projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bonding: add and use bond_is_lb()
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / net / bonding / bond_main.c
CommitLineData
1da177e4
LT		 1/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
1da177e4
LT		 32 */
33
34//#define BONDING_DEBUG 1
35
1da177e4
LT		 36#include <linux/kernel.h>
37#include <linux/module.h>
1da177e4
LT		 38#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
169a3e66 44#include <net/ip.h>
1da177e4 45#include <linux/ip.h>
169a3e66
JV		 46#include <linux/tcp.h>
47#include <linux/udp.h>
1da177e4
LT		 48#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
56#include <asm/system.h>
57#include <asm/io.h>
58#include <asm/dma.h>
59#include <asm/uaccess.h>
60#include <linux/errno.h>
61#include <linux/netdevice.h>
62#include <linux/inetdevice.h>
a816c7c7 63#include <linux/igmp.h>
1da177e4
LT		 64#include <linux/etherdevice.h>
65#include <linux/skbuff.h>
66#include <net/sock.h>
67#include <linux/rtnetlink.h>
68#include <linux/proc_fs.h>
69#include <linux/seq_file.h>
70#include <linux/smp.h>
71#include <linux/if_ether.h>
72#include <net/arp.h>
73#include <linux/mii.h>
74#include <linux/ethtool.h>
75#include <linux/if_vlan.h>
76#include <linux/if_bonding.h>
b63bb739 77#include <linux/jiffies.h>
c3ade5ca 78#include <net/route.h>
457c4cbc 79#include <net/net_namespace.h>
1da177e4
LT		 80#include "bonding.h"
81#include "bond_3ad.h"
82#include "bond_alb.h"
83
84/*---------------------------- Module parameters ----------------------------*/
85
86/* monitor all links that often (in milliseconds). <=0 disables monitoring */
87#define BOND_LINK_MON_INTERV 0
88#define BOND_LINK_ARP_INTERV 0
89
90static int max_bonds = BOND_DEFAULT_MAX_BONDS;
7893b249 91static int num_grat_arp = 1;
305d552a 92static int num_unsol_na = 1;
1da177e4
LT		 93static int miimon = BOND_LINK_MON_INTERV;
94static int updelay = 0;
95static int downdelay = 0;
96static int use_carrier = 1;
97static char *mode = NULL;
98static char *primary = NULL;
99static char *lacp_rate = NULL;
fd989c83 100static char *ad_select = NULL;
169a3e66 101static char *xmit_hash_policy = NULL;
1da177e4
LT		 102static int arp_interval = BOND_LINK_ARP_INTERV;
103static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
f5b2b966 104static char *arp_validate = NULL;
3915c1e8 105static char *fail_over_mac = NULL;
12479f9a 106struct bond_params bonding_defaults;
1da177e4
LT		 107
108module_param(max_bonds, int, 0);
109MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
7893b249
MS		 110module_param(num_grat_arp, int, 0644);
111MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
305d552a
BH		 112module_param(num_unsol_na, int, 0644);
113MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
1da177e4
LT		 114module_param(miimon, int, 0);
115MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
116module_param(updelay, int, 0);
117MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
118module_param(downdelay, int, 0);
2ac47660
MW		 119MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
120 "in milliseconds");
1da177e4 121module_param(use_carrier, int, 0);
2ac47660
MW		 122MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
123 "0 for off, 1 for on (default)");
1da177e4 124module_param(mode, charp, 0);
2ac47660
MW		 125MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
126 "1 for active-backup, 2 for balance-xor, "
127 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
128 "6 for balance-alb");
1da177e4
LT		 129module_param(primary, charp, 0);
130MODULE_PARM_DESC(primary, "Primary network device to use");
131module_param(lacp_rate, charp, 0);
2ac47660
MW		 132MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
133 "(slow/fast)");
fd989c83
JV		 134module_param(ad_select, charp, 0);
135MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
169a3e66 136module_param(xmit_hash_policy, charp, 0);
2ac47660
MW		 137MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
138 ", 1 for layer 3+4");
1da177e4
LT		 139module_param(arp_interval, int, 0);
140MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
141module_param_array(arp_ip_target, charp, NULL, 0);
142MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
f5b2b966
JV		 143module_param(arp_validate, charp, 0);
144MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
3915c1e8
JV		 145module_param(fail_over_mac, charp, 0);
146MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
1da177e4
LT		 147
148/*----------------------------- Global variables ----------------------------*/
149
f71e1309 150static const char * const version =
1da177e4
LT		 151 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
152
12479f9a 153LIST_HEAD(bond_dev_list);
1da177e4
LT		 154
155#ifdef CONFIG_PROC_FS
156static struct proc_dir_entry *bond_proc_dir = NULL;
157#endif
158
b76cdba9 159extern struct rw_semaphore bonding_rwsem;
d3bb52b0 160static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
1da177e4 161static int arp_ip_count = 0;
1da177e4 162static int bond_mode = BOND_MODE_ROUNDROBIN;
169a3e66 163static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
1da177e4 164static int lacp_fast = 0;
217df670 165
1da177e4 166
12479f9a 167struct bond_parm_tbl bond_lacp_tbl[] = {
1da177e4
LT		 168{ "slow", AD_LACP_SLOW},
169{ "fast", AD_LACP_FAST},
170{ NULL, -1},
171};
172
12479f9a 173struct bond_parm_tbl bond_mode_tbl[] = {
1da177e4
LT		 174{ "balance-rr", BOND_MODE_ROUNDROBIN},
175{ "active-backup", BOND_MODE_ACTIVEBACKUP},
176{ "balance-xor", BOND_MODE_XOR},
177{ "broadcast", BOND_MODE_BROADCAST},
178{ "802.3ad", BOND_MODE_8023AD},
179{ "balance-tlb", BOND_MODE_TLB},
180{ "balance-alb", BOND_MODE_ALB},
181{ NULL, -1},
182};
183
12479f9a 184struct bond_parm_tbl xmit_hashtype_tbl[] = {
169a3e66
JV		 185{ "layer2", BOND_XMIT_POLICY_LAYER2},
186{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
6f6652be 187{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
169a3e66
JV		 188{ NULL, -1},
189};
190
f5b2b966
JV		 191struct bond_parm_tbl arp_validate_tbl[] = {
192{ "none", BOND_ARP_VALIDATE_NONE},
193{ "active", BOND_ARP_VALIDATE_ACTIVE},
194{ "backup", BOND_ARP_VALIDATE_BACKUP},
195{ "all", BOND_ARP_VALIDATE_ALL},
196{ NULL, -1},
197};
198
3915c1e8
JV		 199struct bond_parm_tbl fail_over_mac_tbl[] = {
200{ "none", BOND_FOM_NONE},
201{ "active", BOND_FOM_ACTIVE},
202{ "follow", BOND_FOM_FOLLOW},
203{ NULL, -1},
204};
205
fd989c83
JV		 206struct bond_parm_tbl ad_select_tbl[] = {
207{ "stable", BOND_AD_STABLE},
208{ "bandwidth", BOND_AD_BANDWIDTH},
209{ "count", BOND_AD_COUNT},
210{ NULL, -1},
211};
212
1da177e4
LT		 213/*-------------------------- Forward declarations ---------------------------*/
214
c3ade5ca 215static void bond_send_gratuitous_arp(struct bonding *bond);
c50b85d0 216static void bond_deinit(struct net_device *bond_dev);
1da177e4
LT		 217
218/*---------------------------- General routines -----------------------------*/
219
4ad072c9 220static const char *bond_mode_name(int mode)
1da177e4
LT		 221{
222 switch (mode) {
223 case BOND_MODE_ROUNDROBIN :
224 return "load balancing (round-robin)";
225 case BOND_MODE_ACTIVEBACKUP :
226 return "fault-tolerance (active-backup)";
227 case BOND_MODE_XOR :
228 return "load balancing (xor)";
229 case BOND_MODE_BROADCAST :
230 return "fault-tolerance (broadcast)";
231 case BOND_MODE_8023AD:
232 return "IEEE 802.3ad Dynamic link aggregation";
233 case BOND_MODE_TLB:
234 return "transmit load balancing";
235 case BOND_MODE_ALB:
236 return "adaptive load balancing";
237 default:
238 return "unknown";
239 }
240}
241
242/*---------------------------------- VLAN -----------------------------------*/
243
244/**
245 * bond_add_vlan - add a new vlan id on bond
246 * @bond: bond that got the notification
247 * @vlan_id: the vlan id to add
248 *
249 * Returns -ENOMEM if allocation failed.
250 */
251static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
252{
253 struct vlan_entry *vlan;
254
255 dprintk("bond: %s, vlan id %d\n",
256 (bond ? bond->dev->name: "None"), vlan_id);
257
305d552a 258 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
1da177e4
LT		 259 if (!vlan) {
260 return -ENOMEM;
261 }
262
263 INIT_LIST_HEAD(&vlan->vlan_list);
264 vlan->vlan_id = vlan_id;
265
266 write_lock_bh(&bond->lock);
267
268 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
269
270 write_unlock_bh(&bond->lock);
271
272 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
273
274 return 0;
275}
276
277/**
278 * bond_del_vlan - delete a vlan id from bond
279 * @bond: bond that got the notification
280 * @vlan_id: the vlan id to delete
281 *
282 * returns -ENODEV if @vlan_id was not found in @bond.
283 */
284static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
285{
0883beca 286 struct vlan_entry *vlan;
1da177e4
LT		 287 int res = -ENODEV;
288
289 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
290
291 write_lock_bh(&bond->lock);
292
0883beca 293 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
1da177e4
LT		 294 if (vlan->vlan_id == vlan_id) {
295 list_del(&vlan->vlan_list);
296
58402054 297 if (bond_is_lb(bond))
1da177e4 298 bond_alb_clear_vlan(bond, vlan_id);
1da177e4
LT		 299
300 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
301 bond->dev->name);
302
303 kfree(vlan);
304
305 if (list_empty(&bond->vlan_list) &&
306 (bond->slave_cnt == 0)) {
307 /* Last VLAN removed and no slaves, so
308 * restore block on adding VLANs. This will
309 * be removed once new slaves that are not
310 * VLAN challenged will be added.
311 */
312 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
313 }
314
315 res = 0;
316 goto out;
317 }
318 }
319
320 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
321 bond->dev->name);
322
323out:
324 write_unlock_bh(&bond->lock);
325 return res;
326}
327
328/**
329 * bond_has_challenged_slaves
330 * @bond: the bond we're working on
331 *
332 * Searches the slave list. Returns 1 if a vlan challenged slave
333 * was found, 0 otherwise.
334 *
335 * Assumes bond->lock is held.
336 */
337static int bond_has_challenged_slaves(struct bonding *bond)
338{
339 struct slave *slave;
340 int i;
341
342 bond_for_each_slave(bond, slave, i) {
343 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
344 dprintk("found VLAN challenged slave - %s\n",
345 slave->dev->name);
346 return 1;
347 }
348 }
349
350 dprintk("no VLAN challenged slaves found\n");
351 return 0;
352}
353
354/**
355 * bond_next_vlan - safely skip to the next item in the vlans list.
356 * @bond: the bond we're working on
357 * @curr: item we're advancing from
358 *
359 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
360 * or @curr->next otherwise (even if it is @curr itself again).
361 *
362 * Caller must hold bond->lock
363 */
364struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
365{
366 struct vlan_entry *next, *last;
367
368 if (list_empty(&bond->vlan_list)) {
369 return NULL;
370 }
371
372 if (!curr) {
373 next = list_entry(bond->vlan_list.next,
374 struct vlan_entry, vlan_list);
375 } else {
376 last = list_entry(bond->vlan_list.prev,
377 struct vlan_entry, vlan_list);
378 if (last == curr) {
379 next = list_entry(bond->vlan_list.next,
380 struct vlan_entry, vlan_list);
381 } else {
382 next = list_entry(curr->vlan_list.next,
383 struct vlan_entry, vlan_list);
384 }
385 }
386
387 return next;
388}
389
390/**
391 * bond_dev_queue_xmit - Prepare skb for xmit.
392 *
393 * @bond: bond device that got this skb for tx.
394 * @skb: hw accel VLAN tagged skb to transmit
395 * @slave_dev: slave that is supposed to xmit this skbuff
396 *
397 * When the bond gets an skb to transmit that is
398 * already hardware accelerated VLAN tagged, and it
399 * needs to relay this skb to a slave that is not
400 * hw accel capable, the skb needs to be "unaccelerated",
401 * i.e. strip the hwaccel tag and re-insert it as part
402 * of the payload.
403 */
404int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
405{
966bc6f4 406 unsigned short uninitialized_var(vlan_id);
1da177e4
LT		 407
408 if (!list_empty(&bond->vlan_list) &&
409 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
410 vlan_get_tag(skb, &vlan_id) == 0) {
411 skb->dev = slave_dev;
412 skb = vlan_put_tag(skb, vlan_id);
413 if (!skb) {
414 /* vlan_put_tag() frees the skb in case of error,
415 * so return success here so the calling functions
416 * won't attempt to free is again.
417 */
418 return 0;
419 }
420 } else {
421 skb->dev = slave_dev;
422 }
423
424 skb->priority = 1;
425 dev_queue_xmit(skb);
426
427 return 0;
428}
429
430/*
431 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
432 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
433 * lock because:
434 * a. This operation is performed in IOCTL context,
435 * b. The operation is protected by the RTNL semaphore in the 8021q code,
436 * c. Holding a lock with BH disabled while directly calling a base driver
437 * entry point is generally a BAD idea.
438 *
439 * The design of synchronization/protection for this operation in the 8021q
440 * module is good for one or more VLAN devices over a single physical device
441 * and cannot be extended for a teaming solution like bonding, so there is a
442 * potential race condition here where a net device from the vlan group might
443 * be referenced (either by a base driver or the 8021q code) while it is being
444 * removed from the system. However, it turns out we're not making matters
445 * worse, and if it works for regular VLAN usage it will work here too.
446*/
447
448/**
449 * bond_vlan_rx_register - Propagates registration to slaves
450 * @bond_dev: bonding net device that got called
451 * @grp: vlan group being registered
452 */
453static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
454{
454d7c9b 455 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 456 struct slave *slave;
457 int i;
458
459 bond->vlgrp = grp;
460
461 bond_for_each_slave(bond, slave, i) {
462 struct net_device *slave_dev = slave->dev;
eb7cc59a 463 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 464
465 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 466 slave_ops->ndo_vlan_rx_register) {
467 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
1da177e4
LT		 468 }
469 }
470}
471
472/**
473 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
474 * @bond_dev: bonding net device that got called
475 * @vid: vlan id being added
476 */
477static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
478{
454d7c9b 479 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 480 struct slave *slave;
481 int i, res;
482
483 bond_for_each_slave(bond, slave, i) {
484 struct net_device *slave_dev = slave->dev;
eb7cc59a 485 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 486
487 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a
SH		 488 slave_ops->ndo_vlan_rx_add_vid) {
489 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
1da177e4
LT		 490 }
491 }
492
493 res = bond_add_vlan(bond, vid);
494 if (res) {
495 printk(KERN_ERR DRV_NAME
4e0952c7 496 ": %s: Error: Failed to add vlan id %d\n",
1da177e4
LT		 497 bond_dev->name, vid);
498 }
499}
500
501/**
502 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
503 * @bond_dev: bonding net device that got called
504 * @vid: vlan id being removed
505 */
506static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
507{
454d7c9b 508 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 509 struct slave *slave;
510 struct net_device *vlan_dev;
511 int i, res;
512
513 bond_for_each_slave(bond, slave, i) {
514 struct net_device *slave_dev = slave->dev;
eb7cc59a 515 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 516
517 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a 518 slave_ops->ndo_vlan_rx_kill_vid) {
1da177e4
LT		 519 /* Save and then restore vlan_dev in the grp array,
520 * since the slave's driver might clear it.
521 */
5c15bdec 522 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
eb7cc59a 523 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
5c15bdec 524 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
1da177e4
LT		 525 }
526 }
527
528 res = bond_del_vlan(bond, vid);
529 if (res) {
530 printk(KERN_ERR DRV_NAME
4e0952c7 531 ": %s: Error: Failed to remove vlan id %d\n",
1da177e4
LT		 532 bond_dev->name, vid);
533 }
534}
535
536static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
537{
538 struct vlan_entry *vlan;
eb7cc59a 539 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 540
541 write_lock_bh(&bond->lock);
542
eb7cc59a 543 if (list_empty(&bond->vlan_list))
1da177e4 544 goto out;
1da177e4
LT		 545
546 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 547 slave_ops->ndo_vlan_rx_register)
548 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
1da177e4
LT		 549
550 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 551 !(slave_ops->ndo_vlan_rx_add_vid))
1da177e4 552 goto out;
1da177e4 553
eb7cc59a
SH		 554 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
555 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
1da177e4
LT		 556
557out:
558 write_unlock_bh(&bond->lock);
559}
560
561static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
562{
eb7cc59a 563 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 564 struct vlan_entry *vlan;
565 struct net_device *vlan_dev;
566
567 write_lock_bh(&bond->lock);
568
eb7cc59a 569 if (list_empty(&bond->vlan_list))
1da177e4 570 goto out;
1da177e4
LT		 571
572 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 573 !(slave_ops->ndo_vlan_rx_kill_vid))
1da177e4 574 goto unreg;
1da177e4
LT		 575
576 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
577 /* Save and then restore vlan_dev in the grp array,
578 * since the slave's driver might clear it.
579 */
5c15bdec 580 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
eb7cc59a 581 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
5c15bdec 582 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
1da177e4
LT		 583 }
584
585unreg:
586 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 587 slave_ops->ndo_vlan_rx_register)
588 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
1da177e4
LT		 589
590out:
591 write_unlock_bh(&bond->lock);
592}
593
594/*------------------------------- Link status -------------------------------*/
595
ff59c456
JV		 596/*
597 * Set the carrier state for the master according to the state of its
598 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
599 * do special 802.3ad magic.
600 *
601 * Returns zero if carrier state does not change, nonzero if it does.
602 */
603static int bond_set_carrier(struct bonding *bond)
604{
605 struct slave *slave;
606 int i;
607
608 if (bond->slave_cnt == 0)
609 goto down;
610
611 if (bond->params.mode == BOND_MODE_8023AD)
612 return bond_3ad_set_carrier(bond);
613
614 bond_for_each_slave(bond, slave, i) {
615 if (slave->link == BOND_LINK_UP) {
616 if (!netif_carrier_ok(bond->dev)) {
617 netif_carrier_on(bond->dev);
618 return 1;
619 }
620 return 0;
621 }
622 }
623
624down:
625 if (netif_carrier_ok(bond->dev)) {
626 netif_carrier_off(bond->dev);
627 return 1;
628 }
629 return 0;
630}
631
1da177e4
LT		 632/*
633 * Get link speed and duplex from the slave's base driver
634 * using ethtool. If for some reason the call fails or the
635 * values are invalid, fake speed and duplex to 100/Full
636 * and return error.
637 */
638static int bond_update_speed_duplex(struct slave *slave)
639{
640 struct net_device *slave_dev = slave->dev;
1da177e4 641 struct ethtool_cmd etool;
61a44b9c 642 int res;
1da177e4
LT		 643
644 /* Fake speed and duplex */
645 slave->speed = SPEED_100;
646 slave->duplex = DUPLEX_FULL;
647
61a44b9c
MW		 648 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
649 return -1;
1da177e4 650
61a44b9c
MW		 651 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
652 if (res < 0)
1da177e4 653 return -1;
1da177e4 654
1da177e4
LT		 655 switch (etool.speed) {
656 case SPEED_10:
657 case SPEED_100:
658 case SPEED_1000:
94dbffd5 659 case SPEED_10000:
1da177e4
LT		 660 break;
661 default:
662 return -1;
663 }
664
665 switch (etool.duplex) {
666 case DUPLEX_FULL:
667 case DUPLEX_HALF:
668 break;
669 default:
670 return -1;
671 }
672
673 slave->speed = etool.speed;
674 slave->duplex = etool.duplex;
675
676 return 0;
677}
678
679/*
680 * if <dev> supports MII link status reporting, check its link status.
681 *
682 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
683 * depening upon the setting of the use_carrier parameter.
684 *
685 * Return either BMSR_LSTATUS, meaning that the link is up (or we
686 * can't tell and just pretend it is), or 0, meaning that the link is
687 * down.
688 *
689 * If reporting is non-zero, instead of faking link up, return -1 if
690 * both ETHTOOL and MII ioctls fail (meaning the device does not
691 * support them). If use_carrier is set, return whatever it says.
692 * It'd be nice if there was a good way to tell if a driver supports
693 * netif_carrier, but there really isn't.
694 */
695static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
696{
eb7cc59a 697 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 698 static int (* ioctl)(struct net_device *, struct ifreq *, int);
699 struct ifreq ifr;
700 struct mii_ioctl_data *mii;
1da177e4 701
eb7cc59a 702 if (bond->params.use_carrier)
1da177e4 703 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1da177e4 704
eb7cc59a 705 ioctl = slave_ops->ndo_do_ioctl;
1da177e4
LT		 706 if (ioctl) {
707 /* TODO: set pointer to correct ioctl on a per team member */
708 /* bases to make this more efficient. that is, once */
709 /* we determine the correct ioctl, we will always */
710 /* call it and not the others for that team */
711 /* member. */
712
713 /*
714 * We cannot assume that SIOCGMIIPHY will also read a
715 * register; not all network drivers (e.g., e100)
716 * support that.
717 */
718
719 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
720 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
721 mii = if_mii(&ifr);
722 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
723 mii->reg_num = MII_BMSR;
724 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
725 return (mii->val_out & BMSR_LSTATUS);
726 }
727 }
728 }
729
61a44b9c
MW		 730 /*
731 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
732 * attempt to get link status from it if the above MII ioctls fail.
733 */
1da177e4
LT		 734 if (slave_dev->ethtool_ops) {
735 if (slave_dev->ethtool_ops->get_link) {
736 u32 link;
737
738 link = slave_dev->ethtool_ops->get_link(slave_dev);
739
740 return link ? BMSR_LSTATUS : 0;
741 }
742 }
743
1da177e4
LT		 744 /*
745 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 746 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 747 * cannot report link status). If not reporting, pretend
748 * we're ok.
749 */
750 return (reporting ? -1 : BMSR_LSTATUS);
751}
752
753/*----------------------------- Multicast list ------------------------------*/
754
755/*
756 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
757 */
758static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
759{
760 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
761 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
762}
763
764/*
765 * returns dmi entry if found, NULL otherwise
766 */
767static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
768{
769 struct dev_mc_list *idmi;
770
771 for (idmi = mc_list; idmi; idmi = idmi->next) {
772 if (bond_is_dmi_same(dmi, idmi)) {
773 return idmi;
774 }
775 }
776
777 return NULL;
778}
779
780/*
781 * Push the promiscuity flag down to appropriate slaves
782 */
7e1a1ac1 783static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 784{
7e1a1ac1 785 int err = 0;
1da177e4
LT		 786 if (USES_PRIMARY(bond->params.mode)) {
787 /* write lock already acquired */
788 if (bond->curr_active_slave) {
7e1a1ac1
WC		 789 err = dev_set_promiscuity(bond->curr_active_slave->dev,
790 inc);
1da177e4
LT		 791 }
792 } else {
793 struct slave *slave;
794 int i;
795 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 796 err = dev_set_promiscuity(slave->dev, inc);
797 if (err)
798 return err;
1da177e4
LT		 799 }
800 }
7e1a1ac1 801 return err;
1da177e4
LT		 802}
803
804/*
805 * Push the allmulti flag down to all slaves
806 */
7e1a1ac1 807static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 808{
7e1a1ac1 809 int err = 0;
1da177e4
LT		 810 if (USES_PRIMARY(bond->params.mode)) {
811 /* write lock already acquired */
812 if (bond->curr_active_slave) {
7e1a1ac1
WC		 813 err = dev_set_allmulti(bond->curr_active_slave->dev,
814 inc);
1da177e4
LT		 815 }
816 } else {
817 struct slave *slave;
818 int i;
819 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 820 err = dev_set_allmulti(slave->dev, inc);
821 if (err)
822 return err;
1da177e4
LT		 823 }
824 }
7e1a1ac1 825 return err;
1da177e4
LT		 826}
827
828/*
829 * Add a Multicast address to slaves
830 * according to mode
831 */
832static void bond_mc_add(struct bonding *bond, void *addr, int alen)
833{
834 if (USES_PRIMARY(bond->params.mode)) {
835 /* write lock already acquired */
836 if (bond->curr_active_slave) {
837 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
838 }
839 } else {
840 struct slave *slave;
841 int i;
842 bond_for_each_slave(bond, slave, i) {
843 dev_mc_add(slave->dev, addr, alen, 0);
844 }
845 }
846}
847
848/*
849 * Remove a multicast address from slave
850 * according to mode
851 */
852static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
853{
854 if (USES_PRIMARY(bond->params.mode)) {
855 /* write lock already acquired */
856 if (bond->curr_active_slave) {
857 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
858 }
859 } else {
860 struct slave *slave;
861 int i;
862 bond_for_each_slave(bond, slave, i) {
863 dev_mc_delete(slave->dev, addr, alen, 0);
864 }
865 }
866}
867
a816c7c7
JV		 868
869/*
870 * Retrieve the list of registered multicast addresses for the bonding
871 * device and retransmit an IGMP JOIN request to the current active
872 * slave.
873 */
874static void bond_resend_igmp_join_requests(struct bonding *bond)
875{
876 struct in_device *in_dev;
877 struct ip_mc_list *im;
878
879 rcu_read_lock();
880 in_dev = __in_dev_get_rcu(bond->dev);
881 if (in_dev) {
882 for (im = in_dev->mc_list; im; im = im->next) {
883 ip_mc_rejoin_group(im);
884 }
885 }
886
887 rcu_read_unlock();
888}
889
1da177e4
LT		 890/*
891 * Totally destroys the mc_list in bond
892 */
893static void bond_mc_list_destroy(struct bonding *bond)
894{
895 struct dev_mc_list *dmi;
896
897 dmi = bond->mc_list;
898 while (dmi) {
899 bond->mc_list = dmi->next;
900 kfree(dmi);
901 dmi = bond->mc_list;
902 }
a816c7c7 903 bond->mc_list = NULL;
1da177e4
LT		 904}
905
906/*
907 * Copy all the Multicast addresses from src to the bonding device dst
908 */
de54f390 909static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 910 gfp_t gfp_flag)
1da177e4
LT		 911{
912 struct dev_mc_list *dmi, *new_dmi;
913
914 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 915 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT		 916
917 if (!new_dmi) {
918 /* FIXME: Potential memory leak !!! */
919 return -ENOMEM;
920 }
921
922 new_dmi->next = bond->mc_list;
923 bond->mc_list = new_dmi;
924 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
925 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
926 new_dmi->dmi_users = dmi->dmi_users;
927 new_dmi->dmi_gusers = dmi->dmi_gusers;
928 }
929
930 return 0;
931}
932
933/*
934 * flush all members of flush->mc_list from device dev->mc_list
935 */
936static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
937{
454d7c9b 938 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 939 struct dev_mc_list *dmi;
940
941 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
942 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
943 }
944
945 if (bond->params.mode == BOND_MODE_8023AD) {
946 /* del lacpdu mc addr from mc list */
947 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
948
949 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
950 }
951}
952
953/*--------------------------- Active slave change ---------------------------*/
954
955/*
956 * Update the mc list and multicast-related flags for the new and
957 * old active slaves (if any) according to the multicast mode, and
958 * promiscuous flags unconditionally.
959 */
960static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
961{
962 struct dev_mc_list *dmi;
963
964 if (!USES_PRIMARY(bond->params.mode)) {
965 /* nothing to do - mc list is already up-to-date on
966 * all slaves
967 */
968 return;
969 }
970
971 if (old_active) {
972 if (bond->dev->flags & IFF_PROMISC) {
973 dev_set_promiscuity(old_active->dev, -1);
974 }
975
976 if (bond->dev->flags & IFF_ALLMULTI) {
977 dev_set_allmulti(old_active->dev, -1);
978 }
979
980 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
981 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
982 }
983 }
984
985 if (new_active) {
7e1a1ac1 986 /* FIXME: Signal errors upstream. */
1da177e4
LT		 987 if (bond->dev->flags & IFF_PROMISC) {
988 dev_set_promiscuity(new_active->dev, 1);
989 }
990
991 if (bond->dev->flags & IFF_ALLMULTI) {
992 dev_set_allmulti(new_active->dev, 1);
993 }
994
995 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
996 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
997 }
a816c7c7 998 bond_resend_igmp_join_requests(bond);
1da177e4
LT		 999 }
1000}
1001
3915c1e8
JV		 1002/*
1003 * bond_do_fail_over_mac
1004 *
1005 * Perform special MAC address swapping for fail_over_mac settings
1006 *
1007 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1008 */
1009static void bond_do_fail_over_mac(struct bonding *bond,
1010 struct slave *new_active,
1011 struct slave *old_active)
1012{
1013 u8 tmp_mac[ETH_ALEN];
1014 struct sockaddr saddr;
1015 int rv;
1016
1017 switch (bond->params.fail_over_mac) {
1018 case BOND_FOM_ACTIVE:
1019 if (new_active)
1020 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1021 new_active->dev->addr_len);
1022 break;
1023 case BOND_FOM_FOLLOW:
1024 /*
1025 * if new_active && old_active, swap them
1026 * if just old_active, do nothing (going to no active slave)
1027 * if just new_active, set new_active to bond's MAC
1028 */
1029 if (!new_active)
1030 return;
1031
1032 write_unlock_bh(&bond->curr_slave_lock);
1033 read_unlock(&bond->lock);
1034
1035 if (old_active) {
1036 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1037 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1038 ETH_ALEN);
1039 saddr.sa_family = new_active->dev->type;
1040 } else {
1041 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1042 saddr.sa_family = bond->dev->type;
1043 }
1044
1045 rv = dev_set_mac_address(new_active->dev, &saddr);
1046 if (rv) {
1047 printk(KERN_ERR DRV_NAME
1048 ": %s: Error %d setting MAC of slave %s\n",
1049 bond->dev->name, -rv, new_active->dev->name);
1050 goto out;
1051 }
1052
1053 if (!old_active)
1054 goto out;
1055
1056 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1057 saddr.sa_family = old_active->dev->type;
1058
1059 rv = dev_set_mac_address(old_active->dev, &saddr);
1060 if (rv)
1061 printk(KERN_ERR DRV_NAME
1062 ": %s: Error %d setting MAC of slave %s\n",
1063 bond->dev->name, -rv, new_active->dev->name);
1064out:
1065 read_lock(&bond->lock);
1066 write_lock_bh(&bond->curr_slave_lock);
1067 break;
1068 default:
1069 printk(KERN_ERR DRV_NAME
1070 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1071 bond->dev->name, bond->params.fail_over_mac);
1072 break;
1073 }
1074
1075}
1076
1077
1da177e4
LT		 1078/**
1079 * find_best_interface - select the best available slave to be the active one
1080 * @bond: our bonding struct
1081 *
1082 * Warning: Caller must hold curr_slave_lock for writing.
1083 */
1084static struct slave *bond_find_best_slave(struct bonding *bond)
1085{
1086 struct slave *new_active, *old_active;
1087 struct slave *bestslave = NULL;
1088 int mintime = bond->params.updelay;
1089 int i;
1090
1091 new_active = old_active = bond->curr_active_slave;
1092
1093 if (!new_active) { /* there were no active slaves left */
1094 if (bond->slave_cnt > 0) { /* found one slave */
1095 new_active = bond->first_slave;
1096 } else {
1097 return NULL; /* still no slave, return NULL */
1098 }
1099 }
1100
1101 /* first try the primary link; if arping, a link must tx/rx traffic
1102 * before it can be considered the curr_active_slave - also, we would skip
1103 * slaves between the curr_active_slave and primary_slave that may be up
1104 * and able to arp
1105 */
1106 if ((bond->primary_slave) &&
1107 (!bond->params.arp_interval) &&
1108 (IS_UP(bond->primary_slave->dev))) {
1109 new_active = bond->primary_slave;
1110 }
1111
1112 /* remember where to stop iterating over the slaves */
1113 old_active = new_active;
1114
1115 bond_for_each_slave_from(bond, new_active, i, old_active) {
1116 if (IS_UP(new_active->dev)) {
1117 if (new_active->link == BOND_LINK_UP) {
1118 return new_active;
1119 } else if (new_active->link == BOND_LINK_BACK) {
1120 /* link up, but waiting for stabilization */
1121 if (new_active->delay < mintime) {
1122 mintime = new_active->delay;
1123 bestslave = new_active;
1124 }
1125 }
1126 }
1127 }
1128
1129 return bestslave;
1130}
1131
1132/**
1133 * change_active_interface - change the active slave into the specified one
1134 * @bond: our bonding struct
1135 * @new: the new slave to make the active one
1136 *
1137 * Set the new slave to the bond's settings and unset them on the old
1138 * curr_active_slave.
1139 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1140 *
1141 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1142 * because it is apparently the best available slave we have, even though its
1143 * updelay hasn't timed out yet.
1144 *
3915c1e8
JV		 1145 * If new_active is not NULL, caller must hold bond->lock for read and
1146 * curr_slave_lock for write_bh.
1da177e4 1147 */
a77b5325 1148void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1149{
1150 struct slave *old_active = bond->curr_active_slave;
1151
1152 if (old_active == new_active) {
1153 return;
1154 }
1155
1156 if (new_active) {
b2220cad
JV		 1157 new_active->jiffies = jiffies;
1158
1da177e4
LT		 1159 if (new_active->link == BOND_LINK_BACK) {
1160 if (USES_PRIMARY(bond->params.mode)) {
1161 printk(KERN_INFO DRV_NAME
1162 ": %s: making interface %s the new "
1163 "active one %d ms earlier.\n",
1164 bond->dev->name, new_active->dev->name,
1165 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1166 }
1167
1168 new_active->delay = 0;
1169 new_active->link = BOND_LINK_UP;
1da177e4
LT		 1170
1171 if (bond->params.mode == BOND_MODE_8023AD) {
1172 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1173 }
1174
58402054 1175 if (bond_is_lb(bond))
1da177e4 1176 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1da177e4
LT		 1177 } else {
1178 if (USES_PRIMARY(bond->params.mode)) {
1179 printk(KERN_INFO DRV_NAME
1180 ": %s: making interface %s the new "
1181 "active one.\n",
1182 bond->dev->name, new_active->dev->name);
1183 }
1184 }
1185 }
1186
1da177e4
LT		 1187 if (USES_PRIMARY(bond->params.mode)) {
1188 bond_mc_swap(bond, new_active, old_active);
1189 }
1190
58402054 1191 if (bond_is_lb(bond)) {
1da177e4 1192 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1193 if (old_active)
1194 bond_set_slave_inactive_flags(old_active);
1195 if (new_active)
1196 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1197 } else {
1198 bond->curr_active_slave = new_active;
1199 }
c3ade5ca
JV		 1200
1201 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1202 if (old_active) {
1203 bond_set_slave_inactive_flags(old_active);
1204 }
1205
1206 if (new_active) {
1207 bond_set_slave_active_flags(new_active);
2ab82852 1208
709f8a45
OG		 1209 if (bond->params.fail_over_mac)
1210 bond_do_fail_over_mac(bond, new_active,
1211 old_active);
3915c1e8 1212
709f8a45 1213 bond->send_grat_arp = bond->params.num_grat_arp;
b59f9f74 1214 bond_send_gratuitous_arp(bond);
01f3109d 1215
305d552a
BH		 1216 bond->send_unsol_na = bond->params.num_unsol_na;
1217 bond_send_unsolicited_na(bond);
1218
01f3109d
OG		 1219 write_unlock_bh(&bond->curr_slave_lock);
1220 read_unlock(&bond->lock);
1221
1222 netdev_bonding_change(bond->dev);
1223
1224 read_lock(&bond->lock);
1225 write_lock_bh(&bond->curr_slave_lock);
7893b249 1226 }
c3ade5ca 1227 }
1da177e4
LT		 1228}
1229
1230/**
1231 * bond_select_active_slave - select a new active slave, if needed
1232 * @bond: our bonding struct
1233 *
1234 * This functions shoud be called when one of the following occurs:
1235 * - The old curr_active_slave has been released or lost its link.
1236 * - The primary_slave has got its link back.
1237 * - A slave has got its link back and there's no old curr_active_slave.
1238 *
3915c1e8 1239 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1240 */
a77b5325 1241void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1242{
1243 struct slave *best_slave;
ff59c456 1244 int rv;
1da177e4
LT		 1245
1246 best_slave = bond_find_best_slave(bond);
1247 if (best_slave != bond->curr_active_slave) {
1248 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1249 rv = bond_set_carrier(bond);
1250 if (!rv)
1251 return;
1252
1253 if (netif_carrier_ok(bond->dev)) {
1254 printk(KERN_INFO DRV_NAME
1255 ": %s: first active interface up!\n",
1256 bond->dev->name);
1257 } else {
1258 printk(KERN_INFO DRV_NAME ": %s: "
1259 "now running without any active interface !\n",
1260 bond->dev->name);
1261 }
1da177e4
LT		 1262 }
1263}
1264
1265/*--------------------------- slave list handling ---------------------------*/
1266
1267/*
1268 * This function attaches the slave to the end of list.
1269 *
1270 * bond->lock held for writing by caller.
1271 */
1272static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1273{
1274 if (bond->first_slave == NULL) { /* attaching the first slave */
1275 new_slave->next = new_slave;
1276 new_slave->prev = new_slave;
1277 bond->first_slave = new_slave;
1278 } else {
1279 new_slave->next = bond->first_slave;
1280 new_slave->prev = bond->first_slave->prev;
1281 new_slave->next->prev = new_slave;
1282 new_slave->prev->next = new_slave;
1283 }
1284
1285 bond->slave_cnt++;
1286}
1287
1288/*
1289 * This function detaches the slave from the list.
1290 * WARNING: no check is made to verify if the slave effectively
1291 * belongs to <bond>.
1292 * Nothing is freed on return, structures are just unchained.
1293 * If any slave pointer in bond was pointing to <slave>,
1294 * it should be changed by the calling function.
1295 *
1296 * bond->lock held for writing by caller.
1297 */
1298static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1299{
1300 if (slave->next) {
1301 slave->next->prev = slave->prev;
1302 }
1303
1304 if (slave->prev) {
1305 slave->prev->next = slave->next;
1306 }
1307
1308 if (bond->first_slave == slave) { /* slave is the first slave */
1309 if (bond->slave_cnt > 1) { /* there are more slave */
1310 bond->first_slave = slave->next;
1311 } else {
1312 bond->first_slave = NULL; /* slave was the last one */
1313 }
1314 }
1315
1316 slave->next = NULL;
1317 slave->prev = NULL;
1318 bond->slave_cnt--;
1319}
1320
1321/*---------------------------------- IOCTL ----------------------------------*/
1322
4ad072c9
AB		 1323static int bond_sethwaddr(struct net_device *bond_dev,
1324 struct net_device *slave_dev)
1da177e4
LT		 1325{
1326 dprintk("bond_dev=%p\n", bond_dev);
1327 dprintk("slave_dev=%p\n", slave_dev);
1328 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1329 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1330 return 0;
1331}
1332
7f353bf2
HX		 1333#define BOND_VLAN_FEATURES \
1334 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1335 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK		 1336
1337/*
8e3babcd 1338 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX		 1339 * feature bits are managed elsewhere, so preserve those feature bits
1340 * on the master device.
8531c5ff
AK		 1341 */
1342static int bond_compute_features(struct bonding *bond)
1343{
8531c5ff
AK		 1344 struct slave *slave;
1345 struct net_device *bond_dev = bond->dev;
7f353bf2 1346 unsigned long features = bond_dev->features;
3158bf7d
MS		 1347 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1348 bond_dev->hard_header_len);
8e3babcd 1349 int i;
8531c5ff 1350
7f353bf2 1351 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
b63365a2
HX		 1352 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1353
1354 if (!bond->first_slave)
1355 goto done;
1356
1357 features &= ~NETIF_F_ONE_FOR_ALL;
7f353bf2 1358
54ef3137 1359 bond_for_each_slave(bond, slave, i) {
b63365a2
HX		 1360 features = netdev_increment_features(features,
1361 slave->dev->features,
1362 NETIF_F_ONE_FOR_ALL);
54ef3137
JV		 1363 if (slave->dev->hard_header_len > max_hard_header_len)
1364 max_hard_header_len = slave->dev->hard_header_len;
1365 }
8531c5ff 1366
b63365a2 1367done:
7f353bf2 1368 features |= (bond_dev->features & BOND_VLAN_FEATURES);
b63365a2 1369 bond_dev->features = netdev_fix_features(features, NULL);
54ef3137 1370 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1371
1372 return 0;
1373}
1374
872254dd
MS		 1375static void bond_setup_by_slave(struct net_device *bond_dev,
1376 struct net_device *slave_dev)
1377{
454d7c9b 1378 struct bonding *bond = netdev_priv(bond_dev);
d90a162a 1379
00829823 1380 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1381
1382 bond_dev->type = slave_dev->type;
1383 bond_dev->hard_header_len = slave_dev->hard_header_len;
1384 bond_dev->addr_len = slave_dev->addr_len;
1385
1386 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1387 slave_dev->addr_len);
d90a162a 1388 bond->setup_by_slave = 1;
872254dd
MS		 1389}
1390
1da177e4 1391/* enslave device <slave> to bond device <master> */
a77b5325 1392int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1393{
454d7c9b 1394 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1395 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 1396 struct slave *new_slave = NULL;
1397 struct dev_mc_list *dmi;
1398 struct sockaddr addr;
1399 int link_reporting;
1400 int old_features = bond_dev->features;
1401 int res = 0;
1402
552709d5 1403 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
eb7cc59a 1404 slave_ops->ndo_do_ioctl == NULL) {
1da177e4 1405 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1406 ": %s: Warning: no link monitoring support for %s\n",
1407 bond_dev->name, slave_dev->name);
1da177e4
LT		 1408 }
1409
1410 /* bond must be initialized by bond_open() before enslaving */
1411 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1412 printk(KERN_WARNING DRV_NAME
1413 " %s: master_dev is not up in bond_enslave\n",
1414 bond_dev->name);
1da177e4
LT		 1415 }
1416
1417 /* already enslaved */
1418 if (slave_dev->flags & IFF_SLAVE) {
1419 dprintk("Error, Device was already enslaved\n");
1420 return -EBUSY;
1421 }
1422
1423 /* vlan challenged mutual exclusion */
1424 /* no need to lock since we're protected by rtnl_lock */
1425 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1426 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1427 if (!list_empty(&bond->vlan_list)) {
1428 printk(KERN_ERR DRV_NAME
4e0952c7 1429 ": %s: Error: cannot enslave VLAN "
1da177e4 1430 "challenged slave %s on VLAN enabled "
4e0952c7 1431 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1432 bond_dev->name);
1433 return -EPERM;
1434 } else {
1435 printk(KERN_WARNING DRV_NAME
4e0952c7 1436 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1437 "slave %s. Adding VLANs will be blocked as "
1438 "long as %s is part of bond %s\n",
4e0952c7 1439 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1440 bond_dev->name);
1441 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1442 }
1443 } else {
1444 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1445 if (bond->slave_cnt == 0) {
1446 /* First slave, and it is not VLAN challenged,
1447 * so remove the block of adding VLANs over the bond.
1448 */
1449 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1450 }
1451 }
1452
217df670
JV		 1453 /*
1454 * Old ifenslave binaries are no longer supported. These can
1455 * be identified with moderate accurary by the state of the slave:
1456 * the current ifenslave will set the interface down prior to
1457 * enslaving it; the old ifenslave will not.
1458 */
1459 if ((slave_dev->flags & IFF_UP)) {
1460 printk(KERN_ERR DRV_NAME ": %s is up. "
1461 "This may be due to an out of date ifenslave.\n",
1462 slave_dev->name);
1463 res = -EPERM;
1464 goto err_undo_flags;
1465 }
1da177e4 1466
872254dd
MS		 1467 /* set bonding device ether type by slave - bonding netdevices are
1468 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1469 * there is a need to override some of the type dependent attribs/funcs.
1470 *
1471 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1472 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1473 */
1474 if (bond->slave_cnt == 0) {
1475 if (slave_dev->type != ARPHRD_ETHER)
1476 bond_setup_by_slave(bond_dev, slave_dev);
1477 } else if (bond_dev->type != slave_dev->type) {
1478 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1479 "from other slaves (%d), can not enslave it.\n",
1480 slave_dev->name,
1481 slave_dev->type, bond_dev->type);
1482 res = -EINVAL;
1483 goto err_undo_flags;
1484 }
1485
eb7cc59a 1486 if (slave_ops->ndo_set_mac_address == NULL) {
2ab82852
MS		 1487 if (bond->slave_cnt == 0) {
1488 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1489 ": %s: Warning: The first slave device "
1490 "specified does not support setting the MAC "
3915c1e8 1491 "address. Setting fail_over_mac to active.",
dd957c57 1492 bond_dev->name);
3915c1e8
JV		 1493 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1494 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1495 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1496 ": %s: Error: The slave device specified "
1497 "does not support setting the MAC address, "
3915c1e8 1498 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1499 , bond_dev->name);
1500 res = -EOPNOTSUPP;
1501 goto err_undo_flags;
1502 }
1da177e4
LT		 1503 }
1504
243cb4e5 1505 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1506 if (!new_slave) {
1507 res = -ENOMEM;
1508 goto err_undo_flags;
1509 }
1510
1da177e4
LT		 1511 /* save slave's original flags before calling
1512 * netdev_set_master and dev_open
1513 */
1514 new_slave->original_flags = slave_dev->flags;
1515
217df670
JV		 1516 /*
1517 * Save slave's original ("permanent") mac address for modes
1518 * that need it, and for restoring it upon release, and then
1519 * set it to the master's address
1520 */
1521 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1522
dd957c57 1523 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1524 /*
1525 * Set slave to master's mac address. The application already
1526 * set the master's mac address to that of the first slave
1527 */
1528 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1529 addr.sa_family = slave_dev->type;
1530 res = dev_set_mac_address(slave_dev, &addr);
1531 if (res) {
1532 dprintk("Error %d calling set_mac_address\n", res);
1533 goto err_free;
1534 }
217df670 1535 }
1da177e4 1536
c2edacf8
JV		 1537 res = netdev_set_master(slave_dev, bond_dev);
1538 if (res) {
1539 dprintk("Error %d calling netdev_set_master\n", res);
569f0c4d 1540 goto err_restore_mac;
c2edacf8 1541 }
217df670
JV		 1542 /* open the slave since the application closed it */
1543 res = dev_open(slave_dev);
1544 if (res) {
1545 dprintk("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1546 goto err_unset_master;
1da177e4
LT		 1547 }
1548
1da177e4 1549 new_slave->dev = slave_dev;
0b680e75 1550 slave_dev->priv_flags |= IFF_BONDING;
1da177e4 1551
58402054 1552 if (bond_is_lb(bond)) {
1da177e4
LT		 1553 /* bond_alb_init_slave() must be called before all other stages since
1554 * it might fail and we do not want to have to undo everything
1555 */
1556 res = bond_alb_init_slave(bond, new_slave);
1557 if (res) {
569f0c4d 1558 goto err_close;
1da177e4
LT		 1559 }
1560 }
1561
1562 /* If the mode USES_PRIMARY, then the new slave gets the
1563 * master's promisc (and mc) settings only if it becomes the
1564 * curr_active_slave, and that is taken care of later when calling
1565 * bond_change_active()
1566 */
1567 if (!USES_PRIMARY(bond->params.mode)) {
1568 /* set promiscuity level to new slave */
1569 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1570 res = dev_set_promiscuity(slave_dev, 1);
1571 if (res)
1572 goto err_close;
1da177e4
LT		 1573 }
1574
1575 /* set allmulti level to new slave */
1576 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1577 res = dev_set_allmulti(slave_dev, 1);
1578 if (res)
1579 goto err_close;
1da177e4
LT		 1580 }
1581
b9e40857 1582 netif_addr_lock_bh(bond_dev);
1da177e4
LT		 1583 /* upload master's mc_list to new slave */
1584 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1585 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1586 }
b9e40857 1587 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1588 }
1589
1590 if (bond->params.mode == BOND_MODE_8023AD) {
1591 /* add lacpdu mc addr to mc list */
1592 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1593
1594 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1595 }
1596
1597 bond_add_vlans_on_slave(bond, slave_dev);
1598
1599 write_lock_bh(&bond->lock);
1600
1601 bond_attach_slave(bond, new_slave);
1602
1603 new_slave->delay = 0;
1604 new_slave->link_failure_count = 0;
1605
8531c5ff
AK		 1606 bond_compute_features(bond);
1607
3915c1e8
JV		 1608 write_unlock_bh(&bond->lock);
1609
1610 read_lock(&bond->lock);
1611
f5b2b966
JV		 1612 new_slave->last_arp_rx = jiffies;
1613
1da177e4
LT		 1614 if (bond->params.miimon && !bond->params.use_carrier) {
1615 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1616
1617 if ((link_reporting == -1) && !bond->params.arp_interval) {
1618 /*
1619 * miimon is set but a bonded network driver
1620 * does not support ETHTOOL/MII and
1621 * arp_interval is not set. Note: if
1622 * use_carrier is enabled, we will never go
1623 * here (because netif_carrier is always
1624 * supported); thus, we don't need to change
1625 * the messages for netif_carrier.
1626 */
1627 printk(KERN_WARNING DRV_NAME
4e0952c7 1628 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1629 "available for interface %s, and "
1630 "arp_interval/arp_ip_target module parameters "
1631 "not specified, thus bonding will not detect "
1632 "link failures! see bonding.txt for details.\n",
4e0952c7 1633 bond_dev->name, slave_dev->name);
1da177e4
LT		 1634 } else if (link_reporting == -1) {
1635 /* unable get link status using mii/ethtool */
1636 printk(KERN_WARNING DRV_NAME
4e0952c7 1637 ": %s: Warning: can't get link status from "
1da177e4
LT		 1638 "interface %s; the network driver associated "
1639 "with this interface does not support MII or "
1640 "ETHTOOL link status reporting, thus miimon "
1641 "has no effect on this interface.\n",
4e0952c7 1642 bond_dev->name, slave_dev->name);
1da177e4
LT		 1643 }
1644 }
1645
1646 /* check for initial state */
1647 if (!bond->params.miimon ||
1648 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1649 if (bond->params.updelay) {
1650 dprintk("Initial state of slave_dev is "
1651 "BOND_LINK_BACK\n");
1652 new_slave->link = BOND_LINK_BACK;
1653 new_slave->delay = bond->params.updelay;
1654 } else {
1655 dprintk("Initial state of slave_dev is "
1656 "BOND_LINK_UP\n");
1657 new_slave->link = BOND_LINK_UP;
1658 }
1659 new_slave->jiffies = jiffies;
1660 } else {
1661 dprintk("Initial state of slave_dev is "
1662 "BOND_LINK_DOWN\n");
1663 new_slave->link = BOND_LINK_DOWN;
1664 }
1665
1666 if (bond_update_speed_duplex(new_slave) &&
1667 (new_slave->link != BOND_LINK_DOWN)) {
1668 printk(KERN_WARNING DRV_NAME
4e0952c7 1669 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1670 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1671 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1672
1673 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1674 printk(KERN_WARNING DRV_NAME
1675 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1676 "support in base driver for proper aggregator "
4e0952c7 1677 "selection.\n", bond_dev->name);
1da177e4
LT		 1678 }
1679 }
1680
1681 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1682 /* if there is a primary slave, remember it */
1683 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1684 bond->primary_slave = new_slave;
1685 }
1686 }
1687
3915c1e8
JV		 1688 write_lock_bh(&bond->curr_slave_lock);
1689
1da177e4
LT		 1690 switch (bond->params.mode) {
1691 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1692 bond_set_slave_inactive_flags(new_slave);
1693 bond_select_active_slave(bond);
1da177e4
LT		 1694 break;
1695 case BOND_MODE_8023AD:
1696 /* in 802.3ad mode, the internal mechanism
1697 * will activate the slaves in the selected
1698 * aggregator
1699 */
1700 bond_set_slave_inactive_flags(new_slave);
1701 /* if this is the first slave */
1702 if (bond->slave_cnt == 1) {
1703 SLAVE_AD_INFO(new_slave).id = 1;
1704 /* Initialize AD with the number of times that the AD timer is called in 1 second
1705 * can be called only after the mac address of the bond is set
1706 */
1707 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1708 bond->params.lacp_fast);
1709 } else {
1710 SLAVE_AD_INFO(new_slave).id =
1711 SLAVE_AD_INFO(new_slave->prev).id + 1;
1712 }
1713
1714 bond_3ad_bind_slave(new_slave);
1715 break;
1716 case BOND_MODE_TLB:
1717 case BOND_MODE_ALB:
1718 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1719 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1720 break;
1721 default:
1722 dprintk("This slave is always active in trunk mode\n");
1723
1724 /* always active in trunk mode */
1725 new_slave->state = BOND_STATE_ACTIVE;
1726
1727 /* In trunking mode there is little meaning to curr_active_slave
1728 * anyway (it holds no special properties of the bond device),
1729 * so we can change it without calling change_active_interface()
1730 */
1731 if (!bond->curr_active_slave) {
1732 bond->curr_active_slave = new_slave;
1733 }
1734 break;
1735 } /* switch(bond_mode) */
1736
3915c1e8
JV		 1737 write_unlock_bh(&bond->curr_slave_lock);
1738
ff59c456
JV		 1739 bond_set_carrier(bond);
1740
3915c1e8 1741 read_unlock(&bond->lock);
1da177e4 1742
b76cdba9
MW		 1743 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1744 if (res)
569f0c4d 1745 goto err_close;
b76cdba9 1746
1da177e4
LT		 1747 printk(KERN_INFO DRV_NAME
1748 ": %s: enslaving %s as a%s interface with a%s link.\n",
1749 bond_dev->name, slave_dev->name,
1750 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1751 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1752
1753 /* enslave is successful */
1754 return 0;
1755
1756/* Undo stages on error */
1da177e4
LT		 1757err_close:
1758 dev_close(slave_dev);
1759
569f0c4d
JV		 1760err_unset_master:
1761 netdev_set_master(slave_dev, NULL);
1762
1da177e4 1763err_restore_mac:
dd957c57 1764 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1765 /* XXX TODO - fom follow mode needs to change master's
1766 * MAC if this slave's MAC is in use by the bond, or at
1767 * least print a warning.
1768 */
2ab82852
MS		 1769 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1770 addr.sa_family = slave_dev->type;
1771 dev_set_mac_address(slave_dev, &addr);
1772 }
1da177e4
LT		 1773
1774err_free:
1775 kfree(new_slave);
1776
1777err_undo_flags:
1778 bond_dev->features = old_features;
8531c5ff 1779
1da177e4
LT		 1780 return res;
1781}
1782
1783/*
1784 * Try to release the slave device <slave> from the bond device <master>
1785 * It is legal to access curr_active_slave without a lock because all the function
1786 * is write-locked.
1787 *
1788 * The rules for slave state should be:
1789 * for Active/Backup:
1790 * Active stays on all backups go down
1791 * for Bonded connections:
1792 * The first up interface should be left on and all others downed.
1793 */
a77b5325 1794int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1795{
454d7c9b 1796 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1797 struct slave *slave, *oldcurrent;
1798 struct sockaddr addr;
1799 int mac_addr_differ;
1800
1801 /* slave is not a slave or master is not master of this slave */
1802 if (!(slave_dev->flags & IFF_SLAVE) ||
1803 (slave_dev->master != bond_dev)) {
1804 printk(KERN_ERR DRV_NAME
4e0952c7 1805 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1806 bond_dev->name, slave_dev->name);
1807 return -EINVAL;
1808 }
1809
1810 write_lock_bh(&bond->lock);
1811
1812 slave = bond_get_slave_by_dev(bond, slave_dev);
1813 if (!slave) {
1814 /* not a slave of this bond */
1815 printk(KERN_INFO DRV_NAME
1816 ": %s: %s not enslaved\n",
1817 bond_dev->name, slave_dev->name);
f5e2a7b2 1818 write_unlock_bh(&bond->lock);
1da177e4
LT		 1819 return -EINVAL;
1820 }
1821
3915c1e8
JV		 1822 if (!bond->params.fail_over_mac) {
1823 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1824 ETH_ALEN);
1825 if (!mac_addr_differ && (bond->slave_cnt > 1))
1826 printk(KERN_WARNING DRV_NAME
1827 ": %s: Warning: the permanent HWaddr of %s - "
e174961c 1828 "%pM - is still in use by %s. "
3915c1e8
JV		 1829 "Set the HWaddr of %s to a different address "
1830 "to avoid conflicts.\n",
1831 bond_dev->name, slave_dev->name,
e174961c 1832 slave->perm_hwaddr,
3915c1e8 1833 bond_dev->name, slave_dev->name);
1da177e4
LT		 1834 }
1835
1836 /* Inform AD package of unbinding of slave. */
1837 if (bond->params.mode == BOND_MODE_8023AD) {
1838 /* must be called before the slave is
1839 * detached from the list
1840 */
1841 bond_3ad_unbind_slave(slave);
1842 }
1843
1844 printk(KERN_INFO DRV_NAME
1845 ": %s: releasing %s interface %s\n",
1846 bond_dev->name,
1847 (slave->state == BOND_STATE_ACTIVE)
1848 ? "active" : "backup",
1849 slave_dev->name);
1850
1851 oldcurrent = bond->curr_active_slave;
1852
1853 bond->current_arp_slave = NULL;
1854
1855 /* release the slave from its bond */
1856 bond_detach_slave(bond, slave);
1857
8531c5ff
AK		 1858 bond_compute_features(bond);
1859
1da177e4
LT		 1860 if (bond->primary_slave == slave) {
1861 bond->primary_slave = NULL;
1862 }
1863
1864 if (oldcurrent == slave) {
1865 bond_change_active_slave(bond, NULL);
1866 }
1867
58402054 1868 if (bond_is_lb(bond)) {
1da177e4
LT		 1869 /* Must be called only after the slave has been
1870 * detached from the list and the curr_active_slave
1871 * has been cleared (if our_slave == old_current),
1872 * but before a new active slave is selected.
1873 */
2543331d 1874 write_unlock_bh(&bond->lock);
1da177e4 1875 bond_alb_deinit_slave(bond, slave);
2543331d 1876 write_lock_bh(&bond->lock);
1da177e4
LT		 1877 }
1878
059fe7a5
JV		 1879 if (oldcurrent == slave) {
1880 /*
1881 * Note that we hold RTNL over this sequence, so there
1882 * is no concern that another slave add/remove event
1883 * will interfere.
1884 */
1885 write_unlock_bh(&bond->lock);
1886 read_lock(&bond->lock);
1887 write_lock_bh(&bond->curr_slave_lock);
1888
1da177e4
LT		 1889 bond_select_active_slave(bond);
1890
059fe7a5
JV		 1891 write_unlock_bh(&bond->curr_slave_lock);
1892 read_unlock(&bond->lock);
1893 write_lock_bh(&bond->lock);
1894 }
1895
1da177e4 1896 if (bond->slave_cnt == 0) {
ff59c456
JV		 1897 bond_set_carrier(bond);
1898
1da177e4
LT		 1899 /* if the last slave was removed, zero the mac address
1900 * of the master so it will be set by the application
1901 * to the mac address of the first slave
1902 */
1903 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1904
1905 if (list_empty(&bond->vlan_list)) {
1906 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1907 } else {
1908 printk(KERN_WARNING DRV_NAME
4e0952c7 1909 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1910 "still has VLANs.\n",
4e0952c7 1911 bond_dev->name, bond_dev->name);
1da177e4 1912 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1913 ": %s: When re-adding slaves, make sure the bond's "
1914 "HW address matches its VLANs'.\n",
1915 bond_dev->name);
1da177e4
LT		 1916 }
1917 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1918 !bond_has_challenged_slaves(bond)) {
1919 printk(KERN_INFO DRV_NAME
4e0952c7 1920 ": %s: last VLAN challenged slave %s "
1da177e4 1921 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1922 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1923 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1924 }
1925
1926 write_unlock_bh(&bond->lock);
1927
b76cdba9
MW		 1928 /* must do this from outside any spinlocks */
1929 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1930
1da177e4
LT		 1931 bond_del_vlans_from_slave(bond, slave_dev);
1932
1933 /* If the mode USES_PRIMARY, then we should only remove its
1934 * promisc and mc settings if it was the curr_active_slave, but that was
1935 * already taken care of above when we detached the slave
1936 */
1937 if (!USES_PRIMARY(bond->params.mode)) {
1938 /* unset promiscuity level from slave */
1939 if (bond_dev->flags & IFF_PROMISC) {
1940 dev_set_promiscuity(slave_dev, -1);
1941 }
1942
1943 /* unset allmulti level from slave */
1944 if (bond_dev->flags & IFF_ALLMULTI) {
1945 dev_set_allmulti(slave_dev, -1);
1946 }
1947
1948 /* flush master's mc_list from slave */
b9e40857 1949 netif_addr_lock_bh(bond_dev);
1da177e4 1950 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 1951 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1952 }
1953
1954 netdev_set_master(slave_dev, NULL);
1955
1956 /* close slave before restoring its mac address */
1957 dev_close(slave_dev);
1958
3915c1e8 1959 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1960 /* restore original ("permanent") mac address */
1961 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1962 addr.sa_family = slave_dev->type;
1963 dev_set_mac_address(slave_dev, &addr);
1964 }
1da177e4 1965
8f903c70 1966 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1967 IFF_SLAVE_INACTIVE | IFF_BONDING |
1968 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1969
1970 kfree(slave);
1971
1972 return 0; /* deletion OK */
1973}
1974
d90a162a
MS		 1975/*
1976* Destroy a bonding device.
1977* Must be under rtnl_lock when this function is called.
1978*/
1979void bond_destroy(struct bonding *bond)
1980{
1981 bond_deinit(bond->dev);
1982 bond_destroy_sysfs_entry(bond);
8cbdeec6 1983 unregister_netdevice(bond->dev);
d90a162a
MS		 1984}
1985
a434e43f
JV		 1986static void bond_destructor(struct net_device *bond_dev)
1987{
244ef9b9 1988 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 1989
1990 if (bond->wq)
1991 destroy_workqueue(bond->wq);
1992
1993 netif_addr_lock_bh(bond_dev);
1994 bond_mc_list_destroy(bond);
1995 netif_addr_unlock_bh(bond_dev);
1996
1997 free_netdev(bond_dev);
1998}
1999
d90a162a
MS		 2000/*
2001* First release a slave and than destroy the bond if no more slaves iare left.
2002* Must be under rtnl_lock when this function is called.
2003*/
2004int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2005{
454d7c9b 2006 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 2007 int ret;
2008
2009 ret = bond_release(bond_dev, slave_dev);
2010 if ((ret == 0) && (bond->slave_cnt == 0)) {
2011 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2012 bond_dev->name, bond_dev->name);
2013 bond_destroy(bond);
2014 }
2015 return ret;
2016}
2017
1da177e4
LT		 2018/*
2019 * This function releases all slaves.
2020 */
2021static int bond_release_all(struct net_device *bond_dev)
2022{
454d7c9b 2023 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2024 struct slave *slave;
2025 struct net_device *slave_dev;
2026 struct sockaddr addr;
2027
2028 write_lock_bh(&bond->lock);
2029
ff59c456
JV		 2030 netif_carrier_off(bond_dev);
2031
1da177e4
LT		 2032 if (bond->slave_cnt == 0) {
2033 goto out;
2034 }
2035
2036 bond->current_arp_slave = NULL;
2037 bond->primary_slave = NULL;
2038 bond_change_active_slave(bond, NULL);
2039
2040 while ((slave = bond->first_slave) != NULL) {
2041 /* Inform AD package of unbinding of slave
2042 * before slave is detached from the list.
2043 */
2044 if (bond->params.mode == BOND_MODE_8023AD) {
2045 bond_3ad_unbind_slave(slave);
2046 }
2047
2048 slave_dev = slave->dev;
2049 bond_detach_slave(bond, slave);
2050
2543331d
JV		 2051 /* now that the slave is detached, unlock and perform
2052 * all the undo steps that should not be called from
2053 * within a lock.
2054 */
2055 write_unlock_bh(&bond->lock);
2056
58402054 2057 if (bond_is_lb(bond)) {
1da177e4
LT		 2058 /* must be called only after the slave
2059 * has been detached from the list
2060 */
2061 bond_alb_deinit_slave(bond, slave);
2062 }
2063
8531c5ff
AK		 2064 bond_compute_features(bond);
2065
b76cdba9 2066 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2067 bond_del_vlans_from_slave(bond, slave_dev);
2068
2069 /* If the mode USES_PRIMARY, then we should only remove its
2070 * promisc and mc settings if it was the curr_active_slave, but that was
2071 * already taken care of above when we detached the slave
2072 */
2073 if (!USES_PRIMARY(bond->params.mode)) {
2074 /* unset promiscuity level from slave */
2075 if (bond_dev->flags & IFF_PROMISC) {
2076 dev_set_promiscuity(slave_dev, -1);
2077 }
2078
2079 /* unset allmulti level from slave */
2080 if (bond_dev->flags & IFF_ALLMULTI) {
2081 dev_set_allmulti(slave_dev, -1);
2082 }
2083
2084 /* flush master's mc_list from slave */
b9e40857 2085 netif_addr_lock_bh(bond_dev);
1da177e4 2086 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2087 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2088 }
2089
2090 netdev_set_master(slave_dev, NULL);
2091
2092 /* close slave before restoring its mac address */
2093 dev_close(slave_dev);
2094
dd957c57 2095 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2096 /* restore original ("permanent") mac address*/
2097 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2098 addr.sa_family = slave_dev->type;
2099 dev_set_mac_address(slave_dev, &addr);
2100 }
1da177e4 2101
8f903c70
JV		 2102 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2103 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2104
2105 kfree(slave);
2106
2107 /* re-acquire the lock before getting the next slave */
2108 write_lock_bh(&bond->lock);
2109 }
2110
2111 /* zero the mac address of the master so it will be
2112 * set by the application to the mac address of the
2113 * first slave
2114 */
2115 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2116
2117 if (list_empty(&bond->vlan_list)) {
2118 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2119 } else {
2120 printk(KERN_WARNING DRV_NAME
4e0952c7 2121 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2122 "still has VLANs.\n",
4e0952c7 2123 bond_dev->name, bond_dev->name);
1da177e4 2124 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2125 ": %s: When re-adding slaves, make sure the bond's "
2126 "HW address matches its VLANs'.\n",
2127 bond_dev->name);
1da177e4
LT		 2128 }
2129
2130 printk(KERN_INFO DRV_NAME
2131 ": %s: released all slaves\n",
2132 bond_dev->name);
2133
2134out:
2135 write_unlock_bh(&bond->lock);
2136
2137 return 0;
2138}
2139
2140/*
2141 * This function changes the active slave to slave <slave_dev>.
2142 * It returns -EINVAL in the following cases.
2143 * - <slave_dev> is not found in the list.
2144 * - There is not active slave now.
2145 * - <slave_dev> is already active.
2146 * - The link state of <slave_dev> is not BOND_LINK_UP.
2147 * - <slave_dev> is not running.
2148 * In these cases, this fuction does nothing.
2149 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2150 */
2151static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2152{
454d7c9b 2153 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2154 struct slave *old_active = NULL;
2155 struct slave *new_active = NULL;
2156 int res = 0;
2157
2158 if (!USES_PRIMARY(bond->params.mode)) {
2159 return -EINVAL;
2160 }
2161
2162 /* Verify that master_dev is indeed the master of slave_dev */
2163 if (!(slave_dev->flags & IFF_SLAVE) ||
2164 (slave_dev->master != bond_dev)) {
2165 return -EINVAL;
2166 }
2167
059fe7a5 2168 read_lock(&bond->lock);
1da177e4 2169
059fe7a5 2170 read_lock(&bond->curr_slave_lock);
1da177e4 2171 old_active = bond->curr_active_slave;
059fe7a5
JV		 2172 read_unlock(&bond->curr_slave_lock);
2173
1da177e4
LT		 2174 new_active = bond_get_slave_by_dev(bond, slave_dev);
2175
2176 /*
2177 * Changing to the current active: do nothing; return success.
2178 */
2179 if (new_active && (new_active == old_active)) {
059fe7a5 2180 read_unlock(&bond->lock);
1da177e4
LT		 2181 return 0;
2182 }
2183
2184 if ((new_active) &&
2185 (old_active) &&
2186 (new_active->link == BOND_LINK_UP) &&
2187 IS_UP(new_active->dev)) {
059fe7a5 2188 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2189 bond_change_active_slave(bond, new_active);
059fe7a5 2190 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2191 } else {
2192 res = -EINVAL;
2193 }
2194
059fe7a5 2195 read_unlock(&bond->lock);
1da177e4
LT		 2196
2197 return res;
2198}
2199
1da177e4
LT		 2200static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2201{
454d7c9b 2202 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2203
2204 info->bond_mode = bond->params.mode;
2205 info->miimon = bond->params.miimon;
2206
6603a6f2 2207 read_lock(&bond->lock);
1da177e4 2208 info->num_slaves = bond->slave_cnt;
6603a6f2 2209 read_unlock(&bond->lock);
1da177e4
LT		 2210
2211 return 0;
2212}
2213
2214static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2215{
454d7c9b 2216 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2217 struct slave *slave;
2218 int i, found = 0;
2219
2220 if (info->slave_id < 0) {
2221 return -ENODEV;
2222 }
2223
6603a6f2 2224 read_lock(&bond->lock);
1da177e4
LT		 2225
2226 bond_for_each_slave(bond, slave, i) {
2227 if (i == (int)info->slave_id) {
2228 found = 1;
2229 break;
2230 }
2231 }
2232
6603a6f2 2233 read_unlock(&bond->lock);
1da177e4
LT		 2234
2235 if (found) {
2236 strcpy(info->slave_name, slave->dev->name);
2237 info->link = slave->link;
2238 info->state = slave->state;
2239 info->link_failure_count = slave->link_failure_count;
2240 } else {
2241 return -ENODEV;
2242 }
2243
2244 return 0;
2245}
2246
2247/*-------------------------------- Monitoring -------------------------------*/
2248
1da177e4 2249
f0c76d61
JV		 2250static int bond_miimon_inspect(struct bonding *bond)
2251{
2252 struct slave *slave;
2253 int i, link_state, commit = 0;
1da177e4
LT		 2254
2255 bond_for_each_slave(bond, slave, i) {
f0c76d61 2256 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2257
f0c76d61 2258 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2259
2260 switch (slave->link) {
f0c76d61
JV		 2261 case BOND_LINK_UP:
2262 if (link_state)
2263 continue;
1da177e4 2264
f0c76d61
JV		 2265 slave->link = BOND_LINK_FAIL;
2266 slave->delay = bond->params.downdelay;
2267 if (slave->delay) {
2268 printk(KERN_INFO DRV_NAME
2269 ": %s: link status down for %s"
2270 "interface %s, disabling it in %d ms.\n",
2271 bond->dev->name,
2272 (bond->params.mode ==
2273 BOND_MODE_ACTIVEBACKUP) ?
2274 ((slave->state == BOND_STATE_ACTIVE) ?
2275 "active " : "backup ") : "",
2276 slave->dev->name,
2277 bond->params.downdelay * bond->params.miimon);
1da177e4 2278 }
f0c76d61
JV		 2279 /*FALLTHRU*/
2280 case BOND_LINK_FAIL:
2281 if (link_state) {
2282 /*
2283 * recovered before downdelay expired
2284 */
2285 slave->link = BOND_LINK_UP;
1da177e4
LT		 2286 slave->jiffies = jiffies;
2287 printk(KERN_INFO DRV_NAME
2288 ": %s: link status up again after %d "
2289 "ms for interface %s.\n",
1b76b316 2290 bond->dev->name,
f0c76d61
JV		 2291 (bond->params.downdelay - slave->delay) *
2292 bond->params.miimon,
2293 slave->dev->name);
2294 continue;
1da177e4 2295 }
f0c76d61
JV		 2296
2297 if (slave->delay <= 0) {
2298 slave->new_link = BOND_LINK_DOWN;
2299 commit++;
2300 continue;
1da177e4 2301 }
1da177e4 2302
f0c76d61
JV		 2303 slave->delay--;
2304 break;
2305
2306 case BOND_LINK_DOWN:
2307 if (!link_state)
2308 continue;
2309
2310 slave->link = BOND_LINK_BACK;
2311 slave->delay = bond->params.updelay;
2312
2313 if (slave->delay) {
2314 printk(KERN_INFO DRV_NAME
2315 ": %s: link status up for "
2316 "interface %s, enabling it in %d ms.\n",
2317 bond->dev->name, slave->dev->name,
2318 bond->params.updelay *
2319 bond->params.miimon);
2320 }
2321 /*FALLTHRU*/
2322 case BOND_LINK_BACK:
2323 if (!link_state) {
2324 slave->link = BOND_LINK_DOWN;
1da177e4
LT		 2325 printk(KERN_INFO DRV_NAME
2326 ": %s: link status down again after %d "
2327 "ms for interface %s.\n",
1b76b316 2328 bond->dev->name,
f0c76d61
JV		 2329 (bond->params.updelay - slave->delay) *
2330 bond->params.miimon,
2331 slave->dev->name);
2332
2333 continue;
2334 }
2335
2336 if (slave->delay <= 0) {
2337 slave->new_link = BOND_LINK_UP;
2338 commit++;
2339 continue;
1da177e4 2340 }
f0c76d61
JV		 2341
2342 slave->delay--;
1da177e4 2343 break;
f0c76d61
JV		 2344 }
2345 }
1da177e4 2346
f0c76d61
JV		 2347 return commit;
2348}
1da177e4 2349
f0c76d61
JV		 2350static void bond_miimon_commit(struct bonding *bond)
2351{
2352 struct slave *slave;
2353 int i;
2354
2355 bond_for_each_slave(bond, slave, i) {
2356 switch (slave->new_link) {
2357 case BOND_LINK_NOCHANGE:
2358 continue;
1da177e4 2359
f0c76d61
JV		 2360 case BOND_LINK_UP:
2361 slave->link = BOND_LINK_UP;
2362 slave->jiffies = jiffies;
2363
2364 if (bond->params.mode == BOND_MODE_8023AD) {
2365 /* prevent it from being the active one */
2366 slave->state = BOND_STATE_BACKUP;
2367 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2368 /* make it immediately active */
2369 slave->state = BOND_STATE_ACTIVE;
2370 } else if (slave != bond->primary_slave) {
2371 /* prevent it from being the active one */
2372 slave->state = BOND_STATE_BACKUP;
1da177e4 2373 }
1da177e4 2374
f0c76d61
JV		 2375 printk(KERN_INFO DRV_NAME
2376 ": %s: link status definitely "
2377 "up for interface %s.\n",
2378 bond->dev->name, slave->dev->name);
1da177e4 2379
f0c76d61
JV		 2380 /* notify ad that the link status has changed */
2381 if (bond->params.mode == BOND_MODE_8023AD)
2382 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2383
58402054 2384 if (bond_is_lb(bond))
f0c76d61
JV		 2385 bond_alb_handle_link_change(bond, slave,
2386 BOND_LINK_UP);
1da177e4 2387
f0c76d61
JV		 2388 if (!bond->curr_active_slave ||
2389 (slave == bond->primary_slave))
2390 goto do_failover;
1da177e4 2391
f0c76d61 2392 continue;
059fe7a5 2393
f0c76d61 2394 case BOND_LINK_DOWN:
fba4acda
JV		 2395 if (slave->link_failure_count < UINT_MAX)
2396 slave->link_failure_count++;
2397
f0c76d61 2398 slave->link = BOND_LINK_DOWN;
1da177e4 2399
f0c76d61
JV		 2400 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2401 bond->params.mode == BOND_MODE_8023AD)
2402 bond_set_slave_inactive_flags(slave);
2403
2404 printk(KERN_INFO DRV_NAME
2405 ": %s: link status definitely down for "
2406 "interface %s, disabling it\n",
2407 bond->dev->name, slave->dev->name);
2408
2409 if (bond->params.mode == BOND_MODE_8023AD)
2410 bond_3ad_handle_link_change(slave,
2411 BOND_LINK_DOWN);
2412
2413 if (bond->params.mode == BOND_MODE_TLB ||
2414 bond->params.mode == BOND_MODE_ALB)
2415 bond_alb_handle_link_change(bond, slave,
2416 BOND_LINK_DOWN);
2417
2418 if (slave == bond->curr_active_slave)
2419 goto do_failover;
2420
2421 continue;
2422
2423 default:
2424 printk(KERN_ERR DRV_NAME
2425 ": %s: invalid new link %d on slave %s\n",
2426 bond->dev->name, slave->new_link,
2427 slave->dev->name);
2428 slave->new_link = BOND_LINK_NOCHANGE;
2429
2430 continue;
2431 }
2432
2433do_failover:
2434 ASSERT_RTNL();
2435 write_lock_bh(&bond->curr_slave_lock);
2436 bond_select_active_slave(bond);
2437 write_unlock_bh(&bond->curr_slave_lock);
2438 }
2439
2440 bond_set_carrier(bond);
1da177e4
LT		 2441}
2442
0b0eef66
JV		 2443/*
2444 * bond_mii_monitor
2445 *
2446 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2447 * inspection, then (if inspection indicates something needs to be done)
2448 * an acquisition of appropriate locks followed by a commit phase to
2449 * implement whatever link state changes are indicated.
0b0eef66
JV		 2450 */
2451void bond_mii_monitor(struct work_struct *work)
2452{
2453 struct bonding *bond = container_of(work, struct bonding,
2454 mii_work.work);
0b0eef66
JV		 2455
2456 read_lock(&bond->lock);
f0c76d61
JV		 2457 if (bond->kill_timers)
2458 goto out;
2459
2460 if (bond->slave_cnt == 0)
2461 goto re_arm;
b59f9f74
JV		 2462
2463 if (bond->send_grat_arp) {
2464 read_lock(&bond->curr_slave_lock);
2465 bond_send_gratuitous_arp(bond);
2466 read_unlock(&bond->curr_slave_lock);
2467 }
2468
305d552a
BH		 2469 if (bond->send_unsol_na) {
2470 read_lock(&bond->curr_slave_lock);
2471 bond_send_unsolicited_na(bond);
2472 read_unlock(&bond->curr_slave_lock);
2473 }
2474
f0c76d61 2475 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2476 read_unlock(&bond->lock);
2477 rtnl_lock();
2478 read_lock(&bond->lock);
f0c76d61
JV		 2479
2480 bond_miimon_commit(bond);
2481
5655662d
JV		 2482 read_unlock(&bond->lock);
2483 rtnl_unlock(); /* might sleep, hold no other locks */
2484 read_lock(&bond->lock);
0b0eef66
JV		 2485 }
2486
f0c76d61
JV		 2487re_arm:
2488 if (bond->params.miimon)
2489 queue_delayed_work(bond->wq, &bond->mii_work,
2490 msecs_to_jiffies(bond->params.miimon));
2491out:
0b0eef66 2492 read_unlock(&bond->lock);
0b0eef66 2493}
c3ade5ca 2494
d3bb52b0 2495static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2496{
2497 struct in_device *idev;
2498 struct in_ifaddr *ifa;
a144ea4b 2499 __be32 addr = 0;
c3ade5ca
JV		 2500
2501 if (!dev)
2502 return 0;
2503
2504 rcu_read_lock();
e5ed6399 2505 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2506 if (!idev)
2507 goto out;
2508
2509 ifa = idev->ifa_list;
2510 if (!ifa)
2511 goto out;
2512
2513 addr = ifa->ifa_local;
2514out:
2515 rcu_read_unlock();
2516 return addr;
2517}
2518
d3bb52b0 2519static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2520{
0883beca 2521 struct vlan_entry *vlan;
f5b2b966
JV		 2522
2523 if (ip == bond->master_ip)
2524 return 1;
2525
0883beca 2526 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2527 if (ip == vlan->vlan_ip)
2528 return 1;
2529 }
2530
2531 return 0;
2532}
2533
c3ade5ca
JV		 2534/*
2535 * We go to the (large) trouble of VLAN tagging ARP frames because
2536 * switches in VLAN mode (especially if ports are configured as
2537 * "native" to a VLAN) might not pass non-tagged frames.
2538 */
d3bb52b0 2539static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2540{
2541 struct sk_buff *skb;
2542
2543 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2544 slave_dev->name, dest_ip, src_ip, vlan_id);
2545
2546 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2547 NULL, slave_dev->dev_addr, NULL);
2548
2549 if (!skb) {
2550 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2551 return;
2552 }
2553 if (vlan_id) {
2554 skb = vlan_put_tag(skb, vlan_id);
2555 if (!skb) {
2556 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2557 return;
2558 }
2559 }
2560 arp_xmit(skb);
2561}
2562
2563
1da177e4
LT		 2564static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2565{
c3ade5ca 2566 int i, vlan_id, rv;
d3bb52b0 2567 __be32 *targets = bond->params.arp_targets;
0883beca 2568 struct vlan_entry *vlan;
c3ade5ca
JV		 2569 struct net_device *vlan_dev;
2570 struct flowi fl;
2571 struct rtable *rt;
1da177e4 2572
6b780567
MW		 2573 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2574 if (!targets[i])
2575 continue;
c3ade5ca
JV		 2576 dprintk("basa: target %x\n", targets[i]);
2577 if (list_empty(&bond->vlan_list)) {
2578 dprintk("basa: empty vlan: arp_send\n");
2579 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2580 bond->master_ip, 0);
2581 continue;
2582 }
2583
2584 /*
2585 * If VLANs are configured, we do a route lookup to
2586 * determine which VLAN interface would be used, so we
2587 * can tag the ARP with the proper VLAN tag.
2588 */
2589 memset(&fl, 0, sizeof(fl));
2590 fl.fl4_dst = targets[i];
2591 fl.fl4_tos = RTO_ONLINK;
2592
f206351a 2593 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2594 if (rv) {
2595 if (net_ratelimit()) {
2596 printk(KERN_WARNING DRV_NAME
63779436
HH		 2597 ": %s: no route to arp_ip_target %pI4\n",
2598 bond->dev->name, &fl.fl4_dst);
c3ade5ca
JV		 2599 }
2600 continue;
2601 }
2602
2603 /*
2604 * This target is not on a VLAN
2605 */
2606 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2607 ip_rt_put(rt);
c3ade5ca
JV		 2608 dprintk("basa: rtdev == bond->dev: arp_send\n");
2609 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2610 bond->master_ip, 0);
2611 continue;
2612 }
2613
2614 vlan_id = 0;
0883beca 2615 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2616 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2617 if (vlan_dev == rt->u.dst.dev) {
2618 vlan_id = vlan->vlan_id;
2619 dprintk("basa: vlan match on %s %d\n",
2620 vlan_dev->name, vlan_id);
2621 break;
2622 }
2623 }
2624
2625 if (vlan_id) {
ed4b9f80 2626 ip_rt_put(rt);
c3ade5ca
JV		 2627 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2628 vlan->vlan_ip, vlan_id);
2629 continue;
2630 }
2631
2632 if (net_ratelimit()) {
2633 printk(KERN_WARNING DRV_NAME
63779436
HH		 2634 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2635 bond->dev->name, &fl.fl4_dst,
c3ade5ca
JV		 2636 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2637 }
ed4b9f80 2638 ip_rt_put(rt);
c3ade5ca
JV		 2639 }
2640}
2641
2642/*
2643 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2644 * for each VLAN above us.
b59f9f74
JV		 2645 *
2646 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2647 */
2648static void bond_send_gratuitous_arp(struct bonding *bond)
2649{
2650 struct slave *slave = bond->curr_active_slave;
2651 struct vlan_entry *vlan;
2652 struct net_device *vlan_dev;
2653
2654 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2655 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2656
2657 if (!slave || !bond->send_grat_arp ||
2658 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2659 return;
2660
b59f9f74
JV		 2661 bond->send_grat_arp--;
2662
c3ade5ca
JV		 2663 if (bond->master_ip) {
2664 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2665 bond->master_ip, 0);
c3ade5ca
JV		 2666 }
2667
2668 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2669 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2670 if (vlan->vlan_ip) {
2671 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2672 vlan->vlan_ip, vlan->vlan_id);
2673 }
1da177e4
LT		 2674 }
2675}
2676
d3bb52b0 2677static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2678{
2679 int i;
d3bb52b0 2680 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2681
2682 targets = bond->params.arp_targets;
2683 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
63779436
HH		 2684 dprintk("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2685 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
f5b2b966
JV		 2686 if (sip == targets[i]) {
2687 if (bond_has_this_ip(bond, tip))
2688 slave->last_arp_rx = jiffies;
2689 return;
2690 }
2691 }
2692}
2693
2694static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2695{
2696 struct arphdr *arp;
2697 struct slave *slave;
2698 struct bonding *bond;
2699 unsigned char *arp_ptr;
d3bb52b0 2700 __be32 sip, tip;
f5b2b966 2701
c346dca1 2702 if (dev_net(dev) != &init_net)
e730c155
EB		 2703 goto out;
2704
f5b2b966
JV		 2705 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2706 goto out;
2707
454d7c9b 2708 bond = netdev_priv(dev);
f5b2b966
JV		 2709 read_lock(&bond->lock);
2710
2711 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2712 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2713 orig_dev ? orig_dev->name : "NULL");
2714
2715 slave = bond_get_slave_by_dev(bond, orig_dev);
2716 if (!slave || !slave_do_arp_validate(bond, slave))
2717 goto out_unlock;
2718
988b7050 2719 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2720 goto out_unlock;
2721
d0a92be0 2722 arp = arp_hdr(skb);
f5b2b966
JV		 2723 if (arp->ar_hln != dev->addr_len ||
2724 skb->pkt_type == PACKET_OTHERHOST ||
2725 skb->pkt_type == PACKET_LOOPBACK ||
2726 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2727 arp->ar_pro != htons(ETH_P_IP) ||
2728 arp->ar_pln != 4)
2729 goto out_unlock;
2730
2731 arp_ptr = (unsigned char *)(arp + 1);
2732 arp_ptr += dev->addr_len;
2733 memcpy(&sip, arp_ptr, 4);
2734 arp_ptr += 4 + dev->addr_len;
2735 memcpy(&tip, arp_ptr, 4);
2736
63779436
HH		 2737 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2738 bond->dev->name, slave->dev->name, slave->state,
2739 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2740 &sip, &tip);
f5b2b966
JV		 2741
2742 /*
2743 * Backup slaves won't see the ARP reply, but do come through
2744 * here for each ARP probe (so we swap the sip/tip to validate
2745 * the probe). In a "redundant switch, common router" type of
2746 * configuration, the ARP probe will (hopefully) travel from
2747 * the active, through one switch, the router, then the other
2748 * switch before reaching the backup.
2749 */
2750 if (slave->state == BOND_STATE_ACTIVE)
2751 bond_validate_arp(bond, slave, sip, tip);
2752 else
2753 bond_validate_arp(bond, slave, tip, sip);
2754
2755out_unlock:
2756 read_unlock(&bond->lock);
2757out:
2758 dev_kfree_skb(skb);
2759 return NET_RX_SUCCESS;
2760}
2761
1da177e4
LT		 2762/*
2763 * this function is called regularly to monitor each slave's link
2764 * ensuring that traffic is being sent and received when arp monitoring
2765 * is used in load-balancing mode. if the adapter has been dormant, then an
2766 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2767 * arp monitoring in active backup mode.
2768 */
1b76b316 2769void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2770{
1b76b316
JV		 2771 struct bonding *bond = container_of(work, struct bonding,
2772 arp_work.work);
1da177e4
LT		 2773 struct slave *slave, *oldcurrent;
2774 int do_failover = 0;
2775 int delta_in_ticks;
2776 int i;
2777
2778 read_lock(&bond->lock);
2779
5ce0da8f 2780 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2781
2782 if (bond->kill_timers) {
2783 goto out;
2784 }
2785
2786 if (bond->slave_cnt == 0) {
2787 goto re_arm;
2788 }
2789
2790 read_lock(&bond->curr_slave_lock);
2791 oldcurrent = bond->curr_active_slave;
2792 read_unlock(&bond->curr_slave_lock);
2793
2794 /* see if any of the previous devices are up now (i.e. they have
2795 * xmt and rcv traffic). the curr_active_slave does not come into
2796 * the picture unless it is null. also, slave->jiffies is not needed
2797 * here because we send an arp on each slave and give a slave as
2798 * long as it needs to get the tx/rx within the delta.
2799 * TODO: what about up/down delay in arp mode? it wasn't here before
2800 * so it can wait
2801 */
2802 bond_for_each_slave(bond, slave, i) {
2803 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2804 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2805 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2806
2807 slave->link = BOND_LINK_UP;
2808 slave->state = BOND_STATE_ACTIVE;
2809
2810 /* primary_slave has no meaning in round-robin
2811 * mode. the window of a slave being up and
2812 * curr_active_slave being null after enslaving
2813 * is closed.
2814 */
2815 if (!oldcurrent) {
2816 printk(KERN_INFO DRV_NAME
2817 ": %s: link status definitely "
2818 "up for interface %s, ",
1b76b316 2819 bond->dev->name,
1da177e4
LT		 2820 slave->dev->name);
2821 do_failover = 1;
2822 } else {
2823 printk(KERN_INFO DRV_NAME
2824 ": %s: interface %s is now up\n",
1b76b316 2825 bond->dev->name,
1da177e4
LT		 2826 slave->dev->name);
2827 }
2828 }
2829 } else {
2830 /* slave->link == BOND_LINK_UP */
2831
2832 /* not all switches will respond to an arp request
2833 * when the source ip is 0, so don't take the link down
2834 * if we don't know our ip yet
2835 */
b63bb739 2836 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2837 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2838
2839 slave->link = BOND_LINK_DOWN;
2840 slave->state = BOND_STATE_BACKUP;
2841
2842 if (slave->link_failure_count < UINT_MAX) {
2843 slave->link_failure_count++;
2844 }
2845
2846 printk(KERN_INFO DRV_NAME
2847 ": %s: interface %s is now down.\n",
1b76b316 2848 bond->dev->name,
1da177e4
LT		 2849 slave->dev->name);
2850
2851 if (slave == oldcurrent) {
2852 do_failover = 1;
2853 }
2854 }
2855 }
2856
2857 /* note: if switch is in round-robin mode, all links
2858 * must tx arp to ensure all links rx an arp - otherwise
2859 * links may oscillate or not come up at all; if switch is
2860 * in something like xor mode, there is nothing we can
2861 * do - all replies will be rx'ed on same link causing slaves
2862 * to be unstable during low/no traffic periods
2863 */
2864 if (IS_UP(slave->dev)) {
2865 bond_arp_send_all(bond, slave);
2866 }
2867 }
2868
2869 if (do_failover) {
059fe7a5 2870 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2871
2872 bond_select_active_slave(bond);
2873
059fe7a5 2874 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2875 }
2876
2877re_arm:
1b76b316
JV		 2878 if (bond->params.arp_interval)
2879 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2880out:
2881 read_unlock(&bond->lock);
2882}
2883
2884/*
b2220cad
JV		 2885 * Called to inspect slaves for active-backup mode ARP monitor link state
2886 * changes. Sets new_link in slaves to specify what action should take
2887 * place for the slave. Returns 0 if no changes are found, >0 if changes
2888 * to link states must be committed.
2889 *
2890 * Called with bond->lock held for read.
1da177e4 2891 */
b2220cad 2892static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2893{
1da177e4 2894 struct slave *slave;
b2220cad 2895 int i, commit = 0;
1da177e4 2896
b2220cad
JV		 2897 bond_for_each_slave(bond, slave, i) {
2898 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2899
b2220cad
JV		 2900 if (slave->link != BOND_LINK_UP) {
2901 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2902 delta_in_ticks)) {
2903 slave->new_link = BOND_LINK_UP;
2904 commit++;
2905 }
1da177e4 2906
b2220cad
JV		 2907 continue;
2908 }
1da177e4 2909
b2220cad
JV		 2910 /*
2911 * Give slaves 2*delta after being enslaved or made
2912 * active. This avoids bouncing, as the last receive
2913 * times need a full ARP monitor cycle to be updated.
2914 */
2915 if (!time_after_eq(jiffies, slave->jiffies +
2916 2 * delta_in_ticks))
2917 continue;
2918
2919 /*
2920 * Backup slave is down if:
2921 * - No current_arp_slave AND
2922 * - more than 3*delta since last receive AND
2923 * - the bond has an IP address
2924 *
2925 * Note: a non-null current_arp_slave indicates
2926 * the curr_active_slave went down and we are
2927 * searching for a new one; under this condition
2928 * we only take the curr_active_slave down - this
2929 * gives each slave a chance to tx/rx traffic
2930 * before being taken out
2931 */
2932 if (slave->state == BOND_STATE_BACKUP &&
2933 !bond->current_arp_slave &&
2934 time_after(jiffies, slave_last_rx(bond, slave) +
2935 3 * delta_in_ticks)) {
2936 slave->new_link = BOND_LINK_DOWN;
2937 commit++;
2938 }
2939
2940 /*
2941 * Active slave is down if:
2942 * - more than 2*delta since transmitting OR
2943 * - (more than 2*delta since receive AND
2944 * the bond has an IP address)
2945 */
2946 if ((slave->state == BOND_STATE_ACTIVE) &&
2947 (time_after_eq(jiffies, slave->dev->trans_start +
2948 2 * delta_in_ticks) ||
2949 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2950 + 2 * delta_in_ticks)))) {
2951 slave->new_link = BOND_LINK_DOWN;
2952 commit++;
2953 }
1da177e4
LT		 2954 }
2955
b2220cad
JV		 2956 read_lock(&bond->curr_slave_lock);
2957
2958 /*
2959 * Trigger a commit if the primary option setting has changed.
1da177e4 2960 */
b2220cad
JV		 2961 if (bond->primary_slave &&
2962 (bond->primary_slave != bond->curr_active_slave) &&
2963 (bond->primary_slave->link == BOND_LINK_UP))
2964 commit++;
1da177e4 2965
b2220cad 2966 read_unlock(&bond->curr_slave_lock);
1da177e4 2967
b2220cad
JV		 2968 return commit;
2969}
1da177e4 2970
b2220cad
JV		 2971/*
2972 * Called to commit link state changes noted by inspection step of
2973 * active-backup mode ARP monitor.
2974 *
2975 * Called with RTNL and bond->lock for read.
2976 */
2977static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2978{
2979 struct slave *slave;
2980 int i;
1da177e4 2981
b2220cad
JV		 2982 bond_for_each_slave(bond, slave, i) {
2983 switch (slave->new_link) {
2984 case BOND_LINK_NOCHANGE:
2985 continue;
ff59c456 2986
b2220cad
JV		 2987 case BOND_LINK_UP:
2988 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2989
b2220cad
JV		 2990 if (!bond->curr_active_slave &&
2991 time_before_eq(jiffies, slave->dev->trans_start +
2992 delta_in_ticks)) {
2993 slave->link = BOND_LINK_UP;
2994 bond_change_active_slave(bond, slave);
2995 bond->current_arp_slave = NULL;
2996
2997 printk(KERN_INFO DRV_NAME
2998 ": %s: %s is up and now the "
2999 "active interface\n",
3000 bond->dev->name, slave->dev->name);
3001
3002 } else if (bond->curr_active_slave != slave) {
3003 /* this slave has just come up but we
3004 * already have a current slave; this can
3005 * also happen if bond_enslave adds a new
3006 * slave that is up while we are searching
3007 * for a new slave
1da177e4 3008 */
b2220cad
JV		 3009 slave->link = BOND_LINK_UP;
3010 bond_set_slave_inactive_flags(slave);
3011 bond->current_arp_slave = NULL;
3012
3013 printk(KERN_INFO DRV_NAME
3014 ": %s: backup interface %s is now up\n",
3015 bond->dev->name, slave->dev->name);
3016 }
1da177e4 3017
b2220cad 3018 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3019
b2220cad 3020 break;
1da177e4 3021
b2220cad
JV		 3022 case BOND_LINK_DOWN:
3023 if (slave->link_failure_count < UINT_MAX)
3024 slave->link_failure_count++;
3025
3026 slave->link = BOND_LINK_DOWN;
3027
3028 if (slave == bond->curr_active_slave) {
3029 printk(KERN_INFO DRV_NAME
3030 ": %s: link status down for active "
3031 "interface %s, disabling it\n",
3032 bond->dev->name, slave->dev->name);
1da177e4
LT		 3033
3034 bond_set_slave_inactive_flags(slave);
3035
b2220cad
JV		 3036 write_lock_bh(&bond->curr_slave_lock);
3037
3038 bond_select_active_slave(bond);
3039 if (bond->curr_active_slave)
3040 bond->curr_active_slave->jiffies =
3041 jiffies;
3042
3043 write_unlock_bh(&bond->curr_slave_lock);
3044
3045 bond->current_arp_slave = NULL;
3046
3047 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3048 printk(KERN_INFO DRV_NAME
3049 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3050 bond->dev->name, slave->dev->name);
3051
3052 bond_set_slave_inactive_flags(slave);
1da177e4 3053 }
b2220cad
JV		 3054 break;
3055
3056 default:
3057 printk(KERN_ERR DRV_NAME
3058 ": %s: impossible: new_link %d on slave %s\n",
3059 bond->dev->name, slave->new_link,
3060 slave->dev->name);
1da177e4
LT		 3061 }
3062 }
3063
b2220cad
JV		 3064 /*
3065 * No race with changes to primary via sysfs, as we hold rtnl.
3066 */
3067 if (bond->primary_slave &&
3068 (bond->primary_slave != bond->curr_active_slave) &&
3069 (bond->primary_slave->link == BOND_LINK_UP)) {
3070 write_lock_bh(&bond->curr_slave_lock);
3071 bond_change_active_slave(bond, bond->primary_slave);
3072 write_unlock_bh(&bond->curr_slave_lock);
3073 }
1da177e4 3074
b2220cad
JV		 3075 bond_set_carrier(bond);
3076}
1da177e4 3077
b2220cad
JV		 3078/*
3079 * Send ARP probes for active-backup mode ARP monitor.
3080 *
3081 * Called with bond->lock held for read.
3082 */
3083static void bond_ab_arp_probe(struct bonding *bond)
3084{
3085 struct slave *slave;
3086 int i;
1da177e4 3087
b2220cad 3088 read_lock(&bond->curr_slave_lock);
1da177e4 3089
b2220cad
JV		 3090 if (bond->current_arp_slave && bond->curr_active_slave)
3091 printk("PROBE: c_arp %s && cas %s BAD\n",
3092 bond->current_arp_slave->dev->name,
3093 bond->curr_active_slave->dev->name);
1da177e4 3094
b2220cad
JV		 3095 if (bond->curr_active_slave) {
3096 bond_arp_send_all(bond, bond->curr_active_slave);
3097 read_unlock(&bond->curr_slave_lock);
3098 return;
3099 }
1da177e4 3100
b2220cad 3101 read_unlock(&bond->curr_slave_lock);
059fe7a5 3102
b2220cad
JV		 3103 /* if we don't have a curr_active_slave, search for the next available
3104 * backup slave from the current_arp_slave and make it the candidate
3105 * for becoming the curr_active_slave
3106 */
1da177e4 3107
b2220cad
JV		 3108 if (!bond->current_arp_slave) {
3109 bond->current_arp_slave = bond->first_slave;
3110 if (!bond->current_arp_slave)
3111 return;
3112 }
1da177e4 3113
b2220cad 3114 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3115
b2220cad
JV		 3116 /* search for next candidate */
3117 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3118 if (IS_UP(slave->dev)) {
3119 slave->link = BOND_LINK_BACK;
3120 bond_set_slave_active_flags(slave);
3121 bond_arp_send_all(bond, slave);
1da177e4 3122 slave->jiffies = jiffies;
b2220cad
JV		 3123 bond->current_arp_slave = slave;
3124 break;
1da177e4
LT		 3125 }
3126
b2220cad
JV		 3127 /* if the link state is up at this point, we
3128 * mark it down - this can happen if we have
3129 * simultaneous link failures and
3130 * reselect_active_interface doesn't make this
3131 * one the current slave so it is still marked
3132 * up when it is actually down
1da177e4 3133 */
b2220cad
JV		 3134 if (slave->link == BOND_LINK_UP) {
3135 slave->link = BOND_LINK_DOWN;
3136 if (slave->link_failure_count < UINT_MAX)
3137 slave->link_failure_count++;
1da177e4 3138
b2220cad
JV		 3139 bond_set_slave_inactive_flags(slave);
3140
3141 printk(KERN_INFO DRV_NAME
3142 ": %s: backup interface %s is now down.\n",
3143 bond->dev->name, slave->dev->name);
1da177e4 3144 }
b2220cad
JV		 3145 }
3146}
1da177e4 3147
b2220cad
JV		 3148void bond_activebackup_arp_mon(struct work_struct *work)
3149{
3150 struct bonding *bond = container_of(work, struct bonding,
3151 arp_work.work);
3152 int delta_in_ticks;
1da177e4 3153
b2220cad 3154 read_lock(&bond->lock);
1da177e4 3155
b2220cad
JV		 3156 if (bond->kill_timers)
3157 goto out;
1da177e4 3158
b2220cad 3159 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3160
b2220cad
JV		 3161 if (bond->slave_cnt == 0)
3162 goto re_arm;
3163
b59f9f74
JV		 3164 if (bond->send_grat_arp) {
3165 read_lock(&bond->curr_slave_lock);
3166 bond_send_gratuitous_arp(bond);
3167 read_unlock(&bond->curr_slave_lock);
3168 }
3169
305d552a
BH		 3170 if (bond->send_unsol_na) {
3171 read_lock(&bond->curr_slave_lock);
3172 bond_send_unsolicited_na(bond);
3173 read_unlock(&bond->curr_slave_lock);
3174 }
3175
b2220cad
JV		 3176 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3177 read_unlock(&bond->lock);
3178 rtnl_lock();
3179 read_lock(&bond->lock);
3180
3181 bond_ab_arp_commit(bond, delta_in_ticks);
3182
3183 read_unlock(&bond->lock);
3184 rtnl_unlock();
3185 read_lock(&bond->lock);
1da177e4
LT		 3186 }
3187
b2220cad
JV		 3188 bond_ab_arp_probe(bond);
3189
1da177e4
LT		 3190re_arm:
3191 if (bond->params.arp_interval) {
1b76b316 3192 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3193 }
3194out:
3195 read_unlock(&bond->lock);
3196}
3197
3198/*------------------------------ proc/seq_file-------------------------------*/
3199
3200#ifdef CONFIG_PROC_FS
3201
1da177e4
LT		 3202static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3203{
3204 struct bonding *bond = seq->private;
3205 loff_t off = 0;
3206 struct slave *slave;
3207 int i;
3208
3209 /* make sure the bond won't be taken away */
3210 read_lock(&dev_base_lock);
6603a6f2 3211 read_lock(&bond->lock);
1da177e4
LT		 3212
3213 if (*pos == 0) {
3214 return SEQ_START_TOKEN;
3215 }
3216
3217 bond_for_each_slave(bond, slave, i) {
3218 if (++off == *pos) {
3219 return slave;
3220 }
3221 }
3222
3223 return NULL;
3224}
3225
3226static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3227{
3228 struct bonding *bond = seq->private;
3229 struct slave *slave = v;
3230
3231 ++*pos;
3232 if (v == SEQ_START_TOKEN) {
3233 return bond->first_slave;
3234 }
3235
3236 slave = slave->next;
3237
3238 return (slave == bond->first_slave) ? NULL : slave;
3239}
3240
3241static void bond_info_seq_stop(struct seq_file *seq, void *v)
3242{
3243 struct bonding *bond = seq->private;
3244
6603a6f2 3245 read_unlock(&bond->lock);
1da177e4
LT		 3246 read_unlock(&dev_base_lock);
3247}
3248
3249static void bond_info_show_master(struct seq_file *seq)
3250{
3251 struct bonding *bond = seq->private;
3252 struct slave *curr;
4756b02f 3253 int i;
1da177e4
LT		 3254
3255 read_lock(&bond->curr_slave_lock);
3256 curr = bond->curr_active_slave;
3257 read_unlock(&bond->curr_slave_lock);
3258
dd957c57 3259 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3260 bond_mode_name(bond->params.mode));
3261
dd957c57
JV		 3262 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3263 bond->params.fail_over_mac)
3915c1e8
JV		 3264 seq_printf(seq, " (fail_over_mac %s)",
3265 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3266
3267 seq_printf(seq, "\n");
3268
c61b75ad
MW		 3269 if (bond->params.mode == BOND_MODE_XOR ||
3270 bond->params.mode == BOND_MODE_8023AD) {
3271 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3272 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3273 bond->params.xmit_policy);
3274 }
3275
1da177e4
LT		 3276 if (USES_PRIMARY(bond->params.mode)) {
3277 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3278 (bond->primary_slave) ?
3279 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3280
3281 seq_printf(seq, "Currently Active Slave: %s\n",
3282 (curr) ? curr->dev->name : "None");
3283 }
3284
ff59c456
JV		 3285 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3286 "up" : "down");
1da177e4
LT		 3287 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3288 seq_printf(seq, "Up Delay (ms): %d\n",
3289 bond->params.updelay * bond->params.miimon);
3290 seq_printf(seq, "Down Delay (ms): %d\n",
3291 bond->params.downdelay * bond->params.miimon);
3292
4756b02f
MW		 3293
3294 /* ARP information */
3295 if(bond->params.arp_interval > 0) {
3296 int printed=0;
3297 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3298 bond->params.arp_interval);
3299
3300 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3301
3302 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3303 if (!bond->params.arp_targets[i])
3304 continue;
3305 if (printed)
3306 seq_printf(seq, ",");
8cf14e38 3307 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
4756b02f
MW		 3308 printed = 1;
3309 }
3310 seq_printf(seq, "\n");
3311 }
3312
1da177e4
LT		 3313 if (bond->params.mode == BOND_MODE_8023AD) {
3314 struct ad_info ad_info;
3315
3316 seq_puts(seq, "\n802.3ad info\n");
3317 seq_printf(seq, "LACP rate: %s\n",
3318 (bond->params.lacp_fast) ? "fast" : "slow");
fd989c83
JV		 3319 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3320 ad_select_tbl[bond->params.ad_select].modename);
1da177e4
LT		 3321
3322 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3323 seq_printf(seq, "bond %s has no active aggregator\n",
3324 bond->dev->name);
3325 } else {
3326 seq_printf(seq, "Active Aggregator Info:\n");
3327
3328 seq_printf(seq, "\tAggregator ID: %d\n",
3329 ad_info.aggregator_id);
3330 seq_printf(seq, "\tNumber of ports: %d\n",
3331 ad_info.ports);
3332 seq_printf(seq, "\tActor Key: %d\n",
3333 ad_info.actor_key);
3334 seq_printf(seq, "\tPartner Key: %d\n",
3335 ad_info.partner_key);
e174961c
JB		 3336 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3337 ad_info.partner_system);
1da177e4
LT		 3338 }
3339 }
3340}
3341
3342static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3343{
3344 struct bonding *bond = seq->private;
3345
3346 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3347 seq_printf(seq, "MII Status: %s\n",
3348 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3349 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3350 slave->link_failure_count);
3351
e174961c 3352 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
1da177e4
LT		 3353
3354 if (bond->params.mode == BOND_MODE_8023AD) {
3355 const struct aggregator *agg
3356 = SLAVE_AD_INFO(slave).port.aggregator;
3357
3358 if (agg) {
3359 seq_printf(seq, "Aggregator ID: %d\n",
3360 agg->aggregator_identifier);
3361 } else {
3362 seq_puts(seq, "Aggregator ID: N/A\n");
3363 }
3364 }
3365}
3366
3367static int bond_info_seq_show(struct seq_file *seq, void *v)
3368{
3369 if (v == SEQ_START_TOKEN) {
3370 seq_printf(seq, "%s\n", version);
3371 bond_info_show_master(seq);
3372 } else {
3373 bond_info_show_slave(seq, v);
3374 }
3375
3376 return 0;
3377}
3378
3379static struct seq_operations bond_info_seq_ops = {
3380 .start = bond_info_seq_start,
3381 .next = bond_info_seq_next,
3382 .stop = bond_info_seq_stop,
3383 .show = bond_info_seq_show,
3384};
3385
3386static int bond_info_open(struct inode *inode, struct file *file)
3387{
3388 struct seq_file *seq;
3389 struct proc_dir_entry *proc;
3390 int res;
3391
3392 res = seq_open(file, &bond_info_seq_ops);
3393 if (!res) {
3394 /* recover the pointer buried in proc_dir_entry data */
3395 seq = file->private_data;
3396 proc = PDE(inode);
3397 seq->private = proc->data;
3398 }
3399
3400 return res;
3401}
3402
d54b1fdb 3403static const struct file_operations bond_info_fops = {
1da177e4
LT		 3404 .owner = THIS_MODULE,
3405 .open = bond_info_open,
3406 .read = seq_read,
3407 .llseek = seq_lseek,
3408 .release = seq_release,
3409};
3410
3411static int bond_create_proc_entry(struct bonding *bond)
3412{
3413 struct net_device *bond_dev = bond->dev;
3414
3415 if (bond_proc_dir) {
a95609cb
DL		 3416 bond->proc_entry = proc_create_data(bond_dev->name,
3417 S_IRUGO, bond_proc_dir,
3418 &bond_info_fops, bond);
1da177e4
LT		 3419 if (bond->proc_entry == NULL) {
3420 printk(KERN_WARNING DRV_NAME
3421 ": Warning: Cannot create /proc/net/%s/%s\n",
3422 DRV_NAME, bond_dev->name);
3423 } else {
1da177e4
LT		 3424 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3425 }
3426 }
3427
3428 return 0;
3429}
3430
3431static void bond_remove_proc_entry(struct bonding *bond)
3432{
3433 if (bond_proc_dir && bond->proc_entry) {
3434 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3435 memset(bond->proc_file_name, 0, IFNAMSIZ);
3436 bond->proc_entry = NULL;
3437 }
3438}
3439
3440/* Create the bonding directory under /proc/net, if doesn't exist yet.
3441 * Caller must hold rtnl_lock.
3442 */
3443static void bond_create_proc_dir(void)
3444{
3445 int len = strlen(DRV_NAME);
3446
457c4cbc 3447 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3448 bond_proc_dir = bond_proc_dir->next) {
3449 if ((bond_proc_dir->namelen == len) &&
3450 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3451 break;
3452 }
3453 }
3454
3455 if (!bond_proc_dir) {
457c4cbc 3456 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3457 if (bond_proc_dir) {
3458 bond_proc_dir->owner = THIS_MODULE;
3459 } else {
3460 printk(KERN_WARNING DRV_NAME
3461 ": Warning: cannot create /proc/net/%s\n",
3462 DRV_NAME);
3463 }
3464 }
3465}
3466
3467/* Destroy the bonding directory under /proc/net, if empty.
3468 * Caller must hold rtnl_lock.
3469 */
3470static void bond_destroy_proc_dir(void)
3471{
3472 struct proc_dir_entry *de;
3473
3474 if (!bond_proc_dir) {
3475 return;
3476 }
3477
3478 /* verify that the /proc dir is empty */
3479 for (de = bond_proc_dir->subdir; de; de = de->next) {
3480 /* ignore . and .. */
3481 if (*(de->name) != '.') {
3482 break;
3483 }
3484 }
3485
3486 if (de) {
3487 if (bond_proc_dir->owner == THIS_MODULE) {
3488 bond_proc_dir->owner = NULL;
3489 }
3490 } else {
457c4cbc 3491 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3492 bond_proc_dir = NULL;
3493 }
3494}
3495#endif /* CONFIG_PROC_FS */
3496
3497/*-------------------------- netdev event handling --------------------------*/
3498
3499/*
3500 * Change device name
3501 */
3502static int bond_event_changename(struct bonding *bond)
3503{
3504#ifdef CONFIG_PROC_FS
3505 bond_remove_proc_entry(bond);
3506 bond_create_proc_entry(bond);
3507#endif
b76cdba9
MW		 3508 down_write(&(bonding_rwsem));
3509 bond_destroy_sysfs_entry(bond);
3510 bond_create_sysfs_entry(bond);
3511 up_write(&(bonding_rwsem));
1da177e4
LT		 3512 return NOTIFY_DONE;
3513}
3514
3515static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3516{
454d7c9b 3517 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 3518
3519 switch (event) {
3520 case NETDEV_CHANGENAME:
3521 return bond_event_changename(event_bond);
3522 case NETDEV_UNREGISTER:
3b96c858 3523 bond_release_all(event_bond->dev);
1da177e4
LT		 3524 break;
3525 default:
3526 break;
3527 }
3528
3529 return NOTIFY_DONE;
3530}
3531
3532static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3533{
3534 struct net_device *bond_dev = slave_dev->master;
454d7c9b 3535 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3536
3537 switch (event) {
3538 case NETDEV_UNREGISTER:
3539 if (bond_dev) {
1284cd3a
JV		 3540 if (bond->setup_by_slave)
3541 bond_release_and_destroy(bond_dev, slave_dev);
3542 else
3543 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3544 }
3545 break;
3546 case NETDEV_CHANGE:
3547 /*
3548 * TODO: is this what we get if somebody
3549 * sets up a hierarchical bond, then rmmod's
3550 * one of the slave bonding devices?
3551 */
3552 break;
3553 case NETDEV_DOWN:
3554 /*
3555 * ... Or is it this?
3556 */
3557 break;
3558 case NETDEV_CHANGEMTU:
3559 /*
3560 * TODO: Should slaves be allowed to
3561 * independently alter their MTU? For
3562 * an active-backup bond, slaves need
3563 * not be the same type of device, so
3564 * MTUs may vary. For other modes,
3565 * slaves arguably should have the
3566 * same MTUs. To do this, we'd need to
3567 * take over the slave's change_mtu
3568 * function for the duration of their
3569 * servitude.
3570 */
3571 break;
3572 case NETDEV_CHANGENAME:
3573 /*
3574 * TODO: handle changing the primary's name
3575 */
3576 break;
8531c5ff
AK		 3577 case NETDEV_FEAT_CHANGE:
3578 bond_compute_features(bond);
3579 break;
1da177e4
LT		 3580 default:
3581 break;
3582 }
3583
3584 return NOTIFY_DONE;
3585}
3586
3587/*
3588 * bond_netdev_event: handle netdev notifier chain events.
3589 *
3590 * This function receives events for the netdev chain. The caller (an
e041c683 3591 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3592 * locks for us to safely manipulate the slave devices (RTNL lock,
3593 * dev_probe_lock).
3594 */
3595static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3596{
3597 struct net_device *event_dev = (struct net_device *)ptr;
3598
c346dca1 3599 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3600 return NOTIFY_DONE;
3601
1da177e4
LT		 3602 dprintk("event_dev: %s, event: %lx\n",
3603 (event_dev ? event_dev->name : "None"),
3604 event);
3605
0b680e75
JV		 3606 if (!(event_dev->priv_flags & IFF_BONDING))
3607 return NOTIFY_DONE;
3608
1da177e4
LT		 3609 if (event_dev->flags & IFF_MASTER) {
3610 dprintk("IFF_MASTER\n");
3611 return bond_master_netdev_event(event, event_dev);
3612 }
3613
3614 if (event_dev->flags & IFF_SLAVE) {
3615 dprintk("IFF_SLAVE\n");
3616 return bond_slave_netdev_event(event, event_dev);
3617 }
3618
3619 return NOTIFY_DONE;
3620}
3621
c3ade5ca
JV		 3622/*
3623 * bond_inetaddr_event: handle inetaddr notifier chain events.
3624 *
3625 * We keep track of device IPs primarily to use as source addresses in
3626 * ARP monitor probes (rather than spewing out broadcasts all the time).
3627 *
3628 * We track one IP for the main device (if it has one), plus one per VLAN.
3629 */
3630static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3631{
3632 struct in_ifaddr *ifa = ptr;
3633 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3634 struct bonding *bond;
3635 struct vlan_entry *vlan;
c3ade5ca 3636
c346dca1 3637 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3638 return NOTIFY_DONE;
3639
0883beca 3640 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3641 if (bond->dev == event_dev) {
3642 switch (event) {
3643 case NETDEV_UP:
3644 bond->master_ip = ifa->ifa_local;
3645 return NOTIFY_OK;
3646 case NETDEV_DOWN:
3647 bond->master_ip = bond_glean_dev_ip(bond->dev);
3648 return NOTIFY_OK;
3649 default:
3650 return NOTIFY_DONE;
3651 }
3652 }
3653
0883beca 3654 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3655 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3656 if (vlan_dev == event_dev) {
3657 switch (event) {
3658 case NETDEV_UP:
3659 vlan->vlan_ip = ifa->ifa_local;
3660 return NOTIFY_OK;
3661 case NETDEV_DOWN:
3662 vlan->vlan_ip =
3663 bond_glean_dev_ip(vlan_dev);
3664 return NOTIFY_OK;
3665 default:
3666 return NOTIFY_DONE;
3667 }
3668 }
3669 }
3670 }
3671 return NOTIFY_DONE;
3672}
3673
1da177e4
LT		 3674static struct notifier_block bond_netdev_notifier = {
3675 .notifier_call = bond_netdev_event,
3676};
3677
c3ade5ca
JV		 3678static struct notifier_block bond_inetaddr_notifier = {
3679 .notifier_call = bond_inetaddr_event,
3680};
3681
1da177e4
LT		 3682/*-------------------------- Packet type handling ---------------------------*/
3683
3684/* register to receive lacpdus on a bond */
3685static void bond_register_lacpdu(struct bonding *bond)
3686{
3687 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3688
3689 /* initialize packet type */
3690 pk_type->type = PKT_TYPE_LACPDU;
3691 pk_type->dev = bond->dev;
3692 pk_type->func = bond_3ad_lacpdu_recv;
3693
3694 dev_add_pack(pk_type);
3695}
3696
3697/* unregister to receive lacpdus on a bond */
3698static void bond_unregister_lacpdu(struct bonding *bond)
3699{
3700 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3701}
3702
f5b2b966
JV		 3703void bond_register_arp(struct bonding *bond)
3704{
3705 struct packet_type *pt = &bond->arp_mon_pt;
3706
c4f283b1
JV		 3707 if (pt->type)
3708 return;
3709
f5b2b966 3710 pt->type = htons(ETH_P_ARP);
e245cb71 3711 pt->dev = bond->dev;
f5b2b966
JV		 3712 pt->func = bond_arp_rcv;
3713 dev_add_pack(pt);
3714}
3715
3716void bond_unregister_arp(struct bonding *bond)
3717{
c4f283b1
JV		 3718 struct packet_type *pt = &bond->arp_mon_pt;
3719
3720 dev_remove_pack(pt);
3721 pt->type = 0;
f5b2b966
JV		 3722}
3723
169a3e66
JV		 3724/*---------------------------- Hashing Policies -----------------------------*/
3725
6f6652be
JV		 3726/*
3727 * Hash for the output device based upon layer 2 and layer 3 data. If
3728 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3729 */
3730static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3731 struct net_device *bond_dev, int count)
3732{
3733 struct ethhdr *data = (struct ethhdr *)skb->data;
3734 struct iphdr *iph = ip_hdr(skb);
3735
f14c4e4e 3736 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be
JV		 3737 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3738 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3739 }
3740
3741 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3742}
3743
169a3e66 3744/*
59c51591 3745 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3746 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3747 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3748 */
3749static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3750 struct net_device *bond_dev, int count)
3751{
3752 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3753 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3754 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3755 int layer4_xor = 0;
3756
f14c4e4e
BH		 3757 if (skb->protocol == htons(ETH_P_IP)) {
3758 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3759 (iph->protocol == IPPROTO_TCP ||
3760 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3761 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3762 }
3763 return (layer4_xor ^
3764 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3765
3766 }
3767
3768 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3769}
3770
3771/*
3772 * Hash for the output device based upon layer 2 data
3773 */
3774static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3775 struct net_device *bond_dev, int count)
3776{
3777 struct ethhdr *data = (struct ethhdr *)skb->data;
3778
3779 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3780}
3781
1da177e4
LT		 3782/*-------------------------- Device entry points ----------------------------*/
3783
3784static int bond_open(struct net_device *bond_dev)
3785{
454d7c9b 3786 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3787
3788 bond->kill_timers = 0;
3789
58402054 3790 if (bond_is_lb(bond)) {
1da177e4
LT		 3791 /* bond_alb_initialize must be called before the timer
3792 * is started.
3793 */
3794 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3795 /* something went wrong - fail the open operation */
3796 return -1;
3797 }
3798
1b76b316
JV		 3799 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3800 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3801 }
3802
3803 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3804 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3805 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3806 }
3807
3808 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3809 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3810 INIT_DELAYED_WORK(&bond->arp_work,
3811 bond_activebackup_arp_mon);
3812 else
3813 INIT_DELAYED_WORK(&bond->arp_work,
3814 bond_loadbalance_arp_mon);
3815
3816 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3817 if (bond->params.arp_validate)
3818 bond_register_arp(bond);
1da177e4
LT		 3819 }
3820
3821 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3822 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3823 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3824 /* register to receive LACPDUs */
3825 bond_register_lacpdu(bond);
fd989c83 3826 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3827 }
3828
3829 return 0;
3830}
3831
3832static int bond_close(struct net_device *bond_dev)
3833{
454d7c9b 3834 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3835
3836 if (bond->params.mode == BOND_MODE_8023AD) {
3837 /* Unregister the receive of LACPDUs */
3838 bond_unregister_lacpdu(bond);
3839 }
3840
f5b2b966
JV		 3841 if (bond->params.arp_validate)
3842 bond_unregister_arp(bond);
3843
1da177e4
LT		 3844 write_lock_bh(&bond->lock);
3845
b59f9f74 3846 bond->send_grat_arp = 0;
305d552a 3847 bond->send_unsol_na = 0;
1da177e4
LT		 3848
3849 /* signal timers not to re-arm */
3850 bond->kill_timers = 1;
3851
3852 write_unlock_bh(&bond->lock);
3853
1da177e4 3854 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3855 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3856 }
3857
3858 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3859 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3860 }
3861
3862 switch (bond->params.mode) {
3863 case BOND_MODE_8023AD:
1b76b316 3864 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3865 break;
3866 case BOND_MODE_TLB:
3867 case BOND_MODE_ALB:
1b76b316 3868 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3869 break;
3870 default:
3871 break;
3872 }
3873
1da177e4 3874
58402054 3875 if (bond_is_lb(bond)) {
1da177e4
LT		 3876 /* Must be called only after all
3877 * slaves have been released
3878 */
3879 bond_alb_deinitialize(bond);
3880 }
3881
3882 return 0;
3883}
3884
3885static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3886{
454d7c9b 3887 struct bonding *bond = netdev_priv(bond_dev);
eeda3fd6 3888 struct net_device_stats *stats = &bond->stats;
2439f9eb 3889 struct net_device_stats local_stats;
1da177e4
LT		 3890 struct slave *slave;
3891 int i;
3892
2439f9eb 3893 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3894
3895 read_lock_bh(&bond->lock);
3896
3897 bond_for_each_slave(bond, slave, i) {
eeda3fd6
SH		 3898 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3899
2439f9eb
AG		 3900 local_stats.rx_packets += sstats->rx_packets;
3901 local_stats.rx_bytes += sstats->rx_bytes;
3902 local_stats.rx_errors += sstats->rx_errors;
3903 local_stats.rx_dropped += sstats->rx_dropped;
3904
3905 local_stats.tx_packets += sstats->tx_packets;
3906 local_stats.tx_bytes += sstats->tx_bytes;
3907 local_stats.tx_errors += sstats->tx_errors;
3908 local_stats.tx_dropped += sstats->tx_dropped;
3909
3910 local_stats.multicast += sstats->multicast;
3911 local_stats.collisions += sstats->collisions;
3912
3913 local_stats.rx_length_errors += sstats->rx_length_errors;
3914 local_stats.rx_over_errors += sstats->rx_over_errors;
3915 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3916 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3917 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3918 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3919
3920 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3921 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3922 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3923 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3924 local_stats.tx_window_errors += sstats->tx_window_errors;
3925 }
3926
3927 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3928
3929 read_unlock_bh(&bond->lock);
3930
3931 return stats;
3932}
3933
3934static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3935{
3936 struct net_device *slave_dev = NULL;
3937 struct ifbond k_binfo;
3938 struct ifbond __user *u_binfo = NULL;
3939 struct ifslave k_sinfo;
3940 struct ifslave __user *u_sinfo = NULL;
3941 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3942 int res = 0;
3943
3944 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3945 bond_dev->name, cmd);
3946
3947 switch (cmd) {
1da177e4
LT		 3948 case SIOCGMIIPHY:
3949 mii = if_mii(ifr);
3950 if (!mii) {
3951 return -EINVAL;
3952 }
3953 mii->phy_id = 0;
3954 /* Fall Through */
3955 case SIOCGMIIREG:
3956 /*
3957 * We do this again just in case we were called by SIOCGMIIREG
3958 * instead of SIOCGMIIPHY.
3959 */
3960 mii = if_mii(ifr);
3961 if (!mii) {
3962 return -EINVAL;
3963 }
3964
3965 if (mii->reg_num == 1) {
454d7c9b 3966 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3967 mii->val_out = 0;
6603a6f2 3968 read_lock(&bond->lock);
1da177e4 3969 read_lock(&bond->curr_slave_lock);
4e140079 3970 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3971 mii->val_out = BMSR_LSTATUS;
3972 }
3973 read_unlock(&bond->curr_slave_lock);
6603a6f2 3974 read_unlock(&bond->lock);
1da177e4
LT		 3975 }
3976
3977 return 0;
3978 case BOND_INFO_QUERY_OLD:
3979 case SIOCBONDINFOQUERY:
3980 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3981
3982 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3983 return -EFAULT;
3984 }
3985
3986 res = bond_info_query(bond_dev, &k_binfo);
3987 if (res == 0) {
3988 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3989 return -EFAULT;
3990 }
3991 }
3992
3993 return res;
3994 case BOND_SLAVE_INFO_QUERY_OLD:
3995 case SIOCBONDSLAVEINFOQUERY:
3996 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3997
3998 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3999 return -EFAULT;
4000 }
4001
4002 res = bond_slave_info_query(bond_dev, &k_sinfo);
4003 if (res == 0) {
4004 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4005 return -EFAULT;
4006 }
4007 }
4008
4009 return res;
4010 default:
4011 /* Go on */
4012 break;
4013 }
4014
4015 if (!capable(CAP_NET_ADMIN)) {
4016 return -EPERM;
4017 }
4018
b76cdba9 4019 down_write(&(bonding_rwsem));
881d966b 4020 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4
LT		 4021
4022 dprintk("slave_dev=%p: \n", slave_dev);
4023
4024 if (!slave_dev) {
4025 res = -ENODEV;
4026 } else {
4027 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4028 switch (cmd) {
4029 case BOND_ENSLAVE_OLD:
4030 case SIOCBONDENSLAVE:
4031 res = bond_enslave(bond_dev, slave_dev);
4032 break;
4033 case BOND_RELEASE_OLD:
4034 case SIOCBONDRELEASE:
4035 res = bond_release(bond_dev, slave_dev);
4036 break;
4037 case BOND_SETHWADDR_OLD:
4038 case SIOCBONDSETHWADDR:
4039 res = bond_sethwaddr(bond_dev, slave_dev);
4040 break;
4041 case BOND_CHANGE_ACTIVE_OLD:
4042 case SIOCBONDCHANGEACTIVE:
4043 res = bond_ioctl_change_active(bond_dev, slave_dev);
4044 break;
4045 default:
4046 res = -EOPNOTSUPP;
4047 }
4048
4049 dev_put(slave_dev);
4050 }
4051
b76cdba9 4052 up_write(&(bonding_rwsem));
1da177e4
LT		 4053 return res;
4054}
4055
4056static void bond_set_multicast_list(struct net_device *bond_dev)
4057{
454d7c9b 4058 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4059 struct dev_mc_list *dmi;
4060
1da177e4
LT		 4061 /*
4062 * Do promisc before checking multicast_mode
4063 */
4064 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4065 /*
4066 * FIXME: Need to handle the error when one of the multi-slaves
4067 * encounters error.
4068 */
1da177e4
LT		 4069 bond_set_promiscuity(bond, 1);
4070 }
4071
4072 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4073 bond_set_promiscuity(bond, -1);
4074 }
4075
4076 /* set allmulti flag to slaves */
4077 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4078 /*
4079 * FIXME: Need to handle the error when one of the multi-slaves
4080 * encounters error.
4081 */
1da177e4
LT		 4082 bond_set_allmulti(bond, 1);
4083 }
4084
4085 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4086 bond_set_allmulti(bond, -1);
4087 }
4088
80ee5ad2
JV		 4089 read_lock(&bond->lock);
4090
1da177e4
LT		 4091 bond->flags = bond_dev->flags;
4092
4093 /* looking for addresses to add to slaves' mc list */
4094 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4095 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4096 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4097 }
4098 }
4099
4100 /* looking for addresses to delete from slaves' list */
4101 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4102 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4103 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4104 }
4105 }
4106
4107 /* save master's multicast list */
4108 bond_mc_list_destroy(bond);
4109 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4110
80ee5ad2 4111 read_unlock(&bond->lock);
1da177e4
LT		 4112}
4113
00829823
SH		 4114static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4115{
4116 struct bonding *bond = netdev_priv(dev);
4117 struct slave *slave = bond->first_slave;
4118
4119 if (slave) {
4120 const struct net_device_ops *slave_ops
4121 = slave->dev->netdev_ops;
4122 if (slave_ops->ndo_neigh_setup)
4123 return slave_ops->ndo_neigh_setup(dev, parms);
4124 }
4125 return 0;
4126}
4127
1da177e4
LT		 4128/*
4129 * Change the MTU of all of a master's slaves to match the master
4130 */
4131static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4132{
454d7c9b 4133 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4134 struct slave *slave, *stop_at;
4135 int res = 0;
4136 int i;
4137
4138 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4139 (bond_dev ? bond_dev->name : "None"), new_mtu);
4140
4141 /* Can't hold bond->lock with bh disabled here since
4142 * some base drivers panic. On the other hand we can't
4143 * hold bond->lock without bh disabled because we'll
4144 * deadlock. The only solution is to rely on the fact
4145 * that we're under rtnl_lock here, and the slaves
4146 * list won't change. This doesn't solve the problem
4147 * of setting the slave's MTU while it is
4148 * transmitting, but the assumption is that the base
4149 * driver can handle that.
4150 *
4151 * TODO: figure out a way to safely iterate the slaves
4152 * list, but without holding a lock around the actual
4153 * call to the base driver.
4154 */
4155
4156 bond_for_each_slave(bond, slave, i) {
4157 dprintk("s %p s->p %p c_m %p\n", slave,
4158 slave->prev, slave->dev->change_mtu);
e944ef79 4159
1da177e4
LT		 4160 res = dev_set_mtu(slave->dev, new_mtu);
4161
4162 if (res) {
4163 /* If we failed to set the slave's mtu to the new value
4164 * we must abort the operation even in ACTIVE_BACKUP
4165 * mode, because if we allow the backup slaves to have
4166 * different mtu values than the active slave we'll
4167 * need to change their mtu when doing a failover. That
4168 * means changing their mtu from timer context, which
4169 * is probably not a good idea.
4170 */
4171 dprintk("err %d %s\n", res, slave->dev->name);
4172 goto unwind;
4173 }
4174 }
4175
4176 bond_dev->mtu = new_mtu;
4177
4178 return 0;
4179
4180unwind:
4181 /* unwind from head to the slave that failed */
4182 stop_at = slave;
4183 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4184 int tmp_res;
4185
4186 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4187 if (tmp_res) {
4188 dprintk("unwind err %d dev %s\n", tmp_res,
4189 slave->dev->name);
4190 }
4191 }
4192
4193 return res;
4194}
4195
4196/*
4197 * Change HW address
4198 *
4199 * Note that many devices must be down to change the HW address, and
4200 * downing the master releases all slaves. We can make bonds full of
4201 * bonding devices to test this, however.
4202 */
4203static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4204{
454d7c9b 4205 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4206 struct sockaddr *sa = addr, tmp_sa;
4207 struct slave *slave, *stop_at;
4208 int res = 0;
4209 int i;
4210
eb7cc59a
SH		 4211 if (bond->params.mode == BOND_MODE_ALB)
4212 return bond_alb_set_mac_address(bond_dev, addr);
4213
4214
1da177e4
LT		 4215 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4216
dd957c57 4217 /*
3915c1e8
JV		 4218 * If fail_over_mac is set to active, do nothing and return
4219 * success. Returning an error causes ifenslave to fail.
dd957c57 4220 */
3915c1e8 4221 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4222 return 0;
2ab82852 4223
1da177e4
LT		 4224 if (!is_valid_ether_addr(sa->sa_data)) {
4225 return -EADDRNOTAVAIL;
4226 }
4227
4228 /* Can't hold bond->lock with bh disabled here since
4229 * some base drivers panic. On the other hand we can't
4230 * hold bond->lock without bh disabled because we'll
4231 * deadlock. The only solution is to rely on the fact
4232 * that we're under rtnl_lock here, and the slaves
4233 * list won't change. This doesn't solve the problem
4234 * of setting the slave's hw address while it is
4235 * transmitting, but the assumption is that the base
4236 * driver can handle that.
4237 *
4238 * TODO: figure out a way to safely iterate the slaves
4239 * list, but without holding a lock around the actual
4240 * call to the base driver.
4241 */
4242
4243 bond_for_each_slave(bond, slave, i) {
eb7cc59a 4244 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
1da177e4
LT		 4245 dprintk("slave %p %s\n", slave, slave->dev->name);
4246
eb7cc59a 4247 if (slave_ops->ndo_set_mac_address == NULL) {
1da177e4
LT		 4248 res = -EOPNOTSUPP;
4249 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4250 goto unwind;
4251 }
4252
4253 res = dev_set_mac_address(slave->dev, addr);
4254 if (res) {
4255 /* TODO: consider downing the slave
4256 * and retry ?
4257 * User should expect communications
4258 * breakage anyway until ARP finish
4259 * updating, so...
4260 */
4261 dprintk("err %d %s\n", res, slave->dev->name);
4262 goto unwind;
4263 }
4264 }
4265
4266 /* success */
4267 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4268 return 0;
4269
4270unwind:
4271 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4272 tmp_sa.sa_family = bond_dev->type;
4273
4274 /* unwind from head to the slave that failed */
4275 stop_at = slave;
4276 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4277 int tmp_res;
4278
4279 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4280 if (tmp_res) {
4281 dprintk("unwind err %d dev %s\n", tmp_res,
4282 slave->dev->name);
4283 }
4284 }
4285
4286 return res;
4287}
4288
4289static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4290{
454d7c9b 4291 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4292 struct slave *slave, *start_at;
cf5f9044 4293 int i, slave_no, res = 1;
1da177e4
LT		 4294
4295 read_lock(&bond->lock);
4296
4297 if (!BOND_IS_OK(bond)) {
4298 goto out;
4299 }
4300
cf5f9044
JV		 4301 /*
4302 * Concurrent TX may collide on rr_tx_counter; we accept that
4303 * as being rare enough not to justify using an atomic op here
4304 */
4305 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4306
cf5f9044
JV		 4307 bond_for_each_slave(bond, slave, i) {
4308 slave_no--;
4309 if (slave_no < 0) {
4310 break;
4311 }
1da177e4
LT		 4312 }
4313
cf5f9044 4314 start_at = slave;
1da177e4
LT		 4315 bond_for_each_slave_from(bond, slave, i, start_at) {
4316 if (IS_UP(slave->dev) &&
4317 (slave->link == BOND_LINK_UP) &&
4318 (slave->state == BOND_STATE_ACTIVE)) {
4319 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4320 break;
4321 }
4322 }
4323
1da177e4
LT		 4324out:
4325 if (res) {
4326 /* no suitable interface, frame not sent */
4327 dev_kfree_skb(skb);
4328 }
4329 read_unlock(&bond->lock);
4330 return 0;
4331}
4332
075897ce 4333
1da177e4
LT		 4334/*
4335 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4336 * the bond has a usable interface.
4337 */
4338static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4339{
454d7c9b 4340 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4341 int res = 1;
4342
1da177e4
LT		 4343 read_lock(&bond->lock);
4344 read_lock(&bond->curr_slave_lock);
4345
4346 if (!BOND_IS_OK(bond)) {
4347 goto out;
4348 }
4349
075897ce
JL		 4350 if (!bond->curr_active_slave)
4351 goto out;
4352
075897ce
JL		 4353 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4354
1da177e4
LT		 4355out:
4356 if (res) {
4357 /* no suitable interface, frame not sent */
4358 dev_kfree_skb(skb);
4359 }
4360 read_unlock(&bond->curr_slave_lock);
4361 read_unlock(&bond->lock);
4362 return 0;
4363}
4364
4365/*
169a3e66
JV		 4366 * In bond_xmit_xor() , we determine the output device by using a pre-
4367 * determined xmit_hash_policy(), If the selected device is not enabled,
4368 * find the next active slave.
1da177e4
LT		 4369 */
4370static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4371{
454d7c9b 4372 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4373 struct slave *slave, *start_at;
4374 int slave_no;
4375 int i;
4376 int res = 1;
4377
4378 read_lock(&bond->lock);
4379
4380 if (!BOND_IS_OK(bond)) {
4381 goto out;
4382 }
4383
169a3e66 4384 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4385
4386 bond_for_each_slave(bond, slave, i) {
4387 slave_no--;
4388 if (slave_no < 0) {
4389 break;
4390 }
4391 }
4392
4393 start_at = slave;
4394
4395 bond_for_each_slave_from(bond, slave, i, start_at) {
4396 if (IS_UP(slave->dev) &&
4397 (slave->link == BOND_LINK_UP) &&
4398 (slave->state == BOND_STATE_ACTIVE)) {
4399 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4400 break;
4401 }
4402 }
4403
4404out:
4405 if (res) {
4406 /* no suitable interface, frame not sent */
4407 dev_kfree_skb(skb);
4408 }
4409 read_unlock(&bond->lock);
4410 return 0;
4411}
4412
4413/*
4414 * in broadcast mode, we send everything to all usable interfaces.
4415 */
4416static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4417{
454d7c9b 4418 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4419 struct slave *slave, *start_at;
4420 struct net_device *tx_dev = NULL;
4421 int i;
4422 int res = 1;
4423
4424 read_lock(&bond->lock);
4425
4426 if (!BOND_IS_OK(bond)) {
4427 goto out;
4428 }
4429
4430 read_lock(&bond->curr_slave_lock);
4431 start_at = bond->curr_active_slave;
4432 read_unlock(&bond->curr_slave_lock);
4433
4434 if (!start_at) {
4435 goto out;
4436 }
4437
4438 bond_for_each_slave_from(bond, slave, i, start_at) {
4439 if (IS_UP(slave->dev) &&
4440 (slave->link == BOND_LINK_UP) &&
4441 (slave->state == BOND_STATE_ACTIVE)) {
4442 if (tx_dev) {
4443 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4444 if (!skb2) {
4445 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4446 ": %s: Error: bond_xmit_broadcast(): "
4447 "skb_clone() failed\n",
4448 bond_dev->name);
1da177e4
LT		 4449 continue;
4450 }
4451
4452 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4453 if (res) {
4454 dev_kfree_skb(skb2);
4455 continue;
4456 }
4457 }
4458 tx_dev = slave->dev;
4459 }
4460 }
4461
4462 if (tx_dev) {
4463 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4464 }
4465
4466out:
4467 if (res) {
4468 /* no suitable interface, frame not sent */
4469 dev_kfree_skb(skb);
4470 }
4471 /* frame sent to all suitable interfaces */
4472 read_unlock(&bond->lock);
4473 return 0;
4474}
4475
4476/*------------------------- Device initialization ---------------------------*/
4477
6f6652be
JV		 4478static void bond_set_xmit_hash_policy(struct bonding *bond)
4479{
4480 switch (bond->params.xmit_policy) {
4481 case BOND_XMIT_POLICY_LAYER23:
4482 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4483 break;
4484 case BOND_XMIT_POLICY_LAYER34:
4485 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4486 break;
4487 case BOND_XMIT_POLICY_LAYER2:
4488 default:
4489 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4490 break;
4491 }
4492}
4493
00829823
SH		 4494static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4495{
4496 const struct bonding *bond = netdev_priv(dev);
4497
4498 switch (bond->params.mode) {
4499 case BOND_MODE_ROUNDROBIN:
4500 return bond_xmit_roundrobin(skb, dev);
4501 case BOND_MODE_ACTIVEBACKUP:
4502 return bond_xmit_activebackup(skb, dev);
4503 case BOND_MODE_XOR:
4504 return bond_xmit_xor(skb, dev);
4505 case BOND_MODE_BROADCAST:
4506 return bond_xmit_broadcast(skb, dev);
4507 case BOND_MODE_8023AD:
4508 return bond_3ad_xmit_xor(skb, dev);
4509 case BOND_MODE_ALB:
4510 case BOND_MODE_TLB:
4511 return bond_alb_xmit(skb, dev);
4512 default:
4513 /* Should never happen, mode already checked */
4514 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4515 dev->name, bond->params.mode);
4516 WARN_ON_ONCE(1);
4517 dev_kfree_skb(skb);
4518 return NETDEV_TX_OK;
4519 }
4520}
4521
4522
1da177e4
LT		 4523/*
4524 * set bond mode specific net device operations
4525 */
a77b5325 4526void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4527{
169a3e66
JV		 4528 struct net_device *bond_dev = bond->dev;
4529
1da177e4
LT		 4530 switch (mode) {
4531 case BOND_MODE_ROUNDROBIN:
1da177e4
LT		 4532 break;
4533 case BOND_MODE_ACTIVEBACKUP:
1da177e4
LT		 4534 break;
4535 case BOND_MODE_XOR:
6f6652be 4536 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4537 break;
4538 case BOND_MODE_BROADCAST:
1da177e4
LT		 4539 break;
4540 case BOND_MODE_8023AD:
8f903c70 4541 bond_set_master_3ad_flags(bond);
6f6652be 4542 bond_set_xmit_hash_policy(bond);
1da177e4 4543 break;
1da177e4 4544 case BOND_MODE_ALB:
8f903c70
JV		 4545 bond_set_master_alb_flags(bond);
4546 /* FALLTHRU */
4547 case BOND_MODE_TLB:
1da177e4
LT		 4548 break;
4549 default:
4550 /* Should never happen, mode already checked */
4551 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4552 ": %s: Error: Unknown bonding mode %d\n",
4553 bond_dev->name,
1da177e4
LT		 4554 mode);
4555 break;
4556 }
4557}
4558
217df670
JV		 4559static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4560 struct ethtool_drvinfo *drvinfo)
4561{
4562 strncpy(drvinfo->driver, DRV_NAME, 32);
4563 strncpy(drvinfo->version, DRV_VERSION, 32);
4564 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4565}
4566
7282d491 4567static const struct ethtool_ops bond_ethtool_ops = {
217df670 4568 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac
SH		 4569 .get_link = ethtool_op_get_link,
4570 .get_tx_csum = ethtool_op_get_tx_csum,
4571 .get_sg = ethtool_op_get_sg,
4572 .get_tso = ethtool_op_get_tso,
4573 .get_ufo = ethtool_op_get_ufo,
4574 .get_flags = ethtool_op_get_flags,
8531c5ff
AK		 4575};
4576
eb7cc59a
SH		 4577static const struct net_device_ops bond_netdev_ops = {
4578 .ndo_open = bond_open,
4579 .ndo_stop = bond_close,
00829823 4580 .ndo_start_xmit = bond_start_xmit,
eb7cc59a
SH		 4581 .ndo_get_stats = bond_get_stats,
4582 .ndo_do_ioctl = bond_do_ioctl,
4583 .ndo_set_multicast_list = bond_set_multicast_list,
4584 .ndo_change_mtu = bond_change_mtu,
eb7cc59a 4585 .ndo_set_mac_address = bond_set_mac_address,
00829823 4586 .ndo_neigh_setup = bond_neigh_setup,
eb7cc59a
SH		 4587 .ndo_vlan_rx_register = bond_vlan_rx_register,
4588 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4589 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4590};
4591
1da177e4
LT		 4592/*
4593 * Does not allocate but creates a /proc entry.
4594 * Allowed to fail.
4595 */
3c535952 4596static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4 4597{
454d7c9b 4598 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4599
4600 dprintk("Begin bond_init for %s\n", bond_dev->name);
4601
4602 /* initialize rwlocks */
4603 rwlock_init(&bond->lock);
4604 rwlock_init(&bond->curr_slave_lock);
4605
4606 bond->params = *params; /* copy params struct */
4607
1b76b316
JV		 4608 bond->wq = create_singlethread_workqueue(bond_dev->name);
4609 if (!bond->wq)
4610 return -ENOMEM;
4611
1da177e4
LT		 4612 /* Initialize pointers */
4613 bond->first_slave = NULL;
4614 bond->curr_active_slave = NULL;
4615 bond->current_arp_slave = NULL;
4616 bond->primary_slave = NULL;
4617 bond->dev = bond_dev;
1053f62c 4618 bond->send_grat_arp = 0;
305d552a 4619 bond->send_unsol_na = 0;
d90a162a 4620 bond->setup_by_slave = 0;
1da177e4
LT		 4621 INIT_LIST_HEAD(&bond->vlan_list);
4622
4623 /* Initialize the device entry points */
eb7cc59a 4624 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 4625 bond_dev->ethtool_ops = &bond_ethtool_ops;
169a3e66 4626 bond_set_mode_ops(bond, bond->params.mode);
1da177e4 4627
a434e43f 4628 bond_dev->destructor = bond_destructor;
1da177e4
LT		 4629
4630 /* Initialize the device options */
4631 bond_dev->tx_queue_len = 0;
4632 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4633 bond_dev->priv_flags |= IFF_BONDING;
6cf3f41e
JV		 4634 if (bond->params.arp_interval)
4635 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
1da177e4
LT		 4636
4637 /* At first, we block adding VLANs. That's the only way to
4638 * prevent problems that occur when adding VLANs over an
4639 * empty bond. The block will be removed once non-challenged
4640 * slaves are enslaved.
4641 */
4642 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4643
932ff279 4644 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4645 * transmitting */
4646 bond_dev->features |= NETIF_F_LLTX;
4647
4648 /* By default, we declare the bond to be fully
4649 * VLAN hardware accelerated capable. Special
4650 * care is taken in the various xmit functions
4651 * when there are slaves that are not hw accel
4652 * capable
4653 */
1da177e4
LT		 4654 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4655 NETIF_F_HW_VLAN_RX |
4656 NETIF_F_HW_VLAN_FILTER);
4657
4658#ifdef CONFIG_PROC_FS
4659 bond_create_proc_entry(bond);
4660#endif
1da177e4
LT		 4661 list_add_tail(&bond->bond_list, &bond_dev_list);
4662
4663 return 0;
4664}
4665
fdaea7a9
JV		 4666static void bond_work_cancel_all(struct bonding *bond)
4667{
4668 write_lock_bh(&bond->lock);
4669 bond->kill_timers = 1;
4670 write_unlock_bh(&bond->lock);
4671
4672 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4673 cancel_delayed_work(&bond->mii_work);
4674
4675 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4676 cancel_delayed_work(&bond->arp_work);
4677
4678 if (bond->params.mode == BOND_MODE_ALB &&
4679 delayed_work_pending(&bond->alb_work))
4680 cancel_delayed_work(&bond->alb_work);
4681
4682 if (bond->params.mode == BOND_MODE_8023AD &&
4683 delayed_work_pending(&bond->ad_work))
4684 cancel_delayed_work(&bond->ad_work);
4685}
4686
a434e43f
JV		 4687/* De-initialize device specific data.
4688 * Caller must hold rtnl_lock.
4689 */
4690static void bond_deinit(struct net_device *bond_dev)
4691{
454d7c9b 4692 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 4693
4694 list_del(&bond->bond_list);
4695
4696 bond_work_cancel_all(bond);
4697
4698#ifdef CONFIG_PROC_FS
4699 bond_remove_proc_entry(bond);
4700#endif
4701}
4702
1da177e4
LT		 4703/* Unregister and free all bond devices.
4704 * Caller must hold rtnl_lock.
4705 */
4706static void bond_free_all(void)
4707{
4708 struct bonding *bond, *nxt;
4709
4710 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4711 struct net_device *bond_dev = bond->dev;
4712
fdaea7a9 4713 bond_work_cancel_all(bond);
70298705 4714 /* Release the bonded slaves */
4715 bond_release_all(bond_dev);
92b41daa 4716 bond_destroy(bond);
1da177e4
LT		 4717 }
4718
4719#ifdef CONFIG_PROC_FS
4720 bond_destroy_proc_dir();
4721#endif
4722}
4723
4724/*------------------------- Module initialization ---------------------------*/
4725
4726/*
4727 * Convert string input module parms. Accept either the
ece95f7f
JV		 4728 * number of the mode or its string name. A bit complicated because
4729 * some mode names are substrings of other names, and calls from sysfs
4730 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4731 */
ece95f7f 4732int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
1da177e4 4733{
ece95f7f 4734 int mode = -1, i, rv;
a42e534f 4735 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4736
a42e534f
JV		 4737 for (p = (char *)buf; *p; p++)
4738 if (!(isdigit(*p) || isspace(*p)))
4739 break;
4740
4741 if (*p)
ece95f7f 4742 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4743 else
4744 rv = sscanf(buf, "%d", &mode);
4745
4746 if (!rv)
4747 return -1;
1da177e4
LT		 4748
4749 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4750 if (mode == tbl[i].mode)
4751 return tbl[i].mode;
4752 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4753 return tbl[i].mode;
1da177e4
LT		 4754 }
4755
4756 return -1;
4757}
4758
4759static int bond_check_params(struct bond_params *params)
4760{
3915c1e8 4761 int arp_validate_value, fail_over_mac_value;
f5b2b966 4762
1da177e4
LT		 4763 /*
4764 * Convert string parameters.
4765 */
4766 if (mode) {
4767 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4768 if (bond_mode == -1) {
4769 printk(KERN_ERR DRV_NAME
4770 ": Error: Invalid bonding mode \"%s\"\n",
4771 mode == NULL ? "NULL" : mode);
4772 return -EINVAL;
4773 }
4774 }
4775
169a3e66
JV		 4776 if (xmit_hash_policy) {
4777 if ((bond_mode != BOND_MODE_XOR) &&
4778 (bond_mode != BOND_MODE_8023AD)) {
4779 printk(KERN_INFO DRV_NAME
4780 ": xor_mode param is irrelevant in mode %s\n",
4781 bond_mode_name(bond_mode));
4782 } else {
4783 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4784 xmit_hashtype_tbl);
4785 if (xmit_hashtype == -1) {
4786 printk(KERN_ERR DRV_NAME
4787 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4788 xmit_hash_policy == NULL ? "NULL" :
4789 xmit_hash_policy);
4790 return -EINVAL;
4791 }
4792 }
4793 }
4794
1da177e4
LT		 4795 if (lacp_rate) {
4796 if (bond_mode != BOND_MODE_8023AD) {
4797 printk(KERN_INFO DRV_NAME
4798 ": lacp_rate param is irrelevant in mode %s\n",
4799 bond_mode_name(bond_mode));
4800 } else {
4801 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4802 if (lacp_fast == -1) {
4803 printk(KERN_ERR DRV_NAME
4804 ": Error: Invalid lacp rate \"%s\"\n",
4805 lacp_rate == NULL ? "NULL" : lacp_rate);
4806 return -EINVAL;
4807 }
4808 }
4809 }
4810
fd989c83
JV		 4811 if (ad_select) {
4812 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4813 if (params->ad_select == -1) {
4814 printk(KERN_ERR DRV_NAME
4815 ": Error: Invalid ad_select \"%s\"\n",
4816 ad_select == NULL ? "NULL" : ad_select);
4817 return -EINVAL;
4818 }
4819
4820 if (bond_mode != BOND_MODE_8023AD) {
4821 printk(KERN_WARNING DRV_NAME
4822 ": ad_select param only affects 802.3ad mode\n");
4823 }
4824 } else {
4825 params->ad_select = BOND_AD_STABLE;
4826 }
4827
b8a9787e 4828 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4829 printk(KERN_WARNING DRV_NAME
4830 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4831 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4832 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4833 max_bonds = BOND_DEFAULT_MAX_BONDS;
4834 }
4835
4836 if (miimon < 0) {
4837 printk(KERN_WARNING DRV_NAME
4838 ": Warning: miimon module parameter (%d), "
4839 "not in range 0-%d, so it was reset to %d\n",
4840 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4841 miimon = BOND_LINK_MON_INTERV;
4842 }
4843
4844 if (updelay < 0) {
4845 printk(KERN_WARNING DRV_NAME
4846 ": Warning: updelay module parameter (%d), "
4847 "not in range 0-%d, so it was reset to 0\n",
4848 updelay, INT_MAX);
4849 updelay = 0;
4850 }
4851
4852 if (downdelay < 0) {
4853 printk(KERN_WARNING DRV_NAME
4854 ": Warning: downdelay module parameter (%d), "
4855 "not in range 0-%d, so it was reset to 0\n",
4856 downdelay, INT_MAX);
4857 downdelay = 0;
4858 }
4859
4860 if ((use_carrier != 0) && (use_carrier != 1)) {
4861 printk(KERN_WARNING DRV_NAME
4862 ": Warning: use_carrier module parameter (%d), "
4863 "not of valid value (0/1), so it was set to 1\n",
4864 use_carrier);
4865 use_carrier = 1;
4866 }
4867
7893b249
MS		 4868 if (num_grat_arp < 0 || num_grat_arp > 255) {
4869 printk(KERN_WARNING DRV_NAME
4870 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4871 "was reset to 1 \n", num_grat_arp);
4872 num_grat_arp = 1;
4873 }
4874
305d552a
BH		 4875 if (num_unsol_na < 0 || num_unsol_na > 255) {
4876 printk(KERN_WARNING DRV_NAME
4877 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4878 "was reset to 1 \n", num_unsol_na);
4879 num_unsol_na = 1;
4880 }
4881
1da177e4
LT		 4882 /* reset values for 802.3ad */
4883 if (bond_mode == BOND_MODE_8023AD) {
4884 if (!miimon) {
4885 printk(KERN_WARNING DRV_NAME
4886 ": Warning: miimon must be specified, "
4887 "otherwise bonding will not detect link "
4888 "failure, speed and duplex which are "
4889 "essential for 802.3ad operation\n");
4890 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4891 miimon = 100;
4892 }
4893 }
4894
4895 /* reset values for TLB/ALB */
4896 if ((bond_mode == BOND_MODE_TLB) ||
4897 (bond_mode == BOND_MODE_ALB)) {
4898 if (!miimon) {
4899 printk(KERN_WARNING DRV_NAME
4900 ": Warning: miimon must be specified, "
4901 "otherwise bonding will not detect link "
4902 "failure and link speed which are essential "
4903 "for TLB/ALB load balancing\n");
4904 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4905 miimon = 100;
4906 }
4907 }
4908
4909 if (bond_mode == BOND_MODE_ALB) {
4910 printk(KERN_NOTICE DRV_NAME
4911 ": In ALB mode you might experience client "
4912 "disconnections upon reconnection of a link if the "
4913 "bonding module updelay parameter (%d msec) is "
4914 "incompatible with the forwarding delay time of the "
4915 "switch\n",
4916 updelay);
4917 }
4918
4919 if (!miimon) {
4920 if (updelay || downdelay) {
4921 /* just warn the user the up/down delay will have
4922 * no effect since miimon is zero...
4923 */
4924 printk(KERN_WARNING DRV_NAME
4925 ": Warning: miimon module parameter not set "
4926 "and updelay (%d) or downdelay (%d) module "
4927 "parameter is set; updelay and downdelay have "
4928 "no effect unless miimon is set\n",
4929 updelay, downdelay);
4930 }
4931 } else {
4932 /* don't allow arp monitoring */
4933 if (arp_interval) {
4934 printk(KERN_WARNING DRV_NAME
4935 ": Warning: miimon (%d) and arp_interval (%d) "
4936 "can't be used simultaneously, disabling ARP "
4937 "monitoring\n",
4938 miimon, arp_interval);
4939 arp_interval = 0;
4940 }
4941
4942 if ((updelay % miimon) != 0) {
4943 printk(KERN_WARNING DRV_NAME
4944 ": Warning: updelay (%d) is not a multiple "
4945 "of miimon (%d), updelay rounded to %d ms\n",
4946 updelay, miimon, (updelay / miimon) * miimon);
4947 }
4948
4949 updelay /= miimon;
4950
4951 if ((downdelay % miimon) != 0) {
4952 printk(KERN_WARNING DRV_NAME
4953 ": Warning: downdelay (%d) is not a multiple "
4954 "of miimon (%d), downdelay rounded to %d ms\n",
4955 downdelay, miimon,
4956 (downdelay / miimon) * miimon);
4957 }
4958
4959 downdelay /= miimon;
4960 }
4961
4962 if (arp_interval < 0) {
4963 printk(KERN_WARNING DRV_NAME
4964 ": Warning: arp_interval module parameter (%d) "
4965 ", not in range 0-%d, so it was reset to %d\n",
4966 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4967 arp_interval = BOND_LINK_ARP_INTERV;
4968 }
4969
4970 for (arp_ip_count = 0;
4971 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4972 arp_ip_count++) {
4973 /* not complete check, but should be good enough to
4974 catch mistakes */
4975 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4976 printk(KERN_WARNING DRV_NAME
4977 ": Warning: bad arp_ip_target module parameter "
4978 "(%s), ARP monitoring will not be performed\n",
4979 arp_ip_target[arp_ip_count]);
4980 arp_interval = 0;
4981 } else {
d3bb52b0 4982 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4983 arp_target[arp_ip_count] = ip;
4984 }
4985 }
4986
4987 if (arp_interval && !arp_ip_count) {
4988 /* don't allow arping if no arp_ip_target given... */
4989 printk(KERN_WARNING DRV_NAME
4990 ": Warning: arp_interval module parameter (%d) "
4991 "specified without providing an arp_ip_target "
4992 "parameter, arp_interval was reset to 0\n",
4993 arp_interval);
4994 arp_interval = 0;
4995 }
4996
f5b2b966
JV		 4997 if (arp_validate) {
4998 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4999 printk(KERN_ERR DRV_NAME
5000 ": arp_validate only supported in active-backup mode\n");
5001 return -EINVAL;
5002 }
5003 if (!arp_interval) {
5004 printk(KERN_ERR DRV_NAME
5005 ": arp_validate requires arp_interval\n");
5006 return -EINVAL;
5007 }
5008
5009 arp_validate_value = bond_parse_parm(arp_validate,
5010 arp_validate_tbl);
5011 if (arp_validate_value == -1) {
5012 printk(KERN_ERR DRV_NAME
5013 ": Error: invalid arp_validate \"%s\"\n",
5014 arp_validate == NULL ? "NULL" : arp_validate);
5015 return -EINVAL;
5016 }
5017 } else
5018 arp_validate_value = 0;
5019
1da177e4
LT		 5020 if (miimon) {
5021 printk(KERN_INFO DRV_NAME
5022 ": MII link monitoring set to %d ms\n",
5023 miimon);
5024 } else if (arp_interval) {
5025 int i;
5026
5027 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 5028 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5029 arp_interval,
5030 arp_validate_tbl[arp_validate_value].modename,
5031 arp_ip_count);
1da177e4
LT		 5032
5033 for (i = 0; i < arp_ip_count; i++)
5034 printk (" %s", arp_ip_target[i]);
5035
5036 printk("\n");
5037
b8a9787e 5038 } else if (max_bonds) {
1da177e4
LT		 5039 /* miimon and arp_interval not set, we need one so things
5040 * work as expected, see bonding.txt for details
5041 */
5042 printk(KERN_WARNING DRV_NAME
5043 ": Warning: either miimon or arp_interval and "
5044 "arp_ip_target module parameters must be specified, "
5045 "otherwise bonding will not detect link failures! see "
5046 "bonding.txt for details.\n");
5047 }
5048
5049 if (primary && !USES_PRIMARY(bond_mode)) {
5050 /* currently, using a primary only makes sense
5051 * in active backup, TLB or ALB modes
5052 */
5053 printk(KERN_WARNING DRV_NAME
5054 ": Warning: %s primary device specified but has no "
5055 "effect in %s mode\n",
5056 primary, bond_mode_name(bond_mode));
5057 primary = NULL;
5058 }
5059
3915c1e8
JV		 5060 if (fail_over_mac) {
5061 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5062 fail_over_mac_tbl);
5063 if (fail_over_mac_value == -1) {
5064 printk(KERN_ERR DRV_NAME
5065 ": Error: invalid fail_over_mac \"%s\"\n",
5066 arp_validate == NULL ? "NULL" : arp_validate);
5067 return -EINVAL;
5068 }
5069
5070 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5071 printk(KERN_WARNING DRV_NAME
5072 ": Warning: fail_over_mac only affects "
5073 "active-backup mode.\n");
5074 } else {
5075 fail_over_mac_value = BOND_FOM_NONE;
5076 }
dd957c57 5077
1da177e4
LT		 5078 /* fill params struct with the proper values */
5079 params->mode = bond_mode;
169a3e66 5080 params->xmit_policy = xmit_hashtype;
1da177e4 5081 params->miimon = miimon;
7893b249 5082 params->num_grat_arp = num_grat_arp;
305d552a 5083 params->num_unsol_na = num_unsol_na;
1da177e4 5084 params->arp_interval = arp_interval;
f5b2b966 5085 params->arp_validate = arp_validate_value;
1da177e4
LT		 5086 params->updelay = updelay;
5087 params->downdelay = downdelay;
5088 params->use_carrier = use_carrier;
5089 params->lacp_fast = lacp_fast;
5090 params->primary[0] = 0;
3915c1e8 5091 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5092
5093 if (primary) {
5094 strncpy(params->primary, primary, IFNAMSIZ);
5095 params->primary[IFNAMSIZ - 1] = 0;
5096 }
5097
5098 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5099
5100 return 0;
5101}
5102
0daa2303 5103static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 5104static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 5105
e8a0464c
DM		 5106static void bond_set_lockdep_class_one(struct net_device *dev,
5107 struct netdev_queue *txq,
5108 void *_unused)
c773e847
DM		 5109{
5110 lockdep_set_class(&txq->_xmit_lock,
5111 &bonding_netdev_xmit_lock_key);
5112}
5113
5114static void bond_set_lockdep_class(struct net_device *dev)
5115{
cf508b12
DM		 5116 lockdep_set_class(&dev->addr_list_lock,
5117 &bonding_netdev_addr_lock_key);
e8a0464c 5118 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 5119}
5120
dfe60397 5121/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5122 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5123 * Caller must NOT hold rtnl_lock; we need to release it here before we
5124 * set up our sysfs entries.
5125 */
0dd646fe 5126int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5127{
5128 struct net_device *bond_dev;
0883beca 5129 struct bonding *bond;
dfe60397
MW		 5130 int res;
5131
5132 rtnl_lock();
027ea041
JV		 5133 down_write(&bonding_rwsem);
5134
5135 /* Check to see if the bond already exists. */
4fe4763c 5136 if (name) {
0883beca 5137 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5138 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5139 printk(KERN_ERR DRV_NAME
027ea041 5140 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5141 name);
5142 res = -EPERM;
5143 goto out_rtnl;
5144 }
5145 }
027ea041 5146
e4b91c48
JV		 5147 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5148 ether_setup);
dfe60397
MW		 5149 if (!bond_dev) {
5150 printk(KERN_ERR DRV_NAME
5151 ": %s: eek! can't alloc netdev!\n",
5152 name);
5153 res = -ENOMEM;
5154 goto out_rtnl;
5155 }
5156
e4b91c48
JV		 5157 if (!name) {
5158 res = dev_alloc_name(bond_dev, "bond%d");
5159 if (res < 0)
5160 goto out_netdev;
5161 }
5162
dfe60397
MW		 5163 /* bond_init() must be called after dev_alloc_name() (for the
5164 * /proc files), but before register_netdevice(), because we
5165 * need to set function pointers.
5166 */
5167
5168 res = bond_init(bond_dev, params);
5169 if (res < 0) {
5170 goto out_netdev;
5171 }
5172
dfe60397
MW		 5173 res = register_netdevice(bond_dev);
5174 if (res < 0) {
5175 goto out_bond;
5176 }
0daa2303 5177
c773e847 5178 bond_set_lockdep_class(bond_dev);
0daa2303 5179
ff59c456
JV		 5180 netif_carrier_off(bond_dev);
5181
027ea041 5182 up_write(&bonding_rwsem);
dfe60397 5183 rtnl_unlock(); /* allows sysfs registration of net device */
454d7c9b 5184 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
09c89279
JV		 5185 if (res < 0) {
5186 rtnl_lock();
027ea041 5187 down_write(&bonding_rwsem);
822973ba
PE		 5188 bond_deinit(bond_dev);
5189 unregister_netdevice(bond_dev);
5190 goto out_rtnl;
09c89279
JV		 5191 }
5192
5193 return 0;
5194
dfe60397
MW		 5195out_bond:
5196 bond_deinit(bond_dev);
5197out_netdev:
5198 free_netdev(bond_dev);
5199out_rtnl:
027ea041 5200 up_write(&bonding_rwsem);
dfe60397 5201 rtnl_unlock();
dfe60397
MW		 5202 return res;
5203}
5204
1da177e4
LT		 5205static int __init bonding_init(void)
5206{
1da177e4
LT		 5207 int i;
5208 int res;
0883beca 5209 struct bonding *bond;
1da177e4
LT		 5210
5211 printk(KERN_INFO "%s", version);
5212
dfe60397 5213 res = bond_check_params(&bonding_defaults);
1da177e4 5214 if (res) {
dfe60397 5215 goto out;
1da177e4
LT		 5216 }
5217
1da177e4
LT		 5218#ifdef CONFIG_PROC_FS
5219 bond_create_proc_dir();
5220#endif
027ea041
JV		 5221
5222 init_rwsem(&bonding_rwsem);
5223
1da177e4 5224 for (i = 0; i < max_bonds; i++) {
0dd646fe 5225 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5226 if (res)
5227 goto err;
1da177e4
LT		 5228 }
5229
b76cdba9
MW		 5230 res = bond_create_sysfs();
5231 if (res)
5232 goto err;
5233
1da177e4 5234 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5235 register_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5236 bond_register_ipv6_notifier();
1da177e4 5237
dfe60397
MW		 5238 goto out;
5239err:
0883beca 5240 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5241 bond_work_cancel_all(bond);
5242 destroy_workqueue(bond->wq);
5243 }
5244
ae68c398
PE		 5245 bond_destroy_sysfs();
5246
40abc270 5247 rtnl_lock();
1da177e4 5248 bond_free_all();
1da177e4 5249 rtnl_unlock();
dfe60397 5250out:
1da177e4 5251 return res;
dfe60397 5252
1da177e4
LT		 5253}
5254
5255static void __exit bonding_exit(void)
5256{
5257 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5258 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5259 bond_unregister_ipv6_notifier();
1da177e4 5260
ae68c398
PE		 5261 bond_destroy_sysfs();
5262
1da177e4
LT		 5263 rtnl_lock();
5264 bond_free_all();
5265 rtnl_unlock();
5266}
5267
5268module_init(bonding_init);
5269module_exit(bonding_exit);
5270MODULE_LICENSE("GPL");
5271MODULE_VERSION(DRV_VERSION);
5272MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5273MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5274MODULE_SUPPORTED_DEVICE("most ethernet devices");
5275
5276/*
5277 * Local variables:
5278 * c-indent-level: 8
5279 * c-basic-offset: 8
5280 * tab-width: 8
5281 * End:
5282 */
5283
kernel source for telekom puls (alcatel/mediatek)