]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/bonding/bond_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bonding: fix destructor
[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;
d2991f75 104static struct 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
29112f4e
JP		 698 /* Try to get link status using Ethtool first. */
699 if (slave_dev->ethtool_ops) {
700 if (slave_dev->ethtool_ops->get_link) {
701 u32 link;
702
703 link = slave_dev->ethtool_ops->get_link(slave_dev);
704
705 return link ? BMSR_LSTATUS : 0;
706 }
707 }
708
709 /* Ethtool can't be used, fallback to MII ioclts. */
eb7cc59a 710 ioctl = slave_ops->ndo_do_ioctl;
1da177e4
LT		 711 if (ioctl) {
712 /* TODO: set pointer to correct ioctl on a per team member */
713 /* bases to make this more efficient. that is, once */
714 /* we determine the correct ioctl, we will always */
715 /* call it and not the others for that team */
716 /* member. */
717
718 /*
719 * We cannot assume that SIOCGMIIPHY will also read a
720 * register; not all network drivers (e.g., e100)
721 * support that.
722 */
723
724 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
725 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
726 mii = if_mii(&ifr);
727 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
728 mii->reg_num = MII_BMSR;
729 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
730 return (mii->val_out & BMSR_LSTATUS);
731 }
732 }
733 }
734
1da177e4
LT		 735 /*
736 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 737 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 738 * cannot report link status). If not reporting, pretend
739 * we're ok.
740 */
741 return (reporting ? -1 : BMSR_LSTATUS);
742}
743
744/*----------------------------- Multicast list ------------------------------*/
745
746/*
747 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
748 */
749static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
750{
751 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
752 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
753}
754
755/*
756 * returns dmi entry if found, NULL otherwise
757 */
758static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
759{
760 struct dev_mc_list *idmi;
761
762 for (idmi = mc_list; idmi; idmi = idmi->next) {
763 if (bond_is_dmi_same(dmi, idmi)) {
764 return idmi;
765 }
766 }
767
768 return NULL;
769}
770
771/*
772 * Push the promiscuity flag down to appropriate slaves
773 */
7e1a1ac1 774static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 775{
7e1a1ac1 776 int err = 0;
1da177e4
LT		 777 if (USES_PRIMARY(bond->params.mode)) {
778 /* write lock already acquired */
779 if (bond->curr_active_slave) {
7e1a1ac1
WC		 780 err = dev_set_promiscuity(bond->curr_active_slave->dev,
781 inc);
1da177e4
LT		 782 }
783 } else {
784 struct slave *slave;
785 int i;
786 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 787 err = dev_set_promiscuity(slave->dev, inc);
788 if (err)
789 return err;
1da177e4
LT		 790 }
791 }
7e1a1ac1 792 return err;
1da177e4
LT		 793}
794
795/*
796 * Push the allmulti flag down to all slaves
797 */
7e1a1ac1 798static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 799{
7e1a1ac1 800 int err = 0;
1da177e4
LT		 801 if (USES_PRIMARY(bond->params.mode)) {
802 /* write lock already acquired */
803 if (bond->curr_active_slave) {
7e1a1ac1
WC		 804 err = dev_set_allmulti(bond->curr_active_slave->dev,
805 inc);
1da177e4
LT		 806 }
807 } else {
808 struct slave *slave;
809 int i;
810 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 811 err = dev_set_allmulti(slave->dev, inc);
812 if (err)
813 return err;
1da177e4
LT		 814 }
815 }
7e1a1ac1 816 return err;
1da177e4
LT		 817}
818
819/*
820 * Add a Multicast address to slaves
821 * according to mode
822 */
823static void bond_mc_add(struct bonding *bond, void *addr, int alen)
824{
825 if (USES_PRIMARY(bond->params.mode)) {
826 /* write lock already acquired */
827 if (bond->curr_active_slave) {
828 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
829 }
830 } else {
831 struct slave *slave;
832 int i;
833 bond_for_each_slave(bond, slave, i) {
834 dev_mc_add(slave->dev, addr, alen, 0);
835 }
836 }
837}
838
839/*
840 * Remove a multicast address from slave
841 * according to mode
842 */
843static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
844{
845 if (USES_PRIMARY(bond->params.mode)) {
846 /* write lock already acquired */
847 if (bond->curr_active_slave) {
848 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
849 }
850 } else {
851 struct slave *slave;
852 int i;
853 bond_for_each_slave(bond, slave, i) {
854 dev_mc_delete(slave->dev, addr, alen, 0);
855 }
856 }
857}
858
a816c7c7
JV		 859
860/*
861 * Retrieve the list of registered multicast addresses for the bonding
862 * device and retransmit an IGMP JOIN request to the current active
863 * slave.
864 */
865static void bond_resend_igmp_join_requests(struct bonding *bond)
866{
867 struct in_device *in_dev;
868 struct ip_mc_list *im;
869
870 rcu_read_lock();
871 in_dev = __in_dev_get_rcu(bond->dev);
872 if (in_dev) {
873 for (im = in_dev->mc_list; im; im = im->next) {
874 ip_mc_rejoin_group(im);
875 }
876 }
877
878 rcu_read_unlock();
879}
880
1da177e4
LT		 881/*
882 * Totally destroys the mc_list in bond
883 */
884static void bond_mc_list_destroy(struct bonding *bond)
885{
886 struct dev_mc_list *dmi;
887
888 dmi = bond->mc_list;
889 while (dmi) {
890 bond->mc_list = dmi->next;
891 kfree(dmi);
892 dmi = bond->mc_list;
893 }
a816c7c7 894 bond->mc_list = NULL;
1da177e4
LT		 895}
896
897/*
898 * Copy all the Multicast addresses from src to the bonding device dst
899 */
de54f390 900static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
dd0fc66f 901 gfp_t gfp_flag)
1da177e4
LT		 902{
903 struct dev_mc_list *dmi, *new_dmi;
904
905 for (dmi = mc_list; dmi; dmi = dmi->next) {
de54f390 906 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1da177e4
LT		 907
908 if (!new_dmi) {
909 /* FIXME: Potential memory leak !!! */
910 return -ENOMEM;
911 }
912
913 new_dmi->next = bond->mc_list;
914 bond->mc_list = new_dmi;
915 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
916 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
917 new_dmi->dmi_users = dmi->dmi_users;
918 new_dmi->dmi_gusers = dmi->dmi_gusers;
919 }
920
921 return 0;
922}
923
924/*
925 * flush all members of flush->mc_list from device dev->mc_list
926 */
927static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
928{
454d7c9b 929 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 930 struct dev_mc_list *dmi;
931
932 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
933 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
934 }
935
936 if (bond->params.mode == BOND_MODE_8023AD) {
937 /* del lacpdu mc addr from mc list */
938 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
939
940 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
941 }
942}
943
944/*--------------------------- Active slave change ---------------------------*/
945
946/*
947 * Update the mc list and multicast-related flags for the new and
948 * old active slaves (if any) according to the multicast mode, and
949 * promiscuous flags unconditionally.
950 */
951static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
952{
953 struct dev_mc_list *dmi;
954
955 if (!USES_PRIMARY(bond->params.mode)) {
956 /* nothing to do - mc list is already up-to-date on
957 * all slaves
958 */
959 return;
960 }
961
962 if (old_active) {
963 if (bond->dev->flags & IFF_PROMISC) {
964 dev_set_promiscuity(old_active->dev, -1);
965 }
966
967 if (bond->dev->flags & IFF_ALLMULTI) {
968 dev_set_allmulti(old_active->dev, -1);
969 }
970
971 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
972 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
973 }
974 }
975
976 if (new_active) {
7e1a1ac1 977 /* FIXME: Signal errors upstream. */
1da177e4
LT		 978 if (bond->dev->flags & IFF_PROMISC) {
979 dev_set_promiscuity(new_active->dev, 1);
980 }
981
982 if (bond->dev->flags & IFF_ALLMULTI) {
983 dev_set_allmulti(new_active->dev, 1);
984 }
985
986 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
987 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
988 }
a816c7c7 989 bond_resend_igmp_join_requests(bond);
1da177e4
LT		 990 }
991}
992
3915c1e8
JV		 993/*
994 * bond_do_fail_over_mac
995 *
996 * Perform special MAC address swapping for fail_over_mac settings
997 *
998 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
999 */
1000static void bond_do_fail_over_mac(struct bonding *bond,
1001 struct slave *new_active,
1002 struct slave *old_active)
1f78d9f9
HE		 1003 __releases(&bond->curr_slave_lock)
1004 __releases(&bond->lock)
1005 __acquires(&bond->lock)
1006 __acquires(&bond->curr_slave_lock)
3915c1e8
JV		 1007{
1008 u8 tmp_mac[ETH_ALEN];
1009 struct sockaddr saddr;
1010 int rv;
1011
1012 switch (bond->params.fail_over_mac) {
1013 case BOND_FOM_ACTIVE:
1014 if (new_active)
1015 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1016 new_active->dev->addr_len);
1017 break;
1018 case BOND_FOM_FOLLOW:
1019 /*
1020 * if new_active && old_active, swap them
1021 * if just old_active, do nothing (going to no active slave)
1022 * if just new_active, set new_active to bond's MAC
1023 */
1024 if (!new_active)
1025 return;
1026
1027 write_unlock_bh(&bond->curr_slave_lock);
1028 read_unlock(&bond->lock);
1029
1030 if (old_active) {
1031 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1032 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1033 ETH_ALEN);
1034 saddr.sa_family = new_active->dev->type;
1035 } else {
1036 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1037 saddr.sa_family = bond->dev->type;
1038 }
1039
1040 rv = dev_set_mac_address(new_active->dev, &saddr);
1041 if (rv) {
1042 printk(KERN_ERR DRV_NAME
1043 ": %s: Error %d setting MAC of slave %s\n",
1044 bond->dev->name, -rv, new_active->dev->name);
1045 goto out;
1046 }
1047
1048 if (!old_active)
1049 goto out;
1050
1051 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1052 saddr.sa_family = old_active->dev->type;
1053
1054 rv = dev_set_mac_address(old_active->dev, &saddr);
1055 if (rv)
1056 printk(KERN_ERR DRV_NAME
1057 ": %s: Error %d setting MAC of slave %s\n",
1058 bond->dev->name, -rv, new_active->dev->name);
1059out:
1060 read_lock(&bond->lock);
1061 write_lock_bh(&bond->curr_slave_lock);
1062 break;
1063 default:
1064 printk(KERN_ERR DRV_NAME
1065 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1066 bond->dev->name, bond->params.fail_over_mac);
1067 break;
1068 }
1069
1070}
1071
1072
1da177e4
LT		 1073/**
1074 * find_best_interface - select the best available slave to be the active one
1075 * @bond: our bonding struct
1076 *
1077 * Warning: Caller must hold curr_slave_lock for writing.
1078 */
1079static struct slave *bond_find_best_slave(struct bonding *bond)
1080{
1081 struct slave *new_active, *old_active;
1082 struct slave *bestslave = NULL;
1083 int mintime = bond->params.updelay;
1084 int i;
1085
1086 new_active = old_active = bond->curr_active_slave;
1087
1088 if (!new_active) { /* there were no active slaves left */
1089 if (bond->slave_cnt > 0) { /* found one slave */
1090 new_active = bond->first_slave;
1091 } else {
1092 return NULL; /* still no slave, return NULL */
1093 }
1094 }
1095
1096 /* first try the primary link; if arping, a link must tx/rx traffic
1097 * before it can be considered the curr_active_slave - also, we would skip
1098 * slaves between the curr_active_slave and primary_slave that may be up
1099 * and able to arp
1100 */
1101 if ((bond->primary_slave) &&
1102 (!bond->params.arp_interval) &&
1103 (IS_UP(bond->primary_slave->dev))) {
1104 new_active = bond->primary_slave;
1105 }
1106
1107 /* remember where to stop iterating over the slaves */
1108 old_active = new_active;
1109
1110 bond_for_each_slave_from(bond, new_active, i, old_active) {
1111 if (IS_UP(new_active->dev)) {
1112 if (new_active->link == BOND_LINK_UP) {
1113 return new_active;
1114 } else if (new_active->link == BOND_LINK_BACK) {
1115 /* link up, but waiting for stabilization */
1116 if (new_active->delay < mintime) {
1117 mintime = new_active->delay;
1118 bestslave = new_active;
1119 }
1120 }
1121 }
1122 }
1123
1124 return bestslave;
1125}
1126
1127/**
1128 * change_active_interface - change the active slave into the specified one
1129 * @bond: our bonding struct
1130 * @new: the new slave to make the active one
1131 *
1132 * Set the new slave to the bond's settings and unset them on the old
1133 * curr_active_slave.
1134 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1135 *
1136 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1137 * because it is apparently the best available slave we have, even though its
1138 * updelay hasn't timed out yet.
1139 *
3915c1e8
JV		 1140 * If new_active is not NULL, caller must hold bond->lock for read and
1141 * curr_slave_lock for write_bh.
1da177e4 1142 */
a77b5325 1143void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1144{
1145 struct slave *old_active = bond->curr_active_slave;
1146
1147 if (old_active == new_active) {
1148 return;
1149 }
1150
1151 if (new_active) {
b2220cad
JV		 1152 new_active->jiffies = jiffies;
1153
1da177e4
LT		 1154 if (new_active->link == BOND_LINK_BACK) {
1155 if (USES_PRIMARY(bond->params.mode)) {
1156 printk(KERN_INFO DRV_NAME
1157 ": %s: making interface %s the new "
1158 "active one %d ms earlier.\n",
1159 bond->dev->name, new_active->dev->name,
1160 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1161 }
1162
1163 new_active->delay = 0;
1164 new_active->link = BOND_LINK_UP;
1da177e4
LT		 1165
1166 if (bond->params.mode == BOND_MODE_8023AD) {
1167 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1168 }
1169
58402054 1170 if (bond_is_lb(bond))
1da177e4 1171 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1da177e4
LT		 1172 } else {
1173 if (USES_PRIMARY(bond->params.mode)) {
1174 printk(KERN_INFO DRV_NAME
1175 ": %s: making interface %s the new "
1176 "active one.\n",
1177 bond->dev->name, new_active->dev->name);
1178 }
1179 }
1180 }
1181
1da177e4
LT		 1182 if (USES_PRIMARY(bond->params.mode)) {
1183 bond_mc_swap(bond, new_active, old_active);
1184 }
1185
58402054 1186 if (bond_is_lb(bond)) {
1da177e4 1187 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1188 if (old_active)
1189 bond_set_slave_inactive_flags(old_active);
1190 if (new_active)
1191 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1192 } else {
1193 bond->curr_active_slave = new_active;
1194 }
c3ade5ca
JV		 1195
1196 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1197 if (old_active) {
1198 bond_set_slave_inactive_flags(old_active);
1199 }
1200
1201 if (new_active) {
1202 bond_set_slave_active_flags(new_active);
2ab82852 1203
709f8a45
OG		 1204 if (bond->params.fail_over_mac)
1205 bond_do_fail_over_mac(bond, new_active,
1206 old_active);
3915c1e8 1207
709f8a45 1208 bond->send_grat_arp = bond->params.num_grat_arp;
b59f9f74 1209 bond_send_gratuitous_arp(bond);
01f3109d 1210
305d552a
BH		 1211 bond->send_unsol_na = bond->params.num_unsol_na;
1212 bond_send_unsolicited_na(bond);
1213
01f3109d
OG		 1214 write_unlock_bh(&bond->curr_slave_lock);
1215 read_unlock(&bond->lock);
1216
1217 netdev_bonding_change(bond->dev);
1218
1219 read_lock(&bond->lock);
1220 write_lock_bh(&bond->curr_slave_lock);
7893b249 1221 }
c3ade5ca 1222 }
1da177e4
LT		 1223}
1224
1225/**
1226 * bond_select_active_slave - select a new active slave, if needed
1227 * @bond: our bonding struct
1228 *
1229 * This functions shoud be called when one of the following occurs:
1230 * - The old curr_active_slave has been released or lost its link.
1231 * - The primary_slave has got its link back.
1232 * - A slave has got its link back and there's no old curr_active_slave.
1233 *
3915c1e8 1234 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1235 */
a77b5325 1236void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1237{
1238 struct slave *best_slave;
ff59c456 1239 int rv;
1da177e4
LT		 1240
1241 best_slave = bond_find_best_slave(bond);
1242 if (best_slave != bond->curr_active_slave) {
1243 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1244 rv = bond_set_carrier(bond);
1245 if (!rv)
1246 return;
1247
1248 if (netif_carrier_ok(bond->dev)) {
1249 printk(KERN_INFO DRV_NAME
1250 ": %s: first active interface up!\n",
1251 bond->dev->name);
1252 } else {
1253 printk(KERN_INFO DRV_NAME ": %s: "
1254 "now running without any active interface !\n",
1255 bond->dev->name);
1256 }
1da177e4
LT		 1257 }
1258}
1259
1260/*--------------------------- slave list handling ---------------------------*/
1261
1262/*
1263 * This function attaches the slave to the end of list.
1264 *
1265 * bond->lock held for writing by caller.
1266 */
1267static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1268{
1269 if (bond->first_slave == NULL) { /* attaching the first slave */
1270 new_slave->next = new_slave;
1271 new_slave->prev = new_slave;
1272 bond->first_slave = new_slave;
1273 } else {
1274 new_slave->next = bond->first_slave;
1275 new_slave->prev = bond->first_slave->prev;
1276 new_slave->next->prev = new_slave;
1277 new_slave->prev->next = new_slave;
1278 }
1279
1280 bond->slave_cnt++;
1281}
1282
1283/*
1284 * This function detaches the slave from the list.
1285 * WARNING: no check is made to verify if the slave effectively
1286 * belongs to <bond>.
1287 * Nothing is freed on return, structures are just unchained.
1288 * If any slave pointer in bond was pointing to <slave>,
1289 * it should be changed by the calling function.
1290 *
1291 * bond->lock held for writing by caller.
1292 */
1293static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1294{
1295 if (slave->next) {
1296 slave->next->prev = slave->prev;
1297 }
1298
1299 if (slave->prev) {
1300 slave->prev->next = slave->next;
1301 }
1302
1303 if (bond->first_slave == slave) { /* slave is the first slave */
1304 if (bond->slave_cnt > 1) { /* there are more slave */
1305 bond->first_slave = slave->next;
1306 } else {
1307 bond->first_slave = NULL; /* slave was the last one */
1308 }
1309 }
1310
1311 slave->next = NULL;
1312 slave->prev = NULL;
1313 bond->slave_cnt--;
1314}
1315
1316/*---------------------------------- IOCTL ----------------------------------*/
1317
4ad072c9
AB		 1318static int bond_sethwaddr(struct net_device *bond_dev,
1319 struct net_device *slave_dev)
1da177e4 1320{
5a03cdb7
HE		 1321 pr_debug("bond_dev=%p\n", bond_dev);
1322 pr_debug("slave_dev=%p\n", slave_dev);
1323 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1da177e4
LT		 1324 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1325 return 0;
1326}
1327
7f353bf2
HX		 1328#define BOND_VLAN_FEATURES \
1329 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1330 NETIF_F_HW_VLAN_FILTER)
8531c5ff
AK		 1331
1332/*
8e3babcd 1333 * Compute the common dev->feature set available to all slaves. Some
7f353bf2
HX		 1334 * feature bits are managed elsewhere, so preserve those feature bits
1335 * on the master device.
8531c5ff
AK		 1336 */
1337static int bond_compute_features(struct bonding *bond)
1338{
8531c5ff
AK		 1339 struct slave *slave;
1340 struct net_device *bond_dev = bond->dev;
7f353bf2 1341 unsigned long features = bond_dev->features;
3158bf7d
MS		 1342 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1343 bond_dev->hard_header_len);
8e3babcd 1344 int i;
8531c5ff 1345
7f353bf2 1346 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
b63365a2
HX		 1347 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1348
1349 if (!bond->first_slave)
1350 goto done;
1351
1352 features &= ~NETIF_F_ONE_FOR_ALL;
7f353bf2 1353
54ef3137 1354 bond_for_each_slave(bond, slave, i) {
b63365a2
HX		 1355 features = netdev_increment_features(features,
1356 slave->dev->features,
1357 NETIF_F_ONE_FOR_ALL);
54ef3137
JV		 1358 if (slave->dev->hard_header_len > max_hard_header_len)
1359 max_hard_header_len = slave->dev->hard_header_len;
1360 }
8531c5ff 1361
b63365a2 1362done:
7f353bf2 1363 features |= (bond_dev->features & BOND_VLAN_FEATURES);
b63365a2 1364 bond_dev->features = netdev_fix_features(features, NULL);
54ef3137 1365 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff
AK		 1366
1367 return 0;
1368}
1369
872254dd
MS		 1370static void bond_setup_by_slave(struct net_device *bond_dev,
1371 struct net_device *slave_dev)
1372{
454d7c9b 1373 struct bonding *bond = netdev_priv(bond_dev);
d90a162a 1374
00829823 1375 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1376
1377 bond_dev->type = slave_dev->type;
1378 bond_dev->hard_header_len = slave_dev->hard_header_len;
1379 bond_dev->addr_len = slave_dev->addr_len;
1380
1381 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1382 slave_dev->addr_len);
d90a162a 1383 bond->setup_by_slave = 1;
872254dd
MS		 1384}
1385
1da177e4 1386/* enslave device <slave> to bond device <master> */
a77b5325 1387int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1388{
454d7c9b 1389 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1390 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 1391 struct slave *new_slave = NULL;
1392 struct dev_mc_list *dmi;
1393 struct sockaddr addr;
1394 int link_reporting;
1395 int old_features = bond_dev->features;
1396 int res = 0;
1397
552709d5 1398 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
eb7cc59a 1399 slave_ops->ndo_do_ioctl == NULL) {
1da177e4 1400 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1401 ": %s: Warning: no link monitoring support for %s\n",
1402 bond_dev->name, slave_dev->name);
1da177e4
LT		 1403 }
1404
1405 /* bond must be initialized by bond_open() before enslaving */
1406 if (!(bond_dev->flags & IFF_UP)) {
6b1bf096
MS		 1407 printk(KERN_WARNING DRV_NAME
1408 " %s: master_dev is not up in bond_enslave\n",
1409 bond_dev->name);
1da177e4
LT		 1410 }
1411
1412 /* already enslaved */
1413 if (slave_dev->flags & IFF_SLAVE) {
5a03cdb7 1414 pr_debug("Error, Device was already enslaved\n");
1da177e4
LT		 1415 return -EBUSY;
1416 }
1417
1418 /* vlan challenged mutual exclusion */
1419 /* no need to lock since we're protected by rtnl_lock */
1420 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
5a03cdb7 1421 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1da177e4
LT		 1422 if (!list_empty(&bond->vlan_list)) {
1423 printk(KERN_ERR DRV_NAME
4e0952c7 1424 ": %s: Error: cannot enslave VLAN "
1da177e4 1425 "challenged slave %s on VLAN enabled "
4e0952c7 1426 "bond %s\n", bond_dev->name, slave_dev->name,
1da177e4
LT		 1427 bond_dev->name);
1428 return -EPERM;
1429 } else {
1430 printk(KERN_WARNING DRV_NAME
4e0952c7 1431 ": %s: Warning: enslaved VLAN challenged "
1da177e4
LT		 1432 "slave %s. Adding VLANs will be blocked as "
1433 "long as %s is part of bond %s\n",
4e0952c7 1434 bond_dev->name, slave_dev->name, slave_dev->name,
1da177e4
LT		 1435 bond_dev->name);
1436 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1437 }
1438 } else {
5a03cdb7 1439 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1da177e4
LT		 1440 if (bond->slave_cnt == 0) {
1441 /* First slave, and it is not VLAN challenged,
1442 * so remove the block of adding VLANs over the bond.
1443 */
1444 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1445 }
1446 }
1447
217df670
JV		 1448 /*
1449 * Old ifenslave binaries are no longer supported. These can
1450 * be identified with moderate accurary by the state of the slave:
1451 * the current ifenslave will set the interface down prior to
1452 * enslaving it; the old ifenslave will not.
1453 */
1454 if ((slave_dev->flags & IFF_UP)) {
1455 printk(KERN_ERR DRV_NAME ": %s is up. "
1456 "This may be due to an out of date ifenslave.\n",
1457 slave_dev->name);
1458 res = -EPERM;
1459 goto err_undo_flags;
1460 }
1da177e4 1461
872254dd
MS		 1462 /* set bonding device ether type by slave - bonding netdevices are
1463 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1464 * there is a need to override some of the type dependent attribs/funcs.
1465 *
1466 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1467 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1468 */
1469 if (bond->slave_cnt == 0) {
1470 if (slave_dev->type != ARPHRD_ETHER)
1471 bond_setup_by_slave(bond_dev, slave_dev);
1472 } else if (bond_dev->type != slave_dev->type) {
1473 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1474 "from other slaves (%d), can not enslave it.\n",
1475 slave_dev->name,
1476 slave_dev->type, bond_dev->type);
1477 res = -EINVAL;
1478 goto err_undo_flags;
1479 }
1480
eb7cc59a 1481 if (slave_ops->ndo_set_mac_address == NULL) {
2ab82852
MS		 1482 if (bond->slave_cnt == 0) {
1483 printk(KERN_WARNING DRV_NAME
dd957c57
JV		 1484 ": %s: Warning: The first slave device "
1485 "specified does not support setting the MAC "
3915c1e8 1486 "address. Setting fail_over_mac to active.",
dd957c57 1487 bond_dev->name);
3915c1e8
JV		 1488 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1489 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852 1490 printk(KERN_ERR DRV_NAME
dd957c57
JV		 1491 ": %s: Error: The slave device specified "
1492 "does not support setting the MAC address, "
3915c1e8 1493 "but fail_over_mac is not set to active.\n"
2ab82852
MS		 1494 , bond_dev->name);
1495 res = -EOPNOTSUPP;
1496 goto err_undo_flags;
1497 }
1da177e4
LT		 1498 }
1499
243cb4e5 1500 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1501 if (!new_slave) {
1502 res = -ENOMEM;
1503 goto err_undo_flags;
1504 }
1505
1da177e4
LT		 1506 /* save slave's original flags before calling
1507 * netdev_set_master and dev_open
1508 */
1509 new_slave->original_flags = slave_dev->flags;
1510
217df670
JV		 1511 /*
1512 * Save slave's original ("permanent") mac address for modes
1513 * that need it, and for restoring it upon release, and then
1514 * set it to the master's address
1515 */
1516 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1517
dd957c57 1518 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1519 /*
1520 * Set slave to master's mac address. The application already
1521 * set the master's mac address to that of the first slave
1522 */
1523 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1524 addr.sa_family = slave_dev->type;
1525 res = dev_set_mac_address(slave_dev, &addr);
1526 if (res) {
5a03cdb7 1527 pr_debug("Error %d calling set_mac_address\n", res);
2ab82852
MS		 1528 goto err_free;
1529 }
217df670 1530 }
1da177e4 1531
c2edacf8
JV		 1532 res = netdev_set_master(slave_dev, bond_dev);
1533 if (res) {
5a03cdb7 1534 pr_debug("Error %d calling netdev_set_master\n", res);
569f0c4d 1535 goto err_restore_mac;
c2edacf8 1536 }
217df670
JV		 1537 /* open the slave since the application closed it */
1538 res = dev_open(slave_dev);
1539 if (res) {
5a03cdb7 1540 pr_debug("Openning slave %s failed\n", slave_dev->name);
569f0c4d 1541 goto err_unset_master;
1da177e4
LT		 1542 }
1543
1da177e4 1544 new_slave->dev = slave_dev;
0b680e75 1545 slave_dev->priv_flags |= IFF_BONDING;
1da177e4 1546
58402054 1547 if (bond_is_lb(bond)) {
1da177e4
LT		 1548 /* bond_alb_init_slave() must be called before all other stages since
1549 * it might fail and we do not want to have to undo everything
1550 */
1551 res = bond_alb_init_slave(bond, new_slave);
1552 if (res) {
569f0c4d 1553 goto err_close;
1da177e4
LT		 1554 }
1555 }
1556
1557 /* If the mode USES_PRIMARY, then the new slave gets the
1558 * master's promisc (and mc) settings only if it becomes the
1559 * curr_active_slave, and that is taken care of later when calling
1560 * bond_change_active()
1561 */
1562 if (!USES_PRIMARY(bond->params.mode)) {
1563 /* set promiscuity level to new slave */
1564 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1565 res = dev_set_promiscuity(slave_dev, 1);
1566 if (res)
1567 goto err_close;
1da177e4
LT		 1568 }
1569
1570 /* set allmulti level to new slave */
1571 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1572 res = dev_set_allmulti(slave_dev, 1);
1573 if (res)
1574 goto err_close;
1da177e4
LT		 1575 }
1576
b9e40857 1577 netif_addr_lock_bh(bond_dev);
1da177e4
LT		 1578 /* upload master's mc_list to new slave */
1579 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1580 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1581 }
b9e40857 1582 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1583 }
1584
1585 if (bond->params.mode == BOND_MODE_8023AD) {
1586 /* add lacpdu mc addr to mc list */
1587 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1588
1589 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1590 }
1591
1592 bond_add_vlans_on_slave(bond, slave_dev);
1593
1594 write_lock_bh(&bond->lock);
1595
1596 bond_attach_slave(bond, new_slave);
1597
1598 new_slave->delay = 0;
1599 new_slave->link_failure_count = 0;
1600
8531c5ff
AK		 1601 bond_compute_features(bond);
1602
3915c1e8
JV		 1603 write_unlock_bh(&bond->lock);
1604
1605 read_lock(&bond->lock);
1606
f5b2b966
JV		 1607 new_slave->last_arp_rx = jiffies;
1608
1da177e4
LT		 1609 if (bond->params.miimon && !bond->params.use_carrier) {
1610 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1611
1612 if ((link_reporting == -1) && !bond->params.arp_interval) {
1613 /*
1614 * miimon is set but a bonded network driver
1615 * does not support ETHTOOL/MII and
1616 * arp_interval is not set. Note: if
1617 * use_carrier is enabled, we will never go
1618 * here (because netif_carrier is always
1619 * supported); thus, we don't need to change
1620 * the messages for netif_carrier.
1621 */
1622 printk(KERN_WARNING DRV_NAME
4e0952c7 1623 ": %s: Warning: MII and ETHTOOL support not "
1da177e4
LT		 1624 "available for interface %s, and "
1625 "arp_interval/arp_ip_target module parameters "
1626 "not specified, thus bonding will not detect "
1627 "link failures! see bonding.txt for details.\n",
4e0952c7 1628 bond_dev->name, slave_dev->name);
1da177e4
LT		 1629 } else if (link_reporting == -1) {
1630 /* unable get link status using mii/ethtool */
1631 printk(KERN_WARNING DRV_NAME
4e0952c7 1632 ": %s: Warning: can't get link status from "
1da177e4
LT		 1633 "interface %s; the network driver associated "
1634 "with this interface does not support MII or "
1635 "ETHTOOL link status reporting, thus miimon "
1636 "has no effect on this interface.\n",
4e0952c7 1637 bond_dev->name, slave_dev->name);
1da177e4
LT		 1638 }
1639 }
1640
1641 /* check for initial state */
1642 if (!bond->params.miimon ||
1643 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1644 if (bond->params.updelay) {
5a03cdb7 1645 pr_debug("Initial state of slave_dev is "
1da177e4
LT		 1646 "BOND_LINK_BACK\n");
1647 new_slave->link = BOND_LINK_BACK;
1648 new_slave->delay = bond->params.updelay;
1649 } else {
5a03cdb7 1650 pr_debug("Initial state of slave_dev is "
1da177e4
LT		 1651 "BOND_LINK_UP\n");
1652 new_slave->link = BOND_LINK_UP;
1653 }
1654 new_slave->jiffies = jiffies;
1655 } else {
5a03cdb7 1656 pr_debug("Initial state of slave_dev is "
1da177e4
LT		 1657 "BOND_LINK_DOWN\n");
1658 new_slave->link = BOND_LINK_DOWN;
1659 }
1660
1661 if (bond_update_speed_duplex(new_slave) &&
1662 (new_slave->link != BOND_LINK_DOWN)) {
1663 printk(KERN_WARNING DRV_NAME
4e0952c7 1664 ": %s: Warning: failed to get speed and duplex from %s, "
1da177e4 1665 "assumed to be 100Mb/sec and Full.\n",
4e0952c7 1666 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1667
1668 if (bond->params.mode == BOND_MODE_8023AD) {
4e0952c7
MW		 1669 printk(KERN_WARNING DRV_NAME
1670 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1da177e4 1671 "support in base driver for proper aggregator "
4e0952c7 1672 "selection.\n", bond_dev->name);
1da177e4
LT		 1673 }
1674 }
1675
1676 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1677 /* if there is a primary slave, remember it */
1678 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1679 bond->primary_slave = new_slave;
1680 }
1681 }
1682
3915c1e8
JV		 1683 write_lock_bh(&bond->curr_slave_lock);
1684
1da177e4
LT		 1685 switch (bond->params.mode) {
1686 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1687 bond_set_slave_inactive_flags(new_slave);
1688 bond_select_active_slave(bond);
1da177e4
LT		 1689 break;
1690 case BOND_MODE_8023AD:
1691 /* in 802.3ad mode, the internal mechanism
1692 * will activate the slaves in the selected
1693 * aggregator
1694 */
1695 bond_set_slave_inactive_flags(new_slave);
1696 /* if this is the first slave */
1697 if (bond->slave_cnt == 1) {
1698 SLAVE_AD_INFO(new_slave).id = 1;
1699 /* Initialize AD with the number of times that the AD timer is called in 1 second
1700 * can be called only after the mac address of the bond is set
1701 */
1702 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1703 bond->params.lacp_fast);
1704 } else {
1705 SLAVE_AD_INFO(new_slave).id =
1706 SLAVE_AD_INFO(new_slave->prev).id + 1;
1707 }
1708
1709 bond_3ad_bind_slave(new_slave);
1710 break;
1711 case BOND_MODE_TLB:
1712 case BOND_MODE_ALB:
1713 new_slave->state = BOND_STATE_ACTIVE;
059fe7a5 1714 bond_set_slave_inactive_flags(new_slave);
5a29f789 1715 bond_select_active_slave(bond);
1da177e4
LT		 1716 break;
1717 default:
5a03cdb7 1718 pr_debug("This slave is always active in trunk mode\n");
1da177e4
LT		 1719
1720 /* always active in trunk mode */
1721 new_slave->state = BOND_STATE_ACTIVE;
1722
1723 /* In trunking mode there is little meaning to curr_active_slave
1724 * anyway (it holds no special properties of the bond device),
1725 * so we can change it without calling change_active_interface()
1726 */
1727 if (!bond->curr_active_slave) {
1728 bond->curr_active_slave = new_slave;
1729 }
1730 break;
1731 } /* switch(bond_mode) */
1732
3915c1e8
JV		 1733 write_unlock_bh(&bond->curr_slave_lock);
1734
ff59c456
JV		 1735 bond_set_carrier(bond);
1736
3915c1e8 1737 read_unlock(&bond->lock);
1da177e4 1738
b76cdba9
MW		 1739 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1740 if (res)
569f0c4d 1741 goto err_close;
b76cdba9 1742
1da177e4
LT		 1743 printk(KERN_INFO DRV_NAME
1744 ": %s: enslaving %s as a%s interface with a%s link.\n",
1745 bond_dev->name, slave_dev->name,
1746 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1747 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1748
1749 /* enslave is successful */
1750 return 0;
1751
1752/* Undo stages on error */
1da177e4
LT		 1753err_close:
1754 dev_close(slave_dev);
1755
569f0c4d
JV		 1756err_unset_master:
1757 netdev_set_master(slave_dev, NULL);
1758
1da177e4 1759err_restore_mac:
dd957c57 1760 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1761 /* XXX TODO - fom follow mode needs to change master's
1762 * MAC if this slave's MAC is in use by the bond, or at
1763 * least print a warning.
1764 */
2ab82852
MS		 1765 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1766 addr.sa_family = slave_dev->type;
1767 dev_set_mac_address(slave_dev, &addr);
1768 }
1da177e4
LT		 1769
1770err_free:
1771 kfree(new_slave);
1772
1773err_undo_flags:
1774 bond_dev->features = old_features;
8531c5ff 1775
1da177e4
LT		 1776 return res;
1777}
1778
1779/*
1780 * Try to release the slave device <slave> from the bond device <master>
1781 * It is legal to access curr_active_slave without a lock because all the function
1782 * is write-locked.
1783 *
1784 * The rules for slave state should be:
1785 * for Active/Backup:
1786 * Active stays on all backups go down
1787 * for Bonded connections:
1788 * The first up interface should be left on and all others downed.
1789 */
a77b5325 1790int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1791{
454d7c9b 1792 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1793 struct slave *slave, *oldcurrent;
1794 struct sockaddr addr;
1795 int mac_addr_differ;
1796
1797 /* slave is not a slave or master is not master of this slave */
1798 if (!(slave_dev->flags & IFF_SLAVE) ||
1799 (slave_dev->master != bond_dev)) {
1800 printk(KERN_ERR DRV_NAME
4e0952c7 1801 ": %s: Error: cannot release %s.\n",
1da177e4
LT		 1802 bond_dev->name, slave_dev->name);
1803 return -EINVAL;
1804 }
1805
1806 write_lock_bh(&bond->lock);
1807
1808 slave = bond_get_slave_by_dev(bond, slave_dev);
1809 if (!slave) {
1810 /* not a slave of this bond */
1811 printk(KERN_INFO DRV_NAME
1812 ": %s: %s not enslaved\n",
1813 bond_dev->name, slave_dev->name);
f5e2a7b2 1814 write_unlock_bh(&bond->lock);
1da177e4
LT		 1815 return -EINVAL;
1816 }
1817
3915c1e8
JV		 1818 if (!bond->params.fail_over_mac) {
1819 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1820 ETH_ALEN);
1821 if (!mac_addr_differ && (bond->slave_cnt > 1))
1822 printk(KERN_WARNING DRV_NAME
1823 ": %s: Warning: the permanent HWaddr of %s - "
e174961c 1824 "%pM - is still in use by %s. "
3915c1e8
JV		 1825 "Set the HWaddr of %s to a different address "
1826 "to avoid conflicts.\n",
1827 bond_dev->name, slave_dev->name,
e174961c 1828 slave->perm_hwaddr,
3915c1e8 1829 bond_dev->name, slave_dev->name);
1da177e4
LT		 1830 }
1831
1832 /* Inform AD package of unbinding of slave. */
1833 if (bond->params.mode == BOND_MODE_8023AD) {
1834 /* must be called before the slave is
1835 * detached from the list
1836 */
1837 bond_3ad_unbind_slave(slave);
1838 }
1839
1840 printk(KERN_INFO DRV_NAME
1841 ": %s: releasing %s interface %s\n",
1842 bond_dev->name,
1843 (slave->state == BOND_STATE_ACTIVE)
1844 ? "active" : "backup",
1845 slave_dev->name);
1846
1847 oldcurrent = bond->curr_active_slave;
1848
1849 bond->current_arp_slave = NULL;
1850
1851 /* release the slave from its bond */
1852 bond_detach_slave(bond, slave);
1853
8531c5ff
AK		 1854 bond_compute_features(bond);
1855
1da177e4
LT		 1856 if (bond->primary_slave == slave) {
1857 bond->primary_slave = NULL;
1858 }
1859
1860 if (oldcurrent == slave) {
1861 bond_change_active_slave(bond, NULL);
1862 }
1863
58402054 1864 if (bond_is_lb(bond)) {
1da177e4
LT		 1865 /* Must be called only after the slave has been
1866 * detached from the list and the curr_active_slave
1867 * has been cleared (if our_slave == old_current),
1868 * but before a new active slave is selected.
1869 */
2543331d 1870 write_unlock_bh(&bond->lock);
1da177e4 1871 bond_alb_deinit_slave(bond, slave);
2543331d 1872 write_lock_bh(&bond->lock);
1da177e4
LT		 1873 }
1874
059fe7a5
JV		 1875 if (oldcurrent == slave) {
1876 /*
1877 * Note that we hold RTNL over this sequence, so there
1878 * is no concern that another slave add/remove event
1879 * will interfere.
1880 */
1881 write_unlock_bh(&bond->lock);
1882 read_lock(&bond->lock);
1883 write_lock_bh(&bond->curr_slave_lock);
1884
1da177e4
LT		 1885 bond_select_active_slave(bond);
1886
059fe7a5
JV		 1887 write_unlock_bh(&bond->curr_slave_lock);
1888 read_unlock(&bond->lock);
1889 write_lock_bh(&bond->lock);
1890 }
1891
1da177e4 1892 if (bond->slave_cnt == 0) {
ff59c456
JV		 1893 bond_set_carrier(bond);
1894
1da177e4
LT		 1895 /* if the last slave was removed, zero the mac address
1896 * of the master so it will be set by the application
1897 * to the mac address of the first slave
1898 */
1899 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1900
1901 if (list_empty(&bond->vlan_list)) {
1902 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1903 } else {
1904 printk(KERN_WARNING DRV_NAME
4e0952c7 1905 ": %s: Warning: clearing HW address of %s while it "
1da177e4 1906 "still has VLANs.\n",
4e0952c7 1907 bond_dev->name, bond_dev->name);
1da177e4 1908 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 1909 ": %s: When re-adding slaves, make sure the bond's "
1910 "HW address matches its VLANs'.\n",
1911 bond_dev->name);
1da177e4
LT		 1912 }
1913 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1914 !bond_has_challenged_slaves(bond)) {
1915 printk(KERN_INFO DRV_NAME
4e0952c7 1916 ": %s: last VLAN challenged slave %s "
1da177e4 1917 "left bond %s. VLAN blocking is removed\n",
4e0952c7 1918 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1919 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1920 }
1921
1922 write_unlock_bh(&bond->lock);
1923
b76cdba9
MW		 1924 /* must do this from outside any spinlocks */
1925 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1926
1da177e4
LT		 1927 bond_del_vlans_from_slave(bond, slave_dev);
1928
1929 /* If the mode USES_PRIMARY, then we should only remove its
1930 * promisc and mc settings if it was the curr_active_slave, but that was
1931 * already taken care of above when we detached the slave
1932 */
1933 if (!USES_PRIMARY(bond->params.mode)) {
1934 /* unset promiscuity level from slave */
1935 if (bond_dev->flags & IFF_PROMISC) {
1936 dev_set_promiscuity(slave_dev, -1);
1937 }
1938
1939 /* unset allmulti level from slave */
1940 if (bond_dev->flags & IFF_ALLMULTI) {
1941 dev_set_allmulti(slave_dev, -1);
1942 }
1943
1944 /* flush master's mc_list from slave */
b9e40857 1945 netif_addr_lock_bh(bond_dev);
1da177e4 1946 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 1947 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1948 }
1949
1950 netdev_set_master(slave_dev, NULL);
1951
1952 /* close slave before restoring its mac address */
1953 dev_close(slave_dev);
1954
3915c1e8 1955 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 1956 /* restore original ("permanent") mac address */
1957 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1958 addr.sa_family = slave_dev->type;
1959 dev_set_mac_address(slave_dev, &addr);
1960 }
1da177e4 1961
8f903c70 1962 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
f5b2b966
JV		 1963 IFF_SLAVE_INACTIVE | IFF_BONDING |
1964 IFF_SLAVE_NEEDARP);
1da177e4
LT		 1965
1966 kfree(slave);
1967
1968 return 0; /* deletion OK */
1969}
1970
d90a162a
MS		 1971/*
1972* Destroy a bonding device.
1973* Must be under rtnl_lock when this function is called.
1974*/
9e71626c 1975static void bond_uninit(struct net_device *bond_dev)
a434e43f 1976{
244ef9b9 1977 struct bonding *bond = netdev_priv(bond_dev);
a434e43f 1978
9e71626c
SH		 1979 bond_deinit(bond_dev);
1980 bond_destroy_sysfs_entry(bond);
1981
a434e43f
JV		 1982 if (bond->wq)
1983 destroy_workqueue(bond->wq);
1984
1985 netif_addr_lock_bh(bond_dev);
1986 bond_mc_list_destroy(bond);
1987 netif_addr_unlock_bh(bond_dev);
a434e43f
JV		 1988}
1989
d90a162a
MS		 1990/*
1991* First release a slave and than destroy the bond if no more slaves iare left.
1992* Must be under rtnl_lock when this function is called.
1993*/
1994int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1995{
454d7c9b 1996 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 1997 int ret;
1998
1999 ret = bond_release(bond_dev, slave_dev);
2000 if ((ret == 0) && (bond->slave_cnt == 0)) {
2001 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2002 bond_dev->name, bond_dev->name);
9e71626c 2003 unregister_netdevice(bond_dev);
d90a162a
MS		 2004 }
2005 return ret;
2006}
2007
1da177e4
LT		 2008/*
2009 * This function releases all slaves.
2010 */
2011static int bond_release_all(struct net_device *bond_dev)
2012{
454d7c9b 2013 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2014 struct slave *slave;
2015 struct net_device *slave_dev;
2016 struct sockaddr addr;
2017
2018 write_lock_bh(&bond->lock);
2019
ff59c456
JV		 2020 netif_carrier_off(bond_dev);
2021
1da177e4
LT		 2022 if (bond->slave_cnt == 0) {
2023 goto out;
2024 }
2025
2026 bond->current_arp_slave = NULL;
2027 bond->primary_slave = NULL;
2028 bond_change_active_slave(bond, NULL);
2029
2030 while ((slave = bond->first_slave) != NULL) {
2031 /* Inform AD package of unbinding of slave
2032 * before slave is detached from the list.
2033 */
2034 if (bond->params.mode == BOND_MODE_8023AD) {
2035 bond_3ad_unbind_slave(slave);
2036 }
2037
2038 slave_dev = slave->dev;
2039 bond_detach_slave(bond, slave);
2040
2543331d
JV		 2041 /* now that the slave is detached, unlock and perform
2042 * all the undo steps that should not be called from
2043 * within a lock.
2044 */
2045 write_unlock_bh(&bond->lock);
2046
58402054 2047 if (bond_is_lb(bond)) {
1da177e4
LT		 2048 /* must be called only after the slave
2049 * has been detached from the list
2050 */
2051 bond_alb_deinit_slave(bond, slave);
2052 }
2053
8531c5ff
AK		 2054 bond_compute_features(bond);
2055
b76cdba9 2056 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2057 bond_del_vlans_from_slave(bond, slave_dev);
2058
2059 /* If the mode USES_PRIMARY, then we should only remove its
2060 * promisc and mc settings if it was the curr_active_slave, but that was
2061 * already taken care of above when we detached the slave
2062 */
2063 if (!USES_PRIMARY(bond->params.mode)) {
2064 /* unset promiscuity level from slave */
2065 if (bond_dev->flags & IFF_PROMISC) {
2066 dev_set_promiscuity(slave_dev, -1);
2067 }
2068
2069 /* unset allmulti level from slave */
2070 if (bond_dev->flags & IFF_ALLMULTI) {
2071 dev_set_allmulti(slave_dev, -1);
2072 }
2073
2074 /* flush master's mc_list from slave */
b9e40857 2075 netif_addr_lock_bh(bond_dev);
1da177e4 2076 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2077 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2078 }
2079
2080 netdev_set_master(slave_dev, NULL);
2081
2082 /* close slave before restoring its mac address */
2083 dev_close(slave_dev);
2084
dd957c57 2085 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2086 /* restore original ("permanent") mac address*/
2087 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2088 addr.sa_family = slave_dev->type;
2089 dev_set_mac_address(slave_dev, &addr);
2090 }
1da177e4 2091
8f903c70
JV		 2092 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2093 IFF_SLAVE_INACTIVE);
1da177e4
LT		 2094
2095 kfree(slave);
2096
2097 /* re-acquire the lock before getting the next slave */
2098 write_lock_bh(&bond->lock);
2099 }
2100
2101 /* zero the mac address of the master so it will be
2102 * set by the application to the mac address of the
2103 * first slave
2104 */
2105 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2106
2107 if (list_empty(&bond->vlan_list)) {
2108 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2109 } else {
2110 printk(KERN_WARNING DRV_NAME
4e0952c7 2111 ": %s: Warning: clearing HW address of %s while it "
1da177e4 2112 "still has VLANs.\n",
4e0952c7 2113 bond_dev->name, bond_dev->name);
1da177e4 2114 printk(KERN_WARNING DRV_NAME
4e0952c7
MW		 2115 ": %s: When re-adding slaves, make sure the bond's "
2116 "HW address matches its VLANs'.\n",
2117 bond_dev->name);
1da177e4
LT		 2118 }
2119
2120 printk(KERN_INFO DRV_NAME
2121 ": %s: released all slaves\n",
2122 bond_dev->name);
2123
2124out:
2125 write_unlock_bh(&bond->lock);
2126
2127 return 0;
2128}
2129
2130/*
2131 * This function changes the active slave to slave <slave_dev>.
2132 * It returns -EINVAL in the following cases.
2133 * - <slave_dev> is not found in the list.
2134 * - There is not active slave now.
2135 * - <slave_dev> is already active.
2136 * - The link state of <slave_dev> is not BOND_LINK_UP.
2137 * - <slave_dev> is not running.
2138 * In these cases, this fuction does nothing.
2139 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2140 */
2141static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2142{
454d7c9b 2143 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2144 struct slave *old_active = NULL;
2145 struct slave *new_active = NULL;
2146 int res = 0;
2147
2148 if (!USES_PRIMARY(bond->params.mode)) {
2149 return -EINVAL;
2150 }
2151
2152 /* Verify that master_dev is indeed the master of slave_dev */
2153 if (!(slave_dev->flags & IFF_SLAVE) ||
2154 (slave_dev->master != bond_dev)) {
2155 return -EINVAL;
2156 }
2157
059fe7a5 2158 read_lock(&bond->lock);
1da177e4 2159
059fe7a5 2160 read_lock(&bond->curr_slave_lock);
1da177e4 2161 old_active = bond->curr_active_slave;
059fe7a5
JV		 2162 read_unlock(&bond->curr_slave_lock);
2163
1da177e4
LT		 2164 new_active = bond_get_slave_by_dev(bond, slave_dev);
2165
2166 /*
2167 * Changing to the current active: do nothing; return success.
2168 */
2169 if (new_active && (new_active == old_active)) {
059fe7a5 2170 read_unlock(&bond->lock);
1da177e4
LT		 2171 return 0;
2172 }
2173
2174 if ((new_active) &&
2175 (old_active) &&
2176 (new_active->link == BOND_LINK_UP) &&
2177 IS_UP(new_active->dev)) {
059fe7a5 2178 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2179 bond_change_active_slave(bond, new_active);
059fe7a5 2180 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2181 } else {
2182 res = -EINVAL;
2183 }
2184
059fe7a5 2185 read_unlock(&bond->lock);
1da177e4
LT		 2186
2187 return res;
2188}
2189
1da177e4
LT		 2190static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2191{
454d7c9b 2192 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2193
2194 info->bond_mode = bond->params.mode;
2195 info->miimon = bond->params.miimon;
2196
6603a6f2 2197 read_lock(&bond->lock);
1da177e4 2198 info->num_slaves = bond->slave_cnt;
6603a6f2 2199 read_unlock(&bond->lock);
1da177e4
LT		 2200
2201 return 0;
2202}
2203
2204static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2205{
454d7c9b 2206 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 2207 struct slave *slave;
689c96cc 2208 int i, res = -ENODEV;
1da177e4 2209
6603a6f2 2210 read_lock(&bond->lock);
1da177e4
LT		 2211
2212 bond_for_each_slave(bond, slave, i) {
2213 if (i == (int)info->slave_id) {
689c96cc
ED		 2214 res = 0;
2215 strcpy(info->slave_name, slave->dev->name);
2216 info->link = slave->link;
2217 info->state = slave->state;
2218 info->link_failure_count = slave->link_failure_count;
1da177e4
LT		 2219 break;
2220 }
2221 }
2222
6603a6f2 2223 read_unlock(&bond->lock);
1da177e4 2224
689c96cc 2225 return res;
1da177e4
LT		 2226}
2227
2228/*-------------------------------- Monitoring -------------------------------*/
2229
1da177e4 2230
f0c76d61
JV		 2231static int bond_miimon_inspect(struct bonding *bond)
2232{
2233 struct slave *slave;
2234 int i, link_state, commit = 0;
41f89100
JP		 2235 bool ignore_updelay;
2236
2237 ignore_updelay = !bond->curr_active_slave ? true : false;
1da177e4
LT		 2238
2239 bond_for_each_slave(bond, slave, i) {
f0c76d61 2240 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2241
f0c76d61 2242 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2243
2244 switch (slave->link) {
f0c76d61
JV		 2245 case BOND_LINK_UP:
2246 if (link_state)
2247 continue;
1da177e4 2248
f0c76d61
JV		 2249 slave->link = BOND_LINK_FAIL;
2250 slave->delay = bond->params.downdelay;
2251 if (slave->delay) {
2252 printk(KERN_INFO DRV_NAME
2253 ": %s: link status down for %s"
2254 "interface %s, disabling it in %d ms.\n",
2255 bond->dev->name,
2256 (bond->params.mode ==
2257 BOND_MODE_ACTIVEBACKUP) ?
2258 ((slave->state == BOND_STATE_ACTIVE) ?
2259 "active " : "backup ") : "",
2260 slave->dev->name,
2261 bond->params.downdelay * bond->params.miimon);
1da177e4 2262 }
f0c76d61
JV		 2263 /*FALLTHRU*/
2264 case BOND_LINK_FAIL:
2265 if (link_state) {
2266 /*
2267 * recovered before downdelay expired
2268 */
2269 slave->link = BOND_LINK_UP;
1da177e4
LT		 2270 slave->jiffies = jiffies;
2271 printk(KERN_INFO DRV_NAME
2272 ": %s: link status up again after %d "
2273 "ms for interface %s.\n",
1b76b316 2274 bond->dev->name,
f0c76d61
JV		 2275 (bond->params.downdelay - slave->delay) *
2276 bond->params.miimon,
2277 slave->dev->name);
2278 continue;
1da177e4 2279 }
f0c76d61
JV		 2280
2281 if (slave->delay <= 0) {
2282 slave->new_link = BOND_LINK_DOWN;
2283 commit++;
2284 continue;
1da177e4 2285 }
1da177e4 2286
f0c76d61
JV		 2287 slave->delay--;
2288 break;
2289
2290 case BOND_LINK_DOWN:
2291 if (!link_state)
2292 continue;
2293
2294 slave->link = BOND_LINK_BACK;
2295 slave->delay = bond->params.updelay;
2296
2297 if (slave->delay) {
2298 printk(KERN_INFO DRV_NAME
2299 ": %s: link status up for "
2300 "interface %s, enabling it in %d ms.\n",
2301 bond->dev->name, slave->dev->name,
41f89100 2302 ignore_updelay ? 0 :
f0c76d61
JV		 2303 bond->params.updelay *
2304 bond->params.miimon);
2305 }
2306 /*FALLTHRU*/
2307 case BOND_LINK_BACK:
2308 if (!link_state) {
2309 slave->link = BOND_LINK_DOWN;
1da177e4
LT		 2310 printk(KERN_INFO DRV_NAME
2311 ": %s: link status down again after %d "
2312 "ms for interface %s.\n",
1b76b316 2313 bond->dev->name,
f0c76d61
JV		 2314 (bond->params.updelay - slave->delay) *
2315 bond->params.miimon,
2316 slave->dev->name);
2317
2318 continue;
2319 }
2320
41f89100
JP		 2321 if (ignore_updelay)
2322 slave->delay = 0;
2323
f0c76d61
JV		 2324 if (slave->delay <= 0) {
2325 slave->new_link = BOND_LINK_UP;
2326 commit++;
41f89100 2327 ignore_updelay = false;
f0c76d61 2328 continue;
1da177e4 2329 }
f0c76d61
JV		 2330
2331 slave->delay--;
1da177e4 2332 break;
f0c76d61
JV		 2333 }
2334 }
1da177e4 2335
f0c76d61
JV		 2336 return commit;
2337}
1da177e4 2338
f0c76d61
JV		 2339static void bond_miimon_commit(struct bonding *bond)
2340{
2341 struct slave *slave;
2342 int i;
2343
2344 bond_for_each_slave(bond, slave, i) {
2345 switch (slave->new_link) {
2346 case BOND_LINK_NOCHANGE:
2347 continue;
1da177e4 2348
f0c76d61
JV		 2349 case BOND_LINK_UP:
2350 slave->link = BOND_LINK_UP;
2351 slave->jiffies = jiffies;
2352
2353 if (bond->params.mode == BOND_MODE_8023AD) {
2354 /* prevent it from being the active one */
2355 slave->state = BOND_STATE_BACKUP;
2356 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2357 /* make it immediately active */
2358 slave->state = BOND_STATE_ACTIVE;
2359 } else if (slave != bond->primary_slave) {
2360 /* prevent it from being the active one */
2361 slave->state = BOND_STATE_BACKUP;
1da177e4 2362 }
1da177e4 2363
f0c76d61
JV		 2364 printk(KERN_INFO DRV_NAME
2365 ": %s: link status definitely "
2366 "up for interface %s.\n",
2367 bond->dev->name, slave->dev->name);
1da177e4 2368
f0c76d61
JV		 2369 /* notify ad that the link status has changed */
2370 if (bond->params.mode == BOND_MODE_8023AD)
2371 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2372
58402054 2373 if (bond_is_lb(bond))
f0c76d61
JV		 2374 bond_alb_handle_link_change(bond, slave,
2375 BOND_LINK_UP);
1da177e4 2376
f0c76d61
JV		 2377 if (!bond->curr_active_slave ||
2378 (slave == bond->primary_slave))
2379 goto do_failover;
1da177e4 2380
f0c76d61 2381 continue;
059fe7a5 2382
f0c76d61 2383 case BOND_LINK_DOWN:
fba4acda
JV		 2384 if (slave->link_failure_count < UINT_MAX)
2385 slave->link_failure_count++;
2386
f0c76d61 2387 slave->link = BOND_LINK_DOWN;
1da177e4 2388
f0c76d61
JV		 2389 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2390 bond->params.mode == BOND_MODE_8023AD)
2391 bond_set_slave_inactive_flags(slave);
2392
2393 printk(KERN_INFO DRV_NAME
2394 ": %s: link status definitely down for "
2395 "interface %s, disabling it\n",
2396 bond->dev->name, slave->dev->name);
2397
2398 if (bond->params.mode == BOND_MODE_8023AD)
2399 bond_3ad_handle_link_change(slave,
2400 BOND_LINK_DOWN);
2401
ae63e808 2402 if (bond_is_lb(bond))
f0c76d61
JV		 2403 bond_alb_handle_link_change(bond, slave,
2404 BOND_LINK_DOWN);
2405
2406 if (slave == bond->curr_active_slave)
2407 goto do_failover;
2408
2409 continue;
2410
2411 default:
2412 printk(KERN_ERR DRV_NAME
2413 ": %s: invalid new link %d on slave %s\n",
2414 bond->dev->name, slave->new_link,
2415 slave->dev->name);
2416 slave->new_link = BOND_LINK_NOCHANGE;
2417
2418 continue;
2419 }
2420
2421do_failover:
2422 ASSERT_RTNL();
2423 write_lock_bh(&bond->curr_slave_lock);
2424 bond_select_active_slave(bond);
2425 write_unlock_bh(&bond->curr_slave_lock);
2426 }
2427
2428 bond_set_carrier(bond);
1da177e4
LT		 2429}
2430
0b0eef66
JV		 2431/*
2432 * bond_mii_monitor
2433 *
2434 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2435 * inspection, then (if inspection indicates something needs to be done)
2436 * an acquisition of appropriate locks followed by a commit phase to
2437 * implement whatever link state changes are indicated.
0b0eef66
JV		 2438 */
2439void bond_mii_monitor(struct work_struct *work)
2440{
2441 struct bonding *bond = container_of(work, struct bonding,
2442 mii_work.work);
0b0eef66
JV		 2443
2444 read_lock(&bond->lock);
f0c76d61
JV		 2445 if (bond->kill_timers)
2446 goto out;
2447
2448 if (bond->slave_cnt == 0)
2449 goto re_arm;
b59f9f74
JV		 2450
2451 if (bond->send_grat_arp) {
2452 read_lock(&bond->curr_slave_lock);
2453 bond_send_gratuitous_arp(bond);
2454 read_unlock(&bond->curr_slave_lock);
2455 }
2456
305d552a
BH		 2457 if (bond->send_unsol_na) {
2458 read_lock(&bond->curr_slave_lock);
2459 bond_send_unsolicited_na(bond);
2460 read_unlock(&bond->curr_slave_lock);
2461 }
2462
f0c76d61 2463 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2464 read_unlock(&bond->lock);
2465 rtnl_lock();
2466 read_lock(&bond->lock);
f0c76d61
JV		 2467
2468 bond_miimon_commit(bond);
2469
5655662d
JV		 2470 read_unlock(&bond->lock);
2471 rtnl_unlock(); /* might sleep, hold no other locks */
2472 read_lock(&bond->lock);
0b0eef66
JV		 2473 }
2474
f0c76d61
JV		 2475re_arm:
2476 if (bond->params.miimon)
2477 queue_delayed_work(bond->wq, &bond->mii_work,
2478 msecs_to_jiffies(bond->params.miimon));
2479out:
0b0eef66 2480 read_unlock(&bond->lock);
0b0eef66 2481}
c3ade5ca 2482
d3bb52b0 2483static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2484{
2485 struct in_device *idev;
2486 struct in_ifaddr *ifa;
a144ea4b 2487 __be32 addr = 0;
c3ade5ca
JV		 2488
2489 if (!dev)
2490 return 0;
2491
2492 rcu_read_lock();
e5ed6399 2493 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2494 if (!idev)
2495 goto out;
2496
2497 ifa = idev->ifa_list;
2498 if (!ifa)
2499 goto out;
2500
2501 addr = ifa->ifa_local;
2502out:
2503 rcu_read_unlock();
2504 return addr;
2505}
2506
d3bb52b0 2507static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2508{
0883beca 2509 struct vlan_entry *vlan;
f5b2b966
JV		 2510
2511 if (ip == bond->master_ip)
2512 return 1;
2513
0883beca 2514 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2515 if (ip == vlan->vlan_ip)
2516 return 1;
2517 }
2518
2519 return 0;
2520}
2521
c3ade5ca
JV		 2522/*
2523 * We go to the (large) trouble of VLAN tagging ARP frames because
2524 * switches in VLAN mode (especially if ports are configured as
2525 * "native" to a VLAN) might not pass non-tagged frames.
2526 */
d3bb52b0 2527static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2528{
2529 struct sk_buff *skb;
2530
5a03cdb7 2531 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
c3ade5ca
JV		 2532 slave_dev->name, dest_ip, src_ip, vlan_id);
2533
2534 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2535 NULL, slave_dev->dev_addr, NULL);
2536
2537 if (!skb) {
2538 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2539 return;
2540 }
2541 if (vlan_id) {
2542 skb = vlan_put_tag(skb, vlan_id);
2543 if (!skb) {
2544 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2545 return;
2546 }
2547 }
2548 arp_xmit(skb);
2549}
2550
2551
1da177e4
LT		 2552static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2553{
c3ade5ca 2554 int i, vlan_id, rv;
d3bb52b0 2555 __be32 *targets = bond->params.arp_targets;
0883beca 2556 struct vlan_entry *vlan;
c3ade5ca
JV		 2557 struct net_device *vlan_dev;
2558 struct flowi fl;
2559 struct rtable *rt;
1da177e4 2560
6b780567
MW		 2561 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2562 if (!targets[i])
5a31bec0 2563 break;
5a03cdb7 2564 pr_debug("basa: target %x\n", targets[i]);
c3ade5ca 2565 if (list_empty(&bond->vlan_list)) {
5a03cdb7 2566 pr_debug("basa: empty vlan: arp_send\n");
c3ade5ca
JV		 2567 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2568 bond->master_ip, 0);
2569 continue;
2570 }
2571
2572 /*
2573 * If VLANs are configured, we do a route lookup to
2574 * determine which VLAN interface would be used, so we
2575 * can tag the ARP with the proper VLAN tag.
2576 */
2577 memset(&fl, 0, sizeof(fl));
2578 fl.fl4_dst = targets[i];
2579 fl.fl4_tos = RTO_ONLINK;
2580
f206351a 2581 rv = ip_route_output_key(&init_net, &rt, &fl);
c3ade5ca
JV		 2582 if (rv) {
2583 if (net_ratelimit()) {
2584 printk(KERN_WARNING DRV_NAME
63779436
HH		 2585 ": %s: no route to arp_ip_target %pI4\n",
2586 bond->dev->name, &fl.fl4_dst);
c3ade5ca
JV		 2587 }
2588 continue;
2589 }
2590
2591 /*
2592 * This target is not on a VLAN
2593 */
2594 if (rt->u.dst.dev == bond->dev) {
ed4b9f80 2595 ip_rt_put(rt);
5a03cdb7 2596 pr_debug("basa: rtdev == bond->dev: arp_send\n");
c3ade5ca
JV		 2597 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2598 bond->master_ip, 0);
2599 continue;
2600 }
2601
2602 vlan_id = 0;
0883beca 2603 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2604 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2605 if (vlan_dev == rt->u.dst.dev) {
2606 vlan_id = vlan->vlan_id;
5a03cdb7 2607 pr_debug("basa: vlan match on %s %d\n",
c3ade5ca
JV		 2608 vlan_dev->name, vlan_id);
2609 break;
2610 }
2611 }
2612
2613 if (vlan_id) {
ed4b9f80 2614 ip_rt_put(rt);
c3ade5ca
JV		 2615 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2616 vlan->vlan_ip, vlan_id);
2617 continue;
2618 }
2619
2620 if (net_ratelimit()) {
2621 printk(KERN_WARNING DRV_NAME
63779436
HH		 2622 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2623 bond->dev->name, &fl.fl4_dst,
c3ade5ca
JV		 2624 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2625 }
ed4b9f80 2626 ip_rt_put(rt);
c3ade5ca
JV		 2627 }
2628}
2629
2630/*
2631 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2632 * for each VLAN above us.
b59f9f74
JV		 2633 *
2634 * Caller must hold curr_slave_lock for read or better
c3ade5ca
JV		 2635 */
2636static void bond_send_gratuitous_arp(struct bonding *bond)
2637{
2638 struct slave *slave = bond->curr_active_slave;
2639 struct vlan_entry *vlan;
2640 struct net_device *vlan_dev;
2641
5a03cdb7 2642 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
c3ade5ca 2643 slave ? slave->dev->name : "NULL");
b59f9f74
JV		 2644
2645 if (!slave || !bond->send_grat_arp ||
2646 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
c3ade5ca
JV		 2647 return;
2648
b59f9f74
JV		 2649 bond->send_grat_arp--;
2650
c3ade5ca
JV		 2651 if (bond->master_ip) {
2652 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
1053f62c 2653 bond->master_ip, 0);
c3ade5ca
JV		 2654 }
2655
2656 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2657 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 2658 if (vlan->vlan_ip) {
2659 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2660 vlan->vlan_ip, vlan->vlan_id);
2661 }
1da177e4
LT		 2662 }
2663}
2664
d3bb52b0 2665static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2666{
2667 int i;
d3bb52b0 2668 __be32 *targets = bond->params.arp_targets;
f5b2b966 2669
f5b2b966 2670 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
5a03cdb7 2671 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
63779436 2672 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
f5b2b966
JV		 2673 if (sip == targets[i]) {
2674 if (bond_has_this_ip(bond, tip))
2675 slave->last_arp_rx = jiffies;
2676 return;
2677 }
2678 }
2679}
2680
2681static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2682{
2683 struct arphdr *arp;
2684 struct slave *slave;
2685 struct bonding *bond;
2686 unsigned char *arp_ptr;
d3bb52b0 2687 __be32 sip, tip;
f5b2b966 2688
c346dca1 2689 if (dev_net(dev) != &init_net)
e730c155
EB		 2690 goto out;
2691
f5b2b966
JV		 2692 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2693 goto out;
2694
454d7c9b 2695 bond = netdev_priv(dev);
f5b2b966
JV		 2696 read_lock(&bond->lock);
2697
5a03cdb7 2698 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
f5b2b966
JV		 2699 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2700 orig_dev ? orig_dev->name : "NULL");
2701
2702 slave = bond_get_slave_by_dev(bond, orig_dev);
2703 if (!slave || !slave_do_arp_validate(bond, slave))
2704 goto out_unlock;
2705
988b7050 2706 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
f5b2b966
JV		 2707 goto out_unlock;
2708
d0a92be0 2709 arp = arp_hdr(skb);
f5b2b966
JV		 2710 if (arp->ar_hln != dev->addr_len ||
2711 skb->pkt_type == PACKET_OTHERHOST ||
2712 skb->pkt_type == PACKET_LOOPBACK ||
2713 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2714 arp->ar_pro != htons(ETH_P_IP) ||
2715 arp->ar_pln != 4)
2716 goto out_unlock;
2717
2718 arp_ptr = (unsigned char *)(arp + 1);
2719 arp_ptr += dev->addr_len;
2720 memcpy(&sip, arp_ptr, 4);
2721 arp_ptr += 4 + dev->addr_len;
2722 memcpy(&tip, arp_ptr, 4);
2723
5a03cdb7 2724 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
63779436
HH		 2725 bond->dev->name, slave->dev->name, slave->state,
2726 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2727 &sip, &tip);
f5b2b966
JV		 2728
2729 /*
2730 * Backup slaves won't see the ARP reply, but do come through
2731 * here for each ARP probe (so we swap the sip/tip to validate
2732 * the probe). In a "redundant switch, common router" type of
2733 * configuration, the ARP probe will (hopefully) travel from
2734 * the active, through one switch, the router, then the other
2735 * switch before reaching the backup.
2736 */
2737 if (slave->state == BOND_STATE_ACTIVE)
2738 bond_validate_arp(bond, slave, sip, tip);
2739 else
2740 bond_validate_arp(bond, slave, tip, sip);
2741
2742out_unlock:
2743 read_unlock(&bond->lock);
2744out:
2745 dev_kfree_skb(skb);
2746 return NET_RX_SUCCESS;
2747}
2748
1da177e4
LT		 2749/*
2750 * this function is called regularly to monitor each slave's link
2751 * ensuring that traffic is being sent and received when arp monitoring
2752 * is used in load-balancing mode. if the adapter has been dormant, then an
2753 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2754 * arp monitoring in active backup mode.
2755 */
1b76b316 2756void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2757{
1b76b316
JV		 2758 struct bonding *bond = container_of(work, struct bonding,
2759 arp_work.work);
1da177e4
LT		 2760 struct slave *slave, *oldcurrent;
2761 int do_failover = 0;
2762 int delta_in_ticks;
2763 int i;
2764
2765 read_lock(&bond->lock);
2766
5ce0da8f 2767 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4
LT		 2768
2769 if (bond->kill_timers) {
2770 goto out;
2771 }
2772
2773 if (bond->slave_cnt == 0) {
2774 goto re_arm;
2775 }
2776
2777 read_lock(&bond->curr_slave_lock);
2778 oldcurrent = bond->curr_active_slave;
2779 read_unlock(&bond->curr_slave_lock);
2780
2781 /* see if any of the previous devices are up now (i.e. they have
2782 * xmt and rcv traffic). the curr_active_slave does not come into
2783 * the picture unless it is null. also, slave->jiffies is not needed
2784 * here because we send an arp on each slave and give a slave as
2785 * long as it needs to get the tx/rx within the delta.
2786 * TODO: what about up/down delay in arp mode? it wasn't here before
2787 * so it can wait
2788 */
2789 bond_for_each_slave(bond, slave, i) {
2790 if (slave->link != BOND_LINK_UP) {
9d21493b 2791 if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
b63bb739 2792 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2793
2794 slave->link = BOND_LINK_UP;
2795 slave->state = BOND_STATE_ACTIVE;
2796
2797 /* primary_slave has no meaning in round-robin
2798 * mode. the window of a slave being up and
2799 * curr_active_slave being null after enslaving
2800 * is closed.
2801 */
2802 if (!oldcurrent) {
2803 printk(KERN_INFO DRV_NAME
2804 ": %s: link status definitely "
2805 "up for interface %s, ",
1b76b316 2806 bond->dev->name,
1da177e4
LT		 2807 slave->dev->name);
2808 do_failover = 1;
2809 } else {
2810 printk(KERN_INFO DRV_NAME
2811 ": %s: interface %s is now up\n",
1b76b316 2812 bond->dev->name,
1da177e4
LT		 2813 slave->dev->name);
2814 }
2815 }
2816 } else {
2817 /* slave->link == BOND_LINK_UP */
2818
2819 /* not all switches will respond to an arp request
2820 * when the source ip is 0, so don't take the link down
2821 * if we don't know our ip yet
2822 */
9d21493b 2823 if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
4b8a9239 2824 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
1da177e4
LT		 2825
2826 slave->link = BOND_LINK_DOWN;
2827 slave->state = BOND_STATE_BACKUP;
2828
2829 if (slave->link_failure_count < UINT_MAX) {
2830 slave->link_failure_count++;
2831 }
2832
2833 printk(KERN_INFO DRV_NAME
2834 ": %s: interface %s is now down.\n",
1b76b316 2835 bond->dev->name,
1da177e4
LT		 2836 slave->dev->name);
2837
2838 if (slave == oldcurrent) {
2839 do_failover = 1;
2840 }
2841 }
2842 }
2843
2844 /* note: if switch is in round-robin mode, all links
2845 * must tx arp to ensure all links rx an arp - otherwise
2846 * links may oscillate or not come up at all; if switch is
2847 * in something like xor mode, there is nothing we can
2848 * do - all replies will be rx'ed on same link causing slaves
2849 * to be unstable during low/no traffic periods
2850 */
2851 if (IS_UP(slave->dev)) {
2852 bond_arp_send_all(bond, slave);
2853 }
2854 }
2855
2856 if (do_failover) {
059fe7a5 2857 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2858
2859 bond_select_active_slave(bond);
2860
059fe7a5 2861 write_unlock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2862 }
2863
2864re_arm:
1b76b316
JV		 2865 if (bond->params.arp_interval)
2866 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2867out:
2868 read_unlock(&bond->lock);
2869}
2870
2871/*
b2220cad
JV		 2872 * Called to inspect slaves for active-backup mode ARP monitor link state
2873 * changes. Sets new_link in slaves to specify what action should take
2874 * place for the slave. Returns 0 if no changes are found, >0 if changes
2875 * to link states must be committed.
2876 *
2877 * Called with bond->lock held for read.
1da177e4 2878 */
b2220cad 2879static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2880{
1da177e4 2881 struct slave *slave;
b2220cad 2882 int i, commit = 0;
1da177e4 2883
b2220cad
JV		 2884 bond_for_each_slave(bond, slave, i) {
2885 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2886
b2220cad
JV		 2887 if (slave->link != BOND_LINK_UP) {
2888 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2889 delta_in_ticks)) {
2890 slave->new_link = BOND_LINK_UP;
2891 commit++;
2892 }
1da177e4 2893
b2220cad
JV		 2894 continue;
2895 }
1da177e4 2896
b2220cad
JV		 2897 /*
2898 * Give slaves 2*delta after being enslaved or made
2899 * active. This avoids bouncing, as the last receive
2900 * times need a full ARP monitor cycle to be updated.
2901 */
2902 if (!time_after_eq(jiffies, slave->jiffies +
2903 2 * delta_in_ticks))
2904 continue;
2905
2906 /*
2907 * Backup slave is down if:
2908 * - No current_arp_slave AND
2909 * - more than 3*delta since last receive AND
2910 * - the bond has an IP address
2911 *
2912 * Note: a non-null current_arp_slave indicates
2913 * the curr_active_slave went down and we are
2914 * searching for a new one; under this condition
2915 * we only take the curr_active_slave down - this
2916 * gives each slave a chance to tx/rx traffic
2917 * before being taken out
2918 */
2919 if (slave->state == BOND_STATE_BACKUP &&
2920 !bond->current_arp_slave &&
2921 time_after(jiffies, slave_last_rx(bond, slave) +
2922 3 * delta_in_ticks)) {
2923 slave->new_link = BOND_LINK_DOWN;
2924 commit++;
2925 }
2926
2927 /*
2928 * Active slave is down if:
2929 * - more than 2*delta since transmitting OR
2930 * - (more than 2*delta since receive AND
2931 * the bond has an IP address)
2932 */
2933 if ((slave->state == BOND_STATE_ACTIVE) &&
9d21493b 2934 (time_after_eq(jiffies, dev_trans_start(slave->dev) +
b2220cad
JV		 2935 2 * delta_in_ticks) ||
2936 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2937 + 2 * delta_in_ticks)))) {
2938 slave->new_link = BOND_LINK_DOWN;
2939 commit++;
2940 }
1da177e4
LT		 2941 }
2942
b2220cad
JV		 2943 read_lock(&bond->curr_slave_lock);
2944
2945 /*
2946 * Trigger a commit if the primary option setting has changed.
1da177e4 2947 */
b2220cad
JV		 2948 if (bond->primary_slave &&
2949 (bond->primary_slave != bond->curr_active_slave) &&
2950 (bond->primary_slave->link == BOND_LINK_UP))
2951 commit++;
1da177e4 2952
b2220cad 2953 read_unlock(&bond->curr_slave_lock);
1da177e4 2954
b2220cad
JV		 2955 return commit;
2956}
1da177e4 2957
b2220cad
JV		 2958/*
2959 * Called to commit link state changes noted by inspection step of
2960 * active-backup mode ARP monitor.
2961 *
2962 * Called with RTNL and bond->lock for read.
2963 */
2964static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2965{
2966 struct slave *slave;
2967 int i;
1da177e4 2968
b2220cad
JV		 2969 bond_for_each_slave(bond, slave, i) {
2970 switch (slave->new_link) {
2971 case BOND_LINK_NOCHANGE:
2972 continue;
ff59c456 2973
b2220cad
JV		 2974 case BOND_LINK_UP:
2975 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2976
b2220cad 2977 if (!bond->curr_active_slave &&
9d21493b 2978 time_before_eq(jiffies, dev_trans_start(slave->dev) +
b2220cad
JV		 2979 delta_in_ticks)) {
2980 slave->link = BOND_LINK_UP;
2981 bond_change_active_slave(bond, slave);
2982 bond->current_arp_slave = NULL;
2983
2984 printk(KERN_INFO DRV_NAME
2985 ": %s: %s is up and now the "
2986 "active interface\n",
2987 bond->dev->name, slave->dev->name);
2988
2989 } else if (bond->curr_active_slave != slave) {
2990 /* this slave has just come up but we
2991 * already have a current slave; this can
2992 * also happen if bond_enslave adds a new
2993 * slave that is up while we are searching
2994 * for a new slave
1da177e4 2995 */
b2220cad
JV		 2996 slave->link = BOND_LINK_UP;
2997 bond_set_slave_inactive_flags(slave);
2998 bond->current_arp_slave = NULL;
2999
3000 printk(KERN_INFO DRV_NAME
3001 ": %s: backup interface %s is now up\n",
3002 bond->dev->name, slave->dev->name);
3003 }
1da177e4 3004
b2220cad 3005 write_unlock_bh(&bond->curr_slave_lock);
1da177e4 3006
b2220cad 3007 break;
1da177e4 3008
b2220cad
JV		 3009 case BOND_LINK_DOWN:
3010 if (slave->link_failure_count < UINT_MAX)
3011 slave->link_failure_count++;
3012
3013 slave->link = BOND_LINK_DOWN;
3014
3015 if (slave == bond->curr_active_slave) {
3016 printk(KERN_INFO DRV_NAME
3017 ": %s: link status down for active "
3018 "interface %s, disabling it\n",
3019 bond->dev->name, slave->dev->name);
1da177e4
LT		 3020
3021 bond_set_slave_inactive_flags(slave);
3022
b2220cad
JV		 3023 write_lock_bh(&bond->curr_slave_lock);
3024
3025 bond_select_active_slave(bond);
3026 if (bond->curr_active_slave)
3027 bond->curr_active_slave->jiffies =
3028 jiffies;
3029
3030 write_unlock_bh(&bond->curr_slave_lock);
3031
3032 bond->current_arp_slave = NULL;
3033
3034 } else if (slave->state == BOND_STATE_BACKUP) {
1da177e4
LT		 3035 printk(KERN_INFO DRV_NAME
3036 ": %s: backup interface %s is now down\n",
b2220cad
JV		 3037 bond->dev->name, slave->dev->name);
3038
3039 bond_set_slave_inactive_flags(slave);
1da177e4 3040 }
b2220cad
JV		 3041 break;
3042
3043 default:
3044 printk(KERN_ERR DRV_NAME
3045 ": %s: impossible: new_link %d on slave %s\n",
3046 bond->dev->name, slave->new_link,
3047 slave->dev->name);
1da177e4
LT		 3048 }
3049 }
3050
b2220cad
JV		 3051 /*
3052 * No race with changes to primary via sysfs, as we hold rtnl.
3053 */
3054 if (bond->primary_slave &&
3055 (bond->primary_slave != bond->curr_active_slave) &&
3056 (bond->primary_slave->link == BOND_LINK_UP)) {
3057 write_lock_bh(&bond->curr_slave_lock);
3058 bond_change_active_slave(bond, bond->primary_slave);
3059 write_unlock_bh(&bond->curr_slave_lock);
3060 }
1da177e4 3061
b2220cad
JV		 3062 bond_set_carrier(bond);
3063}
1da177e4 3064
b2220cad
JV		 3065/*
3066 * Send ARP probes for active-backup mode ARP monitor.
3067 *
3068 * Called with bond->lock held for read.
3069 */
3070static void bond_ab_arp_probe(struct bonding *bond)
3071{
3072 struct slave *slave;
3073 int i;
1da177e4 3074
b2220cad 3075 read_lock(&bond->curr_slave_lock);
1da177e4 3076
b2220cad
JV		 3077 if (bond->current_arp_slave && bond->curr_active_slave)
3078 printk("PROBE: c_arp %s && cas %s BAD\n",
3079 bond->current_arp_slave->dev->name,
3080 bond->curr_active_slave->dev->name);
1da177e4 3081
b2220cad
JV		 3082 if (bond->curr_active_slave) {
3083 bond_arp_send_all(bond, bond->curr_active_slave);
3084 read_unlock(&bond->curr_slave_lock);
3085 return;
3086 }
1da177e4 3087
b2220cad 3088 read_unlock(&bond->curr_slave_lock);
059fe7a5 3089
b2220cad
JV		 3090 /* if we don't have a curr_active_slave, search for the next available
3091 * backup slave from the current_arp_slave and make it the candidate
3092 * for becoming the curr_active_slave
3093 */
1da177e4 3094
b2220cad
JV		 3095 if (!bond->current_arp_slave) {
3096 bond->current_arp_slave = bond->first_slave;
3097 if (!bond->current_arp_slave)
3098 return;
3099 }
1da177e4 3100
b2220cad 3101 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3102
b2220cad
JV		 3103 /* search for next candidate */
3104 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3105 if (IS_UP(slave->dev)) {
3106 slave->link = BOND_LINK_BACK;
3107 bond_set_slave_active_flags(slave);
3108 bond_arp_send_all(bond, slave);
1da177e4 3109 slave->jiffies = jiffies;
b2220cad
JV		 3110 bond->current_arp_slave = slave;
3111 break;
1da177e4
LT		 3112 }
3113
b2220cad
JV		 3114 /* if the link state is up at this point, we
3115 * mark it down - this can happen if we have
3116 * simultaneous link failures and
3117 * reselect_active_interface doesn't make this
3118 * one the current slave so it is still marked
3119 * up when it is actually down
1da177e4 3120 */
b2220cad
JV		 3121 if (slave->link == BOND_LINK_UP) {
3122 slave->link = BOND_LINK_DOWN;
3123 if (slave->link_failure_count < UINT_MAX)
3124 slave->link_failure_count++;
1da177e4 3125
b2220cad
JV		 3126 bond_set_slave_inactive_flags(slave);
3127
3128 printk(KERN_INFO DRV_NAME
3129 ": %s: backup interface %s is now down.\n",
3130 bond->dev->name, slave->dev->name);
1da177e4 3131 }
b2220cad
JV		 3132 }
3133}
1da177e4 3134
b2220cad
JV		 3135void bond_activebackup_arp_mon(struct work_struct *work)
3136{
3137 struct bonding *bond = container_of(work, struct bonding,
3138 arp_work.work);
3139 int delta_in_ticks;
1da177e4 3140
b2220cad 3141 read_lock(&bond->lock);
1da177e4 3142
b2220cad
JV		 3143 if (bond->kill_timers)
3144 goto out;
1da177e4 3145
b2220cad 3146 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3147
b2220cad
JV		 3148 if (bond->slave_cnt == 0)
3149 goto re_arm;
3150
b59f9f74
JV		 3151 if (bond->send_grat_arp) {
3152 read_lock(&bond->curr_slave_lock);
3153 bond_send_gratuitous_arp(bond);
3154 read_unlock(&bond->curr_slave_lock);
3155 }
3156
305d552a
BH		 3157 if (bond->send_unsol_na) {
3158 read_lock(&bond->curr_slave_lock);
3159 bond_send_unsolicited_na(bond);
3160 read_unlock(&bond->curr_slave_lock);
3161 }
3162
b2220cad
JV		 3163 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3164 read_unlock(&bond->lock);
3165 rtnl_lock();
3166 read_lock(&bond->lock);
3167
3168 bond_ab_arp_commit(bond, delta_in_ticks);
3169
3170 read_unlock(&bond->lock);
3171 rtnl_unlock();
3172 read_lock(&bond->lock);
1da177e4
LT		 3173 }
3174
b2220cad
JV		 3175 bond_ab_arp_probe(bond);
3176
1da177e4
LT		 3177re_arm:
3178 if (bond->params.arp_interval) {
1b76b316 3179 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3180 }
3181out:
3182 read_unlock(&bond->lock);
3183}
3184
3185/*------------------------------ proc/seq_file-------------------------------*/
3186
3187#ifdef CONFIG_PROC_FS
3188
1da177e4 3189static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
1f78d9f9
HE		 3190 __acquires(&dev_base_lock)
3191 __acquires(&bond->lock)
1da177e4
LT		 3192{
3193 struct bonding *bond = seq->private;
3194 loff_t off = 0;
3195 struct slave *slave;
3196 int i;
3197
3198 /* make sure the bond won't be taken away */
3199 read_lock(&dev_base_lock);
6603a6f2 3200 read_lock(&bond->lock);
1da177e4
LT		 3201
3202 if (*pos == 0) {
3203 return SEQ_START_TOKEN;
3204 }
3205
3206 bond_for_each_slave(bond, slave, i) {
3207 if (++off == *pos) {
3208 return slave;
3209 }
3210 }
3211
3212 return NULL;
3213}
3214
3215static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3216{
3217 struct bonding *bond = seq->private;
3218 struct slave *slave = v;
3219
3220 ++*pos;
3221 if (v == SEQ_START_TOKEN) {
3222 return bond->first_slave;
3223 }
3224
3225 slave = slave->next;
3226
3227 return (slave == bond->first_slave) ? NULL : slave;
3228}
3229
3230static void bond_info_seq_stop(struct seq_file *seq, void *v)
1f78d9f9
HE		 3231 __releases(&bond->lock)
3232 __releases(&dev_base_lock)
1da177e4
LT		 3233{
3234 struct bonding *bond = seq->private;
3235
6603a6f2 3236 read_unlock(&bond->lock);
1da177e4
LT		 3237 read_unlock(&dev_base_lock);
3238}
3239
3240static void bond_info_show_master(struct seq_file *seq)
3241{
3242 struct bonding *bond = seq->private;
3243 struct slave *curr;
4756b02f 3244 int i;
1da177e4
LT		 3245
3246 read_lock(&bond->curr_slave_lock);
3247 curr = bond->curr_active_slave;
3248 read_unlock(&bond->curr_slave_lock);
3249
dd957c57 3250 seq_printf(seq, "Bonding Mode: %s",
1da177e4
LT		 3251 bond_mode_name(bond->params.mode));
3252
dd957c57
JV		 3253 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3254 bond->params.fail_over_mac)
3915c1e8
JV		 3255 seq_printf(seq, " (fail_over_mac %s)",
3256 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
dd957c57
JV		 3257
3258 seq_printf(seq, "\n");
3259
c61b75ad
MW		 3260 if (bond->params.mode == BOND_MODE_XOR ||
3261 bond->params.mode == BOND_MODE_8023AD) {
3262 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3263 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3264 bond->params.xmit_policy);
3265 }
3266
1da177e4
LT		 3267 if (USES_PRIMARY(bond->params.mode)) {
3268 seq_printf(seq, "Primary Slave: %s\n",
0f418b2a
MW		 3269 (bond->primary_slave) ?
3270 bond->primary_slave->dev->name : "None");
1da177e4
LT		 3271
3272 seq_printf(seq, "Currently Active Slave: %s\n",
3273 (curr) ? curr->dev->name : "None");
3274 }
3275
ff59c456
JV		 3276 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3277 "up" : "down");
1da177e4
LT		 3278 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3279 seq_printf(seq, "Up Delay (ms): %d\n",
3280 bond->params.updelay * bond->params.miimon);
3281 seq_printf(seq, "Down Delay (ms): %d\n",
3282 bond->params.downdelay * bond->params.miimon);
3283
4756b02f
MW		 3284
3285 /* ARP information */
3286 if(bond->params.arp_interval > 0) {
3287 int printed=0;
3288 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3289 bond->params.arp_interval);
3290
3291 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3292
3293 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3294 if (!bond->params.arp_targets[i])
5a31bec0 3295 break;
4756b02f
MW		 3296 if (printed)
3297 seq_printf(seq, ",");
8cf14e38 3298 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
4756b02f
MW		 3299 printed = 1;
3300 }
3301 seq_printf(seq, "\n");
3302 }
3303
1da177e4
LT		 3304 if (bond->params.mode == BOND_MODE_8023AD) {
3305 struct ad_info ad_info;
3306
3307 seq_puts(seq, "\n802.3ad info\n");
3308 seq_printf(seq, "LACP rate: %s\n",
3309 (bond->params.lacp_fast) ? "fast" : "slow");
fd989c83
JV		 3310 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3311 ad_select_tbl[bond->params.ad_select].modename);
1da177e4
LT		 3312
3313 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3314 seq_printf(seq, "bond %s has no active aggregator\n",
3315 bond->dev->name);
3316 } else {
3317 seq_printf(seq, "Active Aggregator Info:\n");
3318
3319 seq_printf(seq, "\tAggregator ID: %d\n",
3320 ad_info.aggregator_id);
3321 seq_printf(seq, "\tNumber of ports: %d\n",
3322 ad_info.ports);
3323 seq_printf(seq, "\tActor Key: %d\n",
3324 ad_info.actor_key);
3325 seq_printf(seq, "\tPartner Key: %d\n",
3326 ad_info.partner_key);
e174961c
JB		 3327 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3328 ad_info.partner_system);
1da177e4
LT		 3329 }
3330 }
3331}
3332
3333static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3334{
3335 struct bonding *bond = seq->private;
3336
3337 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3338 seq_printf(seq, "MII Status: %s\n",
3339 (slave->link == BOND_LINK_UP) ? "up" : "down");
65509645 3340 seq_printf(seq, "Link Failure Count: %u\n",
1da177e4
LT		 3341 slave->link_failure_count);
3342
e174961c 3343 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
1da177e4
LT		 3344
3345 if (bond->params.mode == BOND_MODE_8023AD) {
3346 const struct aggregator *agg
3347 = SLAVE_AD_INFO(slave).port.aggregator;
3348
3349 if (agg) {
3350 seq_printf(seq, "Aggregator ID: %d\n",
3351 agg->aggregator_identifier);
3352 } else {
3353 seq_puts(seq, "Aggregator ID: N/A\n");
3354 }
3355 }
3356}
3357
3358static int bond_info_seq_show(struct seq_file *seq, void *v)
3359{
3360 if (v == SEQ_START_TOKEN) {
3361 seq_printf(seq, "%s\n", version);
3362 bond_info_show_master(seq);
3363 } else {
3364 bond_info_show_slave(seq, v);
3365 }
3366
3367 return 0;
3368}
3369
4101dec9 3370static const struct seq_operations bond_info_seq_ops = {
1da177e4
LT		 3371 .start = bond_info_seq_start,
3372 .next = bond_info_seq_next,
3373 .stop = bond_info_seq_stop,
3374 .show = bond_info_seq_show,
3375};
3376
3377static int bond_info_open(struct inode *inode, struct file *file)
3378{
3379 struct seq_file *seq;
3380 struct proc_dir_entry *proc;
3381 int res;
3382
3383 res = seq_open(file, &bond_info_seq_ops);
3384 if (!res) {
3385 /* recover the pointer buried in proc_dir_entry data */
3386 seq = file->private_data;
3387 proc = PDE(inode);
3388 seq->private = proc->data;
3389 }
3390
3391 return res;
3392}
3393
d54b1fdb 3394static const struct file_operations bond_info_fops = {
1da177e4
LT		 3395 .owner = THIS_MODULE,
3396 .open = bond_info_open,
3397 .read = seq_read,
3398 .llseek = seq_lseek,
3399 .release = seq_release,
3400};
3401
3402static int bond_create_proc_entry(struct bonding *bond)
3403{
3404 struct net_device *bond_dev = bond->dev;
3405
3406 if (bond_proc_dir) {
a95609cb
DL		 3407 bond->proc_entry = proc_create_data(bond_dev->name,
3408 S_IRUGO, bond_proc_dir,
3409 &bond_info_fops, bond);
1da177e4
LT		 3410 if (bond->proc_entry == NULL) {
3411 printk(KERN_WARNING DRV_NAME
3412 ": Warning: Cannot create /proc/net/%s/%s\n",
3413 DRV_NAME, bond_dev->name);
3414 } else {
1da177e4
LT		 3415 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3416 }
3417 }
3418
3419 return 0;
3420}
3421
3422static void bond_remove_proc_entry(struct bonding *bond)
3423{
3424 if (bond_proc_dir && bond->proc_entry) {
3425 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3426 memset(bond->proc_file_name, 0, IFNAMSIZ);
3427 bond->proc_entry = NULL;
3428 }
3429}
3430
3431/* Create the bonding directory under /proc/net, if doesn't exist yet.
3432 * Caller must hold rtnl_lock.
3433 */
3434static void bond_create_proc_dir(void)
3435{
1da177e4 3436 if (!bond_proc_dir) {
457c4cbc 3437 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
99b76233 3438 if (!bond_proc_dir)
1da177e4
LT		 3439 printk(KERN_WARNING DRV_NAME
3440 ": Warning: cannot create /proc/net/%s\n",
3441 DRV_NAME);
1da177e4
LT		 3442 }
3443}
3444
3445/* Destroy the bonding directory under /proc/net, if empty.
3446 * Caller must hold rtnl_lock.
3447 */
3448static void bond_destroy_proc_dir(void)
3449{
99b76233 3450 if (bond_proc_dir) {
457c4cbc 3451 remove_proc_entry(DRV_NAME, init_net.proc_net);
1da177e4
LT		 3452 bond_proc_dir = NULL;
3453 }
3454}
aee64faf
JP		 3455
3456#else /* !CONFIG_PROC_FS */
3457
3458static int bond_create_proc_entry(struct bonding *bond)
3459{
3460}
3461
3462static void bond_remove_proc_entry(struct bonding *bond)
3463{
3464}
3465
3466static void bond_create_proc_dir(void)
3467{
3468}
3469
3470static void bond_destroy_proc_dir(void)
3471{
3472}
3473
1da177e4
LT		 3474#endif /* CONFIG_PROC_FS */
3475
aee64faf 3476
1da177e4
LT		 3477/*-------------------------- netdev event handling --------------------------*/
3478
3479/*
3480 * Change device name
3481 */
3482static int bond_event_changename(struct bonding *bond)
3483{
1da177e4
LT		 3484 bond_remove_proc_entry(bond);
3485 bond_create_proc_entry(bond);
7e083840 3486
b76cdba9
MW		 3487 bond_destroy_sysfs_entry(bond);
3488 bond_create_sysfs_entry(bond);
7e083840 3489
1da177e4
LT		 3490 return NOTIFY_DONE;
3491}
3492
3493static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3494{
454d7c9b 3495 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 3496
3497 switch (event) {
3498 case NETDEV_CHANGENAME:
3499 return bond_event_changename(event_bond);
3500 case NETDEV_UNREGISTER:
3b96c858 3501 bond_release_all(event_bond->dev);
1da177e4
LT		 3502 break;
3503 default:
3504 break;
3505 }
3506
3507 return NOTIFY_DONE;
3508}
3509
3510static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3511{
3512 struct net_device *bond_dev = slave_dev->master;
454d7c9b 3513 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3514
3515 switch (event) {
3516 case NETDEV_UNREGISTER:
3517 if (bond_dev) {
1284cd3a
JV		 3518 if (bond->setup_by_slave)
3519 bond_release_and_destroy(bond_dev, slave_dev);
3520 else
3521 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3522 }
3523 break;
3524 case NETDEV_CHANGE:
17d04500
JV		 3525 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3526 struct slave *slave;
3527
3528 slave = bond_get_slave_by_dev(bond, slave_dev);
3529 if (slave) {
3530 u16 old_speed = slave->speed;
3531 u16 old_duplex = slave->duplex;
3532
3533 bond_update_speed_duplex(slave);
3534
3535 if (bond_is_lb(bond))
3536 break;
3537
3538 if (old_speed != slave->speed)
3539 bond_3ad_adapter_speed_changed(slave);
3540 if (old_duplex != slave->duplex)
3541 bond_3ad_adapter_duplex_changed(slave);
3542 }
3543 }
3544
1da177e4
LT		 3545 break;
3546 case NETDEV_DOWN:
3547 /*
3548 * ... Or is it this?
3549 */
3550 break;
3551 case NETDEV_CHANGEMTU:
3552 /*
3553 * TODO: Should slaves be allowed to
3554 * independently alter their MTU? For
3555 * an active-backup bond, slaves need
3556 * not be the same type of device, so
3557 * MTUs may vary. For other modes,
3558 * slaves arguably should have the
3559 * same MTUs. To do this, we'd need to
3560 * take over the slave's change_mtu
3561 * function for the duration of their
3562 * servitude.
3563 */
3564 break;
3565 case NETDEV_CHANGENAME:
3566 /*
3567 * TODO: handle changing the primary's name
3568 */
3569 break;
8531c5ff
AK		 3570 case NETDEV_FEAT_CHANGE:
3571 bond_compute_features(bond);
3572 break;
1da177e4
LT		 3573 default:
3574 break;
3575 }
3576
3577 return NOTIFY_DONE;
3578}
3579
3580/*
3581 * bond_netdev_event: handle netdev notifier chain events.
3582 *
3583 * This function receives events for the netdev chain. The caller (an
e041c683 3584 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3585 * locks for us to safely manipulate the slave devices (RTNL lock,
3586 * dev_probe_lock).
3587 */
3588static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3589{
3590 struct net_device *event_dev = (struct net_device *)ptr;
3591
c346dca1 3592 if (dev_net(event_dev) != &init_net)
e9dc8653
EB		 3593 return NOTIFY_DONE;
3594
5a03cdb7 3595 pr_debug("event_dev: %s, event: %lx\n",
1da177e4
LT		 3596 (event_dev ? event_dev->name : "None"),
3597 event);
3598
0b680e75
JV		 3599 if (!(event_dev->priv_flags & IFF_BONDING))
3600 return NOTIFY_DONE;
3601
1da177e4 3602 if (event_dev->flags & IFF_MASTER) {
5a03cdb7 3603 pr_debug("IFF_MASTER\n");
1da177e4
LT		 3604 return bond_master_netdev_event(event, event_dev);
3605 }
3606
3607 if (event_dev->flags & IFF_SLAVE) {
5a03cdb7 3608 pr_debug("IFF_SLAVE\n");
1da177e4
LT		 3609 return bond_slave_netdev_event(event, event_dev);
3610 }
3611
3612 return NOTIFY_DONE;
3613}
3614
c3ade5ca
JV		 3615/*
3616 * bond_inetaddr_event: handle inetaddr notifier chain events.
3617 *
3618 * We keep track of device IPs primarily to use as source addresses in
3619 * ARP monitor probes (rather than spewing out broadcasts all the time).
3620 *
3621 * We track one IP for the main device (if it has one), plus one per VLAN.
3622 */
3623static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3624{
3625 struct in_ifaddr *ifa = ptr;
3626 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
0883beca
PE		 3627 struct bonding *bond;
3628 struct vlan_entry *vlan;
c3ade5ca 3629
c346dca1 3630 if (dev_net(ifa->ifa_dev->dev) != &init_net)
6133fb1a
DL		 3631 return NOTIFY_DONE;
3632
0883beca 3633 list_for_each_entry(bond, &bond_dev_list, bond_list) {
c3ade5ca
JV		 3634 if (bond->dev == event_dev) {
3635 switch (event) {
3636 case NETDEV_UP:
3637 bond->master_ip = ifa->ifa_local;
3638 return NOTIFY_OK;
3639 case NETDEV_DOWN:
3640 bond->master_ip = bond_glean_dev_ip(bond->dev);
3641 return NOTIFY_OK;
3642 default:
3643 return NOTIFY_DONE;
3644 }
3645 }
3646
0883beca 3647 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 3648 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3649 if (vlan_dev == event_dev) {
3650 switch (event) {
3651 case NETDEV_UP:
3652 vlan->vlan_ip = ifa->ifa_local;
3653 return NOTIFY_OK;
3654 case NETDEV_DOWN:
3655 vlan->vlan_ip =
3656 bond_glean_dev_ip(vlan_dev);
3657 return NOTIFY_OK;
3658 default:
3659 return NOTIFY_DONE;
3660 }
3661 }
3662 }
3663 }
3664 return NOTIFY_DONE;
3665}
3666
1da177e4
LT		 3667static struct notifier_block bond_netdev_notifier = {
3668 .notifier_call = bond_netdev_event,
3669};
3670
c3ade5ca
JV		 3671static struct notifier_block bond_inetaddr_notifier = {
3672 .notifier_call = bond_inetaddr_event,
3673};
3674
1da177e4
LT		 3675/*-------------------------- Packet type handling ---------------------------*/
3676
3677/* register to receive lacpdus on a bond */
3678static void bond_register_lacpdu(struct bonding *bond)
3679{
3680 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3681
3682 /* initialize packet type */
3683 pk_type->type = PKT_TYPE_LACPDU;
3684 pk_type->dev = bond->dev;
3685 pk_type->func = bond_3ad_lacpdu_recv;
3686
3687 dev_add_pack(pk_type);
3688}
3689
3690/* unregister to receive lacpdus on a bond */
3691static void bond_unregister_lacpdu(struct bonding *bond)
3692{
3693 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3694}
3695
f5b2b966
JV		 3696void bond_register_arp(struct bonding *bond)
3697{
3698 struct packet_type *pt = &bond->arp_mon_pt;
3699
c4f283b1
JV		 3700 if (pt->type)
3701 return;
3702
f5b2b966 3703 pt->type = htons(ETH_P_ARP);
e245cb71 3704 pt->dev = bond->dev;
f5b2b966
JV		 3705 pt->func = bond_arp_rcv;
3706 dev_add_pack(pt);
3707}
3708
3709void bond_unregister_arp(struct bonding *bond)
3710{
c4f283b1
JV		 3711 struct packet_type *pt = &bond->arp_mon_pt;
3712
3713 dev_remove_pack(pt);
3714 pt->type = 0;
f5b2b966
JV		 3715}
3716
169a3e66
JV		 3717/*---------------------------- Hashing Policies -----------------------------*/
3718
6f6652be
JV		 3719/*
3720 * Hash for the output device based upon layer 2 and layer 3 data. If
3721 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3722 */
3723static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3724 struct net_device *bond_dev, int count)
3725{
3726 struct ethhdr *data = (struct ethhdr *)skb->data;
3727 struct iphdr *iph = ip_hdr(skb);
3728
f14c4e4e 3729 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be
JV		 3730 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3731 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3732 }
3733
3734 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3735}
3736
169a3e66 3737/*
59c51591 3738 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3739 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3740 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3741 */
3742static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3743 struct net_device *bond_dev, int count)
3744{
3745 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3746 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3747 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3748 int layer4_xor = 0;
3749
f14c4e4e
BH		 3750 if (skb->protocol == htons(ETH_P_IP)) {
3751 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3752 (iph->protocol == IPPROTO_TCP ||
3753 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3754 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3755 }
3756 return (layer4_xor ^
3757 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3758
3759 }
3760
3761 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3762}
3763
3764/*
3765 * Hash for the output device based upon layer 2 data
3766 */
3767static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3768 struct net_device *bond_dev, int count)
3769{
3770 struct ethhdr *data = (struct ethhdr *)skb->data;
3771
3772 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3773}
3774
1da177e4
LT		 3775/*-------------------------- Device entry points ----------------------------*/
3776
3777static int bond_open(struct net_device *bond_dev)
3778{
454d7c9b 3779 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3780
3781 bond->kill_timers = 0;
3782
58402054 3783 if (bond_is_lb(bond)) {
1da177e4
LT		 3784 /* bond_alb_initialize must be called before the timer
3785 * is started.
3786 */
3787 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3788 /* something went wrong - fail the open operation */
3789 return -1;
3790 }
3791
1b76b316
JV		 3792 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3793 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3794 }
3795
3796 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3797 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3798 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3799 }
3800
3801 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3802 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3803 INIT_DELAYED_WORK(&bond->arp_work,
3804 bond_activebackup_arp_mon);
3805 else
3806 INIT_DELAYED_WORK(&bond->arp_work,
3807 bond_loadbalance_arp_mon);
3808
3809 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966
JV		 3810 if (bond->params.arp_validate)
3811 bond_register_arp(bond);
1da177e4
LT		 3812 }
3813
3814 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3815 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3816 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4
LT		 3817 /* register to receive LACPDUs */
3818 bond_register_lacpdu(bond);
fd989c83 3819 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3820 }
3821
3822 return 0;
3823}
3824
3825static int bond_close(struct net_device *bond_dev)
3826{
454d7c9b 3827 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3828
3829 if (bond->params.mode == BOND_MODE_8023AD) {
3830 /* Unregister the receive of LACPDUs */
3831 bond_unregister_lacpdu(bond);
3832 }
3833
f5b2b966
JV		 3834 if (bond->params.arp_validate)
3835 bond_unregister_arp(bond);
3836
1da177e4
LT		 3837 write_lock_bh(&bond->lock);
3838
b59f9f74 3839 bond->send_grat_arp = 0;
305d552a 3840 bond->send_unsol_na = 0;
1da177e4
LT		 3841
3842 /* signal timers not to re-arm */
3843 bond->kill_timers = 1;
3844
3845 write_unlock_bh(&bond->lock);
3846
1da177e4 3847 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3848 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3849 }
3850
3851 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3852 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3853 }
3854
3855 switch (bond->params.mode) {
3856 case BOND_MODE_8023AD:
1b76b316 3857 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3858 break;
3859 case BOND_MODE_TLB:
3860 case BOND_MODE_ALB:
1b76b316 3861 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3862 break;
3863 default:
3864 break;
3865 }
3866
1da177e4 3867
58402054 3868 if (bond_is_lb(bond)) {
1da177e4
LT		 3869 /* Must be called only after all
3870 * slaves have been released
3871 */
3872 bond_alb_deinitialize(bond);
3873 }
3874
3875 return 0;
3876}
3877
3878static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3879{
454d7c9b 3880 struct bonding *bond = netdev_priv(bond_dev);
eeda3fd6 3881 struct net_device_stats *stats = &bond->stats;
2439f9eb 3882 struct net_device_stats local_stats;
1da177e4
LT		 3883 struct slave *slave;
3884 int i;
3885
2439f9eb 3886 memset(&local_stats, 0, sizeof(struct net_device_stats));
1da177e4
LT		 3887
3888 read_lock_bh(&bond->lock);
3889
3890 bond_for_each_slave(bond, slave, i) {
eeda3fd6
SH		 3891 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3892
2439f9eb
AG		 3893 local_stats.rx_packets += sstats->rx_packets;
3894 local_stats.rx_bytes += sstats->rx_bytes;
3895 local_stats.rx_errors += sstats->rx_errors;
3896 local_stats.rx_dropped += sstats->rx_dropped;
3897
3898 local_stats.tx_packets += sstats->tx_packets;
3899 local_stats.tx_bytes += sstats->tx_bytes;
3900 local_stats.tx_errors += sstats->tx_errors;
3901 local_stats.tx_dropped += sstats->tx_dropped;
3902
3903 local_stats.multicast += sstats->multicast;
3904 local_stats.collisions += sstats->collisions;
3905
3906 local_stats.rx_length_errors += sstats->rx_length_errors;
3907 local_stats.rx_over_errors += sstats->rx_over_errors;
3908 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3909 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3910 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3911 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3912
3913 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3914 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3915 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3916 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3917 local_stats.tx_window_errors += sstats->tx_window_errors;
3918 }
3919
3920 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
1da177e4
LT		 3921
3922 read_unlock_bh(&bond->lock);
3923
3924 return stats;
3925}
3926
3927static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3928{
3929 struct net_device *slave_dev = NULL;
3930 struct ifbond k_binfo;
3931 struct ifbond __user *u_binfo = NULL;
3932 struct ifslave k_sinfo;
3933 struct ifslave __user *u_sinfo = NULL;
3934 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3935 int res = 0;
3936
5a03cdb7 3937 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
1da177e4
LT		 3938 bond_dev->name, cmd);
3939
3940 switch (cmd) {
1da177e4
LT		 3941 case SIOCGMIIPHY:
3942 mii = if_mii(ifr);
3943 if (!mii) {
3944 return -EINVAL;
3945 }
3946 mii->phy_id = 0;
3947 /* Fall Through */
3948 case SIOCGMIIREG:
3949 /*
3950 * We do this again just in case we were called by SIOCGMIIREG
3951 * instead of SIOCGMIIPHY.
3952 */
3953 mii = if_mii(ifr);
3954 if (!mii) {
3955 return -EINVAL;
3956 }
3957
3958 if (mii->reg_num == 1) {
454d7c9b 3959 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3960 mii->val_out = 0;
6603a6f2 3961 read_lock(&bond->lock);
1da177e4 3962 read_lock(&bond->curr_slave_lock);
4e140079 3963 if (netif_carrier_ok(bond->dev)) {
1da177e4
LT		 3964 mii->val_out = BMSR_LSTATUS;
3965 }
3966 read_unlock(&bond->curr_slave_lock);
6603a6f2 3967 read_unlock(&bond->lock);
1da177e4
LT		 3968 }
3969
3970 return 0;
3971 case BOND_INFO_QUERY_OLD:
3972 case SIOCBONDINFOQUERY:
3973 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3974
3975 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3976 return -EFAULT;
3977 }
3978
3979 res = bond_info_query(bond_dev, &k_binfo);
3980 if (res == 0) {
3981 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3982 return -EFAULT;
3983 }
3984 }
3985
3986 return res;
3987 case BOND_SLAVE_INFO_QUERY_OLD:
3988 case SIOCBONDSLAVEINFOQUERY:
3989 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3990
3991 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3992 return -EFAULT;
3993 }
3994
3995 res = bond_slave_info_query(bond_dev, &k_sinfo);
3996 if (res == 0) {
3997 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3998 return -EFAULT;
3999 }
4000 }
4001
4002 return res;
4003 default:
4004 /* Go on */
4005 break;
4006 }
4007
4008 if (!capable(CAP_NET_ADMIN)) {
4009 return -EPERM;
4010 }
4011
881d966b 4012 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
1da177e4 4013
5a03cdb7 4014 pr_debug("slave_dev=%p: \n", slave_dev);
1da177e4
LT		 4015
4016 if (!slave_dev) {
4017 res = -ENODEV;
4018 } else {
5a03cdb7 4019 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
1da177e4
LT		 4020 switch (cmd) {
4021 case BOND_ENSLAVE_OLD:
4022 case SIOCBONDENSLAVE:
4023 res = bond_enslave(bond_dev, slave_dev);
4024 break;
4025 case BOND_RELEASE_OLD:
4026 case SIOCBONDRELEASE:
4027 res = bond_release(bond_dev, slave_dev);
4028 break;
4029 case BOND_SETHWADDR_OLD:
4030 case SIOCBONDSETHWADDR:
4031 res = bond_sethwaddr(bond_dev, slave_dev);
4032 break;
4033 case BOND_CHANGE_ACTIVE_OLD:
4034 case SIOCBONDCHANGEACTIVE:
4035 res = bond_ioctl_change_active(bond_dev, slave_dev);
4036 break;
4037 default:
4038 res = -EOPNOTSUPP;
4039 }
4040
4041 dev_put(slave_dev);
4042 }
4043
1da177e4
LT		 4044 return res;
4045}
4046
4047static void bond_set_multicast_list(struct net_device *bond_dev)
4048{
454d7c9b 4049 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4050 struct dev_mc_list *dmi;
4051
1da177e4
LT		 4052 /*
4053 * Do promisc before checking multicast_mode
4054 */
4055 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
7e1a1ac1
WC		 4056 /*
4057 * FIXME: Need to handle the error when one of the multi-slaves
4058 * encounters error.
4059 */
1da177e4
LT		 4060 bond_set_promiscuity(bond, 1);
4061 }
4062
4063 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4064 bond_set_promiscuity(bond, -1);
4065 }
4066
4067 /* set allmulti flag to slaves */
4068 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
7e1a1ac1
WC		 4069 /*
4070 * FIXME: Need to handle the error when one of the multi-slaves
4071 * encounters error.
4072 */
1da177e4
LT		 4073 bond_set_allmulti(bond, 1);
4074 }
4075
4076 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4077 bond_set_allmulti(bond, -1);
4078 }
4079
80ee5ad2
JV		 4080 read_lock(&bond->lock);
4081
1da177e4
LT		 4082 bond->flags = bond_dev->flags;
4083
4084 /* looking for addresses to add to slaves' mc list */
4085 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4086 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4087 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4088 }
4089 }
4090
4091 /* looking for addresses to delete from slaves' list */
4092 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4093 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4094 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4095 }
4096 }
4097
4098 /* save master's multicast list */
4099 bond_mc_list_destroy(bond);
4100 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4101
80ee5ad2 4102 read_unlock(&bond->lock);
1da177e4
LT		 4103}
4104
00829823
SH		 4105static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4106{
4107 struct bonding *bond = netdev_priv(dev);
4108 struct slave *slave = bond->first_slave;
4109
4110 if (slave) {
4111 const struct net_device_ops *slave_ops
4112 = slave->dev->netdev_ops;
4113 if (slave_ops->ndo_neigh_setup)
72e2240f 4114 return slave_ops->ndo_neigh_setup(slave->dev, parms);
00829823
SH		 4115 }
4116 return 0;
4117}
4118
1da177e4
LT		 4119/*
4120 * Change the MTU of all of a master's slaves to match the master
4121 */
4122static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4123{
454d7c9b 4124 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4125 struct slave *slave, *stop_at;
4126 int res = 0;
4127 int i;
4128
5a03cdb7 4129 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
1da177e4
LT		 4130 (bond_dev ? bond_dev->name : "None"), new_mtu);
4131
4132 /* Can't hold bond->lock with bh disabled here since
4133 * some base drivers panic. On the other hand we can't
4134 * hold bond->lock without bh disabled because we'll
4135 * deadlock. The only solution is to rely on the fact
4136 * that we're under rtnl_lock here, and the slaves
4137 * list won't change. This doesn't solve the problem
4138 * of setting the slave's MTU while it is
4139 * transmitting, but the assumption is that the base
4140 * driver can handle that.
4141 *
4142 * TODO: figure out a way to safely iterate the slaves
4143 * list, but without holding a lock around the actual
4144 * call to the base driver.
4145 */
4146
4147 bond_for_each_slave(bond, slave, i) {
5a03cdb7 4148 pr_debug("s %p s->p %p c_m %p\n", slave,
53a3294e 4149 slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
e944ef79 4150
1da177e4
LT		 4151 res = dev_set_mtu(slave->dev, new_mtu);
4152
4153 if (res) {
4154 /* If we failed to set the slave's mtu to the new value
4155 * we must abort the operation even in ACTIVE_BACKUP
4156 * mode, because if we allow the backup slaves to have
4157 * different mtu values than the active slave we'll
4158 * need to change their mtu when doing a failover. That
4159 * means changing their mtu from timer context, which
4160 * is probably not a good idea.
4161 */
5a03cdb7 4162 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 4163 goto unwind;
4164 }
4165 }
4166
4167 bond_dev->mtu = new_mtu;
4168
4169 return 0;
4170
4171unwind:
4172 /* unwind from head to the slave that failed */
4173 stop_at = slave;
4174 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4175 int tmp_res;
4176
4177 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4178 if (tmp_res) {
5a03cdb7 4179 pr_debug("unwind err %d dev %s\n", tmp_res,
1da177e4
LT		 4180 slave->dev->name);
4181 }
4182 }
4183
4184 return res;
4185}
4186
4187/*
4188 * Change HW address
4189 *
4190 * Note that many devices must be down to change the HW address, and
4191 * downing the master releases all slaves. We can make bonds full of
4192 * bonding devices to test this, however.
4193 */
4194static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4195{
454d7c9b 4196 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4197 struct sockaddr *sa = addr, tmp_sa;
4198 struct slave *slave, *stop_at;
4199 int res = 0;
4200 int i;
4201
eb7cc59a
SH		 4202 if (bond->params.mode == BOND_MODE_ALB)
4203 return bond_alb_set_mac_address(bond_dev, addr);
4204
4205
5a03cdb7 4206 pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
1da177e4 4207
dd957c57 4208 /*
3915c1e8
JV		 4209 * If fail_over_mac is set to active, do nothing and return
4210 * success. Returning an error causes ifenslave to fail.
dd957c57 4211 */
3915c1e8 4212 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 4213 return 0;
2ab82852 4214
1da177e4
LT		 4215 if (!is_valid_ether_addr(sa->sa_data)) {
4216 return -EADDRNOTAVAIL;
4217 }
4218
4219 /* Can't hold bond->lock with bh disabled here since
4220 * some base drivers panic. On the other hand we can't
4221 * hold bond->lock without bh disabled because we'll
4222 * deadlock. The only solution is to rely on the fact
4223 * that we're under rtnl_lock here, and the slaves
4224 * list won't change. This doesn't solve the problem
4225 * of setting the slave's hw address while it is
4226 * transmitting, but the assumption is that the base
4227 * driver can handle that.
4228 *
4229 * TODO: figure out a way to safely iterate the slaves
4230 * list, but without holding a lock around the actual
4231 * call to the base driver.
4232 */
4233
4234 bond_for_each_slave(bond, slave, i) {
eb7cc59a 4235 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
5a03cdb7 4236 pr_debug("slave %p %s\n", slave, slave->dev->name);
1da177e4 4237
eb7cc59a 4238 if (slave_ops->ndo_set_mac_address == NULL) {
1da177e4 4239 res = -EOPNOTSUPP;
5a03cdb7 4240 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
1da177e4
LT		 4241 goto unwind;
4242 }
4243
4244 res = dev_set_mac_address(slave->dev, addr);
4245 if (res) {
4246 /* TODO: consider downing the slave
4247 * and retry ?
4248 * User should expect communications
4249 * breakage anyway until ARP finish
4250 * updating, so...
4251 */
5a03cdb7 4252 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 4253 goto unwind;
4254 }
4255 }
4256
4257 /* success */
4258 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4259 return 0;
4260
4261unwind:
4262 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4263 tmp_sa.sa_family = bond_dev->type;
4264
4265 /* unwind from head to the slave that failed */
4266 stop_at = slave;
4267 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4268 int tmp_res;
4269
4270 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4271 if (tmp_res) {
5a03cdb7 4272 pr_debug("unwind err %d dev %s\n", tmp_res,
1da177e4
LT		 4273 slave->dev->name);
4274 }
4275 }
4276
4277 return res;
4278}
4279
4280static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4281{
454d7c9b 4282 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4283 struct slave *slave, *start_at;
cf5f9044 4284 int i, slave_no, res = 1;
1da177e4
LT		 4285
4286 read_lock(&bond->lock);
4287
4288 if (!BOND_IS_OK(bond)) {
4289 goto out;
4290 }
4291
cf5f9044
JV		 4292 /*
4293 * Concurrent TX may collide on rr_tx_counter; we accept that
4294 * as being rare enough not to justify using an atomic op here
4295 */
4296 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
1da177e4 4297
cf5f9044
JV		 4298 bond_for_each_slave(bond, slave, i) {
4299 slave_no--;
4300 if (slave_no < 0) {
4301 break;
4302 }
1da177e4
LT		 4303 }
4304
cf5f9044 4305 start_at = slave;
1da177e4
LT		 4306 bond_for_each_slave_from(bond, slave, i, start_at) {
4307 if (IS_UP(slave->dev) &&
4308 (slave->link == BOND_LINK_UP) &&
4309 (slave->state == BOND_STATE_ACTIVE)) {
4310 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4311 break;
4312 }
4313 }
4314
1da177e4
LT		 4315out:
4316 if (res) {
4317 /* no suitable interface, frame not sent */
4318 dev_kfree_skb(skb);
4319 }
4320 read_unlock(&bond->lock);
4321 return 0;
4322}
4323
075897ce 4324
1da177e4
LT		 4325/*
4326 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4327 * the bond has a usable interface.
4328 */
4329static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4330{
454d7c9b 4331 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4332 int res = 1;
4333
1da177e4
LT		 4334 read_lock(&bond->lock);
4335 read_lock(&bond->curr_slave_lock);
4336
4337 if (!BOND_IS_OK(bond)) {
4338 goto out;
4339 }
4340
075897ce
JL		 4341 if (!bond->curr_active_slave)
4342 goto out;
4343
075897ce
JL		 4344 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4345
1da177e4
LT		 4346out:
4347 if (res) {
4348 /* no suitable interface, frame not sent */
4349 dev_kfree_skb(skb);
4350 }
4351 read_unlock(&bond->curr_slave_lock);
4352 read_unlock(&bond->lock);
4353 return 0;
4354}
4355
4356/*
169a3e66
JV		 4357 * In bond_xmit_xor() , we determine the output device by using a pre-
4358 * determined xmit_hash_policy(), If the selected device is not enabled,
4359 * find the next active slave.
1da177e4
LT		 4360 */
4361static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4362{
454d7c9b 4363 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4364 struct slave *slave, *start_at;
4365 int slave_no;
4366 int i;
4367 int res = 1;
4368
4369 read_lock(&bond->lock);
4370
4371 if (!BOND_IS_OK(bond)) {
4372 goto out;
4373 }
4374
169a3e66 4375 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
1da177e4
LT		 4376
4377 bond_for_each_slave(bond, slave, i) {
4378 slave_no--;
4379 if (slave_no < 0) {
4380 break;
4381 }
4382 }
4383
4384 start_at = slave;
4385
4386 bond_for_each_slave_from(bond, slave, i, start_at) {
4387 if (IS_UP(slave->dev) &&
4388 (slave->link == BOND_LINK_UP) &&
4389 (slave->state == BOND_STATE_ACTIVE)) {
4390 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4391 break;
4392 }
4393 }
4394
4395out:
4396 if (res) {
4397 /* no suitable interface, frame not sent */
4398 dev_kfree_skb(skb);
4399 }
4400 read_unlock(&bond->lock);
4401 return 0;
4402}
4403
4404/*
4405 * in broadcast mode, we send everything to all usable interfaces.
4406 */
4407static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4408{
454d7c9b 4409 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4410 struct slave *slave, *start_at;
4411 struct net_device *tx_dev = NULL;
4412 int i;
4413 int res = 1;
4414
4415 read_lock(&bond->lock);
4416
4417 if (!BOND_IS_OK(bond)) {
4418 goto out;
4419 }
4420
4421 read_lock(&bond->curr_slave_lock);
4422 start_at = bond->curr_active_slave;
4423 read_unlock(&bond->curr_slave_lock);
4424
4425 if (!start_at) {
4426 goto out;
4427 }
4428
4429 bond_for_each_slave_from(bond, slave, i, start_at) {
4430 if (IS_UP(slave->dev) &&
4431 (slave->link == BOND_LINK_UP) &&
4432 (slave->state == BOND_STATE_ACTIVE)) {
4433 if (tx_dev) {
4434 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4435 if (!skb2) {
4436 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4437 ": %s: Error: bond_xmit_broadcast(): "
4438 "skb_clone() failed\n",
4439 bond_dev->name);
1da177e4
LT		 4440 continue;
4441 }
4442
4443 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4444 if (res) {
4445 dev_kfree_skb(skb2);
4446 continue;
4447 }
4448 }
4449 tx_dev = slave->dev;
4450 }
4451 }
4452
4453 if (tx_dev) {
4454 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4455 }
4456
4457out:
4458 if (res) {
4459 /* no suitable interface, frame not sent */
4460 dev_kfree_skb(skb);
4461 }
4462 /* frame sent to all suitable interfaces */
4463 read_unlock(&bond->lock);
4464 return 0;
4465}
4466
4467/*------------------------- Device initialization ---------------------------*/
4468
6f6652be
JV		 4469static void bond_set_xmit_hash_policy(struct bonding *bond)
4470{
4471 switch (bond->params.xmit_policy) {
4472 case BOND_XMIT_POLICY_LAYER23:
4473 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4474 break;
4475 case BOND_XMIT_POLICY_LAYER34:
4476 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4477 break;
4478 case BOND_XMIT_POLICY_LAYER2:
4479 default:
4480 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4481 break;
4482 }
4483}
4484
00829823
SH		 4485static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4486{
4487 const struct bonding *bond = netdev_priv(dev);
4488
4489 switch (bond->params.mode) {
4490 case BOND_MODE_ROUNDROBIN:
4491 return bond_xmit_roundrobin(skb, dev);
4492 case BOND_MODE_ACTIVEBACKUP:
4493 return bond_xmit_activebackup(skb, dev);
4494 case BOND_MODE_XOR:
4495 return bond_xmit_xor(skb, dev);
4496 case BOND_MODE_BROADCAST:
4497 return bond_xmit_broadcast(skb, dev);
4498 case BOND_MODE_8023AD:
4499 return bond_3ad_xmit_xor(skb, dev);
4500 case BOND_MODE_ALB:
4501 case BOND_MODE_TLB:
4502 return bond_alb_xmit(skb, dev);
4503 default:
4504 /* Should never happen, mode already checked */
4505 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4506 dev->name, bond->params.mode);
4507 WARN_ON_ONCE(1);
4508 dev_kfree_skb(skb);
4509 return NETDEV_TX_OK;
4510 }
4511}
4512
4513
1da177e4
LT		 4514/*
4515 * set bond mode specific net device operations
4516 */
a77b5325 4517void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4518{
169a3e66
JV		 4519 struct net_device *bond_dev = bond->dev;
4520
1da177e4
LT		 4521 switch (mode) {
4522 case BOND_MODE_ROUNDROBIN:
1da177e4
LT		 4523 break;
4524 case BOND_MODE_ACTIVEBACKUP:
1da177e4
LT		 4525 break;
4526 case BOND_MODE_XOR:
6f6652be 4527 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4528 break;
4529 case BOND_MODE_BROADCAST:
1da177e4
LT		 4530 break;
4531 case BOND_MODE_8023AD:
8f903c70 4532 bond_set_master_3ad_flags(bond);
6f6652be 4533 bond_set_xmit_hash_policy(bond);
1da177e4 4534 break;
1da177e4 4535 case BOND_MODE_ALB:
8f903c70
JV		 4536 bond_set_master_alb_flags(bond);
4537 /* FALLTHRU */
4538 case BOND_MODE_TLB:
1da177e4
LT		 4539 break;
4540 default:
4541 /* Should never happen, mode already checked */
4542 printk(KERN_ERR DRV_NAME
4e0952c7
MW		 4543 ": %s: Error: Unknown bonding mode %d\n",
4544 bond_dev->name,
1da177e4
LT		 4545 mode);
4546 break;
4547 }
4548}
4549
217df670
JV		 4550static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4551 struct ethtool_drvinfo *drvinfo)
4552{
4553 strncpy(drvinfo->driver, DRV_NAME, 32);
4554 strncpy(drvinfo->version, DRV_VERSION, 32);
4555 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4556}
4557
7282d491 4558static const struct ethtool_ops bond_ethtool_ops = {
217df670 4559 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac
SH		 4560 .get_link = ethtool_op_get_link,
4561 .get_tx_csum = ethtool_op_get_tx_csum,
4562 .get_sg = ethtool_op_get_sg,
4563 .get_tso = ethtool_op_get_tso,
4564 .get_ufo = ethtool_op_get_ufo,
4565 .get_flags = ethtool_op_get_flags,
8531c5ff
AK		 4566};
4567
eb7cc59a 4568static const struct net_device_ops bond_netdev_ops = {
9e71626c 4569 .ndo_uninit = bond_uninit,
eb7cc59a
SH		 4570 .ndo_open = bond_open,
4571 .ndo_stop = bond_close,
00829823 4572 .ndo_start_xmit = bond_start_xmit,
eb7cc59a
SH		 4573 .ndo_get_stats = bond_get_stats,
4574 .ndo_do_ioctl = bond_do_ioctl,
4575 .ndo_set_multicast_list = bond_set_multicast_list,
4576 .ndo_change_mtu = bond_change_mtu,
eb7cc59a 4577 .ndo_set_mac_address = bond_set_mac_address,
00829823 4578 .ndo_neigh_setup = bond_neigh_setup,
eb7cc59a
SH		 4579 .ndo_vlan_rx_register = bond_vlan_rx_register,
4580 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4581 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4582};
4583
1da177e4
LT		 4584/*
4585 * Does not allocate but creates a /proc entry.
4586 * Allowed to fail.
4587 */
d2991f75 4588static int bond_init(struct net_device *bond_dev)
1da177e4 4589{
454d7c9b 4590 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4591
5a03cdb7 4592 pr_debug("Begin bond_init for %s\n", bond_dev->name);
1da177e4
LT		 4593
4594 /* initialize rwlocks */
4595 rwlock_init(&bond->lock);
4596 rwlock_init(&bond->curr_slave_lock);
4597
d2991f75 4598 bond->params = bonding_defaults;
1da177e4 4599
1b76b316
JV		 4600 bond->wq = create_singlethread_workqueue(bond_dev->name);
4601 if (!bond->wq)
4602 return -ENOMEM;
4603
1da177e4
LT		 4604 /* Initialize pointers */
4605 bond->first_slave = NULL;
4606 bond->curr_active_slave = NULL;
4607 bond->current_arp_slave = NULL;
4608 bond->primary_slave = NULL;
4609 bond->dev = bond_dev;
1053f62c 4610 bond->send_grat_arp = 0;
305d552a 4611 bond->send_unsol_na = 0;
d90a162a 4612 bond->setup_by_slave = 0;
1da177e4
LT		 4613 INIT_LIST_HEAD(&bond->vlan_list);
4614
4615 /* Initialize the device entry points */
eb7cc59a 4616 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 4617 bond_dev->ethtool_ops = &bond_ethtool_ops;
169a3e66 4618 bond_set_mode_ops(bond, bond->params.mode);
1da177e4 4619
9e71626c 4620 bond_dev->destructor = free_netdev;
1da177e4
LT		 4621
4622 /* Initialize the device options */
4623 bond_dev->tx_queue_len = 0;
4624 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4625 bond_dev->priv_flags |= IFF_BONDING;
6cf3f41e
JV		 4626 if (bond->params.arp_interval)
4627 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
1da177e4
LT		 4628
4629 /* At first, we block adding VLANs. That's the only way to
4630 * prevent problems that occur when adding VLANs over an
4631 * empty bond. The block will be removed once non-challenged
4632 * slaves are enslaved.
4633 */
4634 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4635
932ff279 4636 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4637 * transmitting */
4638 bond_dev->features |= NETIF_F_LLTX;
4639
4640 /* By default, we declare the bond to be fully
4641 * VLAN hardware accelerated capable. Special
4642 * care is taken in the various xmit functions
4643 * when there are slaves that are not hw accel
4644 * capable
4645 */
1da177e4
LT		 4646 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4647 NETIF_F_HW_VLAN_RX |
4648 NETIF_F_HW_VLAN_FILTER);
4649
1da177e4 4650 bond_create_proc_entry(bond);
1da177e4
LT		 4651 list_add_tail(&bond->bond_list, &bond_dev_list);
4652
4653 return 0;
4654}
4655
fdaea7a9
JV		 4656static void bond_work_cancel_all(struct bonding *bond)
4657{
4658 write_lock_bh(&bond->lock);
4659 bond->kill_timers = 1;
4660 write_unlock_bh(&bond->lock);
4661
4662 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4663 cancel_delayed_work(&bond->mii_work);
4664
4665 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4666 cancel_delayed_work(&bond->arp_work);
4667
4668 if (bond->params.mode == BOND_MODE_ALB &&
4669 delayed_work_pending(&bond->alb_work))
4670 cancel_delayed_work(&bond->alb_work);
4671
4672 if (bond->params.mode == BOND_MODE_8023AD &&
4673 delayed_work_pending(&bond->ad_work))
4674 cancel_delayed_work(&bond->ad_work);
4675}
4676
a434e43f
JV		 4677/* De-initialize device specific data.
4678 * Caller must hold rtnl_lock.
4679 */
4680static void bond_deinit(struct net_device *bond_dev)
4681{
454d7c9b 4682 struct bonding *bond = netdev_priv(bond_dev);
a434e43f
JV		 4683
4684 list_del(&bond->bond_list);
4685
4686 bond_work_cancel_all(bond);
4687
a434e43f 4688 bond_remove_proc_entry(bond);
a434e43f
JV		 4689}
4690
1da177e4
LT		 4691/* Unregister and free all bond devices.
4692 * Caller must hold rtnl_lock.
4693 */
4694static void bond_free_all(void)
4695{
4696 struct bonding *bond, *nxt;
4697
4698 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4699 struct net_device *bond_dev = bond->dev;
4700
fdaea7a9 4701 bond_work_cancel_all(bond);
70298705 4702 /* Release the bonded slaves */
4703 bond_release_all(bond_dev);
9e71626c 4704 unregister_netdevice(bond_dev);
1da177e4
LT		 4705 }
4706
1da177e4 4707 bond_destroy_proc_dir();
1da177e4
LT		 4708}
4709
4710/*------------------------- Module initialization ---------------------------*/
4711
4712/*
4713 * Convert string input module parms. Accept either the
ece95f7f
JV		 4714 * number of the mode or its string name. A bit complicated because
4715 * some mode names are substrings of other names, and calls from sysfs
4716 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4717 */
325dcf7a 4718int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
1da177e4 4719{
54b87323 4720 int modeint = -1, i, rv;
a42e534f 4721 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4722
a42e534f
JV		 4723 for (p = (char *)buf; *p; p++)
4724 if (!(isdigit(*p) || isspace(*p)))
4725 break;
4726
4727 if (*p)
ece95f7f 4728 rv = sscanf(buf, "%20s", modestr);
a42e534f 4729 else
54b87323 4730 rv = sscanf(buf, "%d", &modeint);
a42e534f
JV		 4731
4732 if (!rv)
4733 return -1;
1da177e4
LT		 4734
4735 for (i = 0; tbl[i].modename; i++) {
54b87323 4736 if (modeint == tbl[i].mode)
ece95f7f
JV		 4737 return tbl[i].mode;
4738 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4739 return tbl[i].mode;
1da177e4
LT		 4740 }
4741
4742 return -1;
4743}
4744
4745static int bond_check_params(struct bond_params *params)
4746{
3915c1e8 4747 int arp_validate_value, fail_over_mac_value;
f5b2b966 4748
1da177e4
LT		 4749 /*
4750 * Convert string parameters.
4751 */
4752 if (mode) {
4753 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4754 if (bond_mode == -1) {
4755 printk(KERN_ERR DRV_NAME
4756 ": Error: Invalid bonding mode \"%s\"\n",
4757 mode == NULL ? "NULL" : mode);
4758 return -EINVAL;
4759 }
4760 }
4761
169a3e66
JV		 4762 if (xmit_hash_policy) {
4763 if ((bond_mode != BOND_MODE_XOR) &&
4764 (bond_mode != BOND_MODE_8023AD)) {
4765 printk(KERN_INFO DRV_NAME
4766 ": xor_mode param is irrelevant in mode %s\n",
4767 bond_mode_name(bond_mode));
4768 } else {
4769 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4770 xmit_hashtype_tbl);
4771 if (xmit_hashtype == -1) {
4772 printk(KERN_ERR DRV_NAME
4773 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4774 xmit_hash_policy == NULL ? "NULL" :
4775 xmit_hash_policy);
4776 return -EINVAL;
4777 }
4778 }
4779 }
4780
1da177e4
LT		 4781 if (lacp_rate) {
4782 if (bond_mode != BOND_MODE_8023AD) {
4783 printk(KERN_INFO DRV_NAME
4784 ": lacp_rate param is irrelevant in mode %s\n",
4785 bond_mode_name(bond_mode));
4786 } else {
4787 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4788 if (lacp_fast == -1) {
4789 printk(KERN_ERR DRV_NAME
4790 ": Error: Invalid lacp rate \"%s\"\n",
4791 lacp_rate == NULL ? "NULL" : lacp_rate);
4792 return -EINVAL;
4793 }
4794 }
4795 }
4796
fd989c83
JV		 4797 if (ad_select) {
4798 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4799 if (params->ad_select == -1) {
4800 printk(KERN_ERR DRV_NAME
4801 ": Error: Invalid ad_select \"%s\"\n",
4802 ad_select == NULL ? "NULL" : ad_select);
4803 return -EINVAL;
4804 }
4805
4806 if (bond_mode != BOND_MODE_8023AD) {
4807 printk(KERN_WARNING DRV_NAME
4808 ": ad_select param only affects 802.3ad mode\n");
4809 }
4810 } else {
4811 params->ad_select = BOND_AD_STABLE;
4812 }
4813
b8a9787e 4814 if (max_bonds < 0 || max_bonds > INT_MAX) {
1da177e4
LT		 4815 printk(KERN_WARNING DRV_NAME
4816 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4e0952c7 4817 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
b8a9787e 4818 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4819 max_bonds = BOND_DEFAULT_MAX_BONDS;
4820 }
4821
4822 if (miimon < 0) {
4823 printk(KERN_WARNING DRV_NAME
4824 ": Warning: miimon module parameter (%d), "
4825 "not in range 0-%d, so it was reset to %d\n",
4826 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4827 miimon = BOND_LINK_MON_INTERV;
4828 }
4829
4830 if (updelay < 0) {
4831 printk(KERN_WARNING DRV_NAME
4832 ": Warning: updelay module parameter (%d), "
4833 "not in range 0-%d, so it was reset to 0\n",
4834 updelay, INT_MAX);
4835 updelay = 0;
4836 }
4837
4838 if (downdelay < 0) {
4839 printk(KERN_WARNING DRV_NAME
4840 ": Warning: downdelay module parameter (%d), "
4841 "not in range 0-%d, so it was reset to 0\n",
4842 downdelay, INT_MAX);
4843 downdelay = 0;
4844 }
4845
4846 if ((use_carrier != 0) && (use_carrier != 1)) {
4847 printk(KERN_WARNING DRV_NAME
4848 ": Warning: use_carrier module parameter (%d), "
4849 "not of valid value (0/1), so it was set to 1\n",
4850 use_carrier);
4851 use_carrier = 1;
4852 }
4853
7893b249
MS		 4854 if (num_grat_arp < 0 || num_grat_arp > 255) {
4855 printk(KERN_WARNING DRV_NAME
4856 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4857 "was reset to 1 \n", num_grat_arp);
4858 num_grat_arp = 1;
4859 }
4860
305d552a
BH		 4861 if (num_unsol_na < 0 || num_unsol_na > 255) {
4862 printk(KERN_WARNING DRV_NAME
4863 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4864 "was reset to 1 \n", num_unsol_na);
4865 num_unsol_na = 1;
4866 }
4867
1da177e4
LT		 4868 /* reset values for 802.3ad */
4869 if (bond_mode == BOND_MODE_8023AD) {
4870 if (!miimon) {
4871 printk(KERN_WARNING DRV_NAME
4872 ": Warning: miimon must be specified, "
4873 "otherwise bonding will not detect link "
4874 "failure, speed and duplex which are "
4875 "essential for 802.3ad operation\n");
4876 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4877 miimon = 100;
4878 }
4879 }
4880
4881 /* reset values for TLB/ALB */
4882 if ((bond_mode == BOND_MODE_TLB) ||
4883 (bond_mode == BOND_MODE_ALB)) {
4884 if (!miimon) {
4885 printk(KERN_WARNING DRV_NAME
4886 ": Warning: miimon must be specified, "
4887 "otherwise bonding will not detect link "
4888 "failure and link speed which are essential "
4889 "for TLB/ALB load balancing\n");
4890 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4891 miimon = 100;
4892 }
4893 }
4894
4895 if (bond_mode == BOND_MODE_ALB) {
4896 printk(KERN_NOTICE DRV_NAME
4897 ": In ALB mode you might experience client "
4898 "disconnections upon reconnection of a link if the "
4899 "bonding module updelay parameter (%d msec) is "
4900 "incompatible with the forwarding delay time of the "
4901 "switch\n",
4902 updelay);
4903 }
4904
4905 if (!miimon) {
4906 if (updelay || downdelay) {
4907 /* just warn the user the up/down delay will have
4908 * no effect since miimon is zero...
4909 */
4910 printk(KERN_WARNING DRV_NAME
4911 ": Warning: miimon module parameter not set "
4912 "and updelay (%d) or downdelay (%d) module "
4913 "parameter is set; updelay and downdelay have "
4914 "no effect unless miimon is set\n",
4915 updelay, downdelay);
4916 }
4917 } else {
4918 /* don't allow arp monitoring */
4919 if (arp_interval) {
4920 printk(KERN_WARNING DRV_NAME
4921 ": Warning: miimon (%d) and arp_interval (%d) "
4922 "can't be used simultaneously, disabling ARP "
4923 "monitoring\n",
4924 miimon, arp_interval);
4925 arp_interval = 0;
4926 }
4927
4928 if ((updelay % miimon) != 0) {
4929 printk(KERN_WARNING DRV_NAME
4930 ": Warning: updelay (%d) is not a multiple "
4931 "of miimon (%d), updelay rounded to %d ms\n",
4932 updelay, miimon, (updelay / miimon) * miimon);
4933 }
4934
4935 updelay /= miimon;
4936
4937 if ((downdelay % miimon) != 0) {
4938 printk(KERN_WARNING DRV_NAME
4939 ": Warning: downdelay (%d) is not a multiple "
4940 "of miimon (%d), downdelay rounded to %d ms\n",
4941 downdelay, miimon,
4942 (downdelay / miimon) * miimon);
4943 }
4944
4945 downdelay /= miimon;
4946 }
4947
4948 if (arp_interval < 0) {
4949 printk(KERN_WARNING DRV_NAME
4950 ": Warning: arp_interval module parameter (%d) "
4951 ", not in range 0-%d, so it was reset to %d\n",
4952 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4953 arp_interval = BOND_LINK_ARP_INTERV;
4954 }
4955
4956 for (arp_ip_count = 0;
4957 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4958 arp_ip_count++) {
4959 /* not complete check, but should be good enough to
4960 catch mistakes */
4961 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4962 printk(KERN_WARNING DRV_NAME
4963 ": Warning: bad arp_ip_target module parameter "
4964 "(%s), ARP monitoring will not be performed\n",
4965 arp_ip_target[arp_ip_count]);
4966 arp_interval = 0;
4967 } else {
d3bb52b0 4968 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4969 arp_target[arp_ip_count] = ip;
4970 }
4971 }
4972
4973 if (arp_interval && !arp_ip_count) {
4974 /* don't allow arping if no arp_ip_target given... */
4975 printk(KERN_WARNING DRV_NAME
4976 ": Warning: arp_interval module parameter (%d) "
4977 "specified without providing an arp_ip_target "
4978 "parameter, arp_interval was reset to 0\n",
4979 arp_interval);
4980 arp_interval = 0;
4981 }
4982
f5b2b966
JV		 4983 if (arp_validate) {
4984 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4985 printk(KERN_ERR DRV_NAME
4986 ": arp_validate only supported in active-backup mode\n");
4987 return -EINVAL;
4988 }
4989 if (!arp_interval) {
4990 printk(KERN_ERR DRV_NAME
4991 ": arp_validate requires arp_interval\n");
4992 return -EINVAL;
4993 }
4994
4995 arp_validate_value = bond_parse_parm(arp_validate,
4996 arp_validate_tbl);
4997 if (arp_validate_value == -1) {
4998 printk(KERN_ERR DRV_NAME
4999 ": Error: invalid arp_validate \"%s\"\n",
5000 arp_validate == NULL ? "NULL" : arp_validate);
5001 return -EINVAL;
5002 }
5003 } else
5004 arp_validate_value = 0;
5005
1da177e4
LT		 5006 if (miimon) {
5007 printk(KERN_INFO DRV_NAME
5008 ": MII link monitoring set to %d ms\n",
5009 miimon);
5010 } else if (arp_interval) {
5011 int i;
5012
5013 printk(KERN_INFO DRV_NAME
f5b2b966
JV		 5014 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5015 arp_interval,
5016 arp_validate_tbl[arp_validate_value].modename,
5017 arp_ip_count);
1da177e4
LT		 5018
5019 for (i = 0; i < arp_ip_count; i++)
5020 printk (" %s", arp_ip_target[i]);
5021
5022 printk("\n");
5023
b8a9787e 5024 } else if (max_bonds) {
1da177e4
LT		 5025 /* miimon and arp_interval not set, we need one so things
5026 * work as expected, see bonding.txt for details
5027 */
5028 printk(KERN_WARNING DRV_NAME
5029 ": Warning: either miimon or arp_interval and "
5030 "arp_ip_target module parameters must be specified, "
5031 "otherwise bonding will not detect link failures! see "
5032 "bonding.txt for details.\n");
5033 }
5034
5035 if (primary && !USES_PRIMARY(bond_mode)) {
5036 /* currently, using a primary only makes sense
5037 * in active backup, TLB or ALB modes
5038 */
5039 printk(KERN_WARNING DRV_NAME
5040 ": Warning: %s primary device specified but has no "
5041 "effect in %s mode\n",
5042 primary, bond_mode_name(bond_mode));
5043 primary = NULL;
5044 }
5045
3915c1e8
JV		 5046 if (fail_over_mac) {
5047 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5048 fail_over_mac_tbl);
5049 if (fail_over_mac_value == -1) {
5050 printk(KERN_ERR DRV_NAME
5051 ": Error: invalid fail_over_mac \"%s\"\n",
5052 arp_validate == NULL ? "NULL" : arp_validate);
5053 return -EINVAL;
5054 }
5055
5056 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5057 printk(KERN_WARNING DRV_NAME
5058 ": Warning: fail_over_mac only affects "
5059 "active-backup mode.\n");
5060 } else {
5061 fail_over_mac_value = BOND_FOM_NONE;
5062 }
dd957c57 5063
1da177e4
LT		 5064 /* fill params struct with the proper values */
5065 params->mode = bond_mode;
169a3e66 5066 params->xmit_policy = xmit_hashtype;
1da177e4 5067 params->miimon = miimon;
7893b249 5068 params->num_grat_arp = num_grat_arp;
305d552a 5069 params->num_unsol_na = num_unsol_na;
1da177e4 5070 params->arp_interval = arp_interval;
f5b2b966 5071 params->arp_validate = arp_validate_value;
1da177e4
LT		 5072 params->updelay = updelay;
5073 params->downdelay = downdelay;
5074 params->use_carrier = use_carrier;
5075 params->lacp_fast = lacp_fast;
5076 params->primary[0] = 0;
3915c1e8 5077 params->fail_over_mac = fail_over_mac_value;
1da177e4
LT		 5078
5079 if (primary) {
5080 strncpy(params->primary, primary, IFNAMSIZ);
5081 params->primary[IFNAMSIZ - 1] = 0;
5082 }
5083
5084 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5085
5086 return 0;
5087}
5088
0daa2303 5089static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 5090static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 5091
e8a0464c
DM		 5092static void bond_set_lockdep_class_one(struct net_device *dev,
5093 struct netdev_queue *txq,
5094 void *_unused)
c773e847
DM		 5095{
5096 lockdep_set_class(&txq->_xmit_lock,
5097 &bonding_netdev_xmit_lock_key);
5098}
5099
5100static void bond_set_lockdep_class(struct net_device *dev)
5101{
cf508b12
DM		 5102 lockdep_set_class(&dev->addr_list_lock,
5103 &bonding_netdev_addr_lock_key);
e8a0464c 5104 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 5105}
5106
dfe60397 5107/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 5108 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 5109 * Caller must NOT hold rtnl_lock; we need to release it here before we
5110 * set up our sysfs entries.
5111 */
d2991f75 5112int bond_create(const char *name)
dfe60397
MW		 5113{
5114 struct net_device *bond_dev;
0883beca 5115 struct bonding *bond;
dfe60397
MW		 5116 int res;
5117
5118 rtnl_lock();
027ea041 5119 /* Check to see if the bond already exists. */
4fe4763c 5120 if (name) {
0883beca 5121 list_for_each_entry(bond, &bond_dev_list, bond_list)
4fe4763c
JV		 5122 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5123 printk(KERN_ERR DRV_NAME
027ea041 5124 ": cannot add bond %s; it already exists\n",
4fe4763c
JV		 5125 name);
5126 res = -EPERM;
5127 goto out_rtnl;
5128 }
5129 }
027ea041 5130
e4b91c48
JV		 5131 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5132 ether_setup);
dfe60397
MW		 5133 if (!bond_dev) {
5134 printk(KERN_ERR DRV_NAME
5135 ": %s: eek! can't alloc netdev!\n",
5136 name);
5137 res = -ENOMEM;
5138 goto out_rtnl;
5139 }
5140
93f154b5 5141 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
e4b91c48
JV		 5142 if (!name) {
5143 res = dev_alloc_name(bond_dev, "bond%d");
5144 if (res < 0)
5145 goto out_netdev;
5146 }
5147
dfe60397
MW		 5148 /* bond_init() must be called after dev_alloc_name() (for the
5149 * /proc files), but before register_netdevice(), because we
5150 * need to set function pointers.
5151 */
5152
d2991f75 5153 res = bond_init(bond_dev);
dfe60397
MW		 5154 if (res < 0) {
5155 goto out_netdev;
5156 }
5157
d9321605
SH		 5158 bond_set_lockdep_class(bond_dev);
5159
5160 netif_carrier_off(bond_dev);
5161
dfe60397
MW		 5162 res = register_netdevice(bond_dev);
5163 if (res < 0) {
5164 goto out_bond;
5165 }
0daa2303 5166
454d7c9b 5167 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
1363d9b1
JP		 5168 if (res < 0)
5169 goto out_unreg;
09c89279 5170
7e083840 5171 rtnl_unlock();
09c89279
JV		 5172 return 0;
5173
1363d9b1 5174out_unreg:
1363d9b1 5175 unregister_netdevice(bond_dev);
dfe60397
MW		 5176out_bond:
5177 bond_deinit(bond_dev);
5178out_netdev:
5179 free_netdev(bond_dev);
5180out_rtnl:
5181 rtnl_unlock();
dfe60397
MW		 5182 return res;
5183}
5184
1da177e4
LT		 5185static int __init bonding_init(void)
5186{
1da177e4
LT		 5187 int i;
5188 int res;
5189
5190 printk(KERN_INFO "%s", version);
5191
dfe60397 5192 res = bond_check_params(&bonding_defaults);
1da177e4 5193 if (res) {
dfe60397 5194 goto out;
1da177e4
LT		 5195 }
5196
1da177e4 5197 bond_create_proc_dir();
027ea041 5198
1da177e4 5199 for (i = 0; i < max_bonds; i++) {
d2991f75 5200 res = bond_create(NULL);
dfe60397
MW		 5201 if (res)
5202 goto err;
1da177e4
LT		 5203 }
5204
b76cdba9
MW		 5205 res = bond_create_sysfs();
5206 if (res)
5207 goto err;
5208
1da177e4 5209 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5210 register_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5211 bond_register_ipv6_notifier();
1da177e4 5212
dfe60397
MW		 5213 goto out;
5214err:
40abc270 5215 rtnl_lock();
1da177e4 5216 bond_free_all();
1da177e4 5217 rtnl_unlock();
dfe60397 5218out:
1da177e4 5219 return res;
dfe60397 5220
1da177e4
LT		 5221}
5222
5223static void __exit bonding_exit(void)
5224{
5225 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 5226 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
305d552a 5227 bond_unregister_ipv6_notifier();
1da177e4 5228
ae68c398
PE		 5229 bond_destroy_sysfs();
5230
1da177e4
LT		 5231 rtnl_lock();
5232 bond_free_all();
5233 rtnl_unlock();
5234}
5235
5236module_init(bonding_init);
5237module_exit(bonding_exit);
5238MODULE_LICENSE("GPL");
5239MODULE_VERSION(DRV_VERSION);
5240MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5241MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5242MODULE_SUPPORTED_DEVICE("most ethernet devices");
5243
5244/*
5245 * Local variables:
5246 * c-indent-level: 8
5247 * c-basic-offset: 8
5248 * tab-width: 8
5249 * End:
5250 */
5251
Clone of davem's net-next-2.6 tree