]> 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
pkt_sched: sch_htb: Optimize WARN_ONs in htb_dequeue_tree() etc.
[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		 1380static void bond_setup_by_slave(struct net_device *bond_dev,
1381 struct net_device *slave_dev)
1382{
454d7c9b 1383 struct bonding *bond = netdev_priv(bond_dev);
d90a162a 1384
00829823 1385 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1386
1387 bond_dev->type = slave_dev->type;
1388 bond_dev->hard_header_len = slave_dev->hard_header_len;
1389 bond_dev->addr_len = slave_dev->addr_len;
1390
1391 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1392 slave_dev->addr_len);
d90a162a 1393 bond->setup_by_slave = 1;
872254dd
MS		 1394}
1395
1da177e4 1396/* enslave device <slave> to bond device <master> */
a77b5325 1397int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1398{
454d7c9b 1399 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1400 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 1401 struct slave *new_slave = NULL;
1402 struct dev_mc_list *dmi;
1403 struct sockaddr addr;
1404 int link_reporting;
1405 int old_features = bond_dev->features;
1406 int res = 0;
1407
552709d5 1408 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
eb7cc59a 1409 slave_ops->ndo_do_ioctl == NULL) {
1da177e4 1410 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1411 ": %s: Warning: no link monitoring support for %s\n",
1412 bond_dev->name, slave_dev->name);
1da177e4
LT		 1413 }
1414
1415 /* bond must be initialized by bond_open() before enslaving */
1416 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1417 printk(KERN_WARNING DRV_NAME
1418 " %s: master_dev is not up in bond_enslave\n",
1419 bond_dev->name);
1da177e4
LT		 1420 }
1421
1422 /* already enslaved */
1423 if (slave_dev->flags & IFF_SLAVE) {
1424 dprintk("Error, Device was already enslaved\n");
1425 return -EBUSY;
1426 }
1427
1428 /* vlan challenged mutual exclusion */
1429 /* no need to lock since we're protected by rtnl_lock */
1430 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1431 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1432 if (!list_empty(&bond->vlan_list)) {
1433 printk(KERN_ERR DRV_NAME
4e0952c7 1434 ": %s: Error: cannot enslave VLAN "
1da177e4 1435 "challenged slave %s on VLAN enabled "
4e0952c7 1436 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1437 bond_dev->name);
1438 return -EPERM;
1439 } else {
1440 printk(KERN_WARNING DRV_NAME
4e0952c7 1441 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1442 "slave %s. Adding VLANs will be blocked as "
1443 "long as %s is part of bond %s\n",
4e0952c7 1444 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1445 bond_dev->name);
1446 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1447 }
1448 } else {
1449 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1450 if (bond->slave_cnt == 0) {
1451 /* First slave, and it is not VLAN challenged,
1452 * so remove the block of adding VLANs over the bond.
1453 */
1454 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1455 }
1456 }
1457
217df670
JV		 1458 /*
1459 * Old ifenslave binaries are no longer supported. These can
1460 * be identified with moderate accurary by the state of the slave:
1461 * the current ifenslave will set the interface down prior to
1462 * enslaving it; the old ifenslave will not.
1463 */
1464 if ((slave_dev->flags & IFF_UP)) {
1465 printk(KERN_ERR DRV_NAME ": %s is up. "
1466 "This may be due to an out of date ifenslave.\n",
1467 slave_dev->name);
1468 res = -EPERM;
1469 goto err_undo_flags;
1470 }
1da177e4 1471
872254dd
MS		 1472 /* set bonding device ether type by slave - bonding netdevices are
1473 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1474 * there is a need to override some of the type dependent attribs/funcs.
1475 *
1476 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1477 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1478 */
1479 if (bond->slave_cnt == 0) {
1480 if (slave_dev->type != ARPHRD_ETHER)
1481 bond_setup_by_slave(bond_dev, slave_dev);
1482 } else if (bond_dev->type != slave_dev->type) {
1483 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1484 "from other slaves (%d), can not enslave it.\n",
1485 slave_dev->name,
1486 slave_dev->type, bond_dev->type);
1487 res = -EINVAL;
1488 goto err_undo_flags;
1489 }
1490
eb7cc59a 1491 if (slave_ops->ndo_set_mac_address == NULL) {
2ab82852
MS		 1492 if (bond->slave_cnt == 0) {
1493 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1494 ": %s: Warning: The first slave device "
1495 "specified does not support setting the MAC "
3915c1e8 1496 "address. Setting fail_over_mac to active.",
dd957c57 1497 bond_dev->name);
3915c1e8
JV		 1498 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1499 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1500 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1501 ": %s: Error: The slave device specified "
1502 "does not support setting the MAC address, "
3915c1e8 1503 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1504 , bond_dev->name);
1505 res = -EOPNOTSUPP;
1506 goto err_undo_flags;
1507 }
1da177e4
LT		 1508 }
1509
243cb4e5 1510 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1511 if (!new_slave) {
1512 res = -ENOMEM;
1513 goto err_undo_flags;
1514 }
1515
1da177e4
LT		 1516 /* save slave's original flags before calling
1517 * netdev_set_master and dev_open
1518 */
1519 new_slave->original_flags = slave_dev->flags;
1520
217df670
JV		 1521 /*
1522 * Save slave's original ("permanent") mac address for modes
1523 * that need it, and for restoring it upon release, and then
1524 * set it to the master's address
1525 */
1526 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1527
dd957c57 1528 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1529 /*
1530 * Set slave to master's mac address. The application already
1531 * set the master's mac address to that of the first slave
1532 */
1533 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1534 addr.sa_family = slave_dev->type;
1535 res = dev_set_mac_address(slave_dev, &addr);
1536 if (res) {
1537 dprintk("Error %d calling set_mac_address\n", res);
1538 goto err_free;
1539 }
217df670 1540 }
1da177e4 1541
c2edacf8
JV		 1542 res = netdev_set_master(slave_dev, bond_dev);
1543 if (res) {
1544 dprintk("Error %d calling netdev_set_master\n", res);
569f0c4d 1545 goto err_restore_mac;
c2edacf8 1546 }
217df670
JV		 1547 /* open the slave since the application closed it */
1548 res = dev_open(slave_dev);
1549 if (res) {
1550 dprintk("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1551 goto err_unset_master;
1da177e4
LT		 1552 }
1553
1da177e4 1554 new_slave->dev = slave_dev;
0b680e75 1555 slave_dev->priv_flags |= IFF_BONDING;
1da177e4
LT		 1556
1557 if ((bond->params.mode == BOND_MODE_TLB) ||
1558 (bond->params.mode == BOND_MODE_ALB)) {
1559 /* bond_alb_init_slave() must be called before all other stages since
1560 * it might fail and we do not want to have to undo everything
1561 */
1562 res = bond_alb_init_slave(bond, new_slave);
1563 if (res) {
569f0c4d 1564 goto err_close;
1da177e4
LT		 1565 }
1566 }
1567
1568 /* If the mode USES_PRIMARY, then the new slave gets the
1569 * master's promisc (and mc) settings only if it becomes the
1570 * curr_active_slave, and that is taken care of later when calling
1571 * bond_change_active()
1572 */
1573 if (!USES_PRIMARY(bond->params.mode)) {
1574 /* set promiscuity level to new slave */
1575 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1576 res = dev_set_promiscuity(slave_dev, 1);
1577 if (res)
1578 goto err_close;
1da177e4
LT		 1579 }
1580
1581 /* set allmulti level to new slave */
1582 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1583 res = dev_set_allmulti(slave_dev, 1);
1584 if (res)
1585 goto err_close;
1da177e4
LT		 1586 }
1587
b9e40857 1588 netif_addr_lock_bh(bond_dev);
1da177e4
LT		 1589 /* upload master's mc_list to new slave */
1590 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1591 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1592 }
b9e40857 1593 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1594 }
1595
1596 if (bond->params.mode == BOND_MODE_8023AD) {
1597 /* add lacpdu mc addr to mc list */
1598 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1599
1600 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1601 }
1602
1603 bond_add_vlans_on_slave(bond, slave_dev);
1604
1605 write_lock_bh(&bond->lock);
1606
1607 bond_attach_slave(bond, new_slave);
1608
1609 new_slave->delay = 0;
1610 new_slave->link_failure_count = 0;
1611
8531c5ff
AK		 1612 bond_compute_features(bond);
1613
3915c1e8
JV		 1614 write_unlock_bh(&bond->lock);
1615
1616 read_lock(&bond->lock);
1617
f5b2b966
JV		 1618 new_slave->last_arp_rx = jiffies;
1619
1da177e4
LT		 1620 if (bond->params.miimon && !bond->params.use_carrier) {
1621 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1622
1623 if ((link_reporting == -1) && !bond->params.arp_interval) {
1624 /*
1625 * miimon is set but a bonded network driver
1626 * does not support ETHTOOL/MII and
1627 * arp_interval is not set. Note: if
1628 * use_carrier is enabled, we will never go
1629 * here (because netif_carrier is always
1630 * supported); thus, we don't need to change
1631 * the messages for netif_carrier.
1632 */
1633 printk(KERN_WARNING DRV_NAME
4e0952c7 1634 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1635 "available for interface %s, and "
1636 "arp_interval/arp_ip_target module parameters "
1637 "not specified, thus bonding will not detect "
1638 "link failures! see bonding.txt for details.\n",
4e0952c7 1639 bond_dev->name, slave_dev->name);
1da177e4
LT		 1640 } else if (link_reporting == -1) {
1641 /* unable get link status using mii/ethtool */
1642 printk(KERN_WARNING DRV_NAME
4e0952c7 1643 ": %s: Warning: can't get link status from "
1da177e4
LT		 1644 "interface %s; the network driver associated "
1645 "with this interface does not support MII or "
1646 "ETHTOOL link status reporting, thus miimon "
1647 "has no effect on this interface.\n",
4e0952c7 1648 bond_dev->name, slave_dev->name);
1da177e4
LT		 1649 }
1650 }
1651
1652 /* check for initial state */
1653 if (!bond->params.miimon ||
1654 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1655 if (bond->params.updelay) {
1656 dprintk("Initial state of slave_dev is "
1657 "BOND_LINK_BACK\n");
1658 new_slave->link = BOND_LINK_BACK;
1659 new_slave->delay = bond->params.updelay;
1660 } else {
1661 dprintk("Initial state of slave_dev is "
1662 "BOND_LINK_UP\n");
1663 new_slave->link = BOND_LINK_UP;
1664 }
1665 new_slave->jiffies = jiffies;
1666 } else {
1667 dprintk("Initial state of slave_dev is "
1668 "BOND_LINK_DOWN\n");
1669 new_slave->link = BOND_LINK_DOWN;
1670 }
1671
1672 if (bond_update_speed_duplex(new_slave) &&
1673 (new_slave->link != BOND_LINK_DOWN)) {
1674 printk(KERN_WARNING DRV_NAME
4e0952c7 1675 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1676 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1677 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1678
1679 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1680 printk(KERN_WARNING DRV_NAME
1681 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1682 "support in base driver for proper aggregator "
4e0952c7 1683 "selection.\n", bond_dev->name);
1da177e4
LT		 1684 }
1685 }
1686
1687 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1688 /* if there is a primary slave, remember it */
1689 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1690 bond->primary_slave = new_slave;
1691 }
1692 }
1693
3915c1e8
JV		 1694 write_lock_bh(&bond->curr_slave_lock);
1695
1da177e4
LT		 1696 switch (bond->params.mode) {
1697 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1698 bond_set_slave_inactive_flags(new_slave);
1699 bond_select_active_slave(bond);
1da177e4
LT		 1700 break;
1701 case BOND_MODE_8023AD:
1702 /* in 802.3ad mode, the internal mechanism
1703 * will activate the slaves in the selected
1704 * aggregator
1705 */
1706 bond_set_slave_inactive_flags(new_slave);
1707 /* if this is the first slave */
1708 if (bond->slave_cnt == 1) {
1709 SLAVE_AD_INFO(new_slave).id = 1;
1710 /* Initialize AD with the number of times that the AD timer is called in 1 second
1711 * can be called only after the mac address of the bond is set
1712 */
1713 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1714 bond->params.lacp_fast);
1715 } else {
1716 SLAVE_AD_INFO(new_slave).id =
1717 SLAVE_AD_INFO(new_slave->prev).id + 1;
1718 }
1719
1720 bond_3ad_bind_slave(new_slave);
1721 break;
1722 case BOND_MODE_TLB:
1723 case BOND_MODE_ALB:
1724 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1725 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1726 break;
1727 default:
1728 dprintk("This slave is always active in trunk mode\n");
1729
1730 /* always active in trunk mode */
1731 new_slave->state = BOND_STATE_ACTIVE;
1732
1733 /* In trunking mode there is little meaning to curr_active_slave
1734 * anyway (it holds no special properties of the bond device),
1735 * so we can change it without calling change_active_interface()
1736 */
1737 if (!bond->curr_active_slave) {
1738 bond->curr_active_slave = new_slave;
1739 }
1740 break;
1741 } /* switch(bond_mode) */
1742
3915c1e8
JV		 1743 write_unlock_bh(&bond->curr_slave_lock);
1744
ff59c456
JV		 1745 bond_set_carrier(bond);
1746
3915c1e8 1747 read_unlock(&bond->lock);
1da177e4 1748
b76cdba9
MW		 1749 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1750 if (res)
569f0c4d 1751 goto err_close;
b76cdba9 1752
1da177e4
LT		 1753 printk(KERN_INFO DRV_NAME
1754 ": %s: enslaving %s as a%s interface with a%s link.\n",
1755 bond_dev->name, slave_dev->name,
1756 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1757 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1758
1759 /* enslave is successful */
1760 return 0;
1761
1762/* Undo stages on error */
1da177e4
LT		 1763err_close:
1764 dev_close(slave_dev);
1765
569f0c4d
JV		 1766err_unset_master:
1767 netdev_set_master(slave_dev, NULL);
1768
1da177e4 1769err_restore_mac:
dd957c57 1770 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1771 /* XXX TODO - fom follow mode needs to change master's
1772 * MAC if this slave's MAC is in use by the bond, or at
1773 * least print a warning.
1774 */
2ab82852
MS		 1775 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1776 addr.sa_family = slave_dev->type;
1777 dev_set_mac_address(slave_dev, &addr);
1778 }
1da177e4
LT		 1779
1780err_free:
1781 kfree(new_slave);
1782
1783err_undo_flags:
1784 bond_dev->features = old_features;
8531c5ff 1785
1da177e4
LT		 1786 return res;
1787}
1788
1789/*
1790 * Try to release the slave device <slave> from the bond device <master>
1791 * It is legal to access curr_active_slave without a lock because all the function
1792 * is write-locked.
1793 *
1794 * The rules for slave state should be:
1795 * for Active/Backup:
1796 * Active stays on all backups go down
1797 * for Bonded connections:
1798 * The first up interface should be left on and all others downed.
1799 */
a77b5325 1800int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1801{
454d7c9b 1802 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1803 struct slave *slave, *oldcurrent;
1804 struct sockaddr addr;
1805 int mac_addr_differ;
1806
1807 /* slave is not a slave or master is not master of this slave */
1808 if (!(slave_dev->flags & IFF_SLAVE) ||
1809 (slave_dev->master != bond_dev)) {
1810 printk(KERN_ERR DRV_NAME
4e0952c7 1811 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1812 bond_dev->name, slave_dev->name);
1813 return -EINVAL;
1814 }
1815
1816 write_lock_bh(&bond->lock);
1817
1818 slave = bond_get_slave_by_dev(bond, slave_dev);
1819 if (!slave) {
1820 /* not a slave of this bond */
1821 printk(KERN_INFO DRV_NAME
1822 ": %s: %s not enslaved\n",
1823 bond_dev->name, slave_dev->name);
f5e2a7b2 1824 write_unlock_bh(&bond->lock);
1da177e4
LT		 1825 return -EINVAL;
1826 }
1827
3915c1e8
JV		 1828 if (!bond->params.fail_over_mac) {
1829 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1830 ETH_ALEN);
1831 if (!mac_addr_differ && (bond->slave_cnt > 1))
1832 printk(KERN_WARNING DRV_NAME
1833 ": %s: Warning: the permanent HWaddr of %s - "
e174961c 1834 "%pM - is still in use by %s. "
3915c1e8
JV		 1835 "Set the HWaddr of %s to a different address "
1836 "to avoid conflicts.\n",
1837 bond_dev->name, slave_dev->name,
e174961c 1838 slave->perm_hwaddr,
3915c1e8 1839 bond_dev->name, slave_dev->name);
1da177e4
LT		 1840 }
1841
1842 /* Inform AD package of unbinding of slave. */
1843 if (bond->params.mode == BOND_MODE_8023AD) {
1844 /* must be called before the slave is
1845 * detached from the list
1846 */
1847 bond_3ad_unbind_slave(slave);
1848 }
1849
1850 printk(KERN_INFO DRV_NAME
1851 ": %s: releasing %s interface %s\n",
1852 bond_dev->name,
1853 (slave->state == BOND_STATE_ACTIVE)
1854 ? "active" : "backup",
1855 slave_dev->name);
1856
1857 oldcurrent = bond->curr_active_slave;
1858
1859 bond->current_arp_slave = NULL;
1860
1861 /* release the slave from its bond */
1862 bond_detach_slave(bond, slave);
1863
8531c5ff
AK		 1864 bond_compute_features(bond);
1865
1da177e4
LT		 1866 if (bond->primary_slave == slave) {
1867 bond->primary_slave = NULL;
1868 }
1869
1870 if (oldcurrent == slave) {
1871 bond_change_active_slave(bond, NULL);
1872 }
1873
1874 if ((bond->params.mode == BOND_MODE_TLB) ||
1875 (bond->params.mode == BOND_MODE_ALB)) {
1876 /* Must be called only after the slave has been
1877 * detached from the list and the curr_active_slave
1878 * has been cleared (if our_slave == old_current),
1879 * but before a new active slave is selected.
1880 */
2543331d 1881 write_unlock_bh(&bond->lock);
1da177e4 1882 bond_alb_deinit_slave(bond, slave);
2543331d 1883 write_lock_bh(&bond->lock);
1da177e4
LT		 1884 }
1885
059fe7a5
JV		 1886 if (oldcurrent == slave) {
1887 /*
1888 * Note that we hold RTNL over this sequence, so there
1889 * is no concern that another slave add/remove event
1890 * will interfere.
1891 */
1892 write_unlock_bh(&bond->lock);
1893 read_lock(&bond->lock);
1894 write_lock_bh(&bond->curr_slave_lock);
1895
1da177e4
LT		 1896 bond_select_active_slave(bond);
1897
059fe7a5
JV		 1898 write_unlock_bh(&bond->curr_slave_lock);
1899 read_unlock(&bond->lock);
1900 write_lock_bh(&bond->lock);
1901 }
1902
1da177e4 1903 if (bond->slave_cnt == 0) {
ff59c456
JV		 1904 bond_set_carrier(bond);
1905
1da177e4
LT		 1906 /* if the last slave was removed, zero the mac address
1907 * of the master so it will be set by the application
1908 * to the mac address of the first slave
1909 */
1910 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1911
1912 if (list_empty(&bond->vlan_list)) {
1913 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1914 } else {
1915 printk(KERN_WARNING DRV_NAME
4e0952c7 1916 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1917 "still has VLANs.\n",
4e0952c7 1918 bond_dev->name, bond_dev->name);
1da177e4 1919 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1920 ": %s: When re-adding slaves, make sure the bond's "
1921 "HW address matches its VLANs'.\n",
1922 bond_dev->name);
1da177e4
LT		 1923 }
1924 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1925 !bond_has_challenged_slaves(bond)) {
1926 printk(KERN_INFO DRV_NAME
4e0952c7 1927 ": %s: last VLAN challenged slave %s "
1da177e4 1928 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1929 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1930 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1931 }
1932
1933 write_unlock_bh(&bond->lock);
1934
b76cdba9
MW		 1935 /* must do this from outside any spinlocks */
1936 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1937
1da177e4
LT		 1938 bond_del_vlans_from_slave(bond, slave_dev);
1939
1940 /* If the mode USES_PRIMARY, then we should only remove its
1941 * promisc and mc settings if it was the curr_active_slave, but that was
1942 * already taken care of above when we detached the slave
1943 */
1944 if (!USES_PRIMARY(bond->params.mode)) {
1945 /* unset promiscuity level from slave */
1946 if (bond_dev->flags & IFF_PROMISC) {
1947 dev_set_promiscuity(slave_dev, -1);
1948 }
1949
1950 /* unset allmulti level from slave */
1951 if (bond_dev->flags & IFF_ALLMULTI) {
1952 dev_set_allmulti(slave_dev, -1);
1953 }
1954
1955 /* flush master's mc_list from slave */
b9e40857 1956 netif_addr_lock_bh(bond_dev);
1da177e4 1957 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 1958 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1959 }
1960
1961 netdev_set_master(slave_dev, NULL);
1962
1963 /* close slave before restoring its mac address */
1964 dev_close(slave_dev);
1965
3915c1e8 1966 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1967 /* restore original ("permanent") mac address */
1968 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1969 addr.sa_family = slave_dev->type;
1970 dev_set_mac_address(slave_dev, &addr);
1971 }
1da177e4 1972
8f903c70 1973 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1974 IFF_SLAVE_INACTIVE | IFF_BONDING |
1975 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1976
1977 kfree(slave);
1978
1979 return 0; /* deletion OK */
1980}
1981
d90a162a
MS		 1982/*
1983* Destroy a bonding device.
1984* Must be under rtnl_lock when this function is called.
1985*/
1986void bond_destroy(struct bonding *bond)
1987{
1988 bond_deinit(bond->dev);
1989 bond_destroy_sysfs_entry(bond);
8cbdeec6 1990 unregister_netdevice(bond->dev);
d90a162a
MS		 1991}
1992
a434e43f
JV		 1993static void bond_destructor(struct net_device *bond_dev)
1994{
244ef9b9 1995 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 1996
1997 if (bond->wq)
1998 destroy_workqueue(bond->wq);
1999
2000 netif_addr_lock_bh(bond_dev);
2001 bond_mc_list_destroy(bond);
2002 netif_addr_unlock_bh(bond_dev);
2003
2004 free_netdev(bond_dev);
2005}
2006
d90a162a
MS		 2007/*
2008* First release a slave and than destroy the bond if no more slaves iare left.
2009* Must be under rtnl_lock when this function is called.
2010*/
2011int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2012{
454d7c9b 2013 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 2014 int ret;
2015
2016 ret = bond_release(bond_dev, slave_dev);
2017 if ((ret == 0) && (bond->slave_cnt == 0)) {
2018 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2019 bond_dev->name, bond_dev->name);
2020 bond_destroy(bond);
2021 }
2022 return ret;
2023}
2024
1da177e4
LT		 2025/*
2026 * This function releases all slaves.
2027 */
2028static int bond_release_all(struct net_device *bond_dev)
2029{
454d7c9b 2030 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2031 struct slave *slave;
2032 struct net_device *slave_dev;
2033 struct sockaddr addr;
2034
2035 write_lock_bh(&bond->lock);
2036
ff59c456
JV		 2037 netif_carrier_off(bond_dev);
2038
1da177e4
LT		 2039 if (bond->slave_cnt == 0) {
2040 goto out;
2041 }
2042
2043 bond->current_arp_slave = NULL;
2044 bond->primary_slave = NULL;
2045 bond_change_active_slave(bond, NULL);
2046
2047 while ((slave = bond->first_slave) != NULL) {
2048 /* Inform AD package of unbinding of slave
2049 * before slave is detached from the list.
2050 */
2051 if (bond->params.mode == BOND_MODE_8023AD) {
2052 bond_3ad_unbind_slave(slave);
2053 }
2054
2055 slave_dev = slave->dev;
2056 bond_detach_slave(bond, slave);
2057
2543331d
JV		 2058 /* now that the slave is detached, unlock and perform
2059 * all the undo steps that should not be called from
2060 * within a lock.
2061 */
2062 write_unlock_bh(&bond->lock);
2063
1da177e4
LT		 2064 if ((bond->params.mode == BOND_MODE_TLB) ||
2065 (bond->params.mode == BOND_MODE_ALB)) {
2066 /* must be called only after the slave
2067 * has been detached from the list
2068 */
2069 bond_alb_deinit_slave(bond, slave);
2070 }
2071
8531c5ff
AK		 2072 bond_compute_features(bond);
2073
b76cdba9 2074 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2075 bond_del_vlans_from_slave(bond, slave_dev);
2076
2077 /* If the mode USES_PRIMARY, then we should only remove its
2078 * promisc and mc settings if it was the curr_active_slave, but that was
2079 * already taken care of above when we detached the slave
2080 */
2081 if (!USES_PRIMARY(bond->params.mode)) {
2082 /* unset promiscuity level from slave */
2083 if (bond_dev->flags & IFF_PROMISC) {
2084 dev_set_promiscuity(slave_dev, -1);
2085 }
2086
2087 /* unset allmulti level from slave */
2088 if (bond_dev->flags & IFF_ALLMULTI) {
2089 dev_set_allmulti(slave_dev, -1);
2090 }
2091
2092 /* flush master's mc_list from slave */
b9e40857 2093 netif_addr_lock_bh(bond_dev);
1da177e4 2094 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2095 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2096 }
2097
2098 netdev_set_master(slave_dev, NULL);
2099
2100 /* close slave before restoring its mac address */
2101 dev_close(slave_dev);
2102
dd957c57 2103 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2104 /* restore original ("permanent") mac address*/
2105 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2106 addr.sa_family = slave_dev->type;
2107 dev_set_mac_address(slave_dev, &addr);
2108 }
1da177e4 2109
8f903c70
JV		 2110 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2111 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2112
2113 kfree(slave);
2114
2115 /* re-acquire the lock before getting the next slave */
2116 write_lock_bh(&bond->lock);
2117 }
2118
2119 /* zero the mac address of the master so it will be
2120 * set by the application to the mac address of the
2121 * first slave
2122 */
2123 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2124
2125 if (list_empty(&bond->vlan_list)) {
2126 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2127 } else {
2128 printk(KERN_WARNING DRV_NAME
4e0952c7 2129 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2130 "still has VLANs.\n",
4e0952c7 2131 bond_dev->name, bond_dev->name);
1da177e4 2132 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2133 ": %s: When re-adding slaves, make sure the bond's "
2134 "HW address matches its VLANs'.\n",
2135 bond_dev->name);
1da177e4
LT		 2136 }
2137
2138 printk(KERN_INFO DRV_NAME
2139 ": %s: released all slaves\n",
2140 bond_dev->name);
2141
2142out:
2143 write_unlock_bh(&bond->lock);
2144
2145 return 0;
2146}
2147
2148/*
2149 * This function changes the active slave to slave <slave_dev>.
2150 * It returns -EINVAL in the following cases.
2151 * - <slave_dev> is not found in the list.
2152 * - There is not active slave now.
2153 * - <slave_dev> is already active.
2154 * - The link state of <slave_dev> is not BOND_LINK_UP.
2155 * - <slave_dev> is not running.
2156 * In these cases, this fuction does nothing.
2157 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2158 */
2159static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2160{
454d7c9b 2161 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2162 struct slave *old_active = NULL;
2163 struct slave *new_active = NULL;
2164 int res = 0;
2165
2166 if (!USES_PRIMARY(bond->params.mode)) {
2167 return -EINVAL;
2168 }
2169
2170 /* Verify that master_dev is indeed the master of slave_dev */
2171 if (!(slave_dev->flags & IFF_SLAVE) ||
2172 (slave_dev->master != bond_dev)) {
2173 return -EINVAL;
2174 }
2175
059fe7a5 2176 read_lock(&bond->lock);
1da177e4 2177
059fe7a5 2178 read_lock(&bond->curr_slave_lock);
1da177e4 2179 old_active = bond->curr_active_slave;
059fe7a5
JV		 2180 read_unlock(&bond->curr_slave_lock);
2181
1da177e4
LT		 2182 new_active = bond_get_slave_by_dev(bond, slave_dev);
2183
2184 /*
2185 * Changing to the current active: do nothing; return success.
2186 */
2187 if (new_active && (new_active == old_active)) {
059fe7a5 2188 read_unlock(&bond->lock);
1da177e4
LT		 2189 return 0;
2190 }
2191
2192 if ((new_active) &&
2193 (old_active) &&
2194 (new_active->link == BOND_LINK_UP) &&
2195 IS_UP(new_active->dev)) {
059fe7a5 2196 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2197 bond_change_active_slave(bond, new_active);
059fe7a5 2198 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2199 } else {
2200 res = -EINVAL;
2201 }
2202
059fe7a5 2203 read_unlock(&bond->lock);
1da177e4
LT		 2204
2205 return res;
2206}
2207
1da177e4
LT		 2208static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2209{
454d7c9b 2210 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2211
2212 info->bond_mode = bond->params.mode;
2213 info->miimon = bond->params.miimon;
2214
6603a6f2 2215 read_lock(&bond->lock);
1da177e4 2216 info->num_slaves = bond->slave_cnt;
6603a6f2 2217 read_unlock(&bond->lock);
1da177e4
LT		 2218
2219 return 0;
2220}
2221
2222static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2223{
454d7c9b 2224 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2225 struct slave *slave;
2226 int i, found = 0;
2227
2228 if (info->slave_id < 0) {
2229 return -ENODEV;
2230 }
2231
6603a6f2 2232 read_lock(&bond->lock);
1da177e4
LT		 2233
2234 bond_for_each_slave(bond, slave, i) {
2235 if (i == (int)info->slave_id) {
2236 found = 1;
2237 break;
2238 }
2239 }
2240
6603a6f2 2241 read_unlock(&bond->lock);
1da177e4
LT		 2242
2243 if (found) {
2244 strcpy(info->slave_name, slave->dev->name);
2245 info->link = slave->link;
2246 info->state = slave->state;
2247 info->link_failure_count = slave->link_failure_count;
2248 } else {
2249 return -ENODEV;
2250 }
2251
2252 return 0;
2253}
2254
2255/*-------------------------------- Monitoring -------------------------------*/
2256
1da177e4 2257
f0c76d61
JV		 2258static int bond_miimon_inspect(struct bonding *bond)
2259{
2260 struct slave *slave;
2261 int i, link_state, commit = 0;
1da177e4
LT		 2262
2263 bond_for_each_slave(bond, slave, i) {
f0c76d61 2264 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2265
f0c76d61 2266 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2267
2268 switch (slave->link) {
f0c76d61
JV		 2269 case BOND_LINK_UP:
2270 if (link_state)
2271 continue;
1da177e4 2272
f0c76d61
JV		 2273 slave->link = BOND_LINK_FAIL;
2274 slave->delay = bond->params.downdelay;
2275 if (slave->delay) {
2276 printk(KERN_INFO DRV_NAME
2277 ": %s: link status down for %s"
2278 "interface %s, disabling it in %d ms.\n",
2279 bond->dev->name,
2280 (bond->params.mode ==
2281 BOND_MODE_ACTIVEBACKUP) ?
2282 ((slave->state == BOND_STATE_ACTIVE) ?
2283 "active " : "backup ") : "",
2284 slave->dev->name,
2285 bond->params.downdelay * bond->params.miimon);
1da177e4 2286 }
f0c76d61
JV		 2287 /*FALLTHRU*/
2288 case BOND_LINK_FAIL:
2289 if (link_state) {
2290 /*
2291 * recovered before downdelay expired
2292 */
2293 slave->link = BOND_LINK_UP;
1da177e4
LT		 2294 slave->jiffies = jiffies;
2295 printk(KERN_INFO DRV_NAME
2296 ": %s: link status up again after %d "
2297 "ms for interface %s.\n",
1b76b316 2298 bond->dev->name,
f0c76d61
JV		 2299 (bond->params.downdelay - slave->delay) *
2300 bond->params.miimon,
2301 slave->dev->name);
2302 continue;
1da177e4 2303 }
f0c76d61
JV		 2304
2305 if (slave->delay <= 0) {
2306 slave->new_link = BOND_LINK_DOWN;
2307 commit++;
2308 continue;
1da177e4 2309 }
1da177e4 2310
f0c76d61
JV		 2311 slave->delay--;
2312 break;
2313
2314 case BOND_LINK_DOWN:
2315 if (!link_state)
2316 continue;
2317
2318 slave->link = BOND_LINK_BACK;
2319 slave->delay = bond->params.updelay;
2320
2321 if (slave->delay) {
2322 printk(KERN_INFO DRV_NAME
2323 ": %s: link status up for "
2324 "interface %s, enabling it in %d ms.\n",
2325 bond->dev->name, slave->dev->name,
2326 bond->params.updelay *
2327 bond->params.miimon);
2328 }
2329 /*FALLTHRU*/
2330 case BOND_LINK_BACK:
2331 if (!link_state) {
2332 slave->link = BOND_LINK_DOWN;
1da177e4
LT		 2333 printk(KERN_INFO DRV_NAME
2334 ": %s: link status down again after %d "
2335 "ms for interface %s.\n",
1b76b316 2336 bond->dev->name,
f0c76d61
JV		 2337 (bond->params.updelay - slave->delay) *
2338 bond->params.miimon,
2339 slave->dev->name);
2340
2341 continue;
2342 }
2343
2344 if (slave->delay <= 0) {
2345 slave->new_link = BOND_LINK_UP;
2346 commit++;
2347 continue;
1da177e4 2348 }
f0c76d61
JV		 2349
2350 slave->delay--;
1da177e4 2351 break;
f0c76d61
JV		 2352 }
2353 }
1da177e4 2354
f0c76d61
JV		 2355 return commit;
2356}
1da177e4 2357
f0c76d61
JV		 2358static void bond_miimon_commit(struct bonding *bond)
2359{
2360 struct slave *slave;
2361 int i;
2362
2363 bond_for_each_slave(bond, slave, i) {
2364 switch (slave->new_link) {
2365 case BOND_LINK_NOCHANGE:
2366 continue;
1da177e4 2367
f0c76d61
JV		 2368 case BOND_LINK_UP:
2369 slave->link = BOND_LINK_UP;
2370 slave->jiffies = jiffies;
2371
2372 if (bond->params.mode == BOND_MODE_8023AD) {
2373 /* prevent it from being the active one */
2374 slave->state = BOND_STATE_BACKUP;
2375 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2376 /* make it immediately active */
2377 slave->state = BOND_STATE_ACTIVE;
2378 } else if (slave != bond->primary_slave) {
2379 /* prevent it from being the active one */
2380 slave->state = BOND_STATE_BACKUP;
1da177e4 2381 }
1da177e4 2382
f0c76d61
JV		 2383 printk(KERN_INFO DRV_NAME
2384 ": %s: link status definitely "
2385 "up for interface %s.\n",
2386 bond->dev->name, slave->dev->name);
1da177e4 2387
f0c76d61
JV		 2388 /* notify ad that the link status has changed */
2389 if (bond->params.mode == BOND_MODE_8023AD)
2390 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2391
f0c76d61
JV		 2392 if ((bond->params.mode == BOND_MODE_TLB) ||
2393 (bond->params.mode == BOND_MODE_ALB))
2394 bond_alb_handle_link_change(bond, slave,
2395 BOND_LINK_UP);
1da177e4 2396
f0c76d61
JV		 2397 if (!bond->curr_active_slave ||
2398 (slave == bond->primary_slave))
2399 goto do_failover;
1da177e4 2400
f0c76d61 2401 continue;
059fe7a5 2402
f0c76d61 2403 case BOND_LINK_DOWN:
fba4acda
JV		 2404 if (slave->link_failure_count < UINT_MAX)
2405 slave->link_failure_count++;
2406
f0c76d61 2407 slave->link = BOND_LINK_DOWN;
1da177e4 2408
f0c76d61
JV		 2409 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2410 bond->params.mode == BOND_MODE_8023AD)
2411 bond_set_slave_inactive_flags(slave);
2412
2413 printk(KERN_INFO DRV_NAME
2414 ": %s: link status definitely down for "
2415 "interface %s, disabling it\n",
2416 bond->dev->name, slave->dev->name);
2417
2418 if (bond->params.mode == BOND_MODE_8023AD)
2419 bond_3ad_handle_link_change(slave,
2420 BOND_LINK_DOWN);
2421
2422 if (bond->params.mode == BOND_MODE_TLB ||
2423 bond->params.mode == BOND_MODE_ALB)
2424 bond_alb_handle_link_change(bond, slave,
2425 BOND_LINK_DOWN);
2426
2427 if (slave == bond->curr_active_slave)
2428 goto do_failover;
2429
2430 continue;
2431
2432 default:
2433 printk(KERN_ERR DRV_NAME
2434 ": %s: invalid new link %d on slave %s\n",
2435 bond->dev->name, slave->new_link,
2436 slave->dev->name);
2437 slave->new_link = BOND_LINK_NOCHANGE;
2438
2439 continue;
2440 }
2441
2442do_failover:
2443 ASSERT_RTNL();
2444 write_lock_bh(&bond->curr_slave_lock);
2445 bond_select_active_slave(bond);
2446 write_unlock_bh(&bond->curr_slave_lock);
2447 }
2448
2449 bond_set_carrier(bond);
1da177e4
LT		 2450}
2451
0b0eef66
JV		 2452/*
2453 * bond_mii_monitor
2454 *
2455 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2456 * inspection, then (if inspection indicates something needs to be done)
2457 * an acquisition of appropriate locks followed by a commit phase to
2458 * implement whatever link state changes are indicated.
0b0eef66
JV		 2459 */
2460void bond_mii_monitor(struct work_struct *work)
2461{
2462 struct bonding *bond = container_of(work, struct bonding,
2463 mii_work.work);
0b0eef66
JV		 2464
2465 read_lock(&bond->lock);
f0c76d61
JV		 2466 if (bond->kill_timers)
2467 goto out;
2468
2469 if (bond->slave_cnt == 0)
2470 goto re_arm;
b59f9f74
JV		 2471
2472 if (bond->send_grat_arp) {
2473 read_lock(&bond->curr_slave_lock);
2474 bond_send_gratuitous_arp(bond);
2475 read_unlock(&bond->curr_slave_lock);
2476 }
2477
305d552a
BH		 2478 if (bond->send_unsol_na) {
2479 read_lock(&bond->curr_slave_lock);
2480 bond_send_unsolicited_na(bond);
2481 read_unlock(&bond->curr_slave_lock);
2482 }
2483
f0c76d61 2484 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2485 read_unlock(&bond->lock);
2486 rtnl_lock();
2487 read_lock(&bond->lock);
f0c76d61
JV		 2488
2489 bond_miimon_commit(bond);
2490
5655662d
JV		 2491 read_unlock(&bond->lock);
2492 rtnl_unlock(); /* might sleep, hold no other locks */
2493 read_lock(&bond->lock);
0b0eef66
JV		 2494 }
2495
f0c76d61
JV		 2496re_arm:
2497 if (bond->params.miimon)
2498 queue_delayed_work(bond->wq, &bond->mii_work,
2499 msecs_to_jiffies(bond->params.miimon));
2500out:
0b0eef66 2501 read_unlock(&bond->lock);
0b0eef66 2502}
c3ade5ca 2503
d3bb52b0 2504static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2505{
2506 struct in_device *idev;
2507 struct in_ifaddr *ifa;
a144ea4b 2508 __be32 addr = 0;
c3ade5ca
JV		 2509
2510 if (!dev)
2511 return 0;
2512
2513 rcu_read_lock();
e5ed6399 2514 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2515 if (!idev)
2516 goto out;
2517
2518 ifa = idev->ifa_list;
2519 if (!ifa)
2520 goto out;
2521
2522 addr = ifa->ifa_local;
2523out:
2524 rcu_read_unlock();
2525 return addr;
2526}
2527
d3bb52b0 2528static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2529{
0883beca 2530 struct vlan_entry *vlan;
f5b2b966
JV		 2531
2532 if (ip == bond->master_ip)
2533 return 1;
2534
0883beca 2535 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2536 if (ip == vlan->vlan_ip)
2537 return 1;
2538 }
2539
2540 return 0;
2541}
2542
c3ade5ca
JV		 2543/*
2544 * We go to the (large) trouble of VLAN tagging ARP frames because
2545 * switches in VLAN mode (especially if ports are configured as
2546 * "native" to a VLAN) might not pass non-tagged frames.
2547 */
d3bb52b0 2548static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2549{
2550 struct sk_buff *skb;
2551
2552 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2553 slave_dev->name, dest_ip, src_ip, vlan_id);
2554
2555 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2556 NULL, slave_dev->dev_addr, NULL);
2557
2558 if (!skb) {
2559 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2560 return;
2561 }
2562 if (vlan_id) {
2563 skb = vlan_put_tag(skb, vlan_id);
2564 if (!skb) {
2565 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2566 return;
2567 }
2568 }
2569 arp_xmit(skb);
2570}
2571
2572
1da177e4
LT		 2573static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2574{
c3ade5ca 2575 int i, vlan_id, rv;
d3bb52b0 2576 __be32 *targets = bond->params.arp_targets;
0883beca 2577 struct vlan_entry *vlan;
c3ade5ca
JV		 2578 struct net_device *vlan_dev;
2579 struct flowi fl;
2580 struct rtable *rt;
1da177e4 2581
6b780567
MW		 2582 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2583 if (!targets[i])
2584 continue;
c3ade5ca
JV		 2585 dprintk("basa: target %x\n", targets[i]);
2586 if (list_empty(&bond->vlan_list)) {
2587 dprintk("basa: empty vlan: arp_send\n");
2588 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2589 bond->master_ip, 0);
2590 continue;
2591 }
2592
2593 /*
2594 * If VLANs are configured, we do a route lookup to
2595 * determine which VLAN interface would be used, so we
2596 * can tag the ARP with the proper VLAN tag.
2597 */
2598 memset(&fl, 0, sizeof(fl));
2599 fl.fl4_dst = targets[i];
2600 fl.fl4_tos = RTO_ONLINK;
2601
f206351a 2602 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2603 if (rv) {
2604 if (net_ratelimit()) {
2605 printk(KERN_WARNING DRV_NAME
63779436
HH		 2606 ": %s: no route to arp_ip_target %pI4\n",
2607 bond->dev->name, &fl.fl4_dst);
c3ade5ca
JV		 2608 }
2609 continue;
2610 }
2611
2612 /*
2613 * This target is not on a VLAN
2614 */
2615 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2616 ip_rt_put(rt);
c3ade5ca
JV		 2617 dprintk("basa: rtdev == bond->dev: arp_send\n");
2618 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2619 bond->master_ip, 0);
2620 continue;
2621 }
2622
2623 vlan_id = 0;
0883beca 2624 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2625 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2626 if (vlan_dev == rt->u.dst.dev) {
2627 vlan_id = vlan->vlan_id;
2628 dprintk("basa: vlan match on %s %d\n",
2629 vlan_dev->name, vlan_id);
2630 break;
2631 }
2632 }
2633
2634 if (vlan_id) {
ed4b9f80 2635 ip_rt_put(rt);
c3ade5ca
JV		 2636 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2637 vlan->vlan_ip, vlan_id);
2638 continue;
2639 }
2640
2641 if (net_ratelimit()) {
2642 printk(KERN_WARNING DRV_NAME
63779436
HH		 2643 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2644 bond->dev->name, &fl.fl4_dst,
c3ade5ca
JV		 2645 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2646 }
ed4b9f80 2647 ip_rt_put(rt);
c3ade5ca
JV		 2648 }
2649}
2650
2651/*
2652 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2653 * for each VLAN above us.
b59f9f74
JV		 2654 *
2655 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2656 */
2657static void bond_send_gratuitous_arp(struct bonding *bond)
2658{
2659 struct slave *slave = bond->curr_active_slave;
2660 struct vlan_entry *vlan;
2661 struct net_device *vlan_dev;
2662
2663 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2664 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2665
2666 if (!slave || !bond->send_grat_arp ||
2667 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2668 return;
2669
b59f9f74
JV		 2670 bond->send_grat_arp--;
2671
c3ade5ca
JV		 2672 if (bond->master_ip) {
2673 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2674 bond->master_ip, 0);
c3ade5ca
JV		 2675 }
2676
2677 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2678 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2679 if (vlan->vlan_ip) {
2680 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2681 vlan->vlan_ip, vlan->vlan_id);
2682 }
1da177e4
LT		 2683 }
2684}
2685
d3bb52b0 2686static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2687{
2688 int i;
d3bb52b0 2689 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2690
2691 targets = bond->params.arp_targets;
2692 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
63779436
HH		 2693 dprintk("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2694 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
f5b2b966
JV		 2695 if (sip == targets[i]) {
2696 if (bond_has_this_ip(bond, tip))
2697 slave->last_arp_rx = jiffies;
2698 return;
2699 }
2700 }
2701}
2702
2703static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2704{
2705 struct arphdr *arp;
2706 struct slave *slave;
2707 struct bonding *bond;
2708 unsigned char *arp_ptr;
d3bb52b0 2709 __be32 sip, tip;
f5b2b966 2710
c346dca1 2711 if (dev_net(dev) != &init_net)
e730c155
EB		 2712 goto out;
2713
f5b2b966
JV		 2714 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2715 goto out;
2716
454d7c9b 2717 bond = netdev_priv(dev);
f5b2b966
JV		 2718 read_lock(&bond->lock);
2719
2720 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2721 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2722 orig_dev ? orig_dev->name : "NULL");
2723
2724 slave = bond_get_slave_by_dev(bond, orig_dev);
2725 if (!slave || !slave_do_arp_validate(bond, slave))
2726 goto out_unlock;
2727
988b7050 2728 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2729 goto out_unlock;
2730
d0a92be0 2731 arp = arp_hdr(skb);
f5b2b966
JV		 2732 if (arp->ar_hln != dev->addr_len ||
2733 skb->pkt_type == PACKET_OTHERHOST ||
2734 skb->pkt_type == PACKET_LOOPBACK ||
2735 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2736 arp->ar_pro != htons(ETH_P_IP) ||
2737 arp->ar_pln != 4)
2738 goto out_unlock;
2739
2740 arp_ptr = (unsigned char *)(arp + 1);
2741 arp_ptr += dev->addr_len;
2742 memcpy(&sip, arp_ptr, 4);
2743 arp_ptr += 4 + dev->addr_len;
2744 memcpy(&tip, arp_ptr, 4);
2745
63779436
HH		 2746 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2747 bond->dev->name, slave->dev->name, slave->state,
2748 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2749 &sip, &tip);
f5b2b966
JV		 2750
2751 /*
2752 * Backup slaves won't see the ARP reply, but do come through
2753 * here for each ARP probe (so we swap the sip/tip to validate
2754 * the probe). In a "redundant switch, common router" type of
2755 * configuration, the ARP probe will (hopefully) travel from
2756 * the active, through one switch, the router, then the other
2757 * switch before reaching the backup.
2758 */
2759 if (slave->state == BOND_STATE_ACTIVE)
2760 bond_validate_arp(bond, slave, sip, tip);
2761 else
2762 bond_validate_arp(bond, slave, tip, sip);
2763
2764out_unlock:
2765 read_unlock(&bond->lock);
2766out:
2767 dev_kfree_skb(skb);
2768 return NET_RX_SUCCESS;
2769}
2770
1da177e4
LT		 2771/*
2772 * this function is called regularly to monitor each slave's link
2773 * ensuring that traffic is being sent and received when arp monitoring
2774 * is used in load-balancing mode. if the adapter has been dormant, then an
2775 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2776 * arp monitoring in active backup mode.
2777 */
1b76b316 2778void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2779{
1b76b316
JV		 2780 struct bonding *bond = container_of(work, struct bonding,
2781 arp_work.work);
1da177e4
LT		 2782 struct slave *slave, *oldcurrent;
2783 int do_failover = 0;
2784 int delta_in_ticks;
2785 int i;
2786
2787 read_lock(&bond->lock);
2788
5ce0da8f 2789 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2790
2791 if (bond->kill_timers) {
2792 goto out;
2793 }
2794
2795 if (bond->slave_cnt == 0) {
2796 goto re_arm;
2797 }
2798
2799 read_lock(&bond->curr_slave_lock);
2800 oldcurrent = bond->curr_active_slave;
2801 read_unlock(&bond->curr_slave_lock);
2802
2803 /* see if any of the previous devices are up now (i.e. they have
2804 * xmt and rcv traffic). the curr_active_slave does not come into
2805 * the picture unless it is null. also, slave->jiffies is not needed
2806 * here because we send an arp on each slave and give a slave as
2807 * long as it needs to get the tx/rx within the delta.
2808 * TODO: what about up/down delay in arp mode? it wasn't here before
2809 * so it can wait
2810 */
2811 bond_for_each_slave(bond, slave, i) {
2812 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2813 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2814 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2815
2816 slave->link = BOND_LINK_UP;
2817 slave->state = BOND_STATE_ACTIVE;
2818
2819 /* primary_slave has no meaning in round-robin
2820 * mode. the window of a slave being up and
2821 * curr_active_slave being null after enslaving
2822 * is closed.
2823 */
2824 if (!oldcurrent) {
2825 printk(KERN_INFO DRV_NAME
2826 ": %s: link status definitely "
2827 "up for interface %s, ",
1b76b316 2828 bond->dev->name,
1da177e4
LT		 2829 slave->dev->name);
2830 do_failover = 1;
2831 } else {
2832 printk(KERN_INFO DRV_NAME
2833 ": %s: interface %s is now up\n",
1b76b316 2834 bond->dev->name,
1da177e4
LT		 2835 slave->dev->name);
2836 }
2837 }
2838 } else {
2839 /* slave->link == BOND_LINK_UP */
2840
2841 /* not all switches will respond to an arp request
2842 * when the source ip is 0, so don't take the link down
2843 * if we don't know our ip yet
2844 */
b63bb739 2845 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2846 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2847
2848 slave->link = BOND_LINK_DOWN;
2849 slave->state = BOND_STATE_BACKUP;
2850
2851 if (slave->link_failure_count < UINT_MAX) {
2852 slave->link_failure_count++;
2853 }
2854
2855 printk(KERN_INFO DRV_NAME
2856 ": %s: interface %s is now down.\n",
1b76b316 2857 bond->dev->name,
1da177e4
LT		 2858 slave->dev->name);
2859
2860 if (slave == oldcurrent) {
2861 do_failover = 1;
2862 }
2863 }
2864 }
2865
2866 /* note: if switch is in round-robin mode, all links
2867 * must tx arp to ensure all links rx an arp - otherwise
2868 * links may oscillate or not come up at all; if switch is
2869 * in something like xor mode, there is nothing we can
2870 * do - all replies will be rx'ed on same link causing slaves
2871 * to be unstable during low/no traffic periods
2872 */
2873 if (IS_UP(slave->dev)) {
2874 bond_arp_send_all(bond, slave);
2875 }
2876 }
2877
2878 if (do_failover) {
059fe7a5 2879 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2880
2881 bond_select_active_slave(bond);
2882
059fe7a5 2883 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2884 }
2885
2886re_arm:
1b76b316
JV		 2887 if (bond->params.arp_interval)
2888 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2889out:
2890 read_unlock(&bond->lock);
2891}
2892
2893/*
b2220cad
JV		 2894 * Called to inspect slaves for active-backup mode ARP monitor link state
2895 * changes. Sets new_link in slaves to specify what action should take
2896 * place for the slave. Returns 0 if no changes are found, >0 if changes
2897 * to link states must be committed.
2898 *
2899 * Called with bond->lock held for read.
1da177e4 2900 */
b2220cad 2901static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2902{
1da177e4 2903 struct slave *slave;
b2220cad 2904 int i, commit = 0;
1da177e4 2905
b2220cad
JV		 2906 bond_for_each_slave(bond, slave, i) {
2907 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2908
b2220cad
JV		 2909 if (slave->link != BOND_LINK_UP) {
2910 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2911 delta_in_ticks)) {
2912 slave->new_link = BOND_LINK_UP;
2913 commit++;
2914 }
1da177e4 2915
b2220cad
JV		 2916 continue;
2917 }
1da177e4 2918
b2220cad
JV		 2919 /*
2920 * Give slaves 2*delta after being enslaved or made
2921 * active. This avoids bouncing, as the last receive
2922 * times need a full ARP monitor cycle to be updated.
2923 */
2924 if (!time_after_eq(jiffies, slave->jiffies +
2925 2 * delta_in_ticks))
2926 continue;
2927
2928 /*
2929 * Backup slave is down if:
2930 * - No current_arp_slave AND
2931 * - more than 3*delta since last receive AND
2932 * - the bond has an IP address
2933 *
2934 * Note: a non-null current_arp_slave indicates
2935 * the curr_active_slave went down and we are
2936 * searching for a new one; under this condition
2937 * we only take the curr_active_slave down - this
2938 * gives each slave a chance to tx/rx traffic
2939 * before being taken out
2940 */
2941 if (slave->state == BOND_STATE_BACKUP &&
2942 !bond->current_arp_slave &&
2943 time_after(jiffies, slave_last_rx(bond, slave) +
2944 3 * delta_in_ticks)) {
2945 slave->new_link = BOND_LINK_DOWN;
2946 commit++;
2947 }
2948
2949 /*
2950 * Active slave is down if:
2951 * - more than 2*delta since transmitting OR
2952 * - (more than 2*delta since receive AND
2953 * the bond has an IP address)
2954 */
2955 if ((slave->state == BOND_STATE_ACTIVE) &&
2956 (time_after_eq(jiffies, slave->dev->trans_start +
2957 2 * delta_in_ticks) ||
2958 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2959 + 2 * delta_in_ticks)))) {
2960 slave->new_link = BOND_LINK_DOWN;
2961 commit++;
2962 }
1da177e4
LT		 2963 }
2964
b2220cad
JV		 2965 read_lock(&bond->curr_slave_lock);
2966
2967 /*
2968 * Trigger a commit if the primary option setting has changed.
1da177e4 2969 */
b2220cad
JV		 2970 if (bond->primary_slave &&
2971 (bond->primary_slave != bond->curr_active_slave) &&
2972 (bond->primary_slave->link == BOND_LINK_UP))
2973 commit++;
1da177e4 2974
b2220cad 2975 read_unlock(&bond->curr_slave_lock);
1da177e4 2976
b2220cad
JV		 2977 return commit;
2978}
1da177e4 2979
b2220cad
JV		 2980/*
2981 * Called to commit link state changes noted by inspection step of
2982 * active-backup mode ARP monitor.
2983 *
2984 * Called with RTNL and bond->lock for read.
2985 */
2986static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2987{
2988 struct slave *slave;
2989 int i;
1da177e4 2990
b2220cad
JV		 2991 bond_for_each_slave(bond, slave, i) {
2992 switch (slave->new_link) {
2993 case BOND_LINK_NOCHANGE:
2994 continue;
ff59c456 2995
b2220cad
JV		 2996 case BOND_LINK_UP:
2997 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2998
b2220cad
JV		 2999 if (!bond->curr_active_slave &&
3000 time_before_eq(jiffies, slave->dev->trans_start +
3001 delta_in_ticks)) {
3002 slave->link = BOND_LINK_UP;
3003 bond_change_active_slave(bond, slave);
3004 bond->current_arp_slave = NULL;
3005
3006 printk(KERN_INFO DRV_NAME
3007 ": %s: %s is up and now the "
3008 "active interface\n",
3009 bond->dev->name, slave->dev->name);
3010
3011 } else if (bond->curr_active_slave != slave) {
3012 /* this slave has just come up but we
3013 * already have a current slave; this can
3014 * also happen if bond_enslave adds a new
3015 * slave that is up while we are searching
3016 * for a new slave
1da177e4 3017 */
b2220cad
JV		 3018 slave->link = BOND_LINK_UP;
3019 bond_set_slave_inactive_flags(slave);
3020 bond->current_arp_slave = NULL;
3021
3022 printk(KERN_INFO DRV_NAME
3023 ": %s: backup interface %s is now up\n",
3024 bond->dev->name, slave->dev->name);
3025 }
1da177e4 3026
b2220cad 3027 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3028
b2220cad 3029 break;
1da177e4 3030
b2220cad
JV		 3031 case BOND_LINK_DOWN:
3032 if (slave->link_failure_count < UINT_MAX)
3033 slave->link_failure_count++;
3034
3035 slave->link = BOND_LINK_DOWN;
3036
3037 if (slave == bond->curr_active_slave) {
3038 printk(KERN_INFO DRV_NAME
3039 ": %s: link status down for active "
3040 "interface %s, disabling it\n",
3041 bond->dev->name, slave->dev->name);
1da177e4
LT		 3042
3043 bond_set_slave_inactive_flags(slave);
3044
b2220cad
JV		 3045 write_lock_bh(&bond->curr_slave_lock);
3046
3047 bond_select_active_slave(bond);
3048 if (bond->curr_active_slave)
3049 bond->curr_active_slave->jiffies =
3050 jiffies;
3051
3052 write_unlock_bh(&bond->curr_slave_lock);
3053
3054 bond->current_arp_slave = NULL;
3055
3056 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3057 printk(KERN_INFO DRV_NAME
3058 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3059 bond->dev->name, slave->dev->name);
3060
3061 bond_set_slave_inactive_flags(slave);
1da177e4 3062 }
b2220cad
JV		 3063 break;
3064
3065 default:
3066 printk(KERN_ERR DRV_NAME
3067 ": %s: impossible: new_link %d on slave %s\n",
3068 bond->dev->name, slave->new_link,
3069 slave->dev->name);
1da177e4
LT		 3070 }
3071 }
3072
b2220cad
JV		 3073 /*
3074 * No race with changes to primary via sysfs, as we hold rtnl.
3075 */
3076 if (bond->primary_slave &&
3077 (bond->primary_slave != bond->curr_active_slave) &&
3078 (bond->primary_slave->link == BOND_LINK_UP)) {
3079 write_lock_bh(&bond->curr_slave_lock);
3080 bond_change_active_slave(bond, bond->primary_slave);
3081 write_unlock_bh(&bond->curr_slave_lock);
3082 }
1da177e4 3083
b2220cad
JV		 3084 bond_set_carrier(bond);
3085}
1da177e4 3086
b2220cad
JV		 3087/*
3088 * Send ARP probes for active-backup mode ARP monitor.
3089 *
3090 * Called with bond->lock held for read.
3091 */
3092static void bond_ab_arp_probe(struct bonding *bond)
3093{
3094 struct slave *slave;
3095 int i;
1da177e4 3096
b2220cad 3097 read_lock(&bond->curr_slave_lock);
1da177e4 3098
b2220cad
JV		 3099 if (bond->current_arp_slave && bond->curr_active_slave)
3100 printk("PROBE: c_arp %s && cas %s BAD\n",
3101 bond->current_arp_slave->dev->name,
3102 bond->curr_active_slave->dev->name);
1da177e4 3103
b2220cad
JV		 3104 if (bond->curr_active_slave) {
3105 bond_arp_send_all(bond, bond->curr_active_slave);
3106 read_unlock(&bond->curr_slave_lock);
3107 return;
3108 }
1da177e4 3109
b2220cad 3110 read_unlock(&bond->curr_slave_lock);
059fe7a5 3111
b2220cad
JV		 3112 /* if we don't have a curr_active_slave, search for the next available
3113 * backup slave from the current_arp_slave and make it the candidate
3114 * for becoming the curr_active_slave
3115 */
1da177e4 3116
b2220cad
JV		 3117 if (!bond->current_arp_slave) {
3118 bond->current_arp_slave = bond->first_slave;
3119 if (!bond->current_arp_slave)
3120 return;
3121 }
1da177e4 3122
b2220cad 3123 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3124
b2220cad
JV		 3125 /* search for next candidate */
3126 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3127 if (IS_UP(slave->dev)) {
3128 slave->link = BOND_LINK_BACK;
3129 bond_set_slave_active_flags(slave);
3130 bond_arp_send_all(bond, slave);
1da177e4 3131 slave->jiffies = jiffies;
b2220cad
JV		 3132 bond->current_arp_slave = slave;
3133 break;
1da177e4
LT		 3134 }
3135
b2220cad
JV		 3136 /* if the link state is up at this point, we
3137 * mark it down - this can happen if we have
3138 * simultaneous link failures and
3139 * reselect_active_interface doesn't make this
3140 * one the current slave so it is still marked
3141 * up when it is actually down
1da177e4 3142 */
b2220cad
JV		 3143 if (slave->link == BOND_LINK_UP) {
3144 slave->link = BOND_LINK_DOWN;
3145 if (slave->link_failure_count < UINT_MAX)
3146 slave->link_failure_count++;
1da177e4 3147
b2220cad
JV		 3148 bond_set_slave_inactive_flags(slave);
3149
3150 printk(KERN_INFO DRV_NAME
3151 ": %s: backup interface %s is now down.\n",
3152 bond->dev->name, slave->dev->name);
1da177e4 3153 }
b2220cad
JV		 3154 }
3155}
1da177e4 3156
b2220cad
JV		 3157void bond_activebackup_arp_mon(struct work_struct *work)
3158{
3159 struct bonding *bond = container_of(work, struct bonding,
3160 arp_work.work);
3161 int delta_in_ticks;
1da177e4 3162
b2220cad 3163 read_lock(&bond->lock);
1da177e4 3164
b2220cad
JV		 3165 if (bond->kill_timers)
3166 goto out;
1da177e4 3167
b2220cad 3168 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3169
b2220cad
JV		 3170 if (bond->slave_cnt == 0)
3171 goto re_arm;
3172
b59f9f74
JV		 3173 if (bond->send_grat_arp) {
3174 read_lock(&bond->curr_slave_lock);
3175 bond_send_gratuitous_arp(bond);
3176 read_unlock(&bond->curr_slave_lock);
3177 }
3178
305d552a
BH		 3179 if (bond->send_unsol_na) {
3180 read_lock(&bond->curr_slave_lock);
3181 bond_send_unsolicited_na(bond);
3182 read_unlock(&bond->curr_slave_lock);
3183 }
3184
b2220cad
JV		 3185 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3186 read_unlock(&bond->lock);
3187 rtnl_lock();
3188 read_lock(&bond->lock);
3189
3190 bond_ab_arp_commit(bond, delta_in_ticks);
3191
3192 read_unlock(&bond->lock);
3193 rtnl_unlock();
3194 read_lock(&bond->lock);
1da177e4
LT		 3195 }
3196
b2220cad
JV		 3197 bond_ab_arp_probe(bond);
3198
1da177e4
LT		 3199re_arm:
3200 if (bond->params.arp_interval) {
1b76b316 3201 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3202 }
3203out:
3204 read_unlock(&bond->lock);
3205}
3206
3207/*------------------------------ proc/seq_file-------------------------------*/
3208
3209#ifdef CONFIG_PROC_FS
3210
1da177e4
LT		 3211static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3212{
3213 struct bonding *bond = seq->private;
3214 loff_t off = 0;
3215 struct slave *slave;
3216 int i;
3217
3218 /* make sure the bond won't be taken away */
3219 read_lock(&dev_base_lock);
6603a6f2 3220 read_lock(&bond->lock);
1da177e4
LT		 3221
3222 if (*pos == 0) {
3223 return SEQ_START_TOKEN;
3224 }
3225
3226 bond_for_each_slave(bond, slave, i) {
3227 if (++off == *pos) {
3228 return slave;
3229 }
3230 }
3231
3232 return NULL;
3233}
3234
3235static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3236{
3237 struct bonding *bond = seq->private;
3238 struct slave *slave = v;
3239
3240 ++*pos;
3241 if (v == SEQ_START_TOKEN) {
3242 return bond->first_slave;
3243 }
3244
3245 slave = slave->next;
3246
3247 return (slave == bond->first_slave) ? NULL : slave;
3248}
3249
3250static void bond_info_seq_stop(struct seq_file *seq, void *v)
3251{
3252 struct bonding *bond = seq->private;
3253
6603a6f2 3254 read_unlock(&bond->lock);
1da177e4
LT		 3255 read_unlock(&dev_base_lock);
3256}
3257
3258static void bond_info_show_master(struct seq_file *seq)
3259{
3260 struct bonding *bond = seq->private;
3261 struct slave *curr;
4756b02f 3262 int i;
1da177e4
LT		 3263
3264 read_lock(&bond->curr_slave_lock);
3265 curr = bond->curr_active_slave;
3266 read_unlock(&bond->curr_slave_lock);
3267
dd957c57 3268 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3269 bond_mode_name(bond->params.mode));
3270
dd957c57
JV		 3271 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3272 bond->params.fail_over_mac)
3915c1e8
JV		 3273 seq_printf(seq, " (fail_over_mac %s)",
3274 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3275
3276 seq_printf(seq, "\n");
3277
c61b75ad
MW		 3278 if (bond->params.mode == BOND_MODE_XOR ||
3279 bond->params.mode == BOND_MODE_8023AD) {
3280 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3281 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3282 bond->params.xmit_policy);
3283 }
3284
1da177e4
LT		 3285 if (USES_PRIMARY(bond->params.mode)) {
3286 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3287 (bond->primary_slave) ?
3288 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3289
3290 seq_printf(seq, "Currently Active Slave: %s\n",
3291 (curr) ? curr->dev->name : "None");
3292 }
3293
ff59c456
JV		 3294 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3295 "up" : "down");
1da177e4
LT		 3296 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3297 seq_printf(seq, "Up Delay (ms): %d\n",
3298 bond->params.updelay * bond->params.miimon);
3299 seq_printf(seq, "Down Delay (ms): %d\n",
3300 bond->params.downdelay * bond->params.miimon);
3301
4756b02f
MW		 3302
3303 /* ARP information */
3304 if(bond->params.arp_interval > 0) {
3305 int printed=0;
3306 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3307 bond->params.arp_interval);
3308
3309 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3310
3311 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3312 if (!bond->params.arp_targets[i])
3313 continue;
3314 if (printed)
3315 seq_printf(seq, ",");
8cf14e38 3316 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
4756b02f
MW		 3317 printed = 1;
3318 }
3319 seq_printf(seq, "\n");
3320 }
3321
1da177e4
LT		 3322 if (bond->params.mode == BOND_MODE_8023AD) {
3323 struct ad_info ad_info;
3324
3325 seq_puts(seq, "\n802.3ad info\n");
3326 seq_printf(seq, "LACP rate: %s\n",
3327 (bond->params.lacp_fast) ? "fast" : "slow");
fd989c83
JV		 3328 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3329 ad_select_tbl[bond->params.ad_select].modename);
1da177e4
LT		 3330
3331 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3332 seq_printf(seq, "bond %s has no active aggregator\n",
3333 bond->dev->name);
3334 } else {
3335 seq_printf(seq, "Active Aggregator Info:\n");
3336
3337 seq_printf(seq, "\tAggregator ID: %d\n",
3338 ad_info.aggregator_id);
3339 seq_printf(seq, "\tNumber of ports: %d\n",
3340 ad_info.ports);
3341 seq_printf(seq, "\tActor Key: %d\n",
3342 ad_info.actor_key);
3343 seq_printf(seq, "\tPartner Key: %d\n",
3344 ad_info.partner_key);
e174961c
JB		 3345 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3346 ad_info.partner_system);
1da177e4
LT		 3347 }
3348 }
3349}
3350
3351static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3352{
3353 struct bonding *bond = seq->private;
3354
3355 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3356 seq_printf(seq, "MII Status: %s\n",
3357 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3358 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3359 slave->link_failure_count);
3360
e174961c 3361 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
1da177e4
LT		 3362
3363 if (bond->params.mode == BOND_MODE_8023AD) {
3364 const struct aggregator *agg
3365 = SLAVE_AD_INFO(slave).port.aggregator;
3366
3367 if (agg) {
3368 seq_printf(seq, "Aggregator ID: %d\n",
3369 agg->aggregator_identifier);
3370 } else {
3371 seq_puts(seq, "Aggregator ID: N/A\n");
3372 }
3373 }
3374}
3375
3376static int bond_info_seq_show(struct seq_file *seq, void *v)
3377{
3378 if (v == SEQ_START_TOKEN) {
3379 seq_printf(seq, "%s\n", version);
3380 bond_info_show_master(seq);
3381 } else {
3382 bond_info_show_slave(seq, v);
3383 }
3384
3385 return 0;
3386}
3387
3388static struct seq_operations bond_info_seq_ops = {
3389 .start = bond_info_seq_start,
3390 .next = bond_info_seq_next,
3391 .stop = bond_info_seq_stop,
3392 .show = bond_info_seq_show,
3393};
3394
3395static int bond_info_open(struct inode *inode, struct file *file)
3396{
3397 struct seq_file *seq;
3398 struct proc_dir_entry *proc;
3399 int res;
3400
3401 res = seq_open(file, &bond_info_seq_ops);
3402 if (!res) {
3403 /* recover the pointer buried in proc_dir_entry data */
3404 seq = file->private_data;
3405 proc = PDE(inode);
3406 seq->private = proc->data;
3407 }
3408
3409 return res;
3410}
3411
d54b1fdb 3412static const struct file_operations bond_info_fops = {
1da177e4
LT		 3413 .owner = THIS_MODULE,
3414 .open = bond_info_open,
3415 .read = seq_read,
3416 .llseek = seq_lseek,
3417 .release = seq_release,
3418};
3419
3420static int bond_create_proc_entry(struct bonding *bond)
3421{
3422 struct net_device *bond_dev = bond->dev;
3423
3424 if (bond_proc_dir) {
a95609cb
DL		 3425 bond->proc_entry = proc_create_data(bond_dev->name,
3426 S_IRUGO, bond_proc_dir,
3427 &bond_info_fops, bond);
1da177e4
LT		 3428 if (bond->proc_entry == NULL) {
3429 printk(KERN_WARNING DRV_NAME
3430 ": Warning: Cannot create /proc/net/%s/%s\n",
3431 DRV_NAME, bond_dev->name);
3432 } else {
1da177e4
LT		 3433 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3434 }
3435 }
3436
3437 return 0;
3438}
3439
3440static void bond_remove_proc_entry(struct bonding *bond)
3441{
3442 if (bond_proc_dir && bond->proc_entry) {
3443 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3444 memset(bond->proc_file_name, 0, IFNAMSIZ);
3445 bond->proc_entry = NULL;
3446 }
3447}
3448
3449/* Create the bonding directory under /proc/net, if doesn't exist yet.
3450 * Caller must hold rtnl_lock.
3451 */
3452static void bond_create_proc_dir(void)
3453{
3454 int len = strlen(DRV_NAME);
3455
457c4cbc 3456 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3457 bond_proc_dir = bond_proc_dir->next) {
3458 if ((bond_proc_dir->namelen == len) &&
3459 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3460 break;
3461 }
3462 }
3463
3464 if (!bond_proc_dir) {
457c4cbc 3465 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3466 if (bond_proc_dir) {
3467 bond_proc_dir->owner = THIS_MODULE;
3468 } else {
3469 printk(KERN_WARNING DRV_NAME
3470 ": Warning: cannot create /proc/net/%s\n",
3471 DRV_NAME);
3472 }
3473 }
3474}
3475
3476/* Destroy the bonding directory under /proc/net, if empty.
3477 * Caller must hold rtnl_lock.
3478 */
3479static void bond_destroy_proc_dir(void)
3480{
3481 struct proc_dir_entry *de;
3482
3483 if (!bond_proc_dir) {
3484 return;
3485 }
3486
3487 /* verify that the /proc dir is empty */
3488 for (de = bond_proc_dir->subdir; de; de = de->next) {
3489 /* ignore . and .. */
3490 if (*(de->name) != '.') {
3491 break;
3492 }
3493 }
3494
3495 if (de) {
3496 if (bond_proc_dir->owner == THIS_MODULE) {
3497 bond_proc_dir->owner = NULL;
3498 }
3499 } else {
457c4cbc 3500 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3501 bond_proc_dir = NULL;
3502 }
3503}
3504#endif /* CONFIG_PROC_FS */
3505
3506/*-------------------------- netdev event handling --------------------------*/
3507
3508/*
3509 * Change device name
3510 */
3511static int bond_event_changename(struct bonding *bond)
3512{
3513#ifdef CONFIG_PROC_FS
3514 bond_remove_proc_entry(bond);
3515 bond_create_proc_entry(bond);
3516#endif
b76cdba9
MW		 3517 down_write(&(bonding_rwsem));
3518 bond_destroy_sysfs_entry(bond);
3519 bond_create_sysfs_entry(bond);
3520 up_write(&(bonding_rwsem));
1da177e4
LT		 3521 return NOTIFY_DONE;
3522}
3523
3524static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3525{
454d7c9b 3526 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 3527
3528 switch (event) {
3529 case NETDEV_CHANGENAME:
3530 return bond_event_changename(event_bond);
3531 case NETDEV_UNREGISTER:
3b96c858 3532 bond_release_all(event_bond->dev);
1da177e4
LT		 3533 break;
3534 default:
3535 break;
3536 }
3537
3538 return NOTIFY_DONE;
3539}
3540
3541static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3542{
3543 struct net_device *bond_dev = slave_dev->master;
454d7c9b 3544 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3545
3546 switch (event) {
3547 case NETDEV_UNREGISTER:
3548 if (bond_dev) {
1284cd3a
JV		 3549 if (bond->setup_by_slave)
3550 bond_release_and_destroy(bond_dev, slave_dev);
3551 else
3552 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3553 }
3554 break;
3555 case NETDEV_CHANGE:
3556 /*
3557 * TODO: is this what we get if somebody
3558 * sets up a hierarchical bond, then rmmod's
3559 * one of the slave bonding devices?
3560 */
3561 break;
3562 case NETDEV_DOWN:
3563 /*
3564 * ... Or is it this?
3565 */
3566 break;
3567 case NETDEV_CHANGEMTU:
3568 /*
3569 * TODO: Should slaves be allowed to
3570 * independently alter their MTU? For
3571 * an active-backup bond, slaves need
3572 * not be the same type of device, so
3573 * MTUs may vary. For other modes,
3574 * slaves arguably should have the
3575 * same MTUs. To do this, we'd need to
3576 * take over the slave's change_mtu
3577 * function for the duration of their
3578 * servitude.
3579 */
3580 break;
3581 case NETDEV_CHANGENAME:
3582 /*
3583 * TODO: handle changing the primary's name
3584 */
3585 break;
8531c5ff
AK		 3586 case NETDEV_FEAT_CHANGE:
3587 bond_compute_features(bond);
3588 break;
1da177e4
LT		 3589 default:
3590 break;
3591 }
3592
3593 return NOTIFY_DONE;
3594}
3595
3596/*
3597 * bond_netdev_event: handle netdev notifier chain events.
3598 *
3599 * This function receives events for the netdev chain. The caller (an
e041c683 3600 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3601 * locks for us to safely manipulate the slave devices (RTNL lock,
3602 * dev_probe_lock).
3603 */
3604static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3605{
3606 struct net_device *event_dev = (struct net_device *)ptr;
3607
c346dca1 3608 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3609 return NOTIFY_DONE;
3610
1da177e4
LT		 3611 dprintk("event_dev: %s, event: %lx\n",
3612 (event_dev ? event_dev->name : "None"),
3613 event);
3614
0b680e75
JV		 3615 if (!(event_dev->priv_flags & IFF_BONDING))
3616 return NOTIFY_DONE;
3617
1da177e4
LT		 3618 if (event_dev->flags & IFF_MASTER) {
3619 dprintk("IFF_MASTER\n");
3620 return bond_master_netdev_event(event, event_dev);
3621 }
3622
3623 if (event_dev->flags & IFF_SLAVE) {
3624 dprintk("IFF_SLAVE\n");
3625 return bond_slave_netdev_event(event, event_dev);
3626 }
3627
3628 return NOTIFY_DONE;
3629}
3630
c3ade5ca
JV		 3631/*
3632 * bond_inetaddr_event: handle inetaddr notifier chain events.
3633 *
3634 * We keep track of device IPs primarily to use as source addresses in
3635 * ARP monitor probes (rather than spewing out broadcasts all the time).
3636 *
3637 * We track one IP for the main device (if it has one), plus one per VLAN.
3638 */
3639static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3640{
3641 struct in_ifaddr *ifa = ptr;
3642 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3643 struct bonding *bond;
3644 struct vlan_entry *vlan;
c3ade5ca 3645
c346dca1 3646 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3647 return NOTIFY_DONE;
3648
0883beca 3649 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3650 if (bond->dev == event_dev) {
3651 switch (event) {
3652 case NETDEV_UP:
3653 bond->master_ip = ifa->ifa_local;
3654 return NOTIFY_OK;
3655 case NETDEV_DOWN:
3656 bond->master_ip = bond_glean_dev_ip(bond->dev);
3657 return NOTIFY_OK;
3658 default:
3659 return NOTIFY_DONE;
3660 }
3661 }
3662
0883beca 3663 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3664 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3665 if (vlan_dev == event_dev) {
3666 switch (event) {
3667 case NETDEV_UP:
3668 vlan->vlan_ip = ifa->ifa_local;
3669 return NOTIFY_OK;
3670 case NETDEV_DOWN:
3671 vlan->vlan_ip =
3672 bond_glean_dev_ip(vlan_dev);
3673 return NOTIFY_OK;
3674 default:
3675 return NOTIFY_DONE;
3676 }
3677 }
3678 }
3679 }
3680 return NOTIFY_DONE;
3681}
3682
1da177e4
LT		 3683static struct notifier_block bond_netdev_notifier = {
3684 .notifier_call = bond_netdev_event,
3685};
3686
c3ade5ca
JV		 3687static struct notifier_block bond_inetaddr_notifier = {
3688 .notifier_call = bond_inetaddr_event,
3689};
3690
1da177e4
LT		 3691/*-------------------------- Packet type handling ---------------------------*/
3692
3693/* register to receive lacpdus on a bond */
3694static void bond_register_lacpdu(struct bonding *bond)
3695{
3696 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3697
3698 /* initialize packet type */
3699 pk_type->type = PKT_TYPE_LACPDU;
3700 pk_type->dev = bond->dev;
3701 pk_type->func = bond_3ad_lacpdu_recv;
3702
3703 dev_add_pack(pk_type);
3704}
3705
3706/* unregister to receive lacpdus on a bond */
3707static void bond_unregister_lacpdu(struct bonding *bond)
3708{
3709 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3710}
3711
f5b2b966
JV		 3712void bond_register_arp(struct bonding *bond)
3713{
3714 struct packet_type *pt = &bond->arp_mon_pt;
3715
c4f283b1
JV		 3716 if (pt->type)
3717 return;
3718
f5b2b966 3719 pt->type = htons(ETH_P_ARP);
e245cb71 3720 pt->dev = bond->dev;
f5b2b966
JV		 3721 pt->func = bond_arp_rcv;
3722 dev_add_pack(pt);
3723}
3724
3725void bond_unregister_arp(struct bonding *bond)
3726{
c4f283b1
JV		 3727 struct packet_type *pt = &bond->arp_mon_pt;
3728
3729 dev_remove_pack(pt);
3730 pt->type = 0;
f5b2b966
JV		 3731}
3732
169a3e66
JV		 3733/*---------------------------- Hashing Policies -----------------------------*/
3734
6f6652be
JV		 3735/*
3736 * Hash for the output device based upon layer 2 and layer 3 data. If
3737 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3738 */
3739static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3740 struct net_device *bond_dev, int count)
3741{
3742 struct ethhdr *data = (struct ethhdr *)skb->data;
3743 struct iphdr *iph = ip_hdr(skb);
3744
f14c4e4e 3745 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be
JV		 3746 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3747 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3748 }
3749
3750 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3751}
3752
169a3e66 3753/*
59c51591 3754 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3755 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3756 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3757 */
3758static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3759 struct net_device *bond_dev, int count)
3760{
3761 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3762 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3763 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3764 int layer4_xor = 0;
3765
f14c4e4e
BH		 3766 if (skb->protocol == htons(ETH_P_IP)) {
3767 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3768 (iph->protocol == IPPROTO_TCP ||
3769 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3770 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3771 }
3772 return (layer4_xor ^
3773 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3774
3775 }
3776
3777 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3778}
3779
3780/*
3781 * Hash for the output device based upon layer 2 data
3782 */
3783static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3784 struct net_device *bond_dev, int count)
3785{
3786 struct ethhdr *data = (struct ethhdr *)skb->data;
3787
3788 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3789}
3790
1da177e4
LT		 3791/*-------------------------- Device entry points ----------------------------*/
3792
3793static int bond_open(struct net_device *bond_dev)
3794{
454d7c9b 3795 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3796
3797 bond->kill_timers = 0;
3798
3799 if ((bond->params.mode == BOND_MODE_TLB) ||
3800 (bond->params.mode == BOND_MODE_ALB)) {
1da177e4
LT		 3801 /* bond_alb_initialize must be called before the timer
3802 * is started.
3803 */
3804 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3805 /* something went wrong - fail the open operation */
3806 return -1;
3807 }
3808
1b76b316
JV		 3809 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3810 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3811 }
3812
3813 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3814 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3815 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3816 }
3817
3818 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3819 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3820 INIT_DELAYED_WORK(&bond->arp_work,
3821 bond_activebackup_arp_mon);
3822 else
3823 INIT_DELAYED_WORK(&bond->arp_work,
3824 bond_loadbalance_arp_mon);
3825
3826 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3827 if (bond->params.arp_validate)
3828 bond_register_arp(bond);
1da177e4
LT		 3829 }
3830
3831 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3832 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3833 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3834 /* register to receive LACPDUs */
3835 bond_register_lacpdu(bond);
fd989c83 3836 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3837 }
3838
3839 return 0;
3840}
3841
3842static int bond_close(struct net_device *bond_dev)
3843{
454d7c9b 3844 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3845
3846 if (bond->params.mode == BOND_MODE_8023AD) {
3847 /* Unregister the receive of LACPDUs */
3848 bond_unregister_lacpdu(bond);
3849 }
3850
f5b2b966
JV		 3851 if (bond->params.arp_validate)
3852 bond_unregister_arp(bond);
3853
1da177e4
LT		 3854 write_lock_bh(&bond->lock);
3855
b59f9f74 3856 bond->send_grat_arp = 0;
305d552a 3857 bond->send_unsol_na = 0;
1da177e4
LT		 3858
3859 /* signal timers not to re-arm */
3860 bond->kill_timers = 1;
3861
3862 write_unlock_bh(&bond->lock);
3863
1da177e4 3864 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3865 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3866 }
3867
3868 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3869 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3870 }
3871
3872 switch (bond->params.mode) {
3873 case BOND_MODE_8023AD:
1b76b316 3874 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3875 break;
3876 case BOND_MODE_TLB:
3877 case BOND_MODE_ALB:
1b76b316 3878 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3879 break;
3880 default:
3881 break;
3882 }
3883
1da177e4
LT		 3884
3885 if ((bond->params.mode == BOND_MODE_TLB) ||
3886 (bond->params.mode == BOND_MODE_ALB)) {
3887 /* Must be called only after all
3888 * slaves have been released
3889 */
3890 bond_alb_deinitialize(bond);
3891 }
3892
3893 return 0;
3894}
3895
3896static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3897{
454d7c9b 3898 struct bonding *bond = netdev_priv(bond_dev);
eeda3fd6 3899 struct net_device_stats *stats = &bond->stats;
2439f9eb 3900 struct net_device_stats local_stats;
1da177e4
LT		 3901 struct slave *slave;
3902 int i;
3903
2439f9eb 3904 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3905
3906 read_lock_bh(&bond->lock);
3907
3908 bond_for_each_slave(bond, slave, i) {
eeda3fd6
SH		 3909 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3910
2439f9eb
AG		 3911 local_stats.rx_packets += sstats->rx_packets;
3912 local_stats.rx_bytes += sstats->rx_bytes;
3913 local_stats.rx_errors += sstats->rx_errors;
3914 local_stats.rx_dropped += sstats->rx_dropped;
3915
3916 local_stats.tx_packets += sstats->tx_packets;
3917 local_stats.tx_bytes += sstats->tx_bytes;
3918 local_stats.tx_errors += sstats->tx_errors;
3919 local_stats.tx_dropped += sstats->tx_dropped;
3920
3921 local_stats.multicast += sstats->multicast;
3922 local_stats.collisions += sstats->collisions;
3923
3924 local_stats.rx_length_errors += sstats->rx_length_errors;
3925 local_stats.rx_over_errors += sstats->rx_over_errors;
3926 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3927 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3928 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3929 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3930
3931 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3932 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3933 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3934 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3935 local_stats.tx_window_errors += sstats->tx_window_errors;
3936 }
3937
3938 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3939
3940 read_unlock_bh(&bond->lock);
3941
3942 return stats;
3943}
3944
3945static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3946{
3947 struct net_device *slave_dev = NULL;
3948 struct ifbond k_binfo;
3949 struct ifbond __user *u_binfo = NULL;
3950 struct ifslave k_sinfo;
3951 struct ifslave __user *u_sinfo = NULL;
3952 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3953 int res = 0;
3954
3955 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3956 bond_dev->name, cmd);
3957
3958 switch (cmd) {
1da177e4
LT		 3959 case SIOCGMIIPHY:
3960 mii = if_mii(ifr);
3961 if (!mii) {
3962 return -EINVAL;
3963 }
3964 mii->phy_id = 0;
3965 /* Fall Through */
3966 case SIOCGMIIREG:
3967 /*
3968 * We do this again just in case we were called by SIOCGMIIREG
3969 * instead of SIOCGMIIPHY.
3970 */
3971 mii = if_mii(ifr);
3972 if (!mii) {
3973 return -EINVAL;
3974 }
3975
3976 if (mii->reg_num == 1) {
454d7c9b 3977 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3978 mii->val_out = 0;
6603a6f2 3979 read_lock(&bond->lock);
1da177e4 3980 read_lock(&bond->curr_slave_lock);
4e140079 3981 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3982 mii->val_out = BMSR_LSTATUS;
3983 }
3984 read_unlock(&bond->curr_slave_lock);
6603a6f2 3985 read_unlock(&bond->lock);
1da177e4
LT		 3986 }
3987
3988 return 0;
3989 case BOND_INFO_QUERY_OLD:
3990 case SIOCBONDINFOQUERY:
3991 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3992
3993 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3994 return -EFAULT;
3995 }
3996
3997 res = bond_info_query(bond_dev, &k_binfo);
3998 if (res == 0) {
3999 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
4000 return -EFAULT;
4001 }
4002 }
4003
4004 return res;
4005 case BOND_SLAVE_INFO_QUERY_OLD:
4006 case SIOCBONDSLAVEINFOQUERY:
4007 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4008
4009 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4010 return -EFAULT;
4011 }
4012
4013 res = bond_slave_info_query(bond_dev, &k_sinfo);
4014 if (res == 0) {
4015 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4016 return -EFAULT;
4017 }
4018 }
4019
4020 return res;
4021 default:
4022 /* Go on */
4023 break;
4024 }
4025
4026 if (!capable(CAP_NET_ADMIN)) {
4027 return -EPERM;
4028 }
4029
b76cdba9 4030 down_write(&(bonding_rwsem));
881d966b 4031 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT		 4032
4033 dprintk("slave_dev=%p: \n", slave_dev);
4034
4035 if (!slave_dev) {
4036 res = -ENODEV;
4037 } else {
4038 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4039 switch (cmd) {
4040 case BOND_ENSLAVE_OLD:
4041 case SIOCBONDENSLAVE:
4042 res = bond_enslave(bond_dev, slave_dev);
4043 break;
4044 case BOND_RELEASE_OLD:
4045 case SIOCBONDRELEASE:
4046 res = bond_release(bond_dev, slave_dev);
4047 break;
4048 case BOND_SETHWADDR_OLD:
4049 case SIOCBONDSETHWADDR:
4050 res = bond_sethwaddr(bond_dev, slave_dev);
4051 break;
4052 case BOND_CHANGE_ACTIVE_OLD:
4053 case SIOCBONDCHANGEACTIVE:
4054 res = bond_ioctl_change_active(bond_dev, slave_dev);
4055 break;
4056 default:
4057 res = -EOPNOTSUPP;
4058 }
4059
4060 dev_put(slave_dev);
4061 }
4062
b76cdba9 4063 up_write(&(bonding_rwsem));
1da177e4
LT		 4064 return res;
4065}
4066
4067static void bond_set_multicast_list(struct net_device *bond_dev)
4068{
454d7c9b 4069 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4070 struct dev_mc_list *dmi;
4071
1da177e4
LT		 4072 /*
4073 * Do promisc before checking multicast_mode
4074 */
4075 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4076 /*
4077 * FIXME: Need to handle the error when one of the multi-slaves
4078 * encounters error.
4079 */
1da177e4
LT		 4080 bond_set_promiscuity(bond, 1);
4081 }
4082
4083 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4084 bond_set_promiscuity(bond, -1);
4085 }
4086
4087 /* set allmulti flag to slaves */
4088 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4089 /*
4090 * FIXME: Need to handle the error when one of the multi-slaves
4091 * encounters error.
4092 */
1da177e4
LT		 4093 bond_set_allmulti(bond, 1);
4094 }
4095
4096 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4097 bond_set_allmulti(bond, -1);
4098 }
4099
80ee5ad2
JV		 4100 read_lock(&bond->lock);
4101
1da177e4
LT		 4102 bond->flags = bond_dev->flags;
4103
4104 /* looking for addresses to add to slaves' mc list */
4105 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4106 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4107 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4108 }
4109 }
4110
4111 /* looking for addresses to delete from slaves' list */
4112 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4113 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4114 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4115 }
4116 }
4117
4118 /* save master's multicast list */
4119 bond_mc_list_destroy(bond);
4120 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4121
80ee5ad2 4122 read_unlock(&bond->lock);
1da177e4
LT		 4123}
4124
00829823
SH		 4125static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4126{
4127 struct bonding *bond = netdev_priv(dev);
4128 struct slave *slave = bond->first_slave;
4129
4130 if (slave) {
4131 const struct net_device_ops *slave_ops
4132 = slave->dev->netdev_ops;
4133 if (slave_ops->ndo_neigh_setup)
4134 return slave_ops->ndo_neigh_setup(dev, parms);
4135 }
4136 return 0;
4137}
4138
1da177e4
LT		 4139/*
4140 * Change the MTU of all of a master's slaves to match the master
4141 */
4142static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4143{
454d7c9b 4144 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4145 struct slave *slave, *stop_at;
4146 int res = 0;
4147 int i;
4148
4149 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4150 (bond_dev ? bond_dev->name : "None"), new_mtu);
4151
4152 /* Can't hold bond->lock with bh disabled here since
4153 * some base drivers panic. On the other hand we can't
4154 * hold bond->lock without bh disabled because we'll
4155 * deadlock. The only solution is to rely on the fact
4156 * that we're under rtnl_lock here, and the slaves
4157 * list won't change. This doesn't solve the problem
4158 * of setting the slave's MTU while it is
4159 * transmitting, but the assumption is that the base
4160 * driver can handle that.
4161 *
4162 * TODO: figure out a way to safely iterate the slaves
4163 * list, but without holding a lock around the actual
4164 * call to the base driver.
4165 */
4166
4167 bond_for_each_slave(bond, slave, i) {
4168 dprintk("s %p s->p %p c_m %p\n", slave,
4169 slave->prev, slave->dev->change_mtu);
e944ef79 4170
1da177e4
LT		 4171 res = dev_set_mtu(slave->dev, new_mtu);
4172
4173 if (res) {
4174 /* If we failed to set the slave's mtu to the new value
4175 * we must abort the operation even in ACTIVE_BACKUP
4176 * mode, because if we allow the backup slaves to have
4177 * different mtu values than the active slave we'll
4178 * need to change their mtu when doing a failover. That
4179 * means changing their mtu from timer context, which
4180 * is probably not a good idea.
4181 */
4182 dprintk("err %d %s\n", res, slave->dev->name);
4183 goto unwind;
4184 }
4185 }
4186
4187 bond_dev->mtu = new_mtu;
4188
4189 return 0;
4190
4191unwind:
4192 /* unwind from head to the slave that failed */
4193 stop_at = slave;
4194 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4195 int tmp_res;
4196
4197 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4198 if (tmp_res) {
4199 dprintk("unwind err %d dev %s\n", tmp_res,
4200 slave->dev->name);
4201 }
4202 }
4203
4204 return res;
4205}
4206
4207/*
4208 * Change HW address
4209 *
4210 * Note that many devices must be down to change the HW address, and
4211 * downing the master releases all slaves. We can make bonds full of
4212 * bonding devices to test this, however.
4213 */
4214static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4215{
454d7c9b 4216 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4217 struct sockaddr *sa = addr, tmp_sa;
4218 struct slave *slave, *stop_at;
4219 int res = 0;
4220 int i;
4221
eb7cc59a
SH		 4222 if (bond->params.mode == BOND_MODE_ALB)
4223 return bond_alb_set_mac_address(bond_dev, addr);
4224
4225
1da177e4
LT		 4226 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4227
dd957c57 4228 /*
3915c1e8
JV		 4229 * If fail_over_mac is set to active, do nothing and return
4230 * success. Returning an error causes ifenslave to fail.
dd957c57 4231 */
3915c1e8 4232 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4233 return 0;
2ab82852 4234
1da177e4
LT		 4235 if (!is_valid_ether_addr(sa->sa_data)) {
4236 return -EADDRNOTAVAIL;
4237 }
4238
4239 /* Can't hold bond->lock with bh disabled here since
4240 * some base drivers panic. On the other hand we can't
4241 * hold bond->lock without bh disabled because we'll
4242 * deadlock. The only solution is to rely on the fact
4243 * that we're under rtnl_lock here, and the slaves
4244 * list won't change. This doesn't solve the problem
4245 * of setting the slave's hw address while it is
4246 * transmitting, but the assumption is that the base
4247 * driver can handle that.
4248 *
4249 * TODO: figure out a way to safely iterate the slaves
4250 * list, but without holding a lock around the actual
4251 * call to the base driver.
4252 */
4253
4254 bond_for_each_slave(bond, slave, i) {
eb7cc59a 4255 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
1da177e4
LT		 4256 dprintk("slave %p %s\n", slave, slave->dev->name);
4257
eb7cc59a 4258 if (slave_ops->ndo_set_mac_address == NULL) {
1da177e4
LT		 4259 res = -EOPNOTSUPP;
4260 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4261 goto unwind;
4262 }
4263
4264 res = dev_set_mac_address(slave->dev, addr);
4265 if (res) {
4266 /* TODO: consider downing the slave
4267 * and retry ?
4268 * User should expect communications
4269 * breakage anyway until ARP finish
4270 * updating, so...
4271 */
4272 dprintk("err %d %s\n", res, slave->dev->name);
4273 goto unwind;
4274 }
4275 }
4276
4277 /* success */
4278 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4279 return 0;
4280
4281unwind:
4282 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4283 tmp_sa.sa_family = bond_dev->type;
4284
4285 /* unwind from head to the slave that failed */
4286 stop_at = slave;
4287 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4288 int tmp_res;
4289
4290 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4291 if (tmp_res) {
4292 dprintk("unwind err %d dev %s\n", tmp_res,
4293 slave->dev->name);
4294 }
4295 }
4296
4297 return res;
4298}
4299
4300static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4301{
454d7c9b 4302 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4303 struct slave *slave, *start_at;
cf5f9044 4304 int i, slave_no, res = 1;
1da177e4
LT		 4305
4306 read_lock(&bond->lock);
4307
4308 if (!BOND_IS_OK(bond)) {
4309 goto out;
4310 }
4311
cf5f9044
JV		 4312 /*
4313 * Concurrent TX may collide on rr_tx_counter; we accept that
4314 * as being rare enough not to justify using an atomic op here
4315 */
4316 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4317
cf5f9044
JV		 4318 bond_for_each_slave(bond, slave, i) {
4319 slave_no--;
4320 if (slave_no < 0) {
4321 break;
4322 }
1da177e4
LT		 4323 }
4324
cf5f9044 4325 start_at = slave;
1da177e4
LT		 4326 bond_for_each_slave_from(bond, slave, i, start_at) {
4327 if (IS_UP(slave->dev) &&
4328 (slave->link == BOND_LINK_UP) &&
4329 (slave->state == BOND_STATE_ACTIVE)) {
4330 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4331 break;
4332 }
4333 }
4334
1da177e4
LT		 4335out:
4336 if (res) {
4337 /* no suitable interface, frame not sent */
4338 dev_kfree_skb(skb);
4339 }
4340 read_unlock(&bond->lock);
4341 return 0;
4342}
4343
075897ce 4344
1da177e4
LT		 4345/*
4346 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4347 * the bond has a usable interface.
4348 */
4349static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4350{
454d7c9b 4351 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4352 int res = 1;
4353
1da177e4
LT		 4354 read_lock(&bond->lock);
4355 read_lock(&bond->curr_slave_lock);
4356
4357 if (!BOND_IS_OK(bond)) {
4358 goto out;
4359 }
4360
075897ce
JL		 4361 if (!bond->curr_active_slave)
4362 goto out;
4363
075897ce
JL		 4364 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4365
1da177e4
LT		 4366out:
4367 if (res) {
4368 /* no suitable interface, frame not sent */
4369 dev_kfree_skb(skb);
4370 }
4371 read_unlock(&bond->curr_slave_lock);
4372 read_unlock(&bond->lock);
4373 return 0;
4374}
4375
4376/*
169a3e66
JV		 4377 * In bond_xmit_xor() , we determine the output device by using a pre-
4378 * determined xmit_hash_policy(), If the selected device is not enabled,
4379 * find the next active slave.
1da177e4
LT		 4380 */
4381static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4382{
454d7c9b 4383 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4384 struct slave *slave, *start_at;
4385 int slave_no;
4386 int i;
4387 int res = 1;
4388
4389 read_lock(&bond->lock);
4390
4391 if (!BOND_IS_OK(bond)) {
4392 goto out;
4393 }
4394
169a3e66 4395 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4396
4397 bond_for_each_slave(bond, slave, i) {
4398 slave_no--;
4399 if (slave_no < 0) {
4400 break;
4401 }
4402 }
4403
4404 start_at = slave;
4405
4406 bond_for_each_slave_from(bond, slave, i, start_at) {
4407 if (IS_UP(slave->dev) &&
4408 (slave->link == BOND_LINK_UP) &&
4409 (slave->state == BOND_STATE_ACTIVE)) {
4410 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4411 break;
4412 }
4413 }
4414
4415out:
4416 if (res) {
4417 /* no suitable interface, frame not sent */
4418 dev_kfree_skb(skb);
4419 }
4420 read_unlock(&bond->lock);
4421 return 0;
4422}
4423
4424/*
4425 * in broadcast mode, we send everything to all usable interfaces.
4426 */
4427static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4428{
454d7c9b 4429 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4430 struct slave *slave, *start_at;
4431 struct net_device *tx_dev = NULL;
4432 int i;
4433 int res = 1;
4434
4435 read_lock(&bond->lock);
4436
4437 if (!BOND_IS_OK(bond)) {
4438 goto out;
4439 }
4440
4441 read_lock(&bond->curr_slave_lock);
4442 start_at = bond->curr_active_slave;
4443 read_unlock(&bond->curr_slave_lock);
4444
4445 if (!start_at) {
4446 goto out;
4447 }
4448
4449 bond_for_each_slave_from(bond, slave, i, start_at) {
4450 if (IS_UP(slave->dev) &&
4451 (slave->link == BOND_LINK_UP) &&
4452 (slave->state == BOND_STATE_ACTIVE)) {
4453 if (tx_dev) {
4454 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4455 if (!skb2) {
4456 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4457 ": %s: Error: bond_xmit_broadcast(): "
4458 "skb_clone() failed\n",
4459 bond_dev->name);
1da177e4
LT		 4460 continue;
4461 }
4462
4463 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4464 if (res) {
4465 dev_kfree_skb(skb2);
4466 continue;
4467 }
4468 }
4469 tx_dev = slave->dev;
4470 }
4471 }
4472
4473 if (tx_dev) {
4474 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4475 }
4476
4477out:
4478 if (res) {
4479 /* no suitable interface, frame not sent */
4480 dev_kfree_skb(skb);
4481 }
4482 /* frame sent to all suitable interfaces */
4483 read_unlock(&bond->lock);
4484 return 0;
4485}
4486
4487/*------------------------- Device initialization ---------------------------*/
4488
6f6652be
JV		 4489static void bond_set_xmit_hash_policy(struct bonding *bond)
4490{
4491 switch (bond->params.xmit_policy) {
4492 case BOND_XMIT_POLICY_LAYER23:
4493 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4494 break;
4495 case BOND_XMIT_POLICY_LAYER34:
4496 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4497 break;
4498 case BOND_XMIT_POLICY_LAYER2:
4499 default:
4500 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4501 break;
4502 }
4503}
4504
00829823
SH		 4505static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4506{
4507 const struct bonding *bond = netdev_priv(dev);
4508
4509 switch (bond->params.mode) {
4510 case BOND_MODE_ROUNDROBIN:
4511 return bond_xmit_roundrobin(skb, dev);
4512 case BOND_MODE_ACTIVEBACKUP:
4513 return bond_xmit_activebackup(skb, dev);
4514 case BOND_MODE_XOR:
4515 return bond_xmit_xor(skb, dev);
4516 case BOND_MODE_BROADCAST:
4517 return bond_xmit_broadcast(skb, dev);
4518 case BOND_MODE_8023AD:
4519 return bond_3ad_xmit_xor(skb, dev);
4520 case BOND_MODE_ALB:
4521 case BOND_MODE_TLB:
4522 return bond_alb_xmit(skb, dev);
4523 default:
4524 /* Should never happen, mode already checked */
4525 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4526 dev->name, bond->params.mode);
4527 WARN_ON_ONCE(1);
4528 dev_kfree_skb(skb);
4529 return NETDEV_TX_OK;
4530 }
4531}
4532
4533
1da177e4
LT		 4534/*
4535 * set bond mode specific net device operations
4536 */
a77b5325 4537void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4538{
169a3e66
JV		 4539 struct net_device *bond_dev = bond->dev;
4540
1da177e4
LT		 4541 switch (mode) {
4542 case BOND_MODE_ROUNDROBIN:
1da177e4
LT		 4543 break;
4544 case BOND_MODE_ACTIVEBACKUP:
1da177e4
LT		 4545 break;
4546 case BOND_MODE_XOR:
6f6652be 4547 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4548 break;
4549 case BOND_MODE_BROADCAST:
1da177e4
LT		 4550 break;
4551 case BOND_MODE_8023AD:
8f903c70 4552 bond_set_master_3ad_flags(bond);
6f6652be 4553 bond_set_xmit_hash_policy(bond);
1da177e4 4554 break;
1da177e4 4555 case BOND_MODE_ALB:
8f903c70
JV		 4556 bond_set_master_alb_flags(bond);
4557 /* FALLTHRU */
4558 case BOND_MODE_TLB:
1da177e4
LT		 4559 break;
4560 default:
4561 /* Should never happen, mode already checked */
4562 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4563 ": %s: Error: Unknown bonding mode %d\n",
4564 bond_dev->name,
1da177e4
LT		 4565 mode);
4566 break;
4567 }
4568}
4569
217df670
JV		 4570static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4571 struct ethtool_drvinfo *drvinfo)
4572{
4573 strncpy(drvinfo->driver, DRV_NAME, 32);
4574 strncpy(drvinfo->version, DRV_VERSION, 32);
4575 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4576}
4577
7282d491 4578static const struct ethtool_ops bond_ethtool_ops = {
217df670 4579 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac
SH		 4580 .get_link = ethtool_op_get_link,
4581 .get_tx_csum = ethtool_op_get_tx_csum,
4582 .get_sg = ethtool_op_get_sg,
4583 .get_tso = ethtool_op_get_tso,
4584 .get_ufo = ethtool_op_get_ufo,
4585 .get_flags = ethtool_op_get_flags,
8531c5ff
AK		 4586};
4587
eb7cc59a
SH		 4588static const struct net_device_ops bond_netdev_ops = {
4589 .ndo_open = bond_open,
4590 .ndo_stop = bond_close,
00829823 4591 .ndo_start_xmit = bond_start_xmit,
eb7cc59a
SH		 4592 .ndo_get_stats = bond_get_stats,
4593 .ndo_do_ioctl = bond_do_ioctl,
4594 .ndo_set_multicast_list = bond_set_multicast_list,
4595 .ndo_change_mtu = bond_change_mtu,
eb7cc59a 4596 .ndo_set_mac_address = bond_set_mac_address,
00829823 4597 .ndo_neigh_setup = bond_neigh_setup,
eb7cc59a
SH		 4598 .ndo_vlan_rx_register = bond_vlan_rx_register,
4599 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4600 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4601};
4602
1da177e4
LT		 4603/*
4604 * Does not allocate but creates a /proc entry.
4605 * Allowed to fail.
4606 */
3c535952 4607static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4 4608{
454d7c9b 4609 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4610
4611 dprintk("Begin bond_init for %s\n", bond_dev->name);
4612
4613 /* initialize rwlocks */
4614 rwlock_init(&bond->lock);
4615 rwlock_init(&bond->curr_slave_lock);
4616
4617 bond->params = *params; /* copy params struct */
4618
1b76b316
JV		 4619 bond->wq = create_singlethread_workqueue(bond_dev->name);
4620 if (!bond->wq)
4621 return -ENOMEM;
4622
1da177e4
LT		 4623 /* Initialize pointers */
4624 bond->first_slave = NULL;
4625 bond->curr_active_slave = NULL;
4626 bond->current_arp_slave = NULL;
4627 bond->primary_slave = NULL;
4628 bond->dev = bond_dev;
1053f62c 4629 bond->send_grat_arp = 0;
305d552a 4630 bond->send_unsol_na = 0;
d90a162a 4631 bond->setup_by_slave = 0;
1da177e4
LT		 4632 INIT_LIST_HEAD(&bond->vlan_list);
4633
4634 /* Initialize the device entry points */
eb7cc59a 4635 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 4636 bond_dev->ethtool_ops = &bond_ethtool_ops;
169a3e66 4637 bond_set_mode_ops(bond, bond->params.mode);
1da177e4 4638
a434e43f 4639 bond_dev->destructor = bond_destructor;
1da177e4
LT		 4640
4641 /* Initialize the device options */
4642 bond_dev->tx_queue_len = 0;
4643 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4644 bond_dev->priv_flags |= IFF_BONDING;
6cf3f41e
JV		 4645 if (bond->params.arp_interval)
4646 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
1da177e4
LT		 4647
4648 /* At first, we block adding VLANs. That's the only way to
4649 * prevent problems that occur when adding VLANs over an
4650 * empty bond. The block will be removed once non-challenged
4651 * slaves are enslaved.
4652 */
4653 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4654
932ff279 4655 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4656 * transmitting */
4657 bond_dev->features |= NETIF_F_LLTX;
4658
4659 /* By default, we declare the bond to be fully
4660 * VLAN hardware accelerated capable. Special
4661 * care is taken in the various xmit functions
4662 * when there are slaves that are not hw accel
4663 * capable
4664 */
1da177e4
LT		 4665 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4666 NETIF_F_HW_VLAN_RX |
4667 NETIF_F_HW_VLAN_FILTER);
4668
4669#ifdef CONFIG_PROC_FS
4670 bond_create_proc_entry(bond);
4671#endif
1da177e4
LT		 4672 list_add_tail(&bond->bond_list, &bond_dev_list);
4673
4674 return 0;
4675}
4676
fdaea7a9
JV		 4677static void bond_work_cancel_all(struct bonding *bond)
4678{
4679 write_lock_bh(&bond->lock);
4680 bond->kill_timers = 1;
4681 write_unlock_bh(&bond->lock);
4682
4683 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4684 cancel_delayed_work(&bond->mii_work);
4685
4686 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4687 cancel_delayed_work(&bond->arp_work);
4688
4689 if (bond->params.mode == BOND_MODE_ALB &&
4690 delayed_work_pending(&bond->alb_work))
4691 cancel_delayed_work(&bond->alb_work);
4692
4693 if (bond->params.mode == BOND_MODE_8023AD &&
4694 delayed_work_pending(&bond->ad_work))
4695 cancel_delayed_work(&bond->ad_work);
4696}
4697
a434e43f
JV		 4698/* De-initialize device specific data.
4699 * Caller must hold rtnl_lock.
4700 */
4701static void bond_deinit(struct net_device *bond_dev)
4702{
454d7c9b 4703 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 4704
4705 list_del(&bond->bond_list);
4706
4707 bond_work_cancel_all(bond);
4708
4709#ifdef CONFIG_PROC_FS
4710 bond_remove_proc_entry(bond);
4711#endif
4712}
4713
1da177e4
LT		 4714/* Unregister and free all bond devices.
4715 * Caller must hold rtnl_lock.
4716 */
4717static void bond_free_all(void)
4718{
4719 struct bonding *bond, *nxt;
4720
4721 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4722 struct net_device *bond_dev = bond->dev;
4723
fdaea7a9 4724 bond_work_cancel_all(bond);
70298705 4725 /* Release the bonded slaves */
4726 bond_release_all(bond_dev);
92b41daa 4727 bond_destroy(bond);
1da177e4
LT		 4728 }
4729
4730#ifdef CONFIG_PROC_FS
4731 bond_destroy_proc_dir();
4732#endif
4733}
4734
4735/*------------------------- Module initialization ---------------------------*/
4736
4737/*
4738 * Convert string input module parms. Accept either the
ece95f7f
JV		 4739 * number of the mode or its string name. A bit complicated because
4740 * some mode names are substrings of other names, and calls from sysfs
4741 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4742 */
ece95f7f 4743int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4744{
ece95f7f 4745 int mode = -1, i, rv;
a42e534f 4746 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4747
a42e534f
JV		 4748 for (p = (char *)buf; *p; p++)
4749 if (!(isdigit(*p) || isspace(*p)))
4750 break;
4751
4752 if (*p)
ece95f7f 4753 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4754 else
4755 rv = sscanf(buf, "%d", &mode);
4756
4757 if (!rv)
4758 return -1;
1da177e4
LT		 4759
4760 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4761 if (mode == tbl[i].mode)
4762 return tbl[i].mode;
4763 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4764 return tbl[i].mode;
1da177e4
LT		 4765 }
4766
4767 return -1;
4768}
4769
4770static int bond_check_params(struct bond_params *params)
4771{
3915c1e8 4772 int arp_validate_value, fail_over_mac_value;
f5b2b966 4773
1da177e4
LT		 4774 /*
4775 * Convert string parameters.
4776 */
4777 if (mode) {
4778 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4779 if (bond_mode == -1) {
4780 printk(KERN_ERR DRV_NAME
4781 ": Error: Invalid bonding mode \"%s\"\n",
4782 mode == NULL ? "NULL" : mode);
4783 return -EINVAL;
4784 }
4785 }
4786
169a3e66
JV		 4787 if (xmit_hash_policy) {
4788 if ((bond_mode != BOND_MODE_XOR) &&
4789 (bond_mode != BOND_MODE_8023AD)) {
4790 printk(KERN_INFO DRV_NAME
4791 ": xor_mode param is irrelevant in mode %s\n",
4792 bond_mode_name(bond_mode));
4793 } else {
4794 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4795 xmit_hashtype_tbl);
4796 if (xmit_hashtype == -1) {
4797 printk(KERN_ERR DRV_NAME
4798 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4799 xmit_hash_policy == NULL ? "NULL" :
4800 xmit_hash_policy);
4801 return -EINVAL;
4802 }
4803 }
4804 }
4805
1da177e4
LT		 4806 if (lacp_rate) {
4807 if (bond_mode != BOND_MODE_8023AD) {
4808 printk(KERN_INFO DRV_NAME
4809 ": lacp_rate param is irrelevant in mode %s\n",
4810 bond_mode_name(bond_mode));
4811 } else {
4812 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4813 if (lacp_fast == -1) {
4814 printk(KERN_ERR DRV_NAME
4815 ": Error: Invalid lacp rate \"%s\"\n",
4816 lacp_rate == NULL ? "NULL" : lacp_rate);
4817 return -EINVAL;
4818 }
4819 }
4820 }
4821
fd989c83
JV		 4822 if (ad_select) {
4823 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4824 if (params->ad_select == -1) {
4825 printk(KERN_ERR DRV_NAME
4826 ": Error: Invalid ad_select \"%s\"\n",
4827 ad_select == NULL ? "NULL" : ad_select);
4828 return -EINVAL;
4829 }
4830
4831 if (bond_mode != BOND_MODE_8023AD) {
4832 printk(KERN_WARNING DRV_NAME
4833 ": ad_select param only affects 802.3ad mode\n");
4834 }
4835 } else {
4836 params->ad_select = BOND_AD_STABLE;
4837 }
4838
b8a9787e 4839 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4840 printk(KERN_WARNING DRV_NAME
4841 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4842 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4843 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4844 max_bonds = BOND_DEFAULT_MAX_BONDS;
4845 }
4846
4847 if (miimon < 0) {
4848 printk(KERN_WARNING DRV_NAME
4849 ": Warning: miimon module parameter (%d), "
4850 "not in range 0-%d, so it was reset to %d\n",
4851 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4852 miimon = BOND_LINK_MON_INTERV;
4853 }
4854
4855 if (updelay < 0) {
4856 printk(KERN_WARNING DRV_NAME
4857 ": Warning: updelay module parameter (%d), "
4858 "not in range 0-%d, so it was reset to 0\n",
4859 updelay, INT_MAX);
4860 updelay = 0;
4861 }
4862
4863 if (downdelay < 0) {
4864 printk(KERN_WARNING DRV_NAME
4865 ": Warning: downdelay module parameter (%d), "
4866 "not in range 0-%d, so it was reset to 0\n",
4867 downdelay, INT_MAX);
4868 downdelay = 0;
4869 }
4870
4871 if ((use_carrier != 0) && (use_carrier != 1)) {
4872 printk(KERN_WARNING DRV_NAME
4873 ": Warning: use_carrier module parameter (%d), "
4874 "not of valid value (0/1), so it was set to 1\n",
4875 use_carrier);
4876 use_carrier = 1;
4877 }
4878
7893b249
MS		 4879 if (num_grat_arp < 0 || num_grat_arp > 255) {
4880 printk(KERN_WARNING DRV_NAME
4881 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4882 "was reset to 1 \n", num_grat_arp);
4883 num_grat_arp = 1;
4884 }
4885
305d552a
BH		 4886 if (num_unsol_na < 0 || num_unsol_na > 255) {
4887 printk(KERN_WARNING DRV_NAME
4888 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4889 "was reset to 1 \n", num_unsol_na);
4890 num_unsol_na = 1;
4891 }
4892
1da177e4
LT		 4893 /* reset values for 802.3ad */
4894 if (bond_mode == BOND_MODE_8023AD) {
4895 if (!miimon) {
4896 printk(KERN_WARNING DRV_NAME
4897 ": Warning: miimon must be specified, "
4898 "otherwise bonding will not detect link "
4899 "failure, speed and duplex which are "
4900 "essential for 802.3ad operation\n");
4901 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4902 miimon = 100;
4903 }
4904 }
4905
4906 /* reset values for TLB/ALB */
4907 if ((bond_mode == BOND_MODE_TLB) ||
4908 (bond_mode == BOND_MODE_ALB)) {
4909 if (!miimon) {
4910 printk(KERN_WARNING DRV_NAME
4911 ": Warning: miimon must be specified, "
4912 "otherwise bonding will not detect link "
4913 "failure and link speed which are essential "
4914 "for TLB/ALB load balancing\n");
4915 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4916 miimon = 100;
4917 }
4918 }
4919
4920 if (bond_mode == BOND_MODE_ALB) {
4921 printk(KERN_NOTICE DRV_NAME
4922 ": In ALB mode you might experience client "
4923 "disconnections upon reconnection of a link if the "
4924 "bonding module updelay parameter (%d msec) is "
4925 "incompatible with the forwarding delay time of the "
4926 "switch\n",
4927 updelay);
4928 }
4929
4930 if (!miimon) {
4931 if (updelay || downdelay) {
4932 /* just warn the user the up/down delay will have
4933 * no effect since miimon is zero...
4934 */
4935 printk(KERN_WARNING DRV_NAME
4936 ": Warning: miimon module parameter not set "
4937 "and updelay (%d) or downdelay (%d) module "
4938 "parameter is set; updelay and downdelay have "
4939 "no effect unless miimon is set\n",
4940 updelay, downdelay);
4941 }
4942 } else {
4943 /* don't allow arp monitoring */
4944 if (arp_interval) {
4945 printk(KERN_WARNING DRV_NAME
4946 ": Warning: miimon (%d) and arp_interval (%d) "
4947 "can't be used simultaneously, disabling ARP "
4948 "monitoring\n",
4949 miimon, arp_interval);
4950 arp_interval = 0;
4951 }
4952
4953 if ((updelay % miimon) != 0) {
4954 printk(KERN_WARNING DRV_NAME
4955 ": Warning: updelay (%d) is not a multiple "
4956 "of miimon (%d), updelay rounded to %d ms\n",
4957 updelay, miimon, (updelay / miimon) * miimon);
4958 }
4959
4960 updelay /= miimon;
4961
4962 if ((downdelay % miimon) != 0) {
4963 printk(KERN_WARNING DRV_NAME
4964 ": Warning: downdelay (%d) is not a multiple "
4965 "of miimon (%d), downdelay rounded to %d ms\n",
4966 downdelay, miimon,
4967 (downdelay / miimon) * miimon);
4968 }
4969
4970 downdelay /= miimon;
4971 }
4972
4973 if (arp_interval < 0) {
4974 printk(KERN_WARNING DRV_NAME
4975 ": Warning: arp_interval module parameter (%d) "
4976 ", not in range 0-%d, so it was reset to %d\n",
4977 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4978 arp_interval = BOND_LINK_ARP_INTERV;
4979 }
4980
4981 for (arp_ip_count = 0;
4982 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4983 arp_ip_count++) {
4984 /* not complete check, but should be good enough to
4985 catch mistakes */
4986 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4987 printk(KERN_WARNING DRV_NAME
4988 ": Warning: bad arp_ip_target module parameter "
4989 "(%s), ARP monitoring will not be performed\n",
4990 arp_ip_target[arp_ip_count]);
4991 arp_interval = 0;
4992 } else {
d3bb52b0 4993 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4994 arp_target[arp_ip_count] = ip;
4995 }
4996 }
4997
4998 if (arp_interval && !arp_ip_count) {
4999 /* don't allow arping if no arp_ip_target given... */
5000 printk(KERN_WARNING DRV_NAME
5001 ": Warning: arp_interval module parameter (%d) "
5002 "specified without providing an arp_ip_target "
5003 "parameter, arp_interval was reset to 0\n",
5004 arp_interval);
5005 arp_interval = 0;
5006 }
5007
f5b2b966
JV		 5008 if (arp_validate) {
5009 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5010 printk(KERN_ERR DRV_NAME
5011 ": arp_validate only supported in active-backup mode\n");
5012 return -EINVAL;
5013 }
5014 if (!arp_interval) {
5015 printk(KERN_ERR DRV_NAME
5016 ": arp_validate requires arp_interval\n");
5017 return -EINVAL;
5018 }
5019
5020 arp_validate_value = bond_parse_parm(arp_validate,
5021 arp_validate_tbl);
5022 if (arp_validate_value == -1) {
5023 printk(KERN_ERR DRV_NAME
5024 ": Error: invalid arp_validate \"%s\"\n",
5025 arp_validate == NULL ? "NULL" : arp_validate);
5026 return -EINVAL;
5027 }
5028 } else
5029 arp_validate_value = 0;
5030
1da177e4
LT		 5031 if (miimon) {
5032 printk(KERN_INFO DRV_NAME
5033 ": MII link monitoring set to %d ms\n",
5034 miimon);
5035 } else if (arp_interval) {
5036 int i;
5037
5038 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 5039 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5040 arp_interval,
5041 arp_validate_tbl[arp_validate_value].modename,
5042 arp_ip_count);
1da177e4
LT		 5043
5044 for (i = 0; i < arp_ip_count; i++)
5045 printk (" %s", arp_ip_target[i]);
5046
5047 printk("\n");
5048
b8a9787e 5049 } else if (max_bonds) {
1da177e4
LT		 5050 /* miimon and arp_interval not set, we need one so things
5051 * work as expected, see bonding.txt for details
5052 */
5053 printk(KERN_WARNING DRV_NAME
5054 ": Warning: either miimon or arp_interval and "
5055 "arp_ip_target module parameters must be specified, "
5056 "otherwise bonding will not detect link failures! see "
5057 "bonding.txt for details.\n");
5058 }
5059
5060 if (primary && !USES_PRIMARY(bond_mode)) {
5061 /* currently, using a primary only makes sense
5062 * in active backup, TLB or ALB modes
5063 */
5064 printk(KERN_WARNING DRV_NAME
5065 ": Warning: %s primary device specified but has no "
5066 "effect in %s mode\n",
5067 primary, bond_mode_name(bond_mode));
5068 primary = NULL;
5069 }
5070
3915c1e8
JV		 5071 if (fail_over_mac) {
5072 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5073 fail_over_mac_tbl);
5074 if (fail_over_mac_value == -1) {
5075 printk(KERN_ERR DRV_NAME
5076 ": Error: invalid fail_over_mac \"%s\"\n",
5077 arp_validate == NULL ? "NULL" : arp_validate);
5078 return -EINVAL;
5079 }
5080
5081 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5082 printk(KERN_WARNING DRV_NAME
5083 ": Warning: fail_over_mac only affects "
5084 "active-backup mode.\n");
5085 } else {
5086 fail_over_mac_value = BOND_FOM_NONE;
5087 }
dd957c57 5088
1da177e4
LT		 5089 /* fill params struct with the proper values */
5090 params->mode = bond_mode;
169a3e66 5091 params->xmit_policy = xmit_hashtype;
1da177e4 5092 params->miimon = miimon;
7893b249 5093 params->num_grat_arp = num_grat_arp;
305d552a 5094 params->num_unsol_na = num_unsol_na;
1da177e4 5095 params->arp_interval = arp_interval;
f5b2b966 5096 params->arp_validate = arp_validate_value;
1da177e4
LT		 5097 params->updelay = updelay;
5098 params->downdelay = downdelay;
5099 params->use_carrier = use_carrier;
5100 params->lacp_fast = lacp_fast;
5101 params->primary[0] = 0;
3915c1e8 5102 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5103
5104 if (primary) {
5105 strncpy(params->primary, primary, IFNAMSIZ);
5106 params->primary[IFNAMSIZ - 1] = 0;
5107 }
5108
5109 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5110
5111 return 0;
5112}
5113
0daa2303 5114static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 5115static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 5116
e8a0464c
DM		 5117static void bond_set_lockdep_class_one(struct net_device *dev,
5118 struct netdev_queue *txq,
5119 void *_unused)
c773e847
DM		 5120{
5121 lockdep_set_class(&txq->_xmit_lock,
5122 &bonding_netdev_xmit_lock_key);
5123}
5124
5125static void bond_set_lockdep_class(struct net_device *dev)
5126{
cf508b12
DM		 5127 lockdep_set_class(&dev->addr_list_lock,
5128 &bonding_netdev_addr_lock_key);
e8a0464c 5129 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 5130}
5131
dfe60397 5132/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5133 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5134 * Caller must NOT hold rtnl_lock; we need to release it here before we
5135 * set up our sysfs entries.
5136 */
0dd646fe 5137int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5138{
5139 struct net_device *bond_dev;
0883beca 5140 struct bonding *bond;
dfe60397
MW		 5141 int res;
5142
5143 rtnl_lock();
027ea041
JV		 5144 down_write(&bonding_rwsem);
5145
5146 /* Check to see if the bond already exists. */
4fe4763c 5147 if (name) {
0883beca 5148 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5149 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5150 printk(KERN_ERR DRV_NAME
027ea041 5151 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5152 name);
5153 res = -EPERM;
5154 goto out_rtnl;
5155 }
5156 }
027ea041 5157
e4b91c48
JV		 5158 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5159 ether_setup);
dfe60397
MW		 5160 if (!bond_dev) {
5161 printk(KERN_ERR DRV_NAME
5162 ": %s: eek! can't alloc netdev!\n",
5163 name);
5164 res = -ENOMEM;
5165 goto out_rtnl;
5166 }
5167
e4b91c48
JV		 5168 if (!name) {
5169 res = dev_alloc_name(bond_dev, "bond%d");
5170 if (res < 0)
5171 goto out_netdev;
5172 }
5173
dfe60397
MW		 5174 /* bond_init() must be called after dev_alloc_name() (for the
5175 * /proc files), but before register_netdevice(), because we
5176 * need to set function pointers.
5177 */
5178
5179 res = bond_init(bond_dev, params);
5180 if (res < 0) {
5181 goto out_netdev;
5182 }
5183
dfe60397
MW		 5184 res = register_netdevice(bond_dev);
5185 if (res < 0) {
5186 goto out_bond;
5187 }
0daa2303 5188
c773e847 5189 bond_set_lockdep_class(bond_dev);
0daa2303 5190
ff59c456
JV		 5191 netif_carrier_off(bond_dev);
5192
027ea041 5193 up_write(&bonding_rwsem);
dfe60397 5194 rtnl_unlock(); /* allows sysfs registration of net device */
454d7c9b 5195 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
09c89279
JV		 5196 if (res < 0) {
5197 rtnl_lock();
027ea041 5198 down_write(&bonding_rwsem);
822973ba
PE		 5199 bond_deinit(bond_dev);
5200 unregister_netdevice(bond_dev);
5201 goto out_rtnl;
09c89279
JV		 5202 }
5203
5204 return 0;
5205
dfe60397
MW		 5206out_bond:
5207 bond_deinit(bond_dev);
5208out_netdev:
5209 free_netdev(bond_dev);
5210out_rtnl:
027ea041 5211 up_write(&bonding_rwsem);
dfe60397 5212 rtnl_unlock();
dfe60397
MW		 5213 return res;
5214}
5215
1da177e4
LT		 5216static int __init bonding_init(void)
5217{
1da177e4
LT		 5218 int i;
5219 int res;
0883beca 5220 struct bonding *bond;
1da177e4
LT		 5221
5222 printk(KERN_INFO "%s", version);
5223
dfe60397 5224 res = bond_check_params(&bonding_defaults);
1da177e4 5225 if (res) {
dfe60397 5226 goto out;
1da177e4
LT		 5227 }
5228
1da177e4
LT		 5229#ifdef CONFIG_PROC_FS
5230 bond_create_proc_dir();
5231#endif
027ea041
JV		 5232
5233 init_rwsem(&bonding_rwsem);
5234
1da177e4 5235 for (i = 0; i < max_bonds; i++) {
0dd646fe 5236 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5237 if (res)
5238 goto err;
1da177e4
LT		 5239 }
5240
b76cdba9
MW		 5241 res = bond_create_sysfs();
5242 if (res)
5243 goto err;
5244
1da177e4 5245 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5246 register_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5247 bond_register_ipv6_notifier();
1da177e4 5248
dfe60397
MW		 5249 goto out;
5250err:
0883beca 5251 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5252 bond_work_cancel_all(bond);
5253 destroy_workqueue(bond->wq);
5254 }
5255
ae68c398
PE		 5256 bond_destroy_sysfs();
5257
40abc270 5258 rtnl_lock();
1da177e4 5259 bond_free_all();
1da177e4 5260 rtnl_unlock();
dfe60397 5261out:
1da177e4 5262 return res;
dfe60397 5263
1da177e4
LT		 5264}
5265
5266static void __exit bonding_exit(void)
5267{
5268 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5269 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5270 bond_unregister_ipv6_notifier();
1da177e4 5271
ae68c398
PE		 5272 bond_destroy_sysfs();
5273
1da177e4
LT		 5274 rtnl_lock();
5275 bond_free_all();
5276 rtnl_unlock();
5277}
5278
5279module_init(bonding_init);
5280module_exit(bonding_exit);
5281MODULE_LICENSE("GPL");
5282MODULE_VERSION(DRV_VERSION);
5283MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5284MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5285MODULE_SUPPORTED_DEVICE("most ethernet devices");
5286
5287/*
5288 * Local variables:
5289 * c-indent-level: 8
5290 * c-basic-offset: 8
5291 * tab-width: 8
5292 * End:
5293 */
5294
Clone of davem's net-next-2.6 tree