]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/bonding/bond_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drivers/net/bonding: fix sparse warnings: context imbalance
[net-next-2.6.git] / drivers / net / bonding / bond_main.c
CommitLineData
1da177e4
LT		 1/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
1da177e4
LT		 32 */
33
1da177e4
LT		 34#include <linux/kernel.h>
35#include <linux/module.h>
1da177e4
LT		 36#include <linux/types.h>
37#include <linux/fcntl.h>
38#include <linux/interrupt.h>
39#include <linux/ptrace.h>
40#include <linux/ioport.h>
41#include <linux/in.h>
169a3e66 42#include <net/ip.h>
1da177e4 43#include <linux/ip.h>
169a3e66
JV		 44#include <linux/tcp.h>
45#include <linux/udp.h>
1da177e4
LT		 46#include <linux/slab.h>
47#include <linux/string.h>
48#include <linux/init.h>
49#include <linux/timer.h>
50#include <linux/socket.h>
51#include <linux/ctype.h>
52#include <linux/inet.h>
53#include <linux/bitops.h>
54#include <asm/system.h>
55#include <asm/io.h>
56#include <asm/dma.h>
57#include <asm/uaccess.h>
58#include <linux/errno.h>
59#include <linux/netdevice.h>
60#include <linux/inetdevice.h>
a816c7c7 61#include <linux/igmp.h>
1da177e4
LT		 62#include <linux/etherdevice.h>
63#include <linux/skbuff.h>
64#include <net/sock.h>
65#include <linux/rtnetlink.h>
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68#include <linux/smp.h>
69#include <linux/if_ether.h>
70#include <net/arp.h>
71#include <linux/mii.h>
72#include <linux/ethtool.h>
73#include <linux/if_vlan.h>
74#include <linux/if_bonding.h>
b63bb739 75#include <linux/jiffies.h>
c3ade5ca 76#include <net/route.h>
457c4cbc 77#include <net/net_namespace.h>
1da177e4
LT		 78#include "bonding.h"
79#include "bond_3ad.h"
80#include "bond_alb.h"
81
82/*---------------------------- Module parameters ----------------------------*/
83
84/* monitor all links that often (in milliseconds). <=0 disables monitoring */
85#define BOND_LINK_MON_INTERV 0
86#define BOND_LINK_ARP_INTERV 0
87
88static int max_bonds = BOND_DEFAULT_MAX_BONDS;
7893b249 89static int num_grat_arp = 1;
305d552a 90static int num_unsol_na = 1;
1da177e4
LT		 91static int miimon = BOND_LINK_MON_INTERV;
92static int updelay = 0;
93static int downdelay = 0;
94static int use_carrier = 1;
95static char *mode = NULL;
96static char *primary = NULL;
97static char *lacp_rate = NULL;
fd989c83 98static char *ad_select = NULL;
169a3e66 99static char *xmit_hash_policy = NULL;
1da177e4
LT		 100static int arp_interval = BOND_LINK_ARP_INTERV;
101static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
f5b2b966 102static char *arp_validate = NULL;
3915c1e8 103static char *fail_over_mac = NULL;
12479f9a 104struct bond_params bonding_defaults;
1da177e4
LT		 105
106module_param(max_bonds, int, 0);
107MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
7893b249
MS		 108module_param(num_grat_arp, int, 0644);
109MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
305d552a
BH		 110module_param(num_unsol_na, int, 0644);
111MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
1da177e4
LT		 112module_param(miimon, int, 0);
113MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
114module_param(updelay, int, 0);
115MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
116module_param(downdelay, int, 0);
2ac47660
MW		 117MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
118 "in milliseconds");
1da177e4 119module_param(use_carrier, int, 0);
2ac47660
MW		 120MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
121 "0 for off, 1 for on (default)");
1da177e4 122module_param(mode, charp, 0);
2ac47660
MW		 123MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
124 "1 for active-backup, 2 for balance-xor, "
125 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
126 "6 for balance-alb");
1da177e4
LT		 127module_param(primary, charp, 0);
128MODULE_PARM_DESC(primary, "Primary network device to use");
129module_param(lacp_rate, charp, 0);
2ac47660
MW		 130MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
131 "(slow/fast)");
fd989c83
JV		 132module_param(ad_select, charp, 0);
133MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
169a3e66 134module_param(xmit_hash_policy, charp, 0);
2ac47660
MW		 135MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
136 ", 1 for layer 3+4");
1da177e4
LT		 137module_param(arp_interval, int, 0);
138MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
139module_param_array(arp_ip_target, charp, NULL, 0);
140MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
f5b2b966
JV		 141module_param(arp_validate, charp, 0);
142MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
3915c1e8
JV		 143module_param(fail_over_mac, charp, 0);
144MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
1da177e4
LT		 145
146/*----------------------------- Global variables ----------------------------*/
147
f71e1309 148static const char * const version =
1da177e4
LT		 149 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
150
12479f9a 151LIST_HEAD(bond_dev_list);
1da177e4
LT		 152
153#ifdef CONFIG_PROC_FS
154static struct proc_dir_entry *bond_proc_dir = NULL;
155#endif
156
d3bb52b0 157static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
1da177e4 158static int arp_ip_count = 0;
1da177e4 159static int bond_mode = BOND_MODE_ROUNDROBIN;
169a3e66 160static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
1da177e4 161static int lacp_fast = 0;
217df670 162
1da177e4 163
e97fd7c6 164const struct bond_parm_tbl bond_lacp_tbl[] = {
1da177e4
LT		 165{ "slow", AD_LACP_SLOW},
166{ "fast", AD_LACP_FAST},
167{ NULL, -1},
168};
169
e97fd7c6 170const struct bond_parm_tbl bond_mode_tbl[] = {
1da177e4
LT		 171{ "balance-rr", BOND_MODE_ROUNDROBIN},
172{ "active-backup", BOND_MODE_ACTIVEBACKUP},
173{ "balance-xor", BOND_MODE_XOR},
174{ "broadcast", BOND_MODE_BROADCAST},
175{ "802.3ad", BOND_MODE_8023AD},
176{ "balance-tlb", BOND_MODE_TLB},
177{ "balance-alb", BOND_MODE_ALB},
178{ NULL, -1},
179};
180
e97fd7c6 181const struct bond_parm_tbl xmit_hashtype_tbl[] = {
169a3e66
JV		 182{ "layer2", BOND_XMIT_POLICY_LAYER2},
183{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
6f6652be 184{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
169a3e66
JV		 185{ NULL, -1},
186};
187
e97fd7c6 188const struct bond_parm_tbl arp_validate_tbl[] = {
f5b2b966
JV		 189{ "none", BOND_ARP_VALIDATE_NONE},
190{ "active", BOND_ARP_VALIDATE_ACTIVE},
191{ "backup", BOND_ARP_VALIDATE_BACKUP},
192{ "all", BOND_ARP_VALIDATE_ALL},
193{ NULL, -1},
194};
195
e97fd7c6 196const struct bond_parm_tbl fail_over_mac_tbl[] = {
3915c1e8
JV		 197{ "none", BOND_FOM_NONE},
198{ "active", BOND_FOM_ACTIVE},
199{ "follow", BOND_FOM_FOLLOW},
200{ NULL, -1},
201};
202
fd989c83
JV		 203struct bond_parm_tbl ad_select_tbl[] = {
204{ "stable", BOND_AD_STABLE},
205{ "bandwidth", BOND_AD_BANDWIDTH},
206{ "count", BOND_AD_COUNT},
207{ NULL, -1},
208};
209
1da177e4
LT		 210/*-------------------------- Forward declarations ---------------------------*/
211
c3ade5ca 212static void bond_send_gratuitous_arp(struct bonding *bond);
c50b85d0 213static void bond_deinit(struct net_device *bond_dev);
1da177e4
LT		 214
215/*---------------------------- General routines -----------------------------*/
216
4ad072c9 217static const char *bond_mode_name(int mode)
1da177e4 218{
77afc92b
HE		 219 static const char *names[] = {
220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
224 [BOND_MODE_8023AD]= "IEEE 802.3ad Dynamic link aggregation",
225 [BOND_MODE_TLB] = "transmit load balancing",
226 [BOND_MODE_ALB] = "adaptive load balancing",
227 };
228
229 if (mode < 0 || mode > BOND_MODE_ALB)
1da177e4 230 return "unknown";
77afc92b
HE		 231
232 return names[mode];
1da177e4
LT		 233}
234
235/*---------------------------------- VLAN -----------------------------------*/
236
237/**
238 * bond_add_vlan - add a new vlan id on bond
239 * @bond: bond that got the notification
240 * @vlan_id: the vlan id to add
241 *
242 * Returns -ENOMEM if allocation failed.
243 */
244static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
245{
246 struct vlan_entry *vlan;
247
5a03cdb7 248 pr_debug("bond: %s, vlan id %d\n",
1da177e4
LT		 249 (bond ? bond->dev->name: "None"), vlan_id);
250
305d552a 251 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
1da177e4
LT		 252 if (!vlan) {
253 return -ENOMEM;
254 }
255
256 INIT_LIST_HEAD(&vlan->vlan_list);
257 vlan->vlan_id = vlan_id;
258
259 write_lock_bh(&bond->lock);
260
261 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
262
263 write_unlock_bh(&bond->lock);
264
5a03cdb7 265 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
1da177e4
LT		 266
267 return 0;
268}
269
270/**
271 * bond_del_vlan - delete a vlan id from bond
272 * @bond: bond that got the notification
273 * @vlan_id: the vlan id to delete
274 *
275 * returns -ENODEV if @vlan_id was not found in @bond.
276 */
277static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
278{
0883beca 279 struct vlan_entry *vlan;
1da177e4
LT		 280 int res = -ENODEV;
281
5a03cdb7 282 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
1da177e4
LT		 283
284 write_lock_bh(&bond->lock);
285
0883beca 286 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
1da177e4
LT		 287 if (vlan->vlan_id == vlan_id) {
288 list_del(&vlan->vlan_list);
289
58402054 290 if (bond_is_lb(bond))
1da177e4 291 bond_alb_clear_vlan(bond, vlan_id);
1da177e4 292
5a03cdb7 293 pr_debug("removed VLAN ID %d from bond %s\n", vlan_id,
1da177e4
LT		 294 bond->dev->name);
295
296 kfree(vlan);
297
298 if (list_empty(&bond->vlan_list) &&
299 (bond->slave_cnt == 0)) {
300 /* Last VLAN removed and no slaves, so
301 * restore block on adding VLANs. This will
302 * be removed once new slaves that are not
303 * VLAN challenged will be added.
304 */
305 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
306 }
307
308 res = 0;
309 goto out;
310 }
311 }
312
5a03cdb7 313 pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id,
1da177e4
LT		 314 bond->dev->name);
315
316out:
317 write_unlock_bh(&bond->lock);
318 return res;
319}
320
321/**
322 * bond_has_challenged_slaves
323 * @bond: the bond we're working on
324 *
325 * Searches the slave list. Returns 1 if a vlan challenged slave
326 * was found, 0 otherwise.
327 *
328 * Assumes bond->lock is held.
329 */
330static int bond_has_challenged_slaves(struct bonding *bond)
331{
332 struct slave *slave;
333 int i;
334
335 bond_for_each_slave(bond, slave, i) {
336 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
5a03cdb7 337 pr_debug("found VLAN challenged slave - %s\n",
1da177e4
LT		 338 slave->dev->name);
339 return 1;
340 }
341 }
342
5a03cdb7 343 pr_debug("no VLAN challenged slaves found\n");
1da177e4
LT		 344 return 0;
345}
346
347/**
348 * bond_next_vlan - safely skip to the next item in the vlans list.
349 * @bond: the bond we're working on
350 * @curr: item we're advancing from
351 *
352 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
353 * or @curr->next otherwise (even if it is @curr itself again).
354 *
355 * Caller must hold bond->lock
356 */
357struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
358{
359 struct vlan_entry *next, *last;
360
361 if (list_empty(&bond->vlan_list)) {
362 return NULL;
363 }
364
365 if (!curr) {
366 next = list_entry(bond->vlan_list.next,
367 struct vlan_entry, vlan_list);
368 } else {
369 last = list_entry(bond->vlan_list.prev,
370 struct vlan_entry, vlan_list);
371 if (last == curr) {
372 next = list_entry(bond->vlan_list.next,
373 struct vlan_entry, vlan_list);
374 } else {
375 next = list_entry(curr->vlan_list.next,
376 struct vlan_entry, vlan_list);
377 }
378 }
379
380 return next;
381}
382
383/**
384 * bond_dev_queue_xmit - Prepare skb for xmit.
385 *
386 * @bond: bond device that got this skb for tx.
387 * @skb: hw accel VLAN tagged skb to transmit
388 * @slave_dev: slave that is supposed to xmit this skbuff
389 *
390 * When the bond gets an skb to transmit that is
391 * already hardware accelerated VLAN tagged, and it
392 * needs to relay this skb to a slave that is not
393 * hw accel capable, the skb needs to be "unaccelerated",
394 * i.e. strip the hwaccel tag and re-insert it as part
395 * of the payload.
396 */
397int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
398{
966bc6f4 399 unsigned short uninitialized_var(vlan_id);
1da177e4
LT		 400
401 if (!list_empty(&bond->vlan_list) &&
402 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
403 vlan_get_tag(skb, &vlan_id) == 0) {
404 skb->dev = slave_dev;
405 skb = vlan_put_tag(skb, vlan_id);
406 if (!skb) {
407 /* vlan_put_tag() frees the skb in case of error,
408 * so return success here so the calling functions
409 * won't attempt to free is again.
410 */
411 return 0;
412 }
413 } else {
414 skb->dev = slave_dev;
415 }
416
417 skb->priority = 1;
418 dev_queue_xmit(skb);
419
420 return 0;
421}
422
423/*
424 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
425 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
426 * lock because:
427 * a. This operation is performed in IOCTL context,
428 * b. The operation is protected by the RTNL semaphore in the 8021q code,
429 * c. Holding a lock with BH disabled while directly calling a base driver
430 * entry point is generally a BAD idea.
431 *
432 * The design of synchronization/protection for this operation in the 8021q
433 * module is good for one or more VLAN devices over a single physical device
434 * and cannot be extended for a teaming solution like bonding, so there is a
435 * potential race condition here where a net device from the vlan group might
436 * be referenced (either by a base driver or the 8021q code) while it is being
437 * removed from the system. However, it turns out we're not making matters
438 * worse, and if it works for regular VLAN usage it will work here too.
439*/
440
441/**
442 * bond_vlan_rx_register - Propagates registration to slaves
443 * @bond_dev: bonding net device that got called
444 * @grp: vlan group being registered
445 */
446static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
447{
454d7c9b 448 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 449 struct slave *slave;
450 int i;
451
452 bond->vlgrp = grp;
453
454 bond_for_each_slave(bond, slave, i) {
455 struct net_device *slave_dev = slave->dev;
eb7cc59a 456 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 457
458 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 459 slave_ops->ndo_vlan_rx_register) {
460 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
1da177e4
LT		 461 }
462 }
463}
464
465/**
466 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
467 * @bond_dev: bonding net device that got called
468 * @vid: vlan id being added
469 */
470static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
471{
454d7c9b 472 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 473 struct slave *slave;
474 int i, res;
475
476 bond_for_each_slave(bond, slave, i) {
477 struct net_device *slave_dev = slave->dev;
eb7cc59a 478 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 479
480 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a
SH		 481 slave_ops->ndo_vlan_rx_add_vid) {
482 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
1da177e4
LT		 483 }
484 }
485
486 res = bond_add_vlan(bond, vid);
487 if (res) {
488 printk(KERN_ERR DRV_NAME
4e0952c7 489 ": %s: Error: Failed to add vlan id %d\n",
1da177e4
LT		 490 bond_dev->name, vid);
491 }
492}
493
494/**
495 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
496 * @bond_dev: bonding net device that got called
497 * @vid: vlan id being removed
498 */
499static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
500{
454d7c9b 501 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 502 struct slave *slave;
503 struct net_device *vlan_dev;
504 int i, res;
505
506 bond_for_each_slave(bond, slave, i) {
507 struct net_device *slave_dev = slave->dev;
eb7cc59a 508 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 509
510 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a 511 slave_ops->ndo_vlan_rx_kill_vid) {
1da177e4
LT		 512 /* Save and then restore vlan_dev in the grp array,
513 * since the slave's driver might clear it.
514 */
5c15bdec 515 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
eb7cc59a 516 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
5c15bdec 517 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
1da177e4
LT		 518 }
519 }
520
521 res = bond_del_vlan(bond, vid);
522 if (res) {
523 printk(KERN_ERR DRV_NAME
4e0952c7 524 ": %s: Error: Failed to remove vlan id %d\n",
1da177e4
LT		 525 bond_dev->name, vid);
526 }
527}
528
529static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
530{
531 struct vlan_entry *vlan;
eb7cc59a 532 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 533
534 write_lock_bh(&bond->lock);
535
eb7cc59a 536 if (list_empty(&bond->vlan_list))
1da177e4 537 goto out;
1da177e4
LT		 538
539 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 540 slave_ops->ndo_vlan_rx_register)
541 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
1da177e4
LT		 542
543 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 544 !(slave_ops->ndo_vlan_rx_add_vid))
1da177e4 545 goto out;
1da177e4 546
eb7cc59a
SH		 547 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
548 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
1da177e4
LT		 549
550out:
551 write_unlock_bh(&bond->lock);
552}
553
554static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
555{
eb7cc59a 556 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 557 struct vlan_entry *vlan;
558 struct net_device *vlan_dev;
559
560 write_lock_bh(&bond->lock);
561
eb7cc59a 562 if (list_empty(&bond->vlan_list))
1da177e4 563 goto out;
1da177e4
LT		 564
565 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 566 !(slave_ops->ndo_vlan_rx_kill_vid))
1da177e4 567 goto unreg;
1da177e4
LT		 568
569 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
570 /* Save and then restore vlan_dev in the grp array,
571 * since the slave's driver might clear it.
572 */
5c15bdec 573 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
eb7cc59a 574 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
5c15bdec 575 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
1da177e4
LT		 576 }
577
578unreg:
579 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 580 slave_ops->ndo_vlan_rx_register)
581 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
1da177e4
LT		 582
583out:
584 write_unlock_bh(&bond->lock);
585}
586
587/*------------------------------- Link status -------------------------------*/
588
ff59c456
JV		 589/*
590 * Set the carrier state for the master according to the state of its
591 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
592 * do special 802.3ad magic.
593 *
594 * Returns zero if carrier state does not change, nonzero if it does.
595 */
596static int bond_set_carrier(struct bonding *bond)
597{
598 struct slave *slave;
599 int i;
600
601 if (bond->slave_cnt == 0)
602 goto down;
603
604 if (bond->params.mode == BOND_MODE_8023AD)
605 return bond_3ad_set_carrier(bond);
606
607 bond_for_each_slave(bond, slave, i) {
608 if (slave->link == BOND_LINK_UP) {
609 if (!netif_carrier_ok(bond->dev)) {
610 netif_carrier_on(bond->dev);
611 return 1;
612 }
613 return 0;
614 }
615 }
616
617down:
618 if (netif_carrier_ok(bond->dev)) {
619 netif_carrier_off(bond->dev);
620 return 1;
621 }
622 return 0;
623}
624
1da177e4
LT		 625/*
626 * Get link speed and duplex from the slave's base driver
627 * using ethtool. If for some reason the call fails or the
628 * values are invalid, fake speed and duplex to 100/Full
629 * and return error.
630 */
631static int bond_update_speed_duplex(struct slave *slave)
632{
633 struct net_device *slave_dev = slave->dev;
1da177e4 634 struct ethtool_cmd etool;
61a44b9c 635 int res;
1da177e4
LT		 636
637 /* Fake speed and duplex */
638 slave->speed = SPEED_100;
639 slave->duplex = DUPLEX_FULL;
640
61a44b9c
MW		 641 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
642 return -1;
1da177e4 643
61a44b9c
MW		 644 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
645 if (res < 0)
1da177e4 646 return -1;
1da177e4 647
1da177e4
LT		 648 switch (etool.speed) {
649 case SPEED_10:
650 case SPEED_100:
651 case SPEED_1000:
94dbffd5 652 case SPEED_10000:
1da177e4
LT		 653 break;
654 default:
655 return -1;
656 }
657
658 switch (etool.duplex) {
659 case DUPLEX_FULL:
660 case DUPLEX_HALF:
661 break;
662 default:
663 return -1;
664 }
665
666 slave->speed = etool.speed;
667 slave->duplex = etool.duplex;
668
669 return 0;
670}
671
672/*
673 * if <dev> supports MII link status reporting, check its link status.
674 *
675 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
676 * depening upon the setting of the use_carrier parameter.
677 *
678 * Return either BMSR_LSTATUS, meaning that the link is up (or we
679 * can't tell and just pretend it is), or 0, meaning that the link is
680 * down.
681 *
682 * If reporting is non-zero, instead of faking link up, return -1 if
683 * both ETHTOOL and MII ioctls fail (meaning the device does not
684 * support them). If use_carrier is set, return whatever it says.
685 * It'd be nice if there was a good way to tell if a driver supports
686 * netif_carrier, but there really isn't.
687 */
688static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
689{
eb7cc59a 690 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 691 static int (* ioctl)(struct net_device *, struct ifreq *, int);
692 struct ifreq ifr;
693 struct mii_ioctl_data *mii;
1da177e4 694
eb7cc59a 695 if (bond->params.use_carrier)
1da177e4 696 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1da177e4 697
eb7cc59a 698 ioctl = slave_ops->ndo_do_ioctl;
1da177e4
LT		 699 if (ioctl) {
700 /* TODO: set pointer to correct ioctl on a per team member */
701 /* bases to make this more efficient. that is, once */
702 /* we determine the correct ioctl, we will always */
703 /* call it and not the others for that team */
704 /* member. */
705
706 /*
707 * We cannot assume that SIOCGMIIPHY will also read a
708 * register; not all network drivers (e.g., e100)
709 * support that.
710 */
711
712 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
713 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
714 mii = if_mii(&ifr);
715 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
716 mii->reg_num = MII_BMSR;
717 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
718 return (mii->val_out & BMSR_LSTATUS);
719 }
720 }
721 }
722
61a44b9c
MW		 723 /*
724 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
725 * attempt to get link status from it if the above MII ioctls fail.
726 */
1da177e4
LT		 727 if (slave_dev->ethtool_ops) {
728 if (slave_dev->ethtool_ops->get_link) {
729 u32 link;
730
731 link = slave_dev->ethtool_ops->get_link(slave_dev);
732
733 return link ? BMSR_LSTATUS : 0;
734 }
735 }
736
1da177e4
LT		 737 /*
738 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 739 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 740 * cannot report link status). If not reporting, pretend
741 * we're ok.
742 */
743 return (reporting ? -1 : BMSR_LSTATUS);
744}
745
746/*----------------------------- Multicast list ------------------------------*/
747
748/*
749 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
750 */
751static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
752{
753 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
754 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
755}
756
757/*
758 * returns dmi entry if found, NULL otherwise
759 */
760static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
761{
762 struct dev_mc_list *idmi;
763
764 for (idmi = mc_list; idmi; idmi = idmi->next) {
765 if (bond_is_dmi_same(dmi, idmi)) {
766 return idmi;
767 }
768 }
769
770 return NULL;
771}
772
773/*
774 * Push the promiscuity flag down to appropriate slaves
775 */
7e1a1ac1 776static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 777{
7e1a1ac1 778 int err = 0;
1da177e4
LT		 779 if (USES_PRIMARY(bond->params.mode)) {
780 /* write lock already acquired */
781 if (bond->curr_active_slave) {
7e1a1ac1
WC		 782 err = dev_set_promiscuity(bond->curr_active_slave->dev,
783 inc);
1da177e4
LT		 784 }
785 } else {
786 struct slave *slave;
787 int i;
788 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 789 err = dev_set_promiscuity(slave->dev, inc);
790 if (err)
791 return err;
1da177e4
LT		 792 }
793 }
7e1a1ac1 794 return err;
1da177e4
LT		 795}
796
797/*
798 * Push the allmulti flag down to all slaves
799 */
7e1a1ac1 800static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 801{
7e1a1ac1 802 int err = 0;
1da177e4
LT		 803 if (USES_PRIMARY(bond->params.mode)) {
804 /* write lock already acquired */
805 if (bond->curr_active_slave) {
7e1a1ac1
WC		 806 err = dev_set_allmulti(bond->curr_active_slave->dev,
807 inc);
1da177e4
LT		 808 }
809 } else {
810 struct slave *slave;
811 int i;
812 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 813 err = dev_set_allmulti(slave->dev, inc);
814 if (err)
815 return err;
1da177e4
LT		 816 }
817 }
7e1a1ac1 818 return err;
1da177e4
LT		 819}
820
821/*
822 * Add a Multicast address to slaves
823 * according to mode
824 */
825static void bond_mc_add(struct bonding *bond, void *addr, int alen)
826{
827 if (USES_PRIMARY(bond->params.mode)) {
828 /* write lock already acquired */
829 if (bond->curr_active_slave) {
830 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
831 }
832 } else {
833 struct slave *slave;
834 int i;
835 bond_for_each_slave(bond, slave, i) {
836 dev_mc_add(slave->dev, addr, alen, 0);
837 }
838 }
839}
840
841/*
842 * Remove a multicast address from slave
843 * according to mode
844 */
845static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
846{
847 if (USES_PRIMARY(bond->params.mode)) {
848 /* write lock already acquired */
849 if (bond->curr_active_slave) {
850 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
851 }
852 } else {
853 struct slave *slave;
854 int i;
855 bond_for_each_slave(bond, slave, i) {
856 dev_mc_delete(slave->dev, addr, alen, 0);
857 }
858 }
859}
860
a816c7c7
JV		 861
862/*
863 * Retrieve the list of registered multicast addresses for the bonding
864 * device and retransmit an IGMP JOIN request to the current active
865 * slave.
866 */
867static void bond_resend_igmp_join_requests(struct bonding *bond)
868{
869 struct in_device *in_dev;
870 struct ip_mc_list *im;
871
872 rcu_read_lock();
873 in_dev = __in_dev_get_rcu(bond->dev);
874 if (in_dev) {
875 for (im = in_dev->mc_list; im; im = im->next) {
876 ip_mc_rejoin_group(im);
877 }
878 }
879
880 rcu_read_unlock();
881}
882
1da177e4
LT		 883/*
884 * Totally destroys the mc_list in bond
885 */
886static void bond_mc_list_destroy(struct bonding *bond)
887{
888 struct dev_mc_list *dmi;
889
890 dmi = bond->mc_list;
891 while (dmi) {
892 bond->mc_list = dmi->next;
893 kfree(dmi);
894 dmi = bond->mc_list;
895 }
a816c7c7 896 bond->mc_list = NULL;
1da177e4
LT		 897}
898
899/*
900 * Copy all the Multicast addresses from src to the bonding device dst
901 */
de54f390 902static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 903 gfp_t gfp_flag)
1da177e4
LT		 904{
905 struct dev_mc_list *dmi, *new_dmi;
906
907 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 908 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT		 909
910 if (!new_dmi) {
911 /* FIXME: Potential memory leak !!! */
912 return -ENOMEM;
913 }
914
915 new_dmi->next = bond->mc_list;
916 bond->mc_list = new_dmi;
917 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
918 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
919 new_dmi->dmi_users = dmi->dmi_users;
920 new_dmi->dmi_gusers = dmi->dmi_gusers;
921 }
922
923 return 0;
924}
925
926/*
927 * flush all members of flush->mc_list from device dev->mc_list
928 */
929static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
930{
454d7c9b 931 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 932 struct dev_mc_list *dmi;
933
934 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
935 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
936 }
937
938 if (bond->params.mode == BOND_MODE_8023AD) {
939 /* del lacpdu mc addr from mc list */
940 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
941
942 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
943 }
944}
945
946/*--------------------------- Active slave change ---------------------------*/
947
948/*
949 * Update the mc list and multicast-related flags for the new and
950 * old active slaves (if any) according to the multicast mode, and
951 * promiscuous flags unconditionally.
952 */
953static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
954{
955 struct dev_mc_list *dmi;
956
957 if (!USES_PRIMARY(bond->params.mode)) {
958 /* nothing to do - mc list is already up-to-date on
959 * all slaves
960 */
961 return;
962 }
963
964 if (old_active) {
965 if (bond->dev->flags & IFF_PROMISC) {
966 dev_set_promiscuity(old_active->dev, -1);
967 }
968
969 if (bond->dev->flags & IFF_ALLMULTI) {
970 dev_set_allmulti(old_active->dev, -1);
971 }
972
973 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
974 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
975 }
976 }
977
978 if (new_active) {
7e1a1ac1 979 /* FIXME: Signal errors upstream. */
1da177e4
LT		 980 if (bond->dev->flags & IFF_PROMISC) {
981 dev_set_promiscuity(new_active->dev, 1);
982 }
983
984 if (bond->dev->flags & IFF_ALLMULTI) {
985 dev_set_allmulti(new_active->dev, 1);
986 }
987
988 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
989 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
990 }
a816c7c7 991 bond_resend_igmp_join_requests(bond);
1da177e4
LT		 992 }
993}
994
3915c1e8
JV		 995/*
996 * bond_do_fail_over_mac
997 *
998 * Perform special MAC address swapping for fail_over_mac settings
999 *
1000 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1001 */
1002static void bond_do_fail_over_mac(struct bonding *bond,
1003 struct slave *new_active,
1004 struct slave *old_active)
1f78d9f9
HE		 1005 __releases(&bond->curr_slave_lock)
1006 __releases(&bond->lock)
1007 __acquires(&bond->lock)
1008 __acquires(&bond->curr_slave_lock)
3915c1e8
JV		 1009{
1010 u8 tmp_mac[ETH_ALEN];
1011 struct sockaddr saddr;
1012 int rv;
1013
1014 switch (bond->params.fail_over_mac) {
1015 case BOND_FOM_ACTIVE:
1016 if (new_active)
1017 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1018 new_active->dev->addr_len);
1019 break;
1020 case BOND_FOM_FOLLOW:
1021 /*
1022 * if new_active && old_active, swap them
1023 * if just old_active, do nothing (going to no active slave)
1024 * if just new_active, set new_active to bond's MAC
1025 */
1026 if (!new_active)
1027 return;
1028
1029 write_unlock_bh(&bond->curr_slave_lock);
1030 read_unlock(&bond->lock);
1031
1032 if (old_active) {
1033 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1034 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1035 ETH_ALEN);
1036 saddr.sa_family = new_active->dev->type;
1037 } else {
1038 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1039 saddr.sa_family = bond->dev->type;
1040 }
1041
1042 rv = dev_set_mac_address(new_active->dev, &saddr);
1043 if (rv) {
1044 printk(KERN_ERR DRV_NAME
1045 ": %s: Error %d setting MAC of slave %s\n",
1046 bond->dev->name, -rv, new_active->dev->name);
1047 goto out;
1048 }
1049
1050 if (!old_active)
1051 goto out;
1052
1053 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1054 saddr.sa_family = old_active->dev->type;
1055
1056 rv = dev_set_mac_address(old_active->dev, &saddr);
1057 if (rv)
1058 printk(KERN_ERR DRV_NAME
1059 ": %s: Error %d setting MAC of slave %s\n",
1060 bond->dev->name, -rv, new_active->dev->name);
1061out:
1062 read_lock(&bond->lock);
1063 write_lock_bh(&bond->curr_slave_lock);
1064 break;
1065 default:
1066 printk(KERN_ERR DRV_NAME
1067 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1068 bond->dev->name, bond->params.fail_over_mac);
1069 break;
1070 }
1071
1072}
1073
1074
1da177e4
LT		 1075/**
1076 * find_best_interface - select the best available slave to be the active one
1077 * @bond: our bonding struct
1078 *
1079 * Warning: Caller must hold curr_slave_lock for writing.
1080 */
1081static struct slave *bond_find_best_slave(struct bonding *bond)
1082{
1083 struct slave *new_active, *old_active;
1084 struct slave *bestslave = NULL;
1085 int mintime = bond->params.updelay;
1086 int i;
1087
1088 new_active = old_active = bond->curr_active_slave;
1089
1090 if (!new_active) { /* there were no active slaves left */
1091 if (bond->slave_cnt > 0) { /* found one slave */
1092 new_active = bond->first_slave;
1093 } else {
1094 return NULL; /* still no slave, return NULL */
1095 }
1096 }
1097
1098 /* first try the primary link; if arping, a link must tx/rx traffic
1099 * before it can be considered the curr_active_slave - also, we would skip
1100 * slaves between the curr_active_slave and primary_slave that may be up
1101 * and able to arp
1102 */
1103 if ((bond->primary_slave) &&
1104 (!bond->params.arp_interval) &&
1105 (IS_UP(bond->primary_slave->dev))) {
1106 new_active = bond->primary_slave;
1107 }
1108
1109 /* remember where to stop iterating over the slaves */
1110 old_active = new_active;
1111
1112 bond_for_each_slave_from(bond, new_active, i, old_active) {
1113 if (IS_UP(new_active->dev)) {
1114 if (new_active->link == BOND_LINK_UP) {
1115 return new_active;
1116 } else if (new_active->link == BOND_LINK_BACK) {
1117 /* link up, but waiting for stabilization */
1118 if (new_active->delay < mintime) {
1119 mintime = new_active->delay;
1120 bestslave = new_active;
1121 }
1122 }
1123 }
1124 }
1125
1126 return bestslave;
1127}
1128
1129/**
1130 * change_active_interface - change the active slave into the specified one
1131 * @bond: our bonding struct
1132 * @new: the new slave to make the active one
1133 *
1134 * Set the new slave to the bond's settings and unset them on the old
1135 * curr_active_slave.
1136 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1137 *
1138 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1139 * because it is apparently the best available slave we have, even though its
1140 * updelay hasn't timed out yet.
1141 *
3915c1e8
JV		 1142 * If new_active is not NULL, caller must hold bond->lock for read and
1143 * curr_slave_lock for write_bh.
1da177e4 1144 */
a77b5325 1145void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1146{
1147 struct slave *old_active = bond->curr_active_slave;
1148
1149 if (old_active == new_active) {
1150 return;
1151 }
1152
1153 if (new_active) {
b2220cad
JV		 1154 new_active->jiffies = jiffies;
1155
1da177e4
LT		 1156 if (new_active->link == BOND_LINK_BACK) {
1157 if (USES_PRIMARY(bond->params.mode)) {
1158 printk(KERN_INFO DRV_NAME
1159 ": %s: making interface %s the new "
1160 "active one %d ms earlier.\n",
1161 bond->dev->name, new_active->dev->name,
1162 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1163 }
1164
1165 new_active->delay = 0;
1166 new_active->link = BOND_LINK_UP;
1da177e4
LT		 1167
1168 if (bond->params.mode == BOND_MODE_8023AD) {
1169 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1170 }
1171
58402054 1172 if (bond_is_lb(bond))
1da177e4 1173 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1da177e4
LT		 1174 } else {
1175 if (USES_PRIMARY(bond->params.mode)) {
1176 printk(KERN_INFO DRV_NAME
1177 ": %s: making interface %s the new "
1178 "active one.\n",
1179 bond->dev->name, new_active->dev->name);
1180 }
1181 }
1182 }
1183
1da177e4
LT		 1184 if (USES_PRIMARY(bond->params.mode)) {
1185 bond_mc_swap(bond, new_active, old_active);
1186 }
1187
58402054 1188 if (bond_is_lb(bond)) {
1da177e4 1189 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1190 if (old_active)
1191 bond_set_slave_inactive_flags(old_active);
1192 if (new_active)
1193 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1194 } else {
1195 bond->curr_active_slave = new_active;
1196 }
c3ade5ca
JV		 1197
1198 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1199 if (old_active) {
1200 bond_set_slave_inactive_flags(old_active);
1201 }
1202
1203 if (new_active) {
1204 bond_set_slave_active_flags(new_active);
2ab82852 1205
709f8a45
OG		 1206 if (bond->params.fail_over_mac)
1207 bond_do_fail_over_mac(bond, new_active,
1208 old_active);
3915c1e8 1209
709f8a45 1210 bond->send_grat_arp = bond->params.num_grat_arp;
b59f9f74 1211 bond_send_gratuitous_arp(bond);
01f3109d 1212
305d552a
BH		 1213 bond->send_unsol_na = bond->params.num_unsol_na;
1214 bond_send_unsolicited_na(bond);
1215
01f3109d
OG		 1216 write_unlock_bh(&bond->curr_slave_lock);
1217 read_unlock(&bond->lock);
1218
1219 netdev_bonding_change(bond->dev);
1220
1221 read_lock(&bond->lock);
1222 write_lock_bh(&bond->curr_slave_lock);
7893b249 1223 }
c3ade5ca 1224 }
1da177e4
LT		 1225}
1226
1227/**
1228 * bond_select_active_slave - select a new active slave, if needed
1229 * @bond: our bonding struct
1230 *
1231 * This functions shoud be called when one of the following occurs:
1232 * - The old curr_active_slave has been released or lost its link.
1233 * - The primary_slave has got its link back.
1234 * - A slave has got its link back and there's no old curr_active_slave.
1235 *
3915c1e8 1236 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1237 */
a77b5325 1238void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1239{
1240 struct slave *best_slave;
ff59c456 1241 int rv;
1da177e4
LT		 1242
1243 best_slave = bond_find_best_slave(bond);
1244 if (best_slave != bond->curr_active_slave) {
1245 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1246 rv = bond_set_carrier(bond);
1247 if (!rv)
1248 return;
1249
1250 if (netif_carrier_ok(bond->dev)) {
1251 printk(KERN_INFO DRV_NAME
1252 ": %s: first active interface up!\n",
1253 bond->dev->name);
1254 } else {
1255 printk(KERN_INFO DRV_NAME ": %s: "
1256 "now running without any active interface !\n",
1257 bond->dev->name);
1258 }
1da177e4
LT		 1259 }
1260}
1261
1262/*--------------------------- slave list handling ---------------------------*/
1263
1264/*
1265 * This function attaches the slave to the end of list.
1266 *
1267 * bond->lock held for writing by caller.
1268 */
1269static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1270{
1271 if (bond->first_slave == NULL) { /* attaching the first slave */
1272 new_slave->next = new_slave;
1273 new_slave->prev = new_slave;
1274 bond->first_slave = new_slave;
1275 } else {
1276 new_slave->next = bond->first_slave;
1277 new_slave->prev = bond->first_slave->prev;
1278 new_slave->next->prev = new_slave;
1279 new_slave->prev->next = new_slave;
1280 }
1281
1282 bond->slave_cnt++;
1283}
1284
1285/*
1286 * This function detaches the slave from the list.
1287 * WARNING: no check is made to verify if the slave effectively
1288 * belongs to <bond>.
1289 * Nothing is freed on return, structures are just unchained.
1290 * If any slave pointer in bond was pointing to <slave>,
1291 * it should be changed by the calling function.
1292 *
1293 * bond->lock held for writing by caller.
1294 */
1295static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1296{
1297 if (slave->next) {
1298 slave->next->prev = slave->prev;
1299 }
1300
1301 if (slave->prev) {
1302 slave->prev->next = slave->next;
1303 }
1304
1305 if (bond->first_slave == slave) { /* slave is the first slave */
1306 if (bond->slave_cnt > 1) { /* there are more slave */
1307 bond->first_slave = slave->next;
1308 } else {
1309 bond->first_slave = NULL; /* slave was the last one */
1310 }
1311 }
1312
1313 slave->next = NULL;
1314 slave->prev = NULL;
1315 bond->slave_cnt--;
1316}
1317
1318/*---------------------------------- IOCTL ----------------------------------*/
1319
4ad072c9
AB		 1320static int bond_sethwaddr(struct net_device *bond_dev,
1321 struct net_device *slave_dev)
1da177e4 1322{
5a03cdb7
HE		 1323 pr_debug("bond_dev=%p\n", bond_dev);
1324 pr_debug("slave_dev=%p\n", slave_dev);
1325 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1da177e4
LT		 1326 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1327 return 0;
1328}
1329
7f353bf2
HX		 1330#define BOND_VLAN_FEATURES \
1331 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1332 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK		 1333
1334/*
8e3babcd 1335 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX		 1336 * feature bits are managed elsewhere, so preserve those feature bits
1337 * on the master device.
8531c5ff
AK		 1338 */
1339static int bond_compute_features(struct bonding *bond)
1340{
8531c5ff
AK		 1341 struct slave *slave;
1342 struct net_device *bond_dev = bond->dev;
7f353bf2 1343 unsigned long features = bond_dev->features;
3158bf7d
MS		 1344 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1345 bond_dev->hard_header_len);
8e3babcd 1346 int i;
8531c5ff 1347
7f353bf2 1348 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
b63365a2
HX		 1349 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1350
1351 if (!bond->first_slave)
1352 goto done;
1353
1354 features &= ~NETIF_F_ONE_FOR_ALL;
7f353bf2 1355
54ef3137 1356 bond_for_each_slave(bond, slave, i) {
b63365a2
HX		 1357 features = netdev_increment_features(features,
1358 slave->dev->features,
1359 NETIF_F_ONE_FOR_ALL);
54ef3137
JV		 1360 if (slave->dev->hard_header_len > max_hard_header_len)
1361 max_hard_header_len = slave->dev->hard_header_len;
1362 }
8531c5ff 1363
b63365a2 1364done:
7f353bf2 1365 features |= (bond_dev->features & BOND_VLAN_FEATURES);
b63365a2 1366 bond_dev->features = netdev_fix_features(features, NULL);
54ef3137 1367 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1368
1369 return 0;
1370}
1371
872254dd
MS		 1372static void bond_setup_by_slave(struct net_device *bond_dev,
1373 struct net_device *slave_dev)
1374{
454d7c9b 1375 struct bonding *bond = netdev_priv(bond_dev);
d90a162a 1376
00829823 1377 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1378
1379 bond_dev->type = slave_dev->type;
1380 bond_dev->hard_header_len = slave_dev->hard_header_len;
1381 bond_dev->addr_len = slave_dev->addr_len;
1382
1383 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1384 slave_dev->addr_len);
d90a162a 1385 bond->setup_by_slave = 1;
872254dd
MS		 1386}
1387
1da177e4 1388/* enslave device <slave> to bond device <master> */
a77b5325 1389int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1390{
454d7c9b 1391 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1392 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 1393 struct slave *new_slave = NULL;
1394 struct dev_mc_list *dmi;
1395 struct sockaddr addr;
1396 int link_reporting;
1397 int old_features = bond_dev->features;
1398 int res = 0;
1399
552709d5 1400 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
eb7cc59a 1401 slave_ops->ndo_do_ioctl == NULL) {
1da177e4 1402 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1403 ": %s: Warning: no link monitoring support for %s\n",
1404 bond_dev->name, slave_dev->name);
1da177e4
LT		 1405 }
1406
1407 /* bond must be initialized by bond_open() before enslaving */
1408 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1409 printk(KERN_WARNING DRV_NAME
1410 " %s: master_dev is not up in bond_enslave\n",
1411 bond_dev->name);
1da177e4
LT		 1412 }
1413
1414 /* already enslaved */
1415 if (slave_dev->flags & IFF_SLAVE) {
5a03cdb7 1416 pr_debug("Error, Device was already enslaved\n");
1da177e4
LT		 1417 return -EBUSY;
1418 }
1419
1420 /* vlan challenged mutual exclusion */
1421 /* no need to lock since we're protected by rtnl_lock */
1422 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
5a03cdb7 1423 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1da177e4
LT		 1424 if (!list_empty(&bond->vlan_list)) {
1425 printk(KERN_ERR DRV_NAME
4e0952c7 1426 ": %s: Error: cannot enslave VLAN "
1da177e4 1427 "challenged slave %s on VLAN enabled "
4e0952c7 1428 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1429 bond_dev->name);
1430 return -EPERM;
1431 } else {
1432 printk(KERN_WARNING DRV_NAME
4e0952c7 1433 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1434 "slave %s. Adding VLANs will be blocked as "
1435 "long as %s is part of bond %s\n",
4e0952c7 1436 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1437 bond_dev->name);
1438 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1439 }
1440 } else {
5a03cdb7 1441 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1da177e4
LT		 1442 if (bond->slave_cnt == 0) {
1443 /* First slave, and it is not VLAN challenged,
1444 * so remove the block of adding VLANs over the bond.
1445 */
1446 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1447 }
1448 }
1449
217df670
JV		 1450 /*
1451 * Old ifenslave binaries are no longer supported. These can
1452 * be identified with moderate accurary by the state of the slave:
1453 * the current ifenslave will set the interface down prior to
1454 * enslaving it; the old ifenslave will not.
1455 */
1456 if ((slave_dev->flags & IFF_UP)) {
1457 printk(KERN_ERR DRV_NAME ": %s is up. "
1458 "This may be due to an out of date ifenslave.\n",
1459 slave_dev->name);
1460 res = -EPERM;
1461 goto err_undo_flags;
1462 }
1da177e4 1463
872254dd
MS		 1464 /* set bonding device ether type by slave - bonding netdevices are
1465 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1466 * there is a need to override some of the type dependent attribs/funcs.
1467 *
1468 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1469 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1470 */
1471 if (bond->slave_cnt == 0) {
1472 if (slave_dev->type != ARPHRD_ETHER)
1473 bond_setup_by_slave(bond_dev, slave_dev);
1474 } else if (bond_dev->type != slave_dev->type) {
1475 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1476 "from other slaves (%d), can not enslave it.\n",
1477 slave_dev->name,
1478 slave_dev->type, bond_dev->type);
1479 res = -EINVAL;
1480 goto err_undo_flags;
1481 }
1482
eb7cc59a 1483 if (slave_ops->ndo_set_mac_address == NULL) {
2ab82852
MS		 1484 if (bond->slave_cnt == 0) {
1485 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1486 ": %s: Warning: The first slave device "
1487 "specified does not support setting the MAC "
3915c1e8 1488 "address. Setting fail_over_mac to active.",
dd957c57 1489 bond_dev->name);
3915c1e8
JV		 1490 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1491 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1492 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1493 ": %s: Error: The slave device specified "
1494 "does not support setting the MAC address, "
3915c1e8 1495 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1496 , bond_dev->name);
1497 res = -EOPNOTSUPP;
1498 goto err_undo_flags;
1499 }
1da177e4
LT		 1500 }
1501
243cb4e5 1502 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1503 if (!new_slave) {
1504 res = -ENOMEM;
1505 goto err_undo_flags;
1506 }
1507
1da177e4
LT		 1508 /* save slave's original flags before calling
1509 * netdev_set_master and dev_open
1510 */
1511 new_slave->original_flags = slave_dev->flags;
1512
217df670
JV		 1513 /*
1514 * Save slave's original ("permanent") mac address for modes
1515 * that need it, and for restoring it upon release, and then
1516 * set it to the master's address
1517 */
1518 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1519
dd957c57 1520 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1521 /*
1522 * Set slave to master's mac address. The application already
1523 * set the master's mac address to that of the first slave
1524 */
1525 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1526 addr.sa_family = slave_dev->type;
1527 res = dev_set_mac_address(slave_dev, &addr);
1528 if (res) {
5a03cdb7 1529 pr_debug("Error %d calling set_mac_address\n", res);
2ab82852
MS		 1530 goto err_free;
1531 }
217df670 1532 }
1da177e4 1533
c2edacf8
JV		 1534 res = netdev_set_master(slave_dev, bond_dev);
1535 if (res) {
5a03cdb7 1536 pr_debug("Error %d calling netdev_set_master\n", res);
569f0c4d 1537 goto err_restore_mac;
c2edacf8 1538 }
217df670
JV		 1539 /* open the slave since the application closed it */
1540 res = dev_open(slave_dev);
1541 if (res) {
5a03cdb7 1542 pr_debug("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1543 goto err_unset_master;
1da177e4
LT		 1544 }
1545
1da177e4 1546 new_slave->dev = slave_dev;
0b680e75 1547 slave_dev->priv_flags |= IFF_BONDING;
1da177e4 1548
58402054 1549 if (bond_is_lb(bond)) {
1da177e4
LT		 1550 /* bond_alb_init_slave() must be called before all other stages since
1551 * it might fail and we do not want to have to undo everything
1552 */
1553 res = bond_alb_init_slave(bond, new_slave);
1554 if (res) {
569f0c4d 1555 goto err_close;
1da177e4
LT		 1556 }
1557 }
1558
1559 /* If the mode USES_PRIMARY, then the new slave gets the
1560 * master's promisc (and mc) settings only if it becomes the
1561 * curr_active_slave, and that is taken care of later when calling
1562 * bond_change_active()
1563 */
1564 if (!USES_PRIMARY(bond->params.mode)) {
1565 /* set promiscuity level to new slave */
1566 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1567 res = dev_set_promiscuity(slave_dev, 1);
1568 if (res)
1569 goto err_close;
1da177e4
LT		 1570 }
1571
1572 /* set allmulti level to new slave */
1573 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1574 res = dev_set_allmulti(slave_dev, 1);
1575 if (res)
1576 goto err_close;
1da177e4
LT		 1577 }
1578
b9e40857 1579 netif_addr_lock_bh(bond_dev);
1da177e4
LT		 1580 /* upload master's mc_list to new slave */
1581 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1582 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1583 }
b9e40857 1584 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1585 }
1586
1587 if (bond->params.mode == BOND_MODE_8023AD) {
1588 /* add lacpdu mc addr to mc list */
1589 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1590
1591 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1592 }
1593
1594 bond_add_vlans_on_slave(bond, slave_dev);
1595
1596 write_lock_bh(&bond->lock);
1597
1598 bond_attach_slave(bond, new_slave);
1599
1600 new_slave->delay = 0;
1601 new_slave->link_failure_count = 0;
1602
8531c5ff
AK		 1603 bond_compute_features(bond);
1604
3915c1e8
JV		 1605 write_unlock_bh(&bond->lock);
1606
1607 read_lock(&bond->lock);
1608
f5b2b966
JV		 1609 new_slave->last_arp_rx = jiffies;
1610
1da177e4
LT		 1611 if (bond->params.miimon && !bond->params.use_carrier) {
1612 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1613
1614 if ((link_reporting == -1) && !bond->params.arp_interval) {
1615 /*
1616 * miimon is set but a bonded network driver
1617 * does not support ETHTOOL/MII and
1618 * arp_interval is not set. Note: if
1619 * use_carrier is enabled, we will never go
1620 * here (because netif_carrier is always
1621 * supported); thus, we don't need to change
1622 * the messages for netif_carrier.
1623 */
1624 printk(KERN_WARNING DRV_NAME
4e0952c7 1625 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1626 "available for interface %s, and "
1627 "arp_interval/arp_ip_target module parameters "
1628 "not specified, thus bonding will not detect "
1629 "link failures! see bonding.txt for details.\n",
4e0952c7 1630 bond_dev->name, slave_dev->name);
1da177e4
LT		 1631 } else if (link_reporting == -1) {
1632 /* unable get link status using mii/ethtool */
1633 printk(KERN_WARNING DRV_NAME
4e0952c7 1634 ": %s: Warning: can't get link status from "
1da177e4
LT		 1635 "interface %s; the network driver associated "
1636 "with this interface does not support MII or "
1637 "ETHTOOL link status reporting, thus miimon "
1638 "has no effect on this interface.\n",
4e0952c7 1639 bond_dev->name, slave_dev->name);
1da177e4
LT		 1640 }
1641 }
1642
1643 /* check for initial state */
1644 if (!bond->params.miimon ||
1645 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1646 if (bond->params.updelay) {
5a03cdb7 1647 pr_debug("Initial state of slave_dev is "
1da177e4
LT		 1648 "BOND_LINK_BACK\n");
1649 new_slave->link = BOND_LINK_BACK;
1650 new_slave->delay = bond->params.updelay;
1651 } else {
5a03cdb7 1652 pr_debug("Initial state of slave_dev is "
1da177e4
LT		 1653 "BOND_LINK_UP\n");
1654 new_slave->link = BOND_LINK_UP;
1655 }
1656 new_slave->jiffies = jiffies;
1657 } else {
5a03cdb7 1658 pr_debug("Initial state of slave_dev is "
1da177e4
LT		 1659 "BOND_LINK_DOWN\n");
1660 new_slave->link = BOND_LINK_DOWN;
1661 }
1662
1663 if (bond_update_speed_duplex(new_slave) &&
1664 (new_slave->link != BOND_LINK_DOWN)) {
1665 printk(KERN_WARNING DRV_NAME
4e0952c7 1666 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1667 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1668 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1669
1670 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1671 printk(KERN_WARNING DRV_NAME
1672 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1673 "support in base driver for proper aggregator "
4e0952c7 1674 "selection.\n", bond_dev->name);
1da177e4
LT		 1675 }
1676 }
1677
1678 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1679 /* if there is a primary slave, remember it */
1680 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1681 bond->primary_slave = new_slave;
1682 }
1683 }
1684
3915c1e8
JV		 1685 write_lock_bh(&bond->curr_slave_lock);
1686
1da177e4
LT		 1687 switch (bond->params.mode) {
1688 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1689 bond_set_slave_inactive_flags(new_slave);
1690 bond_select_active_slave(bond);
1da177e4
LT		 1691 break;
1692 case BOND_MODE_8023AD:
1693 /* in 802.3ad mode, the internal mechanism
1694 * will activate the slaves in the selected
1695 * aggregator
1696 */
1697 bond_set_slave_inactive_flags(new_slave);
1698 /* if this is the first slave */
1699 if (bond->slave_cnt == 1) {
1700 SLAVE_AD_INFO(new_slave).id = 1;
1701 /* Initialize AD with the number of times that the AD timer is called in 1 second
1702 * can be called only after the mac address of the bond is set
1703 */
1704 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1705 bond->params.lacp_fast);
1706 } else {
1707 SLAVE_AD_INFO(new_slave).id =
1708 SLAVE_AD_INFO(new_slave->prev).id + 1;
1709 }
1710
1711 bond_3ad_bind_slave(new_slave);
1712 break;
1713 case BOND_MODE_TLB:
1714 case BOND_MODE_ALB:
1715 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1716 bond_set_slave_inactive_flags(new_slave);
1da177e4
LT		 1717 break;
1718 default:
5a03cdb7 1719 pr_debug("This slave is always active in trunk mode\n");
1da177e4
LT		 1720
1721 /* always active in trunk mode */
1722 new_slave->state = BOND_STATE_ACTIVE;
1723
1724 /* In trunking mode there is little meaning to curr_active_slave
1725 * anyway (it holds no special properties of the bond device),
1726 * so we can change it without calling change_active_interface()
1727 */
1728 if (!bond->curr_active_slave) {
1729 bond->curr_active_slave = new_slave;
1730 }
1731 break;
1732 } /* switch(bond_mode) */
1733
3915c1e8
JV		 1734 write_unlock_bh(&bond->curr_slave_lock);
1735
ff59c456
JV		 1736 bond_set_carrier(bond);
1737
3915c1e8 1738 read_unlock(&bond->lock);
1da177e4 1739
b76cdba9
MW		 1740 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1741 if (res)
569f0c4d 1742 goto err_close;
b76cdba9 1743
1da177e4
LT		 1744 printk(KERN_INFO DRV_NAME
1745 ": %s: enslaving %s as a%s interface with a%s link.\n",
1746 bond_dev->name, slave_dev->name,
1747 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1748 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1749
1750 /* enslave is successful */
1751 return 0;
1752
1753/* Undo stages on error */
1da177e4
LT		 1754err_close:
1755 dev_close(slave_dev);
1756
569f0c4d
JV		 1757err_unset_master:
1758 netdev_set_master(slave_dev, NULL);
1759
1da177e4 1760err_restore_mac:
dd957c57 1761 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1762 /* XXX TODO - fom follow mode needs to change master's
1763 * MAC if this slave's MAC is in use by the bond, or at
1764 * least print a warning.
1765 */
2ab82852
MS		 1766 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1767 addr.sa_family = slave_dev->type;
1768 dev_set_mac_address(slave_dev, &addr);
1769 }
1da177e4
LT		 1770
1771err_free:
1772 kfree(new_slave);
1773
1774err_undo_flags:
1775 bond_dev->features = old_features;
8531c5ff 1776
1da177e4
LT		 1777 return res;
1778}
1779
1780/*
1781 * Try to release the slave device <slave> from the bond device <master>
1782 * It is legal to access curr_active_slave without a lock because all the function
1783 * is write-locked.
1784 *
1785 * The rules for slave state should be:
1786 * for Active/Backup:
1787 * Active stays on all backups go down
1788 * for Bonded connections:
1789 * The first up interface should be left on and all others downed.
1790 */
a77b5325 1791int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1792{
454d7c9b 1793 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1794 struct slave *slave, *oldcurrent;
1795 struct sockaddr addr;
1796 int mac_addr_differ;
1797
1798 /* slave is not a slave or master is not master of this slave */
1799 if (!(slave_dev->flags & IFF_SLAVE) ||
1800 (slave_dev->master != bond_dev)) {
1801 printk(KERN_ERR DRV_NAME
4e0952c7 1802 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1803 bond_dev->name, slave_dev->name);
1804 return -EINVAL;
1805 }
1806
1807 write_lock_bh(&bond->lock);
1808
1809 slave = bond_get_slave_by_dev(bond, slave_dev);
1810 if (!slave) {
1811 /* not a slave of this bond */
1812 printk(KERN_INFO DRV_NAME
1813 ": %s: %s not enslaved\n",
1814 bond_dev->name, slave_dev->name);
f5e2a7b2 1815 write_unlock_bh(&bond->lock);
1da177e4
LT		 1816 return -EINVAL;
1817 }
1818
3915c1e8
JV		 1819 if (!bond->params.fail_over_mac) {
1820 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1821 ETH_ALEN);
1822 if (!mac_addr_differ && (bond->slave_cnt > 1))
1823 printk(KERN_WARNING DRV_NAME
1824 ": %s: Warning: the permanent HWaddr of %s - "
e174961c 1825 "%pM - is still in use by %s. "
3915c1e8
JV		 1826 "Set the HWaddr of %s to a different address "
1827 "to avoid conflicts.\n",
1828 bond_dev->name, slave_dev->name,
e174961c 1829 slave->perm_hwaddr,
3915c1e8 1830 bond_dev->name, slave_dev->name);
1da177e4
LT		 1831 }
1832
1833 /* Inform AD package of unbinding of slave. */
1834 if (bond->params.mode == BOND_MODE_8023AD) {
1835 /* must be called before the slave is
1836 * detached from the list
1837 */
1838 bond_3ad_unbind_slave(slave);
1839 }
1840
1841 printk(KERN_INFO DRV_NAME
1842 ": %s: releasing %s interface %s\n",
1843 bond_dev->name,
1844 (slave->state == BOND_STATE_ACTIVE)
1845 ? "active" : "backup",
1846 slave_dev->name);
1847
1848 oldcurrent = bond->curr_active_slave;
1849
1850 bond->current_arp_slave = NULL;
1851
1852 /* release the slave from its bond */
1853 bond_detach_slave(bond, slave);
1854
8531c5ff
AK		 1855 bond_compute_features(bond);
1856
1da177e4
LT		 1857 if (bond->primary_slave == slave) {
1858 bond->primary_slave = NULL;
1859 }
1860
1861 if (oldcurrent == slave) {
1862 bond_change_active_slave(bond, NULL);
1863 }
1864
58402054 1865 if (bond_is_lb(bond)) {
1da177e4
LT		 1866 /* Must be called only after the slave has been
1867 * detached from the list and the curr_active_slave
1868 * has been cleared (if our_slave == old_current),
1869 * but before a new active slave is selected.
1870 */
2543331d 1871 write_unlock_bh(&bond->lock);
1da177e4 1872 bond_alb_deinit_slave(bond, slave);
2543331d 1873 write_lock_bh(&bond->lock);
1da177e4
LT		 1874 }
1875
059fe7a5
JV		 1876 if (oldcurrent == slave) {
1877 /*
1878 * Note that we hold RTNL over this sequence, so there
1879 * is no concern that another slave add/remove event
1880 * will interfere.
1881 */
1882 write_unlock_bh(&bond->lock);
1883 read_lock(&bond->lock);
1884 write_lock_bh(&bond->curr_slave_lock);
1885
1da177e4
LT		 1886 bond_select_active_slave(bond);
1887
059fe7a5
JV		 1888 write_unlock_bh(&bond->curr_slave_lock);
1889 read_unlock(&bond->lock);
1890 write_lock_bh(&bond->lock);
1891 }
1892
1da177e4 1893 if (bond->slave_cnt == 0) {
ff59c456
JV		 1894 bond_set_carrier(bond);
1895
1da177e4
LT		 1896 /* if the last slave was removed, zero the mac address
1897 * of the master so it will be set by the application
1898 * to the mac address of the first slave
1899 */
1900 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1901
1902 if (list_empty(&bond->vlan_list)) {
1903 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1904 } else {
1905 printk(KERN_WARNING DRV_NAME
4e0952c7 1906 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1907 "still has VLANs.\n",
4e0952c7 1908 bond_dev->name, bond_dev->name);
1da177e4 1909 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1910 ": %s: When re-adding slaves, make sure the bond's "
1911 "HW address matches its VLANs'.\n",
1912 bond_dev->name);
1da177e4
LT		 1913 }
1914 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1915 !bond_has_challenged_slaves(bond)) {
1916 printk(KERN_INFO DRV_NAME
4e0952c7 1917 ": %s: last VLAN challenged slave %s "
1da177e4 1918 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1919 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1920 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1921 }
1922
1923 write_unlock_bh(&bond->lock);
1924
b76cdba9
MW		 1925 /* must do this from outside any spinlocks */
1926 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1927
1da177e4
LT		 1928 bond_del_vlans_from_slave(bond, slave_dev);
1929
1930 /* If the mode USES_PRIMARY, then we should only remove its
1931 * promisc and mc settings if it was the curr_active_slave, but that was
1932 * already taken care of above when we detached the slave
1933 */
1934 if (!USES_PRIMARY(bond->params.mode)) {
1935 /* unset promiscuity level from slave */
1936 if (bond_dev->flags & IFF_PROMISC) {
1937 dev_set_promiscuity(slave_dev, -1);
1938 }
1939
1940 /* unset allmulti level from slave */
1941 if (bond_dev->flags & IFF_ALLMULTI) {
1942 dev_set_allmulti(slave_dev, -1);
1943 }
1944
1945 /* flush master's mc_list from slave */
b9e40857 1946 netif_addr_lock_bh(bond_dev);
1da177e4 1947 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 1948 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1949 }
1950
1951 netdev_set_master(slave_dev, NULL);
1952
1953 /* close slave before restoring its mac address */
1954 dev_close(slave_dev);
1955
3915c1e8 1956 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1957 /* restore original ("permanent") mac address */
1958 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1959 addr.sa_family = slave_dev->type;
1960 dev_set_mac_address(slave_dev, &addr);
1961 }
1da177e4 1962
8f903c70 1963 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1964 IFF_SLAVE_INACTIVE | IFF_BONDING |
1965 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1966
1967 kfree(slave);
1968
1969 return 0; /* deletion OK */
1970}
1971
d90a162a
MS		 1972/*
1973* Destroy a bonding device.
1974* Must be under rtnl_lock when this function is called.
1975*/
1976void bond_destroy(struct bonding *bond)
1977{
1978 bond_deinit(bond->dev);
1979 bond_destroy_sysfs_entry(bond);
8cbdeec6 1980 unregister_netdevice(bond->dev);
d90a162a
MS		 1981}
1982
a434e43f
JV		 1983static void bond_destructor(struct net_device *bond_dev)
1984{
244ef9b9 1985 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 1986
1987 if (bond->wq)
1988 destroy_workqueue(bond->wq);
1989
1990 netif_addr_lock_bh(bond_dev);
1991 bond_mc_list_destroy(bond);
1992 netif_addr_unlock_bh(bond_dev);
1993
1994 free_netdev(bond_dev);
1995}
1996
d90a162a
MS		 1997/*
1998* First release a slave and than destroy the bond if no more slaves iare left.
1999* Must be under rtnl_lock when this function is called.
2000*/
2001int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2002{
454d7c9b 2003 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 2004 int ret;
2005
2006 ret = bond_release(bond_dev, slave_dev);
2007 if ((ret == 0) && (bond->slave_cnt == 0)) {
2008 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2009 bond_dev->name, bond_dev->name);
2010 bond_destroy(bond);
2011 }
2012 return ret;
2013}
2014
1da177e4
LT		 2015/*
2016 * This function releases all slaves.
2017 */
2018static int bond_release_all(struct net_device *bond_dev)
2019{
454d7c9b 2020 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2021 struct slave *slave;
2022 struct net_device *slave_dev;
2023 struct sockaddr addr;
2024
2025 write_lock_bh(&bond->lock);
2026
ff59c456
JV		 2027 netif_carrier_off(bond_dev);
2028
1da177e4
LT		 2029 if (bond->slave_cnt == 0) {
2030 goto out;
2031 }
2032
2033 bond->current_arp_slave = NULL;
2034 bond->primary_slave = NULL;
2035 bond_change_active_slave(bond, NULL);
2036
2037 while ((slave = bond->first_slave) != NULL) {
2038 /* Inform AD package of unbinding of slave
2039 * before slave is detached from the list.
2040 */
2041 if (bond->params.mode == BOND_MODE_8023AD) {
2042 bond_3ad_unbind_slave(slave);
2043 }
2044
2045 slave_dev = slave->dev;
2046 bond_detach_slave(bond, slave);
2047
2543331d
JV		 2048 /* now that the slave is detached, unlock and perform
2049 * all the undo steps that should not be called from
2050 * within a lock.
2051 */
2052 write_unlock_bh(&bond->lock);
2053
58402054 2054 if (bond_is_lb(bond)) {
1da177e4
LT		 2055 /* must be called only after the slave
2056 * has been detached from the list
2057 */
2058 bond_alb_deinit_slave(bond, slave);
2059 }
2060
8531c5ff
AK		 2061 bond_compute_features(bond);
2062
b76cdba9 2063 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2064 bond_del_vlans_from_slave(bond, slave_dev);
2065
2066 /* If the mode USES_PRIMARY, then we should only remove its
2067 * promisc and mc settings if it was the curr_active_slave, but that was
2068 * already taken care of above when we detached the slave
2069 */
2070 if (!USES_PRIMARY(bond->params.mode)) {
2071 /* unset promiscuity level from slave */
2072 if (bond_dev->flags & IFF_PROMISC) {
2073 dev_set_promiscuity(slave_dev, -1);
2074 }
2075
2076 /* unset allmulti level from slave */
2077 if (bond_dev->flags & IFF_ALLMULTI) {
2078 dev_set_allmulti(slave_dev, -1);
2079 }
2080
2081 /* flush master's mc_list from slave */
b9e40857 2082 netif_addr_lock_bh(bond_dev);
1da177e4 2083 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2084 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2085 }
2086
2087 netdev_set_master(slave_dev, NULL);
2088
2089 /* close slave before restoring its mac address */
2090 dev_close(slave_dev);
2091
dd957c57 2092 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2093 /* restore original ("permanent") mac address*/
2094 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2095 addr.sa_family = slave_dev->type;
2096 dev_set_mac_address(slave_dev, &addr);
2097 }
1da177e4 2098
8f903c70
JV		 2099 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2100 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2101
2102 kfree(slave);
2103
2104 /* re-acquire the lock before getting the next slave */
2105 write_lock_bh(&bond->lock);
2106 }
2107
2108 /* zero the mac address of the master so it will be
2109 * set by the application to the mac address of the
2110 * first slave
2111 */
2112 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2113
2114 if (list_empty(&bond->vlan_list)) {
2115 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2116 } else {
2117 printk(KERN_WARNING DRV_NAME
4e0952c7 2118 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2119 "still has VLANs.\n",
4e0952c7 2120 bond_dev->name, bond_dev->name);
1da177e4 2121 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2122 ": %s: When re-adding slaves, make sure the bond's "
2123 "HW address matches its VLANs'.\n",
2124 bond_dev->name);
1da177e4
LT		 2125 }
2126
2127 printk(KERN_INFO DRV_NAME
2128 ": %s: released all slaves\n",
2129 bond_dev->name);
2130
2131out:
2132 write_unlock_bh(&bond->lock);
2133
2134 return 0;
2135}
2136
2137/*
2138 * This function changes the active slave to slave <slave_dev>.
2139 * It returns -EINVAL in the following cases.
2140 * - <slave_dev> is not found in the list.
2141 * - There is not active slave now.
2142 * - <slave_dev> is already active.
2143 * - The link state of <slave_dev> is not BOND_LINK_UP.
2144 * - <slave_dev> is not running.
2145 * In these cases, this fuction does nothing.
2146 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2147 */
2148static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2149{
454d7c9b 2150 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2151 struct slave *old_active = NULL;
2152 struct slave *new_active = NULL;
2153 int res = 0;
2154
2155 if (!USES_PRIMARY(bond->params.mode)) {
2156 return -EINVAL;
2157 }
2158
2159 /* Verify that master_dev is indeed the master of slave_dev */
2160 if (!(slave_dev->flags & IFF_SLAVE) ||
2161 (slave_dev->master != bond_dev)) {
2162 return -EINVAL;
2163 }
2164
059fe7a5 2165 read_lock(&bond->lock);
1da177e4 2166
059fe7a5 2167 read_lock(&bond->curr_slave_lock);
1da177e4 2168 old_active = bond->curr_active_slave;
059fe7a5
JV		 2169 read_unlock(&bond->curr_slave_lock);
2170
1da177e4
LT		 2171 new_active = bond_get_slave_by_dev(bond, slave_dev);
2172
2173 /*
2174 * Changing to the current active: do nothing; return success.
2175 */
2176 if (new_active && (new_active == old_active)) {
059fe7a5 2177 read_unlock(&bond->lock);
1da177e4
LT		 2178 return 0;
2179 }
2180
2181 if ((new_active) &&
2182 (old_active) &&
2183 (new_active->link == BOND_LINK_UP) &&
2184 IS_UP(new_active->dev)) {
059fe7a5 2185 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2186 bond_change_active_slave(bond, new_active);
059fe7a5 2187 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2188 } else {
2189 res = -EINVAL;
2190 }
2191
059fe7a5 2192 read_unlock(&bond->lock);
1da177e4
LT		 2193
2194 return res;
2195}
2196
1da177e4
LT		 2197static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2198{
454d7c9b 2199 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2200
2201 info->bond_mode = bond->params.mode;
2202 info->miimon = bond->params.miimon;
2203
6603a6f2 2204 read_lock(&bond->lock);
1da177e4 2205 info->num_slaves = bond->slave_cnt;
6603a6f2 2206 read_unlock(&bond->lock);
1da177e4
LT		 2207
2208 return 0;
2209}
2210
2211static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2212{
454d7c9b 2213 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2214 struct slave *slave;
2215 int i, found = 0;
2216
2217 if (info->slave_id < 0) {
2218 return -ENODEV;
2219 }
2220
6603a6f2 2221 read_lock(&bond->lock);
1da177e4
LT		 2222
2223 bond_for_each_slave(bond, slave, i) {
2224 if (i == (int)info->slave_id) {
2225 found = 1;
2226 break;
2227 }
2228 }
2229
6603a6f2 2230 read_unlock(&bond->lock);
1da177e4
LT		 2231
2232 if (found) {
2233 strcpy(info->slave_name, slave->dev->name);
2234 info->link = slave->link;
2235 info->state = slave->state;
2236 info->link_failure_count = slave->link_failure_count;
2237 } else {
2238 return -ENODEV;
2239 }
2240
2241 return 0;
2242}
2243
2244/*-------------------------------- Monitoring -------------------------------*/
2245
1da177e4 2246
f0c76d61
JV		 2247static int bond_miimon_inspect(struct bonding *bond)
2248{
2249 struct slave *slave;
2250 int i, link_state, commit = 0;
1da177e4
LT		 2251
2252 bond_for_each_slave(bond, slave, i) {
f0c76d61 2253 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2254
f0c76d61 2255 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2256
2257 switch (slave->link) {
f0c76d61
JV		 2258 case BOND_LINK_UP:
2259 if (link_state)
2260 continue;
1da177e4 2261
f0c76d61
JV		 2262 slave->link = BOND_LINK_FAIL;
2263 slave->delay = bond->params.downdelay;
2264 if (slave->delay) {
2265 printk(KERN_INFO DRV_NAME
2266 ": %s: link status down for %s"
2267 "interface %s, disabling it in %d ms.\n",
2268 bond->dev->name,
2269 (bond->params.mode ==
2270 BOND_MODE_ACTIVEBACKUP) ?
2271 ((slave->state == BOND_STATE_ACTIVE) ?
2272 "active " : "backup ") : "",
2273 slave->dev->name,
2274 bond->params.downdelay * bond->params.miimon);
1da177e4 2275 }
f0c76d61
JV		 2276 /*FALLTHRU*/
2277 case BOND_LINK_FAIL:
2278 if (link_state) {
2279 /*
2280 * recovered before downdelay expired
2281 */
2282 slave->link = BOND_LINK_UP;
1da177e4
LT		 2283 slave->jiffies = jiffies;
2284 printk(KERN_INFO DRV_NAME
2285 ": %s: link status up again after %d "
2286 "ms for interface %s.\n",
1b76b316 2287 bond->dev->name,
f0c76d61
JV		 2288 (bond->params.downdelay - slave->delay) *
2289 bond->params.miimon,
2290 slave->dev->name);
2291 continue;
1da177e4 2292 }
f0c76d61
JV		 2293
2294 if (slave->delay <= 0) {
2295 slave->new_link = BOND_LINK_DOWN;
2296 commit++;
2297 continue;
1da177e4 2298 }
1da177e4 2299
f0c76d61
JV		 2300 slave->delay--;
2301 break;
2302
2303 case BOND_LINK_DOWN:
2304 if (!link_state)
2305 continue;
2306
2307 slave->link = BOND_LINK_BACK;
2308 slave->delay = bond->params.updelay;
2309
2310 if (slave->delay) {
2311 printk(KERN_INFO DRV_NAME
2312 ": %s: link status up for "
2313 "interface %s, enabling it in %d ms.\n",
2314 bond->dev->name, slave->dev->name,
2315 bond->params.updelay *
2316 bond->params.miimon);
2317 }
2318 /*FALLTHRU*/
2319 case BOND_LINK_BACK:
2320 if (!link_state) {
2321 slave->link = BOND_LINK_DOWN;
1da177e4
LT		 2322 printk(KERN_INFO DRV_NAME
2323 ": %s: link status down again after %d "
2324 "ms for interface %s.\n",
1b76b316 2325 bond->dev->name,
f0c76d61
JV		 2326 (bond->params.updelay - slave->delay) *
2327 bond->params.miimon,
2328 slave->dev->name);
2329
2330 continue;
2331 }
2332
2333 if (slave->delay <= 0) {
2334 slave->new_link = BOND_LINK_UP;
2335 commit++;
2336 continue;
1da177e4 2337 }
f0c76d61
JV		 2338
2339 slave->delay--;
1da177e4 2340 break;
f0c76d61
JV		 2341 }
2342 }
1da177e4 2343
f0c76d61
JV		 2344 return commit;
2345}
1da177e4 2346
f0c76d61
JV		 2347static void bond_miimon_commit(struct bonding *bond)
2348{
2349 struct slave *slave;
2350 int i;
2351
2352 bond_for_each_slave(bond, slave, i) {
2353 switch (slave->new_link) {
2354 case BOND_LINK_NOCHANGE:
2355 continue;
1da177e4 2356
f0c76d61
JV		 2357 case BOND_LINK_UP:
2358 slave->link = BOND_LINK_UP;
2359 slave->jiffies = jiffies;
2360
2361 if (bond->params.mode == BOND_MODE_8023AD) {
2362 /* prevent it from being the active one */
2363 slave->state = BOND_STATE_BACKUP;
2364 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2365 /* make it immediately active */
2366 slave->state = BOND_STATE_ACTIVE;
2367 } else if (slave != bond->primary_slave) {
2368 /* prevent it from being the active one */
2369 slave->state = BOND_STATE_BACKUP;
1da177e4 2370 }
1da177e4 2371
f0c76d61
JV		 2372 printk(KERN_INFO DRV_NAME
2373 ": %s: link status definitely "
2374 "up for interface %s.\n",
2375 bond->dev->name, slave->dev->name);
1da177e4 2376
f0c76d61
JV		 2377 /* notify ad that the link status has changed */
2378 if (bond->params.mode == BOND_MODE_8023AD)
2379 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2380
58402054 2381 if (bond_is_lb(bond))
f0c76d61
JV		 2382 bond_alb_handle_link_change(bond, slave,
2383 BOND_LINK_UP);
1da177e4 2384
f0c76d61
JV		 2385 if (!bond->curr_active_slave ||
2386 (slave == bond->primary_slave))
2387 goto do_failover;
1da177e4 2388
f0c76d61 2389 continue;
059fe7a5 2390
f0c76d61 2391 case BOND_LINK_DOWN:
fba4acda
JV		 2392 if (slave->link_failure_count < UINT_MAX)
2393 slave->link_failure_count++;
2394
f0c76d61 2395 slave->link = BOND_LINK_DOWN;
1da177e4 2396
f0c76d61
JV		 2397 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2398 bond->params.mode == BOND_MODE_8023AD)
2399 bond_set_slave_inactive_flags(slave);
2400
2401 printk(KERN_INFO DRV_NAME
2402 ": %s: link status definitely down for "
2403 "interface %s, disabling it\n",
2404 bond->dev->name, slave->dev->name);
2405
2406 if (bond->params.mode == BOND_MODE_8023AD)
2407 bond_3ad_handle_link_change(slave,
2408 BOND_LINK_DOWN);
2409
2410 if (bond->params.mode == BOND_MODE_TLB ||
2411 bond->params.mode == BOND_MODE_ALB)
2412 bond_alb_handle_link_change(bond, slave,
2413 BOND_LINK_DOWN);
2414
2415 if (slave == bond->curr_active_slave)
2416 goto do_failover;
2417
2418 continue;
2419
2420 default:
2421 printk(KERN_ERR DRV_NAME
2422 ": %s: invalid new link %d on slave %s\n",
2423 bond->dev->name, slave->new_link,
2424 slave->dev->name);
2425 slave->new_link = BOND_LINK_NOCHANGE;
2426
2427 continue;
2428 }
2429
2430do_failover:
2431 ASSERT_RTNL();
2432 write_lock_bh(&bond->curr_slave_lock);
2433 bond_select_active_slave(bond);
2434 write_unlock_bh(&bond->curr_slave_lock);
2435 }
2436
2437 bond_set_carrier(bond);
1da177e4
LT		 2438}
2439
0b0eef66
JV		 2440/*
2441 * bond_mii_monitor
2442 *
2443 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2444 * inspection, then (if inspection indicates something needs to be done)
2445 * an acquisition of appropriate locks followed by a commit phase to
2446 * implement whatever link state changes are indicated.
0b0eef66
JV		 2447 */
2448void bond_mii_monitor(struct work_struct *work)
2449{
2450 struct bonding *bond = container_of(work, struct bonding,
2451 mii_work.work);
0b0eef66
JV		 2452
2453 read_lock(&bond->lock);
f0c76d61
JV		 2454 if (bond->kill_timers)
2455 goto out;
2456
2457 if (bond->slave_cnt == 0)
2458 goto re_arm;
b59f9f74
JV		 2459
2460 if (bond->send_grat_arp) {
2461 read_lock(&bond->curr_slave_lock);
2462 bond_send_gratuitous_arp(bond);
2463 read_unlock(&bond->curr_slave_lock);
2464 }
2465
305d552a
BH		 2466 if (bond->send_unsol_na) {
2467 read_lock(&bond->curr_slave_lock);
2468 bond_send_unsolicited_na(bond);
2469 read_unlock(&bond->curr_slave_lock);
2470 }
2471
f0c76d61 2472 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2473 read_unlock(&bond->lock);
2474 rtnl_lock();
2475 read_lock(&bond->lock);
f0c76d61
JV		 2476
2477 bond_miimon_commit(bond);
2478
5655662d
JV		 2479 read_unlock(&bond->lock);
2480 rtnl_unlock(); /* might sleep, hold no other locks */
2481 read_lock(&bond->lock);
0b0eef66
JV		 2482 }
2483
f0c76d61
JV		 2484re_arm:
2485 if (bond->params.miimon)
2486 queue_delayed_work(bond->wq, &bond->mii_work,
2487 msecs_to_jiffies(bond->params.miimon));
2488out:
0b0eef66 2489 read_unlock(&bond->lock);
0b0eef66 2490}
c3ade5ca 2491
d3bb52b0 2492static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2493{
2494 struct in_device *idev;
2495 struct in_ifaddr *ifa;
a144ea4b 2496 __be32 addr = 0;
c3ade5ca
JV		 2497
2498 if (!dev)
2499 return 0;
2500
2501 rcu_read_lock();
e5ed6399 2502 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2503 if (!idev)
2504 goto out;
2505
2506 ifa = idev->ifa_list;
2507 if (!ifa)
2508 goto out;
2509
2510 addr = ifa->ifa_local;
2511out:
2512 rcu_read_unlock();
2513 return addr;
2514}
2515
d3bb52b0 2516static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2517{
0883beca 2518 struct vlan_entry *vlan;
f5b2b966
JV		 2519
2520 if (ip == bond->master_ip)
2521 return 1;
2522
0883beca 2523 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2524 if (ip == vlan->vlan_ip)
2525 return 1;
2526 }
2527
2528 return 0;
2529}
2530
c3ade5ca
JV		 2531/*
2532 * We go to the (large) trouble of VLAN tagging ARP frames because
2533 * switches in VLAN mode (especially if ports are configured as
2534 * "native" to a VLAN) might not pass non-tagged frames.
2535 */
d3bb52b0 2536static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2537{
2538 struct sk_buff *skb;
2539
5a03cdb7 2540 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
c3ade5ca
JV		 2541 slave_dev->name, dest_ip, src_ip, vlan_id);
2542
2543 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2544 NULL, slave_dev->dev_addr, NULL);
2545
2546 if (!skb) {
2547 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2548 return;
2549 }
2550 if (vlan_id) {
2551 skb = vlan_put_tag(skb, vlan_id);
2552 if (!skb) {
2553 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2554 return;
2555 }
2556 }
2557 arp_xmit(skb);
2558}
2559
2560
1da177e4
LT		 2561static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2562{
c3ade5ca 2563 int i, vlan_id, rv;
d3bb52b0 2564 __be32 *targets = bond->params.arp_targets;
0883beca 2565 struct vlan_entry *vlan;
c3ade5ca
JV		 2566 struct net_device *vlan_dev;
2567 struct flowi fl;
2568 struct rtable *rt;
1da177e4 2569
6b780567
MW		 2570 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2571 if (!targets[i])
2572 continue;
5a03cdb7 2573 pr_debug("basa: target %x\n", targets[i]);
c3ade5ca 2574 if (list_empty(&bond->vlan_list)) {
5a03cdb7 2575 pr_debug("basa: empty vlan: arp_send\n");
c3ade5ca
JV		 2576 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2577 bond->master_ip, 0);
2578 continue;
2579 }
2580
2581 /*
2582 * If VLANs are configured, we do a route lookup to
2583 * determine which VLAN interface would be used, so we
2584 * can tag the ARP with the proper VLAN tag.
2585 */
2586 memset(&fl, 0, sizeof(fl));
2587 fl.fl4_dst = targets[i];
2588 fl.fl4_tos = RTO_ONLINK;
2589
f206351a 2590 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2591 if (rv) {
2592 if (net_ratelimit()) {
2593 printk(KERN_WARNING DRV_NAME
63779436
HH		 2594 ": %s: no route to arp_ip_target %pI4\n",
2595 bond->dev->name, &fl.fl4_dst);
c3ade5ca
JV		 2596 }
2597 continue;
2598 }
2599
2600 /*
2601 * This target is not on a VLAN
2602 */
2603 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2604 ip_rt_put(rt);
5a03cdb7 2605 pr_debug("basa: rtdev == bond->dev: arp_send\n");
c3ade5ca
JV		 2606 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2607 bond->master_ip, 0);
2608 continue;
2609 }
2610
2611 vlan_id = 0;
0883beca 2612 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2613 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2614 if (vlan_dev == rt->u.dst.dev) {
2615 vlan_id = vlan->vlan_id;
5a03cdb7 2616 pr_debug("basa: vlan match on %s %d\n",
c3ade5ca
JV		 2617 vlan_dev->name, vlan_id);
2618 break;
2619 }
2620 }
2621
2622 if (vlan_id) {
ed4b9f80 2623 ip_rt_put(rt);
c3ade5ca
JV		 2624 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2625 vlan->vlan_ip, vlan_id);
2626 continue;
2627 }
2628
2629 if (net_ratelimit()) {
2630 printk(KERN_WARNING DRV_NAME
63779436
HH		 2631 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2632 bond->dev->name, &fl.fl4_dst,
c3ade5ca
JV		 2633 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2634 }
ed4b9f80 2635 ip_rt_put(rt);
c3ade5ca
JV		 2636 }
2637}
2638
2639/*
2640 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2641 * for each VLAN above us.
b59f9f74
JV		 2642 *
2643 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2644 */
2645static void bond_send_gratuitous_arp(struct bonding *bond)
2646{
2647 struct slave *slave = bond->curr_active_slave;
2648 struct vlan_entry *vlan;
2649 struct net_device *vlan_dev;
2650
5a03cdb7 2651 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
c3ade5ca 2652 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2653
2654 if (!slave || !bond->send_grat_arp ||
2655 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2656 return;
2657
b59f9f74
JV		 2658 bond->send_grat_arp--;
2659
c3ade5ca
JV		 2660 if (bond->master_ip) {
2661 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2662 bond->master_ip, 0);
c3ade5ca
JV		 2663 }
2664
2665 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2666 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2667 if (vlan->vlan_ip) {
2668 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2669 vlan->vlan_ip, vlan->vlan_id);
2670 }
1da177e4
LT		 2671 }
2672}
2673
d3bb52b0 2674static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2675{
2676 int i;
d3bb52b0 2677 __be32 *targets = bond->params.arp_targets;
f5b2b966
JV		 2678
2679 targets = bond->params.arp_targets;
2680 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
5a03cdb7 2681 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
63779436 2682 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
f5b2b966
JV		 2683 if (sip == targets[i]) {
2684 if (bond_has_this_ip(bond, tip))
2685 slave->last_arp_rx = jiffies;
2686 return;
2687 }
2688 }
2689}
2690
2691static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2692{
2693 struct arphdr *arp;
2694 struct slave *slave;
2695 struct bonding *bond;
2696 unsigned char *arp_ptr;
d3bb52b0 2697 __be32 sip, tip;
f5b2b966 2698
c346dca1 2699 if (dev_net(dev) != &init_net)
e730c155
EB		 2700 goto out;
2701
f5b2b966
JV		 2702 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2703 goto out;
2704
454d7c9b 2705 bond = netdev_priv(dev);
f5b2b966
JV		 2706 read_lock(&bond->lock);
2707
5a03cdb7 2708 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
f5b2b966
JV		 2709 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2710 orig_dev ? orig_dev->name : "NULL");
2711
2712 slave = bond_get_slave_by_dev(bond, orig_dev);
2713 if (!slave || !slave_do_arp_validate(bond, slave))
2714 goto out_unlock;
2715
988b7050 2716 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2717 goto out_unlock;
2718
d0a92be0 2719 arp = arp_hdr(skb);
f5b2b966
JV		 2720 if (arp->ar_hln != dev->addr_len ||
2721 skb->pkt_type == PACKET_OTHERHOST ||
2722 skb->pkt_type == PACKET_LOOPBACK ||
2723 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2724 arp->ar_pro != htons(ETH_P_IP) ||
2725 arp->ar_pln != 4)
2726 goto out_unlock;
2727
2728 arp_ptr = (unsigned char *)(arp + 1);
2729 arp_ptr += dev->addr_len;
2730 memcpy(&sip, arp_ptr, 4);
2731 arp_ptr += 4 + dev->addr_len;
2732 memcpy(&tip, arp_ptr, 4);
2733
5a03cdb7 2734 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
63779436
HH		 2735 bond->dev->name, slave->dev->name, slave->state,
2736 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2737 &sip, &tip);
f5b2b966
JV		 2738
2739 /*
2740 * Backup slaves won't see the ARP reply, but do come through
2741 * here for each ARP probe (so we swap the sip/tip to validate
2742 * the probe). In a "redundant switch, common router" type of
2743 * configuration, the ARP probe will (hopefully) travel from
2744 * the active, through one switch, the router, then the other
2745 * switch before reaching the backup.
2746 */
2747 if (slave->state == BOND_STATE_ACTIVE)
2748 bond_validate_arp(bond, slave, sip, tip);
2749 else
2750 bond_validate_arp(bond, slave, tip, sip);
2751
2752out_unlock:
2753 read_unlock(&bond->lock);
2754out:
2755 dev_kfree_skb(skb);
2756 return NET_RX_SUCCESS;
2757}
2758
1da177e4
LT		 2759/*
2760 * this function is called regularly to monitor each slave's link
2761 * ensuring that traffic is being sent and received when arp monitoring
2762 * is used in load-balancing mode. if the adapter has been dormant, then an
2763 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2764 * arp monitoring in active backup mode.
2765 */
1b76b316 2766void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2767{
1b76b316
JV		 2768 struct bonding *bond = container_of(work, struct bonding,
2769 arp_work.work);
1da177e4
LT		 2770 struct slave *slave, *oldcurrent;
2771 int do_failover = 0;
2772 int delta_in_ticks;
2773 int i;
2774
2775 read_lock(&bond->lock);
2776
5ce0da8f 2777 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2778
2779 if (bond->kill_timers) {
2780 goto out;
2781 }
2782
2783 if (bond->slave_cnt == 0) {
2784 goto re_arm;
2785 }
2786
2787 read_lock(&bond->curr_slave_lock);
2788 oldcurrent = bond->curr_active_slave;
2789 read_unlock(&bond->curr_slave_lock);
2790
2791 /* see if any of the previous devices are up now (i.e. they have
2792 * xmt and rcv traffic). the curr_active_slave does not come into
2793 * the picture unless it is null. also, slave->jiffies is not needed
2794 * here because we send an arp on each slave and give a slave as
2795 * long as it needs to get the tx/rx within the delta.
2796 * TODO: what about up/down delay in arp mode? it wasn't here before
2797 * so it can wait
2798 */
2799 bond_for_each_slave(bond, slave, i) {
2800 if (slave->link != BOND_LINK_UP) {
b63bb739
DS		 2801 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2802 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2803
2804 slave->link = BOND_LINK_UP;
2805 slave->state = BOND_STATE_ACTIVE;
2806
2807 /* primary_slave has no meaning in round-robin
2808 * mode. the window of a slave being up and
2809 * curr_active_slave being null after enslaving
2810 * is closed.
2811 */
2812 if (!oldcurrent) {
2813 printk(KERN_INFO DRV_NAME
2814 ": %s: link status definitely "
2815 "up for interface %s, ",
1b76b316 2816 bond->dev->name,
1da177e4
LT		 2817 slave->dev->name);
2818 do_failover = 1;
2819 } else {
2820 printk(KERN_INFO DRV_NAME
2821 ": %s: interface %s is now up\n",
1b76b316 2822 bond->dev->name,
1da177e4
LT		 2823 slave->dev->name);
2824 }
2825 }
2826 } else {
2827 /* slave->link == BOND_LINK_UP */
2828
2829 /* not all switches will respond to an arp request
2830 * when the source ip is 0, so don't take the link down
2831 * if we don't know our ip yet
2832 */
b63bb739 2833 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
4b8a9239 2834 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2835
2836 slave->link = BOND_LINK_DOWN;
2837 slave->state = BOND_STATE_BACKUP;
2838
2839 if (slave->link_failure_count < UINT_MAX) {
2840 slave->link_failure_count++;
2841 }
2842
2843 printk(KERN_INFO DRV_NAME
2844 ": %s: interface %s is now down.\n",
1b76b316 2845 bond->dev->name,
1da177e4
LT		 2846 slave->dev->name);
2847
2848 if (slave == oldcurrent) {
2849 do_failover = 1;
2850 }
2851 }
2852 }
2853
2854 /* note: if switch is in round-robin mode, all links
2855 * must tx arp to ensure all links rx an arp - otherwise
2856 * links may oscillate or not come up at all; if switch is
2857 * in something like xor mode, there is nothing we can
2858 * do - all replies will be rx'ed on same link causing slaves
2859 * to be unstable during low/no traffic periods
2860 */
2861 if (IS_UP(slave->dev)) {
2862 bond_arp_send_all(bond, slave);
2863 }
2864 }
2865
2866 if (do_failover) {
059fe7a5 2867 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2868
2869 bond_select_active_slave(bond);
2870
059fe7a5 2871 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2872 }
2873
2874re_arm:
1b76b316
JV		 2875 if (bond->params.arp_interval)
2876 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2877out:
2878 read_unlock(&bond->lock);
2879}
2880
2881/*
b2220cad
JV		 2882 * Called to inspect slaves for active-backup mode ARP monitor link state
2883 * changes. Sets new_link in slaves to specify what action should take
2884 * place for the slave. Returns 0 if no changes are found, >0 if changes
2885 * to link states must be committed.
2886 *
2887 * Called with bond->lock held for read.
1da177e4 2888 */
b2220cad 2889static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2890{
1da177e4 2891 struct slave *slave;
b2220cad 2892 int i, commit = 0;
1da177e4 2893
b2220cad
JV		 2894 bond_for_each_slave(bond, slave, i) {
2895 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2896
b2220cad
JV		 2897 if (slave->link != BOND_LINK_UP) {
2898 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2899 delta_in_ticks)) {
2900 slave->new_link = BOND_LINK_UP;
2901 commit++;
2902 }
1da177e4 2903
b2220cad
JV		 2904 continue;
2905 }
1da177e4 2906
b2220cad
JV		 2907 /*
2908 * Give slaves 2*delta after being enslaved or made
2909 * active. This avoids bouncing, as the last receive
2910 * times need a full ARP monitor cycle to be updated.
2911 */
2912 if (!time_after_eq(jiffies, slave->jiffies +
2913 2 * delta_in_ticks))
2914 continue;
2915
2916 /*
2917 * Backup slave is down if:
2918 * - No current_arp_slave AND
2919 * - more than 3*delta since last receive AND
2920 * - the bond has an IP address
2921 *
2922 * Note: a non-null current_arp_slave indicates
2923 * the curr_active_slave went down and we are
2924 * searching for a new one; under this condition
2925 * we only take the curr_active_slave down - this
2926 * gives each slave a chance to tx/rx traffic
2927 * before being taken out
2928 */
2929 if (slave->state == BOND_STATE_BACKUP &&
2930 !bond->current_arp_slave &&
2931 time_after(jiffies, slave_last_rx(bond, slave) +
2932 3 * delta_in_ticks)) {
2933 slave->new_link = BOND_LINK_DOWN;
2934 commit++;
2935 }
2936
2937 /*
2938 * Active slave is down if:
2939 * - more than 2*delta since transmitting OR
2940 * - (more than 2*delta since receive AND
2941 * the bond has an IP address)
2942 */
2943 if ((slave->state == BOND_STATE_ACTIVE) &&
2944 (time_after_eq(jiffies, slave->dev->trans_start +
2945 2 * delta_in_ticks) ||
2946 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2947 + 2 * delta_in_ticks)))) {
2948 slave->new_link = BOND_LINK_DOWN;
2949 commit++;
2950 }
1da177e4
LT		 2951 }
2952
b2220cad
JV		 2953 read_lock(&bond->curr_slave_lock);
2954
2955 /*
2956 * Trigger a commit if the primary option setting has changed.
1da177e4 2957 */
b2220cad
JV		 2958 if (bond->primary_slave &&
2959 (bond->primary_slave != bond->curr_active_slave) &&
2960 (bond->primary_slave->link == BOND_LINK_UP))
2961 commit++;
1da177e4 2962
b2220cad 2963 read_unlock(&bond->curr_slave_lock);
1da177e4 2964
b2220cad
JV		 2965 return commit;
2966}
1da177e4 2967
b2220cad
JV		 2968/*
2969 * Called to commit link state changes noted by inspection step of
2970 * active-backup mode ARP monitor.
2971 *
2972 * Called with RTNL and bond->lock for read.
2973 */
2974static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2975{
2976 struct slave *slave;
2977 int i;
1da177e4 2978
b2220cad
JV		 2979 bond_for_each_slave(bond, slave, i) {
2980 switch (slave->new_link) {
2981 case BOND_LINK_NOCHANGE:
2982 continue;
ff59c456 2983
b2220cad
JV		 2984 case BOND_LINK_UP:
2985 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2986
b2220cad
JV		 2987 if (!bond->curr_active_slave &&
2988 time_before_eq(jiffies, slave->dev->trans_start +
2989 delta_in_ticks)) {
2990 slave->link = BOND_LINK_UP;
2991 bond_change_active_slave(bond, slave);
2992 bond->current_arp_slave = NULL;
2993
2994 printk(KERN_INFO DRV_NAME
2995 ": %s: %s is up and now the "
2996 "active interface\n",
2997 bond->dev->name, slave->dev->name);
2998
2999 } else if (bond->curr_active_slave != slave) {
3000 /* this slave has just come up but we
3001 * already have a current slave; this can
3002 * also happen if bond_enslave adds a new
3003 * slave that is up while we are searching
3004 * for a new slave
1da177e4 3005 */
b2220cad
JV		 3006 slave->link = BOND_LINK_UP;
3007 bond_set_slave_inactive_flags(slave);
3008 bond->current_arp_slave = NULL;
3009
3010 printk(KERN_INFO DRV_NAME
3011 ": %s: backup interface %s is now up\n",
3012 bond->dev->name, slave->dev->name);
3013 }
1da177e4 3014
b2220cad 3015 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3016
b2220cad 3017 break;
1da177e4 3018
b2220cad
JV		 3019 case BOND_LINK_DOWN:
3020 if (slave->link_failure_count < UINT_MAX)
3021 slave->link_failure_count++;
3022
3023 slave->link = BOND_LINK_DOWN;
3024
3025 if (slave == bond->curr_active_slave) {
3026 printk(KERN_INFO DRV_NAME
3027 ": %s: link status down for active "
3028 "interface %s, disabling it\n",
3029 bond->dev->name, slave->dev->name);
1da177e4
LT		 3030
3031 bond_set_slave_inactive_flags(slave);
3032
b2220cad
JV		 3033 write_lock_bh(&bond->curr_slave_lock);
3034
3035 bond_select_active_slave(bond);
3036 if (bond->curr_active_slave)
3037 bond->curr_active_slave->jiffies =
3038 jiffies;
3039
3040 write_unlock_bh(&bond->curr_slave_lock);
3041
3042 bond->current_arp_slave = NULL;
3043
3044 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3045 printk(KERN_INFO DRV_NAME
3046 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3047 bond->dev->name, slave->dev->name);
3048
3049 bond_set_slave_inactive_flags(slave);
1da177e4 3050 }
b2220cad
JV		 3051 break;
3052
3053 default:
3054 printk(KERN_ERR DRV_NAME
3055 ": %s: impossible: new_link %d on slave %s\n",
3056 bond->dev->name, slave->new_link,
3057 slave->dev->name);
1da177e4
LT		 3058 }
3059 }
3060
b2220cad
JV		 3061 /*
3062 * No race with changes to primary via sysfs, as we hold rtnl.
3063 */
3064 if (bond->primary_slave &&
3065 (bond->primary_slave != bond->curr_active_slave) &&
3066 (bond->primary_slave->link == BOND_LINK_UP)) {
3067 write_lock_bh(&bond->curr_slave_lock);
3068 bond_change_active_slave(bond, bond->primary_slave);
3069 write_unlock_bh(&bond->curr_slave_lock);
3070 }
1da177e4 3071
b2220cad
JV		 3072 bond_set_carrier(bond);
3073}
1da177e4 3074
b2220cad
JV		 3075/*
3076 * Send ARP probes for active-backup mode ARP monitor.
3077 *
3078 * Called with bond->lock held for read.
3079 */
3080static void bond_ab_arp_probe(struct bonding *bond)
3081{
3082 struct slave *slave;
3083 int i;
1da177e4 3084
b2220cad 3085 read_lock(&bond->curr_slave_lock);
1da177e4 3086
b2220cad
JV		 3087 if (bond->current_arp_slave && bond->curr_active_slave)
3088 printk("PROBE: c_arp %s && cas %s BAD\n",
3089 bond->current_arp_slave->dev->name,
3090 bond->curr_active_slave->dev->name);
1da177e4 3091
b2220cad
JV		 3092 if (bond->curr_active_slave) {
3093 bond_arp_send_all(bond, bond->curr_active_slave);
3094 read_unlock(&bond->curr_slave_lock);
3095 return;
3096 }
1da177e4 3097
b2220cad 3098 read_unlock(&bond->curr_slave_lock);
059fe7a5 3099
b2220cad
JV		 3100 /* if we don't have a curr_active_slave, search for the next available
3101 * backup slave from the current_arp_slave and make it the candidate
3102 * for becoming the curr_active_slave
3103 */
1da177e4 3104
b2220cad
JV		 3105 if (!bond->current_arp_slave) {
3106 bond->current_arp_slave = bond->first_slave;
3107 if (!bond->current_arp_slave)
3108 return;
3109 }
1da177e4 3110
b2220cad 3111 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3112
b2220cad
JV		 3113 /* search for next candidate */
3114 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3115 if (IS_UP(slave->dev)) {
3116 slave->link = BOND_LINK_BACK;
3117 bond_set_slave_active_flags(slave);
3118 bond_arp_send_all(bond, slave);
1da177e4 3119 slave->jiffies = jiffies;
b2220cad
JV		 3120 bond->current_arp_slave = slave;
3121 break;
1da177e4
LT		 3122 }
3123
b2220cad
JV		 3124 /* if the link state is up at this point, we
3125 * mark it down - this can happen if we have
3126 * simultaneous link failures and
3127 * reselect_active_interface doesn't make this
3128 * one the current slave so it is still marked
3129 * up when it is actually down
1da177e4 3130 */
b2220cad
JV		 3131 if (slave->link == BOND_LINK_UP) {
3132 slave->link = BOND_LINK_DOWN;
3133 if (slave->link_failure_count < UINT_MAX)
3134 slave->link_failure_count++;
1da177e4 3135
b2220cad
JV		 3136 bond_set_slave_inactive_flags(slave);
3137
3138 printk(KERN_INFO DRV_NAME
3139 ": %s: backup interface %s is now down.\n",
3140 bond->dev->name, slave->dev->name);
1da177e4 3141 }
b2220cad
JV		 3142 }
3143}
1da177e4 3144
b2220cad
JV		 3145void bond_activebackup_arp_mon(struct work_struct *work)
3146{
3147 struct bonding *bond = container_of(work, struct bonding,
3148 arp_work.work);
3149 int delta_in_ticks;
1da177e4 3150
b2220cad 3151 read_lock(&bond->lock);
1da177e4 3152
b2220cad
JV		 3153 if (bond->kill_timers)
3154 goto out;
1da177e4 3155
b2220cad 3156 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3157
b2220cad
JV		 3158 if (bond->slave_cnt == 0)
3159 goto re_arm;
3160
b59f9f74
JV		 3161 if (bond->send_grat_arp) {
3162 read_lock(&bond->curr_slave_lock);
3163 bond_send_gratuitous_arp(bond);
3164 read_unlock(&bond->curr_slave_lock);
3165 }
3166
305d552a
BH		 3167 if (bond->send_unsol_na) {
3168 read_lock(&bond->curr_slave_lock);
3169 bond_send_unsolicited_na(bond);
3170 read_unlock(&bond->curr_slave_lock);
3171 }
3172
b2220cad
JV		 3173 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3174 read_unlock(&bond->lock);
3175 rtnl_lock();
3176 read_lock(&bond->lock);
3177
3178 bond_ab_arp_commit(bond, delta_in_ticks);
3179
3180 read_unlock(&bond->lock);
3181 rtnl_unlock();
3182 read_lock(&bond->lock);
1da177e4
LT		 3183 }
3184
b2220cad
JV		 3185 bond_ab_arp_probe(bond);
3186
1da177e4
LT		 3187re_arm:
3188 if (bond->params.arp_interval) {
1b76b316 3189 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3190 }
3191out:
3192 read_unlock(&bond->lock);
3193}
3194
3195/*------------------------------ proc/seq_file-------------------------------*/
3196
3197#ifdef CONFIG_PROC_FS
3198
1da177e4 3199static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
1f78d9f9
HE		 3200 __acquires(&dev_base_lock)
3201 __acquires(&bond->lock)
1da177e4
LT		 3202{
3203 struct bonding *bond = seq->private;
3204 loff_t off = 0;
3205 struct slave *slave;
3206 int i;
3207
3208 /* make sure the bond won't be taken away */
3209 read_lock(&dev_base_lock);
6603a6f2 3210 read_lock(&bond->lock);
1da177e4
LT		 3211
3212 if (*pos == 0) {
3213 return SEQ_START_TOKEN;
3214 }
3215
3216 bond_for_each_slave(bond, slave, i) {
3217 if (++off == *pos) {
3218 return slave;
3219 }
3220 }
3221
3222 return NULL;
3223}
3224
3225static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3226{
3227 struct bonding *bond = seq->private;
3228 struct slave *slave = v;
3229
3230 ++*pos;
3231 if (v == SEQ_START_TOKEN) {
3232 return bond->first_slave;
3233 }
3234
3235 slave = slave->next;
3236
3237 return (slave == bond->first_slave) ? NULL : slave;
3238}
3239
3240static void bond_info_seq_stop(struct seq_file *seq, void *v)
1f78d9f9
HE		 3241 __releases(&bond->lock)
3242 __releases(&dev_base_lock)
1da177e4
LT		 3243{
3244 struct bonding *bond = seq->private;
3245
6603a6f2 3246 read_unlock(&bond->lock);
1da177e4
LT		 3247 read_unlock(&dev_base_lock);
3248}
3249
3250static void bond_info_show_master(struct seq_file *seq)
3251{
3252 struct bonding *bond = seq->private;
3253 struct slave *curr;
4756b02f 3254 int i;
1da177e4
LT		 3255
3256 read_lock(&bond->curr_slave_lock);
3257 curr = bond->curr_active_slave;
3258 read_unlock(&bond->curr_slave_lock);
3259
dd957c57 3260 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3261 bond_mode_name(bond->params.mode));
3262
dd957c57
JV		 3263 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3264 bond->params.fail_over_mac)
3915c1e8
JV		 3265 seq_printf(seq, " (fail_over_mac %s)",
3266 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3267
3268 seq_printf(seq, "\n");
3269
c61b75ad
MW		 3270 if (bond->params.mode == BOND_MODE_XOR ||
3271 bond->params.mode == BOND_MODE_8023AD) {
3272 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3273 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3274 bond->params.xmit_policy);
3275 }
3276
1da177e4
LT		 3277 if (USES_PRIMARY(bond->params.mode)) {
3278 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3279 (bond->primary_slave) ?
3280 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3281
3282 seq_printf(seq, "Currently Active Slave: %s\n",
3283 (curr) ? curr->dev->name : "None");
3284 }
3285
ff59c456
JV		 3286 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3287 "up" : "down");
1da177e4
LT		 3288 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3289 seq_printf(seq, "Up Delay (ms): %d\n",
3290 bond->params.updelay * bond->params.miimon);
3291 seq_printf(seq, "Down Delay (ms): %d\n",
3292 bond->params.downdelay * bond->params.miimon);
3293
4756b02f
MW		 3294
3295 /* ARP information */
3296 if(bond->params.arp_interval > 0) {
3297 int printed=0;
3298 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3299 bond->params.arp_interval);
3300
3301 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3302
3303 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3304 if (!bond->params.arp_targets[i])
3305 continue;
3306 if (printed)
3307 seq_printf(seq, ",");
8cf14e38 3308 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
4756b02f
MW		 3309 printed = 1;
3310 }
3311 seq_printf(seq, "\n");
3312 }
3313
1da177e4
LT		 3314 if (bond->params.mode == BOND_MODE_8023AD) {
3315 struct ad_info ad_info;
3316
3317 seq_puts(seq, "\n802.3ad info\n");
3318 seq_printf(seq, "LACP rate: %s\n",
3319 (bond->params.lacp_fast) ? "fast" : "slow");
fd989c83
JV		 3320 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3321 ad_select_tbl[bond->params.ad_select].modename);
1da177e4
LT		 3322
3323 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3324 seq_printf(seq, "bond %s has no active aggregator\n",
3325 bond->dev->name);
3326 } else {
3327 seq_printf(seq, "Active Aggregator Info:\n");
3328
3329 seq_printf(seq, "\tAggregator ID: %d\n",
3330 ad_info.aggregator_id);
3331 seq_printf(seq, "\tNumber of ports: %d\n",
3332 ad_info.ports);
3333 seq_printf(seq, "\tActor Key: %d\n",
3334 ad_info.actor_key);
3335 seq_printf(seq, "\tPartner Key: %d\n",
3336 ad_info.partner_key);
e174961c
JB		 3337 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3338 ad_info.partner_system);
1da177e4
LT		 3339 }
3340 }
3341}
3342
3343static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3344{
3345 struct bonding *bond = seq->private;
3346
3347 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3348 seq_printf(seq, "MII Status: %s\n",
3349 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3350 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3351 slave->link_failure_count);
3352
e174961c 3353 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
1da177e4
LT		 3354
3355 if (bond->params.mode == BOND_MODE_8023AD) {
3356 const struct aggregator *agg
3357 = SLAVE_AD_INFO(slave).port.aggregator;
3358
3359 if (agg) {
3360 seq_printf(seq, "Aggregator ID: %d\n",
3361 agg->aggregator_identifier);
3362 } else {
3363 seq_puts(seq, "Aggregator ID: N/A\n");
3364 }
3365 }
3366}
3367
3368static int bond_info_seq_show(struct seq_file *seq, void *v)
3369{
3370 if (v == SEQ_START_TOKEN) {
3371 seq_printf(seq, "%s\n", version);
3372 bond_info_show_master(seq);
3373 } else {
3374 bond_info_show_slave(seq, v);
3375 }
3376
3377 return 0;
3378}
3379
4101dec9 3380static const struct seq_operations bond_info_seq_ops = {
1da177e4
LT		 3381 .start = bond_info_seq_start,
3382 .next = bond_info_seq_next,
3383 .stop = bond_info_seq_stop,
3384 .show = bond_info_seq_show,
3385};
3386
3387static int bond_info_open(struct inode *inode, struct file *file)
3388{
3389 struct seq_file *seq;
3390 struct proc_dir_entry *proc;
3391 int res;
3392
3393 res = seq_open(file, &bond_info_seq_ops);
3394 if (!res) {
3395 /* recover the pointer buried in proc_dir_entry data */
3396 seq = file->private_data;
3397 proc = PDE(inode);
3398 seq->private = proc->data;
3399 }
3400
3401 return res;
3402}
3403
d54b1fdb 3404static const struct file_operations bond_info_fops = {
1da177e4
LT		 3405 .owner = THIS_MODULE,
3406 .open = bond_info_open,
3407 .read = seq_read,
3408 .llseek = seq_lseek,
3409 .release = seq_release,
3410};
3411
3412static int bond_create_proc_entry(struct bonding *bond)
3413{
3414 struct net_device *bond_dev = bond->dev;
3415
3416 if (bond_proc_dir) {
a95609cb
DL		 3417 bond->proc_entry = proc_create_data(bond_dev->name,
3418 S_IRUGO, bond_proc_dir,
3419 &bond_info_fops, bond);
1da177e4
LT		 3420 if (bond->proc_entry == NULL) {
3421 printk(KERN_WARNING DRV_NAME
3422 ": Warning: Cannot create /proc/net/%s/%s\n",
3423 DRV_NAME, bond_dev->name);
3424 } else {
1da177e4
LT		 3425 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3426 }
3427 }
3428
3429 return 0;
3430}
3431
3432static void bond_remove_proc_entry(struct bonding *bond)
3433{
3434 if (bond_proc_dir && bond->proc_entry) {
3435 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3436 memset(bond->proc_file_name, 0, IFNAMSIZ);
3437 bond->proc_entry = NULL;
3438 }
3439}
3440
3441/* Create the bonding directory under /proc/net, if doesn't exist yet.
3442 * Caller must hold rtnl_lock.
3443 */
3444static void bond_create_proc_dir(void)
3445{
3446 int len = strlen(DRV_NAME);
3447
457c4cbc 3448 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
1da177e4
LT		 3449 bond_proc_dir = bond_proc_dir->next) {
3450 if ((bond_proc_dir->namelen == len) &&
3451 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3452 break;
3453 }
3454 }
3455
3456 if (!bond_proc_dir) {
457c4cbc 3457 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3458 if (bond_proc_dir) {
3459 bond_proc_dir->owner = THIS_MODULE;
3460 } else {
3461 printk(KERN_WARNING DRV_NAME
3462 ": Warning: cannot create /proc/net/%s\n",
3463 DRV_NAME);
3464 }
3465 }
3466}
3467
3468/* Destroy the bonding directory under /proc/net, if empty.
3469 * Caller must hold rtnl_lock.
3470 */
3471static void bond_destroy_proc_dir(void)
3472{
3473 struct proc_dir_entry *de;
3474
3475 if (!bond_proc_dir) {
3476 return;
3477 }
3478
3479 /* verify that the /proc dir is empty */
3480 for (de = bond_proc_dir->subdir; de; de = de->next) {
3481 /* ignore . and .. */
3482 if (*(de->name) != '.') {
3483 break;
3484 }
3485 }
3486
3487 if (de) {
3488 if (bond_proc_dir->owner == THIS_MODULE) {
3489 bond_proc_dir->owner = NULL;
3490 }
3491 } else {
457c4cbc 3492 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3493 bond_proc_dir = NULL;
3494 }
3495}
3496#endif /* CONFIG_PROC_FS */
3497
3498/*-------------------------- netdev event handling --------------------------*/
3499
3500/*
3501 * Change device name
3502 */
3503static int bond_event_changename(struct bonding *bond)
3504{
3505#ifdef CONFIG_PROC_FS
3506 bond_remove_proc_entry(bond);
3507 bond_create_proc_entry(bond);
3508#endif
b76cdba9
MW		 3509 down_write(&(bonding_rwsem));
3510 bond_destroy_sysfs_entry(bond);
3511 bond_create_sysfs_entry(bond);
3512 up_write(&(bonding_rwsem));
1da177e4
LT		 3513 return NOTIFY_DONE;
3514}
3515
3516static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3517{
454d7c9b 3518 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 3519
3520 switch (event) {
3521 case NETDEV_CHANGENAME:
3522 return bond_event_changename(event_bond);
3523 case NETDEV_UNREGISTER:
3b96c858 3524 bond_release_all(event_bond->dev);
1da177e4
LT		 3525 break;
3526 default:
3527 break;
3528 }
3529
3530 return NOTIFY_DONE;
3531}
3532
3533static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3534{
3535 struct net_device *bond_dev = slave_dev->master;
454d7c9b 3536 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3537
3538 switch (event) {
3539 case NETDEV_UNREGISTER:
3540 if (bond_dev) {
1284cd3a
JV		 3541 if (bond->setup_by_slave)
3542 bond_release_and_destroy(bond_dev, slave_dev);
3543 else
3544 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3545 }
3546 break;
3547 case NETDEV_CHANGE:
3548 /*
3549 * TODO: is this what we get if somebody
3550 * sets up a hierarchical bond, then rmmod's
3551 * one of the slave bonding devices?
3552 */
3553 break;
3554 case NETDEV_DOWN:
3555 /*
3556 * ... Or is it this?
3557 */
3558 break;
3559 case NETDEV_CHANGEMTU:
3560 /*
3561 * TODO: Should slaves be allowed to
3562 * independently alter their MTU? For
3563 * an active-backup bond, slaves need
3564 * not be the same type of device, so
3565 * MTUs may vary. For other modes,
3566 * slaves arguably should have the
3567 * same MTUs. To do this, we'd need to
3568 * take over the slave's change_mtu
3569 * function for the duration of their
3570 * servitude.
3571 */
3572 break;
3573 case NETDEV_CHANGENAME:
3574 /*
3575 * TODO: handle changing the primary's name
3576 */
3577 break;
8531c5ff
AK		 3578 case NETDEV_FEAT_CHANGE:
3579 bond_compute_features(bond);
3580 break;
1da177e4
LT		 3581 default:
3582 break;
3583 }
3584
3585 return NOTIFY_DONE;
3586}
3587
3588/*
3589 * bond_netdev_event: handle netdev notifier chain events.
3590 *
3591 * This function receives events for the netdev chain. The caller (an
e041c683 3592 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3593 * locks for us to safely manipulate the slave devices (RTNL lock,
3594 * dev_probe_lock).
3595 */
3596static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3597{
3598 struct net_device *event_dev = (struct net_device *)ptr;
3599
c346dca1 3600 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3601 return NOTIFY_DONE;
3602
5a03cdb7 3603 pr_debug("event_dev: %s, event: %lx\n",
1da177e4
LT		 3604 (event_dev ? event_dev->name : "None"),
3605 event);
3606
0b680e75
JV		 3607 if (!(event_dev->priv_flags & IFF_BONDING))
3608 return NOTIFY_DONE;
3609
1da177e4 3610 if (event_dev->flags & IFF_MASTER) {
5a03cdb7 3611 pr_debug("IFF_MASTER\n");
1da177e4
LT		 3612 return bond_master_netdev_event(event, event_dev);
3613 }
3614
3615 if (event_dev->flags & IFF_SLAVE) {
5a03cdb7 3616 pr_debug("IFF_SLAVE\n");
1da177e4
LT		 3617 return bond_slave_netdev_event(event, event_dev);
3618 }
3619
3620 return NOTIFY_DONE;
3621}
3622
c3ade5ca
JV		 3623/*
3624 * bond_inetaddr_event: handle inetaddr notifier chain events.
3625 *
3626 * We keep track of device IPs primarily to use as source addresses in
3627 * ARP monitor probes (rather than spewing out broadcasts all the time).
3628 *
3629 * We track one IP for the main device (if it has one), plus one per VLAN.
3630 */
3631static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3632{
3633 struct in_ifaddr *ifa = ptr;
3634 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3635 struct bonding *bond;
3636 struct vlan_entry *vlan;
c3ade5ca 3637
c346dca1 3638 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3639 return NOTIFY_DONE;
3640
0883beca 3641 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3642 if (bond->dev == event_dev) {
3643 switch (event) {
3644 case NETDEV_UP:
3645 bond->master_ip = ifa->ifa_local;
3646 return NOTIFY_OK;
3647 case NETDEV_DOWN:
3648 bond->master_ip = bond_glean_dev_ip(bond->dev);
3649 return NOTIFY_OK;
3650 default:
3651 return NOTIFY_DONE;
3652 }
3653 }
3654
0883beca 3655 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3656 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3657 if (vlan_dev == event_dev) {
3658 switch (event) {
3659 case NETDEV_UP:
3660 vlan->vlan_ip = ifa->ifa_local;
3661 return NOTIFY_OK;
3662 case NETDEV_DOWN:
3663 vlan->vlan_ip =
3664 bond_glean_dev_ip(vlan_dev);
3665 return NOTIFY_OK;
3666 default:
3667 return NOTIFY_DONE;
3668 }
3669 }
3670 }
3671 }
3672 return NOTIFY_DONE;
3673}
3674
1da177e4
LT		 3675static struct notifier_block bond_netdev_notifier = {
3676 .notifier_call = bond_netdev_event,
3677};
3678
c3ade5ca
JV		 3679static struct notifier_block bond_inetaddr_notifier = {
3680 .notifier_call = bond_inetaddr_event,
3681};
3682
1da177e4
LT		 3683/*-------------------------- Packet type handling ---------------------------*/
3684
3685/* register to receive lacpdus on a bond */
3686static void bond_register_lacpdu(struct bonding *bond)
3687{
3688 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3689
3690 /* initialize packet type */
3691 pk_type->type = PKT_TYPE_LACPDU;
3692 pk_type->dev = bond->dev;
3693 pk_type->func = bond_3ad_lacpdu_recv;
3694
3695 dev_add_pack(pk_type);
3696}
3697
3698/* unregister to receive lacpdus on a bond */
3699static void bond_unregister_lacpdu(struct bonding *bond)
3700{
3701 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3702}
3703
f5b2b966
JV		 3704void bond_register_arp(struct bonding *bond)
3705{
3706 struct packet_type *pt = &bond->arp_mon_pt;
3707
c4f283b1
JV		 3708 if (pt->type)
3709 return;
3710
f5b2b966 3711 pt->type = htons(ETH_P_ARP);
e245cb71 3712 pt->dev = bond->dev;
f5b2b966
JV		 3713 pt->func = bond_arp_rcv;
3714 dev_add_pack(pt);
3715}
3716
3717void bond_unregister_arp(struct bonding *bond)
3718{
c4f283b1
JV		 3719 struct packet_type *pt = &bond->arp_mon_pt;
3720
3721 dev_remove_pack(pt);
3722 pt->type = 0;
f5b2b966
JV		 3723}
3724
169a3e66
JV		 3725/*---------------------------- Hashing Policies -----------------------------*/
3726
6f6652be
JV		 3727/*
3728 * Hash for the output device based upon layer 2 and layer 3 data. If
3729 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3730 */
3731static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3732 struct net_device *bond_dev, int count)
3733{
3734 struct ethhdr *data = (struct ethhdr *)skb->data;
3735 struct iphdr *iph = ip_hdr(skb);
3736
f14c4e4e 3737 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be
JV		 3738 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3739 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3740 }
3741
3742 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3743}
3744
169a3e66 3745/*
59c51591 3746 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3747 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3748 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3749 */
3750static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3751 struct net_device *bond_dev, int count)
3752{
3753 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3754 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3755 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3756 int layer4_xor = 0;
3757
f14c4e4e
BH		 3758 if (skb->protocol == htons(ETH_P_IP)) {
3759 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3760 (iph->protocol == IPPROTO_TCP ||
3761 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3762 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3763 }
3764 return (layer4_xor ^
3765 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3766
3767 }
3768
3769 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3770}
3771
3772/*
3773 * Hash for the output device based upon layer 2 data
3774 */
3775static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3776 struct net_device *bond_dev, int count)
3777{
3778 struct ethhdr *data = (struct ethhdr *)skb->data;
3779
3780 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3781}
3782
1da177e4
LT		 3783/*-------------------------- Device entry points ----------------------------*/
3784
3785static int bond_open(struct net_device *bond_dev)
3786{
454d7c9b 3787 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3788
3789 bond->kill_timers = 0;
3790
58402054 3791 if (bond_is_lb(bond)) {
1da177e4
LT		 3792 /* bond_alb_initialize must be called before the timer
3793 * is started.
3794 */
3795 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3796 /* something went wrong - fail the open operation */
3797 return -1;
3798 }
3799
1b76b316
JV		 3800 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3801 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3802 }
3803
3804 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3805 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3806 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3807 }
3808
3809 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3810 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3811 INIT_DELAYED_WORK(&bond->arp_work,
3812 bond_activebackup_arp_mon);
3813 else
3814 INIT_DELAYED_WORK(&bond->arp_work,
3815 bond_loadbalance_arp_mon);
3816
3817 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3818 if (bond->params.arp_validate)
3819 bond_register_arp(bond);
1da177e4
LT		 3820 }
3821
3822 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3823 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3824 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3825 /* register to receive LACPDUs */
3826 bond_register_lacpdu(bond);
fd989c83 3827 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3828 }
3829
3830 return 0;
3831}
3832
3833static int bond_close(struct net_device *bond_dev)
3834{
454d7c9b 3835 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3836
3837 if (bond->params.mode == BOND_MODE_8023AD) {
3838 /* Unregister the receive of LACPDUs */
3839 bond_unregister_lacpdu(bond);
3840 }
3841
f5b2b966
JV		 3842 if (bond->params.arp_validate)
3843 bond_unregister_arp(bond);
3844
1da177e4
LT		 3845 write_lock_bh(&bond->lock);
3846
b59f9f74 3847 bond->send_grat_arp = 0;
305d552a 3848 bond->send_unsol_na = 0;
1da177e4
LT		 3849
3850 /* signal timers not to re-arm */
3851 bond->kill_timers = 1;
3852
3853 write_unlock_bh(&bond->lock);
3854
1da177e4 3855 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3856 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3857 }
3858
3859 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3860 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3861 }
3862
3863 switch (bond->params.mode) {
3864 case BOND_MODE_8023AD:
1b76b316 3865 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3866 break;
3867 case BOND_MODE_TLB:
3868 case BOND_MODE_ALB:
1b76b316 3869 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3870 break;
3871 default:
3872 break;
3873 }
3874
1da177e4 3875
58402054 3876 if (bond_is_lb(bond)) {
1da177e4
LT		 3877 /* Must be called only after all
3878 * slaves have been released
3879 */
3880 bond_alb_deinitialize(bond);
3881 }
3882
3883 return 0;
3884}
3885
3886static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3887{
454d7c9b 3888 struct bonding *bond = netdev_priv(bond_dev);
eeda3fd6 3889 struct net_device_stats *stats = &bond->stats;
2439f9eb 3890 struct net_device_stats local_stats;
1da177e4
LT		 3891 struct slave *slave;
3892 int i;
3893
2439f9eb 3894 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3895
3896 read_lock_bh(&bond->lock);
3897
3898 bond_for_each_slave(bond, slave, i) {
eeda3fd6
SH		 3899 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3900
2439f9eb
AG		 3901 local_stats.rx_packets += sstats->rx_packets;
3902 local_stats.rx_bytes += sstats->rx_bytes;
3903 local_stats.rx_errors += sstats->rx_errors;
3904 local_stats.rx_dropped += sstats->rx_dropped;
3905
3906 local_stats.tx_packets += sstats->tx_packets;
3907 local_stats.tx_bytes += sstats->tx_bytes;
3908 local_stats.tx_errors += sstats->tx_errors;
3909 local_stats.tx_dropped += sstats->tx_dropped;
3910
3911 local_stats.multicast += sstats->multicast;
3912 local_stats.collisions += sstats->collisions;
3913
3914 local_stats.rx_length_errors += sstats->rx_length_errors;
3915 local_stats.rx_over_errors += sstats->rx_over_errors;
3916 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3917 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3918 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3919 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3920
3921 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3922 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3923 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3924 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3925 local_stats.tx_window_errors += sstats->tx_window_errors;
3926 }
3927
3928 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3929
3930 read_unlock_bh(&bond->lock);
3931
3932 return stats;
3933}
3934
3935static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3936{
3937 struct net_device *slave_dev = NULL;
3938 struct ifbond k_binfo;
3939 struct ifbond __user *u_binfo = NULL;
3940 struct ifslave k_sinfo;
3941 struct ifslave __user *u_sinfo = NULL;
3942 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3943 int res = 0;
3944
5a03cdb7 3945 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
1da177e4
LT		 3946 bond_dev->name, cmd);
3947
3948 switch (cmd) {
1da177e4
LT		 3949 case SIOCGMIIPHY:
3950 mii = if_mii(ifr);
3951 if (!mii) {
3952 return -EINVAL;
3953 }
3954 mii->phy_id = 0;
3955 /* Fall Through */
3956 case SIOCGMIIREG:
3957 /*
3958 * We do this again just in case we were called by SIOCGMIIREG
3959 * instead of SIOCGMIIPHY.
3960 */
3961 mii = if_mii(ifr);
3962 if (!mii) {
3963 return -EINVAL;
3964 }
3965
3966 if (mii->reg_num == 1) {
454d7c9b 3967 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3968 mii->val_out = 0;
6603a6f2 3969 read_lock(&bond->lock);
1da177e4 3970 read_lock(&bond->curr_slave_lock);
4e140079 3971 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3972 mii->val_out = BMSR_LSTATUS;
3973 }
3974 read_unlock(&bond->curr_slave_lock);
6603a6f2 3975 read_unlock(&bond->lock);
1da177e4
LT		 3976 }
3977
3978 return 0;
3979 case BOND_INFO_QUERY_OLD:
3980 case SIOCBONDINFOQUERY:
3981 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3982
3983 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3984 return -EFAULT;
3985 }
3986
3987 res = bond_info_query(bond_dev, &k_binfo);
3988 if (res == 0) {
3989 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3990 return -EFAULT;
3991 }
3992 }
3993
3994 return res;
3995 case BOND_SLAVE_INFO_QUERY_OLD:
3996 case SIOCBONDSLAVEINFOQUERY:
3997 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3998
3999 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4000 return -EFAULT;
4001 }
4002
4003 res = bond_slave_info_query(bond_dev, &k_sinfo);
4004 if (res == 0) {
4005 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4006 return -EFAULT;
4007 }
4008 }
4009
4010 return res;
4011 default:
4012 /* Go on */
4013 break;
4014 }
4015
4016 if (!capable(CAP_NET_ADMIN)) {
4017 return -EPERM;
4018 }
4019
b76cdba9 4020 down_write(&(bonding_rwsem));
881d966b 4021 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4 4022
5a03cdb7 4023 pr_debug("slave_dev=%p: \n", slave_dev);
1da177e4
LT		 4024
4025 if (!slave_dev) {
4026 res = -ENODEV;
4027 } else {
5a03cdb7 4028 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
1da177e4
LT		 4029 switch (cmd) {
4030 case BOND_ENSLAVE_OLD:
4031 case SIOCBONDENSLAVE:
4032 res = bond_enslave(bond_dev, slave_dev);
4033 break;
4034 case BOND_RELEASE_OLD:
4035 case SIOCBONDRELEASE:
4036 res = bond_release(bond_dev, slave_dev);
4037 break;
4038 case BOND_SETHWADDR_OLD:
4039 case SIOCBONDSETHWADDR:
4040 res = bond_sethwaddr(bond_dev, slave_dev);
4041 break;
4042 case BOND_CHANGE_ACTIVE_OLD:
4043 case SIOCBONDCHANGEACTIVE:
4044 res = bond_ioctl_change_active(bond_dev, slave_dev);
4045 break;
4046 default:
4047 res = -EOPNOTSUPP;
4048 }
4049
4050 dev_put(slave_dev);
4051 }
4052
b76cdba9 4053 up_write(&(bonding_rwsem));
1da177e4
LT		 4054 return res;
4055}
4056
4057static void bond_set_multicast_list(struct net_device *bond_dev)
4058{
454d7c9b 4059 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4060 struct dev_mc_list *dmi;
4061
1da177e4
LT		 4062 /*
4063 * Do promisc before checking multicast_mode
4064 */
4065 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4066 /*
4067 * FIXME: Need to handle the error when one of the multi-slaves
4068 * encounters error.
4069 */
1da177e4
LT		 4070 bond_set_promiscuity(bond, 1);
4071 }
4072
4073 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4074 bond_set_promiscuity(bond, -1);
4075 }
4076
4077 /* set allmulti flag to slaves */
4078 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4079 /*
4080 * FIXME: Need to handle the error when one of the multi-slaves
4081 * encounters error.
4082 */
1da177e4
LT		 4083 bond_set_allmulti(bond, 1);
4084 }
4085
4086 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4087 bond_set_allmulti(bond, -1);
4088 }
4089
80ee5ad2
JV		 4090 read_lock(&bond->lock);
4091
1da177e4
LT		 4092 bond->flags = bond_dev->flags;
4093
4094 /* looking for addresses to add to slaves' mc list */
4095 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4096 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4097 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4098 }
4099 }
4100
4101 /* looking for addresses to delete from slaves' list */
4102 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4103 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4104 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4105 }
4106 }
4107
4108 /* save master's multicast list */
4109 bond_mc_list_destroy(bond);
4110 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4111
80ee5ad2 4112 read_unlock(&bond->lock);
1da177e4
LT		 4113}
4114
00829823
SH		 4115static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4116{
4117 struct bonding *bond = netdev_priv(dev);
4118 struct slave *slave = bond->first_slave;
4119
4120 if (slave) {
4121 const struct net_device_ops *slave_ops
4122 = slave->dev->netdev_ops;
4123 if (slave_ops->ndo_neigh_setup)
4124 return slave_ops->ndo_neigh_setup(dev, parms);
4125 }
4126 return 0;
4127}
4128
1da177e4
LT		 4129/*
4130 * Change the MTU of all of a master's slaves to match the master
4131 */
4132static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4133{
454d7c9b 4134 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4135 struct slave *slave, *stop_at;
4136 int res = 0;
4137 int i;
4138
5a03cdb7 4139 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
1da177e4
LT		 4140 (bond_dev ? bond_dev->name : "None"), new_mtu);
4141
4142 /* Can't hold bond->lock with bh disabled here since
4143 * some base drivers panic. On the other hand we can't
4144 * hold bond->lock without bh disabled because we'll
4145 * deadlock. The only solution is to rely on the fact
4146 * that we're under rtnl_lock here, and the slaves
4147 * list won't change. This doesn't solve the problem
4148 * of setting the slave's MTU while it is
4149 * transmitting, but the assumption is that the base
4150 * driver can handle that.
4151 *
4152 * TODO: figure out a way to safely iterate the slaves
4153 * list, but without holding a lock around the actual
4154 * call to the base driver.
4155 */
4156
4157 bond_for_each_slave(bond, slave, i) {
5a03cdb7 4158 pr_debug("s %p s->p %p c_m %p\n", slave,
53a3294e 4159 slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
e944ef79 4160
1da177e4
LT		 4161 res = dev_set_mtu(slave->dev, new_mtu);
4162
4163 if (res) {
4164 /* If we failed to set the slave's mtu to the new value
4165 * we must abort the operation even in ACTIVE_BACKUP
4166 * mode, because if we allow the backup slaves to have
4167 * different mtu values than the active slave we'll
4168 * need to change their mtu when doing a failover. That
4169 * means changing their mtu from timer context, which
4170 * is probably not a good idea.
4171 */
5a03cdb7 4172 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 4173 goto unwind;
4174 }
4175 }
4176
4177 bond_dev->mtu = new_mtu;
4178
4179 return 0;
4180
4181unwind:
4182 /* unwind from head to the slave that failed */
4183 stop_at = slave;
4184 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4185 int tmp_res;
4186
4187 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4188 if (tmp_res) {
5a03cdb7 4189 pr_debug("unwind err %d dev %s\n", tmp_res,
1da177e4
LT		 4190 slave->dev->name);
4191 }
4192 }
4193
4194 return res;
4195}
4196
4197/*
4198 * Change HW address
4199 *
4200 * Note that many devices must be down to change the HW address, and
4201 * downing the master releases all slaves. We can make bonds full of
4202 * bonding devices to test this, however.
4203 */
4204static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4205{
454d7c9b 4206 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4207 struct sockaddr *sa = addr, tmp_sa;
4208 struct slave *slave, *stop_at;
4209 int res = 0;
4210 int i;
4211
eb7cc59a
SH		 4212 if (bond->params.mode == BOND_MODE_ALB)
4213 return bond_alb_set_mac_address(bond_dev, addr);
4214
4215
5a03cdb7 4216 pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
1da177e4 4217
dd957c57 4218 /*
3915c1e8
JV		 4219 * If fail_over_mac is set to active, do nothing and return
4220 * success. Returning an error causes ifenslave to fail.
dd957c57 4221 */
3915c1e8 4222 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4223 return 0;
2ab82852 4224
1da177e4
LT		 4225 if (!is_valid_ether_addr(sa->sa_data)) {
4226 return -EADDRNOTAVAIL;
4227 }
4228
4229 /* Can't hold bond->lock with bh disabled here since
4230 * some base drivers panic. On the other hand we can't
4231 * hold bond->lock without bh disabled because we'll
4232 * deadlock. The only solution is to rely on the fact
4233 * that we're under rtnl_lock here, and the slaves
4234 * list won't change. This doesn't solve the problem
4235 * of setting the slave's hw address while it is
4236 * transmitting, but the assumption is that the base
4237 * driver can handle that.
4238 *
4239 * TODO: figure out a way to safely iterate the slaves
4240 * list, but without holding a lock around the actual
4241 * call to the base driver.
4242 */
4243
4244 bond_for_each_slave(bond, slave, i) {
eb7cc59a 4245 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
5a03cdb7 4246 pr_debug("slave %p %s\n", slave, slave->dev->name);
1da177e4 4247
eb7cc59a 4248 if (slave_ops->ndo_set_mac_address == NULL) {
1da177e4 4249 res = -EOPNOTSUPP;
5a03cdb7 4250 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
1da177e4
LT		 4251 goto unwind;
4252 }
4253
4254 res = dev_set_mac_address(slave->dev, addr);
4255 if (res) {
4256 /* TODO: consider downing the slave
4257 * and retry ?
4258 * User should expect communications
4259 * breakage anyway until ARP finish
4260 * updating, so...
4261 */
5a03cdb7 4262 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 4263 goto unwind;
4264 }
4265 }
4266
4267 /* success */
4268 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4269 return 0;
4270
4271unwind:
4272 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4273 tmp_sa.sa_family = bond_dev->type;
4274
4275 /* unwind from head to the slave that failed */
4276 stop_at = slave;
4277 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4278 int tmp_res;
4279
4280 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4281 if (tmp_res) {
5a03cdb7 4282 pr_debug("unwind err %d dev %s\n", tmp_res,
1da177e4
LT		 4283 slave->dev->name);
4284 }
4285 }
4286
4287 return res;
4288}
4289
4290static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4291{
454d7c9b 4292 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4293 struct slave *slave, *start_at;
cf5f9044 4294 int i, slave_no, res = 1;
1da177e4
LT		 4295
4296 read_lock(&bond->lock);
4297
4298 if (!BOND_IS_OK(bond)) {
4299 goto out;
4300 }
4301
cf5f9044
JV		 4302 /*
4303 * Concurrent TX may collide on rr_tx_counter; we accept that
4304 * as being rare enough not to justify using an atomic op here
4305 */
4306 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4307
cf5f9044
JV		 4308 bond_for_each_slave(bond, slave, i) {
4309 slave_no--;
4310 if (slave_no < 0) {
4311 break;
4312 }
1da177e4
LT		 4313 }
4314
cf5f9044 4315 start_at = slave;
1da177e4
LT		 4316 bond_for_each_slave_from(bond, slave, i, start_at) {
4317 if (IS_UP(slave->dev) &&
4318 (slave->link == BOND_LINK_UP) &&
4319 (slave->state == BOND_STATE_ACTIVE)) {
4320 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4321 break;
4322 }
4323 }
4324
1da177e4
LT		 4325out:
4326 if (res) {
4327 /* no suitable interface, frame not sent */
4328 dev_kfree_skb(skb);
4329 }
4330 read_unlock(&bond->lock);
4331 return 0;
4332}
4333
075897ce 4334
1da177e4
LT		 4335/*
4336 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4337 * the bond has a usable interface.
4338 */
4339static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4340{
454d7c9b 4341 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4342 int res = 1;
4343
1da177e4
LT		 4344 read_lock(&bond->lock);
4345 read_lock(&bond->curr_slave_lock);
4346
4347 if (!BOND_IS_OK(bond)) {
4348 goto out;
4349 }
4350
075897ce
JL		 4351 if (!bond->curr_active_slave)
4352 goto out;
4353
075897ce
JL		 4354 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4355
1da177e4
LT		 4356out:
4357 if (res) {
4358 /* no suitable interface, frame not sent */
4359 dev_kfree_skb(skb);
4360 }
4361 read_unlock(&bond->curr_slave_lock);
4362 read_unlock(&bond->lock);
4363 return 0;
4364}
4365
4366/*
169a3e66
JV		 4367 * In bond_xmit_xor() , we determine the output device by using a pre-
4368 * determined xmit_hash_policy(), If the selected device is not enabled,
4369 * find the next active slave.
1da177e4
LT		 4370 */
4371static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4372{
454d7c9b 4373 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4374 struct slave *slave, *start_at;
4375 int slave_no;
4376 int i;
4377 int res = 1;
4378
4379 read_lock(&bond->lock);
4380
4381 if (!BOND_IS_OK(bond)) {
4382 goto out;
4383 }
4384
169a3e66 4385 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4386
4387 bond_for_each_slave(bond, slave, i) {
4388 slave_no--;
4389 if (slave_no < 0) {
4390 break;
4391 }
4392 }
4393
4394 start_at = slave;
4395
4396 bond_for_each_slave_from(bond, slave, i, start_at) {
4397 if (IS_UP(slave->dev) &&
4398 (slave->link == BOND_LINK_UP) &&
4399 (slave->state == BOND_STATE_ACTIVE)) {
4400 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4401 break;
4402 }
4403 }
4404
4405out:
4406 if (res) {
4407 /* no suitable interface, frame not sent */
4408 dev_kfree_skb(skb);
4409 }
4410 read_unlock(&bond->lock);
4411 return 0;
4412}
4413
4414/*
4415 * in broadcast mode, we send everything to all usable interfaces.
4416 */
4417static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4418{
454d7c9b 4419 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4420 struct slave *slave, *start_at;
4421 struct net_device *tx_dev = NULL;
4422 int i;
4423 int res = 1;
4424
4425 read_lock(&bond->lock);
4426
4427 if (!BOND_IS_OK(bond)) {
4428 goto out;
4429 }
4430
4431 read_lock(&bond->curr_slave_lock);
4432 start_at = bond->curr_active_slave;
4433 read_unlock(&bond->curr_slave_lock);
4434
4435 if (!start_at) {
4436 goto out;
4437 }
4438
4439 bond_for_each_slave_from(bond, slave, i, start_at) {
4440 if (IS_UP(slave->dev) &&
4441 (slave->link == BOND_LINK_UP) &&
4442 (slave->state == BOND_STATE_ACTIVE)) {
4443 if (tx_dev) {
4444 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4445 if (!skb2) {
4446 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4447 ": %s: Error: bond_xmit_broadcast(): "
4448 "skb_clone() failed\n",
4449 bond_dev->name);
1da177e4
LT		 4450 continue;
4451 }
4452
4453 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4454 if (res) {
4455 dev_kfree_skb(skb2);
4456 continue;
4457 }
4458 }
4459 tx_dev = slave->dev;
4460 }
4461 }
4462
4463 if (tx_dev) {
4464 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4465 }
4466
4467out:
4468 if (res) {
4469 /* no suitable interface, frame not sent */
4470 dev_kfree_skb(skb);
4471 }
4472 /* frame sent to all suitable interfaces */
4473 read_unlock(&bond->lock);
4474 return 0;
4475}
4476
4477/*------------------------- Device initialization ---------------------------*/
4478
6f6652be
JV		 4479static void bond_set_xmit_hash_policy(struct bonding *bond)
4480{
4481 switch (bond->params.xmit_policy) {
4482 case BOND_XMIT_POLICY_LAYER23:
4483 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4484 break;
4485 case BOND_XMIT_POLICY_LAYER34:
4486 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4487 break;
4488 case BOND_XMIT_POLICY_LAYER2:
4489 default:
4490 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4491 break;
4492 }
4493}
4494
00829823
SH		 4495static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4496{
4497 const struct bonding *bond = netdev_priv(dev);
4498
4499 switch (bond->params.mode) {
4500 case BOND_MODE_ROUNDROBIN:
4501 return bond_xmit_roundrobin(skb, dev);
4502 case BOND_MODE_ACTIVEBACKUP:
4503 return bond_xmit_activebackup(skb, dev);
4504 case BOND_MODE_XOR:
4505 return bond_xmit_xor(skb, dev);
4506 case BOND_MODE_BROADCAST:
4507 return bond_xmit_broadcast(skb, dev);
4508 case BOND_MODE_8023AD:
4509 return bond_3ad_xmit_xor(skb, dev);
4510 case BOND_MODE_ALB:
4511 case BOND_MODE_TLB:
4512 return bond_alb_xmit(skb, dev);
4513 default:
4514 /* Should never happen, mode already checked */
4515 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4516 dev->name, bond->params.mode);
4517 WARN_ON_ONCE(1);
4518 dev_kfree_skb(skb);
4519 return NETDEV_TX_OK;
4520 }
4521}
4522
4523
1da177e4
LT		 4524/*
4525 * set bond mode specific net device operations
4526 */
a77b5325 4527void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4528{
169a3e66
JV		 4529 struct net_device *bond_dev = bond->dev;
4530
1da177e4
LT		 4531 switch (mode) {
4532 case BOND_MODE_ROUNDROBIN:
1da177e4
LT		 4533 break;
4534 case BOND_MODE_ACTIVEBACKUP:
1da177e4
LT		 4535 break;
4536 case BOND_MODE_XOR:
6f6652be 4537 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4538 break;
4539 case BOND_MODE_BROADCAST:
1da177e4
LT		 4540 break;
4541 case BOND_MODE_8023AD:
8f903c70 4542 bond_set_master_3ad_flags(bond);
6f6652be 4543 bond_set_xmit_hash_policy(bond);
1da177e4 4544 break;
1da177e4 4545 case BOND_MODE_ALB:
8f903c70
JV		 4546 bond_set_master_alb_flags(bond);
4547 /* FALLTHRU */
4548 case BOND_MODE_TLB:
1da177e4
LT		 4549 break;
4550 default:
4551 /* Should never happen, mode already checked */
4552 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4553 ": %s: Error: Unknown bonding mode %d\n",
4554 bond_dev->name,
1da177e4
LT		 4555 mode);
4556 break;
4557 }
4558}
4559
217df670
JV		 4560static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4561 struct ethtool_drvinfo *drvinfo)
4562{
4563 strncpy(drvinfo->driver, DRV_NAME, 32);
4564 strncpy(drvinfo->version, DRV_VERSION, 32);
4565 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4566}
4567
7282d491 4568static const struct ethtool_ops bond_ethtool_ops = {
217df670 4569 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac
SH		 4570 .get_link = ethtool_op_get_link,
4571 .get_tx_csum = ethtool_op_get_tx_csum,
4572 .get_sg = ethtool_op_get_sg,
4573 .get_tso = ethtool_op_get_tso,
4574 .get_ufo = ethtool_op_get_ufo,
4575 .get_flags = ethtool_op_get_flags,
8531c5ff
AK		 4576};
4577
eb7cc59a
SH		 4578static const struct net_device_ops bond_netdev_ops = {
4579 .ndo_open = bond_open,
4580 .ndo_stop = bond_close,
00829823 4581 .ndo_start_xmit = bond_start_xmit,
eb7cc59a
SH		 4582 .ndo_get_stats = bond_get_stats,
4583 .ndo_do_ioctl = bond_do_ioctl,
4584 .ndo_set_multicast_list = bond_set_multicast_list,
4585 .ndo_change_mtu = bond_change_mtu,
eb7cc59a 4586 .ndo_set_mac_address = bond_set_mac_address,
00829823 4587 .ndo_neigh_setup = bond_neigh_setup,
eb7cc59a
SH		 4588 .ndo_vlan_rx_register = bond_vlan_rx_register,
4589 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4590 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4591};
4592
1da177e4
LT		 4593/*
4594 * Does not allocate but creates a /proc entry.
4595 * Allowed to fail.
4596 */
3c535952 4597static int bond_init(struct net_device *bond_dev, struct bond_params *params)
1da177e4 4598{
454d7c9b 4599 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4600
5a03cdb7 4601 pr_debug("Begin bond_init for %s\n", bond_dev->name);
1da177e4
LT		 4602
4603 /* initialize rwlocks */
4604 rwlock_init(&bond->lock);
4605 rwlock_init(&bond->curr_slave_lock);
4606
4607 bond->params = *params; /* copy params struct */
4608
1b76b316
JV		 4609 bond->wq = create_singlethread_workqueue(bond_dev->name);
4610 if (!bond->wq)
4611 return -ENOMEM;
4612
1da177e4
LT		 4613 /* Initialize pointers */
4614 bond->first_slave = NULL;
4615 bond->curr_active_slave = NULL;
4616 bond->current_arp_slave = NULL;
4617 bond->primary_slave = NULL;
4618 bond->dev = bond_dev;
1053f62c 4619 bond->send_grat_arp = 0;
305d552a 4620 bond->send_unsol_na = 0;
d90a162a 4621 bond->setup_by_slave = 0;
1da177e4
LT		 4622 INIT_LIST_HEAD(&bond->vlan_list);
4623
4624 /* Initialize the device entry points */
eb7cc59a 4625 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 4626 bond_dev->ethtool_ops = &bond_ethtool_ops;
169a3e66 4627 bond_set_mode_ops(bond, bond->params.mode);
1da177e4 4628
a434e43f 4629 bond_dev->destructor = bond_destructor;
1da177e4
LT		 4630
4631 /* Initialize the device options */
4632 bond_dev->tx_queue_len = 0;
4633 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4634 bond_dev->priv_flags |= IFF_BONDING;
6cf3f41e
JV		 4635 if (bond->params.arp_interval)
4636 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
1da177e4
LT		 4637
4638 /* At first, we block adding VLANs. That's the only way to
4639 * prevent problems that occur when adding VLANs over an
4640 * empty bond. The block will be removed once non-challenged
4641 * slaves are enslaved.
4642 */
4643 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4644
932ff279 4645 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4646 * transmitting */
4647 bond_dev->features |= NETIF_F_LLTX;
4648
4649 /* By default, we declare the bond to be fully
4650 * VLAN hardware accelerated capable. Special
4651 * care is taken in the various xmit functions
4652 * when there are slaves that are not hw accel
4653 * capable
4654 */
1da177e4
LT		 4655 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4656 NETIF_F_HW_VLAN_RX |
4657 NETIF_F_HW_VLAN_FILTER);
4658
4659#ifdef CONFIG_PROC_FS
4660 bond_create_proc_entry(bond);
4661#endif
1da177e4
LT		 4662 list_add_tail(&bond->bond_list, &bond_dev_list);
4663
4664 return 0;
4665}
4666
fdaea7a9
JV		 4667static void bond_work_cancel_all(struct bonding *bond)
4668{
4669 write_lock_bh(&bond->lock);
4670 bond->kill_timers = 1;
4671 write_unlock_bh(&bond->lock);
4672
4673 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4674 cancel_delayed_work(&bond->mii_work);
4675
4676 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4677 cancel_delayed_work(&bond->arp_work);
4678
4679 if (bond->params.mode == BOND_MODE_ALB &&
4680 delayed_work_pending(&bond->alb_work))
4681 cancel_delayed_work(&bond->alb_work);
4682
4683 if (bond->params.mode == BOND_MODE_8023AD &&
4684 delayed_work_pending(&bond->ad_work))
4685 cancel_delayed_work(&bond->ad_work);
4686}
4687
a434e43f
JV		 4688/* De-initialize device specific data.
4689 * Caller must hold rtnl_lock.
4690 */
4691static void bond_deinit(struct net_device *bond_dev)
4692{
454d7c9b 4693 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 4694
4695 list_del(&bond->bond_list);
4696
4697 bond_work_cancel_all(bond);
4698
4699#ifdef CONFIG_PROC_FS
4700 bond_remove_proc_entry(bond);
4701#endif
4702}
4703
1da177e4
LT		 4704/* Unregister and free all bond devices.
4705 * Caller must hold rtnl_lock.
4706 */
4707static void bond_free_all(void)
4708{
4709 struct bonding *bond, *nxt;
4710
4711 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4712 struct net_device *bond_dev = bond->dev;
4713
fdaea7a9 4714 bond_work_cancel_all(bond);
70298705 4715 /* Release the bonded slaves */
4716 bond_release_all(bond_dev);
92b41daa 4717 bond_destroy(bond);
1da177e4
LT		 4718 }
4719
4720#ifdef CONFIG_PROC_FS
4721 bond_destroy_proc_dir();
4722#endif
4723}
4724
4725/*------------------------- Module initialization ---------------------------*/
4726
4727/*
4728 * Convert string input module parms. Accept either the
ece95f7f
JV		 4729 * number of the mode or its string name. A bit complicated because
4730 * some mode names are substrings of other names, and calls from sysfs
4731 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4732 */
325dcf7a 4733int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
1da177e4 4734{
ece95f7f 4735 int mode = -1, i, rv;
a42e534f 4736 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4737
a42e534f
JV		 4738 for (p = (char *)buf; *p; p++)
4739 if (!(isdigit(*p) || isspace(*p)))
4740 break;
4741
4742 if (*p)
ece95f7f 4743 rv = sscanf(buf, "%20s", modestr);
a42e534f
JV		 4744 else
4745 rv = sscanf(buf, "%d", &mode);
4746
4747 if (!rv)
4748 return -1;
1da177e4
LT		 4749
4750 for (i = 0; tbl[i].modename; i++) {
ece95f7f
JV		 4751 if (mode == tbl[i].mode)
4752 return tbl[i].mode;
4753 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4754 return tbl[i].mode;
1da177e4
LT		 4755 }
4756
4757 return -1;
4758}
4759
4760static int bond_check_params(struct bond_params *params)
4761{
3915c1e8 4762 int arp_validate_value, fail_over_mac_value;
f5b2b966 4763
1da177e4
LT		 4764 /*
4765 * Convert string parameters.
4766 */
4767 if (mode) {
4768 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4769 if (bond_mode == -1) {
4770 printk(KERN_ERR DRV_NAME
4771 ": Error: Invalid bonding mode \"%s\"\n",
4772 mode == NULL ? "NULL" : mode);
4773 return -EINVAL;
4774 }
4775 }
4776
169a3e66
JV		 4777 if (xmit_hash_policy) {
4778 if ((bond_mode != BOND_MODE_XOR) &&
4779 (bond_mode != BOND_MODE_8023AD)) {
4780 printk(KERN_INFO DRV_NAME
4781 ": xor_mode param is irrelevant in mode %s\n",
4782 bond_mode_name(bond_mode));
4783 } else {
4784 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4785 xmit_hashtype_tbl);
4786 if (xmit_hashtype == -1) {
4787 printk(KERN_ERR DRV_NAME
4788 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4789 xmit_hash_policy == NULL ? "NULL" :
4790 xmit_hash_policy);
4791 return -EINVAL;
4792 }
4793 }
4794 }
4795
1da177e4
LT		 4796 if (lacp_rate) {
4797 if (bond_mode != BOND_MODE_8023AD) {
4798 printk(KERN_INFO DRV_NAME
4799 ": lacp_rate param is irrelevant in mode %s\n",
4800 bond_mode_name(bond_mode));
4801 } else {
4802 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4803 if (lacp_fast == -1) {
4804 printk(KERN_ERR DRV_NAME
4805 ": Error: Invalid lacp rate \"%s\"\n",
4806 lacp_rate == NULL ? "NULL" : lacp_rate);
4807 return -EINVAL;
4808 }
4809 }
4810 }
4811
fd989c83
JV		 4812 if (ad_select) {
4813 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4814 if (params->ad_select == -1) {
4815 printk(KERN_ERR DRV_NAME
4816 ": Error: Invalid ad_select \"%s\"\n",
4817 ad_select == NULL ? "NULL" : ad_select);
4818 return -EINVAL;
4819 }
4820
4821 if (bond_mode != BOND_MODE_8023AD) {
4822 printk(KERN_WARNING DRV_NAME
4823 ": ad_select param only affects 802.3ad mode\n");
4824 }
4825 } else {
4826 params->ad_select = BOND_AD_STABLE;
4827 }
4828
b8a9787e 4829 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4830 printk(KERN_WARNING DRV_NAME
4831 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4832 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4833 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4834 max_bonds = BOND_DEFAULT_MAX_BONDS;
4835 }
4836
4837 if (miimon < 0) {
4838 printk(KERN_WARNING DRV_NAME
4839 ": Warning: miimon module parameter (%d), "
4840 "not in range 0-%d, so it was reset to %d\n",
4841 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4842 miimon = BOND_LINK_MON_INTERV;
4843 }
4844
4845 if (updelay < 0) {
4846 printk(KERN_WARNING DRV_NAME
4847 ": Warning: updelay module parameter (%d), "
4848 "not in range 0-%d, so it was reset to 0\n",
4849 updelay, INT_MAX);
4850 updelay = 0;
4851 }
4852
4853 if (downdelay < 0) {
4854 printk(KERN_WARNING DRV_NAME
4855 ": Warning: downdelay module parameter (%d), "
4856 "not in range 0-%d, so it was reset to 0\n",
4857 downdelay, INT_MAX);
4858 downdelay = 0;
4859 }
4860
4861 if ((use_carrier != 0) && (use_carrier != 1)) {
4862 printk(KERN_WARNING DRV_NAME
4863 ": Warning: use_carrier module parameter (%d), "
4864 "not of valid value (0/1), so it was set to 1\n",
4865 use_carrier);
4866 use_carrier = 1;
4867 }
4868
7893b249
MS		 4869 if (num_grat_arp < 0 || num_grat_arp > 255) {
4870 printk(KERN_WARNING DRV_NAME
4871 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4872 "was reset to 1 \n", num_grat_arp);
4873 num_grat_arp = 1;
4874 }
4875
305d552a
BH		 4876 if (num_unsol_na < 0 || num_unsol_na > 255) {
4877 printk(KERN_WARNING DRV_NAME
4878 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4879 "was reset to 1 \n", num_unsol_na);
4880 num_unsol_na = 1;
4881 }
4882
1da177e4
LT		 4883 /* reset values for 802.3ad */
4884 if (bond_mode == BOND_MODE_8023AD) {
4885 if (!miimon) {
4886 printk(KERN_WARNING DRV_NAME
4887 ": Warning: miimon must be specified, "
4888 "otherwise bonding will not detect link "
4889 "failure, speed and duplex which are "
4890 "essential for 802.3ad operation\n");
4891 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4892 miimon = 100;
4893 }
4894 }
4895
4896 /* reset values for TLB/ALB */
4897 if ((bond_mode == BOND_MODE_TLB) ||
4898 (bond_mode == BOND_MODE_ALB)) {
4899 if (!miimon) {
4900 printk(KERN_WARNING DRV_NAME
4901 ": Warning: miimon must be specified, "
4902 "otherwise bonding will not detect link "
4903 "failure and link speed which are essential "
4904 "for TLB/ALB load balancing\n");
4905 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4906 miimon = 100;
4907 }
4908 }
4909
4910 if (bond_mode == BOND_MODE_ALB) {
4911 printk(KERN_NOTICE DRV_NAME
4912 ": In ALB mode you might experience client "
4913 "disconnections upon reconnection of a link if the "
4914 "bonding module updelay parameter (%d msec) is "
4915 "incompatible with the forwarding delay time of the "
4916 "switch\n",
4917 updelay);
4918 }
4919
4920 if (!miimon) {
4921 if (updelay || downdelay) {
4922 /* just warn the user the up/down delay will have
4923 * no effect since miimon is zero...
4924 */
4925 printk(KERN_WARNING DRV_NAME
4926 ": Warning: miimon module parameter not set "
4927 "and updelay (%d) or downdelay (%d) module "
4928 "parameter is set; updelay and downdelay have "
4929 "no effect unless miimon is set\n",
4930 updelay, downdelay);
4931 }
4932 } else {
4933 /* don't allow arp monitoring */
4934 if (arp_interval) {
4935 printk(KERN_WARNING DRV_NAME
4936 ": Warning: miimon (%d) and arp_interval (%d) "
4937 "can't be used simultaneously, disabling ARP "
4938 "monitoring\n",
4939 miimon, arp_interval);
4940 arp_interval = 0;
4941 }
4942
4943 if ((updelay % miimon) != 0) {
4944 printk(KERN_WARNING DRV_NAME
4945 ": Warning: updelay (%d) is not a multiple "
4946 "of miimon (%d), updelay rounded to %d ms\n",
4947 updelay, miimon, (updelay / miimon) * miimon);
4948 }
4949
4950 updelay /= miimon;
4951
4952 if ((downdelay % miimon) != 0) {
4953 printk(KERN_WARNING DRV_NAME
4954 ": Warning: downdelay (%d) is not a multiple "
4955 "of miimon (%d), downdelay rounded to %d ms\n",
4956 downdelay, miimon,
4957 (downdelay / miimon) * miimon);
4958 }
4959
4960 downdelay /= miimon;
4961 }
4962
4963 if (arp_interval < 0) {
4964 printk(KERN_WARNING DRV_NAME
4965 ": Warning: arp_interval module parameter (%d) "
4966 ", not in range 0-%d, so it was reset to %d\n",
4967 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4968 arp_interval = BOND_LINK_ARP_INTERV;
4969 }
4970
4971 for (arp_ip_count = 0;
4972 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4973 arp_ip_count++) {
4974 /* not complete check, but should be good enough to
4975 catch mistakes */
4976 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4977 printk(KERN_WARNING DRV_NAME
4978 ": Warning: bad arp_ip_target module parameter "
4979 "(%s), ARP monitoring will not be performed\n",
4980 arp_ip_target[arp_ip_count]);
4981 arp_interval = 0;
4982 } else {
d3bb52b0 4983 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4984 arp_target[arp_ip_count] = ip;
4985 }
4986 }
4987
4988 if (arp_interval && !arp_ip_count) {
4989 /* don't allow arping if no arp_ip_target given... */
4990 printk(KERN_WARNING DRV_NAME
4991 ": Warning: arp_interval module parameter (%d) "
4992 "specified without providing an arp_ip_target "
4993 "parameter, arp_interval was reset to 0\n",
4994 arp_interval);
4995 arp_interval = 0;
4996 }
4997
f5b2b966
JV		 4998 if (arp_validate) {
4999 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5000 printk(KERN_ERR DRV_NAME
5001 ": arp_validate only supported in active-backup mode\n");
5002 return -EINVAL;
5003 }
5004 if (!arp_interval) {
5005 printk(KERN_ERR DRV_NAME
5006 ": arp_validate requires arp_interval\n");
5007 return -EINVAL;
5008 }
5009
5010 arp_validate_value = bond_parse_parm(arp_validate,
5011 arp_validate_tbl);
5012 if (arp_validate_value == -1) {
5013 printk(KERN_ERR DRV_NAME
5014 ": Error: invalid arp_validate \"%s\"\n",
5015 arp_validate == NULL ? "NULL" : arp_validate);
5016 return -EINVAL;
5017 }
5018 } else
5019 arp_validate_value = 0;
5020
1da177e4
LT		 5021 if (miimon) {
5022 printk(KERN_INFO DRV_NAME
5023 ": MII link monitoring set to %d ms\n",
5024 miimon);
5025 } else if (arp_interval) {
5026 int i;
5027
5028 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 5029 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5030 arp_interval,
5031 arp_validate_tbl[arp_validate_value].modename,
5032 arp_ip_count);
1da177e4
LT		 5033
5034 for (i = 0; i < arp_ip_count; i++)
5035 printk (" %s", arp_ip_target[i]);
5036
5037 printk("\n");
5038
b8a9787e 5039 } else if (max_bonds) {
1da177e4
LT		 5040 /* miimon and arp_interval not set, we need one so things
5041 * work as expected, see bonding.txt for details
5042 */
5043 printk(KERN_WARNING DRV_NAME
5044 ": Warning: either miimon or arp_interval and "
5045 "arp_ip_target module parameters must be specified, "
5046 "otherwise bonding will not detect link failures! see "
5047 "bonding.txt for details.\n");
5048 }
5049
5050 if (primary && !USES_PRIMARY(bond_mode)) {
5051 /* currently, using a primary only makes sense
5052 * in active backup, TLB or ALB modes
5053 */
5054 printk(KERN_WARNING DRV_NAME
5055 ": Warning: %s primary device specified but has no "
5056 "effect in %s mode\n",
5057 primary, bond_mode_name(bond_mode));
5058 primary = NULL;
5059 }
5060
3915c1e8
JV		 5061 if (fail_over_mac) {
5062 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5063 fail_over_mac_tbl);
5064 if (fail_over_mac_value == -1) {
5065 printk(KERN_ERR DRV_NAME
5066 ": Error: invalid fail_over_mac \"%s\"\n",
5067 arp_validate == NULL ? "NULL" : arp_validate);
5068 return -EINVAL;
5069 }
5070
5071 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5072 printk(KERN_WARNING DRV_NAME
5073 ": Warning: fail_over_mac only affects "
5074 "active-backup mode.\n");
5075 } else {
5076 fail_over_mac_value = BOND_FOM_NONE;
5077 }
dd957c57 5078
1da177e4
LT		 5079 /* fill params struct with the proper values */
5080 params->mode = bond_mode;
169a3e66 5081 params->xmit_policy = xmit_hashtype;
1da177e4 5082 params->miimon = miimon;
7893b249 5083 params->num_grat_arp = num_grat_arp;
305d552a 5084 params->num_unsol_na = num_unsol_na;
1da177e4 5085 params->arp_interval = arp_interval;
f5b2b966 5086 params->arp_validate = arp_validate_value;
1da177e4
LT		 5087 params->updelay = updelay;
5088 params->downdelay = downdelay;
5089 params->use_carrier = use_carrier;
5090 params->lacp_fast = lacp_fast;
5091 params->primary[0] = 0;
3915c1e8 5092 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5093
5094 if (primary) {
5095 strncpy(params->primary, primary, IFNAMSIZ);
5096 params->primary[IFNAMSIZ - 1] = 0;
5097 }
5098
5099 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5100
5101 return 0;
5102}
5103
0daa2303 5104static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 5105static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 5106
e8a0464c
DM		 5107static void bond_set_lockdep_class_one(struct net_device *dev,
5108 struct netdev_queue *txq,
5109 void *_unused)
c773e847
DM		 5110{
5111 lockdep_set_class(&txq->_xmit_lock,
5112 &bonding_netdev_xmit_lock_key);
5113}
5114
5115static void bond_set_lockdep_class(struct net_device *dev)
5116{
cf508b12
DM		 5117 lockdep_set_class(&dev->addr_list_lock,
5118 &bonding_netdev_addr_lock_key);
e8a0464c 5119 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 5120}
5121
dfe60397 5122/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5123 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5124 * Caller must NOT hold rtnl_lock; we need to release it here before we
5125 * set up our sysfs entries.
5126 */
0dd646fe 5127int bond_create(char *name, struct bond_params *params)
dfe60397
MW		 5128{
5129 struct net_device *bond_dev;
0883beca 5130 struct bonding *bond;
dfe60397
MW		 5131 int res;
5132
5133 rtnl_lock();
027ea041
JV		 5134 down_write(&bonding_rwsem);
5135
5136 /* Check to see if the bond already exists. */
4fe4763c 5137 if (name) {
0883beca 5138 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5139 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5140 printk(KERN_ERR DRV_NAME
027ea041 5141 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5142 name);
5143 res = -EPERM;
5144 goto out_rtnl;
5145 }
5146 }
027ea041 5147
e4b91c48
JV		 5148 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5149 ether_setup);
dfe60397
MW		 5150 if (!bond_dev) {
5151 printk(KERN_ERR DRV_NAME
5152 ": %s: eek! can't alloc netdev!\n",
5153 name);
5154 res = -ENOMEM;
5155 goto out_rtnl;
5156 }
5157
e4b91c48
JV		 5158 if (!name) {
5159 res = dev_alloc_name(bond_dev, "bond%d");
5160 if (res < 0)
5161 goto out_netdev;
5162 }
5163
dfe60397
MW		 5164 /* bond_init() must be called after dev_alloc_name() (for the
5165 * /proc files), but before register_netdevice(), because we
5166 * need to set function pointers.
5167 */
5168
5169 res = bond_init(bond_dev, params);
5170 if (res < 0) {
5171 goto out_netdev;
5172 }
5173
dfe60397
MW		 5174 res = register_netdevice(bond_dev);
5175 if (res < 0) {
5176 goto out_bond;
5177 }
0daa2303 5178
c773e847 5179 bond_set_lockdep_class(bond_dev);
0daa2303 5180
ff59c456
JV		 5181 netif_carrier_off(bond_dev);
5182
027ea041 5183 up_write(&bonding_rwsem);
dfe60397 5184 rtnl_unlock(); /* allows sysfs registration of net device */
454d7c9b 5185 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
09c89279
JV		 5186 if (res < 0) {
5187 rtnl_lock();
027ea041 5188 down_write(&bonding_rwsem);
822973ba
PE		 5189 bond_deinit(bond_dev);
5190 unregister_netdevice(bond_dev);
5191 goto out_rtnl;
09c89279
JV		 5192 }
5193
5194 return 0;
5195
dfe60397
MW		 5196out_bond:
5197 bond_deinit(bond_dev);
5198out_netdev:
5199 free_netdev(bond_dev);
5200out_rtnl:
027ea041 5201 up_write(&bonding_rwsem);
dfe60397 5202 rtnl_unlock();
dfe60397
MW		 5203 return res;
5204}
5205
1da177e4
LT		 5206static int __init bonding_init(void)
5207{
1da177e4
LT		 5208 int i;
5209 int res;
0883beca 5210 struct bonding *bond;
1da177e4
LT		 5211
5212 printk(KERN_INFO "%s", version);
5213
dfe60397 5214 res = bond_check_params(&bonding_defaults);
1da177e4 5215 if (res) {
dfe60397 5216 goto out;
1da177e4
LT		 5217 }
5218
1da177e4
LT		 5219#ifdef CONFIG_PROC_FS
5220 bond_create_proc_dir();
5221#endif
027ea041
JV		 5222
5223 init_rwsem(&bonding_rwsem);
5224
1da177e4 5225 for (i = 0; i < max_bonds; i++) {
0dd646fe 5226 res = bond_create(NULL, &bonding_defaults);
dfe60397
MW		 5227 if (res)
5228 goto err;
1da177e4
LT		 5229 }
5230
b76cdba9
MW		 5231 res = bond_create_sysfs();
5232 if (res)
5233 goto err;
5234
1da177e4 5235 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5236 register_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5237 bond_register_ipv6_notifier();
1da177e4 5238
dfe60397
MW		 5239 goto out;
5240err:
0883beca 5241 list_for_each_entry(bond, &bond_dev_list, bond_list) {
1b76b316
JV		 5242 bond_work_cancel_all(bond);
5243 destroy_workqueue(bond->wq);
5244 }
5245
ae68c398
PE		 5246 bond_destroy_sysfs();
5247
40abc270 5248 rtnl_lock();
1da177e4 5249 bond_free_all();
1da177e4 5250 rtnl_unlock();
dfe60397 5251out:
1da177e4 5252 return res;
dfe60397 5253
1da177e4
LT		 5254}
5255
5256static void __exit bonding_exit(void)
5257{
5258 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5259 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5260 bond_unregister_ipv6_notifier();
1da177e4 5261
ae68c398
PE		 5262 bond_destroy_sysfs();
5263
1da177e4
LT		 5264 rtnl_lock();
5265 bond_free_all();
5266 rtnl_unlock();
5267}
5268
5269module_init(bonding_init);
5270module_exit(bonding_exit);
5271MODULE_LICENSE("GPL");
5272MODULE_VERSION(DRV_VERSION);
5273MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5274MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5275MODULE_SUPPORTED_DEVICE("most ethernet devices");
5276
5277/*
5278 * Local variables:
5279 * c-indent-level: 8
5280 * c-basic-offset: 8
5281 * tab-width: 8
5282 * End:
5283 */
5284
Clone of davem's net-next-2.6 tree