]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/bonding/bond_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bonding: fix compile error if debug enabled
[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 221{
77afc92b
HE		 222 static const char *names[] = {
223 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
224 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
225 [BOND_MODE_XOR] = "load balancing (xor)",
226 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
227 [BOND_MODE_8023AD]= "IEEE 802.3ad Dynamic link aggregation",
228 [BOND_MODE_TLB] = "transmit load balancing",
229 [BOND_MODE_ALB] = "adaptive load balancing",
230 };
231
232 if (mode < 0 || mode > BOND_MODE_ALB)
1da177e4 233 return "unknown";
77afc92b
HE		 234
235 return names[mode];
1da177e4
LT		 236}
237
238/*---------------------------------- VLAN -----------------------------------*/
239
240/**
241 * bond_add_vlan - add a new vlan id on bond
242 * @bond: bond that got the notification
243 * @vlan_id: the vlan id to add
244 *
245 * Returns -ENOMEM if allocation failed.
246 */
247static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
248{
249 struct vlan_entry *vlan;
250
251 dprintk("bond: %s, vlan id %d\n",
252 (bond ? bond->dev->name: "None"), vlan_id);
253
305d552a 254 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
1da177e4
LT		 255 if (!vlan) {
256 return -ENOMEM;
257 }
258
259 INIT_LIST_HEAD(&vlan->vlan_list);
260 vlan->vlan_id = vlan_id;
261
262 write_lock_bh(&bond->lock);
263
264 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
265
266 write_unlock_bh(&bond->lock);
267
268 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
269
270 return 0;
271}
272
273/**
274 * bond_del_vlan - delete a vlan id from bond
275 * @bond: bond that got the notification
276 * @vlan_id: the vlan id to delete
277 *
278 * returns -ENODEV if @vlan_id was not found in @bond.
279 */
280static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
281{
0883beca 282 struct vlan_entry *vlan;
1da177e4
LT		 283 int res = -ENODEV;
284
285 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
286
287 write_lock_bh(&bond->lock);
288
0883beca 289 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
1da177e4
LT		 290 if (vlan->vlan_id == vlan_id) {
291 list_del(&vlan->vlan_list);
292
58402054 293 if (bond_is_lb(bond))
1da177e4 294 bond_alb_clear_vlan(bond, vlan_id);
1da177e4
LT		 295
296 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
297 bond->dev->name);
298
299 kfree(vlan);
300
301 if (list_empty(&bond->vlan_list) &&
302 (bond->slave_cnt == 0)) {
303 /* Last VLAN removed and no slaves, so
304 * restore block on adding VLANs. This will
305 * be removed once new slaves that are not
306 * VLAN challenged will be added.
307 */
308 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
309 }
310
311 res = 0;
312 goto out;
313 }
314 }
315
316 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
317 bond->dev->name);
318
319out:
320 write_unlock_bh(&bond->lock);
321 return res;
322}
323
324/**
325 * bond_has_challenged_slaves
326 * @bond: the bond we're working on
327 *
328 * Searches the slave list. Returns 1 if a vlan challenged slave
329 * was found, 0 otherwise.
330 *
331 * Assumes bond->lock is held.
332 */
333static int bond_has_challenged_slaves(struct bonding *bond)
334{
335 struct slave *slave;
336 int i;
337
338 bond_for_each_slave(bond, slave, i) {
339 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
340 dprintk("found VLAN challenged slave - %s\n",
341 slave->dev->name);
342 return 1;
343 }
344 }
345
346 dprintk("no VLAN challenged slaves found\n");
347 return 0;
348}
349
350/**
351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
354 *
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
357 *
358 * Caller must hold bond->lock
359 */
360struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
361{
362 struct vlan_entry *next, *last;
363
364 if (list_empty(&bond->vlan_list)) {
365 return NULL;
366 }
367
368 if (!curr) {
369 next = list_entry(bond->vlan_list.next,
370 struct vlan_entry, vlan_list);
371 } else {
372 last = list_entry(bond->vlan_list.prev,
373 struct vlan_entry, vlan_list);
374 if (last == curr) {
375 next = list_entry(bond->vlan_list.next,
376 struct vlan_entry, vlan_list);
377 } else {
378 next = list_entry(curr->vlan_list.next,
379 struct vlan_entry, vlan_list);
380 }
381 }
382
383 return next;
384}
385
386/**
387 * bond_dev_queue_xmit - Prepare skb for xmit.
388 *
389 * @bond: bond device that got this skb for tx.
390 * @skb: hw accel VLAN tagged skb to transmit
391 * @slave_dev: slave that is supposed to xmit this skbuff
392 *
393 * When the bond gets an skb to transmit that is
394 * already hardware accelerated VLAN tagged, and it
395 * needs to relay this skb to a slave that is not
396 * hw accel capable, the skb needs to be "unaccelerated",
397 * i.e. strip the hwaccel tag and re-insert it as part
398 * of the payload.
399 */
400int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
401{
966bc6f4 402 unsigned short uninitialized_var(vlan_id);
1da177e4
LT		 403
404 if (!list_empty(&bond->vlan_list) &&
405 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
406 vlan_get_tag(skb, &vlan_id) == 0) {
407 skb->dev = slave_dev;
408 skb = vlan_put_tag(skb, vlan_id);
409 if (!skb) {
410 /* vlan_put_tag() frees the skb in case of error,
411 * so return success here so the calling functions
412 * won't attempt to free is again.
413 */
414 return 0;
415 }
416 } else {
417 skb->dev = slave_dev;
418 }
419
420 skb->priority = 1;
421 dev_queue_xmit(skb);
422
423 return 0;
424}
425
426/*
427 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
428 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
429 * lock because:
430 * a. This operation is performed in IOCTL context,
431 * b. The operation is protected by the RTNL semaphore in the 8021q code,
432 * c. Holding a lock with BH disabled while directly calling a base driver
433 * entry point is generally a BAD idea.
434 *
435 * The design of synchronization/protection for this operation in the 8021q
436 * module is good for one or more VLAN devices over a single physical device
437 * and cannot be extended for a teaming solution like bonding, so there is a
438 * potential race condition here where a net device from the vlan group might
439 * be referenced (either by a base driver or the 8021q code) while it is being
440 * removed from the system. However, it turns out we're not making matters
441 * worse, and if it works for regular VLAN usage it will work here too.
442*/
443
444/**
445 * bond_vlan_rx_register - Propagates registration to slaves
446 * @bond_dev: bonding net device that got called
447 * @grp: vlan group being registered
448 */
449static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
450{
454d7c9b 451 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 452 struct slave *slave;
453 int i;
454
455 bond->vlgrp = grp;
456
457 bond_for_each_slave(bond, slave, i) {
458 struct net_device *slave_dev = slave->dev;
eb7cc59a 459 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 460
461 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 462 slave_ops->ndo_vlan_rx_register) {
463 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
1da177e4
LT		 464 }
465 }
466}
467
468/**
469 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
470 * @bond_dev: bonding net device that got called
471 * @vid: vlan id being added
472 */
473static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
474{
454d7c9b 475 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 476 struct slave *slave;
477 int i, res;
478
479 bond_for_each_slave(bond, slave, i) {
480 struct net_device *slave_dev = slave->dev;
eb7cc59a 481 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 482
483 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a
SH		 484 slave_ops->ndo_vlan_rx_add_vid) {
485 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
1da177e4
LT		 486 }
487 }
488
489 res = bond_add_vlan(bond, vid);
490 if (res) {
491 printk(KERN_ERR DRV_NAME
4e0952c7 492 ": %s: Error: Failed to add vlan id %d\n",
1da177e4
LT		 493 bond_dev->name, vid);
494 }
495}
496
497/**
498 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
499 * @bond_dev: bonding net device that got called
500 * @vid: vlan id being removed
501 */
502static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
503{
454d7c9b 504 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 505 struct slave *slave;
506 struct net_device *vlan_dev;
507 int i, res;
508
509 bond_for_each_slave(bond, slave, i) {
510 struct net_device *slave_dev = slave->dev;
eb7cc59a 511 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 512
513 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a 514 slave_ops->ndo_vlan_rx_kill_vid) {
1da177e4
LT		 515 /* Save and then restore vlan_dev in the grp array,
516 * since the slave's driver might clear it.
517 */
5c15bdec 518 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
eb7cc59a 519 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
5c15bdec 520 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
1da177e4
LT		 521 }
522 }
523
524 res = bond_del_vlan(bond, vid);
525 if (res) {
526 printk(KERN_ERR DRV_NAME
4e0952c7 527 ": %s: Error: Failed to remove vlan id %d\n",
1da177e4
LT		 528 bond_dev->name, vid);
529 }
530}
531
532static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
533{
534 struct vlan_entry *vlan;
eb7cc59a 535 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 536
537 write_lock_bh(&bond->lock);
538
eb7cc59a 539 if (list_empty(&bond->vlan_list))
1da177e4 540 goto out;
1da177e4
LT		 541
542 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 543 slave_ops->ndo_vlan_rx_register)
544 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
1da177e4
LT		 545
546 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 547 !(slave_ops->ndo_vlan_rx_add_vid))
1da177e4 548 goto out;
1da177e4 549
eb7cc59a
SH		 550 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
551 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
1da177e4
LT		 552
553out:
554 write_unlock_bh(&bond->lock);
555}
556
557static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
558{
eb7cc59a 559 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 560 struct vlan_entry *vlan;
561 struct net_device *vlan_dev;
562
563 write_lock_bh(&bond->lock);
564
eb7cc59a 565 if (list_empty(&bond->vlan_list))
1da177e4 566 goto out;
1da177e4
LT		 567
568 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 569 !(slave_ops->ndo_vlan_rx_kill_vid))
1da177e4 570 goto unreg;
1da177e4
LT		 571
572 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
573 /* Save and then restore vlan_dev in the grp array,
574 * since the slave's driver might clear it.
575 */
5c15bdec 576 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
eb7cc59a 577 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
5c15bdec 578 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
1da177e4
LT		 579 }
580
581unreg:
582 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 583 slave_ops->ndo_vlan_rx_register)
584 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
1da177e4
LT		 585
586out:
587 write_unlock_bh(&bond->lock);
588}
589
590/*------------------------------- Link status -------------------------------*/
591
ff59c456
JV		 592/*
593 * Set the carrier state for the master according to the state of its
594 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
595 * do special 802.3ad magic.
596 *
597 * Returns zero if carrier state does not change, nonzero if it does.
598 */
599static int bond_set_carrier(struct bonding *bond)
600{
601 struct slave *slave;
602 int i;
603
604 if (bond->slave_cnt == 0)
605 goto down;
606
607 if (bond->params.mode == BOND_MODE_8023AD)
608 return bond_3ad_set_carrier(bond);
609
610 bond_for_each_slave(bond, slave, i) {
611 if (slave->link == BOND_LINK_UP) {
612 if (!netif_carrier_ok(bond->dev)) {
613 netif_carrier_on(bond->dev);
614 return 1;
615 }
616 return 0;
617 }
618 }
619
620down:
621 if (netif_carrier_ok(bond->dev)) {
622 netif_carrier_off(bond->dev);
623 return 1;
624 }
625 return 0;
626}
627
1da177e4
LT		 628/*
629 * Get link speed and duplex from the slave's base driver
630 * using ethtool. If for some reason the call fails or the
631 * values are invalid, fake speed and duplex to 100/Full
632 * and return error.
633 */
634static int bond_update_speed_duplex(struct slave *slave)
635{
636 struct net_device *slave_dev = slave->dev;
1da177e4 637 struct ethtool_cmd etool;
61a44b9c 638 int res;
1da177e4
LT		 639
640 /* Fake speed and duplex */
641 slave->speed = SPEED_100;
642 slave->duplex = DUPLEX_FULL;
643
61a44b9c
MW		 644 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
645 return -1;
1da177e4 646
61a44b9c
MW		 647 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
648 if (res < 0)
1da177e4 649 return -1;
1da177e4 650
1da177e4
LT		 651 switch (etool.speed) {
652 case SPEED_10:
653 case SPEED_100:
654 case SPEED_1000:
94dbffd5 655 case SPEED_10000:
1da177e4
LT		 656 break;
657 default:
658 return -1;
659 }
660
661 switch (etool.duplex) {
662 case DUPLEX_FULL:
663 case DUPLEX_HALF:
664 break;
665 default:
666 return -1;
667 }
668
669 slave->speed = etool.speed;
670 slave->duplex = etool.duplex;
671
672 return 0;
673}
674
675/*
676 * if <dev> supports MII link status reporting, check its link status.
677 *
678 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
679 * depening upon the setting of the use_carrier parameter.
680 *
681 * Return either BMSR_LSTATUS, meaning that the link is up (or we
682 * can't tell and just pretend it is), or 0, meaning that the link is
683 * down.
684 *
685 * If reporting is non-zero, instead of faking link up, return -1 if
686 * both ETHTOOL and MII ioctls fail (meaning the device does not
687 * support them). If use_carrier is set, return whatever it says.
688 * It'd be nice if there was a good way to tell if a driver supports
689 * netif_carrier, but there really isn't.
690 */
691static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
692{
eb7cc59a 693 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 694 static int (* ioctl)(struct net_device *, struct ifreq *, int);
695 struct ifreq ifr;
696 struct mii_ioctl_data *mii;
1da177e4 697
eb7cc59a 698 if (bond->params.use_carrier)
1da177e4 699 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1da177e4 700
eb7cc59a 701 ioctl = slave_ops->ndo_do_ioctl;
1da177e4
LT		 702 if (ioctl) {
703 /* TODO: set pointer to correct ioctl on a per team member */
704 /* bases to make this more efficient. that is, once */
705 /* we determine the correct ioctl, we will always */
706 /* call it and not the others for that team */
707 /* member. */
708
709 /*
710 * We cannot assume that SIOCGMIIPHY will also read a
711 * register; not all network drivers (e.g., e100)
712 * support that.
713 */
714
715 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
716 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
717 mii = if_mii(&ifr);
718 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
719 mii->reg_num = MII_BMSR;
720 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
721 return (mii->val_out & BMSR_LSTATUS);
722 }
723 }
724 }
725
61a44b9c
MW		 726 /*
727 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
728 * attempt to get link status from it if the above MII ioctls fail.
729 */
1da177e4
LT		 730 if (slave_dev->ethtool_ops) {
731 if (slave_dev->ethtool_ops->get_link) {
732 u32 link;
733
734 link = slave_dev->ethtool_ops->get_link(slave_dev);
735
736 return link ? BMSR_LSTATUS : 0;
737 }
738 }
739
1da177e4
LT		 740 /*
741 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 742 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 743 * cannot report link status). If not reporting, pretend
744 * we're ok.
745 */
746 return (reporting ? -1 : BMSR_LSTATUS);
747}
748
749/*----------------------------- Multicast list ------------------------------*/
750
751/*
752 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
753 */
754static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
755{
756 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
757 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
758}
759
760/*
761 * returns dmi entry if found, NULL otherwise
762 */
763static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
764{
765 struct dev_mc_list *idmi;
766
767 for (idmi = mc_list; idmi; idmi = idmi->next) {
768 if (bond_is_dmi_same(dmi, idmi)) {
769 return idmi;
770 }
771 }
772
773 return NULL;
774}
775
776/*
777 * Push the promiscuity flag down to appropriate slaves
778 */
7e1a1ac1 779static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 780{
7e1a1ac1 781 int err = 0;
1da177e4
LT		 782 if (USES_PRIMARY(bond->params.mode)) {
783 /* write lock already acquired */
784 if (bond->curr_active_slave) {
7e1a1ac1
WC		 785 err = dev_set_promiscuity(bond->curr_active_slave->dev,
786 inc);
1da177e4
LT		 787 }
788 } else {
789 struct slave *slave;
790 int i;
791 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 792 err = dev_set_promiscuity(slave->dev, inc);
793 if (err)
794 return err;
1da177e4
LT		 795 }
796 }
7e1a1ac1 797 return err;
1da177e4
LT		 798}
799
800/*
801 * Push the allmulti flag down to all slaves
802 */
7e1a1ac1 803static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 804{
7e1a1ac1 805 int err = 0;
1da177e4
LT		 806 if (USES_PRIMARY(bond->params.mode)) {
807 /* write lock already acquired */
808 if (bond->curr_active_slave) {
7e1a1ac1
WC		 809 err = dev_set_allmulti(bond->curr_active_slave->dev,
810 inc);
1da177e4
LT		 811 }
812 } else {
813 struct slave *slave;
814 int i;
815 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 816 err = dev_set_allmulti(slave->dev, inc);
817 if (err)
818 return err;
1da177e4
LT		 819 }
820 }
7e1a1ac1 821 return err;
1da177e4
LT		 822}
823
824/*
825 * Add a Multicast address to slaves
826 * according to mode
827 */
828static void bond_mc_add(struct bonding *bond, void *addr, int alen)
829{
830 if (USES_PRIMARY(bond->params.mode)) {
831 /* write lock already acquired */
832 if (bond->curr_active_slave) {
833 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
834 }
835 } else {
836 struct slave *slave;
837 int i;
838 bond_for_each_slave(bond, slave, i) {
839 dev_mc_add(slave->dev, addr, alen, 0);
840 }
841 }
842}
843
844/*
845 * Remove a multicast address from slave
846 * according to mode
847 */
848static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
849{
850 if (USES_PRIMARY(bond->params.mode)) {
851 /* write lock already acquired */
852 if (bond->curr_active_slave) {
853 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
854 }
855 } else {
856 struct slave *slave;
857 int i;
858 bond_for_each_slave(bond, slave, i) {
859 dev_mc_delete(slave->dev, addr, alen, 0);
860 }
861 }
862}
863
a816c7c7
JV		 864
865/*
866 * Retrieve the list of registered multicast addresses for the bonding
867 * device and retransmit an IGMP JOIN request to the current active
868 * slave.
869 */
870static void bond_resend_igmp_join_requests(struct bonding *bond)
871{
872 struct in_device *in_dev;
873 struct ip_mc_list *im;
874
875 rcu_read_lock();
876 in_dev = __in_dev_get_rcu(bond->dev);
877 if (in_dev) {
878 for (im = in_dev->mc_list; im; im = im->next) {
879 ip_mc_rejoin_group(im);
880 }
881 }
882
883 rcu_read_unlock();
884}
885
1da177e4
LT		 886/*
887 * Totally destroys the mc_list in bond
888 */
889static void bond_mc_list_destroy(struct bonding *bond)
890{
891 struct dev_mc_list *dmi;
892
893 dmi = bond->mc_list;
894 while (dmi) {
895 bond->mc_list = dmi->next;
896 kfree(dmi);
897 dmi = bond->mc_list;
898 }
a816c7c7 899 bond->mc_list = NULL;
1da177e4
LT		 900}
901
902/*
903 * Copy all the Multicast addresses from src to the bonding device dst
904 */
de54f390 905static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 906 gfp_t gfp_flag)
1da177e4
LT		 907{
908 struct dev_mc_list *dmi, *new_dmi;
909
910 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 911 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT		 912
913 if (!new_dmi) {
914 /* FIXME: Potential memory leak !!! */
915 return -ENOMEM;
916 }
917
918 new_dmi->next = bond->mc_list;
919 bond->mc_list = new_dmi;
920 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
921 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
922 new_dmi->dmi_users = dmi->dmi_users;
923 new_dmi->dmi_gusers = dmi->dmi_gusers;
924 }
925
926 return 0;
927}
928
929/*
930 * flush all members of flush->mc_list from device dev->mc_list
931 */
932static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
933{
454d7c9b 934 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 935 struct dev_mc_list *dmi;
936
937 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
938 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
939 }
940
941 if (bond->params.mode == BOND_MODE_8023AD) {
942 /* del lacpdu mc addr from mc list */
943 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
944
945 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
946 }
947}
948
949/*--------------------------- Active slave change ---------------------------*/
950
951/*
952 * Update the mc list and multicast-related flags for the new and
953 * old active slaves (if any) according to the multicast mode, and
954 * promiscuous flags unconditionally.
955 */
956static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
957{
958 struct dev_mc_list *dmi;
959
960 if (!USES_PRIMARY(bond->params.mode)) {
961 /* nothing to do - mc list is already up-to-date on
962 * all slaves
963 */
964 return;
965 }
966
967 if (old_active) {
968 if (bond->dev->flags & IFF_PROMISC) {
969 dev_set_promiscuity(old_active->dev, -1);
970 }
971
972 if (bond->dev->flags & IFF_ALLMULTI) {
973 dev_set_allmulti(old_active->dev, -1);
974 }
975
976 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
977 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
978 }
979 }
980
981 if (new_active) {
7e1a1ac1 982 /* FIXME: Signal errors upstream. */
1da177e4
LT		 983 if (bond->dev->flags & IFF_PROMISC) {
984 dev_set_promiscuity(new_active->dev, 1);
985 }
986
987 if (bond->dev->flags & IFF_ALLMULTI) {
988 dev_set_allmulti(new_active->dev, 1);
989 }
990
991 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
992 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
993 }
a816c7c7 994 bond_resend_igmp_join_requests(bond);
1da177e4
LT		 995 }
996}
997
3915c1e8
JV		 998/*
999 * bond_do_fail_over_mac
1000 *
1001 * Perform special MAC address swapping for fail_over_mac settings
1002 *
1003 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1004 */
1005static void bond_do_fail_over_mac(struct bonding *bond,
1006 struct slave *new_active,
1007 struct slave *old_active)
1008{
1009 u8 tmp_mac[ETH_ALEN];
1010 struct sockaddr saddr;
1011 int rv;
1012
1013 switch (bond->params.fail_over_mac) {
1014 case BOND_FOM_ACTIVE:
1015 if (new_active)
1016 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1017 new_active->dev->addr_len);
1018 break;
1019 case BOND_FOM_FOLLOW:
1020 /*
1021 * if new_active && old_active, swap them
1022 * if just old_active, do nothing (going to no active slave)
1023 * if just new_active, set new_active to bond's MAC
1024 */
1025 if (!new_active)
1026 return;
1027
1028 write_unlock_bh(&bond->curr_slave_lock);
1029 read_unlock(&bond->lock);
1030
1031 if (old_active) {
1032 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1033 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1034 ETH_ALEN);
1035 saddr.sa_family = new_active->dev->type;
1036 } else {
1037 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1038 saddr.sa_family = bond->dev->type;
1039 }
1040
1041 rv = dev_set_mac_address(new_active->dev, &saddr);
1042 if (rv) {
1043 printk(KERN_ERR DRV_NAME
1044 ": %s: Error %d setting MAC of slave %s\n",
1045 bond->dev->name, -rv, new_active->dev->name);
1046 goto out;
1047 }
1048
1049 if (!old_active)
1050 goto out;
1051
1052 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1053 saddr.sa_family = old_active->dev->type;
1054
1055 rv = dev_set_mac_address(old_active->dev, &saddr);
1056 if (rv)
1057 printk(KERN_ERR DRV_NAME
1058 ": %s: Error %d setting MAC of slave %s\n",
1059 bond->dev->name, -rv, new_active->dev->name);
1060out:
1061 read_lock(&bond->lock);
1062 write_lock_bh(&bond->curr_slave_lock);
1063 break;
1064 default:
1065 printk(KERN_ERR DRV_NAME
1066 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1067 bond->dev->name, bond->params.fail_over_mac);
1068 break;
1069 }
1070
1071}
1072
1073
1da177e4
LT		 1074/**
1075 * find_best_interface - select the best available slave to be the active one
1076 * @bond: our bonding struct
1077 *
1078 * Warning: Caller must hold curr_slave_lock for writing.
1079 */
1080static struct slave *bond_find_best_slave(struct bonding *bond)
1081{
1082 struct slave *new_active, *old_active;
1083 struct slave *bestslave = NULL;
1084 int mintime = bond->params.updelay;
1085 int i;
1086
1087 new_active = old_active = bond->curr_active_slave;
1088
1089 if (!new_active) { /* there were no active slaves left */
1090 if (bond->slave_cnt > 0) { /* found one slave */
1091 new_active = bond->first_slave;
1092 } else {
1093 return NULL; /* still no slave, return NULL */
1094 }
1095 }
1096
1097 /* first try the primary link; if arping, a link must tx/rx traffic
1098 * before it can be considered the curr_active_slave - also, we would skip
1099 * slaves between the curr_active_slave and primary_slave that may be up
1100 * and able to arp
1101 */
1102 if ((bond->primary_slave) &&
1103 (!bond->params.arp_interval) &&
1104 (IS_UP(bond->primary_slave->dev))) {
1105 new_active = bond->primary_slave;
1106 }
1107
1108 /* remember where to stop iterating over the slaves */
1109 old_active = new_active;
1110
1111 bond_for_each_slave_from(bond, new_active, i, old_active) {
1112 if (IS_UP(new_active->dev)) {
1113 if (new_active->link == BOND_LINK_UP) {
1114 return new_active;
1115 } else if (new_active->link == BOND_LINK_BACK) {
1116 /* link up, but waiting for stabilization */
1117 if (new_active->delay < mintime) {
1118 mintime = new_active->delay;
1119 bestslave = new_active;
1120 }
1121 }
1122 }
1123 }
1124
1125 return bestslave;
1126}
1127
1128/**
1129 * change_active_interface - change the active slave into the specified one
1130 * @bond: our bonding struct
1131 * @new: the new slave to make the active one
1132 *
1133 * Set the new slave to the bond's settings and unset them on the old
1134 * curr_active_slave.
1135 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1136 *
1137 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1138 * because it is apparently the best available slave we have, even though its
1139 * updelay hasn't timed out yet.
1140 *
3915c1e8
JV		 1141 * If new_active is not NULL, caller must hold bond->lock for read and
1142 * curr_slave_lock for write_bh.
1da177e4 1143 */
a77b5325 1144void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1145{
1146 struct slave *old_active = bond->curr_active_slave;
1147
1148 if (old_active == new_active) {
1149 return;
1150 }
1151
1152 if (new_active) {
b2220cad
JV		 1153 new_active->jiffies = jiffies;
1154
1da177e4
LT		 1155 if (new_active->link == BOND_LINK_BACK) {
1156 if (USES_PRIMARY(bond->params.mode)) {
1157 printk(KERN_INFO DRV_NAME
1158 ": %s: making interface %s the new "
1159 "active one %d ms earlier.\n",
1160 bond->dev->name, new_active->dev->name,
1161 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1162 }
1163
1164 new_active->delay = 0;
1165 new_active->link = BOND_LINK_UP;
1da177e4
LT		 1166
1167 if (bond->params.mode == BOND_MODE_8023AD) {
1168 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1169 }
1170
58402054 1171 if (bond_is_lb(bond))
1da177e4 1172 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1da177e4
LT		 1173 } else {
1174 if (USES_PRIMARY(bond->params.mode)) {
1175 printk(KERN_INFO DRV_NAME
1176 ": %s: making interface %s the new "
1177 "active one.\n",
1178 bond->dev->name, new_active->dev->name);
1179 }
1180 }
1181 }
1182
1da177e4
LT		 1183 if (USES_PRIMARY(bond->params.mode)) {
1184 bond_mc_swap(bond, new_active, old_active);
1185 }
1186
58402054 1187 if (bond_is_lb(bond)) {
1da177e4 1188 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1189 if (old_active)
1190 bond_set_slave_inactive_flags(old_active);
1191 if (new_active)
1192 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1193 } else {
1194 bond->curr_active_slave = new_active;
1195 }
c3ade5ca
JV		 1196
1197 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1198 if (old_active) {
1199 bond_set_slave_inactive_flags(old_active);
1200 }
1201
1202 if (new_active) {
1203 bond_set_slave_active_flags(new_active);
2ab82852 1204
709f8a45
OG		 1205 if (bond->params.fail_over_mac)
1206 bond_do_fail_over_mac(bond, new_active,
1207 old_active);
3915c1e8 1208
709f8a45 1209 bond->send_grat_arp = bond->params.num_grat_arp;
b59f9f74 1210 bond_send_gratuitous_arp(bond);
01f3109d 1211
305d552a
BH		 1212 bond->send_unsol_na = bond->params.num_unsol_na;
1213 bond_send_unsolicited_na(bond);
1214
01f3109d
OG		 1215 write_unlock_bh(&bond->curr_slave_lock);
1216 read_unlock(&bond->lock);
1217
1218 netdev_bonding_change(bond->dev);
1219
1220 read_lock(&bond->lock);
1221 write_lock_bh(&bond->curr_slave_lock);
7893b249 1222 }
c3ade5ca 1223 }
1da177e4
LT		 1224}
1225
1226/**
1227 * bond_select_active_slave - select a new active slave, if needed
1228 * @bond: our bonding struct
1229 *
1230 * This functions shoud be called when one of the following occurs:
1231 * - The old curr_active_slave has been released or lost its link.
1232 * - The primary_slave has got its link back.
1233 * - A slave has got its link back and there's no old curr_active_slave.
1234 *
3915c1e8 1235 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1236 */
a77b5325 1237void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1238{
1239 struct slave *best_slave;
ff59c456 1240 int rv;
1da177e4
LT		 1241
1242 best_slave = bond_find_best_slave(bond);
1243 if (best_slave != bond->curr_active_slave) {
1244 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1245 rv = bond_set_carrier(bond);
1246 if (!rv)
1247 return;
1248
1249 if (netif_carrier_ok(bond->dev)) {
1250 printk(KERN_INFO DRV_NAME
1251 ": %s: first active interface up!\n",
1252 bond->dev->name);
1253 } else {
1254 printk(KERN_INFO DRV_NAME ": %s: "
1255 "now running without any active interface !\n",
1256 bond->dev->name);
1257 }
1da177e4
LT		 1258 }
1259}
1260
1261/*--------------------------- slave list handling ---------------------------*/
1262
1263/*
1264 * This function attaches the slave to the end of list.
1265 *
1266 * bond->lock held for writing by caller.
1267 */
1268static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1269{
1270 if (bond->first_slave == NULL) { /* attaching the first slave */
1271 new_slave->next = new_slave;
1272 new_slave->prev = new_slave;
1273 bond->first_slave = new_slave;
1274 } else {
1275 new_slave->next = bond->first_slave;
1276 new_slave->prev = bond->first_slave->prev;
1277 new_slave->next->prev = new_slave;
1278 new_slave->prev->next = new_slave;
1279 }
1280
1281 bond->slave_cnt++;
1282}
1283
1284/*
1285 * This function detaches the slave from the list.
1286 * WARNING: no check is made to verify if the slave effectively
1287 * belongs to <bond>.
1288 * Nothing is freed on return, structures are just unchained.
1289 * If any slave pointer in bond was pointing to <slave>,
1290 * it should be changed by the calling function.
1291 *
1292 * bond->lock held for writing by caller.
1293 */
1294static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1295{
1296 if (slave->next) {
1297 slave->next->prev = slave->prev;
1298 }
1299
1300 if (slave->prev) {
1301 slave->prev->next = slave->next;
1302 }
1303
1304 if (bond->first_slave == slave) { /* slave is the first slave */
1305 if (bond->slave_cnt > 1) { /* there are more slave */
1306 bond->first_slave = slave->next;
1307 } else {
1308 bond->first_slave = NULL; /* slave was the last one */
1309 }
1310 }
1311
1312 slave->next = NULL;
1313 slave->prev = NULL;
1314 bond->slave_cnt--;
1315}
1316
1317/*---------------------------------- IOCTL ----------------------------------*/
1318
4ad072c9
AB		 1319static int bond_sethwaddr(struct net_device *bond_dev,
1320 struct net_device *slave_dev)
1da177e4
LT		 1321{
1322 dprintk("bond_dev=%p\n", bond_dev);
1323 dprintk("slave_dev=%p\n", slave_dev);
1324 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1325 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1326 return 0;
1327}
1328
7f353bf2
HX		 1329#define BOND_VLAN_FEATURES \
1330 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1331 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK		 1332
1333/*
8e3babcd 1334 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX		 1335 * feature bits are managed elsewhere, so preserve those feature bits
1336 * on the master device.
8531c5ff
AK		 1337 */
1338static int bond_compute_features(struct bonding *bond)
1339{
8531c5ff
AK		 1340 struct slave *slave;
1341 struct net_device *bond_dev = bond->dev;
7f353bf2 1342 unsigned long features = bond_dev->features;
3158bf7d
MS		 1343 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1344 bond_dev->hard_header_len);
8e3babcd 1345 int i;
8531c5ff 1346
7f353bf2 1347 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
b63365a2
HX		 1348 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1349
1350 if (!bond->first_slave)
1351 goto done;
1352
1353 features &= ~NETIF_F_ONE_FOR_ALL;
7f353bf2 1354
54ef3137 1355 bond_for_each_slave(bond, slave, i) {
b63365a2
HX		 1356 features = netdev_increment_features(features,
1357 slave->dev->features,
1358 NETIF_F_ONE_FOR_ALL);
54ef3137
JV		 1359 if (slave->dev->hard_header_len > max_hard_header_len)
1360 max_hard_header_len = slave->dev->hard_header_len;
1361 }
8531c5ff 1362
b63365a2 1363done:
7f353bf2 1364 features |= (bond_dev->features & BOND_VLAN_FEATURES);
b63365a2 1365 bond_dev->features = netdev_fix_features(features, NULL);
54ef3137 1366 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1367
1368 return 0;
1369}
1370
872254dd
MS		 1371static void bond_setup_by_slave(struct net_device *bond_dev,
1372 struct net_device *slave_dev)
1373{
454d7c9b 1374 struct bonding *bond = netdev_priv(bond_dev);
d90a162a 1375
00829823 1376 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1377
1378 bond_dev->type = slave_dev->type;
1379 bond_dev->hard_header_len = slave_dev->hard_header_len;
1380 bond_dev->addr_len = slave_dev->addr_len;
1381
1382 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1383 slave_dev->addr_len);
d90a162a 1384 bond->setup_by_slave = 1;
872254dd
MS		 1385}
1386
1da177e4 1387/* enslave device <slave> to bond device <master> */
a77b5325 1388int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1389{
454d7c9b 1390 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1391 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 1392 struct slave *new_slave = NULL;
1393 struct dev_mc_list *dmi;
1394 struct sockaddr addr;
1395 int link_reporting;
1396 int old_features = bond_dev->features;
1397 int res = 0;
1398
552709d5 1399 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
eb7cc59a 1400 slave_ops->ndo_do_ioctl == NULL) {
1da177e4 1401 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1402 ": %s: Warning: no link monitoring support for %s\n",
1403 bond_dev->name, slave_dev->name);
1da177e4
LT		 1404 }
1405
1406 /* bond must be initialized by bond_open() before enslaving */
1407 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1408 printk(KERN_WARNING DRV_NAME
1409 " %s: master_dev is not up in bond_enslave\n",
1410 bond_dev->name);
1da177e4
LT		 1411 }
1412
1413 /* already enslaved */
1414 if (slave_dev->flags & IFF_SLAVE) {
1415 dprintk("Error, Device was already enslaved\n");
1416 return -EBUSY;
1417 }
1418
1419 /* vlan challenged mutual exclusion */
1420 /* no need to lock since we're protected by rtnl_lock */
1421 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1422 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1423 if (!list_empty(&bond->vlan_list)) {
1424 printk(KERN_ERR DRV_NAME
4e0952c7 1425 ": %s: Error: cannot enslave VLAN "
1da177e4 1426 "challenged slave %s on VLAN enabled "
4e0952c7 1427 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1428 bond_dev->name);
1429 return -EPERM;
1430 } else {
1431 printk(KERN_WARNING DRV_NAME
4e0952c7 1432 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1433 "slave %s. Adding VLANs will be blocked as "
1434 "long as %s is part of bond %s\n",
4e0952c7 1435 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1436 bond_dev->name);
1437 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1438 }
1439 } else {
1440 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1441 if (bond->slave_cnt == 0) {
1442 /* First slave, and it is not VLAN challenged,
1443 * so remove the block of adding VLANs over the bond.
1444 */
1445 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1446 }
1447 }
1448
217df670
JV		 1449 /*
1450 * Old ifenslave binaries are no longer supported. These can
1451 * be identified with moderate accurary by the state of the slave:
1452 * the current ifenslave will set the interface down prior to
1453 * enslaving it; the old ifenslave will not.
1454 */
1455 if ((slave_dev->flags & IFF_UP)) {
1456 printk(KERN_ERR DRV_NAME ": %s is up. "
1457 "This may be due to an out of date ifenslave.\n",
1458 slave_dev->name);
1459 res = -EPERM;
1460 goto err_undo_flags;
1461 }
1da177e4 1462
872254dd
MS		 1463 /* set bonding device ether type by slave - bonding netdevices are
1464 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1465 * there is a need to override some of the type dependent attribs/funcs.
1466 *
1467 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1468 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1469 */
1470 if (bond->slave_cnt == 0) {
1471 if (slave_dev->type != ARPHRD_ETHER)
1472 bond_setup_by_slave(bond_dev, slave_dev);
1473 } else if (bond_dev->type != slave_dev->type) {
1474 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1475 "from other slaves (%d), can not enslave it.\n",
1476 slave_dev->name,
1477 slave_dev->type, bond_dev->type);
1478 res = -EINVAL;
1479 goto err_undo_flags;
1480 }
1481
eb7cc59a 1482 if (slave_ops->ndo_set_mac_address == NULL) {
2ab82852
MS		 1483 if (bond->slave_cnt == 0) {
1484 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1485 ": %s: Warning: The first slave device "
1486 "specified does not support setting the MAC "
3915c1e8 1487 "address. Setting fail_over_mac to active.",
dd957c57 1488 bond_dev->name);
3915c1e8
JV		 1489 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1490 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1491 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1492 ": %s: Error: The slave device specified "
1493 "does not support setting the MAC address, "
3915c1e8 1494 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1495 , bond_dev->name);
1496 res = -EOPNOTSUPP;
1497 goto err_undo_flags;
1498 }
1da177e4
LT		 1499 }
1500
243cb4e5 1501 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1502 if (!new_slave) {
1503 res = -ENOMEM;
1504 goto err_undo_flags;
1505 }
1506
1da177e4
LT		 1507 /* save slave's original flags before calling
1508 * netdev_set_master and dev_open
1509 */
1510 new_slave->original_flags = slave_dev->flags;
1511
217df670
JV		 1512 /*
1513 * Save slave's original ("permanent") mac address for modes
1514 * that need it, and for restoring it upon release, and then
1515 * set it to the master's address
1516 */
1517 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1518
dd957c57 1519 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1520 /*
1521 * Set slave to master's mac address. The application already
1522 * set the master's mac address to that of the first slave
1523 */
1524 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1525 addr.sa_family = slave_dev->type;
1526 res = dev_set_mac_address(slave_dev, &addr);
1527 if (res) {
1528 dprintk("Error %d calling set_mac_address\n", res);
1529 goto err_free;
1530 }
217df670 1531 }
1da177e4 1532
c2edacf8
JV		 1533 res = netdev_set_master(slave_dev, bond_dev);
1534 if (res) {
1535 dprintk("Error %d calling netdev_set_master\n", res);
569f0c4d 1536 goto err_restore_mac;
c2edacf8 1537 }
217df670
JV		 1538 /* open the slave since the application closed it */
1539 res = dev_open(slave_dev);
1540 if (res) {
1541 dprintk("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1542 goto err_unset_master;
1da177e4
LT		 1543 }
1544
1da177e4 1545 new_slave->dev = slave_dev;
0b680e75 1546 slave_dev->priv_flags |= IFF_BONDING;
1da177e4 1547
58402054 1548 if (bond_is_lb(bond)) {
1da177e4
LT		 1549 /* bond_alb_init_slave() must be called before all other stages since
1550 * it might fail and we do not want to have to undo everything
1551 */
1552 res = bond_alb_init_slave(bond, new_slave);
1553 if (res) {
569f0c4d 1554 goto err_close;
1da177e4
LT		 1555 }
1556 }
1557
1558 /* If the mode USES_PRIMARY, then the new slave gets the
1559 * master's promisc (and mc) settings only if it becomes the
1560 * curr_active_slave, and that is taken care of later when calling
1561 * bond_change_active()
1562 */
1563 if (!USES_PRIMARY(bond->params.mode)) {
1564 /* set promiscuity level to new slave */
1565 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1566 res = dev_set_promiscuity(slave_dev, 1);
1567 if (res)
1568 goto err_close;
1da177e4
LT		 1569 }
1570
1571 /* set allmulti level to new slave */
1572 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1573 res = dev_set_allmulti(slave_dev, 1);
1574 if (res)
1575 goto err_close;
1da177e4
LT		 1576 }
1577
b9e40857 1578 netif_addr_lock_bh(bond_dev);
1da177e4
LT		 1579 /* upload master's mc_list to new slave */
1580 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1581 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1582 }
b9e40857 1583 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1584 }
1585
1586 if (bond->params.mode == BOND_MODE_8023AD) {
1587 /* add lacpdu mc addr to mc list */
1588 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1589
1590 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1591 }
1592
1593 bond_add_vlans_on_slave(bond, slave_dev);
1594
1595 write_lock_bh(&bond->lock);
1596
1597 bond_attach_slave(bond, new_slave);
1598
1599 new_slave->delay = 0;
1600 new_slave->link_failure_count = 0;
1601
8531c5ff
AK		 1602 bond_compute_features(bond);
1603
3915c1e8
JV		 1604 write_unlock_bh(&bond->lock);
1605
1606 read_lock(&bond->lock);
1607
f5b2b966
JV		 1608 new_slave->last_arp_rx = jiffies;
1609
1da177e4
LT		 1610 if (bond->params.miimon && !bond->params.use_carrier) {
1611 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1612
1613 if ((link_reporting == -1) && !bond->params.arp_interval) {
1614 /*
1615 * miimon is set but a bonded network driver
1616 * does not support ETHTOOL/MII and
1617 * arp_interval is not set. Note: if
1618 * use_carrier is enabled, we will never go
1619 * here (because netif_carrier is always
1620 * supported); thus, we don't need to change
1621 * the messages for netif_carrier.
1622 */
1623 printk(KERN_WARNING DRV_NAME
4e0952c7 1624 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1625 "available for interface %s, and "
1626 "arp_interval/arp_ip_target module parameters "
1627 "not specified, thus bonding will not detect "
1628 "link failures! see bonding.txt for details.\n",
4e0952c7 1629 bond_dev->name, slave_dev->name);
1da177e4
LT		 1630 } else if (link_reporting == -1) {
1631 /* unable get link status using mii/ethtool */
1632 printk(KERN_WARNING DRV_NAME
4e0952c7 1633 ": %s: Warning: can't get link status from "
1da177e4
LT		 1634 "interface %s; the network driver associated "
1635 "with this interface does not support MII or "
1636 "ETHTOOL link status reporting, thus miimon "
1637 "has no effect on this interface.\n",
4e0952c7 1638 bond_dev->name, slave_dev->name);
1da177e4
LT		 1639 }
1640 }
1641
1642 /* check for initial state */
1643 if (!bond->params.miimon ||
1644 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1645 if (bond->params.updelay) {
1646 dprintk("Initial state of slave_dev is "
1647 "BOND_LINK_BACK\n");
1648 new_slave->link = BOND_LINK_BACK;
1649 new_slave->delay = bond->params.updelay;
1650 } else {
1651 dprintk("Initial state of slave_dev is "
1652 "BOND_LINK_UP\n");
1653 new_slave->link = BOND_LINK_UP;
1654 }
1655 new_slave->jiffies = jiffies;
1656 } else {
1657 dprintk("Initial state of slave_dev is "
1658 "BOND_LINK_DOWN\n");
1659 new_slave->link = BOND_LINK_DOWN;
1660 }
1661
1662 if (bond_update_speed_duplex(new_slave) &&
1663 (new_slave->link != BOND_LINK_DOWN)) {
1664 printk(KERN_WARNING DRV_NAME
4e0952c7 1665 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1666 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1667 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1668
1669 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1670 printk(KERN_WARNING DRV_NAME
1671 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1672 "support in base driver for proper aggregator "
4e0952c7 1673 "selection.\n", bond_dev->name);
1da177e4
LT		 1674 }
1675 }
1676
1677 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1678 /* if there is a primary slave, remember it */
1679 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1680 bond->primary_slave = new_slave;
1681 }
1682 }
1683
3915c1e8
JV		 1684 write_lock_bh(&bond->curr_slave_lock);
1685
1da177e4
LT		 1686 switch (bond->params.mode) {
1687 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1688 bond_set_slave_inactive_flags(new_slave);
1689 bond_select_active_slave(bond);
1da177e4
LT		 1690 break;
1691 case BOND_MODE_8023AD:
1692 /* in 802.3ad mode, the internal mechanism
1693 * will activate the slaves in the selected
1694 * aggregator
1695 */
1696 bond_set_slave_inactive_flags(new_slave);
1697 /* if this is the first slave */
1698 if (bond->slave_cnt == 1) {
1699 SLAVE_AD_INFO(new_slave).id = 1;
1700 /* Initialize AD with the number of times that the AD timer is called in 1 second
1701 * can be called only after the mac address of the bond is set
1702 */
1703 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1704 bond->params.lacp_fast);
1705 } else {
1706 SLAVE_AD_INFO(new_slave).id =
1707 SLAVE_AD_INFO(new_slave->prev).id + 1;
1708 }
1709
1710 bond_3ad_bind_slave(new_slave);
1711 break;
1712 case BOND_MODE_TLB:
1713 case BOND_MODE_ALB:
1714 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1715 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1716 break;
1717 default:
1718 dprintk("This slave is always active in trunk mode\n");
1719
1720 /* always active in trunk mode */
1721 new_slave->state = BOND_STATE_ACTIVE;
1722
1723 /* In trunking mode there is little meaning to curr_active_slave
1724 * anyway (it holds no special properties of the bond device),
1725 * so we can change it without calling change_active_interface()
1726 */
1727 if (!bond->curr_active_slave) {
1728 bond->curr_active_slave = new_slave;
1729 }
1730 break;
1731 } /* switch(bond_mode) */
1732
3915c1e8
JV		 1733 write_unlock_bh(&bond->curr_slave_lock);
1734
ff59c456
JV		 1735 bond_set_carrier(bond);
1736
3915c1e8 1737 read_unlock(&bond->lock);
1da177e4 1738
b76cdba9
MW		 1739 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1740 if (res)
569f0c4d 1741 goto err_close;
b76cdba9 1742
1da177e4
LT		 1743 printk(KERN_INFO DRV_NAME
1744 ": %s: enslaving %s as a%s interface with a%s link.\n",
1745 bond_dev->name, slave_dev->name,
1746 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1747 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1748
1749 /* enslave is successful */
1750 return 0;
1751
1752/* Undo stages on error */
1da177e4
LT		 1753err_close:
1754 dev_close(slave_dev);
1755
569f0c4d
JV		 1756err_unset_master:
1757 netdev_set_master(slave_dev, NULL);
1758
1da177e4 1759err_restore_mac:
dd957c57 1760 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1761 /* XXX TODO - fom follow mode needs to change master's
1762 * MAC if this slave's MAC is in use by the bond, or at
1763 * least print a warning.
1764 */
2ab82852
MS		 1765 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1766 addr.sa_family = slave_dev->type;
1767 dev_set_mac_address(slave_dev, &addr);
1768 }
1da177e4
LT		 1769
1770err_free:
1771 kfree(new_slave);
1772
1773err_undo_flags:
1774 bond_dev->features = old_features;
8531c5ff 1775
1da177e4
LT		 1776 return res;
1777}
1778
1779/*
1780 * Try to release the slave device <slave> from the bond device <master>
1781 * It is legal to access curr_active_slave without a lock because all the function
1782 * is write-locked.
1783 *
1784 * The rules for slave state should be:
1785 * for Active/Backup:
1786 * Active stays on all backups go down
1787 * for Bonded connections:
1788 * The first up interface should be left on and all others downed.
1789 */
a77b5325 1790int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1791{
454d7c9b 1792 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1793 struct slave *slave, *oldcurrent;
1794 struct sockaddr addr;
1795 int mac_addr_differ;
1796
1797 /* slave is not a slave or master is not master of this slave */
1798 if (!(slave_dev->flags & IFF_SLAVE) ||
1799 (slave_dev->master != bond_dev)) {
1800 printk(KERN_ERR DRV_NAME
4e0952c7 1801 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1802 bond_dev->name, slave_dev->name);
1803 return -EINVAL;
1804 }
1805
1806 write_lock_bh(&bond->lock);
1807
1808 slave = bond_get_slave_by_dev(bond, slave_dev);
1809 if (!slave) {
1810 /* not a slave of this bond */
1811 printk(KERN_INFO DRV_NAME
1812 ": %s: %s not enslaved\n",
1813 bond_dev->name, slave_dev->name);
f5e2a7b2 1814 write_unlock_bh(&bond->lock);
1da177e4
LT		 1815 return -EINVAL;
1816 }
1817
3915c1e8
JV		 1818 if (!bond->params.fail_over_mac) {
1819 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1820 ETH_ALEN);
1821 if (!mac_addr_differ && (bond->slave_cnt > 1))
1822 printk(KERN_WARNING DRV_NAME
1823 ": %s: Warning: the permanent HWaddr of %s - "
e174961c 1824 "%pM - is still in use by %s. "
3915c1e8
JV		 1825 "Set the HWaddr of %s to a different address "
1826 "to avoid conflicts.\n",
1827 bond_dev->name, slave_dev->name,
e174961c 1828 slave->perm_hwaddr,
3915c1e8 1829 bond_dev->name, slave_dev->name);
1da177e4
LT		 1830 }
1831
1832 /* Inform AD package of unbinding of slave. */
1833 if (bond->params.mode == BOND_MODE_8023AD) {
1834 /* must be called before the slave is
1835 * detached from the list
1836 */
1837 bond_3ad_unbind_slave(slave);
1838 }
1839
1840 printk(KERN_INFO DRV_NAME
1841 ": %s: releasing %s interface %s\n",
1842 bond_dev->name,
1843 (slave->state == BOND_STATE_ACTIVE)
1844 ? "active" : "backup",
1845 slave_dev->name);
1846
1847 oldcurrent = bond->curr_active_slave;
1848
1849 bond->current_arp_slave = NULL;
1850
1851 /* release the slave from its bond */
1852 bond_detach_slave(bond, slave);
1853
8531c5ff
AK		 1854 bond_compute_features(bond);
1855
1da177e4
LT		 1856 if (bond->primary_slave == slave) {
1857 bond->primary_slave = NULL;
1858 }
1859
1860 if (oldcurrent == slave) {
1861 bond_change_active_slave(bond, NULL);
1862 }
1863
58402054 1864 if (bond_is_lb(bond)) {
1da177e4
LT		 1865 /* Must be called only after the slave has been
1866 * detached from the list and the curr_active_slave
1867 * has been cleared (if our_slave == old_current),
1868 * but before a new active slave is selected.
1869 */
2543331d 1870 write_unlock_bh(&bond->lock);
1da177e4 1871 bond_alb_deinit_slave(bond, slave);
2543331d 1872 write_lock_bh(&bond->lock);
1da177e4
LT		 1873 }
1874
059fe7a5
JV		 1875 if (oldcurrent == slave) {
1876 /*
1877 * Note that we hold RTNL over this sequence, so there
1878 * is no concern that another slave add/remove event
1879 * will interfere.
1880 */
1881 write_unlock_bh(&bond->lock);
1882 read_lock(&bond->lock);
1883 write_lock_bh(&bond->curr_slave_lock);
1884
1da177e4
LT		 1885 bond_select_active_slave(bond);
1886
059fe7a5
JV		 1887 write_unlock_bh(&bond->curr_slave_lock);
1888 read_unlock(&bond->lock);
1889 write_lock_bh(&bond->lock);
1890 }
1891
1da177e4 1892 if (bond->slave_cnt == 0) {
ff59c456
JV		 1893 bond_set_carrier(bond);
1894
1da177e4
LT		 1895 /* if the last slave was removed, zero the mac address
1896 * of the master so it will be set by the application
1897 * to the mac address of the first slave
1898 */
1899 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1900
1901 if (list_empty(&bond->vlan_list)) {
1902 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1903 } else {
1904 printk(KERN_WARNING DRV_NAME
4e0952c7 1905 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1906 "still has VLANs.\n",
4e0952c7 1907 bond_dev->name, bond_dev->name);
1da177e4 1908 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1909 ": %s: When re-adding slaves, make sure the bond's "
1910 "HW address matches its VLANs'.\n",
1911 bond_dev->name);
1da177e4
LT		 1912 }
1913 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1914 !bond_has_challenged_slaves(bond)) {
1915 printk(KERN_INFO DRV_NAME
4e0952c7 1916 ": %s: last VLAN challenged slave %s "
1da177e4 1917 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1918 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1919 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1920 }
1921
1922 write_unlock_bh(&bond->lock);
1923
b76cdba9
MW		 1924 /* must do this from outside any spinlocks */
1925 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1926
1da177e4
LT		 1927 bond_del_vlans_from_slave(bond, slave_dev);
1928
1929 /* If the mode USES_PRIMARY, then we should only remove its
1930 * promisc and mc settings if it was the curr_active_slave, but that was
1931 * already taken care of above when we detached the slave
1932 */
1933 if (!USES_PRIMARY(bond->params.mode)) {
1934 /* unset promiscuity level from slave */
1935 if (bond_dev->flags & IFF_PROMISC) {
1936 dev_set_promiscuity(slave_dev, -1);
1937 }
1938
1939 /* unset allmulti level from slave */
1940 if (bond_dev->flags & IFF_ALLMULTI) {
1941 dev_set_allmulti(slave_dev, -1);
1942 }
1943
1944 /* flush master's mc_list from slave */
b9e40857 1945 netif_addr_lock_bh(bond_dev);
1da177e4 1946 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 1947 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1948 }
1949
1950 netdev_set_master(slave_dev, NULL);
1951
1952 /* close slave before restoring its mac address */
1953 dev_close(slave_dev);
1954
3915c1e8 1955 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1956 /* restore original ("permanent") mac address */
1957 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1958 addr.sa_family = slave_dev->type;
1959 dev_set_mac_address(slave_dev, &addr);
1960 }
1da177e4 1961
8f903c70 1962 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1963 IFF_SLAVE_INACTIVE | IFF_BONDING |
1964 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1965
1966 kfree(slave);
1967
1968 return 0; /* deletion OK */
1969}
1970
d90a162a
MS		 1971/*
1972* Destroy a bonding device.
1973* Must be under rtnl_lock when this function is called.
1974*/
1975void bond_destroy(struct bonding *bond)
1976{
1977 bond_deinit(bond->dev);
1978 bond_destroy_sysfs_entry(bond);
8cbdeec6 1979 unregister_netdevice(bond->dev);
d90a162a
MS		 1980}
1981
a434e43f
JV		 1982static void bond_destructor(struct net_device *bond_dev)
1983{
244ef9b9 1984 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 1985
1986 if (bond->wq)
1987 destroy_workqueue(bond->wq);
1988
1989 netif_addr_lock_bh(bond_dev);
1990 bond_mc_list_destroy(bond);
1991 netif_addr_unlock_bh(bond_dev);
1992
1993 free_netdev(bond_dev);
1994}
1995
d90a162a
MS		 1996/*
1997* First release a slave and than destroy the bond if no more slaves iare left.
1998* Must be under rtnl_lock when this function is called.
1999*/
2000int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2001{
454d7c9b 2002 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 2003 int ret;
2004
2005 ret = bond_release(bond_dev, slave_dev);
2006 if ((ret == 0) && (bond->slave_cnt == 0)) {
2007 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2008 bond_dev->name, bond_dev->name);
2009 bond_destroy(bond);
2010 }
2011 return ret;
2012}
2013
1da177e4
LT		 2014/*
2015 * This function releases all slaves.
2016 */
2017static int bond_release_all(struct net_device *bond_dev)
2018{
454d7c9b 2019 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2020 struct slave *slave;
2021 struct net_device *slave_dev;
2022 struct sockaddr addr;
2023
2024 write_lock_bh(&bond->lock);
2025
ff59c456
JV		 2026 netif_carrier_off(bond_dev);
2027
1da177e4
LT		 2028 if (bond->slave_cnt == 0) {
2029 goto out;
2030 }
2031
2032 bond->current_arp_slave = NULL;
2033 bond->primary_slave = NULL;
2034 bond_change_active_slave(bond, NULL);
2035
2036 while ((slave = bond->first_slave) != NULL) {
2037 /* Inform AD package of unbinding of slave
2038 * before slave is detached from the list.
2039 */
2040 if (bond->params.mode == BOND_MODE_8023AD) {
2041 bond_3ad_unbind_slave(slave);
2042 }
2043
2044 slave_dev = slave->dev;
2045 bond_detach_slave(bond, slave);
2046
2543331d
JV		 2047 /* now that the slave is detached, unlock and perform
2048 * all the undo steps that should not be called from
2049 * within a lock.
2050 */
2051 write_unlock_bh(&bond->lock);
2052
58402054 2053 if (bond_is_lb(bond)) {
1da177e4
LT		 2054 /* must be called only after the slave
2055 * has been detached from the list
2056 */
2057 bond_alb_deinit_slave(bond, slave);
2058 }
2059
8531c5ff
AK		 2060 bond_compute_features(bond);
2061
b76cdba9 2062 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2063 bond_del_vlans_from_slave(bond, slave_dev);
2064
2065 /* If the mode USES_PRIMARY, then we should only remove its
2066 * promisc and mc settings if it was the curr_active_slave, but that was
2067 * already taken care of above when we detached the slave
2068 */
2069 if (!USES_PRIMARY(bond->params.mode)) {
2070 /* unset promiscuity level from slave */
2071 if (bond_dev->flags & IFF_PROMISC) {
2072 dev_set_promiscuity(slave_dev, -1);
2073 }
2074
2075 /* unset allmulti level from slave */
2076 if (bond_dev->flags & IFF_ALLMULTI) {
2077 dev_set_allmulti(slave_dev, -1);
2078 }
2079
2080 /* flush master's mc_list from slave */
b9e40857 2081 netif_addr_lock_bh(bond_dev);
1da177e4 2082 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2083 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2084 }
2085
2086 netdev_set_master(slave_dev, NULL);
2087
2088 /* close slave before restoring its mac address */
2089 dev_close(slave_dev);
2090
dd957c57 2091 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2092 /* restore original ("permanent") mac address*/
2093 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2094 addr.sa_family = slave_dev->type;
2095 dev_set_mac_address(slave_dev, &addr);
2096 }
1da177e4 2097
8f903c70
JV		 2098 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2099 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2100
2101 kfree(slave);
2102
2103 /* re-acquire the lock before getting the next slave */
2104 write_lock_bh(&bond->lock);
2105 }
2106
2107 /* zero the mac address of the master so it will be
2108 * set by the application to the mac address of the
2109 * first slave
2110 */
2111 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2112
2113 if (list_empty(&bond->vlan_list)) {
2114 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2115 } else {
2116 printk(KERN_WARNING DRV_NAME
4e0952c7 2117 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2118 "still has VLANs.\n",
4e0952c7 2119 bond_dev->name, bond_dev->name);
1da177e4 2120 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2121 ": %s: When re-adding slaves, make sure the bond's "
2122 "HW address matches its VLANs'.\n",
2123 bond_dev->name);
1da177e4
LT		 2124 }
2125
2126 printk(KERN_INFO DRV_NAME
2127 ": %s: released all slaves\n",
2128 bond_dev->name);
2129
2130out:
2131 write_unlock_bh(&bond->lock);
2132
2133 return 0;
2134}
2135
2136/*
2137 * This function changes the active slave to slave <slave_dev>.
2138 * It returns -EINVAL in the following cases.
2139 * - <slave_dev> is not found in the list.
2140 * - There is not active slave now.
2141 * - <slave_dev> is already active.
2142 * - The link state of <slave_dev> is not BOND_LINK_UP.
2143 * - <slave_dev> is not running.
2144 * In these cases, this fuction does nothing.
2145 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2146 */
2147static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2148{
454d7c9b 2149 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2150 struct slave *old_active = NULL;
2151 struct slave *new_active = NULL;
2152 int res = 0;
2153
2154 if (!USES_PRIMARY(bond->params.mode)) {
2155 return -EINVAL;
2156 }
2157
2158 /* Verify that master_dev is indeed the master of slave_dev */
2159 if (!(slave_dev->flags & IFF_SLAVE) ||
2160 (slave_dev->master != bond_dev)) {
2161 return -EINVAL;
2162 }
2163
059fe7a5 2164 read_lock(&bond->lock);
1da177e4 2165
059fe7a5 2166 read_lock(&bond->curr_slave_lock);
1da177e4 2167 old_active = bond->curr_active_slave;
059fe7a5
JV		 2168 read_unlock(&bond->curr_slave_lock);
2169
1da177e4
LT		 2170 new_active = bond_get_slave_by_dev(bond, slave_dev);
2171
2172 /*
2173 * Changing to the current active: do nothing; return success.
2174 */
2175 if (new_active && (new_active == old_active)) {
059fe7a5 2176 read_unlock(&bond->lock);
1da177e4
LT		 2177 return 0;
2178 }
2179
2180 if ((new_active) &&
2181 (old_active) &&
2182 (new_active->link == BOND_LINK_UP) &&
2183 IS_UP(new_active->dev)) {
059fe7a5 2184 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2185 bond_change_active_slave(bond, new_active);
059fe7a5 2186 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2187 } else {
2188 res = -EINVAL;
2189 }
2190
059fe7a5 2191 read_unlock(&bond->lock);
1da177e4
LT		 2192
2193 return res;
2194}
2195
1da177e4
LT		 2196static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2197{
454d7c9b 2198 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2199
2200 info->bond_mode = bond->params.mode;
2201 info->miimon = bond->params.miimon;
2202
6603a6f2 2203 read_lock(&bond->lock);
1da177e4 2204 info->num_slaves = bond->slave_cnt;
6603a6f2 2205 read_unlock(&bond->lock);
1da177e4
LT		 2206
2207 return 0;
2208}
2209
2210static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2211{
454d7c9b 2212 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2213 struct slave *slave;
2214 int i, found = 0;
2215
2216 if (info->slave_id < 0) {
2217 return -ENODEV;
2218 }
2219
6603a6f2 2220 read_lock(&bond->lock);
1da177e4
LT		 2221
2222 bond_for_each_slave(bond, slave, i) {
2223 if (i == (int)info->slave_id) {
2224 found = 1;
2225 break;
2226 }
2227 }
2228
6603a6f2 2229 read_unlock(&bond->lock);
1da177e4
LT		 2230
2231 if (found) {
2232 strcpy(info->slave_name, slave->dev->name);
2233 info->link = slave->link;
2234 info->state = slave->state;
2235 info->link_failure_count = slave->link_failure_count;
2236 } else {
2237 return -ENODEV;
2238 }
2239
2240 return 0;
2241}
2242
2243/*-------------------------------- Monitoring -------------------------------*/
2244
1da177e4 2245
f0c76d61
JV		 2246static int bond_miimon_inspect(struct bonding *bond)
2247{
2248 struct slave *slave;
2249 int i, link_state, commit = 0;
1da177e4
LT		 2250
2251 bond_for_each_slave(bond, slave, i) {
f0c76d61 2252 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2253
f0c76d61 2254 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2255
2256 switch (slave->link) {
f0c76d61
JV		 2257 case BOND_LINK_UP:
2258 if (link_state)
2259 continue;
1da177e4 2260
f0c76d61
JV		 2261 slave->link = BOND_LINK_FAIL;
2262 slave->delay = bond->params.downdelay;
2263 if (slave->delay) {
2264 printk(KERN_INFO DRV_NAME
2265 ": %s: link status down for %s"
2266 "interface %s, disabling it in %d ms.\n",
2267 bond->dev->name,
2268 (bond->params.mode ==
2269 BOND_MODE_ACTIVEBACKUP) ?
2270 ((slave->state == BOND_STATE_ACTIVE) ?
2271 "active " : "backup ") : "",
2272 slave->dev->name,
2273 bond->params.downdelay * bond->params.miimon);
1da177e4 2274 }
f0c76d61
JV		 2275 /*FALLTHRU*/
2276 case BOND_LINK_FAIL:
2277 if (link_state) {
2278 /*
2279 * recovered before downdelay expired
2280 */
2281 slave->link = BOND_LINK_UP;
1da177e4
LT		 2282 slave->jiffies = jiffies;
2283 printk(KERN_INFO DRV_NAME
2284 ": %s: link status up again after %d "
2285 "ms for interface %s.\n",
1b76b316 2286 bond->dev->name,
f0c76d61
JV		 2287 (bond->params.downdelay - slave->delay) *
2288 bond->params.miimon,
2289 slave->dev->name);
2290 continue;
1da177e4 2291 }
f0c76d61
JV		 2292
2293 if (slave->delay <= 0) {
2294 slave->new_link = BOND_LINK_DOWN;
2295 commit++;
2296 continue;
1da177e4 2297 }
1da177e4 2298
f0c76d61
JV		 2299 slave->delay--;
2300 break;
2301
2302 case BOND_LINK_DOWN:
2303 if (!link_state)
2304 continue;
2305
2306 slave->link = BOND_LINK_BACK;
2307 slave->delay = bond->params.updelay;
2308
2309 if (slave->delay) {
2310 printk(KERN_INFO DRV_NAME
2311 ": %s: link status up for "
2312 "interface %s, enabling it in %d ms.\n",
2313 bond->dev->name, slave->dev->name,
2314 bond->params.updelay *
2315 bond->params.miimon);
2316 }
2317 /*FALLTHRU*/
2318 case BOND_LINK_BACK:
2319 if (!link_state) {
2320 slave->link = BOND_LINK_DOWN;
1da177e4
LT		 2321 printk(KERN_INFO DRV_NAME
2322 ": %s: link status down again after %d "
2323 "ms for interface %s.\n",
1b76b316 2324 bond->dev->name,
f0c76d61
JV		 2325 (bond->params.updelay - slave->delay) *
2326 bond->params.miimon,
2327 slave->dev->name);
2328
2329 continue;
2330 }
2331
2332 if (slave->delay <= 0) {
2333 slave->new_link = BOND_LINK_UP;
2334 commit++;
2335 continue;
1da177e4 2336 }
f0c76d61
JV		 2337
2338 slave->delay--;
1da177e4 2339 break;
f0c76d61
JV		 2340 }
2341 }
1da177e4 2342
f0c76d61
JV		 2343 return commit;
2344}
1da177e4 2345
f0c76d61
JV		 2346static void bond_miimon_commit(struct bonding *bond)
2347{
2348 struct slave *slave;
2349 int i;
2350
2351 bond_for_each_slave(bond, slave, i) {
2352 switch (slave->new_link) {
2353 case BOND_LINK_NOCHANGE:
2354 continue;
1da177e4 2355
f0c76d61
JV		 2356 case BOND_LINK_UP:
2357 slave->link = BOND_LINK_UP;
2358 slave->jiffies = jiffies;
2359
2360 if (bond->params.mode == BOND_MODE_8023AD) {
2361 /* prevent it from being the active one */
2362 slave->state = BOND_STATE_BACKUP;
2363 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2364 /* make it immediately active */
2365 slave->state = BOND_STATE_ACTIVE;
2366 } else if (slave != bond->primary_slave) {
2367 /* prevent it from being the active one */
2368 slave->state = BOND_STATE_BACKUP;
1da177e4 2369 }
1da177e4 2370
f0c76d61
JV		 2371 printk(KERN_INFO DRV_NAME
2372 ": %s: link status definitely "
2373 "up for interface %s.\n",
2374 bond->dev->name, slave->dev->name);
1da177e4 2375
f0c76d61
JV		 2376 /* notify ad that the link status has changed */
2377 if (bond->params.mode == BOND_MODE_8023AD)
2378 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2379
58402054 2380 if (bond_is_lb(bond))
f0c76d61
JV		 2381 bond_alb_handle_link_change(bond, slave,
2382 BOND_LINK_UP);
1da177e4 2383
f0c76d61
JV		 2384 if (!bond->curr_active_slave ||
2385 (slave == bond->primary_slave))
2386 goto do_failover;
1da177e4 2387
f0c76d61 2388 continue;
059fe7a5 2389
f0c76d61 2390 case BOND_LINK_DOWN:
fba4acda
JV		 2391 if (slave->link_failure_count < UINT_MAX)
2392 slave->link_failure_count++;
2393
f0c76d61 2394 slave->link = BOND_LINK_DOWN;
1da177e4 2395
f0c76d61
JV		 2396 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2397 bond->params.mode == BOND_MODE_8023AD)
2398 bond_set_slave_inactive_flags(slave);
2399
2400 printk(KERN_INFO DRV_NAME
2401 ": %s: link status definitely down for "
2402 "interface %s, disabling it\n",
2403 bond->dev->name, slave->dev->name);
2404
2405 if (bond->params.mode == BOND_MODE_8023AD)
2406 bond_3ad_handle_link_change(slave,
2407 BOND_LINK_DOWN);
2408
2409 if (bond->params.mode == BOND_MODE_TLB ||
2410 bond->params.mode == BOND_MODE_ALB)
2411 bond_alb_handle_link_change(bond, slave,
2412 BOND_LINK_DOWN);
2413
2414 if (slave == bond->curr_active_slave)
2415 goto do_failover;
2416
2417 continue;
2418
2419 default:
2420 printk(KERN_ERR DRV_NAME
2421 ": %s: invalid new link %d on slave %s\n",
2422 bond->dev->name, slave->new_link,
2423 slave->dev->name);
2424 slave->new_link = BOND_LINK_NOCHANGE;
2425
2426 continue;
2427 }
2428
2429do_failover:
2430 ASSERT_RTNL();
2431 write_lock_bh(&bond->curr_slave_lock);
2432 bond_select_active_slave(bond);
2433 write_unlock_bh(&bond->curr_slave_lock);
2434 }
2435
2436 bond_set_carrier(bond);
1da177e4
LT		 2437}
2438
0b0eef66
JV		 2439/*
2440 * bond_mii_monitor
2441 *
2442 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2443 * inspection, then (if inspection indicates something needs to be done)
2444 * an acquisition of appropriate locks followed by a commit phase to
2445 * implement whatever link state changes are indicated.
0b0eef66
JV		 2446 */
2447void bond_mii_monitor(struct work_struct *work)
2448{
2449 struct bonding *bond = container_of(work, struct bonding,
2450 mii_work.work);
0b0eef66
JV		 2451
2452 read_lock(&bond->lock);
f0c76d61
JV		 2453 if (bond->kill_timers)
2454 goto out;
2455
2456 if (bond->slave_cnt == 0)
2457 goto re_arm;
b59f9f74
JV		 2458
2459 if (bond->send_grat_arp) {
2460 read_lock(&bond->curr_slave_lock);
2461 bond_send_gratuitous_arp(bond);
2462 read_unlock(&bond->curr_slave_lock);
2463 }
2464
305d552a
BH		 2465 if (bond->send_unsol_na) {
2466 read_lock(&bond->curr_slave_lock);
2467 bond_send_unsolicited_na(bond);
2468 read_unlock(&bond->curr_slave_lock);
2469 }
2470
f0c76d61 2471 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2472 read_unlock(&bond->lock);
2473 rtnl_lock();
2474 read_lock(&bond->lock);
f0c76d61
JV		 2475
2476 bond_miimon_commit(bond);
2477
5655662d
JV		 2478 read_unlock(&bond->lock);
2479 rtnl_unlock(); /* might sleep, hold no other locks */
2480 read_lock(&bond->lock);
0b0eef66
JV		 2481 }
2482
f0c76d61
JV		 2483re_arm:
2484 if (bond->params.miimon)
2485 queue_delayed_work(bond->wq, &bond->mii_work,
2486 msecs_to_jiffies(bond->params.miimon));
2487out:
0b0eef66 2488 read_unlock(&bond->lock);
0b0eef66 2489}
c3ade5ca 2490
d3bb52b0 2491static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2492{
2493 struct in_device *idev;
2494 struct in_ifaddr *ifa;
a144ea4b 2495 __be32 addr = 0;
c3ade5ca
JV		 2496
2497 if (!dev)
2498 return 0;
2499
2500 rcu_read_lock();
e5ed6399 2501 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2502 if (!idev)
2503 goto out;
2504
2505 ifa = idev->ifa_list;
2506 if (!ifa)
2507 goto out;
2508
2509 addr = ifa->ifa_local;
2510out:
2511 rcu_read_unlock();
2512 return addr;
2513}
2514
d3bb52b0 2515static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2516{
0883beca 2517 struct vlan_entry *vlan;
f5b2b966
JV		 2518
2519 if (ip == bond->master_ip)
2520 return 1;
2521
0883beca 2522 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2523 if (ip == vlan->vlan_ip)
2524 return 1;
2525 }
2526
2527 return 0;
2528}
2529
c3ade5ca
JV		 2530/*
2531 * We go to the (large) trouble of VLAN tagging ARP frames because
2532 * switches in VLAN mode (especially if ports are configured as
2533 * "native" to a VLAN) might not pass non-tagged frames.
2534 */
d3bb52b0 2535static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2536{
2537 struct sk_buff *skb;
2538
2539 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2540 slave_dev->name, dest_ip, src_ip, vlan_id);
2541
2542 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2543 NULL, slave_dev->dev_addr, NULL);
2544
2545 if (!skb) {
2546 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2547 return;
2548 }
2549 if (vlan_id) {
2550 skb = vlan_put_tag(skb, vlan_id);
2551 if (!skb) {
2552 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2553 return;
2554 }
2555 }
2556 arp_xmit(skb);
2557}
2558
2559
1da177e4
LT		 2560static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2561{
c3ade5ca 2562 int i, vlan_id, rv;
d3bb52b0 2563 __be32 *targets = bond->params.arp_targets;
0883beca 2564 struct vlan_entry *vlan;
c3ade5ca
JV		 2565 struct net_device *vlan_dev;
2566 struct flowi fl;
2567 struct rtable *rt;
1da177e4 2568
6b780567
MW		 2569 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2570 if (!targets[i])
2571 continue;
c3ade5ca
JV		 2572 dprintk("basa: target %x\n", targets[i]);
2573 if (list_empty(&bond->vlan_list)) {
2574 dprintk("basa: empty vlan: arp_send\n");
2575 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2576 bond->master_ip, 0);
2577 continue;
2578 }
2579
2580 /*
2581 * If VLANs are configured, we do a route lookup to
2582 * determine which VLAN interface would be used, so we
2583 * can tag the ARP with the proper VLAN tag.
2584 */
2585 memset(&fl, 0, sizeof(fl));
2586 fl.fl4_dst = targets[i];
2587 fl.fl4_tos = RTO_ONLINK;
2588
f206351a 2589 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2590 if (rv) {
2591 if (net_ratelimit()) {
2592 printk(KERN_WARNING DRV_NAME
63779436
HH		 2593 ": %s: no route to arp_ip_target %pI4\n",
2594 bond->dev->name, &fl.fl4_dst);
c3ade5ca
JV		 2595 }
2596 continue;
2597 }
2598
2599 /*
2600 * This target is not on a VLAN
2601 */
2602 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2603 ip_rt_put(rt);
c3ade5ca
JV		 2604 dprintk("basa: rtdev == bond->dev: arp_send\n");
2605 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2606 bond->master_ip, 0);
2607 continue;
2608 }
2609
2610 vlan_id = 0;
0883beca 2611 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2612 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2613 if (vlan_dev == rt->u.dst.dev) {
2614 vlan_id = vlan->vlan_id;
2615 dprintk("basa: vlan match on %s %d\n",
2616 vlan_dev->name, vlan_id);
2617 break;
2618 }
2619 }
2620
2621 if (vlan_id) {
ed4b9f80 2622 ip_rt_put(rt);
c3ade5ca
JV		 2623 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2624 vlan->vlan_ip, vlan_id);
2625 continue;
2626 }
2627
2628 if (net_ratelimit()) {
2629 printk(KERN_WARNING DRV_NAME
63779436
HH		 2630 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2631 bond->dev->name, &fl.fl4_dst,
c3ade5ca
JV		 2632 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2633 }
ed4b9f80 2634 ip_rt_put(rt);
c3ade5ca
JV		 2635 }
2636}
2637
2638/*
2639 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2640 * for each VLAN above us.
b59f9f74
JV		 2641 *
2642 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2643 */
2644static void bond_send_gratuitous_arp(struct bonding *bond)
2645{
2646 struct slave *slave = bond->curr_active_slave;
2647 struct vlan_entry *vlan;
2648 struct net_device *vlan_dev;
2649
2650 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2651 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2652
2653 if (!slave || !bond->send_grat_arp ||
2654 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2655 return;
2656
b59f9f74
JV		 2657 bond->send_grat_arp--;
2658
c3ade5ca
JV		 2659 if (bond->master_ip) {
2660 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2661 bond->master_ip, 0);
c3ade5ca
JV		 2662 }
2663
2664 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2665 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2666 if (vlan->vlan_ip) {
2667 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2668 vlan->vlan_ip, vlan->vlan_id);
2669 }
1da177e4
LT		 2670 }
2671}
2672
d3bb52b0 2673static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2674{
2675 int i;
d3bb52b0 2676 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2677
2678 targets = bond->params.arp_targets;
2679 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
63779436
HH		 2680 dprintk("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2681 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
f5b2b966
JV		 2682 if (sip == targets[i]) {
2683 if (bond_has_this_ip(bond, tip))
2684 slave->last_arp_rx = jiffies;
2685 return;
2686 }
2687 }
2688}
2689
2690static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2691{
2692 struct arphdr *arp;
2693 struct slave *slave;
2694 struct bonding *bond;
2695 unsigned char *arp_ptr;
d3bb52b0 2696 __be32 sip, tip;
f5b2b966 2697
c346dca1 2698 if (dev_net(dev) != &init_net)
e730c155
EB		 2699 goto out;
2700
f5b2b966
JV		 2701 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2702 goto out;
2703
454d7c9b 2704 bond = netdev_priv(dev);
f5b2b966
JV		 2705 read_lock(&bond->lock);
2706
2707 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2708 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2709 orig_dev ? orig_dev->name : "NULL");
2710
2711 slave = bond_get_slave_by_dev(bond, orig_dev);
2712 if (!slave || !slave_do_arp_validate(bond, slave))
2713 goto out_unlock;
2714
988b7050 2715 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2716 goto out_unlock;
2717
d0a92be0 2718 arp = arp_hdr(skb);
f5b2b966
JV		 2719 if (arp->ar_hln != dev->addr_len ||
2720 skb->pkt_type == PACKET_OTHERHOST ||
2721 skb->pkt_type == PACKET_LOOPBACK ||
2722 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2723 arp->ar_pro != htons(ETH_P_IP) ||
2724 arp->ar_pln != 4)
2725 goto out_unlock;
2726
2727 arp_ptr = (unsigned char *)(arp + 1);
2728 arp_ptr += dev->addr_len;
2729 memcpy(&sip, arp_ptr, 4);
2730 arp_ptr += 4 + dev->addr_len;
2731 memcpy(&tip, arp_ptr, 4);
2732
63779436
HH		 2733 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2734 bond->dev->name, slave->dev->name, slave->state,
2735 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2736 &sip, &tip);
f5b2b966
JV		 2737
2738 /*
2739 * Backup slaves won't see the ARP reply, but do come through
2740 * here for each ARP probe (so we swap the sip/tip to validate
2741 * the probe). In a "redundant switch, common router" type of
2742 * configuration, the ARP probe will (hopefully) travel from
2743 * the active, through one switch, the router, then the other
2744 * switch before reaching the backup.
2745 */
2746 if (slave->state == BOND_STATE_ACTIVE)
2747 bond_validate_arp(bond, slave, sip, tip);
2748 else
2749 bond_validate_arp(bond, slave, tip, sip);
2750
2751out_unlock:
2752 read_unlock(&bond->lock);
2753out:
2754 dev_kfree_skb(skb);
2755 return NET_RX_SUCCESS;
2756}
2757
1da177e4
LT		 2758/*
2759 * this function is called regularly to monitor each slave's link
2760 * ensuring that traffic is being sent and received when arp monitoring
2761 * is used in load-balancing mode. if the adapter has been dormant, then an
2762 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2763 * arp monitoring in active backup mode.
2764 */
1b76b316 2765void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2766{
1b76b316
JV		 2767 struct bonding *bond = container_of(work, struct bonding,
2768 arp_work.work);
1da177e4
LT		 2769 struct slave *slave, *oldcurrent;
2770 int do_failover = 0;
2771 int delta_in_ticks;
2772 int i;
2773
2774 read_lock(&bond->lock);
2775
5ce0da8f 2776 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2777
2778 if (bond->kill_timers) {
2779 goto out;
2780 }
2781
2782 if (bond->slave_cnt == 0) {
2783 goto re_arm;
2784 }
2785
2786 read_lock(&bond->curr_slave_lock);
2787 oldcurrent = bond->curr_active_slave;
2788 read_unlock(&bond->curr_slave_lock);
2789
2790 /* see if any of the previous devices are up now (i.e. they have
2791 * xmt and rcv traffic). the curr_active_slave does not come into
2792 * the picture unless it is null. also, slave->jiffies is not needed
2793 * here because we send an arp on each slave and give a slave as
2794 * long as it needs to get the tx/rx within the delta.
2795 * TODO: what about up/down delay in arp mode? it wasn't here before
2796 * so it can wait
2797 */
2798 bond_for_each_slave(bond, slave, i) {
2799 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2800 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2801 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2802
2803 slave->link = BOND_LINK_UP;
2804 slave->state = BOND_STATE_ACTIVE;
2805
2806 /* primary_slave has no meaning in round-robin
2807 * mode. the window of a slave being up and
2808 * curr_active_slave being null after enslaving
2809 * is closed.
2810 */
2811 if (!oldcurrent) {
2812 printk(KERN_INFO DRV_NAME
2813 ": %s: link status definitely "
2814 "up for interface %s, ",
1b76b316 2815 bond->dev->name,
1da177e4
LT		 2816 slave->dev->name);
2817 do_failover = 1;
2818 } else {
2819 printk(KERN_INFO DRV_NAME
2820 ": %s: interface %s is now up\n",
1b76b316 2821 bond->dev->name,
1da177e4
LT		 2822 slave->dev->name);
2823 }
2824 }
2825 } else {
2826 /* slave->link == BOND_LINK_UP */
2827
2828 /* not all switches will respond to an arp request
2829 * when the source ip is 0, so don't take the link down
2830 * if we don't know our ip yet
2831 */
b63bb739 2832 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2833 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2834
2835 slave->link = BOND_LINK_DOWN;
2836 slave->state = BOND_STATE_BACKUP;
2837
2838 if (slave->link_failure_count < UINT_MAX) {
2839 slave->link_failure_count++;
2840 }
2841
2842 printk(KERN_INFO DRV_NAME
2843 ": %s: interface %s is now down.\n",
1b76b316 2844 bond->dev->name,
1da177e4
LT		 2845 slave->dev->name);
2846
2847 if (slave == oldcurrent) {
2848 do_failover = 1;
2849 }
2850 }
2851 }
2852
2853 /* note: if switch is in round-robin mode, all links
2854 * must tx arp to ensure all links rx an arp - otherwise
2855 * links may oscillate or not come up at all; if switch is
2856 * in something like xor mode, there is nothing we can
2857 * do - all replies will be rx'ed on same link causing slaves
2858 * to be unstable during low/no traffic periods
2859 */
2860 if (IS_UP(slave->dev)) {
2861 bond_arp_send_all(bond, slave);
2862 }
2863 }
2864
2865 if (do_failover) {
059fe7a5 2866 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2867
2868 bond_select_active_slave(bond);
2869
059fe7a5 2870 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2871 }
2872
2873re_arm:
1b76b316
JV		 2874 if (bond->params.arp_interval)
2875 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2876out:
2877 read_unlock(&bond->lock);
2878}
2879
2880/*
b2220cad
JV		 2881 * Called to inspect slaves for active-backup mode ARP monitor link state
2882 * changes. Sets new_link in slaves to specify what action should take
2883 * place for the slave. Returns 0 if no changes are found, >0 if changes
2884 * to link states must be committed.
2885 *
2886 * Called with bond->lock held for read.
1da177e4 2887 */
b2220cad 2888static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2889{
1da177e4 2890 struct slave *slave;
b2220cad 2891 int i, commit = 0;
1da177e4 2892
b2220cad
JV		 2893 bond_for_each_slave(bond, slave, i) {
2894 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2895
b2220cad
JV		 2896 if (slave->link != BOND_LINK_UP) {
2897 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2898 delta_in_ticks)) {
2899 slave->new_link = BOND_LINK_UP;
2900 commit++;
2901 }
1da177e4 2902
b2220cad
JV		 2903 continue;
2904 }
1da177e4 2905
b2220cad
JV		 2906 /*
2907 * Give slaves 2*delta after being enslaved or made
2908 * active. This avoids bouncing, as the last receive
2909 * times need a full ARP monitor cycle to be updated.
2910 */
2911 if (!time_after_eq(jiffies, slave->jiffies +
2912 2 * delta_in_ticks))
2913 continue;
2914
2915 /*
2916 * Backup slave is down if:
2917 * - No current_arp_slave AND
2918 * - more than 3*delta since last receive AND
2919 * - the bond has an IP address
2920 *
2921 * Note: a non-null current_arp_slave indicates
2922 * the curr_active_slave went down and we are
2923 * searching for a new one; under this condition
2924 * we only take the curr_active_slave down - this
2925 * gives each slave a chance to tx/rx traffic
2926 * before being taken out
2927 */
2928 if (slave->state == BOND_STATE_BACKUP &&
2929 !bond->current_arp_slave &&
2930 time_after(jiffies, slave_last_rx(bond, slave) +
2931 3 * delta_in_ticks)) {
2932 slave->new_link = BOND_LINK_DOWN;
2933 commit++;
2934 }
2935
2936 /*
2937 * Active slave is down if:
2938 * - more than 2*delta since transmitting OR
2939 * - (more than 2*delta since receive AND
2940 * the bond has an IP address)
2941 */
2942 if ((slave->state == BOND_STATE_ACTIVE) &&
2943 (time_after_eq(jiffies, slave->dev->trans_start +
2944 2 * delta_in_ticks) ||
2945 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2946 + 2 * delta_in_ticks)))) {
2947 slave->new_link = BOND_LINK_DOWN;
2948 commit++;
2949 }
1da177e4
LT		 2950 }
2951
b2220cad
JV		 2952 read_lock(&bond->curr_slave_lock);
2953
2954 /*
2955 * Trigger a commit if the primary option setting has changed.
1da177e4 2956 */
b2220cad
JV		 2957 if (bond->primary_slave &&
2958 (bond->primary_slave != bond->curr_active_slave) &&
2959 (bond->primary_slave->link == BOND_LINK_UP))
2960 commit++;
1da177e4 2961
b2220cad 2962 read_unlock(&bond->curr_slave_lock);
1da177e4 2963
b2220cad
JV		 2964 return commit;
2965}
1da177e4 2966
b2220cad
JV		 2967/*
2968 * Called to commit link state changes noted by inspection step of
2969 * active-backup mode ARP monitor.
2970 *
2971 * Called with RTNL and bond->lock for read.
2972 */
2973static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2974{
2975 struct slave *slave;
2976 int i;
1da177e4 2977
b2220cad
JV		 2978 bond_for_each_slave(bond, slave, i) {
2979 switch (slave->new_link) {
2980 case BOND_LINK_NOCHANGE:
2981 continue;
ff59c456 2982
b2220cad
JV		 2983 case BOND_LINK_UP:
2984 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2985
b2220cad
JV		 2986 if (!bond->curr_active_slave &&
2987 time_before_eq(jiffies, slave->dev->trans_start +
2988 delta_in_ticks)) {
2989 slave->link = BOND_LINK_UP;
2990 bond_change_active_slave(bond, slave);
2991 bond->current_arp_slave = NULL;
2992
2993 printk(KERN_INFO DRV_NAME
2994 ": %s: %s is up and now the "
2995 "active interface\n",
2996 bond->dev->name, slave->dev->name);
2997
2998 } else if (bond->curr_active_slave != slave) {
2999 /* this slave has just come up but we
3000 * already have a current slave; this can
3001 * also happen if bond_enslave adds a new
3002 * slave that is up while we are searching
3003 * for a new slave
1da177e4 3004 */
b2220cad
JV		 3005 slave->link = BOND_LINK_UP;
3006 bond_set_slave_inactive_flags(slave);
3007 bond->current_arp_slave = NULL;
3008
3009 printk(KERN_INFO DRV_NAME
3010 ": %s: backup interface %s is now up\n",
3011 bond->dev->name, slave->dev->name);
3012 }
1da177e4 3013
b2220cad 3014 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3015
b2220cad 3016 break;
1da177e4 3017
b2220cad
JV		 3018 case BOND_LINK_DOWN:
3019 if (slave->link_failure_count < UINT_MAX)
3020 slave->link_failure_count++;
3021
3022 slave->link = BOND_LINK_DOWN;
3023
3024 if (slave == bond->curr_active_slave) {
3025 printk(KERN_INFO DRV_NAME
3026 ": %s: link status down for active "
3027 "interface %s, disabling it\n",
3028 bond->dev->name, slave->dev->name);
1da177e4
LT		 3029
3030 bond_set_slave_inactive_flags(slave);
3031
b2220cad
JV		 3032 write_lock_bh(&bond->curr_slave_lock);
3033
3034 bond_select_active_slave(bond);
3035 if (bond->curr_active_slave)
3036 bond->curr_active_slave->jiffies =
3037 jiffies;
3038
3039 write_unlock_bh(&bond->curr_slave_lock);
3040
3041 bond->current_arp_slave = NULL;
3042
3043 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3044 printk(KERN_INFO DRV_NAME
3045 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3046 bond->dev->name, slave->dev->name);
3047
3048 bond_set_slave_inactive_flags(slave);
1da177e4 3049 }
b2220cad
JV		 3050 break;
3051
3052 default:
3053 printk(KERN_ERR DRV_NAME
3054 ": %s: impossible: new_link %d on slave %s\n",
3055 bond->dev->name, slave->new_link,
3056 slave->dev->name);
1da177e4
LT		 3057 }
3058 }
3059
b2220cad
JV		 3060 /*
3061 * No race with changes to primary via sysfs, as we hold rtnl.
3062 */
3063 if (bond->primary_slave &&
3064 (bond->primary_slave != bond->curr_active_slave) &&
3065 (bond->primary_slave->link == BOND_LINK_UP)) {
3066 write_lock_bh(&bond->curr_slave_lock);
3067 bond_change_active_slave(bond, bond->primary_slave);
3068 write_unlock_bh(&bond->curr_slave_lock);
3069 }
1da177e4 3070
b2220cad
JV		 3071 bond_set_carrier(bond);
3072}
1da177e4 3073
b2220cad
JV		 3074/*
3075 * Send ARP probes for active-backup mode ARP monitor.
3076 *
3077 * Called with bond->lock held for read.
3078 */
3079static void bond_ab_arp_probe(struct bonding *bond)
3080{
3081 struct slave *slave;
3082 int i;
1da177e4 3083
b2220cad 3084 read_lock(&bond->curr_slave_lock);
1da177e4 3085
b2220cad
JV		 3086 if (bond->current_arp_slave && bond->curr_active_slave)
3087 printk("PROBE: c_arp %s && cas %s BAD\n",
3088 bond->current_arp_slave->dev->name,
3089 bond->curr_active_slave->dev->name);
1da177e4 3090
b2220cad
JV		 3091 if (bond->curr_active_slave) {
3092 bond_arp_send_all(bond, bond->curr_active_slave);
3093 read_unlock(&bond->curr_slave_lock);
3094 return;
3095 }
1da177e4 3096
b2220cad 3097 read_unlock(&bond->curr_slave_lock);
059fe7a5 3098
b2220cad
JV		 3099 /* if we don't have a curr_active_slave, search for the next available
3100 * backup slave from the current_arp_slave and make it the candidate
3101 * for becoming the curr_active_slave
3102 */
1da177e4 3103
b2220cad
JV		 3104 if (!bond->current_arp_slave) {
3105 bond->current_arp_slave = bond->first_slave;
3106 if (!bond->current_arp_slave)
3107 return;
3108 }
1da177e4 3109
b2220cad 3110 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3111
b2220cad
JV		 3112 /* search for next candidate */
3113 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3114 if (IS_UP(slave->dev)) {
3115 slave->link = BOND_LINK_BACK;
3116 bond_set_slave_active_flags(slave);
3117 bond_arp_send_all(bond, slave);
1da177e4 3118 slave->jiffies = jiffies;
b2220cad
JV		 3119 bond->current_arp_slave = slave;
3120 break;
1da177e4
LT		 3121 }
3122
b2220cad
JV		 3123 /* if the link state is up at this point, we
3124 * mark it down - this can happen if we have
3125 * simultaneous link failures and
3126 * reselect_active_interface doesn't make this
3127 * one the current slave so it is still marked
3128 * up when it is actually down
1da177e4 3129 */
b2220cad
JV		 3130 if (slave->link == BOND_LINK_UP) {
3131 slave->link = BOND_LINK_DOWN;
3132 if (slave->link_failure_count < UINT_MAX)
3133 slave->link_failure_count++;
1da177e4 3134
b2220cad
JV		 3135 bond_set_slave_inactive_flags(slave);
3136
3137 printk(KERN_INFO DRV_NAME
3138 ": %s: backup interface %s is now down.\n",
3139 bond->dev->name, slave->dev->name);
1da177e4 3140 }
b2220cad
JV		 3141 }
3142}
1da177e4 3143
b2220cad
JV		 3144void bond_activebackup_arp_mon(struct work_struct *work)
3145{
3146 struct bonding *bond = container_of(work, struct bonding,
3147 arp_work.work);
3148 int delta_in_ticks;
1da177e4 3149
b2220cad 3150 read_lock(&bond->lock);
1da177e4 3151
b2220cad
JV		 3152 if (bond->kill_timers)
3153 goto out;
1da177e4 3154
b2220cad 3155 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3156
b2220cad
JV		 3157 if (bond->slave_cnt == 0)
3158 goto re_arm;
3159
b59f9f74
JV		 3160 if (bond->send_grat_arp) {
3161 read_lock(&bond->curr_slave_lock);
3162 bond_send_gratuitous_arp(bond);
3163 read_unlock(&bond->curr_slave_lock);
3164 }
3165
305d552a
BH		 3166 if (bond->send_unsol_na) {
3167 read_lock(&bond->curr_slave_lock);
3168 bond_send_unsolicited_na(bond);
3169 read_unlock(&bond->curr_slave_lock);
3170 }
3171
b2220cad
JV		 3172 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3173 read_unlock(&bond->lock);
3174 rtnl_lock();
3175 read_lock(&bond->lock);
3176
3177 bond_ab_arp_commit(bond, delta_in_ticks);
3178
3179 read_unlock(&bond->lock);
3180 rtnl_unlock();
3181 read_lock(&bond->lock);
1da177e4
LT		 3182 }
3183
b2220cad
JV		 3184 bond_ab_arp_probe(bond);
3185
1da177e4
LT		 3186re_arm:
3187 if (bond->params.arp_interval) {
1b76b316 3188 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3189 }
3190out:
3191 read_unlock(&bond->lock);
3192}
3193
3194/*------------------------------ proc/seq_file-------------------------------*/
3195
3196#ifdef CONFIG_PROC_FS
3197
1da177e4
LT		 3198static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3199{
3200 struct bonding *bond = seq->private;
3201 loff_t off = 0;
3202 struct slave *slave;
3203 int i;
3204
3205 /* make sure the bond won't be taken away */
3206 read_lock(&dev_base_lock);
6603a6f2 3207 read_lock(&bond->lock);
1da177e4
LT		 3208
3209 if (*pos == 0) {
3210 return SEQ_START_TOKEN;
3211 }
3212
3213 bond_for_each_slave(bond, slave, i) {
3214 if (++off == *pos) {
3215 return slave;
3216 }
3217 }
3218
3219 return NULL;
3220}
3221
3222static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3223{
3224 struct bonding *bond = seq->private;
3225 struct slave *slave = v;
3226
3227 ++*pos;
3228 if (v == SEQ_START_TOKEN) {
3229 return bond->first_slave;
3230 }
3231
3232 slave = slave->next;
3233
3234 return (slave == bond->first_slave) ? NULL : slave;
3235}
3236
3237static void bond_info_seq_stop(struct seq_file *seq, void *v)
3238{
3239 struct bonding *bond = seq->private;
3240
6603a6f2 3241 read_unlock(&bond->lock);
1da177e4
LT		 3242 read_unlock(&dev_base_lock);
3243}
3244
3245static void bond_info_show_master(struct seq_file *seq)
3246{
3247 struct bonding *bond = seq->private;
3248 struct slave *curr;
4756b02f 3249 int i;
1da177e4
LT		 3250
3251 read_lock(&bond->curr_slave_lock);
3252 curr = bond->curr_active_slave;
3253 read_unlock(&bond->curr_slave_lock);
3254
dd957c57 3255 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3256 bond_mode_name(bond->params.mode));
3257
dd957c57
JV		 3258 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3259 bond->params.fail_over_mac)
3915c1e8
JV		 3260 seq_printf(seq, " (fail_over_mac %s)",
3261 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3262
3263 seq_printf(seq, "\n");
3264
c61b75ad
MW		 3265 if (bond->params.mode == BOND_MODE_XOR ||
3266 bond->params.mode == BOND_MODE_8023AD) {
3267 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3268 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3269 bond->params.xmit_policy);
3270 }
3271
1da177e4
LT		 3272 if (USES_PRIMARY(bond->params.mode)) {
3273 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3274 (bond->primary_slave) ?
3275 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3276
3277 seq_printf(seq, "Currently Active Slave: %s\n",
3278 (curr) ? curr->dev->name : "None");
3279 }
3280
ff59c456
JV		 3281 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3282 "up" : "down");
1da177e4
LT		 3283 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3284 seq_printf(seq, "Up Delay (ms): %d\n",
3285 bond->params.updelay * bond->params.miimon);
3286 seq_printf(seq, "Down Delay (ms): %d\n",
3287 bond->params.downdelay * bond->params.miimon);
3288
4756b02f
MW		 3289
3290 /* ARP information */
3291 if(bond->params.arp_interval > 0) {
3292 int printed=0;
3293 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3294 bond->params.arp_interval);
3295
3296 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3297
3298 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3299 if (!bond->params.arp_targets[i])
3300 continue;
3301 if (printed)
3302 seq_printf(seq, ",");
8cf14e38 3303 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
4756b02f
MW		 3304 printed = 1;
3305 }
3306 seq_printf(seq, "\n");
3307 }
3308
1da177e4
LT		 3309 if (bond->params.mode == BOND_MODE_8023AD) {
3310 struct ad_info ad_info;
3311
3312 seq_puts(seq, "\n802.3ad info\n");
3313 seq_printf(seq, "LACP rate: %s\n",
3314 (bond->params.lacp_fast) ? "fast" : "slow");
fd989c83
JV		 3315 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3316 ad_select_tbl[bond->params.ad_select].modename);
1da177e4
LT		 3317
3318 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3319 seq_printf(seq, "bond %s has no active aggregator\n",
3320 bond->dev->name);
3321 } else {
3322 seq_printf(seq, "Active Aggregator Info:\n");
3323
3324 seq_printf(seq, "\tAggregator ID: %d\n",
3325 ad_info.aggregator_id);
3326 seq_printf(seq, "\tNumber of ports: %d\n",
3327 ad_info.ports);
3328 seq_printf(seq, "\tActor Key: %d\n",
3329 ad_info.actor_key);
3330 seq_printf(seq, "\tPartner Key: %d\n",
3331 ad_info.partner_key);
e174961c
JB		 3332 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3333 ad_info.partner_system);
1da177e4
LT		 3334 }
3335 }
3336}
3337
3338static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3339{
3340 struct bonding *bond = seq->private;
3341
3342 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3343 seq_printf(seq, "MII Status: %s\n",
3344 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3345 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3346 slave->link_failure_count);
3347
e174961c 3348 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
1da177e4
LT		 3349
3350 if (bond->params.mode == BOND_MODE_8023AD) {
3351 const struct aggregator *agg
3352 = SLAVE_AD_INFO(slave).port.aggregator;
3353
3354 if (agg) {
3355 seq_printf(seq, "Aggregator ID: %d\n",
3356 agg->aggregator_identifier);
3357 } else {
3358 seq_puts(seq, "Aggregator ID: N/A\n");
3359 }
3360 }
3361}
3362
3363static int bond_info_seq_show(struct seq_file *seq, void *v)
3364{
3365 if (v == SEQ_START_TOKEN) {
3366 seq_printf(seq, "%s\n", version);
3367 bond_info_show_master(seq);
3368 } else {
3369 bond_info_show_slave(seq, v);
3370 }
3371
3372 return 0;
3373}
3374
3375static struct seq_operations bond_info_seq_ops = {
3376 .start = bond_info_seq_start,
3377 .next = bond_info_seq_next,
3378 .stop = bond_info_seq_stop,
3379 .show = bond_info_seq_show,
3380};
3381
3382static int bond_info_open(struct inode *inode, struct file *file)
3383{
3384 struct seq_file *seq;
3385 struct proc_dir_entry *proc;
3386 int res;
3387
3388 res = seq_open(file, &bond_info_seq_ops);
3389 if (!res) {
3390 /* recover the pointer buried in proc_dir_entry data */
3391 seq = file->private_data;
3392 proc = PDE(inode);
3393 seq->private = proc->data;
3394 }
3395
3396 return res;
3397}
3398
d54b1fdb 3399static const struct file_operations bond_info_fops = {
1da177e4
LT		 3400 .owner = THIS_MODULE,
3401 .open = bond_info_open,
3402 .read = seq_read,
3403 .llseek = seq_lseek,
3404 .release = seq_release,
3405};
3406
3407static int bond_create_proc_entry(struct bonding *bond)
3408{
3409 struct net_device *bond_dev = bond->dev;
3410
3411 if (bond_proc_dir) {
a95609cb
DL		 3412 bond->proc_entry = proc_create_data(bond_dev->name,
3413 S_IRUGO, bond_proc_dir,
3414 &bond_info_fops, bond);
1da177e4
LT		 3415 if (bond->proc_entry == NULL) {
3416 printk(KERN_WARNING DRV_NAME
3417 ": Warning: Cannot create /proc/net/%s/%s\n",
3418 DRV_NAME, bond_dev->name);
3419 } else {
1da177e4
LT		 3420 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3421 }
3422 }
3423
3424 return 0;
3425}
3426
3427static void bond_remove_proc_entry(struct bonding *bond)
3428{
3429 if (bond_proc_dir && bond->proc_entry) {
3430 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3431 memset(bond->proc_file_name, 0, IFNAMSIZ);
3432 bond->proc_entry = NULL;
3433 }
3434}
3435
3436/* Create the bonding directory under /proc/net, if doesn't exist yet.
3437 * Caller must hold rtnl_lock.
3438 */
3439static void bond_create_proc_dir(void)
3440{
3441 int len = strlen(DRV_NAME);
3442
457c4cbc 3443 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3444 bond_proc_dir = bond_proc_dir->next) {
3445 if ((bond_proc_dir->namelen == len) &&
3446 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3447 break;
3448 }
3449 }
3450
3451 if (!bond_proc_dir) {
457c4cbc 3452 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3453 if (bond_proc_dir) {
3454 bond_proc_dir->owner = THIS_MODULE;
3455 } else {
3456 printk(KERN_WARNING DRV_NAME
3457 ": Warning: cannot create /proc/net/%s\n",
3458 DRV_NAME);
3459 }
3460 }
3461}
3462
3463/* Destroy the bonding directory under /proc/net, if empty.
3464 * Caller must hold rtnl_lock.
3465 */
3466static void bond_destroy_proc_dir(void)
3467{
3468 struct proc_dir_entry *de;
3469
3470 if (!bond_proc_dir) {
3471 return;
3472 }
3473
3474 /* verify that the /proc dir is empty */
3475 for (de = bond_proc_dir->subdir; de; de = de->next) {
3476 /* ignore . and .. */
3477 if (*(de->name) != '.') {
3478 break;
3479 }
3480 }
3481
3482 if (de) {
3483 if (bond_proc_dir->owner == THIS_MODULE) {
3484 bond_proc_dir->owner = NULL;
3485 }
3486 } else {
457c4cbc 3487 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3488 bond_proc_dir = NULL;
3489 }
3490}
3491#endif /* CONFIG_PROC_FS */
3492
3493/*-------------------------- netdev event handling --------------------------*/
3494
3495/*
3496 * Change device name
3497 */
3498static int bond_event_changename(struct bonding *bond)
3499{
3500#ifdef CONFIG_PROC_FS
3501 bond_remove_proc_entry(bond);
3502 bond_create_proc_entry(bond);
3503#endif
b76cdba9
MW		 3504 down_write(&(bonding_rwsem));
3505 bond_destroy_sysfs_entry(bond);
3506 bond_create_sysfs_entry(bond);
3507 up_write(&(bonding_rwsem));
1da177e4
LT		 3508 return NOTIFY_DONE;
3509}
3510
3511static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3512{
454d7c9b 3513 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 3514
3515 switch (event) {
3516 case NETDEV_CHANGENAME:
3517 return bond_event_changename(event_bond);
3518 case NETDEV_UNREGISTER:
3b96c858 3519 bond_release_all(event_bond->dev);
1da177e4
LT		 3520 break;
3521 default:
3522 break;
3523 }
3524
3525 return NOTIFY_DONE;
3526}
3527
3528static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3529{
3530 struct net_device *bond_dev = slave_dev->master;
454d7c9b 3531 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3532
3533 switch (event) {
3534 case NETDEV_UNREGISTER:
3535 if (bond_dev) {
1284cd3a
JV		 3536 if (bond->setup_by_slave)
3537 bond_release_and_destroy(bond_dev, slave_dev);
3538 else
3539 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3540 }
3541 break;
3542 case NETDEV_CHANGE:
3543 /*
3544 * TODO: is this what we get if somebody
3545 * sets up a hierarchical bond, then rmmod's
3546 * one of the slave bonding devices?
3547 */
3548 break;
3549 case NETDEV_DOWN:
3550 /*
3551 * ... Or is it this?
3552 */
3553 break;
3554 case NETDEV_CHANGEMTU:
3555 /*
3556 * TODO: Should slaves be allowed to
3557 * independently alter their MTU? For
3558 * an active-backup bond, slaves need
3559 * not be the same type of device, so
3560 * MTUs may vary. For other modes,
3561 * slaves arguably should have the
3562 * same MTUs. To do this, we'd need to
3563 * take over the slave's change_mtu
3564 * function for the duration of their
3565 * servitude.
3566 */
3567 break;
3568 case NETDEV_CHANGENAME:
3569 /*
3570 * TODO: handle changing the primary's name
3571 */
3572 break;
8531c5ff
AK		 3573 case NETDEV_FEAT_CHANGE:
3574 bond_compute_features(bond);
3575 break;
1da177e4
LT		 3576 default:
3577 break;
3578 }
3579
3580 return NOTIFY_DONE;
3581}
3582
3583/*
3584 * bond_netdev_event: handle netdev notifier chain events.
3585 *
3586 * This function receives events for the netdev chain. The caller (an
e041c683 3587 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3588 * locks for us to safely manipulate the slave devices (RTNL lock,
3589 * dev_probe_lock).
3590 */
3591static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3592{
3593 struct net_device *event_dev = (struct net_device *)ptr;
3594
c346dca1 3595 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3596 return NOTIFY_DONE;
3597
1da177e4
LT		 3598 dprintk("event_dev: %s, event: %lx\n",
3599 (event_dev ? event_dev->name : "None"),
3600 event);
3601
0b680e75
JV		 3602 if (!(event_dev->priv_flags & IFF_BONDING))
3603 return NOTIFY_DONE;
3604
1da177e4
LT		 3605 if (event_dev->flags & IFF_MASTER) {
3606 dprintk("IFF_MASTER\n");
3607 return bond_master_netdev_event(event, event_dev);
3608 }
3609
3610 if (event_dev->flags & IFF_SLAVE) {
3611 dprintk("IFF_SLAVE\n");
3612 return bond_slave_netdev_event(event, event_dev);
3613 }
3614
3615 return NOTIFY_DONE;
3616}
3617
c3ade5ca
JV		 3618/*
3619 * bond_inetaddr_event: handle inetaddr notifier chain events.
3620 *
3621 * We keep track of device IPs primarily to use as source addresses in
3622 * ARP monitor probes (rather than spewing out broadcasts all the time).
3623 *
3624 * We track one IP for the main device (if it has one), plus one per VLAN.
3625 */
3626static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3627{
3628 struct in_ifaddr *ifa = ptr;
3629 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3630 struct bonding *bond;
3631 struct vlan_entry *vlan;
c3ade5ca 3632
c346dca1 3633 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3634 return NOTIFY_DONE;
3635
0883beca 3636 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3637 if (bond->dev == event_dev) {
3638 switch (event) {
3639 case NETDEV_UP:
3640 bond->master_ip = ifa->ifa_local;
3641 return NOTIFY_OK;
3642 case NETDEV_DOWN:
3643 bond->master_ip = bond_glean_dev_ip(bond->dev);
3644 return NOTIFY_OK;
3645 default:
3646 return NOTIFY_DONE;
3647 }
3648 }
3649
0883beca 3650 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3651 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3652 if (vlan_dev == event_dev) {
3653 switch (event) {
3654 case NETDEV_UP:
3655 vlan->vlan_ip = ifa->ifa_local;
3656 return NOTIFY_OK;
3657 case NETDEV_DOWN:
3658 vlan->vlan_ip =
3659 bond_glean_dev_ip(vlan_dev);
3660 return NOTIFY_OK;
3661 default:
3662 return NOTIFY_DONE;
3663 }
3664 }
3665 }
3666 }
3667 return NOTIFY_DONE;
3668}
3669
1da177e4
LT		 3670static struct notifier_block bond_netdev_notifier = {
3671 .notifier_call = bond_netdev_event,
3672};
3673
c3ade5ca
JV		 3674static struct notifier_block bond_inetaddr_notifier = {
3675 .notifier_call = bond_inetaddr_event,
3676};
3677
1da177e4
LT		 3678/*-------------------------- Packet type handling ---------------------------*/
3679
3680/* register to receive lacpdus on a bond */
3681static void bond_register_lacpdu(struct bonding *bond)
3682{
3683 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3684
3685 /* initialize packet type */
3686 pk_type->type = PKT_TYPE_LACPDU;
3687 pk_type->dev = bond->dev;
3688 pk_type->func = bond_3ad_lacpdu_recv;
3689
3690 dev_add_pack(pk_type);
3691}
3692
3693/* unregister to receive lacpdus on a bond */
3694static void bond_unregister_lacpdu(struct bonding *bond)
3695{
3696 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3697}
3698
f5b2b966
JV		 3699void bond_register_arp(struct bonding *bond)
3700{
3701 struct packet_type *pt = &bond->arp_mon_pt;
3702
c4f283b1
JV		 3703 if (pt->type)
3704 return;
3705
f5b2b966 3706 pt->type = htons(ETH_P_ARP);
e245cb71 3707 pt->dev = bond->dev;
f5b2b966
JV		 3708 pt->func = bond_arp_rcv;
3709 dev_add_pack(pt);
3710}
3711
3712void bond_unregister_arp(struct bonding *bond)
3713{
c4f283b1
JV		 3714 struct packet_type *pt = &bond->arp_mon_pt;
3715
3716 dev_remove_pack(pt);
3717 pt->type = 0;
f5b2b966
JV		 3718}
3719
169a3e66
JV		 3720/*---------------------------- Hashing Policies -----------------------------*/
3721
6f6652be
JV		 3722/*
3723 * Hash for the output device based upon layer 2 and layer 3 data. If
3724 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3725 */
3726static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3727 struct net_device *bond_dev, int count)
3728{
3729 struct ethhdr *data = (struct ethhdr *)skb->data;
3730 struct iphdr *iph = ip_hdr(skb);
3731
f14c4e4e 3732 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be
JV		 3733 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3734 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3735 }
3736
3737 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3738}
3739
169a3e66 3740/*
59c51591 3741 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3742 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3743 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3744 */
3745static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3746 struct net_device *bond_dev, int count)
3747{
3748 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3749 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3750 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3751 int layer4_xor = 0;
3752
f14c4e4e
BH		 3753 if (skb->protocol == htons(ETH_P_IP)) {
3754 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3755 (iph->protocol == IPPROTO_TCP ||
3756 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3757 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3758 }
3759 return (layer4_xor ^
3760 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3761
3762 }
3763
3764 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3765}
3766
3767/*
3768 * Hash for the output device based upon layer 2 data
3769 */
3770static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3771 struct net_device *bond_dev, int count)
3772{
3773 struct ethhdr *data = (struct ethhdr *)skb->data;
3774
3775 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3776}
3777
1da177e4
LT		 3778/*-------------------------- Device entry points ----------------------------*/
3779
3780static int bond_open(struct net_device *bond_dev)
3781{
454d7c9b 3782 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3783
3784 bond->kill_timers = 0;
3785
58402054 3786 if (bond_is_lb(bond)) {
1da177e4
LT		 3787 /* bond_alb_initialize must be called before the timer
3788 * is started.
3789 */
3790 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3791 /* something went wrong - fail the open operation */
3792 return -1;
3793 }
3794
1b76b316
JV		 3795 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3796 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3797 }
3798
3799 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3800 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3801 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3802 }
3803
3804 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3805 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3806 INIT_DELAYED_WORK(&bond->arp_work,
3807 bond_activebackup_arp_mon);
3808 else
3809 INIT_DELAYED_WORK(&bond->arp_work,
3810 bond_loadbalance_arp_mon);
3811
3812 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3813 if (bond->params.arp_validate)
3814 bond_register_arp(bond);
1da177e4
LT		 3815 }
3816
3817 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3818 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3819 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3820 /* register to receive LACPDUs */
3821 bond_register_lacpdu(bond);
fd989c83 3822 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3823 }
3824
3825 return 0;
3826}
3827
3828static int bond_close(struct net_device *bond_dev)
3829{
454d7c9b 3830 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3831
3832 if (bond->params.mode == BOND_MODE_8023AD) {
3833 /* Unregister the receive of LACPDUs */
3834 bond_unregister_lacpdu(bond);
3835 }
3836
f5b2b966
JV		 3837 if (bond->params.arp_validate)
3838 bond_unregister_arp(bond);
3839
1da177e4
LT		 3840 write_lock_bh(&bond->lock);
3841
b59f9f74 3842 bond->send_grat_arp = 0;
305d552a 3843 bond->send_unsol_na = 0;
1da177e4
LT		 3844
3845 /* signal timers not to re-arm */
3846 bond->kill_timers = 1;
3847
3848 write_unlock_bh(&bond->lock);
3849
1da177e4 3850 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3851 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3852 }
3853
3854 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3855 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3856 }
3857
3858 switch (bond->params.mode) {
3859 case BOND_MODE_8023AD:
1b76b316 3860 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3861 break;
3862 case BOND_MODE_TLB:
3863 case BOND_MODE_ALB:
1b76b316 3864 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3865 break;
3866 default:
3867 break;
3868 }
3869
1da177e4 3870
58402054 3871 if (bond_is_lb(bond)) {
1da177e4
LT		 3872 /* Must be called only after all
3873 * slaves have been released
3874 */
3875 bond_alb_deinitialize(bond);
3876 }
3877
3878 return 0;
3879}
3880
3881static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3882{
454d7c9b 3883 struct bonding *bond = netdev_priv(bond_dev);
eeda3fd6 3884 struct net_device_stats *stats = &bond->stats;
2439f9eb 3885 struct net_device_stats local_stats;
1da177e4
LT		 3886 struct slave *slave;
3887 int i;
3888
2439f9eb 3889 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3890
3891 read_lock_bh(&bond->lock);
3892
3893 bond_for_each_slave(bond, slave, i) {
eeda3fd6
SH		 3894 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3895
2439f9eb
AG		 3896 local_stats.rx_packets += sstats->rx_packets;
3897 local_stats.rx_bytes += sstats->rx_bytes;
3898 local_stats.rx_errors += sstats->rx_errors;
3899 local_stats.rx_dropped += sstats->rx_dropped;
3900
3901 local_stats.tx_packets += sstats->tx_packets;
3902 local_stats.tx_bytes += sstats->tx_bytes;
3903 local_stats.tx_errors += sstats->tx_errors;
3904 local_stats.tx_dropped += sstats->tx_dropped;
3905
3906 local_stats.multicast += sstats->multicast;
3907 local_stats.collisions += sstats->collisions;
3908
3909 local_stats.rx_length_errors += sstats->rx_length_errors;
3910 local_stats.rx_over_errors += sstats->rx_over_errors;
3911 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3912 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3913 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3914 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3915
3916 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3917 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3918 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3919 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3920 local_stats.tx_window_errors += sstats->tx_window_errors;
3921 }
3922
3923 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3924
3925 read_unlock_bh(&bond->lock);
3926
3927 return stats;
3928}
3929
3930static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3931{
3932 struct net_device *slave_dev = NULL;
3933 struct ifbond k_binfo;
3934 struct ifbond __user *u_binfo = NULL;
3935 struct ifslave k_sinfo;
3936 struct ifslave __user *u_sinfo = NULL;
3937 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3938 int res = 0;
3939
3940 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3941 bond_dev->name, cmd);
3942
3943 switch (cmd) {
1da177e4
LT		 3944 case SIOCGMIIPHY:
3945 mii = if_mii(ifr);
3946 if (!mii) {
3947 return -EINVAL;
3948 }
3949 mii->phy_id = 0;
3950 /* Fall Through */
3951 case SIOCGMIIREG:
3952 /*
3953 * We do this again just in case we were called by SIOCGMIIREG
3954 * instead of SIOCGMIIPHY.
3955 */
3956 mii = if_mii(ifr);
3957 if (!mii) {
3958 return -EINVAL;
3959 }
3960
3961 if (mii->reg_num == 1) {
454d7c9b 3962 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3963 mii->val_out = 0;
6603a6f2 3964 read_lock(&bond->lock);
1da177e4 3965 read_lock(&bond->curr_slave_lock);
4e140079 3966 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3967 mii->val_out = BMSR_LSTATUS;
3968 }
3969 read_unlock(&bond->curr_slave_lock);
6603a6f2 3970 read_unlock(&bond->lock);
1da177e4
LT		 3971 }
3972
3973 return 0;
3974 case BOND_INFO_QUERY_OLD:
3975 case SIOCBONDINFOQUERY:
3976 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3977
3978 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3979 return -EFAULT;
3980 }
3981
3982 res = bond_info_query(bond_dev, &k_binfo);
3983 if (res == 0) {
3984 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3985 return -EFAULT;
3986 }
3987 }
3988
3989 return res;
3990 case BOND_SLAVE_INFO_QUERY_OLD:
3991 case SIOCBONDSLAVEINFOQUERY:
3992 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3993
3994 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3995 return -EFAULT;
3996 }
3997
3998 res = bond_slave_info_query(bond_dev, &k_sinfo);
3999 if (res == 0) {
4000 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4001 return -EFAULT;
4002 }
4003 }
4004
4005 return res;
4006 default:
4007 /* Go on */
4008 break;
4009 }
4010
4011 if (!capable(CAP_NET_ADMIN)) {
4012 return -EPERM;
4013 }
4014
b76cdba9 4015 down_write(&(bonding_rwsem));
881d966b 4016 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT		 4017
4018 dprintk("slave_dev=%p: \n", slave_dev);
4019
4020 if (!slave_dev) {
4021 res = -ENODEV;
4022 } else {
4023 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4024 switch (cmd) {
4025 case BOND_ENSLAVE_OLD:
4026 case SIOCBONDENSLAVE:
4027 res = bond_enslave(bond_dev, slave_dev);
4028 break;
4029 case BOND_RELEASE_OLD:
4030 case SIOCBONDRELEASE:
4031 res = bond_release(bond_dev, slave_dev);
4032 break;
4033 case BOND_SETHWADDR_OLD:
4034 case SIOCBONDSETHWADDR:
4035 res = bond_sethwaddr(bond_dev, slave_dev);
4036 break;
4037 case BOND_CHANGE_ACTIVE_OLD:
4038 case SIOCBONDCHANGEACTIVE:
4039 res = bond_ioctl_change_active(bond_dev, slave_dev);
4040 break;
4041 default:
4042 res = -EOPNOTSUPP;
4043 }
4044
4045 dev_put(slave_dev);
4046 }
4047
b76cdba9 4048 up_write(&(bonding_rwsem));
1da177e4
LT		 4049 return res;
4050}
4051
4052static void bond_set_multicast_list(struct net_device *bond_dev)
4053{
454d7c9b 4054 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4055 struct dev_mc_list *dmi;
4056
1da177e4
LT		 4057 /*
4058 * Do promisc before checking multicast_mode
4059 */
4060 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4061 /*
4062 * FIXME: Need to handle the error when one of the multi-slaves
4063 * encounters error.
4064 */
1da177e4
LT		 4065 bond_set_promiscuity(bond, 1);
4066 }
4067
4068 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4069 bond_set_promiscuity(bond, -1);
4070 }
4071
4072 /* set allmulti flag to slaves */
4073 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4074 /*
4075 * FIXME: Need to handle the error when one of the multi-slaves
4076 * encounters error.
4077 */
1da177e4
LT		 4078 bond_set_allmulti(bond, 1);
4079 }
4080
4081 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4082 bond_set_allmulti(bond, -1);
4083 }
4084
80ee5ad2
JV		 4085 read_lock(&bond->lock);
4086
1da177e4
LT		 4087 bond->flags = bond_dev->flags;
4088
4089 /* looking for addresses to add to slaves' mc list */
4090 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4091 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4092 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4093 }
4094 }
4095
4096 /* looking for addresses to delete from slaves' list */
4097 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4098 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4099 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4100 }
4101 }
4102
4103 /* save master's multicast list */
4104 bond_mc_list_destroy(bond);
4105 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4106
80ee5ad2 4107 read_unlock(&bond->lock);
1da177e4
LT		 4108}
4109
00829823
SH		 4110static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4111{
4112 struct bonding *bond = netdev_priv(dev);
4113 struct slave *slave = bond->first_slave;
4114
4115 if (slave) {
4116 const struct net_device_ops *slave_ops
4117 = slave->dev->netdev_ops;
4118 if (slave_ops->ndo_neigh_setup)
4119 return slave_ops->ndo_neigh_setup(dev, parms);
4120 }
4121 return 0;
4122}
4123
1da177e4
LT		 4124/*
4125 * Change the MTU of all of a master's slaves to match the master
4126 */
4127static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4128{
454d7c9b 4129 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4130 struct slave *slave, *stop_at;
4131 int res = 0;
4132 int i;
4133
4134 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4135 (bond_dev ? bond_dev->name : "None"), new_mtu);
4136
4137 /* Can't hold bond->lock with bh disabled here since
4138 * some base drivers panic. On the other hand we can't
4139 * hold bond->lock without bh disabled because we'll
4140 * deadlock. The only solution is to rely on the fact
4141 * that we're under rtnl_lock here, and the slaves
4142 * list won't change. This doesn't solve the problem
4143 * of setting the slave's MTU while it is
4144 * transmitting, but the assumption is that the base
4145 * driver can handle that.
4146 *
4147 * TODO: figure out a way to safely iterate the slaves
4148 * list, but without holding a lock around the actual
4149 * call to the base driver.
4150 */
4151
4152 bond_for_each_slave(bond, slave, i) {
4153 dprintk("s %p s->p %p c_m %p\n", slave,
4154 slave->prev, slave->dev->change_mtu);
e944ef79 4155
1da177e4
LT		 4156 res = dev_set_mtu(slave->dev, new_mtu);
4157
4158 if (res) {
4159 /* If we failed to set the slave's mtu to the new value
4160 * we must abort the operation even in ACTIVE_BACKUP
4161 * mode, because if we allow the backup slaves to have
4162 * different mtu values than the active slave we'll
4163 * need to change their mtu when doing a failover. That
4164 * means changing their mtu from timer context, which
4165 * is probably not a good idea.
4166 */
4167 dprintk("err %d %s\n", res, slave->dev->name);
4168 goto unwind;
4169 }
4170 }
4171
4172 bond_dev->mtu = new_mtu;
4173
4174 return 0;
4175
4176unwind:
4177 /* unwind from head to the slave that failed */
4178 stop_at = slave;
4179 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4180 int tmp_res;
4181
4182 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4183 if (tmp_res) {
4184 dprintk("unwind err %d dev %s\n", tmp_res,
4185 slave->dev->name);
4186 }
4187 }
4188
4189 return res;
4190}
4191
4192/*
4193 * Change HW address
4194 *
4195 * Note that many devices must be down to change the HW address, and
4196 * downing the master releases all slaves. We can make bonds full of
4197 * bonding devices to test this, however.
4198 */
4199static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4200{
454d7c9b 4201 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4202 struct sockaddr *sa = addr, tmp_sa;
4203 struct slave *slave, *stop_at;
4204 int res = 0;
4205 int i;
4206
eb7cc59a
SH		 4207 if (bond->params.mode == BOND_MODE_ALB)
4208 return bond_alb_set_mac_address(bond_dev, addr);
4209
4210
1da177e4
LT		 4211 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4212
dd957c57 4213 /*
3915c1e8
JV		 4214 * If fail_over_mac is set to active, do nothing and return
4215 * success. Returning an error causes ifenslave to fail.
dd957c57 4216 */
3915c1e8 4217 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4218 return 0;
2ab82852 4219
1da177e4
LT		 4220 if (!is_valid_ether_addr(sa->sa_data)) {
4221 return -EADDRNOTAVAIL;
4222 }
4223
4224 /* Can't hold bond->lock with bh disabled here since
4225 * some base drivers panic. On the other hand we can't
4226 * hold bond->lock without bh disabled because we'll
4227 * deadlock. The only solution is to rely on the fact
4228 * that we're under rtnl_lock here, and the slaves
4229 * list won't change. This doesn't solve the problem
4230 * of setting the slave's hw address while it is
4231 * transmitting, but the assumption is that the base
4232 * driver can handle that.
4233 *
4234 * TODO: figure out a way to safely iterate the slaves
4235 * list, but without holding a lock around the actual
4236 * call to the base driver.
4237 */
4238
4239 bond_for_each_slave(bond, slave, i) {
eb7cc59a 4240 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
1da177e4
LT		 4241 dprintk("slave %p %s\n", slave, slave->dev->name);
4242
eb7cc59a 4243 if (slave_ops->ndo_set_mac_address == NULL) {
1da177e4
LT		 4244 res = -EOPNOTSUPP;
4245 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4246 goto unwind;
4247 }
4248
4249 res = dev_set_mac_address(slave->dev, addr);
4250 if (res) {
4251 /* TODO: consider downing the slave
4252 * and retry ?
4253 * User should expect communications
4254 * breakage anyway until ARP finish
4255 * updating, so...
4256 */
4257 dprintk("err %d %s\n", res, slave->dev->name);
4258 goto unwind;
4259 }
4260 }
4261
4262 /* success */
4263 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4264 return 0;
4265
4266unwind:
4267 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4268 tmp_sa.sa_family = bond_dev->type;
4269
4270 /* unwind from head to the slave that failed */
4271 stop_at = slave;
4272 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4273 int tmp_res;
4274
4275 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4276 if (tmp_res) {
4277 dprintk("unwind err %d dev %s\n", tmp_res,
4278 slave->dev->name);
4279 }
4280 }
4281
4282 return res;
4283}
4284
4285static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4286{
454d7c9b 4287 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4288 struct slave *slave, *start_at;
cf5f9044 4289 int i, slave_no, res = 1;
1da177e4
LT		 4290
4291 read_lock(&bond->lock);
4292
4293 if (!BOND_IS_OK(bond)) {
4294 goto out;
4295 }
4296
cf5f9044
JV		 4297 /*
4298 * Concurrent TX may collide on rr_tx_counter; we accept that
4299 * as being rare enough not to justify using an atomic op here
4300 */
4301 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4302
cf5f9044
JV		 4303 bond_for_each_slave(bond, slave, i) {
4304 slave_no--;
4305 if (slave_no < 0) {
4306 break;
4307 }
1da177e4
LT		 4308 }
4309
cf5f9044 4310 start_at = slave;
1da177e4
LT		 4311 bond_for_each_slave_from(bond, slave, i, start_at) {
4312 if (IS_UP(slave->dev) &&
4313 (slave->link == BOND_LINK_UP) &&
4314 (slave->state == BOND_STATE_ACTIVE)) {
4315 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4316 break;
4317 }
4318 }
4319
1da177e4
LT		 4320out:
4321 if (res) {
4322 /* no suitable interface, frame not sent */
4323 dev_kfree_skb(skb);
4324 }
4325 read_unlock(&bond->lock);
4326 return 0;
4327}
4328
075897ce 4329
1da177e4
LT		 4330/*
4331 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4332 * the bond has a usable interface.
4333 */
4334static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4335{
454d7c9b 4336 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4337 int res = 1;
4338
1da177e4
LT		 4339 read_lock(&bond->lock);
4340 read_lock(&bond->curr_slave_lock);
4341
4342 if (!BOND_IS_OK(bond)) {
4343 goto out;
4344 }
4345
075897ce
JL		 4346 if (!bond->curr_active_slave)
4347 goto out;
4348
075897ce
JL		 4349 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4350
1da177e4
LT		 4351out:
4352 if (res) {
4353 /* no suitable interface, frame not sent */
4354 dev_kfree_skb(skb);
4355 }
4356 read_unlock(&bond->curr_slave_lock);
4357 read_unlock(&bond->lock);
4358 return 0;
4359}
4360
4361/*
169a3e66
JV		 4362 * In bond_xmit_xor() , we determine the output device by using a pre-
4363 * determined xmit_hash_policy(), If the selected device is not enabled,
4364 * find the next active slave.
1da177e4
LT		 4365 */
4366static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4367{
454d7c9b 4368 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4369 struct slave *slave, *start_at;
4370 int slave_no;
4371 int i;
4372 int res = 1;
4373
4374 read_lock(&bond->lock);
4375
4376 if (!BOND_IS_OK(bond)) {
4377 goto out;
4378 }
4379
169a3e66 4380 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4381
4382 bond_for_each_slave(bond, slave, i) {
4383 slave_no--;
4384 if (slave_no < 0) {
4385 break;
4386 }
4387 }
4388
4389 start_at = slave;
4390
4391 bond_for_each_slave_from(bond, slave, i, start_at) {
4392 if (IS_UP(slave->dev) &&
4393 (slave->link == BOND_LINK_UP) &&
4394 (slave->state == BOND_STATE_ACTIVE)) {
4395 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4396 break;
4397 }
4398 }
4399
4400out:
4401 if (res) {
4402 /* no suitable interface, frame not sent */
4403 dev_kfree_skb(skb);
4404 }
4405 read_unlock(&bond->lock);
4406 return 0;
4407}
4408
4409/*
4410 * in broadcast mode, we send everything to all usable interfaces.
4411 */
4412static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4413{
454d7c9b 4414 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4415 struct slave *slave, *start_at;
4416 struct net_device *tx_dev = NULL;
4417 int i;
4418 int res = 1;
4419
4420 read_lock(&bond->lock);
4421
4422 if (!BOND_IS_OK(bond)) {
4423 goto out;
4424 }
4425
4426 read_lock(&bond->curr_slave_lock);
4427 start_at = bond->curr_active_slave;
4428 read_unlock(&bond->curr_slave_lock);
4429
4430 if (!start_at) {
4431 goto out;
4432 }
4433
4434 bond_for_each_slave_from(bond, slave, i, start_at) {
4435 if (IS_UP(slave->dev) &&
4436 (slave->link == BOND_LINK_UP) &&
4437 (slave->state == BOND_STATE_ACTIVE)) {
4438 if (tx_dev) {
4439 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4440 if (!skb2) {
4441 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4442 ": %s: Error: bond_xmit_broadcast(): "
4443 "skb_clone() failed\n",
4444 bond_dev->name);
1da177e4
LT		 4445 continue;
4446 }
4447
4448 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4449 if (res) {
4450 dev_kfree_skb(skb2);
4451 continue;
4452 }
4453 }
4454 tx_dev = slave->dev;
4455 }
4456 }
4457
4458 if (tx_dev) {
4459 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4460 }
4461
4462out:
4463 if (res) {
4464 /* no suitable interface, frame not sent */
4465 dev_kfree_skb(skb);
4466 }
4467 /* frame sent to all suitable interfaces */
4468 read_unlock(&bond->lock);
4469 return 0;
4470}
4471
4472/*------------------------- Device initialization ---------------------------*/
4473
6f6652be
JV		 4474static void bond_set_xmit_hash_policy(struct bonding *bond)
4475{
4476 switch (bond->params.xmit_policy) {
4477 case BOND_XMIT_POLICY_LAYER23:
4478 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4479 break;
4480 case BOND_XMIT_POLICY_LAYER34:
4481 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4482 break;
4483 case BOND_XMIT_POLICY_LAYER2:
4484 default:
4485 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4486 break;
4487 }
4488}
4489
00829823
SH		 4490static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4491{
4492 const struct bonding *bond = netdev_priv(dev);
4493
4494 switch (bond->params.mode) {
4495 case BOND_MODE_ROUNDROBIN:
4496 return bond_xmit_roundrobin(skb, dev);
4497 case BOND_MODE_ACTIVEBACKUP:
4498 return bond_xmit_activebackup(skb, dev);
4499 case BOND_MODE_XOR:
4500 return bond_xmit_xor(skb, dev);
4501 case BOND_MODE_BROADCAST:
4502 return bond_xmit_broadcast(skb, dev);
4503 case BOND_MODE_8023AD:
4504 return bond_3ad_xmit_xor(skb, dev);
4505 case BOND_MODE_ALB:
4506 case BOND_MODE_TLB:
4507 return bond_alb_xmit(skb, dev);
4508 default:
4509 /* Should never happen, mode already checked */
4510 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4511 dev->name, bond->params.mode);
4512 WARN_ON_ONCE(1);
4513 dev_kfree_skb(skb);
4514 return NETDEV_TX_OK;
4515 }
4516}
4517
4518
1da177e4
LT		 4519/*
4520 * set bond mode specific net device operations
4521 */
a77b5325 4522void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4523{
169a3e66
JV		 4524 struct net_device *bond_dev = bond->dev;
4525
1da177e4
LT		 4526 switch (mode) {
4527 case BOND_MODE_ROUNDROBIN:
1da177e4
LT		 4528 break;
4529 case BOND_MODE_ACTIVEBACKUP:
1da177e4
LT		 4530 break;
4531 case BOND_MODE_XOR:
6f6652be 4532 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4533 break;
4534 case BOND_MODE_BROADCAST:
1da177e4
LT		 4535 break;
4536 case BOND_MODE_8023AD:
8f903c70 4537 bond_set_master_3ad_flags(bond);
6f6652be 4538 bond_set_xmit_hash_policy(bond);
1da177e4 4539 break;
1da177e4 4540 case BOND_MODE_ALB:
8f903c70
JV		 4541 bond_set_master_alb_flags(bond);
4542 /* FALLTHRU */
4543 case BOND_MODE_TLB:
1da177e4
LT		 4544 break;
4545 default:
4546 /* Should never happen, mode already checked */
4547 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4548 ": %s: Error: Unknown bonding mode %d\n",
4549 bond_dev->name,
1da177e4
LT		 4550 mode);
4551 break;
4552 }
4553}
4554
217df670
JV		 4555static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4556 struct ethtool_drvinfo *drvinfo)
4557{
4558 strncpy(drvinfo->driver, DRV_NAME, 32);
4559 strncpy(drvinfo->version, DRV_VERSION, 32);
4560 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4561}
4562
7282d491 4563static const struct ethtool_ops bond_ethtool_ops = {
217df670 4564 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac
SH		 4565 .get_link = ethtool_op_get_link,
4566 .get_tx_csum = ethtool_op_get_tx_csum,
4567 .get_sg = ethtool_op_get_sg,
4568 .get_tso = ethtool_op_get_tso,
4569 .get_ufo = ethtool_op_get_ufo,
4570 .get_flags = ethtool_op_get_flags,
8531c5ff
AK		 4571};
4572
eb7cc59a
SH		 4573static const struct net_device_ops bond_netdev_ops = {
4574 .ndo_open = bond_open,
4575 .ndo_stop = bond_close,
00829823 4576 .ndo_start_xmit = bond_start_xmit,
eb7cc59a
SH		 4577 .ndo_get_stats = bond_get_stats,
4578 .ndo_do_ioctl = bond_do_ioctl,
4579 .ndo_set_multicast_list = bond_set_multicast_list,
4580 .ndo_change_mtu = bond_change_mtu,
eb7cc59a 4581 .ndo_set_mac_address = bond_set_mac_address,
00829823 4582 .ndo_neigh_setup = bond_neigh_setup,
eb7cc59a
SH		 4583 .ndo_vlan_rx_register = bond_vlan_rx_register,
4584 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4585 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4586};
4587
1da177e4
LT		 4588/*
4589 * Does not allocate but creates a /proc entry.
4590 * Allowed to fail.
4591 */
3c535952 4592static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4 4593{
454d7c9b 4594 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4595
4596 dprintk("Begin bond_init for %s\n", bond_dev->name);
4597
4598 /* initialize rwlocks */
4599 rwlock_init(&bond->lock);
4600 rwlock_init(&bond->curr_slave_lock);
4601
4602 bond->params = *params; /* copy params struct */
4603
1b76b316
JV		 4604 bond->wq = create_singlethread_workqueue(bond_dev->name);
4605 if (!bond->wq)
4606 return -ENOMEM;
4607
1da177e4
LT		 4608 /* Initialize pointers */
4609 bond->first_slave = NULL;
4610 bond->curr_active_slave = NULL;
4611 bond->current_arp_slave = NULL;
4612 bond->primary_slave = NULL;
4613 bond->dev = bond_dev;
1053f62c 4614 bond->send_grat_arp = 0;
305d552a 4615 bond->send_unsol_na = 0;
d90a162a 4616 bond->setup_by_slave = 0;
1da177e4
LT		 4617 INIT_LIST_HEAD(&bond->vlan_list);
4618
4619 /* Initialize the device entry points */
eb7cc59a 4620 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 4621 bond_dev->ethtool_ops = &bond_ethtool_ops;
169a3e66 4622 bond_set_mode_ops(bond, bond->params.mode);
1da177e4 4623
a434e43f 4624 bond_dev->destructor = bond_destructor;
1da177e4
LT		 4625
4626 /* Initialize the device options */
4627 bond_dev->tx_queue_len = 0;
4628 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4629 bond_dev->priv_flags |= IFF_BONDING;
6cf3f41e
JV		 4630 if (bond->params.arp_interval)
4631 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
1da177e4
LT		 4632
4633 /* At first, we block adding VLANs. That's the only way to
4634 * prevent problems that occur when adding VLANs over an
4635 * empty bond. The block will be removed once non-challenged
4636 * slaves are enslaved.
4637 */
4638 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4639
932ff279 4640 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4641 * transmitting */
4642 bond_dev->features |= NETIF_F_LLTX;
4643
4644 /* By default, we declare the bond to be fully
4645 * VLAN hardware accelerated capable. Special
4646 * care is taken in the various xmit functions
4647 * when there are slaves that are not hw accel
4648 * capable
4649 */
1da177e4
LT		 4650 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4651 NETIF_F_HW_VLAN_RX |
4652 NETIF_F_HW_VLAN_FILTER);
4653
4654#ifdef CONFIG_PROC_FS
4655 bond_create_proc_entry(bond);
4656#endif
1da177e4
LT		 4657 list_add_tail(&bond->bond_list, &bond_dev_list);
4658
4659 return 0;
4660}
4661
fdaea7a9
JV		 4662static void bond_work_cancel_all(struct bonding *bond)
4663{
4664 write_lock_bh(&bond->lock);
4665 bond->kill_timers = 1;
4666 write_unlock_bh(&bond->lock);
4667
4668 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4669 cancel_delayed_work(&bond->mii_work);
4670
4671 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4672 cancel_delayed_work(&bond->arp_work);
4673
4674 if (bond->params.mode == BOND_MODE_ALB &&
4675 delayed_work_pending(&bond->alb_work))
4676 cancel_delayed_work(&bond->alb_work);
4677
4678 if (bond->params.mode == BOND_MODE_8023AD &&
4679 delayed_work_pending(&bond->ad_work))
4680 cancel_delayed_work(&bond->ad_work);
4681}
4682
a434e43f
JV		 4683/* De-initialize device specific data.
4684 * Caller must hold rtnl_lock.
4685 */
4686static void bond_deinit(struct net_device *bond_dev)
4687{
454d7c9b 4688 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 4689
4690 list_del(&bond->bond_list);
4691
4692 bond_work_cancel_all(bond);
4693
4694#ifdef CONFIG_PROC_FS
4695 bond_remove_proc_entry(bond);
4696#endif
4697}
4698
1da177e4
LT		 4699/* Unregister and free all bond devices.
4700 * Caller must hold rtnl_lock.
4701 */
4702static void bond_free_all(void)
4703{
4704 struct bonding *bond, *nxt;
4705
4706 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4707 struct net_device *bond_dev = bond->dev;
4708
fdaea7a9 4709 bond_work_cancel_all(bond);
70298705 4710 /* Release the bonded slaves */
4711 bond_release_all(bond_dev);
92b41daa 4712 bond_destroy(bond);
1da177e4
LT		 4713 }
4714
4715#ifdef CONFIG_PROC_FS
4716 bond_destroy_proc_dir();
4717#endif
4718}
4719
4720/*------------------------- Module initialization ---------------------------*/
4721
4722/*
4723 * Convert string input module parms. Accept either the
ece95f7f
JV		 4724 * number of the mode or its string name. A bit complicated because
4725 * some mode names are substrings of other names, and calls from sysfs
4726 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4727 */
ece95f7f 4728int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4729{
ece95f7f 4730 int mode = -1, i, rv;
a42e534f 4731 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4732
a42e534f
JV		 4733 for (p = (char *)buf; *p; p++)
4734 if (!(isdigit(*p) || isspace(*p)))
4735 break;
4736
4737 if (*p)
ece95f7f 4738 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4739 else
4740 rv = sscanf(buf, "%d", &mode);
4741
4742 if (!rv)
4743 return -1;
1da177e4
LT		 4744
4745 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4746 if (mode == tbl[i].mode)
4747 return tbl[i].mode;
4748 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4749 return tbl[i].mode;
1da177e4
LT		 4750 }
4751
4752 return -1;
4753}
4754
4755static int bond_check_params(struct bond_params *params)
4756{
3915c1e8 4757 int arp_validate_value, fail_over_mac_value;
f5b2b966 4758
1da177e4
LT		 4759 /*
4760 * Convert string parameters.
4761 */
4762 if (mode) {
4763 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4764 if (bond_mode == -1) {
4765 printk(KERN_ERR DRV_NAME
4766 ": Error: Invalid bonding mode \"%s\"\n",
4767 mode == NULL ? "NULL" : mode);
4768 return -EINVAL;
4769 }
4770 }
4771
169a3e66
JV		 4772 if (xmit_hash_policy) {
4773 if ((bond_mode != BOND_MODE_XOR) &&
4774 (bond_mode != BOND_MODE_8023AD)) {
4775 printk(KERN_INFO DRV_NAME
4776 ": xor_mode param is irrelevant in mode %s\n",
4777 bond_mode_name(bond_mode));
4778 } else {
4779 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4780 xmit_hashtype_tbl);
4781 if (xmit_hashtype == -1) {
4782 printk(KERN_ERR DRV_NAME
4783 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4784 xmit_hash_policy == NULL ? "NULL" :
4785 xmit_hash_policy);
4786 return -EINVAL;
4787 }
4788 }
4789 }
4790
1da177e4
LT		 4791 if (lacp_rate) {
4792 if (bond_mode != BOND_MODE_8023AD) {
4793 printk(KERN_INFO DRV_NAME
4794 ": lacp_rate param is irrelevant in mode %s\n",
4795 bond_mode_name(bond_mode));
4796 } else {
4797 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4798 if (lacp_fast == -1) {
4799 printk(KERN_ERR DRV_NAME
4800 ": Error: Invalid lacp rate \"%s\"\n",
4801 lacp_rate == NULL ? "NULL" : lacp_rate);
4802 return -EINVAL;
4803 }
4804 }
4805 }
4806
fd989c83
JV		 4807 if (ad_select) {
4808 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4809 if (params->ad_select == -1) {
4810 printk(KERN_ERR DRV_NAME
4811 ": Error: Invalid ad_select \"%s\"\n",
4812 ad_select == NULL ? "NULL" : ad_select);
4813 return -EINVAL;
4814 }
4815
4816 if (bond_mode != BOND_MODE_8023AD) {
4817 printk(KERN_WARNING DRV_NAME
4818 ": ad_select param only affects 802.3ad mode\n");
4819 }
4820 } else {
4821 params->ad_select = BOND_AD_STABLE;
4822 }
4823
b8a9787e 4824 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4825 printk(KERN_WARNING DRV_NAME
4826 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4827 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4828 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4829 max_bonds = BOND_DEFAULT_MAX_BONDS;
4830 }
4831
4832 if (miimon < 0) {
4833 printk(KERN_WARNING DRV_NAME
4834 ": Warning: miimon module parameter (%d), "
4835 "not in range 0-%d, so it was reset to %d\n",
4836 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4837 miimon = BOND_LINK_MON_INTERV;
4838 }
4839
4840 if (updelay < 0) {
4841 printk(KERN_WARNING DRV_NAME
4842 ": Warning: updelay module parameter (%d), "
4843 "not in range 0-%d, so it was reset to 0\n",
4844 updelay, INT_MAX);
4845 updelay = 0;
4846 }
4847
4848 if (downdelay < 0) {
4849 printk(KERN_WARNING DRV_NAME
4850 ": Warning: downdelay module parameter (%d), "
4851 "not in range 0-%d, so it was reset to 0\n",
4852 downdelay, INT_MAX);
4853 downdelay = 0;
4854 }
4855
4856 if ((use_carrier != 0) && (use_carrier != 1)) {
4857 printk(KERN_WARNING DRV_NAME
4858 ": Warning: use_carrier module parameter (%d), "
4859 "not of valid value (0/1), so it was set to 1\n",
4860 use_carrier);
4861 use_carrier = 1;
4862 }
4863
7893b249
MS		 4864 if (num_grat_arp < 0 || num_grat_arp > 255) {
4865 printk(KERN_WARNING DRV_NAME
4866 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4867 "was reset to 1 \n", num_grat_arp);
4868 num_grat_arp = 1;
4869 }
4870
305d552a
BH		 4871 if (num_unsol_na < 0 || num_unsol_na > 255) {
4872 printk(KERN_WARNING DRV_NAME
4873 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4874 "was reset to 1 \n", num_unsol_na);
4875 num_unsol_na = 1;
4876 }
4877
1da177e4
LT		 4878 /* reset values for 802.3ad */
4879 if (bond_mode == BOND_MODE_8023AD) {
4880 if (!miimon) {
4881 printk(KERN_WARNING DRV_NAME
4882 ": Warning: miimon must be specified, "
4883 "otherwise bonding will not detect link "
4884 "failure, speed and duplex which are "
4885 "essential for 802.3ad operation\n");
4886 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4887 miimon = 100;
4888 }
4889 }
4890
4891 /* reset values for TLB/ALB */
4892 if ((bond_mode == BOND_MODE_TLB) ||
4893 (bond_mode == BOND_MODE_ALB)) {
4894 if (!miimon) {
4895 printk(KERN_WARNING DRV_NAME
4896 ": Warning: miimon must be specified, "
4897 "otherwise bonding will not detect link "
4898 "failure and link speed which are essential "
4899 "for TLB/ALB load balancing\n");
4900 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4901 miimon = 100;
4902 }
4903 }
4904
4905 if (bond_mode == BOND_MODE_ALB) {
4906 printk(KERN_NOTICE DRV_NAME
4907 ": In ALB mode you might experience client "
4908 "disconnections upon reconnection of a link if the "
4909 "bonding module updelay parameter (%d msec) is "
4910 "incompatible with the forwarding delay time of the "
4911 "switch\n",
4912 updelay);
4913 }
4914
4915 if (!miimon) {
4916 if (updelay || downdelay) {
4917 /* just warn the user the up/down delay will have
4918 * no effect since miimon is zero...
4919 */
4920 printk(KERN_WARNING DRV_NAME
4921 ": Warning: miimon module parameter not set "
4922 "and updelay (%d) or downdelay (%d) module "
4923 "parameter is set; updelay and downdelay have "
4924 "no effect unless miimon is set\n",
4925 updelay, downdelay);
4926 }
4927 } else {
4928 /* don't allow arp monitoring */
4929 if (arp_interval) {
4930 printk(KERN_WARNING DRV_NAME
4931 ": Warning: miimon (%d) and arp_interval (%d) "
4932 "can't be used simultaneously, disabling ARP "
4933 "monitoring\n",
4934 miimon, arp_interval);
4935 arp_interval = 0;
4936 }
4937
4938 if ((updelay % miimon) != 0) {
4939 printk(KERN_WARNING DRV_NAME
4940 ": Warning: updelay (%d) is not a multiple "
4941 "of miimon (%d), updelay rounded to %d ms\n",
4942 updelay, miimon, (updelay / miimon) * miimon);
4943 }
4944
4945 updelay /= miimon;
4946
4947 if ((downdelay % miimon) != 0) {
4948 printk(KERN_WARNING DRV_NAME
4949 ": Warning: downdelay (%d) is not a multiple "
4950 "of miimon (%d), downdelay rounded to %d ms\n",
4951 downdelay, miimon,
4952 (downdelay / miimon) * miimon);
4953 }
4954
4955 downdelay /= miimon;
4956 }
4957
4958 if (arp_interval < 0) {
4959 printk(KERN_WARNING DRV_NAME
4960 ": Warning: arp_interval module parameter (%d) "
4961 ", not in range 0-%d, so it was reset to %d\n",
4962 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4963 arp_interval = BOND_LINK_ARP_INTERV;
4964 }
4965
4966 for (arp_ip_count = 0;
4967 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4968 arp_ip_count++) {
4969 /* not complete check, but should be good enough to
4970 catch mistakes */
4971 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4972 printk(KERN_WARNING DRV_NAME
4973 ": Warning: bad arp_ip_target module parameter "
4974 "(%s), ARP monitoring will not be performed\n",
4975 arp_ip_target[arp_ip_count]);
4976 arp_interval = 0;
4977 } else {
d3bb52b0 4978 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4979 arp_target[arp_ip_count] = ip;
4980 }
4981 }
4982
4983 if (arp_interval && !arp_ip_count) {
4984 /* don't allow arping if no arp_ip_target given... */
4985 printk(KERN_WARNING DRV_NAME
4986 ": Warning: arp_interval module parameter (%d) "
4987 "specified without providing an arp_ip_target "
4988 "parameter, arp_interval was reset to 0\n",
4989 arp_interval);
4990 arp_interval = 0;
4991 }
4992
f5b2b966
JV		 4993 if (arp_validate) {
4994 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4995 printk(KERN_ERR DRV_NAME
4996 ": arp_validate only supported in active-backup mode\n");
4997 return -EINVAL;
4998 }
4999 if (!arp_interval) {
5000 printk(KERN_ERR DRV_NAME
5001 ": arp_validate requires arp_interval\n");
5002 return -EINVAL;
5003 }
5004
5005 arp_validate_value = bond_parse_parm(arp_validate,
5006 arp_validate_tbl);
5007 if (arp_validate_value == -1) {
5008 printk(KERN_ERR DRV_NAME
5009 ": Error: invalid arp_validate \"%s\"\n",
5010 arp_validate == NULL ? "NULL" : arp_validate);
5011 return -EINVAL;
5012 }
5013 } else
5014 arp_validate_value = 0;
5015
1da177e4
LT		 5016 if (miimon) {
5017 printk(KERN_INFO DRV_NAME
5018 ": MII link monitoring set to %d ms\n",
5019 miimon);
5020 } else if (arp_interval) {
5021 int i;
5022
5023 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 5024 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5025 arp_interval,
5026 arp_validate_tbl[arp_validate_value].modename,
5027 arp_ip_count);
1da177e4
LT		 5028
5029 for (i = 0; i < arp_ip_count; i++)
5030 printk (" %s", arp_ip_target[i]);
5031
5032 printk("\n");
5033
b8a9787e 5034 } else if (max_bonds) {
1da177e4
LT		 5035 /* miimon and arp_interval not set, we need one so things
5036 * work as expected, see bonding.txt for details
5037 */
5038 printk(KERN_WARNING DRV_NAME
5039 ": Warning: either miimon or arp_interval and "
5040 "arp_ip_target module parameters must be specified, "
5041 "otherwise bonding will not detect link failures! see "
5042 "bonding.txt for details.\n");
5043 }
5044
5045 if (primary && !USES_PRIMARY(bond_mode)) {
5046 /* currently, using a primary only makes sense
5047 * in active backup, TLB or ALB modes
5048 */
5049 printk(KERN_WARNING DRV_NAME
5050 ": Warning: %s primary device specified but has no "
5051 "effect in %s mode\n",
5052 primary, bond_mode_name(bond_mode));
5053 primary = NULL;
5054 }
5055
3915c1e8
JV		 5056 if (fail_over_mac) {
5057 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5058 fail_over_mac_tbl);
5059 if (fail_over_mac_value == -1) {
5060 printk(KERN_ERR DRV_NAME
5061 ": Error: invalid fail_over_mac \"%s\"\n",
5062 arp_validate == NULL ? "NULL" : arp_validate);
5063 return -EINVAL;
5064 }
5065
5066 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5067 printk(KERN_WARNING DRV_NAME
5068 ": Warning: fail_over_mac only affects "
5069 "active-backup mode.\n");
5070 } else {
5071 fail_over_mac_value = BOND_FOM_NONE;
5072 }
dd957c57 5073
1da177e4
LT		 5074 /* fill params struct with the proper values */
5075 params->mode = bond_mode;
169a3e66 5076 params->xmit_policy = xmit_hashtype;
1da177e4 5077 params->miimon = miimon;
7893b249 5078 params->num_grat_arp = num_grat_arp;
305d552a 5079 params->num_unsol_na = num_unsol_na;
1da177e4 5080 params->arp_interval = arp_interval;
f5b2b966 5081 params->arp_validate = arp_validate_value;
1da177e4
LT		 5082 params->updelay = updelay;
5083 params->downdelay = downdelay;
5084 params->use_carrier = use_carrier;
5085 params->lacp_fast = lacp_fast;
5086 params->primary[0] = 0;
3915c1e8 5087 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5088
5089 if (primary) {
5090 strncpy(params->primary, primary, IFNAMSIZ);
5091 params->primary[IFNAMSIZ - 1] = 0;
5092 }
5093
5094 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5095
5096 return 0;
5097}
5098
0daa2303 5099static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 5100static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 5101
e8a0464c
DM		 5102static void bond_set_lockdep_class_one(struct net_device *dev,
5103 struct netdev_queue *txq,
5104 void *_unused)
c773e847
DM		 5105{
5106 lockdep_set_class(&txq->_xmit_lock,
5107 &bonding_netdev_xmit_lock_key);
5108}
5109
5110static void bond_set_lockdep_class(struct net_device *dev)
5111{
cf508b12
DM		 5112 lockdep_set_class(&dev->addr_list_lock,
5113 &bonding_netdev_addr_lock_key);
e8a0464c 5114 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 5115}
5116
dfe60397 5117/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5118 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5119 * Caller must NOT hold rtnl_lock; we need to release it here before we
5120 * set up our sysfs entries.
5121 */
0dd646fe 5122int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5123{
5124 struct net_device *bond_dev;
0883beca 5125 struct bonding *bond;
dfe60397
MW		 5126 int res;
5127
5128 rtnl_lock();
027ea041
JV		 5129 down_write(&bonding_rwsem);
5130
5131 /* Check to see if the bond already exists. */
4fe4763c 5132 if (name) {
0883beca 5133 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5134 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5135 printk(KERN_ERR DRV_NAME
027ea041 5136 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5137 name);
5138 res = -EPERM;
5139 goto out_rtnl;
5140 }
5141 }
027ea041 5142
e4b91c48
JV		 5143 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5144 ether_setup);
dfe60397
MW		 5145 if (!bond_dev) {
5146 printk(KERN_ERR DRV_NAME
5147 ": %s: eek! can't alloc netdev!\n",
5148 name);
5149 res = -ENOMEM;
5150 goto out_rtnl;
5151 }
5152
e4b91c48
JV		 5153 if (!name) {
5154 res = dev_alloc_name(bond_dev, "bond%d");
5155 if (res < 0)
5156 goto out_netdev;
5157 }
5158
dfe60397
MW		 5159 /* bond_init() must be called after dev_alloc_name() (for the
5160 * /proc files), but before register_netdevice(), because we
5161 * need to set function pointers.
5162 */
5163
5164 res = bond_init(bond_dev, params);
5165 if (res < 0) {
5166 goto out_netdev;
5167 }
5168
dfe60397
MW		 5169 res = register_netdevice(bond_dev);
5170 if (res < 0) {
5171 goto out_bond;
5172 }
0daa2303 5173
c773e847 5174 bond_set_lockdep_class(bond_dev);
0daa2303 5175
ff59c456
JV		 5176 netif_carrier_off(bond_dev);
5177
027ea041 5178 up_write(&bonding_rwsem);
dfe60397 5179 rtnl_unlock(); /* allows sysfs registration of net device */
454d7c9b 5180 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
09c89279
JV		 5181 if (res < 0) {
5182 rtnl_lock();
027ea041 5183 down_write(&bonding_rwsem);
822973ba
PE		 5184 bond_deinit(bond_dev);
5185 unregister_netdevice(bond_dev);
5186 goto out_rtnl;
09c89279
JV		 5187 }
5188
5189 return 0;
5190
dfe60397
MW		 5191out_bond:
5192 bond_deinit(bond_dev);
5193out_netdev:
5194 free_netdev(bond_dev);
5195out_rtnl:
027ea041 5196 up_write(&bonding_rwsem);
dfe60397 5197 rtnl_unlock();
dfe60397
MW		 5198 return res;
5199}
5200
1da177e4
LT		 5201static int __init bonding_init(void)
5202{
1da177e4
LT		 5203 int i;
5204 int res;
0883beca 5205 struct bonding *bond;
1da177e4
LT		 5206
5207 printk(KERN_INFO "%s", version);
5208
dfe60397 5209 res = bond_check_params(&bonding_defaults);
1da177e4 5210 if (res) {
dfe60397 5211 goto out;
1da177e4
LT		 5212 }
5213
1da177e4
LT		 5214#ifdef CONFIG_PROC_FS
5215 bond_create_proc_dir();
5216#endif
027ea041
JV		 5217
5218 init_rwsem(&bonding_rwsem);
5219
1da177e4 5220 for (i = 0; i < max_bonds; i++) {
0dd646fe 5221 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5222 if (res)
5223 goto err;
1da177e4
LT		 5224 }
5225
b76cdba9
MW		 5226 res = bond_create_sysfs();
5227 if (res)
5228 goto err;
5229
1da177e4 5230 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5231 register_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5232 bond_register_ipv6_notifier();
1da177e4 5233
dfe60397
MW		 5234 goto out;
5235err:
0883beca 5236 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5237 bond_work_cancel_all(bond);
5238 destroy_workqueue(bond->wq);
5239 }
5240
ae68c398
PE		 5241 bond_destroy_sysfs();
5242
40abc270 5243 rtnl_lock();
1da177e4 5244 bond_free_all();
1da177e4 5245 rtnl_unlock();
dfe60397 5246out:
1da177e4 5247 return res;
dfe60397 5248
1da177e4
LT		 5249}
5250
5251static void __exit bonding_exit(void)
5252{
5253 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5254 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5255 bond_unregister_ipv6_notifier();
1da177e4 5256
ae68c398
PE		 5257 bond_destroy_sysfs();
5258
1da177e4
LT		 5259 rtnl_lock();
5260 bond_free_all();
5261 rtnl_unlock();
5262}
5263
5264module_init(bonding_init);
5265module_exit(bonding_exit);
5266MODULE_LICENSE("GPL");
5267MODULE_VERSION(DRV_VERSION);
5268MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5269MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5270MODULE_SUPPORTED_DEVICE("most ethernet devices");
5271
5272/*
5273 * Local variables:
5274 * c-indent-level: 8
5275 * c-basic-offset: 8
5276 * tab-width: 8
5277 * End:
5278 */
5279
Clone of davem's net-next-2.6 tree