162 arp_work.work);
3163 int delta_in_ticks;
1da177e4 3164
b2220cad 3165 read_lock(&bond->lock);
1da177e4 3166
b2220cad
JV		 3167 if (bond->kill_timers)
3168 goto out;
1da177e4 3169
b2220cad 3170 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3171
b2220cad
JV		 3172 if (bond->slave_cnt == 0)
3173 goto re_arm;
3174
b59f9f74
JV		 3175 if (bond->send_grat_arp) {
3176 read_lock(&bond->curr_slave_lock);
3177 bond_send_gratuitous_arp(bond);
3178 read_unlock(&bond->curr_slave_lock);
3179 }
3180
305d552a
BH		 3181 if (bond->send_unsol_na) {
3182 read_lock(&bond->curr_slave_lock);
3183 bond_send_unsolicited_na(bond);
3184 read_unlock(&bond->curr_slave_lock);
3185 }
3186
b2220cad
JV		 3187 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3188 read_unlock(&bond->lock);
3189 rtnl_lock();
3190 read_lock(&bond->lock);
3191
3192 bond_ab_arp_commit(bond, delta_in_ticks);
3193
3194 read_unlock(&bond->lock);
3195 rtnl_unlock();
3196 read_lock(&bond->lock);
1da177e4
LT		 3197 }
3198
b2220cad
JV		 3199 bond_ab_arp_probe(bond);
3200
1da177e4
LT		 3201re_arm:
3202 if (bond->params.arp_interval) {
1b76b316 3203 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3204 }
3205out:
3206 read_unlock(&bond->lock);
3207}
3208
3209/*------------------------------ proc/seq_file-------------------------------*/
3210
3211#ifdef CONFIG_PROC_FS
3212
1da177e4
LT		 3213static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3214{
3215 struct bonding *bond = seq->private;
3216 loff_t off = 0;
3217 struct slave *slave;
3218 int i;
3219
3220 /* make sure the bond won't be taken away */
3221 read_lock(&dev_base_lock);
6603a6f2 3222 read_lock(&bond->lock);
1da177e4
LT		 3223
3224 if (*pos == 0) {
3225 return SEQ_START_TOKEN;
3226 }
3227
3228 bond_for_each_slave(bond, slave, i) {
3229 if (++off == *pos) {
3230 return slave;
3231 }
3232 }
3233
3234 return NULL;
3235}
3236
3237static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3238{
3239 struct bonding *bond = seq->private;
3240 struct slave *slave = v;
3241
3242 ++*pos;
3243 if (v == SEQ_START_TOKEN) {
3244 return bond->first_slave;
3245 }
3246
3247 slave = slave->next;
3248
3249 return (slave == bond->first_slave) ? NULL : slave;
3250}
3251
3252static void bond_info_seq_stop(struct seq_file *seq, void *v)
3253{
3254 struct bonding *bond = seq->private;
3255
6603a6f2 3256 read_unlock(&bond->lock);
1da177e4
LT		 3257 read_unlock(&dev_base_lock);
3258}
3259
3260static void bond_info_show_master(struct seq_file *seq)
3261{
3262 struct bonding *bond = seq->private;
3263 struct slave *curr;
4756b02f 3264 int i;
1da177e4
LT		 3265
3266 read_lock(&bond->curr_slave_lock);
3267 curr = bond->curr_active_slave;
3268 read_unlock(&bond->curr_slave_lock);
3269
dd957c57 3270 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3271 bond_mode_name(bond->params.mode));
3272
dd957c57
JV		 3273 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3274 bond->params.fail_over_mac)
3915c1e8
JV		 3275 seq_printf(seq, " (fail_over_mac %s)",
3276 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3277
3278 seq_printf(seq, "\n");
3279
c61b75ad
MW		 3280 if (bond->params.mode == BOND_MODE_XOR ||
3281 bond->params.mode == BOND_MODE_8023AD) {
3282 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3283 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3284 bond->params.xmit_policy);
3285 }
3286
1da177e4
LT		 3287 if (USES_PRIMARY(bond->params.mode)) {
3288 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3289 (bond->primary_slave) ?
3290 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3291
3292 seq_printf(seq, "Currently Active Slave: %s\n",
3293 (curr) ? curr->dev->name : "None");
3294 }
3295
ff59c456
JV		 3296 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3297 "up" : "down");
1da177e4
LT		 3298 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3299 seq_printf(seq, "Up Delay (ms): %d\n",
3300 bond->params.updelay * bond->params.miimon);
3301 seq_printf(seq, "Down Delay (ms): %d\n",
3302 bond->params.downdelay * bond->params.miimon);
3303
4756b02f
MW		 3304
3305 /* ARP information */
3306 if(bond->params.arp_interval > 0) {
3307 int printed=0;
3308 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3309 bond->params.arp_interval);
3310
3311 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3312
3313 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3314 if (!bond->params.arp_targets[i])
3315 continue;
3316 if (printed)
3317 seq_printf(seq, ",");
8cf14e38 3318 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
4756b02f
MW		 3319 printed = 1;
3320 }
3321 seq_printf(seq, "\n");
3322 }
3323
1da177e4
LT		 3324 if (bond->params.mode == BOND_MODE_8023AD) {
3325 struct ad_info ad_info;
3326
3327 seq_puts(seq, "\n802.3ad info\n");
3328 seq_printf(seq, "LACP rate: %s\n",
3329 (bond->params.lacp_fast) ? "fast" : "slow");
fd989c83
JV		 3330 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3331 ad_select_tbl[bond->params.ad_select].modename);
1da177e4
LT		 3332
3333 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3334 seq_printf(seq, "bond %s has no active aggregator\n",
3335 bond->dev->name);
3336 } else {
3337 seq_printf(seq, "Active Aggregator Info:\n");
3338
3339 seq_printf(seq, "\tAggregator ID: %d\n",
3340 ad_info.aggregator_id);
3341 seq_printf(seq, "\tNumber of ports: %d\n",
3342 ad_info.ports);
3343 seq_printf(seq, "\tActor Key: %d\n",
3344 ad_info.actor_key);
3345 seq_printf(seq, "\tPartner Key: %d\n",
3346 ad_info.partner_key);
e174961c
JB		 3347 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3348 ad_info.partner_system);
1da177e4
LT		 3349 }
3350 }
3351}
3352
3353static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3354{
3355 struct bonding *bond = seq->private;
3356
3357 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3358 seq_printf(seq, "MII Status: %s\n",
3359 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3360 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3361 slave->link_failure_count);
3362
e174961c 3363 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
1da177e4
LT		 3364
3365 if (bond->params.mode == BOND_MODE_8023AD) {
3366 const struct aggregator *agg
3367 = SLAVE_AD_INFO(slave).port.aggregator;
3368
3369 if (agg) {
3370 seq_printf(seq, "Aggregator ID: %d\n",
3371 agg->aggregator_identifier);
3372 } else {
3373 seq_puts(seq, "Aggregator ID: N/A\n");
3374 }
3375 }
3376}
3377
3378static int bond_info_seq_show(struct seq_file *seq, void *v)
3379{
3380 if (v == SEQ_START_TOKEN) {
3381 seq_printf(seq, "%s\n", version);
3382 bond_info_show_master(seq);
3383 } else {
3384 bond_info_show_slave(seq, v);
3385 }
3386
3387 return 0;
3388}
3389
3390static struct seq_operations bond_info_seq_ops = {
3391 .start = bond_info_seq_start,
3392 .next = bond_info_seq_next,
3393 .stop = bond_info_seq_stop,
3394 .show = bond_info_seq_show,
3395};
3396
3397static int bond_info_open(struct inode *inode, struct file *file)
3398{
3399 struct seq_file *seq;
3400 struct proc_dir_entry *proc;
3401 int res;
3402
3403 res = seq_open(file, &bond_info_seq_ops);
3404 if (!res) {
3405 /* recover the pointer buried in proc_dir_entry data */
3406 seq = file->private_data;
3407 proc = PDE(inode);
3408 seq->private = proc->data;
3409 }
3410
3411 return res;
3412}
3413
d54b1fdb 3414static const struct file_operations bond_info_fops = {
1da177e4
LT		 3415 .owner = THIS_MODULE,
3416 .open = bond_info_open,
3417 .read = seq_read,
3418 .llseek = seq_lseek,
3419 .release = seq_release,
3420};
3421
3422static int bond_create_proc_entry(struct bonding *bond)
3423{
3424 struct net_device *bond_dev = bond->dev;
3425
3426 if (bond_proc_dir) {
a95609cb
DL		 3427 bond->proc_entry = proc_create_data(bond_dev->name,
3428 S_IRUGO, bond_proc_dir,
3429 &bond_info_fops, bond);
1da177e4
LT		 3430 if (bond->proc_entry == NULL) {
3431 printk(KERN_WARNING DRV_NAME
3432 ": Warning: Cannot create /proc/net/%s/%s\n",
3433 DRV_NAME, bond_dev->name);
3434 } else {
1da177e4
LT		 3435 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3436 }
3437 }
3438
3439 return 0;
3440}
3441
3442static void bond_remove_proc_entry(struct bonding *bond)
3443{
3444 if (bond_proc_dir && bond->proc_entry) {
3445 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3446 memset(bond->proc_file_name, 0, IFNAMSIZ);
3447 bond->proc_entry = NULL;
3448 }
3449}
3450
3451/* Create the bonding directory under /proc/net, if doesn't exist yet.
3452 * Caller must hold rtnl_lock.
3453 */
3454static void bond_create_proc_dir(void)
3455{
3456 int len = strlen(DRV_NAME);
3457
457c4cbc 3458 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3459 bond_proc_dir = bond_proc_dir->next) {
3460 if ((bond_proc_dir->namelen == len) &&
3461 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3462 break;
3463 }
3464 }
3465
3466 if (!bond_proc_dir) {
457c4cbc 3467 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3468 if (bond_proc_dir) {
3469 bond_proc_dir->owner = THIS_MODULE;
3470 } else {
3471 printk(KERN_WARNING DRV_NAME
3472 ": Warning: cannot create /proc/net/%s\n",
3473 DRV_NAME);
3474 }
3475 }
3476}
3477
3478/* Destroy the bonding directory under /proc/net, if empty.
3479 * Caller must hold rtnl_lock.
3480 */
3481static void bond_destroy_proc_dir(void)
3482{
3483 struct proc_dir_entry *de;
3484
3485 if (!bond_proc_dir) {
3486 return;
3487 }
3488
3489 /* verify that the /proc dir is empty */
3490 for (de = bond_proc_dir->subdir; de; de = de->next) {
3491 /* ignore . and .. */
3492 if (*(de->name) != '.') {
3493 break;
3494 }
3495 }
3496
3497 if (de) {
3498 if (bond_proc_dir->owner == THIS_MODULE) {
3499 bond_proc_dir->owner = NULL;
3500 }
3501 } else {
457c4cbc 3502 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3503 bond_proc_dir = NULL;
3504 }
3505}
3506#endif /* CONFIG_PROC_FS */
3507
3508/*-------------------------- netdev event handling --------------------------*/
3509
3510/*
3511 * Change device name
3512 */
3513static int bond_event_changename(struct bonding *bond)
3514{
3515#ifdef CONFIG_PROC_FS
3516 bond_remove_proc_entry(bond);
3517 bond_create_proc_entry(bond);
3518#endif
b76cdba9
MW		 3519 down_write(&(bonding_rwsem));
3520 bond_destroy_sysfs_entry(bond);
3521 bond_create_sysfs_entry(bond);
3522 up_write(&(bonding_rwsem));
1da177e4
LT		 3523 return NOTIFY_DONE;
3524}
3525
3526static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3527{
454d7c9b 3528 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 3529
3530 switch (event) {
3531 case NETDEV_CHANGENAME:
3532 return bond_event_changename(event_bond);
3533 case NETDEV_UNREGISTER:
3b96c858 3534 bond_release_all(event_bond->dev);
1da177e4
LT		 3535 break;
3536 default:
3537 break;
3538 }
3539
3540 return NOTIFY_DONE;
3541}
3542
3543static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3544{
3545 struct net_device *bond_dev = slave_dev->master;
454d7c9b 3546 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3547
3548 switch (event) {
3549 case NETDEV_UNREGISTER:
3550 if (bond_dev) {
1284cd3a
JV		 3551 if (bond->setup_by_slave)
3552 bond_release_and_destroy(bond_dev, slave_dev);
3553 else
3554 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3555 }
3556 break;
3557 case NETDEV_CHANGE:
3558 /*
3559 * TODO: is this what we get if somebody
3560 * sets up a hierarchical bond, then rmmod's
3561 * one of the slave bonding devices?
3562 */
3563 break;
3564 case NETDEV_DOWN:
3565 /*
3566 * ... Or is it this?
3567 */
3568 break;
3569 case NETDEV_CHANGEMTU:
3570 /*
3571 * TODO: Should slaves be allowed to
3572 * independently alter their MTU? For
3573 * an active-backup bond, slaves need
3574 * not be the same type of device, so
3575 * MTUs may vary. For other modes,
3576 * slaves arguably should have the
3577 * same MTUs. To do this, we'd need to
3578 * take over the slave's change_mtu
3579 * function for the duration of their
3580 * servitude.
3581 */
3582 break;
3583 case NETDEV_CHANGENAME:
3584 /*
3585 * TODO: handle changing the primary's name
3586 */
3587 break;
8531c5ff
AK		 3588 case NETDEV_FEAT_CHANGE:
3589 bond_compute_features(bond);
3590 break;
1da177e4
LT		 3591 default:
3592 break;
3593 }
3594
3595 return NOTIFY_DONE;
3596}
3597
3598/*
3599 * bond_netdev_event: handle netdev notifier chain events.
3600 *
3601 * This function receives events for the netdev chain. The caller (an
e041c683 3602 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3603 * locks for us to safely manipulate the slave devices (RTNL lock,
3604 * dev_probe_lock).
3605 */
3606static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3607{
3608 struct net_device *event_dev = (struct net_device *)ptr;
3609
c346dca1 3610 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3611 return NOTIFY_DONE;
3612
1da177e4
LT		 3613 dprintk("event_dev: %s, event: %lx\n",
3614 (event_dev ? event_dev->name : "None"),
3615 event);
3616
0b680e75
JV		 3617 if (!(event_dev->priv_flags & IFF_BONDING))
3618 return NOTIFY_DONE;
3619
1da177e4
LT		 3620 if (event_dev->flags & IFF_MASTER) {
3621 dprintk("IFF_MASTER\n");
3622 return bond_master_netdev_event(event, event_dev);
3623 }
3624
3625 if (event_dev->flags & IFF_SLAVE) {
3626 dprintk("IFF_SLAVE\n");
3627 return bond_slave_netdev_event(event, event_dev);
3628 }
3629
3630 return NOTIFY_DONE;
3631}
3632
c3ade5ca
JV		 3633/*
3634 * bond_inetaddr_event: handle inetaddr notifier chain events.
3635 *
3636 * We keep track of device IPs primarily to use as source addresses in
3637 * ARP monitor probes (rather than spewing out broadcasts all the time).
3638 *
3639 * We track one IP for the main device (if it has one), plus one per VLAN.
3640 */
3641static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3642{
3643 struct in_ifaddr *ifa = ptr;
3644 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3645 struct bonding *bond;
3646 struct vlan_entry *vlan;
c3ade5ca 3647
c346dca1 3648 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3649 return NOTIFY_DONE;
3650
0883beca 3651 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3652 if (bond->dev == event_dev) {
3653 switch (event) {
3654 case NETDEV_UP:
3655 bond->master_ip = ifa->ifa_local;
3656 return NOTIFY_OK;
3657 case NETDEV_DOWN:
3658 bond->master_ip = bond_glean_dev_ip(bond->dev);
3659 return NOTIFY_OK;
3660 default:
3661 return NOTIFY_DONE;
3662 }
3663 }
3664
0883beca 3665 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3666 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3667 if (vlan_dev == event_dev) {
3668 switch (event) {
3669 case NETDEV_UP:
3670 vlan->vlan_ip = ifa->ifa_local;
3671 return NOTIFY_OK;
3672 case NETDEV_DOWN:
3673 vlan->vlan_ip =
3674 bond_glean_dev_ip(vlan_dev);
3675 return NOTIFY_OK;
3676 default:
3677 return NOTIFY_DONE;
3678 }
3679 }
3680 }
3681 }
3682 return NOTIFY_DONE;
3683}
3684
1da177e4
LT		 3685static struct notifier_block bond_netdev_notifier = {
3686 .notifier_call = bond_netdev_event,
3687};
3688
c3ade5ca
JV		 3689static struct notifier_block bond_inetaddr_notifier = {
3690 .notifier_call = bond_inetaddr_event,
3691};
3692
1da177e4
LT		 3693/*-------------------------- Packet type handling ---------------------------*/
3694
3695/* register to receive lacpdus on a bond */
3696static void bond_register_lacpdu(struct bonding *bond)
3697{
3698 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3699
3700 /* initialize packet type */
3701 pk_type->type = PKT_TYPE_LACPDU;
3702 pk_type->dev = bond->dev;
3703 pk_type->func = bond_3ad_lacpdu_recv;
3704
3705 dev_add_pack(pk_type);
3706}
3707
3708/* unregister to receive lacpdus on a bond */
3709static void bond_unregister_lacpdu(struct bonding *bond)
3710{
3711 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3712}
3713
f5b2b966
JV		 3714void bond_register_arp(struct bonding *bond)
3715{
3716 struct packet_type *pt = &bond->arp_mon_pt;
3717
c4f283b1
JV		 3718 if (pt->type)
3719 return;
3720
f5b2b966 3721 pt->type = htons(ETH_P_ARP);
e245cb71 3722 pt->dev = bond->dev;
f5b2b966
JV		 3723 pt->func = bond_arp_rcv;
3724 dev_add_pack(pt);
3725}
3726
3727void bond_unregister_arp(struct bonding *bond)
3728{
c4f283b1
JV		 3729 struct packet_type *pt = &bond->arp_mon_pt;
3730
3731 dev_remove_pack(pt);
3732 pt->type = 0;
f5b2b966
JV		 3733}
3734
169a3e66
JV		 3735/*---------------------------- Hashing Policies -----------------------------*/
3736
6f6652be
JV		 3737/*
3738 * Hash for the output device based upon layer 2 and layer 3 data. If
3739 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3740 */
3741static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3742 struct net_device *bond_dev, int count)
3743{
3744 struct ethhdr *data = (struct ethhdr *)skb->data;
3745 struct iphdr *iph = ip_hdr(skb);
3746
f14c4e4e 3747 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be
JV		 3748 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3749 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3750 }
3751
3752 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3753}
3754
169a3e66 3755/*
59c51591 3756 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3757 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3758 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3759 */
3760static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3761 struct net_device *bond_dev, int count)
3762{
3763 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3764 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3765 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3766 int layer4_xor = 0;
3767
f14c4e4e
BH		 3768 if (skb->protocol == htons(ETH_P_IP)) {
3769 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3770 (iph->protocol == IPPROTO_TCP ||
3771 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3772 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3773 }
3774 return (layer4_xor ^
3775 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3776
3777 }
3778
3779 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3780}
3781
3782/*
3783 * Hash for the output device based upon layer 2 data
3784 */
3785static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3786 struct net_device *bond_dev, int count)
3787{
3788 struct ethhdr *data = (struct ethhdr *)skb->data;
3789
3790 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3791}
3792
1da177e4
LT		 3793/*-------------------------- Device entry points ----------------------------*/
3794
3795static int bond_open(struct net_device *bond_dev)
3796{
454d7c9b 3797 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3798
3799 bond->kill_timers = 0;
3800
3801 if ((bond->params.mode == BOND_MODE_TLB) ||
3802 (bond->params.mode == BOND_MODE_ALB)) {
1da177e4
LT		 3803 /* bond_alb_initialize must be called before the timer
3804 * is started.
3805 */
3806 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3807 /* something went wrong - fail the open operation */
3808 return -1;
3809 }
3810
1b76b316
JV		 3811 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3812 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3813 }
3814
3815 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3816 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3817 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3818 }
3819
3820 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3821 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3822 INIT_DELAYED_WORK(&bond->arp_work,
3823 bond_activebackup_arp_mon);
3824 else
3825 INIT_DELAYED_WORK(&bond->arp_work,
3826 bond_loadbalance_arp_mon);
3827
3828 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3829 if (bond->params.arp_validate)
3830 bond_register_arp(bond);
1da177e4
LT		 3831 }
3832
3833 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3834 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3835 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3836 /* register to receive LACPDUs */
3837 bond_register_lacpdu(bond);
fd989c83 3838 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3839 }
3840
3841 return 0;
3842}
3843
3844static int bond_close(struct net_device *bond_dev)
3845{
454d7c9b 3846 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3847
3848 if (bond->params.mode == BOND_MODE_8023AD) {
3849 /* Unregister the receive of LACPDUs */
3850 bond_unregister_lacpdu(bond);
3851 }
3852
f5b2b966
JV		 3853 if (bond->params.arp_validate)
3854 bond_unregister_arp(bond);
3855
1da177e4
LT		 3856 write_lock_bh(&bond->lock);
3857
b59f9f74 3858 bond->send_grat_arp = 0;
305d552a 3859 bond->send_unsol_na = 0;
1da177e4
LT		 3860
3861 /* signal timers not to re-arm */
3862 bond->kill_timers = 1;
3863
3864 write_unlock_bh(&bond->lock);
3865
1da177e4 3866 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3867 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3868 }
3869
3870 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3871 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3872 }
3873
3874 switch (bond->params.mode) {
3875 case BOND_MODE_8023AD:
1b76b316 3876 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3877 break;
3878 case BOND_MODE_TLB:
3879 case BOND_MODE_ALB:
1b76b316 3880 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3881 break;
3882 default:
3883 break;
3884 }
3885
1da177e4
LT		 3886
3887 if ((bond->params.mode == BOND_MODE_TLB) ||
3888 (bond->params.mode == BOND_MODE_ALB)) {
3889 /* Must be called only after all
3890 * slaves have been released
3891 */
3892 bond_alb_deinitialize(bond);
3893 }
3894
3895 return 0;
3896}
3897
3898static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3899{
454d7c9b 3900 struct bonding *bond = netdev_priv(bond_dev);
eeda3fd6 3901 struct net_device_stats *stats = &bond->stats;
2439f9eb 3902 struct net_device_stats local_stats;
1da177e4
LT		 3903 struct slave *slave;
3904 int i;
3905
2439f9eb 3906 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3907
3908 read_lock_bh(&bond->lock);
3909
3910 bond_for_each_slave(bond, slave, i) {
eeda3fd6
SH		 3911 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3912
2439f9eb
AG		 3913 local_stats.rx_packets += sstats->rx_packets;
3914 local_stats.rx_bytes += sstats->rx_bytes;
3915 local_stats.rx_errors += sstats->rx_errors;
3916 local_stats.rx_dropped += sstats->rx_dropped;
3917
3918 local_stats.tx_packets += sstats->tx_packets;
3919 local_stats.tx_bytes += sstats->tx_bytes;
3920 local_stats.tx_errors += sstats->tx_errors;
3921 local_stats.tx_dropped += sstats->tx_dropped;
3922
3923 local_stats.multicast += sstats->multicast;
3924 local_stats.collisions += sstats->collisions;
3925
3926 local_stats.rx_length_errors += sstats->rx_length_errors;
3927 local_stats.rx_over_errors += sstats->rx_over_errors;
3928 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3929 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3930 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3931 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3932
3933 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3934 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3935 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3936 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3937 local_stats.tx_window_errors += sstats->tx_window_errors;
3938 }
3939
3940 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3941
3942 read_unlock_bh(&bond->lock);
3943
3944 return stats;
3945}
3946
3947static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3948{
3949 struct net_device *slave_dev = NULL;
3950 struct ifbond k_binfo;
3951 struct ifbond __user *u_binfo = NULL;
3952 struct ifslave k_sinfo;
3953 struct ifslave __user *u_sinfo = NULL;
3954 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3955 int res = 0;
3956
3957 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3958 bond_dev->name, cmd);
3959
3960 switch (cmd) {
1da177e4
LT		 3961 case SIOCGMIIPHY:
3962 mii = if_mii(ifr);
3963 if (!mii) {
3964 return -EINVAL;
3965 }
3966 mii->phy_id = 0;
3967 /* Fall Through */
3968 case SIOCGMIIREG:
3969 /*
3970 * We do this again just in case we were called by SIOCGMIIREG
3971 * instead of SIOCGMIIPHY.
3972 */
3973 mii = if_mii(ifr);
3974 if (!mii) {
3975 return -EINVAL;
3976 }
3977
3978 if (mii->reg_num == 1) {
454d7c9b 3979 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3980 mii->val_out = 0;
6603a6f2 3981 read_lock(&bond->lock);
1da177e4 3982 read_lock(&bond->curr_slave_lock);
4e140079 3983 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3984 mii->val_out = BMSR_LSTATUS;
3985 }
3986 read_unlock(&bond->curr_slave_lock);
6603a6f2 3987 read_unlock(&bond->lock);
1da177e4
LT		 3988 }
3989
3990 return 0;
3991 case BOND_INFO_QUERY_OLD:
3992 case SIOCBONDINFOQUERY:
3993 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3994
3995 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3996 return -EFAULT;
3997 }
3998
3999 res = bond_info_query(bond_dev, &k_binfo);
4000 if (res == 0) {
4001 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
4002 return -EFAULT;
4003 }
4004 }
4005
4006 return res;
4007 case BOND_SLAVE_INFO_QUERY_OLD:
4008 case SIOCBONDSLAVEINFOQUERY:
4009 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4010
4011 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4012 return -EFAULT;
4013 }
4014
4015 res = bond_slave_info_query(bond_dev, &k_sinfo);
4016 if (res == 0) {
4017 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4018 return -EFAULT;
4019 }
4020 }
4021
4022 return res;
4023 default:
4024 /* Go on */
4025 break;
4026 }
4027
4028 if (!capable(CAP_NET_ADMIN)) {
4029 return -EPERM;
4030 }
4031
b76cdba9 4032 down_write(&(bonding_rwsem));
881d966b 4033 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT		 4034
4035 dprintk("slave_dev=%p: \n", slave_dev);
4036
4037 if (!slave_dev) {
4038 res = -ENODEV;
4039 } else {
4040 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4041 switch (cmd) {
4042 case BOND_ENSLAVE_OLD:
4043 case SIOCBONDENSLAVE:
4044 res = bond_enslave(bond_dev, slave_dev);
4045 break;
4046 case BOND_RELEASE_OLD:
4047 case SIOCBONDRELEASE:
4048 res = bond_release(bond_dev, slave_dev);
4049 break;
4050 case BOND_SETHWADDR_OLD:
4051 case SIOCBONDSETHWADDR:
4052 res = bond_sethwaddr(bond_dev, slave_dev);
4053 break;
4054 case BOND_CHANGE_ACTIVE_OLD:
4055 case SIOCBONDCHANGEACTIVE:
4056 res = bond_ioctl_change_active(bond_dev, slave_dev);
4057 break;
4058 default:
4059 res = -EOPNOTSUPP;
4060 }
4061
4062 dev_put(slave_dev);
4063 }
4064
b76cdba9 4065 up_write(&(bonding_rwsem));
1da177e4
LT		 4066 return res;
4067}
4068
4069static void bond_set_multicast_list(struct net_device *bond_dev)
4070{
454d7c9b 4071 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4072 struct dev_mc_list *dmi;
4073
1da177e4
LT		 4074 /*
4075 * Do promisc before checking multicast_mode
4076 */
4077 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4078 /*
4079 * FIXME: Need to handle the error when one of the multi-slaves
4080 * encounters error.
4081 */
1da177e4
LT		 4082 bond_set_promiscuity(bond, 1);
4083 }
4084
4085 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4086 bond_set_promiscuity(bond, -1);
4087 }
4088
4089 /* set allmulti flag to slaves */
4090 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4091 /*
4092 * FIXME: Need to handle the error when one of the multi-slaves
4093 * encounters error.
4094 */
1da177e4
LT		 4095 bond_set_allmulti(bond, 1);
4096 }
4097
4098 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4099 bond_set_allmulti(bond, -1);
4100 }
4101
80ee5ad2
JV		 4102 read_lock(&bond->lock);
4103
1da177e4
LT		 4104 bond->flags = bond_dev->flags;
4105
4106 /* looking for addresses to add to slaves' mc list */
4107 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4108 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4109 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4110 }
4111 }
4112
4113 /* looking for addresses to delete from slaves' list */
4114 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4115 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4116 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4117 }
4118 }
4119
4120 /* save master's multicast list */
4121 bond_mc_list_destroy(bond);
4122 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4123
80ee5ad2 4124 read_unlock(&bond->lock);
1da177e4
LT		 4125}
4126
4127/*
4128 * Change the MTU of all of a master's slaves to match the master
4129 */
4130static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4131{
454d7c9b 4132 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4133 struct slave *slave, *stop_at;
4134 int res = 0;
4135 int i;
4136
4137 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4138 (bond_dev ? bond_dev->name : "None"), new_mtu);
4139
4140 /* Can't hold bond->lock with bh disabled here since
4141 * some base drivers panic. On the other hand we can't
4142 * hold bond->lock without bh disabled because we'll
4143 * deadlock. The only solution is to rely on the fact
4144 * that we're under rtnl_lock here, and the slaves
4145 * list won't change. This doesn't solve the problem
4146 * of setting the slave's MTU while it is
4147 * transmitting, but the assumption is that the base
4148 * driver can handle that.
4149 *
4150 * TODO: figure out a way to safely iterate the slaves
4151 * list, but without holding a lock around the actual
4152 * call to the base driver.
4153 */
4154
4155 bond_for_each_slave(bond, slave, i) {
4156 dprintk("s %p s->p %p c_m %p\n", slave,
4157 slave->prev, slave->dev->change_mtu);
e944ef79 4158
1da177e4
LT		 4159 res = dev_set_mtu(slave->dev, new_mtu);
4160
4161 if (res) {
4162 /* If we failed to set the slave's mtu to the new value
4163 * we must abort the operation even in ACTIVE_BACKUP
4164 * mode, because if we allow the backup slaves to have
4165 * different mtu values than the active slave we'll
4166 * need to change their mtu when doing a failover. That
4167 * means changing their mtu from timer context, which
4168 * is probably not a good idea.
4169 */
4170 dprintk("err %d %s\n", res, slave->dev->name);
4171 goto unwind;
4172 }
4173 }
4174
4175 bond_dev->mtu = new_mtu;
4176
4177 return 0;
4178
4179unwind:
4180 /* unwind from head to the slave that failed */
4181 stop_at = slave;
4182 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4183 int tmp_res;
4184
4185 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4186 if (tmp_res) {
4187 dprintk("unwind err %d dev %s\n", tmp_res,
4188 slave->dev->name);
4189 }
4190 }
4191
4192 return res;
4193}
4194
4195/*
4196 * Change HW address
4197 *
4198 * Note that many devices must be down to change the HW address, and
4199 * downing the master releases all slaves. We can make bonds full of
4200 * bonding devices to test this, however.
4201 */
4202static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4203{
454d7c9b 4204 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4205 struct sockaddr *sa = addr, tmp_sa;
4206 struct slave *slave, *stop_at;
4207 int res = 0;
4208 int i;
4209
eb7cc59a
SH		 4210 if (bond->params.mode == BOND_MODE_ALB)
4211 return bond_alb_set_mac_address(bond_dev, addr);
4212
4213
1da177e4
LT		 4214 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4215
dd957c57 4216 /*
3915c1e8
JV		 4217 * If fail_over_mac is set to active, do nothing and return
4218 * success. Returning an error causes ifenslave to fail.
dd957c57 4219 */
3915c1e8 4220 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4221 return 0;
2ab82852 4222
1da177e4
LT		 4223 if (!is_valid_ether_addr(sa->sa_data)) {
4224 return -EADDRNOTAVAIL;
4225 }
4226
4227 /* Can't hold bond->lock with bh disabled here since
4228 * some base drivers panic. On the other hand we can't
4229 * hold bond->lock without bh disabled because we'll
4230 * deadlock. The only solution is to rely on the fact
4231 * that we're under rtnl_lock here, and the slaves
4232 * list won't change. This doesn't solve the problem
4233 * of setting the slave's hw address while it is
4234 * transmitting, but the assumption is that the base
4235 * driver can handle that.
4236 *
4237 * TODO: figure out a way to safely iterate the slaves
4238 * list, but without holding a lock around the actual
4239 * call to the base driver.
4240 */
4241
4242 bond_for_each_slave(bond, slave, i) {
eb7cc59a 4243 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
1da177e4
LT		 4244 dprintk("slave %p %s\n", slave, slave->dev->name);
4245
eb7cc59a 4246 if (slave_ops->ndo_set_mac_address == NULL) {
1da177e4
LT		 4247 res = -EOPNOTSUPP;
4248 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4249 goto unwind;
4250 }
4251
4252 res = dev_set_mac_address(slave->dev, addr);
4253 if (res) {
4254 /* TODO: consider downing the slave
4255 * and retry ?
4256 * User should expect communications
4257 * breakage anyway until ARP finish
4258 * updating, so...
4259 */
4260 dprintk("err %d %s\n", res, slave->dev->name);
4261 goto unwind;
4262 }
4263 }
4264
4265 /* success */
4266 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4267 return 0;
4268
4269unwind:
4270 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4271 tmp_sa.sa_family = bond_dev->type;
4272
4273 /* unwind from head to the slave that failed */
4274 stop_at = slave;
4275 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4276 int tmp_res;
4277
4278 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4279 if (tmp_res) {
4280 dprintk("unwind err %d dev %s\n", tmp_res,
4281 slave->dev->name);
4282 }
4283 }
4284
4285 return res;
4286}
4287
4288static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4289{
454d7c9b 4290 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4291 struct slave *slave, *start_at;
cf5f9044 4292 int i, slave_no, res = 1;
1da177e4
LT		 4293
4294 read_lock(&bond->lock);
4295
4296 if (!BOND_IS_OK(bond)) {
4297 goto out;
4298 }
4299
cf5f9044
JV		 4300 /*
4301 * Concurrent TX may collide on rr_tx_counter; we accept that
4302 * as being rare enough not to justify using an atomic op here
4303 */
4304 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4305
cf5f9044
JV		 4306 bond_for_each_slave(bond, slave, i) {
4307 slave_no--;
4308 if (slave_no < 0) {
4309 break;
4310 }
1da177e4
LT		 4311 }
4312
cf5f9044 4313 start_at = slave;
1da177e4
LT		 4314 bond_for_each_slave_from(bond, slave, i, start_at) {
4315 if (IS_UP(slave->dev) &&
4316 (slave->link == BOND_LINK_UP) &&
4317 (slave->state == BOND_STATE_ACTIVE)) {
4318 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4319 break;
4320 }
4321 }
4322
1da177e4
LT		 4323out:
4324 if (res) {
4325 /* no suitable interface, frame not sent */
4326 dev_kfree_skb(skb);
4327 }
4328 read_unlock(&bond->lock);
4329 return 0;
4330}
4331
075897ce 4332
1da177e4
LT		 4333/*
4334 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4335 * the bond has a usable interface.
4336 */
4337static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4338{
454d7c9b 4339 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4340 int res = 1;
4341
1da177e4
LT		 4342 read_lock(&bond->lock);
4343 read_lock(&bond->curr_slave_lock);
4344
4345 if (!BOND_IS_OK(bond)) {
4346 goto out;
4347 }
4348
075897ce
JL		 4349 if (!bond->curr_active_slave)
4350 goto out;
4351
075897ce
JL		 4352 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4353
1da177e4
LT		 4354out:
4355 if (res) {
4356 /* no suitable interface, frame not sent */
4357 dev_kfree_skb(skb);
4358 }
4359 read_unlock(&bond->curr_slave_lock);
4360 read_unlock(&bond->lock);
4361 return 0;
4362}
4363
4364/*
169a3e66
JV		 4365 * In bond_xmit_xor() , we determine the output device by using a pre-
4366 * determined xmit_hash_policy(), If the selected device is not enabled,
4367 * find the next active slave.
1da177e4
LT		 4368 */
4369static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4370{
454d7c9b 4371 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4372 struct slave *slave, *start_at;
4373 int slave_no;
4374 int i;
4375 int res = 1;
4376
4377 read_lock(&bond->lock);
4378
4379 if (!BOND_IS_OK(bond)) {
4380 goto out;
4381 }
4382
169a3e66 4383 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4384
4385 bond_for_each_slave(bond, slave, i) {
4386 slave_no--;
4387 if (slave_no < 0) {
4388 break;
4389 }
4390 }
4391
4392 start_at = slave;
4393
4394 bond_for_each_slave_from(bond, slave, i, start_at) {
4395 if (IS_UP(slave->dev) &&
4396 (slave->link == BOND_LINK_UP) &&
4397 (slave->state == BOND_STATE_ACTIVE)) {
4398 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4399 break;
4400 }
4401 }
4402
4403out:
4404 if (res) {
4405 /* no suitable interface, frame not sent */
4406 dev_kfree_skb(skb);
4407 }
4408 read_unlock(&bond->lock);
4409 return 0;
4410}
4411
4412/*
4413 * in broadcast mode, we send everything to all usable interfaces.
4414 */
4415static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4416{
454d7c9b 4417 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4418 struct slave *slave, *start_at;
4419 struct net_device *tx_dev = NULL;
4420 int i;
4421 int res = 1;
4422
4423 read_lock(&bond->lock);
4424
4425 if (!BOND_IS_OK(bond)) {
4426 goto out;
4427 }
4428
4429 read_lock(&bond->curr_slave_lock);
4430 start_at = bond->curr_active_slave;
4431 read_unlock(&bond->curr_slave_lock);
4432
4433 if (!start_at) {
4434 goto out;
4435 }
4436
4437 bond_for_each_slave_from(bond, slave, i, start_at) {
4438 if (IS_UP(slave->dev) &&
4439 (slave->link == BOND_LINK_UP) &&
4440 (slave->state == BOND_STATE_ACTIVE)) {
4441 if (tx_dev) {
4442 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4443 if (!skb2) {
4444 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4445 ": %s: Error: bond_xmit_broadcast(): "
4446 "skb_clone() failed\n",
4447 bond_dev->name);
1da177e4
LT		 4448 continue;
4449 }
4450
4451 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4452 if (res) {
4453 dev_kfree_skb(skb2);
4454 continue;
4455 }
4456 }
4457 tx_dev = slave->dev;
4458 }
4459 }
4460
4461 if (tx_dev) {
4462 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4463 }
4464
4465out:
4466 if (res) {
4467 /* no suitable interface, frame not sent */
4468 dev_kfree_skb(skb);
4469 }
4470 /* frame sent to all suitable interfaces */
4471 read_unlock(&bond->lock);
4472 return 0;
4473}
4474
4475/*------------------------- Device initialization ---------------------------*/
4476
6f6652be
JV		 4477static void bond_set_xmit_hash_policy(struct bonding *bond)
4478{
4479 switch (bond->params.xmit_policy) {
4480 case BOND_XMIT_POLICY_LAYER23:
4481 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4482 break;
4483 case BOND_XMIT_POLICY_LAYER34:
4484 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4485 break;
4486 case BOND_XMIT_POLICY_LAYER2:
4487 default:
4488 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4489 break;
4490 }
4491}
4492
1da177e4
LT		 4493/*
4494 * set bond mode specific net device operations
4495 */
a77b5325 4496void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4497{
169a3e66
JV		 4498 struct net_device *bond_dev = bond->dev;
4499
1da177e4
LT		 4500 switch (mode) {
4501 case BOND_MODE_ROUNDROBIN:
4502 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4503 break;
4504 case BOND_MODE_ACTIVEBACKUP:
4505 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4506 break;
4507 case BOND_MODE_XOR:
4508 bond_dev->hard_start_xmit = bond_xmit_xor;
6f6652be 4509 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4510 break;
4511 case BOND_MODE_BROADCAST:
4512 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4513 break;
4514 case BOND_MODE_8023AD:
8f903c70 4515 bond_set_master_3ad_flags(bond);
1da177e4 4516 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
6f6652be 4517 bond_set_xmit_hash_policy(bond);
1da177e4 4518 break;
1da177e4 4519 case BOND_MODE_ALB:
8f903c70
JV		 4520 bond_set_master_alb_flags(bond);
4521 /* FALLTHRU */
4522 case BOND_MODE_TLB:
1da177e4 4523 bond_dev->hard_start_xmit = bond_alb_xmit;
1da177e4
LT		 4524 break;
4525 default:
4526 /* Should never happen, mode already checked */
4527 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4528 ": %s: Error: Unknown bonding mode %d\n",
4529 bond_dev->name,
1da177e4
LT		 4530 mode);
4531 break;
4532 }
4533}
4534
217df670
JV		 4535static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4536 struct ethtool_drvinfo *drvinfo)
4537{
4538 strncpy(drvinfo->driver, DRV_NAME, 32);
4539 strncpy(drvinfo->version, DRV_VERSION, 32);
4540 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4541}
4542
7282d491 4543static const struct ethtool_ops bond_ethtool_ops = {
217df670 4544 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac
SH		 4545 .get_link = ethtool_op_get_link,
4546 .get_tx_csum = ethtool_op_get_tx_csum,
4547 .get_sg = ethtool_op_get_sg,
4548 .get_tso = ethtool_op_get_tso,
4549 .get_ufo = ethtool_op_get_ufo,
4550 .get_flags = ethtool_op_get_flags,
8531c5ff
AK		 4551};
4552
eb7cc59a
SH		 4553static const struct net_device_ops bond_netdev_ops = {
4554 .ndo_open = bond_open,
4555 .ndo_stop = bond_close,
4556 .ndo_get_stats = bond_get_stats,
4557 .ndo_do_ioctl = bond_do_ioctl,
4558 .ndo_set_multicast_list = bond_set_multicast_list,
4559 .ndo_change_mtu = bond_change_mtu,
4560 .ndo_validate_addr = NULL,
4561 .ndo_set_mac_address = bond_set_mac_address,
4562 .ndo_vlan_rx_register = bond_vlan_rx_register,
4563 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4564 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4565};
4566
1da177e4
LT		 4567/*
4568 * Does not allocate but creates a /proc entry.
4569 * Allowed to fail.
4570 */
3c535952 4571static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4 4572{
454d7c9b 4573 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4574
4575 dprintk("Begin bond_init for %s\n", bond_dev->name);
4576
4577 /* initialize rwlocks */
4578 rwlock_init(&bond->lock);
4579 rwlock_init(&bond->curr_slave_lock);
4580
4581 bond->params = *params; /* copy params struct */
4582
1b76b316
JV		 4583 bond->wq = create_singlethread_workqueue(bond_dev->name);
4584 if (!bond->wq)
4585 return -ENOMEM;
4586
1da177e4
LT		 4587 /* Initialize pointers */
4588 bond->first_slave = NULL;
4589 bond->curr_active_slave = NULL;
4590 bond->current_arp_slave = NULL;
4591 bond->primary_slave = NULL;
4592 bond->dev = bond_dev;
1053f62c 4593 bond->send_grat_arp = 0;
305d552a 4594 bond->send_unsol_na = 0;
d90a162a 4595 bond->setup_by_slave = 0;
1da177e4
LT		 4596 INIT_LIST_HEAD(&bond->vlan_list);
4597
4598 /* Initialize the device entry points */
eb7cc59a 4599 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 4600 bond_dev->ethtool_ops = &bond_ethtool_ops;
169a3e66 4601 bond_set_mode_ops(bond, bond->params.mode);
1da177e4 4602
a434e43f 4603 bond_dev->destructor = bond_destructor;
1da177e4
LT		 4604
4605 /* Initialize the device options */
4606 bond_dev->tx_queue_len = 0;
4607 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4608 bond_dev->priv_flags |= IFF_BONDING;
6cf3f41e
JV		 4609 if (bond->params.arp_interval)
4610 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
1da177e4
LT		 4611
4612 /* At first, we block adding VLANs. That's the only way to
4613 * prevent problems that occur when adding VLANs over an
4614 * empty bond. The block will be removed once non-challenged
4615 * slaves are enslaved.
4616 */
4617 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4618
932ff279 4619 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4620 * transmitting */
4621 bond_dev->features |= NETIF_F_LLTX;
4622
4623 /* By default, we declare the bond to be fully
4624 * VLAN hardware accelerated capable. Special
4625 * care is taken in the various xmit functions
4626 * when there are slaves that are not hw accel
4627 * capable
4628 */
1da177e4
LT		 4629 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4630 NETIF_F_HW_VLAN_RX |
4631 NETIF_F_HW_VLAN_FILTER);
4632
4633#ifdef CONFIG_PROC_FS
4634 bond_create_proc_entry(bond);
4635#endif
1da177e4
LT		 4636 list_add_tail(&bond->bond_list, &bond_dev_list);
4637
4638 return 0;
4639}
4640
fdaea7a9
JV		 4641static void bond_work_cancel_all(struct bonding *bond)
4642{
4643 write_lock_bh(&bond->lock);
4644 bond->kill_timers = 1;
4645 write_unlock_bh(&bond->lock);
4646
4647 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4648 cancel_delayed_work(&bond->mii_work);
4649
4650 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4651 cancel_delayed_work(&bond->arp_work);
4652
4653 if (bond->params.mode == BOND_MODE_ALB &&
4654 delayed_work_pending(&bond->alb_work))
4655 cancel_delayed_work(&bond->alb_work);
4656
4657 if (bond->params.mode == BOND_MODE_8023AD &&
4658 delayed_work_pending(&bond->ad_work))
4659 cancel_delayed_work(&bond->ad_work);
4660}
4661
a434e43f
JV		 4662/* De-initialize device specific data.
4663 * Caller must hold rtnl_lock.
4664 */
4665static void bond_deinit(struct net_device *bond_dev)
4666{
454d7c9b 4667 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 4668
4669 list_del(&bond->bond_list);
4670
4671 bond_work_cancel_all(bond);
4672
4673#ifdef CONFIG_PROC_FS
4674 bond_remove_proc_entry(bond);
4675#endif
4676}
4677
1da177e4
LT		 4678/* Unregister and free all bond devices.
4679 * Caller must hold rtnl_lock.
4680 */
4681static void bond_free_all(void)
4682{
4683 struct bonding *bond, *nxt;
4684
4685 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4686 struct net_device *bond_dev = bond->dev;
4687
fdaea7a9 4688 bond_work_cancel_all(bond);
70298705 4689 /* Release the bonded slaves */
4690 bond_release_all(bond_dev);
92b41daa 4691 bond_destroy(bond);
1da177e4
LT		 4692 }
4693
4694#ifdef CONFIG_PROC_FS
4695 bond_destroy_proc_dir();
4696#endif
4697}
4698
4699/*------------------------- Module initialization ---------------------------*/
4700
4701/*
4702 * Convert string input module parms. Accept either the
ece95f7f
JV		 4703 * number of the mode or its string name. A bit complicated because
4704 * some mode names are substrings of other names, and calls from sysfs
4705 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4706 */
ece95f7f 4707int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4708{
ece95f7f 4709 int mode = -1, i, rv;
a42e534f 4710 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4711
a42e534f
JV		 4712 for (p = (char *)buf; *p; p++)
4713 if (!(isdigit(*p) || isspace(*p)))
4714 break;
4715
4716 if (*p)
ece95f7f 4717 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4718 else
4719 rv = sscanf(buf, "%d", &mode);
4720
4721 if (!rv)
4722 return -1;
1da177e4
LT		 4723
4724 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4725 if (mode == tbl[i].mode)
4726 return tbl[i].mode;
4727 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4728 return tbl[i].mode;
1da177e4
LT		 4729 }
4730
4731 return -1;
4732}
4733
4734static int bond_check_params(struct bond_params *params)
4735{
3915c1e8 4736 int arp_validate_value, fail_over_mac_value;
f5b2b966 4737
1da177e4
LT		 4738 /*
4739 * Convert string parameters.
4740 */
4741 if (mode) {
4742 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4743 if (bond_mode == -1) {
4744 printk(KERN_ERR DRV_NAME
4745 ": Error: Invalid bonding mode \"%s\"\n",
4746 mode == NULL ? "NULL" : mode);
4747 return -EINVAL;
4748 }
4749 }
4750
169a3e66
JV		 4751 if (xmit_hash_policy) {
4752 if ((bond_mode != BOND_MODE_XOR) &&
4753 (bond_mode != BOND_MODE_8023AD)) {
4754 printk(KERN_INFO DRV_NAME
4755 ": xor_mode param is irrelevant in mode %s\n",
4756 bond_mode_name(bond_mode));
4757 } else {
4758 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4759 xmit_hashtype_tbl);
4760 if (xmit_hashtype == -1) {
4761 printk(KERN_ERR DRV_NAME
4762 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4763 xmit_hash_policy == NULL ? "NULL" :
4764 xmit_hash_policy);
4765 return -EINVAL;
4766 }
4767 }
4768 }
4769
1da177e4
LT		 4770 if (lacp_rate) {
4771 if (bond_mode != BOND_MODE_8023AD) {
4772 printk(KERN_INFO DRV_NAME
4773 ": lacp_rate param is irrelevant in mode %s\n",
4774 bond_mode_name(bond_mode));
4775 } else {
4776 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4777 if (lacp_fast == -1) {
4778 printk(KERN_ERR DRV_NAME
4779 ": Error: Invalid lacp rate \"%s\"\n",
4780 lacp_rate == NULL ? "NULL" : lacp_rate);
4781 return -EINVAL;
4782 }
4783 }
4784 }
4785
fd989c83
JV		 4786 if (ad_select) {
4787 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4788 if (params->ad_select == -1) {
4789 printk(KERN_ERR DRV_NAME
4790 ": Error: Invalid ad_select \"%s\"\n",
4791 ad_select == NULL ? "NULL" : ad_select);
4792 return -EINVAL;
4793 }
4794
4795 if (bond_mode != BOND_MODE_8023AD) {
4796 printk(KERN_WARNING DRV_NAME
4797 ": ad_select param only affects 802.3ad mode\n");
4798 }
4799 } else {
4800 params->ad_select = BOND_AD_STABLE;
4801 }
4802
b8a9787e 4803 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4804 printk(KERN_WARNING DRV_NAME
4805 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4806 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4807 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4808 max_bonds = BOND_DEFAULT_MAX_BONDS;
4809 }
4810
4811 if (miimon < 0) {
4812 printk(KERN_WARNING DRV_NAME
4813 ": Warning: miimon module parameter (%d), "
4814 "not in range 0-%d, so it was reset to %d\n",
4815 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4816 miimon = BOND_LINK_MON_INTERV;
4817 }
4818
4819 if (updelay < 0) {
4820 printk(KERN_WARNING DRV_NAME
4821 ": Warning: updelay module parameter (%d), "
4822 "not in range 0-%d, so it was reset to 0\n",
4823 updelay, INT_MAX);
4824 updelay = 0;
4825 }
4826
4827 if (downdelay < 0) {
4828 printk(KERN_WARNING DRV_NAME
4829 ": Warning: downdelay module parameter (%d), "
4830 "not in range 0-%d, so it was reset to 0\n",
4831 downdelay, INT_MAX);
4832 downdelay = 0;
4833 }
4834
4835 if ((use_carrier != 0) && (use_carrier != 1)) {
4836 printk(KERN_WARNING DRV_NAME
4837 ": Warning: use_carrier module parameter (%d), "
4838 "not of valid value (0/1), so it was set to 1\n",
4839 use_carrier);
4840 use_carrier = 1;
4841 }
4842
7893b249
MS		 4843 if (num_grat_arp < 0 || num_grat_arp > 255) {
4844 printk(KERN_WARNING DRV_NAME
4845 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4846 "was reset to 1 \n", num_grat_arp);
4847 num_grat_arp = 1;
4848 }
4849
305d552a
BH		 4850 if (num_unsol_na < 0 || num_unsol_na > 255) {
4851 printk(KERN_WARNING DRV_NAME
4852 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4853 "was reset to 1 \n", num_unsol_na);
4854 num_unsol_na = 1;
4855 }
4856
1da177e4
LT		 4857 /* reset values for 802.3ad */
4858 if (bond_mode == BOND_MODE_8023AD) {
4859 if (!miimon) {
4860 printk(KERN_WARNING DRV_NAME
4861 ": Warning: miimon must be specified, "
4862 "otherwise bonding will not detect link "
4863 "failure, speed and duplex which are "
4864 "essential for 802.3ad operation\n");
4865 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4866 miimon = 100;
4867 }
4868 }
4869
4870 /* reset values for TLB/ALB */
4871 if ((bond_mode == BOND_MODE_TLB) ||
4872 (bond_mode == BOND_MODE_ALB)) {
4873 if (!miimon) {
4874 printk(KERN_WARNING DRV_NAME
4875 ": Warning: miimon must be specified, "
4876 "otherwise bonding will not detect link "
4877 "failure and link speed which are essential "
4878 "for TLB/ALB load balancing\n");
4879 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4880 miimon = 100;
4881 }
4882 }
4883
4884 if (bond_mode == BOND_MODE_ALB) {
4885 printk(KERN_NOTICE DRV_NAME
4886 ": In ALB mode you might experience client "
4887 "disconnections upon reconnection of a link if the "
4888 "bonding module updelay parameter (%d msec) is "
4889 "incompatible with the forwarding delay time of the "
4890 "switch\n",
4891 updelay);
4892 }
4893
4894 if (!miimon) {
4895 if (updelay || downdelay) {
4896 /* just warn the user the up/down delay will have
4897 * no effect since miimon is zero...
4898 */
4899 printk(KERN_WARNING DRV_NAME
4900 ": Warning: miimon module parameter not set "
4901 "and updelay (%d) or downdelay (%d) module "
4902 "parameter is set; updelay and downdelay have "
4903 "no effect unless miimon is set\n",
4904 updelay, downdelay);
4905 }
4906 } else {
4907 /* don't allow arp monitoring */
4908 if (arp_interval) {
4909 printk(KERN_WARNING DRV_NAME
4910 ": Warning: miimon (%d) and arp_interval (%d) "
4911 "can't be used simultaneously, disabling ARP "
4912 "monitoring\n",
4913 miimon, arp_interval);
4914 arp_interval = 0;
4915 }
4916
4917 if ((updelay % miimon) != 0) {
4918 printk(KERN_WARNING DRV_NAME
4919 ": Warning: updelay (%d) is not a multiple "
4920 "of miimon (%d), updelay rounded to %d ms\n",
4921 updelay, miimon, (updelay / miimon) * miimon);
4922 }
4923
4924 updelay /= miimon;
4925
4926 if ((downdelay % miimon) != 0) {
4927 printk(KERN_WARNING DRV_NAME
4928 ": Warning: downdelay (%d) is not a multiple "
4929 "of miimon (%d), downdelay rounded to %d ms\n",
4930 downdelay, miimon,
4931 (downdelay / miimon) * miimon);
4932 }
4933
4934 downdelay /= miimon;
4935 }
4936
4937 if (arp_interval < 0) {
4938 printk(KERN_WARNING DRV_NAME
4939 ": Warning: arp_interval module parameter (%d) "
4940 ", not in range 0-%d, so it was reset to %d\n",
4941 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4942 arp_interval = BOND_LINK_ARP_INTERV;
4943 }
4944
4945 for (arp_ip_count = 0;
4946 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4947 arp_ip_count++) {
4948 /* not complete check, but should be good enough to
4949 catch mistakes */
4950 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4951 printk(KERN_WARNING DRV_NAME
4952 ": Warning: bad arp_ip_target module parameter "
4953 "(%s), ARP monitoring will not be performed\n",
4954 arp_ip_target[arp_ip_count]);
4955 arp_interval = 0;
4956 } else {
d3bb52b0 4957 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4958 arp_target[arp_ip_count] = ip;
4959 }
4960 }
4961
4962 if (arp_interval && !arp_ip_count) {
4963 /* don't allow arping if no arp_ip_target given... */
4964 printk(KERN_WARNING DRV_NAME
4965 ": Warning: arp_interval module parameter (%d) "
4966 "specified without providing an arp_ip_target "
4967 "parameter, arp_interval was reset to 0\n",
4968 arp_interval);
4969 arp_interval = 0;
4970 }
4971
f5b2b966
JV		 4972 if (arp_validate) {
4973 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4974 printk(KERN_ERR DRV_NAME
4975 ": arp_validate only supported in active-backup mode\n");
4976 return -EINVAL;
4977 }
4978 if (!arp_interval) {
4979 printk(KERN_ERR DRV_NAME
4980 ": arp_validate requires arp_interval\n");
4981 return -EINVAL;
4982 }
4983
4984 arp_validate_value = bond_parse_parm(arp_validate,
4985 arp_validate_tbl);
4986 if (arp_validate_value == -1) {
4987 printk(KERN_ERR DRV_NAME
4988 ": Error: invalid arp_validate \"%s\"\n",
4989 arp_validate == NULL ? "NULL" : arp_validate);
4990 return -EINVAL;
4991 }
4992 } else
4993 arp_validate_value = 0;
4994
1da177e4
LT		 4995 if (miimon) {
4996 printk(KERN_INFO DRV_NAME
4997 ": MII link monitoring set to %d ms\n",
4998 miimon);
4999 } else if (arp_interval) {
5000 int i;
5001
5002 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 5003 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5004 arp_interval,
5005 arp_validate_tbl[arp_validate_value].modename,
5006 arp_ip_count);
1da177e4
LT		 5007
5008 for (i = 0; i < arp_ip_count; i++)
5009 printk (" %s", arp_ip_target[i]);
5010
5011 printk("\n");
5012
b8a9787e 5013 } else if (max_bonds) {
1da177e4
LT		 5014 /* miimon and arp_interval not set, we need one so things
5015 * work as expected, see bonding.txt for details
5016 */
5017 printk(KERN_WARNING DRV_NAME
5018 ": Warning: either miimon or arp_interval and "
5019 "arp_ip_target module parameters must be specified, "
5020 "otherwise bonding will not detect link failures! see "
5021 "bonding.txt for details.\n");
5022 }
5023
5024 if (primary && !USES_PRIMARY(bond_mode)) {
5025 /* currently, using a primary only makes sense
5026 * in active backup, TLB or ALB modes
5027 */
5028 printk(KERN_WARNING DRV_NAME
5029 ": Warning: %s primary device specified but has no "
5030 "effect in %s mode\n",
5031 primary, bond_mode_name(bond_mode));
5032 primary = NULL;
5033 }
5034
3915c1e8
JV		 5035 if (fail_over_mac) {
5036 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5037 fail_over_mac_tbl);
5038 if (fail_over_mac_value == -1) {
5039 printk(KERN_ERR DRV_NAME
5040 ": Error: invalid fail_over_mac \"%s\"\n",
5041 arp_validate == NULL ? "NULL" : arp_validate);
5042 return -EINVAL;
5043 }
5044
5045 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5046 printk(KERN_WARNING DRV_NAME
5047 ": Warning: fail_over_mac only affects "
5048 "active-backup mode.\n");
5049 } else {
5050 fail_over_mac_value = BOND_FOM_NONE;
5051 }
dd957c57 5052
1da177e4
LT		 5053 /* fill params struct with the proper values */
5054 params->mode = bond_mode;
169a3e66 5055 params->xmit_policy = xmit_hashtype;
1da177e4 5056 params->miimon = miimon;
7893b249 5057 params->num_grat_arp = num_grat_arp;
305d552a 5058 params->num_unsol_na = num_unsol_na;
1da177e4 5059 params->arp_interval = arp_interval;
f5b2b966 5060 params->arp_validate = arp_validate_value;
1da177e4
LT		 5061 params->updelay = updelay;
5062 params->downdelay = downdelay;
5063 params->use_carrier = use_carrier;
5064 params->lacp_fast = lacp_fast;
5065 params->primary[0] = 0;
3915c1e8 5066 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5067
5068 if (primary) {
5069 strncpy(params->primary, primary, IFNAMSIZ);
5070 params->primary[IFNAMSIZ - 1] = 0;
5071 }
5072
5073 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5074
5075 return 0;
5076}
5077
0daa2303 5078static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 5079static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 5080
e8a0464c
DM		 5081static void bond_set_lockdep_class_one(struct net_device *dev,
5082 struct netdev_queue *txq,
5083 void *_unused)
c773e847
DM		 5084{
5085 lockdep_set_class(&txq->_xmit_lock,
5086 &bonding_netdev_xmit_lock_key);
5087}
5088
5089static void bond_set_lockdep_class(struct net_device *dev)
5090{
cf508b12
DM		 5091 lockdep_set_class(&dev->addr_list_lock,
5092 &bonding_netdev_addr_lock_key);
e8a0464c 5093 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 5094}
5095
dfe60397 5096/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5097 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5098 * Caller must NOT hold rtnl_lock; we need to release it here before we
5099 * set up our sysfs entries.
5100 */
0dd646fe 5101int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5102{
5103 struct net_device *bond_dev;
0883beca 5104 struct bonding *bond;
dfe60397
MW		 5105 int res;
5106
5107 rtnl_lock();
027ea041
JV		 5108 down_write(&bonding_rwsem);
5109
5110 /* Check to see if the bond already exists. */
4fe4763c 5111 if (name) {
0883beca 5112 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5113 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5114 printk(KERN_ERR DRV_NAME
027ea041 5115 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5116 name);
5117 res = -EPERM;
5118 goto out_rtnl;
5119 }
5120 }
027ea041 5121
e4b91c48
JV		 5122 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5123 ether_setup);
dfe60397
MW		 5124 if (!bond_dev) {
5125 printk(KERN_ERR DRV_NAME
5126 ": %s: eek! can't alloc netdev!\n",
5127 name);
5128 res = -ENOMEM;
5129 goto out_rtnl;
5130 }
5131
e4b91c48
JV		 5132 if (!name) {
5133 res = dev_alloc_name(bond_dev, "bond%d");
5134 if (res < 0)
5135 goto out_netdev;
5136 }
5137
dfe60397
MW		 5138 /* bond_init() must be called after dev_alloc_name() (for the
5139 * /proc files), but before register_netdevice(), because we
5140 * need to set function pointers.
5141 */
5142
5143 res = bond_init(bond_dev, params);
5144 if (res < 0) {
5145 goto out_netdev;
5146 }
5147
dfe60397
MW		 5148 res = register_netdevice(bond_dev);
5149 if (res < 0) {
5150 goto out_bond;
5151 }
0daa2303 5152
c773e847 5153 bond_set_lockdep_class(bond_dev);
0daa2303 5154
ff59c456
JV		 5155 netif_carrier_off(bond_dev);
5156
027ea041 5157 up_write(&bonding_rwsem);
dfe60397 5158 rtnl_unlock(); /* allows sysfs registration of net device */
454d7c9b 5159 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
09c89279
JV		 5160 if (res < 0) {
5161 rtnl_lock();
027ea041 5162 down_write(&bonding_rwsem);
822973ba
PE		 5163 bond_deinit(bond_dev);
5164 unregister_netdevice(bond_dev);
5165 goto out_rtnl;
09c89279
JV		 5166 }
5167
5168 return 0;
5169
dfe60397
MW		 5170out_bond:
5171 bond_deinit(bond_dev);
5172out_netdev:
5173 free_netdev(bond_dev);
5174out_rtnl:
027ea041 5175 up_write(&bonding_rwsem);
dfe60397 5176 rtnl_unlock();
dfe60397
MW		 5177 return res;
5178}
5179
1da177e4
LT		 5180static int __init bonding_init(void)
5181{
1da177e4
LT		 5182 int i;
5183 int res;
0883beca 5184 struct bonding *bond;
1da177e4
LT		 5185
5186 printk(KERN_INFO "%s", version);
5187
dfe60397 5188 res = bond_check_params(&bonding_defaults);
1da177e4 5189 if (res) {
dfe60397 5190 goto out;
1da177e4
LT		 5191 }
5192
1da177e4
LT		 5193#ifdef CONFIG_PROC_FS
5194 bond_create_proc_dir();
5195#endif
027ea041
JV		 5196
5197 init_rwsem(&bonding_rwsem);
5198
1da177e4 5199 for (i = 0; i < max_bonds; i++) {
0dd646fe 5200 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5201 if (res)
5202 goto err;
1da177e4
LT		 5203 }
5204
b76cdba9
MW		 5205 res = bond_create_sysfs();
5206 if (res)
5207 goto err;
5208
1da177e4 5209 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5210 register_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5211 bond_register_ipv6_notifier();
1da177e4 5212
dfe60397
MW		 5213 goto out;
5214err:
0883beca 5215 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5216 bond_work_cancel_all(bond);
5217 destroy_workqueue(bond->wq);
5218 }
5219
ae68c398
PE		 5220 bond_destroy_sysfs();
5221
40abc270 5222 rtnl_lock();
1da177e4 5223 bond_free_all();
1da177e4 5224 rtnl_unlock();
dfe60397 5225out:
1da177e4 5226 return res;
dfe60397 5227
1da177e4
LT		 5228}
5229
5230static void __exit bonding_exit(void)
5231{
5232 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5233 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5234 bond_unregister_ipv6_notifier();
1da177e4 5235
ae68c398
PE		 5236 bond_destroy_sysfs();
5237
1da177e4
LT		 5238 rtnl_lock();
5239 bond_free_all();
5240 rtnl_unlock();
5241}
5242
5243module_init(bonding_init);
5244module_exit(bonding_exit);
5245MODULE_LICENSE("GPL");
5246MODULE_VERSION(DRV_VERSION);
5247MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5248MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5249MODULE_SUPPORTED_DEVICE("most ethernet devices");
5250
5251/*
5252 * Local variables:
5253 * c-indent-level: 8
5254 * c-basic-offset: 8
5255 * tab-width: 8
5256 * End:
5257 */
5258
Clone of davem's net-next-2.6 tree