bonding: fix double dev_add_pack
[firewire-audio.git] / drivers / net / bonding / bond_main.c
blob68afcb5d7257b4fd90ec9cde9c8747e8cff52a8e
1 /*
2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
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 ...
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.
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
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.
34 //#define BONDING_DEBUG 1
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <asm/system.h>
57 #include <asm/io.h>
58 #include <asm/dma.h>
59 #include <asm/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
65 #include <net/sock.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/proc_fs.h>
68 #include <linux/seq_file.h>
69 #include <linux/smp.h>
70 #include <linux/if_ether.h>
71 #include <net/arp.h>
72 #include <linux/mii.h>
73 #include <linux/ethtool.h>
74 #include <linux/if_vlan.h>
75 #include <linux/if_bonding.h>
76 #include <net/route.h>
77 #include "bonding.h"
78 #include "bond_3ad.h"
79 #include "bond_alb.h"
81 /*---------------------------- Module parameters ----------------------------*/
83 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
84 #define BOND_LINK_MON_INTERV 0
85 #define BOND_LINK_ARP_INTERV 0
87 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
88 static int miimon = BOND_LINK_MON_INTERV;
89 static int updelay = 0;
90 static int downdelay = 0;
91 static int use_carrier = 1;
92 static char *mode = NULL;
93 static char *primary = NULL;
94 static char *lacp_rate = NULL;
95 static char *xmit_hash_policy = NULL;
96 static int arp_interval = BOND_LINK_ARP_INTERV;
97 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
98 static char *arp_validate = NULL;
99 struct bond_params bonding_defaults;
101 module_param(max_bonds, int, 0);
102 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
103 module_param(miimon, int, 0);
104 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
105 module_param(updelay, int, 0);
106 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
107 module_param(downdelay, int, 0);
108 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
109 "in milliseconds");
110 module_param(use_carrier, int, 0);
111 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
112 "0 for off, 1 for on (default)");
113 module_param(mode, charp, 0);
114 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
115 "1 for active-backup, 2 for balance-xor, "
116 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
117 "6 for balance-alb");
118 module_param(primary, charp, 0);
119 MODULE_PARM_DESC(primary, "Primary network device to use");
120 module_param(lacp_rate, charp, 0);
121 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
122 "(slow/fast)");
123 module_param(xmit_hash_policy, charp, 0);
124 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
125 ", 1 for layer 3+4");
126 module_param(arp_interval, int, 0);
127 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
128 module_param_array(arp_ip_target, charp, NULL, 0);
129 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
130 module_param(arp_validate, charp, 0);
131 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
133 /*----------------------------- Global variables ----------------------------*/
135 static const char * const version =
136 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
138 LIST_HEAD(bond_dev_list);
140 #ifdef CONFIG_PROC_FS
141 static struct proc_dir_entry *bond_proc_dir = NULL;
142 #endif
144 extern struct rw_semaphore bonding_rwsem;
145 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
146 static int arp_ip_count = 0;
147 static int bond_mode = BOND_MODE_ROUNDROBIN;
148 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
149 static int lacp_fast = 0;
152 struct bond_parm_tbl bond_lacp_tbl[] = {
153 { "slow", AD_LACP_SLOW},
154 { "fast", AD_LACP_FAST},
155 { NULL, -1},
158 struct bond_parm_tbl bond_mode_tbl[] = {
159 { "balance-rr", BOND_MODE_ROUNDROBIN},
160 { "active-backup", BOND_MODE_ACTIVEBACKUP},
161 { "balance-xor", BOND_MODE_XOR},
162 { "broadcast", BOND_MODE_BROADCAST},
163 { "802.3ad", BOND_MODE_8023AD},
164 { "balance-tlb", BOND_MODE_TLB},
165 { "balance-alb", BOND_MODE_ALB},
166 { NULL, -1},
169 struct bond_parm_tbl xmit_hashtype_tbl[] = {
170 { "layer2", BOND_XMIT_POLICY_LAYER2},
171 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
172 { NULL, -1},
175 struct bond_parm_tbl arp_validate_tbl[] = {
176 { "none", BOND_ARP_VALIDATE_NONE},
177 { "active", BOND_ARP_VALIDATE_ACTIVE},
178 { "backup", BOND_ARP_VALIDATE_BACKUP},
179 { "all", BOND_ARP_VALIDATE_ALL},
180 { NULL, -1},
183 /*-------------------------- Forward declarations ---------------------------*/
185 static void bond_send_gratuitous_arp(struct bonding *bond);
187 /*---------------------------- General routines -----------------------------*/
189 const char *bond_mode_name(int mode)
191 switch (mode) {
192 case BOND_MODE_ROUNDROBIN :
193 return "load balancing (round-robin)";
194 case BOND_MODE_ACTIVEBACKUP :
195 return "fault-tolerance (active-backup)";
196 case BOND_MODE_XOR :
197 return "load balancing (xor)";
198 case BOND_MODE_BROADCAST :
199 return "fault-tolerance (broadcast)";
200 case BOND_MODE_8023AD:
201 return "IEEE 802.3ad Dynamic link aggregation";
202 case BOND_MODE_TLB:
203 return "transmit load balancing";
204 case BOND_MODE_ALB:
205 return "adaptive load balancing";
206 default:
207 return "unknown";
211 /*---------------------------------- VLAN -----------------------------------*/
214 * bond_add_vlan - add a new vlan id on bond
215 * @bond: bond that got the notification
216 * @vlan_id: the vlan id to add
218 * Returns -ENOMEM if allocation failed.
220 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
222 struct vlan_entry *vlan;
224 dprintk("bond: %s, vlan id %d\n",
225 (bond ? bond->dev->name: "None"), vlan_id);
227 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
228 if (!vlan) {
229 return -ENOMEM;
232 INIT_LIST_HEAD(&vlan->vlan_list);
233 vlan->vlan_id = vlan_id;
234 vlan->vlan_ip = 0;
236 write_lock_bh(&bond->lock);
238 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
240 write_unlock_bh(&bond->lock);
242 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
244 return 0;
248 * bond_del_vlan - delete a vlan id from bond
249 * @bond: bond that got the notification
250 * @vlan_id: the vlan id to delete
252 * returns -ENODEV if @vlan_id was not found in @bond.
254 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
256 struct vlan_entry *vlan, *next;
257 int res = -ENODEV;
259 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
261 write_lock_bh(&bond->lock);
263 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
264 if (vlan->vlan_id == vlan_id) {
265 list_del(&vlan->vlan_list);
267 if ((bond->params.mode == BOND_MODE_TLB) ||
268 (bond->params.mode == BOND_MODE_ALB)) {
269 bond_alb_clear_vlan(bond, vlan_id);
272 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
273 bond->dev->name);
275 kfree(vlan);
277 if (list_empty(&bond->vlan_list) &&
278 (bond->slave_cnt == 0)) {
279 /* Last VLAN removed and no slaves, so
280 * restore block on adding VLANs. This will
281 * be removed once new slaves that are not
282 * VLAN challenged will be added.
284 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
287 res = 0;
288 goto out;
292 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
293 bond->dev->name);
295 out:
296 write_unlock_bh(&bond->lock);
297 return res;
301 * bond_has_challenged_slaves
302 * @bond: the bond we're working on
304 * Searches the slave list. Returns 1 if a vlan challenged slave
305 * was found, 0 otherwise.
307 * Assumes bond->lock is held.
309 static int bond_has_challenged_slaves(struct bonding *bond)
311 struct slave *slave;
312 int i;
314 bond_for_each_slave(bond, slave, i) {
315 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
316 dprintk("found VLAN challenged slave - %s\n",
317 slave->dev->name);
318 return 1;
322 dprintk("no VLAN challenged slaves found\n");
323 return 0;
327 * bond_next_vlan - safely skip to the next item in the vlans list.
328 * @bond: the bond we're working on
329 * @curr: item we're advancing from
331 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
332 * or @curr->next otherwise (even if it is @curr itself again).
334 * Caller must hold bond->lock
336 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
338 struct vlan_entry *next, *last;
340 if (list_empty(&bond->vlan_list)) {
341 return NULL;
344 if (!curr) {
345 next = list_entry(bond->vlan_list.next,
346 struct vlan_entry, vlan_list);
347 } else {
348 last = list_entry(bond->vlan_list.prev,
349 struct vlan_entry, vlan_list);
350 if (last == curr) {
351 next = list_entry(bond->vlan_list.next,
352 struct vlan_entry, vlan_list);
353 } else {
354 next = list_entry(curr->vlan_list.next,
355 struct vlan_entry, vlan_list);
359 return next;
363 * bond_dev_queue_xmit - Prepare skb for xmit.
365 * @bond: bond device that got this skb for tx.
366 * @skb: hw accel VLAN tagged skb to transmit
367 * @slave_dev: slave that is supposed to xmit this skbuff
369 * When the bond gets an skb to transmit that is
370 * already hardware accelerated VLAN tagged, and it
371 * needs to relay this skb to a slave that is not
372 * hw accel capable, the skb needs to be "unaccelerated",
373 * i.e. strip the hwaccel tag and re-insert it as part
374 * of the payload.
376 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
378 unsigned short vlan_id;
380 if (!list_empty(&bond->vlan_list) &&
381 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
382 vlan_get_tag(skb, &vlan_id) == 0) {
383 skb->dev = slave_dev;
384 skb = vlan_put_tag(skb, vlan_id);
385 if (!skb) {
386 /* vlan_put_tag() frees the skb in case of error,
387 * so return success here so the calling functions
388 * won't attempt to free is again.
390 return 0;
392 } else {
393 skb->dev = slave_dev;
396 skb->priority = 1;
397 dev_queue_xmit(skb);
399 return 0;
403 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
404 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
405 * lock because:
406 * a. This operation is performed in IOCTL context,
407 * b. The operation is protected by the RTNL semaphore in the 8021q code,
408 * c. Holding a lock with BH disabled while directly calling a base driver
409 * entry point is generally a BAD idea.
411 * The design of synchronization/protection for this operation in the 8021q
412 * module is good for one or more VLAN devices over a single physical device
413 * and cannot be extended for a teaming solution like bonding, so there is a
414 * potential race condition here where a net device from the vlan group might
415 * be referenced (either by a base driver or the 8021q code) while it is being
416 * removed from the system. However, it turns out we're not making matters
417 * worse, and if it works for regular VLAN usage it will work here too.
421 * bond_vlan_rx_register - Propagates registration to slaves
422 * @bond_dev: bonding net device that got called
423 * @grp: vlan group being registered
425 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
427 struct bonding *bond = bond_dev->priv;
428 struct slave *slave;
429 int i;
431 bond->vlgrp = grp;
433 bond_for_each_slave(bond, slave, i) {
434 struct net_device *slave_dev = slave->dev;
436 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
437 slave_dev->vlan_rx_register) {
438 slave_dev->vlan_rx_register(slave_dev, grp);
444 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
445 * @bond_dev: bonding net device that got called
446 * @vid: vlan id being added
448 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
450 struct bonding *bond = bond_dev->priv;
451 struct slave *slave;
452 int i, res;
454 bond_for_each_slave(bond, slave, i) {
455 struct net_device *slave_dev = slave->dev;
457 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
458 slave_dev->vlan_rx_add_vid) {
459 slave_dev->vlan_rx_add_vid(slave_dev, vid);
463 res = bond_add_vlan(bond, vid);
464 if (res) {
465 printk(KERN_ERR DRV_NAME
466 ": %s: Error: Failed to add vlan id %d\n",
467 bond_dev->name, vid);
472 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
473 * @bond_dev: bonding net device that got called
474 * @vid: vlan id being removed
476 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
478 struct bonding *bond = bond_dev->priv;
479 struct slave *slave;
480 struct net_device *vlan_dev;
481 int i, res;
483 bond_for_each_slave(bond, slave, i) {
484 struct net_device *slave_dev = slave->dev;
486 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
487 slave_dev->vlan_rx_kill_vid) {
488 /* Save and then restore vlan_dev in the grp array,
489 * since the slave's driver might clear it.
491 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
492 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
493 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
497 res = bond_del_vlan(bond, vid);
498 if (res) {
499 printk(KERN_ERR DRV_NAME
500 ": %s: Error: Failed to remove vlan id %d\n",
501 bond_dev->name, vid);
505 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
507 struct vlan_entry *vlan;
509 write_lock_bh(&bond->lock);
511 if (list_empty(&bond->vlan_list)) {
512 goto out;
515 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
516 slave_dev->vlan_rx_register) {
517 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
520 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
521 !(slave_dev->vlan_rx_add_vid)) {
522 goto out;
525 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
526 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
529 out:
530 write_unlock_bh(&bond->lock);
533 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
535 struct vlan_entry *vlan;
536 struct net_device *vlan_dev;
538 write_lock_bh(&bond->lock);
540 if (list_empty(&bond->vlan_list)) {
541 goto out;
544 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
545 !(slave_dev->vlan_rx_kill_vid)) {
546 goto unreg;
549 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
550 /* Save and then restore vlan_dev in the grp array,
551 * since the slave's driver might clear it.
553 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
554 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
555 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
558 unreg:
559 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
560 slave_dev->vlan_rx_register) {
561 slave_dev->vlan_rx_register(slave_dev, NULL);
564 out:
565 write_unlock_bh(&bond->lock);
568 /*------------------------------- Link status -------------------------------*/
571 * Set the carrier state for the master according to the state of its
572 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
573 * do special 802.3ad magic.
575 * Returns zero if carrier state does not change, nonzero if it does.
577 static int bond_set_carrier(struct bonding *bond)
579 struct slave *slave;
580 int i;
582 if (bond->slave_cnt == 0)
583 goto down;
585 if (bond->params.mode == BOND_MODE_8023AD)
586 return bond_3ad_set_carrier(bond);
588 bond_for_each_slave(bond, slave, i) {
589 if (slave->link == BOND_LINK_UP) {
590 if (!netif_carrier_ok(bond->dev)) {
591 netif_carrier_on(bond->dev);
592 return 1;
594 return 0;
598 down:
599 if (netif_carrier_ok(bond->dev)) {
600 netif_carrier_off(bond->dev);
601 return 1;
603 return 0;
607 * Get link speed and duplex from the slave's base driver
608 * using ethtool. If for some reason the call fails or the
609 * values are invalid, fake speed and duplex to 100/Full
610 * and return error.
612 static int bond_update_speed_duplex(struct slave *slave)
614 struct net_device *slave_dev = slave->dev;
615 static int (* ioctl)(struct net_device *, struct ifreq *, int);
616 struct ifreq ifr;
617 struct ethtool_cmd etool;
619 /* Fake speed and duplex */
620 slave->speed = SPEED_100;
621 slave->duplex = DUPLEX_FULL;
623 if (slave_dev->ethtool_ops) {
624 int res;
626 if (!slave_dev->ethtool_ops->get_settings) {
627 return -1;
630 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
631 if (res < 0) {
632 return -1;
635 goto verify;
638 ioctl = slave_dev->do_ioctl;
639 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
640 etool.cmd = ETHTOOL_GSET;
641 ifr.ifr_data = (char*)&etool;
642 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
643 return -1;
646 verify:
647 switch (etool.speed) {
648 case SPEED_10:
649 case SPEED_100:
650 case SPEED_1000:
651 case SPEED_10000:
652 break;
653 default:
654 return -1;
657 switch (etool.duplex) {
658 case DUPLEX_FULL:
659 case DUPLEX_HALF:
660 break;
661 default:
662 return -1;
665 slave->speed = etool.speed;
666 slave->duplex = etool.duplex;
668 return 0;
672 * if <dev> supports MII link status reporting, check its link status.
674 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
675 * depening upon the setting of the use_carrier parameter.
677 * Return either BMSR_LSTATUS, meaning that the link is up (or we
678 * can't tell and just pretend it is), or 0, meaning that the link is
679 * down.
681 * If reporting is non-zero, instead of faking link up, return -1 if
682 * both ETHTOOL and MII ioctls fail (meaning the device does not
683 * support them). If use_carrier is set, return whatever it says.
684 * It'd be nice if there was a good way to tell if a driver supports
685 * netif_carrier, but there really isn't.
687 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
689 static int (* ioctl)(struct net_device *, struct ifreq *, int);
690 struct ifreq ifr;
691 struct mii_ioctl_data *mii;
692 struct ethtool_value etool;
694 if (bond->params.use_carrier) {
695 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
698 ioctl = slave_dev->do_ioctl;
699 if (ioctl) {
700 /* TODO: set pointer to correct ioctl on a per team member */
701 /* bases to make this more efficient. that is, once */
702 /* we determine the correct ioctl, we will always */
703 /* call it and not the others for that team */
704 /* member. */
707 * We cannot assume that SIOCGMIIPHY will also read a
708 * register; not all network drivers (e.g., e100)
709 * support that.
712 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
713 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
714 mii = if_mii(&ifr);
715 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
716 mii->reg_num = MII_BMSR;
717 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
718 return (mii->val_out & BMSR_LSTATUS);
723 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
724 /* for a period of time so we attempt to get link status */
725 /* from it last if the above MII ioctls fail... */
726 if (slave_dev->ethtool_ops) {
727 if (slave_dev->ethtool_ops->get_link) {
728 u32 link;
730 link = slave_dev->ethtool_ops->get_link(slave_dev);
732 return link ? BMSR_LSTATUS : 0;
736 if (ioctl) {
737 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
738 etool.cmd = ETHTOOL_GLINK;
739 ifr.ifr_data = (char*)&etool;
740 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
741 if (etool.data == 1) {
742 return BMSR_LSTATUS;
743 } else {
744 dprintk("SIOCETHTOOL shows link down\n");
745 return 0;
751 * If reporting, report that either there's no dev->do_ioctl,
752 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
753 * cannot report link status). If not reporting, pretend
754 * we're ok.
756 return (reporting ? -1 : BMSR_LSTATUS);
759 /*----------------------------- Multicast list ------------------------------*/
762 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
764 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
766 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
767 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
771 * returns dmi entry if found, NULL otherwise
773 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
775 struct dev_mc_list *idmi;
777 for (idmi = mc_list; idmi; idmi = idmi->next) {
778 if (bond_is_dmi_same(dmi, idmi)) {
779 return idmi;
783 return NULL;
787 * Push the promiscuity flag down to appropriate slaves
789 static void bond_set_promiscuity(struct bonding *bond, int inc)
791 if (USES_PRIMARY(bond->params.mode)) {
792 /* write lock already acquired */
793 if (bond->curr_active_slave) {
794 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
796 } else {
797 struct slave *slave;
798 int i;
799 bond_for_each_slave(bond, slave, i) {
800 dev_set_promiscuity(slave->dev, inc);
806 * Push the allmulti flag down to all slaves
808 static void bond_set_allmulti(struct bonding *bond, int inc)
810 if (USES_PRIMARY(bond->params.mode)) {
811 /* write lock already acquired */
812 if (bond->curr_active_slave) {
813 dev_set_allmulti(bond->curr_active_slave->dev, inc);
815 } else {
816 struct slave *slave;
817 int i;
818 bond_for_each_slave(bond, slave, i) {
819 dev_set_allmulti(slave->dev, inc);
825 * Add a Multicast address to slaves
826 * according to mode
828 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
830 if (USES_PRIMARY(bond->params.mode)) {
831 /* write lock already acquired */
832 if (bond->curr_active_slave) {
833 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
835 } else {
836 struct slave *slave;
837 int i;
838 bond_for_each_slave(bond, slave, i) {
839 dev_mc_add(slave->dev, addr, alen, 0);
845 * Remove a multicast address from slave
846 * according to mode
848 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
850 if (USES_PRIMARY(bond->params.mode)) {
851 /* write lock already acquired */
852 if (bond->curr_active_slave) {
853 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
855 } else {
856 struct slave *slave;
857 int i;
858 bond_for_each_slave(bond, slave, i) {
859 dev_mc_delete(slave->dev, addr, alen, 0);
865 * Totally destroys the mc_list in bond
867 static void bond_mc_list_destroy(struct bonding *bond)
869 struct dev_mc_list *dmi;
871 dmi = bond->mc_list;
872 while (dmi) {
873 bond->mc_list = dmi->next;
874 kfree(dmi);
875 dmi = bond->mc_list;
880 * Copy all the Multicast addresses from src to the bonding device dst
882 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
883 gfp_t gfp_flag)
885 struct dev_mc_list *dmi, *new_dmi;
887 for (dmi = mc_list; dmi; dmi = dmi->next) {
888 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
890 if (!new_dmi) {
891 /* FIXME: Potential memory leak !!! */
892 return -ENOMEM;
895 new_dmi->next = bond->mc_list;
896 bond->mc_list = new_dmi;
897 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
898 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
899 new_dmi->dmi_users = dmi->dmi_users;
900 new_dmi->dmi_gusers = dmi->dmi_gusers;
903 return 0;
907 * flush all members of flush->mc_list from device dev->mc_list
909 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
911 struct bonding *bond = bond_dev->priv;
912 struct dev_mc_list *dmi;
914 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
915 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
918 if (bond->params.mode == BOND_MODE_8023AD) {
919 /* del lacpdu mc addr from mc list */
920 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
922 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
926 /*--------------------------- Active slave change ---------------------------*/
929 * Update the mc list and multicast-related flags for the new and
930 * old active slaves (if any) according to the multicast mode, and
931 * promiscuous flags unconditionally.
933 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
935 struct dev_mc_list *dmi;
937 if (!USES_PRIMARY(bond->params.mode)) {
938 /* nothing to do - mc list is already up-to-date on
939 * all slaves
941 return;
944 if (old_active) {
945 if (bond->dev->flags & IFF_PROMISC) {
946 dev_set_promiscuity(old_active->dev, -1);
949 if (bond->dev->flags & IFF_ALLMULTI) {
950 dev_set_allmulti(old_active->dev, -1);
953 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
954 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
958 if (new_active) {
959 if (bond->dev->flags & IFF_PROMISC) {
960 dev_set_promiscuity(new_active->dev, 1);
963 if (bond->dev->flags & IFF_ALLMULTI) {
964 dev_set_allmulti(new_active->dev, 1);
967 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
968 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
974 * find_best_interface - select the best available slave to be the active one
975 * @bond: our bonding struct
977 * Warning: Caller must hold curr_slave_lock for writing.
979 static struct slave *bond_find_best_slave(struct bonding *bond)
981 struct slave *new_active, *old_active;
982 struct slave *bestslave = NULL;
983 int mintime = bond->params.updelay;
984 int i;
986 new_active = old_active = bond->curr_active_slave;
988 if (!new_active) { /* there were no active slaves left */
989 if (bond->slave_cnt > 0) { /* found one slave */
990 new_active = bond->first_slave;
991 } else {
992 return NULL; /* still no slave, return NULL */
996 /* first try the primary link; if arping, a link must tx/rx traffic
997 * before it can be considered the curr_active_slave - also, we would skip
998 * slaves between the curr_active_slave and primary_slave that may be up
999 * and able to arp
1001 if ((bond->primary_slave) &&
1002 (!bond->params.arp_interval) &&
1003 (IS_UP(bond->primary_slave->dev))) {
1004 new_active = bond->primary_slave;
1007 /* remember where to stop iterating over the slaves */
1008 old_active = new_active;
1010 bond_for_each_slave_from(bond, new_active, i, old_active) {
1011 if (IS_UP(new_active->dev)) {
1012 if (new_active->link == BOND_LINK_UP) {
1013 return new_active;
1014 } else if (new_active->link == BOND_LINK_BACK) {
1015 /* link up, but waiting for stabilization */
1016 if (new_active->delay < mintime) {
1017 mintime = new_active->delay;
1018 bestslave = new_active;
1024 return bestslave;
1028 * change_active_interface - change the active slave into the specified one
1029 * @bond: our bonding struct
1030 * @new: the new slave to make the active one
1032 * Set the new slave to the bond's settings and unset them on the old
1033 * curr_active_slave.
1034 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1036 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1037 * because it is apparently the best available slave we have, even though its
1038 * updelay hasn't timed out yet.
1040 * Warning: Caller must hold curr_slave_lock for writing.
1042 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1044 struct slave *old_active = bond->curr_active_slave;
1046 if (old_active == new_active) {
1047 return;
1050 if (new_active) {
1051 if (new_active->link == BOND_LINK_BACK) {
1052 if (USES_PRIMARY(bond->params.mode)) {
1053 printk(KERN_INFO DRV_NAME
1054 ": %s: making interface %s the new "
1055 "active one %d ms earlier.\n",
1056 bond->dev->name, new_active->dev->name,
1057 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1060 new_active->delay = 0;
1061 new_active->link = BOND_LINK_UP;
1062 new_active->jiffies = jiffies;
1064 if (bond->params.mode == BOND_MODE_8023AD) {
1065 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1068 if ((bond->params.mode == BOND_MODE_TLB) ||
1069 (bond->params.mode == BOND_MODE_ALB)) {
1070 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1072 } else {
1073 if (USES_PRIMARY(bond->params.mode)) {
1074 printk(KERN_INFO DRV_NAME
1075 ": %s: making interface %s the new "
1076 "active one.\n",
1077 bond->dev->name, new_active->dev->name);
1082 if (USES_PRIMARY(bond->params.mode)) {
1083 bond_mc_swap(bond, new_active, old_active);
1086 if ((bond->params.mode == BOND_MODE_TLB) ||
1087 (bond->params.mode == BOND_MODE_ALB)) {
1088 bond_alb_handle_active_change(bond, new_active);
1089 if (old_active)
1090 bond_set_slave_inactive_flags(old_active);
1091 if (new_active)
1092 bond_set_slave_active_flags(new_active);
1093 } else {
1094 bond->curr_active_slave = new_active;
1097 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1098 if (old_active) {
1099 bond_set_slave_inactive_flags(old_active);
1102 if (new_active) {
1103 bond_set_slave_active_flags(new_active);
1105 bond_send_gratuitous_arp(bond);
1110 * bond_select_active_slave - select a new active slave, if needed
1111 * @bond: our bonding struct
1113 * This functions shoud be called when one of the following occurs:
1114 * - The old curr_active_slave has been released or lost its link.
1115 * - The primary_slave has got its link back.
1116 * - A slave has got its link back and there's no old curr_active_slave.
1118 * Warning: Caller must hold curr_slave_lock for writing.
1120 void bond_select_active_slave(struct bonding *bond)
1122 struct slave *best_slave;
1123 int rv;
1125 best_slave = bond_find_best_slave(bond);
1126 if (best_slave != bond->curr_active_slave) {
1127 bond_change_active_slave(bond, best_slave);
1128 rv = bond_set_carrier(bond);
1129 if (!rv)
1130 return;
1132 if (netif_carrier_ok(bond->dev)) {
1133 printk(KERN_INFO DRV_NAME
1134 ": %s: first active interface up!\n",
1135 bond->dev->name);
1136 } else {
1137 printk(KERN_INFO DRV_NAME ": %s: "
1138 "now running without any active interface !\n",
1139 bond->dev->name);
1144 /*--------------------------- slave list handling ---------------------------*/
1147 * This function attaches the slave to the end of list.
1149 * bond->lock held for writing by caller.
1151 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1153 if (bond->first_slave == NULL) { /* attaching the first slave */
1154 new_slave->next = new_slave;
1155 new_slave->prev = new_slave;
1156 bond->first_slave = new_slave;
1157 } else {
1158 new_slave->next = bond->first_slave;
1159 new_slave->prev = bond->first_slave->prev;
1160 new_slave->next->prev = new_slave;
1161 new_slave->prev->next = new_slave;
1164 bond->slave_cnt++;
1168 * This function detaches the slave from the list.
1169 * WARNING: no check is made to verify if the slave effectively
1170 * belongs to <bond>.
1171 * Nothing is freed on return, structures are just unchained.
1172 * If any slave pointer in bond was pointing to <slave>,
1173 * it should be changed by the calling function.
1175 * bond->lock held for writing by caller.
1177 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1179 if (slave->next) {
1180 slave->next->prev = slave->prev;
1183 if (slave->prev) {
1184 slave->prev->next = slave->next;
1187 if (bond->first_slave == slave) { /* slave is the first slave */
1188 if (bond->slave_cnt > 1) { /* there are more slave */
1189 bond->first_slave = slave->next;
1190 } else {
1191 bond->first_slave = NULL; /* slave was the last one */
1195 slave->next = NULL;
1196 slave->prev = NULL;
1197 bond->slave_cnt--;
1200 /*---------------------------------- IOCTL ----------------------------------*/
1202 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1204 dprintk("bond_dev=%p\n", bond_dev);
1205 dprintk("slave_dev=%p\n", slave_dev);
1206 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1207 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1208 return 0;
1211 #define BOND_INTERSECT_FEATURES \
1212 (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_TSO | NETIF_F_UFO)
1215 * Compute the common dev->feature set available to all slaves. Some
1216 * feature bits are managed elsewhere, so preserve feature bits set on
1217 * master device that are not part of the examined set.
1219 static int bond_compute_features(struct bonding *bond)
1221 unsigned long features = BOND_INTERSECT_FEATURES;
1222 struct slave *slave;
1223 struct net_device *bond_dev = bond->dev;
1224 unsigned short max_hard_header_len = ETH_HLEN;
1225 int i;
1227 bond_for_each_slave(bond, slave, i) {
1228 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1229 if (slave->dev->hard_header_len > max_hard_header_len)
1230 max_hard_header_len = slave->dev->hard_header_len;
1233 if ((features & NETIF_F_SG) &&
1234 !(features & NETIF_F_ALL_CSUM))
1235 features &= ~NETIF_F_SG;
1238 * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1239 * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1240 * implies that all slaves also support scatter-gather
1241 * (NETIF_F_SG), which implies that features also includes
1242 * NETIF_F_SG. So no need to check whether we have an
1243 * illegal combination of NETIF_F_{TSO,UFO} and
1244 * !NETIF_F_SG
1247 features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1248 bond_dev->features = features;
1249 bond_dev->hard_header_len = max_hard_header_len;
1251 return 0;
1254 /* enslave device <slave> to bond device <master> */
1255 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1257 struct bonding *bond = bond_dev->priv;
1258 struct slave *new_slave = NULL;
1259 struct dev_mc_list *dmi;
1260 struct sockaddr addr;
1261 int link_reporting;
1262 int old_features = bond_dev->features;
1263 int res = 0;
1265 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1266 slave_dev->do_ioctl == NULL) {
1267 printk(KERN_WARNING DRV_NAME
1268 ": %s: Warning: no link monitoring support for %s\n",
1269 bond_dev->name, slave_dev->name);
1272 /* bond must be initialized by bond_open() before enslaving */
1273 if (!(bond_dev->flags & IFF_UP)) {
1274 dprintk("Error, master_dev is not up\n");
1275 return -EPERM;
1278 /* already enslaved */
1279 if (slave_dev->flags & IFF_SLAVE) {
1280 dprintk("Error, Device was already enslaved\n");
1281 return -EBUSY;
1284 /* vlan challenged mutual exclusion */
1285 /* no need to lock since we're protected by rtnl_lock */
1286 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1287 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1288 if (!list_empty(&bond->vlan_list)) {
1289 printk(KERN_ERR DRV_NAME
1290 ": %s: Error: cannot enslave VLAN "
1291 "challenged slave %s on VLAN enabled "
1292 "bond %s\n", bond_dev->name, slave_dev->name,
1293 bond_dev->name);
1294 return -EPERM;
1295 } else {
1296 printk(KERN_WARNING DRV_NAME
1297 ": %s: Warning: enslaved VLAN challenged "
1298 "slave %s. Adding VLANs will be blocked as "
1299 "long as %s is part of bond %s\n",
1300 bond_dev->name, slave_dev->name, slave_dev->name,
1301 bond_dev->name);
1302 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1304 } else {
1305 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1306 if (bond->slave_cnt == 0) {
1307 /* First slave, and it is not VLAN challenged,
1308 * so remove the block of adding VLANs over the bond.
1310 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1315 * Old ifenslave binaries are no longer supported. These can
1316 * be identified with moderate accurary by the state of the slave:
1317 * the current ifenslave will set the interface down prior to
1318 * enslaving it; the old ifenslave will not.
1320 if ((slave_dev->flags & IFF_UP)) {
1321 printk(KERN_ERR DRV_NAME ": %s is up. "
1322 "This may be due to an out of date ifenslave.\n",
1323 slave_dev->name);
1324 res = -EPERM;
1325 goto err_undo_flags;
1328 if (slave_dev->set_mac_address == NULL) {
1329 printk(KERN_ERR DRV_NAME
1330 ": %s: Error: The slave device you specified does "
1331 "not support setting the MAC address. "
1332 "Your kernel likely does not support slave "
1333 "devices.\n", bond_dev->name);
1334 res = -EOPNOTSUPP;
1335 goto err_undo_flags;
1338 if (slave_dev->get_stats == NULL) {
1339 printk(KERN_NOTICE DRV_NAME
1340 ": %s: the driver for slave device %s does not provide "
1341 "get_stats function, network statistics will be "
1342 "inaccurate.\n", bond_dev->name, slave_dev->name);
1345 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1346 if (!new_slave) {
1347 res = -ENOMEM;
1348 goto err_undo_flags;
1351 /* save slave's original flags before calling
1352 * netdev_set_master and dev_open
1354 new_slave->original_flags = slave_dev->flags;
1357 * Save slave's original ("permanent") mac address for modes
1358 * that need it, and for restoring it upon release, and then
1359 * set it to the master's address
1361 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1364 * Set slave to master's mac address. The application already
1365 * set the master's mac address to that of the first slave
1367 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1368 addr.sa_family = slave_dev->type;
1369 res = dev_set_mac_address(slave_dev, &addr);
1370 if (res) {
1371 dprintk("Error %d calling set_mac_address\n", res);
1372 goto err_free;
1375 /* open the slave since the application closed it */
1376 res = dev_open(slave_dev);
1377 if (res) {
1378 dprintk("Openning slave %s failed\n", slave_dev->name);
1379 goto err_restore_mac;
1382 res = netdev_set_master(slave_dev, bond_dev);
1383 if (res) {
1384 dprintk("Error %d calling netdev_set_master\n", res);
1385 goto err_close;
1388 new_slave->dev = slave_dev;
1389 slave_dev->priv_flags |= IFF_BONDING;
1391 if ((bond->params.mode == BOND_MODE_TLB) ||
1392 (bond->params.mode == BOND_MODE_ALB)) {
1393 /* bond_alb_init_slave() must be called before all other stages since
1394 * it might fail and we do not want to have to undo everything
1396 res = bond_alb_init_slave(bond, new_slave);
1397 if (res) {
1398 goto err_unset_master;
1402 /* If the mode USES_PRIMARY, then the new slave gets the
1403 * master's promisc (and mc) settings only if it becomes the
1404 * curr_active_slave, and that is taken care of later when calling
1405 * bond_change_active()
1407 if (!USES_PRIMARY(bond->params.mode)) {
1408 /* set promiscuity level to new slave */
1409 if (bond_dev->flags & IFF_PROMISC) {
1410 dev_set_promiscuity(slave_dev, 1);
1413 /* set allmulti level to new slave */
1414 if (bond_dev->flags & IFF_ALLMULTI) {
1415 dev_set_allmulti(slave_dev, 1);
1418 /* upload master's mc_list to new slave */
1419 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1420 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1424 if (bond->params.mode == BOND_MODE_8023AD) {
1425 /* add lacpdu mc addr to mc list */
1426 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1428 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1431 bond_add_vlans_on_slave(bond, slave_dev);
1433 write_lock_bh(&bond->lock);
1435 bond_attach_slave(bond, new_slave);
1437 new_slave->delay = 0;
1438 new_slave->link_failure_count = 0;
1440 bond_compute_features(bond);
1442 new_slave->last_arp_rx = jiffies;
1444 if (bond->params.miimon && !bond->params.use_carrier) {
1445 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1447 if ((link_reporting == -1) && !bond->params.arp_interval) {
1449 * miimon is set but a bonded network driver
1450 * does not support ETHTOOL/MII and
1451 * arp_interval is not set. Note: if
1452 * use_carrier is enabled, we will never go
1453 * here (because netif_carrier is always
1454 * supported); thus, we don't need to change
1455 * the messages for netif_carrier.
1457 printk(KERN_WARNING DRV_NAME
1458 ": %s: Warning: MII and ETHTOOL support not "
1459 "available for interface %s, and "
1460 "arp_interval/arp_ip_target module parameters "
1461 "not specified, thus bonding will not detect "
1462 "link failures! see bonding.txt for details.\n",
1463 bond_dev->name, slave_dev->name);
1464 } else if (link_reporting == -1) {
1465 /* unable get link status using mii/ethtool */
1466 printk(KERN_WARNING DRV_NAME
1467 ": %s: Warning: can't get link status from "
1468 "interface %s; the network driver associated "
1469 "with this interface does not support MII or "
1470 "ETHTOOL link status reporting, thus miimon "
1471 "has no effect on this interface.\n",
1472 bond_dev->name, slave_dev->name);
1476 /* check for initial state */
1477 if (!bond->params.miimon ||
1478 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1479 if (bond->params.updelay) {
1480 dprintk("Initial state of slave_dev is "
1481 "BOND_LINK_BACK\n");
1482 new_slave->link = BOND_LINK_BACK;
1483 new_slave->delay = bond->params.updelay;
1484 } else {
1485 dprintk("Initial state of slave_dev is "
1486 "BOND_LINK_UP\n");
1487 new_slave->link = BOND_LINK_UP;
1489 new_slave->jiffies = jiffies;
1490 } else {
1491 dprintk("Initial state of slave_dev is "
1492 "BOND_LINK_DOWN\n");
1493 new_slave->link = BOND_LINK_DOWN;
1496 if (bond_update_speed_duplex(new_slave) &&
1497 (new_slave->link != BOND_LINK_DOWN)) {
1498 printk(KERN_WARNING DRV_NAME
1499 ": %s: Warning: failed to get speed and duplex from %s, "
1500 "assumed to be 100Mb/sec and Full.\n",
1501 bond_dev->name, new_slave->dev->name);
1503 if (bond->params.mode == BOND_MODE_8023AD) {
1504 printk(KERN_WARNING DRV_NAME
1505 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1506 "support in base driver for proper aggregator "
1507 "selection.\n", bond_dev->name);
1511 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1512 /* if there is a primary slave, remember it */
1513 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1514 bond->primary_slave = new_slave;
1518 switch (bond->params.mode) {
1519 case BOND_MODE_ACTIVEBACKUP:
1520 bond_set_slave_inactive_flags(new_slave);
1521 bond_select_active_slave(bond);
1522 break;
1523 case BOND_MODE_8023AD:
1524 /* in 802.3ad mode, the internal mechanism
1525 * will activate the slaves in the selected
1526 * aggregator
1528 bond_set_slave_inactive_flags(new_slave);
1529 /* if this is the first slave */
1530 if (bond->slave_cnt == 1) {
1531 SLAVE_AD_INFO(new_slave).id = 1;
1532 /* Initialize AD with the number of times that the AD timer is called in 1 second
1533 * can be called only after the mac address of the bond is set
1535 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1536 bond->params.lacp_fast);
1537 } else {
1538 SLAVE_AD_INFO(new_slave).id =
1539 SLAVE_AD_INFO(new_slave->prev).id + 1;
1542 bond_3ad_bind_slave(new_slave);
1543 break;
1544 case BOND_MODE_TLB:
1545 case BOND_MODE_ALB:
1546 new_slave->state = BOND_STATE_ACTIVE;
1547 if ((!bond->curr_active_slave) &&
1548 (new_slave->link != BOND_LINK_DOWN)) {
1549 /* first slave or no active slave yet, and this link
1550 * is OK, so make this interface the active one
1552 bond_change_active_slave(bond, new_slave);
1553 } else {
1554 bond_set_slave_inactive_flags(new_slave);
1556 break;
1557 default:
1558 dprintk("This slave is always active in trunk mode\n");
1560 /* always active in trunk mode */
1561 new_slave->state = BOND_STATE_ACTIVE;
1563 /* In trunking mode there is little meaning to curr_active_slave
1564 * anyway (it holds no special properties of the bond device),
1565 * so we can change it without calling change_active_interface()
1567 if (!bond->curr_active_slave) {
1568 bond->curr_active_slave = new_slave;
1570 break;
1571 } /* switch(bond_mode) */
1573 bond_set_carrier(bond);
1575 write_unlock_bh(&bond->lock);
1577 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1578 if (res)
1579 goto err_unset_master;
1581 printk(KERN_INFO DRV_NAME
1582 ": %s: enslaving %s as a%s interface with a%s link.\n",
1583 bond_dev->name, slave_dev->name,
1584 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1585 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1587 /* enslave is successful */
1588 return 0;
1590 /* Undo stages on error */
1591 err_unset_master:
1592 netdev_set_master(slave_dev, NULL);
1594 err_close:
1595 dev_close(slave_dev);
1597 err_restore_mac:
1598 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1599 addr.sa_family = slave_dev->type;
1600 dev_set_mac_address(slave_dev, &addr);
1602 err_free:
1603 kfree(new_slave);
1605 err_undo_flags:
1606 bond_dev->features = old_features;
1608 return res;
1612 * Try to release the slave device <slave> from the bond device <master>
1613 * It is legal to access curr_active_slave without a lock because all the function
1614 * is write-locked.
1616 * The rules for slave state should be:
1617 * for Active/Backup:
1618 * Active stays on all backups go down
1619 * for Bonded connections:
1620 * The first up interface should be left on and all others downed.
1622 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1624 struct bonding *bond = bond_dev->priv;
1625 struct slave *slave, *oldcurrent;
1626 struct sockaddr addr;
1627 int mac_addr_differ;
1629 /* slave is not a slave or master is not master of this slave */
1630 if (!(slave_dev->flags & IFF_SLAVE) ||
1631 (slave_dev->master != bond_dev)) {
1632 printk(KERN_ERR DRV_NAME
1633 ": %s: Error: cannot release %s.\n",
1634 bond_dev->name, slave_dev->name);
1635 return -EINVAL;
1638 write_lock_bh(&bond->lock);
1640 slave = bond_get_slave_by_dev(bond, slave_dev);
1641 if (!slave) {
1642 /* not a slave of this bond */
1643 printk(KERN_INFO DRV_NAME
1644 ": %s: %s not enslaved\n",
1645 bond_dev->name, slave_dev->name);
1646 write_unlock_bh(&bond->lock);
1647 return -EINVAL;
1650 mac_addr_differ = memcmp(bond_dev->dev_addr,
1651 slave->perm_hwaddr,
1652 ETH_ALEN);
1653 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1654 printk(KERN_WARNING DRV_NAME
1655 ": %s: Warning: the permanent HWaddr of %s "
1656 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1657 "still in use by %s. Set the HWaddr of "
1658 "%s to a different address to avoid "
1659 "conflicts.\n",
1660 bond_dev->name,
1661 slave_dev->name,
1662 slave->perm_hwaddr[0],
1663 slave->perm_hwaddr[1],
1664 slave->perm_hwaddr[2],
1665 slave->perm_hwaddr[3],
1666 slave->perm_hwaddr[4],
1667 slave->perm_hwaddr[5],
1668 bond_dev->name,
1669 slave_dev->name);
1672 /* Inform AD package of unbinding of slave. */
1673 if (bond->params.mode == BOND_MODE_8023AD) {
1674 /* must be called before the slave is
1675 * detached from the list
1677 bond_3ad_unbind_slave(slave);
1680 printk(KERN_INFO DRV_NAME
1681 ": %s: releasing %s interface %s\n",
1682 bond_dev->name,
1683 (slave->state == BOND_STATE_ACTIVE)
1684 ? "active" : "backup",
1685 slave_dev->name);
1687 oldcurrent = bond->curr_active_slave;
1689 bond->current_arp_slave = NULL;
1691 /* release the slave from its bond */
1692 bond_detach_slave(bond, slave);
1694 bond_compute_features(bond);
1696 if (bond->primary_slave == slave) {
1697 bond->primary_slave = NULL;
1700 if (oldcurrent == slave) {
1701 bond_change_active_slave(bond, NULL);
1704 if ((bond->params.mode == BOND_MODE_TLB) ||
1705 (bond->params.mode == BOND_MODE_ALB)) {
1706 /* Must be called only after the slave has been
1707 * detached from the list and the curr_active_slave
1708 * has been cleared (if our_slave == old_current),
1709 * but before a new active slave is selected.
1711 bond_alb_deinit_slave(bond, slave);
1714 if (oldcurrent == slave)
1715 bond_select_active_slave(bond);
1717 if (bond->slave_cnt == 0) {
1718 bond_set_carrier(bond);
1720 /* if the last slave was removed, zero the mac address
1721 * of the master so it will be set by the application
1722 * to the mac address of the first slave
1724 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1726 if (list_empty(&bond->vlan_list)) {
1727 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1728 } else {
1729 printk(KERN_WARNING DRV_NAME
1730 ": %s: Warning: clearing HW address of %s while it "
1731 "still has VLANs.\n",
1732 bond_dev->name, bond_dev->name);
1733 printk(KERN_WARNING DRV_NAME
1734 ": %s: When re-adding slaves, make sure the bond's "
1735 "HW address matches its VLANs'.\n",
1736 bond_dev->name);
1738 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1739 !bond_has_challenged_slaves(bond)) {
1740 printk(KERN_INFO DRV_NAME
1741 ": %s: last VLAN challenged slave %s "
1742 "left bond %s. VLAN blocking is removed\n",
1743 bond_dev->name, slave_dev->name, bond_dev->name);
1744 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1747 write_unlock_bh(&bond->lock);
1749 /* must do this from outside any spinlocks */
1750 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1752 bond_del_vlans_from_slave(bond, slave_dev);
1754 /* If the mode USES_PRIMARY, then we should only remove its
1755 * promisc and mc settings if it was the curr_active_slave, but that was
1756 * already taken care of above when we detached the slave
1758 if (!USES_PRIMARY(bond->params.mode)) {
1759 /* unset promiscuity level from slave */
1760 if (bond_dev->flags & IFF_PROMISC) {
1761 dev_set_promiscuity(slave_dev, -1);
1764 /* unset allmulti level from slave */
1765 if (bond_dev->flags & IFF_ALLMULTI) {
1766 dev_set_allmulti(slave_dev, -1);
1769 /* flush master's mc_list from slave */
1770 bond_mc_list_flush(bond_dev, slave_dev);
1773 netdev_set_master(slave_dev, NULL);
1775 /* close slave before restoring its mac address */
1776 dev_close(slave_dev);
1778 /* restore original ("permanent") mac address */
1779 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1780 addr.sa_family = slave_dev->type;
1781 dev_set_mac_address(slave_dev, &addr);
1783 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1784 IFF_SLAVE_INACTIVE | IFF_BONDING |
1785 IFF_SLAVE_NEEDARP);
1787 kfree(slave);
1789 return 0; /* deletion OK */
1793 * This function releases all slaves.
1795 static int bond_release_all(struct net_device *bond_dev)
1797 struct bonding *bond = bond_dev->priv;
1798 struct slave *slave;
1799 struct net_device *slave_dev;
1800 struct sockaddr addr;
1802 write_lock_bh(&bond->lock);
1804 netif_carrier_off(bond_dev);
1806 if (bond->slave_cnt == 0) {
1807 goto out;
1810 bond->current_arp_slave = NULL;
1811 bond->primary_slave = NULL;
1812 bond_change_active_slave(bond, NULL);
1814 while ((slave = bond->first_slave) != NULL) {
1815 /* Inform AD package of unbinding of slave
1816 * before slave is detached from the list.
1818 if (bond->params.mode == BOND_MODE_8023AD) {
1819 bond_3ad_unbind_slave(slave);
1822 slave_dev = slave->dev;
1823 bond_detach_slave(bond, slave);
1825 if ((bond->params.mode == BOND_MODE_TLB) ||
1826 (bond->params.mode == BOND_MODE_ALB)) {
1827 /* must be called only after the slave
1828 * has been detached from the list
1830 bond_alb_deinit_slave(bond, slave);
1833 bond_compute_features(bond);
1835 /* now that the slave is detached, unlock and perform
1836 * all the undo steps that should not be called from
1837 * within a lock.
1839 write_unlock_bh(&bond->lock);
1841 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1842 bond_del_vlans_from_slave(bond, slave_dev);
1844 /* If the mode USES_PRIMARY, then we should only remove its
1845 * promisc and mc settings if it was the curr_active_slave, but that was
1846 * already taken care of above when we detached the slave
1848 if (!USES_PRIMARY(bond->params.mode)) {
1849 /* unset promiscuity level from slave */
1850 if (bond_dev->flags & IFF_PROMISC) {
1851 dev_set_promiscuity(slave_dev, -1);
1854 /* unset allmulti level from slave */
1855 if (bond_dev->flags & IFF_ALLMULTI) {
1856 dev_set_allmulti(slave_dev, -1);
1859 /* flush master's mc_list from slave */
1860 bond_mc_list_flush(bond_dev, slave_dev);
1863 netdev_set_master(slave_dev, NULL);
1865 /* close slave before restoring its mac address */
1866 dev_close(slave_dev);
1868 /* restore original ("permanent") mac address*/
1869 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1870 addr.sa_family = slave_dev->type;
1871 dev_set_mac_address(slave_dev, &addr);
1873 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1874 IFF_SLAVE_INACTIVE);
1876 kfree(slave);
1878 /* re-acquire the lock before getting the next slave */
1879 write_lock_bh(&bond->lock);
1882 /* zero the mac address of the master so it will be
1883 * set by the application to the mac address of the
1884 * first slave
1886 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1888 if (list_empty(&bond->vlan_list)) {
1889 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1890 } else {
1891 printk(KERN_WARNING DRV_NAME
1892 ": %s: Warning: clearing HW address of %s while it "
1893 "still has VLANs.\n",
1894 bond_dev->name, bond_dev->name);
1895 printk(KERN_WARNING DRV_NAME
1896 ": %s: When re-adding slaves, make sure the bond's "
1897 "HW address matches its VLANs'.\n",
1898 bond_dev->name);
1901 printk(KERN_INFO DRV_NAME
1902 ": %s: released all slaves\n",
1903 bond_dev->name);
1905 out:
1906 write_unlock_bh(&bond->lock);
1908 return 0;
1912 * This function changes the active slave to slave <slave_dev>.
1913 * It returns -EINVAL in the following cases.
1914 * - <slave_dev> is not found in the list.
1915 * - There is not active slave now.
1916 * - <slave_dev> is already active.
1917 * - The link state of <slave_dev> is not BOND_LINK_UP.
1918 * - <slave_dev> is not running.
1919 * In these cases, this fuction does nothing.
1920 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1922 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1924 struct bonding *bond = bond_dev->priv;
1925 struct slave *old_active = NULL;
1926 struct slave *new_active = NULL;
1927 int res = 0;
1929 if (!USES_PRIMARY(bond->params.mode)) {
1930 return -EINVAL;
1933 /* Verify that master_dev is indeed the master of slave_dev */
1934 if (!(slave_dev->flags & IFF_SLAVE) ||
1935 (slave_dev->master != bond_dev)) {
1936 return -EINVAL;
1939 write_lock_bh(&bond->lock);
1941 old_active = bond->curr_active_slave;
1942 new_active = bond_get_slave_by_dev(bond, slave_dev);
1945 * Changing to the current active: do nothing; return success.
1947 if (new_active && (new_active == old_active)) {
1948 write_unlock_bh(&bond->lock);
1949 return 0;
1952 if ((new_active) &&
1953 (old_active) &&
1954 (new_active->link == BOND_LINK_UP) &&
1955 IS_UP(new_active->dev)) {
1956 bond_change_active_slave(bond, new_active);
1957 } else {
1958 res = -EINVAL;
1961 write_unlock_bh(&bond->lock);
1963 return res;
1966 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1968 struct bonding *bond = bond_dev->priv;
1970 info->bond_mode = bond->params.mode;
1971 info->miimon = bond->params.miimon;
1973 read_lock_bh(&bond->lock);
1974 info->num_slaves = bond->slave_cnt;
1975 read_unlock_bh(&bond->lock);
1977 return 0;
1980 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1982 struct bonding *bond = bond_dev->priv;
1983 struct slave *slave;
1984 int i, found = 0;
1986 if (info->slave_id < 0) {
1987 return -ENODEV;
1990 read_lock_bh(&bond->lock);
1992 bond_for_each_slave(bond, slave, i) {
1993 if (i == (int)info->slave_id) {
1994 found = 1;
1995 break;
1999 read_unlock_bh(&bond->lock);
2001 if (found) {
2002 strcpy(info->slave_name, slave->dev->name);
2003 info->link = slave->link;
2004 info->state = slave->state;
2005 info->link_failure_count = slave->link_failure_count;
2006 } else {
2007 return -ENODEV;
2010 return 0;
2013 /*-------------------------------- Monitoring -------------------------------*/
2015 /* this function is called regularly to monitor each slave's link. */
2016 void bond_mii_monitor(struct net_device *bond_dev)
2018 struct bonding *bond = bond_dev->priv;
2019 struct slave *slave, *oldcurrent;
2020 int do_failover = 0;
2021 int delta_in_ticks;
2022 int i;
2024 read_lock(&bond->lock);
2026 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2028 if (bond->kill_timers) {
2029 goto out;
2032 if (bond->slave_cnt == 0) {
2033 goto re_arm;
2036 /* we will try to read the link status of each of our slaves, and
2037 * set their IFF_RUNNING flag appropriately. For each slave not
2038 * supporting MII status, we won't do anything so that a user-space
2039 * program could monitor the link itself if needed.
2042 read_lock(&bond->curr_slave_lock);
2043 oldcurrent = bond->curr_active_slave;
2044 read_unlock(&bond->curr_slave_lock);
2046 bond_for_each_slave(bond, slave, i) {
2047 struct net_device *slave_dev = slave->dev;
2048 int link_state;
2049 u16 old_speed = slave->speed;
2050 u8 old_duplex = slave->duplex;
2052 link_state = bond_check_dev_link(bond, slave_dev, 0);
2054 switch (slave->link) {
2055 case BOND_LINK_UP: /* the link was up */
2056 if (link_state == BMSR_LSTATUS) {
2057 /* link stays up, nothing more to do */
2058 break;
2059 } else { /* link going down */
2060 slave->link = BOND_LINK_FAIL;
2061 slave->delay = bond->params.downdelay;
2063 if (slave->link_failure_count < UINT_MAX) {
2064 slave->link_failure_count++;
2067 if (bond->params.downdelay) {
2068 printk(KERN_INFO DRV_NAME
2069 ": %s: link status down for %s "
2070 "interface %s, disabling it in "
2071 "%d ms.\n",
2072 bond_dev->name,
2073 IS_UP(slave_dev)
2074 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2075 ? ((slave == oldcurrent)
2076 ? "active " : "backup ")
2077 : "")
2078 : "idle ",
2079 slave_dev->name,
2080 bond->params.downdelay * bond->params.miimon);
2083 /* no break ! fall through the BOND_LINK_FAIL test to
2084 ensure proper action to be taken
2086 case BOND_LINK_FAIL: /* the link has just gone down */
2087 if (link_state != BMSR_LSTATUS) {
2088 /* link stays down */
2089 if (slave->delay <= 0) {
2090 /* link down for too long time */
2091 slave->link = BOND_LINK_DOWN;
2093 /* in active/backup mode, we must
2094 * completely disable this interface
2096 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2097 (bond->params.mode == BOND_MODE_8023AD)) {
2098 bond_set_slave_inactive_flags(slave);
2101 printk(KERN_INFO DRV_NAME
2102 ": %s: link status definitely "
2103 "down for interface %s, "
2104 "disabling it\n",
2105 bond_dev->name,
2106 slave_dev->name);
2108 /* notify ad that the link status has changed */
2109 if (bond->params.mode == BOND_MODE_8023AD) {
2110 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2113 if ((bond->params.mode == BOND_MODE_TLB) ||
2114 (bond->params.mode == BOND_MODE_ALB)) {
2115 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2118 if (slave == oldcurrent) {
2119 do_failover = 1;
2121 } else {
2122 slave->delay--;
2124 } else {
2125 /* link up again */
2126 slave->link = BOND_LINK_UP;
2127 slave->jiffies = jiffies;
2128 printk(KERN_INFO DRV_NAME
2129 ": %s: link status up again after %d "
2130 "ms for interface %s.\n",
2131 bond_dev->name,
2132 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2133 slave_dev->name);
2135 break;
2136 case BOND_LINK_DOWN: /* the link was down */
2137 if (link_state != BMSR_LSTATUS) {
2138 /* the link stays down, nothing more to do */
2139 break;
2140 } else { /* link going up */
2141 slave->link = BOND_LINK_BACK;
2142 slave->delay = bond->params.updelay;
2144 if (bond->params.updelay) {
2145 /* if updelay == 0, no need to
2146 advertise about a 0 ms delay */
2147 printk(KERN_INFO DRV_NAME
2148 ": %s: link status up for "
2149 "interface %s, enabling it "
2150 "in %d ms.\n",
2151 bond_dev->name,
2152 slave_dev->name,
2153 bond->params.updelay * bond->params.miimon);
2156 /* no break ! fall through the BOND_LINK_BACK state in
2157 case there's something to do.
2159 case BOND_LINK_BACK: /* the link has just come back */
2160 if (link_state != BMSR_LSTATUS) {
2161 /* link down again */
2162 slave->link = BOND_LINK_DOWN;
2164 printk(KERN_INFO DRV_NAME
2165 ": %s: link status down again after %d "
2166 "ms for interface %s.\n",
2167 bond_dev->name,
2168 (bond->params.updelay - slave->delay) * bond->params.miimon,
2169 slave_dev->name);
2170 } else {
2171 /* link stays up */
2172 if (slave->delay == 0) {
2173 /* now the link has been up for long time enough */
2174 slave->link = BOND_LINK_UP;
2175 slave->jiffies = jiffies;
2177 if (bond->params.mode == BOND_MODE_8023AD) {
2178 /* prevent it from being the active one */
2179 slave->state = BOND_STATE_BACKUP;
2180 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2181 /* make it immediately active */
2182 slave->state = BOND_STATE_ACTIVE;
2183 } else if (slave != bond->primary_slave) {
2184 /* prevent it from being the active one */
2185 slave->state = BOND_STATE_BACKUP;
2188 printk(KERN_INFO DRV_NAME
2189 ": %s: link status definitely "
2190 "up for interface %s.\n",
2191 bond_dev->name,
2192 slave_dev->name);
2194 /* notify ad that the link status has changed */
2195 if (bond->params.mode == BOND_MODE_8023AD) {
2196 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2199 if ((bond->params.mode == BOND_MODE_TLB) ||
2200 (bond->params.mode == BOND_MODE_ALB)) {
2201 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2204 if ((!oldcurrent) ||
2205 (slave == bond->primary_slave)) {
2206 do_failover = 1;
2208 } else {
2209 slave->delay--;
2212 break;
2213 default:
2214 /* Should not happen */
2215 printk(KERN_ERR DRV_NAME
2216 ": %s: Error: %s Illegal value (link=%d)\n",
2217 bond_dev->name,
2218 slave->dev->name,
2219 slave->link);
2220 goto out;
2221 } /* end of switch (slave->link) */
2223 bond_update_speed_duplex(slave);
2225 if (bond->params.mode == BOND_MODE_8023AD) {
2226 if (old_speed != slave->speed) {
2227 bond_3ad_adapter_speed_changed(slave);
2230 if (old_duplex != slave->duplex) {
2231 bond_3ad_adapter_duplex_changed(slave);
2235 } /* end of for */
2237 if (do_failover) {
2238 write_lock(&bond->curr_slave_lock);
2240 bond_select_active_slave(bond);
2242 write_unlock(&bond->curr_slave_lock);
2243 } else
2244 bond_set_carrier(bond);
2246 re_arm:
2247 if (bond->params.miimon) {
2248 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2250 out:
2251 read_unlock(&bond->lock);
2255 static u32 bond_glean_dev_ip(struct net_device *dev)
2257 struct in_device *idev;
2258 struct in_ifaddr *ifa;
2259 __be32 addr = 0;
2261 if (!dev)
2262 return 0;
2264 rcu_read_lock();
2265 idev = __in_dev_get_rcu(dev);
2266 if (!idev)
2267 goto out;
2269 ifa = idev->ifa_list;
2270 if (!ifa)
2271 goto out;
2273 addr = ifa->ifa_local;
2274 out:
2275 rcu_read_unlock();
2276 return addr;
2279 static int bond_has_ip(struct bonding *bond)
2281 struct vlan_entry *vlan, *vlan_next;
2283 if (bond->master_ip)
2284 return 1;
2286 if (list_empty(&bond->vlan_list))
2287 return 0;
2289 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2290 vlan_list) {
2291 if (vlan->vlan_ip)
2292 return 1;
2295 return 0;
2298 static int bond_has_this_ip(struct bonding *bond, u32 ip)
2300 struct vlan_entry *vlan, *vlan_next;
2302 if (ip == bond->master_ip)
2303 return 1;
2305 if (list_empty(&bond->vlan_list))
2306 return 0;
2308 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2309 vlan_list) {
2310 if (ip == vlan->vlan_ip)
2311 return 1;
2314 return 0;
2318 * We go to the (large) trouble of VLAN tagging ARP frames because
2319 * switches in VLAN mode (especially if ports are configured as
2320 * "native" to a VLAN) might not pass non-tagged frames.
2322 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2324 struct sk_buff *skb;
2326 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2327 slave_dev->name, dest_ip, src_ip, vlan_id);
2329 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2330 NULL, slave_dev->dev_addr, NULL);
2332 if (!skb) {
2333 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2334 return;
2336 if (vlan_id) {
2337 skb = vlan_put_tag(skb, vlan_id);
2338 if (!skb) {
2339 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2340 return;
2343 arp_xmit(skb);
2347 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2349 int i, vlan_id, rv;
2350 u32 *targets = bond->params.arp_targets;
2351 struct vlan_entry *vlan, *vlan_next;
2352 struct net_device *vlan_dev;
2353 struct flowi fl;
2354 struct rtable *rt;
2356 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2357 if (!targets[i])
2358 continue;
2359 dprintk("basa: target %x\n", targets[i]);
2360 if (list_empty(&bond->vlan_list)) {
2361 dprintk("basa: empty vlan: arp_send\n");
2362 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2363 bond->master_ip, 0);
2364 continue;
2368 * If VLANs are configured, we do a route lookup to
2369 * determine which VLAN interface would be used, so we
2370 * can tag the ARP with the proper VLAN tag.
2372 memset(&fl, 0, sizeof(fl));
2373 fl.fl4_dst = targets[i];
2374 fl.fl4_tos = RTO_ONLINK;
2376 rv = ip_route_output_key(&rt, &fl);
2377 if (rv) {
2378 if (net_ratelimit()) {
2379 printk(KERN_WARNING DRV_NAME
2380 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2381 bond->dev->name, NIPQUAD(fl.fl4_dst));
2383 continue;
2387 * This target is not on a VLAN
2389 if (rt->u.dst.dev == bond->dev) {
2390 ip_rt_put(rt);
2391 dprintk("basa: rtdev == bond->dev: arp_send\n");
2392 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2393 bond->master_ip, 0);
2394 continue;
2397 vlan_id = 0;
2398 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2399 vlan_list) {
2400 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2401 if (vlan_dev == rt->u.dst.dev) {
2402 vlan_id = vlan->vlan_id;
2403 dprintk("basa: vlan match on %s %d\n",
2404 vlan_dev->name, vlan_id);
2405 break;
2409 if (vlan_id) {
2410 ip_rt_put(rt);
2411 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2412 vlan->vlan_ip, vlan_id);
2413 continue;
2416 if (net_ratelimit()) {
2417 printk(KERN_WARNING DRV_NAME
2418 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2419 bond->dev->name, NIPQUAD(fl.fl4_dst),
2420 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2422 ip_rt_put(rt);
2427 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2428 * for each VLAN above us.
2430 static void bond_send_gratuitous_arp(struct bonding *bond)
2432 struct slave *slave = bond->curr_active_slave;
2433 struct vlan_entry *vlan;
2434 struct net_device *vlan_dev;
2436 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2437 slave ? slave->dev->name : "NULL");
2438 if (!slave)
2439 return;
2441 if (bond->master_ip) {
2442 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2443 bond->master_ip, 0);
2446 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2447 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2448 if (vlan->vlan_ip) {
2449 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2450 vlan->vlan_ip, vlan->vlan_id);
2455 static void bond_validate_arp(struct bonding *bond, struct slave *slave, u32 sip, u32 tip)
2457 int i;
2458 u32 *targets = bond->params.arp_targets;
2460 targets = bond->params.arp_targets;
2461 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2462 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2463 "%u.%u.%u.%u bhti(tip) %d\n",
2464 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2465 bond_has_this_ip(bond, tip));
2466 if (sip == targets[i]) {
2467 if (bond_has_this_ip(bond, tip))
2468 slave->last_arp_rx = jiffies;
2469 return;
2474 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2476 struct arphdr *arp;
2477 struct slave *slave;
2478 struct bonding *bond;
2479 unsigned char *arp_ptr;
2480 u32 sip, tip;
2482 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2483 goto out;
2485 bond = dev->priv;
2486 read_lock(&bond->lock);
2488 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2489 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2490 orig_dev ? orig_dev->name : "NULL");
2492 slave = bond_get_slave_by_dev(bond, orig_dev);
2493 if (!slave || !slave_do_arp_validate(bond, slave))
2494 goto out_unlock;
2496 /* ARP header, plus 2 device addresses, plus 2 IP addresses. */
2497 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2498 (2 * dev->addr_len) +
2499 (2 * sizeof(u32)))))
2500 goto out_unlock;
2502 arp = skb->nh.arph;
2503 if (arp->ar_hln != dev->addr_len ||
2504 skb->pkt_type == PACKET_OTHERHOST ||
2505 skb->pkt_type == PACKET_LOOPBACK ||
2506 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2507 arp->ar_pro != htons(ETH_P_IP) ||
2508 arp->ar_pln != 4)
2509 goto out_unlock;
2511 arp_ptr = (unsigned char *)(arp + 1);
2512 arp_ptr += dev->addr_len;
2513 memcpy(&sip, arp_ptr, 4);
2514 arp_ptr += 4 + dev->addr_len;
2515 memcpy(&tip, arp_ptr, 4);
2517 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2518 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2519 slave->state, bond->params.arp_validate,
2520 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2523 * Backup slaves won't see the ARP reply, but do come through
2524 * here for each ARP probe (so we swap the sip/tip to validate
2525 * the probe). In a "redundant switch, common router" type of
2526 * configuration, the ARP probe will (hopefully) travel from
2527 * the active, through one switch, the router, then the other
2528 * switch before reaching the backup.
2530 if (slave->state == BOND_STATE_ACTIVE)
2531 bond_validate_arp(bond, slave, sip, tip);
2532 else
2533 bond_validate_arp(bond, slave, tip, sip);
2535 out_unlock:
2536 read_unlock(&bond->lock);
2537 out:
2538 dev_kfree_skb(skb);
2539 return NET_RX_SUCCESS;
2543 * this function is called regularly to monitor each slave's link
2544 * ensuring that traffic is being sent and received when arp monitoring
2545 * is used in load-balancing mode. if the adapter has been dormant, then an
2546 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2547 * arp monitoring in active backup mode.
2549 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2551 struct bonding *bond = bond_dev->priv;
2552 struct slave *slave, *oldcurrent;
2553 int do_failover = 0;
2554 int delta_in_ticks;
2555 int i;
2557 read_lock(&bond->lock);
2559 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2561 if (bond->kill_timers) {
2562 goto out;
2565 if (bond->slave_cnt == 0) {
2566 goto re_arm;
2569 read_lock(&bond->curr_slave_lock);
2570 oldcurrent = bond->curr_active_slave;
2571 read_unlock(&bond->curr_slave_lock);
2573 /* see if any of the previous devices are up now (i.e. they have
2574 * xmt and rcv traffic). the curr_active_slave does not come into
2575 * the picture unless it is null. also, slave->jiffies is not needed
2576 * here because we send an arp on each slave and give a slave as
2577 * long as it needs to get the tx/rx within the delta.
2578 * TODO: what about up/down delay in arp mode? it wasn't here before
2579 * so it can wait
2581 bond_for_each_slave(bond, slave, i) {
2582 if (slave->link != BOND_LINK_UP) {
2583 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2584 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2586 slave->link = BOND_LINK_UP;
2587 slave->state = BOND_STATE_ACTIVE;
2589 /* primary_slave has no meaning in round-robin
2590 * mode. the window of a slave being up and
2591 * curr_active_slave being null after enslaving
2592 * is closed.
2594 if (!oldcurrent) {
2595 printk(KERN_INFO DRV_NAME
2596 ": %s: link status definitely "
2597 "up for interface %s, ",
2598 bond_dev->name,
2599 slave->dev->name);
2600 do_failover = 1;
2601 } else {
2602 printk(KERN_INFO DRV_NAME
2603 ": %s: interface %s is now up\n",
2604 bond_dev->name,
2605 slave->dev->name);
2608 } else {
2609 /* slave->link == BOND_LINK_UP */
2611 /* not all switches will respond to an arp request
2612 * when the source ip is 0, so don't take the link down
2613 * if we don't know our ip yet
2615 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2616 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2617 bond_has_ip(bond))) {
2619 slave->link = BOND_LINK_DOWN;
2620 slave->state = BOND_STATE_BACKUP;
2622 if (slave->link_failure_count < UINT_MAX) {
2623 slave->link_failure_count++;
2626 printk(KERN_INFO DRV_NAME
2627 ": %s: interface %s is now down.\n",
2628 bond_dev->name,
2629 slave->dev->name);
2631 if (slave == oldcurrent) {
2632 do_failover = 1;
2637 /* note: if switch is in round-robin mode, all links
2638 * must tx arp to ensure all links rx an arp - otherwise
2639 * links may oscillate or not come up at all; if switch is
2640 * in something like xor mode, there is nothing we can
2641 * do - all replies will be rx'ed on same link causing slaves
2642 * to be unstable during low/no traffic periods
2644 if (IS_UP(slave->dev)) {
2645 bond_arp_send_all(bond, slave);
2649 if (do_failover) {
2650 write_lock(&bond->curr_slave_lock);
2652 bond_select_active_slave(bond);
2654 write_unlock(&bond->curr_slave_lock);
2657 re_arm:
2658 if (bond->params.arp_interval) {
2659 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2661 out:
2662 read_unlock(&bond->lock);
2666 * When using arp monitoring in active-backup mode, this function is
2667 * called to determine if any backup slaves have went down or a new
2668 * current slave needs to be found.
2669 * The backup slaves never generate traffic, they are considered up by merely
2670 * receiving traffic. If the current slave goes down, each backup slave will
2671 * be given the opportunity to tx/rx an arp before being taken down - this
2672 * prevents all slaves from being taken down due to the current slave not
2673 * sending any traffic for the backups to receive. The arps are not necessarily
2674 * necessary, any tx and rx traffic will keep the current slave up. While any
2675 * rx traffic will keep the backup slaves up, the current slave is responsible
2676 * for generating traffic to keep them up regardless of any other traffic they
2677 * may have received.
2678 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2680 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2682 struct bonding *bond = bond_dev->priv;
2683 struct slave *slave;
2684 int delta_in_ticks;
2685 int i;
2687 read_lock(&bond->lock);
2689 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2691 if (bond->kill_timers) {
2692 goto out;
2695 if (bond->slave_cnt == 0) {
2696 goto re_arm;
2699 /* determine if any slave has come up or any backup slave has
2700 * gone down
2701 * TODO: what about up/down delay in arp mode? it wasn't here before
2702 * so it can wait
2704 bond_for_each_slave(bond, slave, i) {
2705 if (slave->link != BOND_LINK_UP) {
2706 if ((jiffies - slave_last_rx(bond, slave)) <=
2707 delta_in_ticks) {
2709 slave->link = BOND_LINK_UP;
2711 write_lock(&bond->curr_slave_lock);
2713 if ((!bond->curr_active_slave) &&
2714 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2715 bond_change_active_slave(bond, slave);
2716 bond->current_arp_slave = NULL;
2717 } else if (bond->curr_active_slave != slave) {
2718 /* this slave has just come up but we
2719 * already have a current slave; this
2720 * can also happen if bond_enslave adds
2721 * a new slave that is up while we are
2722 * searching for a new slave
2724 bond_set_slave_inactive_flags(slave);
2725 bond->current_arp_slave = NULL;
2728 bond_set_carrier(bond);
2730 if (slave == bond->curr_active_slave) {
2731 printk(KERN_INFO DRV_NAME
2732 ": %s: %s is up and now the "
2733 "active interface\n",
2734 bond_dev->name,
2735 slave->dev->name);
2736 netif_carrier_on(bond->dev);
2737 } else {
2738 printk(KERN_INFO DRV_NAME
2739 ": %s: backup interface %s is "
2740 "now up\n",
2741 bond_dev->name,
2742 slave->dev->name);
2745 write_unlock(&bond->curr_slave_lock);
2747 } else {
2748 read_lock(&bond->curr_slave_lock);
2750 if ((slave != bond->curr_active_slave) &&
2751 (!bond->current_arp_slave) &&
2752 (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2753 bond_has_ip(bond))) {
2754 /* a backup slave has gone down; three times
2755 * the delta allows the current slave to be
2756 * taken out before the backup slave.
2757 * note: a non-null current_arp_slave indicates
2758 * the curr_active_slave went down and we are
2759 * searching for a new one; under this
2760 * condition we only take the curr_active_slave
2761 * down - this gives each slave a chance to
2762 * tx/rx traffic before being taken out
2765 read_unlock(&bond->curr_slave_lock);
2767 slave->link = BOND_LINK_DOWN;
2769 if (slave->link_failure_count < UINT_MAX) {
2770 slave->link_failure_count++;
2773 bond_set_slave_inactive_flags(slave);
2775 printk(KERN_INFO DRV_NAME
2776 ": %s: backup interface %s is now down\n",
2777 bond_dev->name,
2778 slave->dev->name);
2779 } else {
2780 read_unlock(&bond->curr_slave_lock);
2785 read_lock(&bond->curr_slave_lock);
2786 slave = bond->curr_active_slave;
2787 read_unlock(&bond->curr_slave_lock);
2789 if (slave) {
2790 /* if we have sent traffic in the past 2*arp_intervals but
2791 * haven't xmit and rx traffic in that time interval, select
2792 * a different slave. slave->jiffies is only updated when
2793 * a slave first becomes the curr_active_slave - not necessarily
2794 * after every arp; this ensures the slave has a full 2*delta
2795 * before being taken out. if a primary is being used, check
2796 * if it is up and needs to take over as the curr_active_slave
2798 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2799 (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2800 bond_has_ip(bond))) &&
2801 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2803 slave->link = BOND_LINK_DOWN;
2805 if (slave->link_failure_count < UINT_MAX) {
2806 slave->link_failure_count++;
2809 printk(KERN_INFO DRV_NAME
2810 ": %s: link status down for active interface "
2811 "%s, disabling it\n",
2812 bond_dev->name,
2813 slave->dev->name);
2815 write_lock(&bond->curr_slave_lock);
2817 bond_select_active_slave(bond);
2818 slave = bond->curr_active_slave;
2820 write_unlock(&bond->curr_slave_lock);
2822 bond->current_arp_slave = slave;
2824 if (slave) {
2825 slave->jiffies = jiffies;
2827 } else if ((bond->primary_slave) &&
2828 (bond->primary_slave != slave) &&
2829 (bond->primary_slave->link == BOND_LINK_UP)) {
2830 /* at this point, slave is the curr_active_slave */
2831 printk(KERN_INFO DRV_NAME
2832 ": %s: changing from interface %s to primary "
2833 "interface %s\n",
2834 bond_dev->name,
2835 slave->dev->name,
2836 bond->primary_slave->dev->name);
2838 /* primary is up so switch to it */
2839 write_lock(&bond->curr_slave_lock);
2840 bond_change_active_slave(bond, bond->primary_slave);
2841 write_unlock(&bond->curr_slave_lock);
2843 slave = bond->primary_slave;
2844 slave->jiffies = jiffies;
2845 } else {
2846 bond->current_arp_slave = NULL;
2849 /* the current slave must tx an arp to ensure backup slaves
2850 * rx traffic
2852 if (slave && bond_has_ip(bond)) {
2853 bond_arp_send_all(bond, slave);
2857 /* if we don't have a curr_active_slave, search for the next available
2858 * backup slave from the current_arp_slave and make it the candidate
2859 * for becoming the curr_active_slave
2861 if (!slave) {
2862 if (!bond->current_arp_slave) {
2863 bond->current_arp_slave = bond->first_slave;
2866 if (bond->current_arp_slave) {
2867 bond_set_slave_inactive_flags(bond->current_arp_slave);
2869 /* search for next candidate */
2870 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2871 if (IS_UP(slave->dev)) {
2872 slave->link = BOND_LINK_BACK;
2873 bond_set_slave_active_flags(slave);
2874 bond_arp_send_all(bond, slave);
2875 slave->jiffies = jiffies;
2876 bond->current_arp_slave = slave;
2877 break;
2880 /* if the link state is up at this point, we
2881 * mark it down - this can happen if we have
2882 * simultaneous link failures and
2883 * reselect_active_interface doesn't make this
2884 * one the current slave so it is still marked
2885 * up when it is actually down
2887 if (slave->link == BOND_LINK_UP) {
2888 slave->link = BOND_LINK_DOWN;
2889 if (slave->link_failure_count < UINT_MAX) {
2890 slave->link_failure_count++;
2893 bond_set_slave_inactive_flags(slave);
2895 printk(KERN_INFO DRV_NAME
2896 ": %s: backup interface %s is "
2897 "now down.\n",
2898 bond_dev->name,
2899 slave->dev->name);
2905 re_arm:
2906 if (bond->params.arp_interval) {
2907 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2909 out:
2910 read_unlock(&bond->lock);
2913 /*------------------------------ proc/seq_file-------------------------------*/
2915 #ifdef CONFIG_PROC_FS
2917 #define SEQ_START_TOKEN ((void *)1)
2919 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2921 struct bonding *bond = seq->private;
2922 loff_t off = 0;
2923 struct slave *slave;
2924 int i;
2926 /* make sure the bond won't be taken away */
2927 read_lock(&dev_base_lock);
2928 read_lock_bh(&bond->lock);
2930 if (*pos == 0) {
2931 return SEQ_START_TOKEN;
2934 bond_for_each_slave(bond, slave, i) {
2935 if (++off == *pos) {
2936 return slave;
2940 return NULL;
2943 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2945 struct bonding *bond = seq->private;
2946 struct slave *slave = v;
2948 ++*pos;
2949 if (v == SEQ_START_TOKEN) {
2950 return bond->first_slave;
2953 slave = slave->next;
2955 return (slave == bond->first_slave) ? NULL : slave;
2958 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2960 struct bonding *bond = seq->private;
2962 read_unlock_bh(&bond->lock);
2963 read_unlock(&dev_base_lock);
2966 static void bond_info_show_master(struct seq_file *seq)
2968 struct bonding *bond = seq->private;
2969 struct slave *curr;
2970 int i;
2971 u32 target;
2973 read_lock(&bond->curr_slave_lock);
2974 curr = bond->curr_active_slave;
2975 read_unlock(&bond->curr_slave_lock);
2977 seq_printf(seq, "Bonding Mode: %s\n",
2978 bond_mode_name(bond->params.mode));
2980 if (bond->params.mode == BOND_MODE_XOR ||
2981 bond->params.mode == BOND_MODE_8023AD) {
2982 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2983 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2984 bond->params.xmit_policy);
2987 if (USES_PRIMARY(bond->params.mode)) {
2988 seq_printf(seq, "Primary Slave: %s\n",
2989 (bond->primary_slave) ?
2990 bond->primary_slave->dev->name : "None");
2992 seq_printf(seq, "Currently Active Slave: %s\n",
2993 (curr) ? curr->dev->name : "None");
2996 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
2997 "up" : "down");
2998 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
2999 seq_printf(seq, "Up Delay (ms): %d\n",
3000 bond->params.updelay * bond->params.miimon);
3001 seq_printf(seq, "Down Delay (ms): %d\n",
3002 bond->params.downdelay * bond->params.miimon);
3005 /* ARP information */
3006 if(bond->params.arp_interval > 0) {
3007 int printed=0;
3008 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3009 bond->params.arp_interval);
3011 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3013 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3014 if (!bond->params.arp_targets[i])
3015 continue;
3016 if (printed)
3017 seq_printf(seq, ",");
3018 target = ntohl(bond->params.arp_targets[i]);
3019 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3020 printed = 1;
3022 seq_printf(seq, "\n");
3025 if (bond->params.mode == BOND_MODE_8023AD) {
3026 struct ad_info ad_info;
3028 seq_puts(seq, "\n802.3ad info\n");
3029 seq_printf(seq, "LACP rate: %s\n",
3030 (bond->params.lacp_fast) ? "fast" : "slow");
3032 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3033 seq_printf(seq, "bond %s has no active aggregator\n",
3034 bond->dev->name);
3035 } else {
3036 seq_printf(seq, "Active Aggregator Info:\n");
3038 seq_printf(seq, "\tAggregator ID: %d\n",
3039 ad_info.aggregator_id);
3040 seq_printf(seq, "\tNumber of ports: %d\n",
3041 ad_info.ports);
3042 seq_printf(seq, "\tActor Key: %d\n",
3043 ad_info.actor_key);
3044 seq_printf(seq, "\tPartner Key: %d\n",
3045 ad_info.partner_key);
3046 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3047 ad_info.partner_system[0],
3048 ad_info.partner_system[1],
3049 ad_info.partner_system[2],
3050 ad_info.partner_system[3],
3051 ad_info.partner_system[4],
3052 ad_info.partner_system[5]);
3057 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3059 struct bonding *bond = seq->private;
3061 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3062 seq_printf(seq, "MII Status: %s\n",
3063 (slave->link == BOND_LINK_UP) ? "up" : "down");
3064 seq_printf(seq, "Link Failure Count: %u\n",
3065 slave->link_failure_count);
3067 seq_printf(seq,
3068 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3069 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3070 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3071 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
3073 if (bond->params.mode == BOND_MODE_8023AD) {
3074 const struct aggregator *agg
3075 = SLAVE_AD_INFO(slave).port.aggregator;
3077 if (agg) {
3078 seq_printf(seq, "Aggregator ID: %d\n",
3079 agg->aggregator_identifier);
3080 } else {
3081 seq_puts(seq, "Aggregator ID: N/A\n");
3086 static int bond_info_seq_show(struct seq_file *seq, void *v)
3088 if (v == SEQ_START_TOKEN) {
3089 seq_printf(seq, "%s\n", version);
3090 bond_info_show_master(seq);
3091 } else {
3092 bond_info_show_slave(seq, v);
3095 return 0;
3098 static struct seq_operations bond_info_seq_ops = {
3099 .start = bond_info_seq_start,
3100 .next = bond_info_seq_next,
3101 .stop = bond_info_seq_stop,
3102 .show = bond_info_seq_show,
3105 static int bond_info_open(struct inode *inode, struct file *file)
3107 struct seq_file *seq;
3108 struct proc_dir_entry *proc;
3109 int res;
3111 res = seq_open(file, &bond_info_seq_ops);
3112 if (!res) {
3113 /* recover the pointer buried in proc_dir_entry data */
3114 seq = file->private_data;
3115 proc = PDE(inode);
3116 seq->private = proc->data;
3119 return res;
3122 static const struct file_operations bond_info_fops = {
3123 .owner = THIS_MODULE,
3124 .open = bond_info_open,
3125 .read = seq_read,
3126 .llseek = seq_lseek,
3127 .release = seq_release,
3130 static int bond_create_proc_entry(struct bonding *bond)
3132 struct net_device *bond_dev = bond->dev;
3134 if (bond_proc_dir) {
3135 bond->proc_entry = create_proc_entry(bond_dev->name,
3136 S_IRUGO,
3137 bond_proc_dir);
3138 if (bond->proc_entry == NULL) {
3139 printk(KERN_WARNING DRV_NAME
3140 ": Warning: Cannot create /proc/net/%s/%s\n",
3141 DRV_NAME, bond_dev->name);
3142 } else {
3143 bond->proc_entry->data = bond;
3144 bond->proc_entry->proc_fops = &bond_info_fops;
3145 bond->proc_entry->owner = THIS_MODULE;
3146 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3150 return 0;
3153 static void bond_remove_proc_entry(struct bonding *bond)
3155 if (bond_proc_dir && bond->proc_entry) {
3156 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3157 memset(bond->proc_file_name, 0, IFNAMSIZ);
3158 bond->proc_entry = NULL;
3162 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3163 * Caller must hold rtnl_lock.
3165 static void bond_create_proc_dir(void)
3167 int len = strlen(DRV_NAME);
3169 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3170 bond_proc_dir = bond_proc_dir->next) {
3171 if ((bond_proc_dir->namelen == len) &&
3172 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3173 break;
3177 if (!bond_proc_dir) {
3178 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3179 if (bond_proc_dir) {
3180 bond_proc_dir->owner = THIS_MODULE;
3181 } else {
3182 printk(KERN_WARNING DRV_NAME
3183 ": Warning: cannot create /proc/net/%s\n",
3184 DRV_NAME);
3189 /* Destroy the bonding directory under /proc/net, if empty.
3190 * Caller must hold rtnl_lock.
3192 static void bond_destroy_proc_dir(void)
3194 struct proc_dir_entry *de;
3196 if (!bond_proc_dir) {
3197 return;
3200 /* verify that the /proc dir is empty */
3201 for (de = bond_proc_dir->subdir; de; de = de->next) {
3202 /* ignore . and .. */
3203 if (*(de->name) != '.') {
3204 break;
3208 if (de) {
3209 if (bond_proc_dir->owner == THIS_MODULE) {
3210 bond_proc_dir->owner = NULL;
3212 } else {
3213 remove_proc_entry(DRV_NAME, proc_net);
3214 bond_proc_dir = NULL;
3217 #endif /* CONFIG_PROC_FS */
3219 /*-------------------------- netdev event handling --------------------------*/
3222 * Change device name
3224 static int bond_event_changename(struct bonding *bond)
3226 #ifdef CONFIG_PROC_FS
3227 bond_remove_proc_entry(bond);
3228 bond_create_proc_entry(bond);
3229 #endif
3230 down_write(&(bonding_rwsem));
3231 bond_destroy_sysfs_entry(bond);
3232 bond_create_sysfs_entry(bond);
3233 up_write(&(bonding_rwsem));
3234 return NOTIFY_DONE;
3237 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3239 struct bonding *event_bond = bond_dev->priv;
3241 switch (event) {
3242 case NETDEV_CHANGENAME:
3243 return bond_event_changename(event_bond);
3244 case NETDEV_UNREGISTER:
3246 * TODO: remove a bond from the list?
3248 break;
3249 default:
3250 break;
3253 return NOTIFY_DONE;
3256 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3258 struct net_device *bond_dev = slave_dev->master;
3259 struct bonding *bond = bond_dev->priv;
3261 switch (event) {
3262 case NETDEV_UNREGISTER:
3263 if (bond_dev) {
3264 bond_release(bond_dev, slave_dev);
3266 break;
3267 case NETDEV_CHANGE:
3269 * TODO: is this what we get if somebody
3270 * sets up a hierarchical bond, then rmmod's
3271 * one of the slave bonding devices?
3273 break;
3274 case NETDEV_DOWN:
3276 * ... Or is it this?
3278 break;
3279 case NETDEV_CHANGEMTU:
3281 * TODO: Should slaves be allowed to
3282 * independently alter their MTU? For
3283 * an active-backup bond, slaves need
3284 * not be the same type of device, so
3285 * MTUs may vary. For other modes,
3286 * slaves arguably should have the
3287 * same MTUs. To do this, we'd need to
3288 * take over the slave's change_mtu
3289 * function for the duration of their
3290 * servitude.
3292 break;
3293 case NETDEV_CHANGENAME:
3295 * TODO: handle changing the primary's name
3297 break;
3298 case NETDEV_FEAT_CHANGE:
3299 bond_compute_features(bond);
3300 break;
3301 default:
3302 break;
3305 return NOTIFY_DONE;
3309 * bond_netdev_event: handle netdev notifier chain events.
3311 * This function receives events for the netdev chain. The caller (an
3312 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3313 * locks for us to safely manipulate the slave devices (RTNL lock,
3314 * dev_probe_lock).
3316 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3318 struct net_device *event_dev = (struct net_device *)ptr;
3320 dprintk("event_dev: %s, event: %lx\n",
3321 (event_dev ? event_dev->name : "None"),
3322 event);
3324 if (!(event_dev->priv_flags & IFF_BONDING))
3325 return NOTIFY_DONE;
3327 if (event_dev->flags & IFF_MASTER) {
3328 dprintk("IFF_MASTER\n");
3329 return bond_master_netdev_event(event, event_dev);
3332 if (event_dev->flags & IFF_SLAVE) {
3333 dprintk("IFF_SLAVE\n");
3334 return bond_slave_netdev_event(event, event_dev);
3337 return NOTIFY_DONE;
3341 * bond_inetaddr_event: handle inetaddr notifier chain events.
3343 * We keep track of device IPs primarily to use as source addresses in
3344 * ARP monitor probes (rather than spewing out broadcasts all the time).
3346 * We track one IP for the main device (if it has one), plus one per VLAN.
3348 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3350 struct in_ifaddr *ifa = ptr;
3351 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3352 struct bonding *bond, *bond_next;
3353 struct vlan_entry *vlan, *vlan_next;
3355 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3356 if (bond->dev == event_dev) {
3357 switch (event) {
3358 case NETDEV_UP:
3359 bond->master_ip = ifa->ifa_local;
3360 return NOTIFY_OK;
3361 case NETDEV_DOWN:
3362 bond->master_ip = bond_glean_dev_ip(bond->dev);
3363 return NOTIFY_OK;
3364 default:
3365 return NOTIFY_DONE;
3369 if (list_empty(&bond->vlan_list))
3370 continue;
3372 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3373 vlan_list) {
3374 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3375 if (vlan_dev == event_dev) {
3376 switch (event) {
3377 case NETDEV_UP:
3378 vlan->vlan_ip = ifa->ifa_local;
3379 return NOTIFY_OK;
3380 case NETDEV_DOWN:
3381 vlan->vlan_ip =
3382 bond_glean_dev_ip(vlan_dev);
3383 return NOTIFY_OK;
3384 default:
3385 return NOTIFY_DONE;
3390 return NOTIFY_DONE;
3393 static struct notifier_block bond_netdev_notifier = {
3394 .notifier_call = bond_netdev_event,
3397 static struct notifier_block bond_inetaddr_notifier = {
3398 .notifier_call = bond_inetaddr_event,
3401 /*-------------------------- Packet type handling ---------------------------*/
3403 /* register to receive lacpdus on a bond */
3404 static void bond_register_lacpdu(struct bonding *bond)
3406 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3408 /* initialize packet type */
3409 pk_type->type = PKT_TYPE_LACPDU;
3410 pk_type->dev = bond->dev;
3411 pk_type->func = bond_3ad_lacpdu_recv;
3413 dev_add_pack(pk_type);
3416 /* unregister to receive lacpdus on a bond */
3417 static void bond_unregister_lacpdu(struct bonding *bond)
3419 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3422 void bond_register_arp(struct bonding *bond)
3424 struct packet_type *pt = &bond->arp_mon_pt;
3426 if (pt->type)
3427 return;
3429 pt->type = htons(ETH_P_ARP);
3430 pt->dev = NULL; /*bond->dev;XXX*/
3431 pt->func = bond_arp_rcv;
3432 dev_add_pack(pt);
3435 void bond_unregister_arp(struct bonding *bond)
3437 struct packet_type *pt = &bond->arp_mon_pt;
3439 dev_remove_pack(pt);
3440 pt->type = 0;
3443 /*---------------------------- Hashing Policies -----------------------------*/
3446 * Hash for the the output device based upon layer 3 and layer 4 data. If
3447 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3448 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3450 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3451 struct net_device *bond_dev, int count)
3453 struct ethhdr *data = (struct ethhdr *)skb->data;
3454 struct iphdr *iph = skb->nh.iph;
3455 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3456 int layer4_xor = 0;
3458 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3459 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3460 (iph->protocol == IPPROTO_TCP ||
3461 iph->protocol == IPPROTO_UDP)) {
3462 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3464 return (layer4_xor ^
3465 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3469 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3473 * Hash for the output device based upon layer 2 data
3475 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3476 struct net_device *bond_dev, int count)
3478 struct ethhdr *data = (struct ethhdr *)skb->data;
3480 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3483 /*-------------------------- Device entry points ----------------------------*/
3485 static int bond_open(struct net_device *bond_dev)
3487 struct bonding *bond = bond_dev->priv;
3488 struct timer_list *mii_timer = &bond->mii_timer;
3489 struct timer_list *arp_timer = &bond->arp_timer;
3491 bond->kill_timers = 0;
3493 if ((bond->params.mode == BOND_MODE_TLB) ||
3494 (bond->params.mode == BOND_MODE_ALB)) {
3495 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3497 /* bond_alb_initialize must be called before the timer
3498 * is started.
3500 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3501 /* something went wrong - fail the open operation */
3502 return -1;
3505 init_timer(alb_timer);
3506 alb_timer->expires = jiffies + 1;
3507 alb_timer->data = (unsigned long)bond;
3508 alb_timer->function = (void *)&bond_alb_monitor;
3509 add_timer(alb_timer);
3512 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3513 init_timer(mii_timer);
3514 mii_timer->expires = jiffies + 1;
3515 mii_timer->data = (unsigned long)bond_dev;
3516 mii_timer->function = (void *)&bond_mii_monitor;
3517 add_timer(mii_timer);
3520 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3521 init_timer(arp_timer);
3522 arp_timer->expires = jiffies + 1;
3523 arp_timer->data = (unsigned long)bond_dev;
3524 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3525 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3526 } else {
3527 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3529 if (bond->params.arp_validate)
3530 bond_register_arp(bond);
3532 add_timer(arp_timer);
3535 if (bond->params.mode == BOND_MODE_8023AD) {
3536 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3537 init_timer(ad_timer);
3538 ad_timer->expires = jiffies + 1;
3539 ad_timer->data = (unsigned long)bond;
3540 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3541 add_timer(ad_timer);
3543 /* register to receive LACPDUs */
3544 bond_register_lacpdu(bond);
3547 return 0;
3550 static int bond_close(struct net_device *bond_dev)
3552 struct bonding *bond = bond_dev->priv;
3554 if (bond->params.mode == BOND_MODE_8023AD) {
3555 /* Unregister the receive of LACPDUs */
3556 bond_unregister_lacpdu(bond);
3559 if (bond->params.arp_validate)
3560 bond_unregister_arp(bond);
3562 write_lock_bh(&bond->lock);
3565 /* signal timers not to re-arm */
3566 bond->kill_timers = 1;
3568 write_unlock_bh(&bond->lock);
3570 /* del_timer_sync must run without holding the bond->lock
3571 * because a running timer might be trying to hold it too
3574 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3575 del_timer_sync(&bond->mii_timer);
3578 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3579 del_timer_sync(&bond->arp_timer);
3582 switch (bond->params.mode) {
3583 case BOND_MODE_8023AD:
3584 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3585 break;
3586 case BOND_MODE_TLB:
3587 case BOND_MODE_ALB:
3588 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3589 break;
3590 default:
3591 break;
3595 if ((bond->params.mode == BOND_MODE_TLB) ||
3596 (bond->params.mode == BOND_MODE_ALB)) {
3597 /* Must be called only after all
3598 * slaves have been released
3600 bond_alb_deinitialize(bond);
3603 return 0;
3606 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3608 struct bonding *bond = bond_dev->priv;
3609 struct net_device_stats *stats = &(bond->stats), *sstats;
3610 struct slave *slave;
3611 int i;
3613 memset(stats, 0, sizeof(struct net_device_stats));
3615 read_lock_bh(&bond->lock);
3617 bond_for_each_slave(bond, slave, i) {
3618 if (slave->dev->get_stats) {
3619 sstats = slave->dev->get_stats(slave->dev);
3621 stats->rx_packets += sstats->rx_packets;
3622 stats->rx_bytes += sstats->rx_bytes;
3623 stats->rx_errors += sstats->rx_errors;
3624 stats->rx_dropped += sstats->rx_dropped;
3626 stats->tx_packets += sstats->tx_packets;
3627 stats->tx_bytes += sstats->tx_bytes;
3628 stats->tx_errors += sstats->tx_errors;
3629 stats->tx_dropped += sstats->tx_dropped;
3631 stats->multicast += sstats->multicast;
3632 stats->collisions += sstats->collisions;
3634 stats->rx_length_errors += sstats->rx_length_errors;
3635 stats->rx_over_errors += sstats->rx_over_errors;
3636 stats->rx_crc_errors += sstats->rx_crc_errors;
3637 stats->rx_frame_errors += sstats->rx_frame_errors;
3638 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3639 stats->rx_missed_errors += sstats->rx_missed_errors;
3641 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3642 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3643 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3644 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3645 stats->tx_window_errors += sstats->tx_window_errors;
3649 read_unlock_bh(&bond->lock);
3651 return stats;
3654 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3656 struct net_device *slave_dev = NULL;
3657 struct ifbond k_binfo;
3658 struct ifbond __user *u_binfo = NULL;
3659 struct ifslave k_sinfo;
3660 struct ifslave __user *u_sinfo = NULL;
3661 struct mii_ioctl_data *mii = NULL;
3662 int res = 0;
3664 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3665 bond_dev->name, cmd);
3667 switch (cmd) {
3668 case SIOCGMIIPHY:
3669 mii = if_mii(ifr);
3670 if (!mii) {
3671 return -EINVAL;
3673 mii->phy_id = 0;
3674 /* Fall Through */
3675 case SIOCGMIIREG:
3677 * We do this again just in case we were called by SIOCGMIIREG
3678 * instead of SIOCGMIIPHY.
3680 mii = if_mii(ifr);
3681 if (!mii) {
3682 return -EINVAL;
3685 if (mii->reg_num == 1) {
3686 struct bonding *bond = bond_dev->priv;
3687 mii->val_out = 0;
3688 read_lock_bh(&bond->lock);
3689 read_lock(&bond->curr_slave_lock);
3690 if (netif_carrier_ok(bond->dev)) {
3691 mii->val_out = BMSR_LSTATUS;
3693 read_unlock(&bond->curr_slave_lock);
3694 read_unlock_bh(&bond->lock);
3697 return 0;
3698 case BOND_INFO_QUERY_OLD:
3699 case SIOCBONDINFOQUERY:
3700 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3702 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3703 return -EFAULT;
3706 res = bond_info_query(bond_dev, &k_binfo);
3707 if (res == 0) {
3708 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3709 return -EFAULT;
3713 return res;
3714 case BOND_SLAVE_INFO_QUERY_OLD:
3715 case SIOCBONDSLAVEINFOQUERY:
3716 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3718 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3719 return -EFAULT;
3722 res = bond_slave_info_query(bond_dev, &k_sinfo);
3723 if (res == 0) {
3724 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3725 return -EFAULT;
3729 return res;
3730 default:
3731 /* Go on */
3732 break;
3735 if (!capable(CAP_NET_ADMIN)) {
3736 return -EPERM;
3739 down_write(&(bonding_rwsem));
3740 slave_dev = dev_get_by_name(ifr->ifr_slave);
3742 dprintk("slave_dev=%p: \n", slave_dev);
3744 if (!slave_dev) {
3745 res = -ENODEV;
3746 } else {
3747 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3748 switch (cmd) {
3749 case BOND_ENSLAVE_OLD:
3750 case SIOCBONDENSLAVE:
3751 res = bond_enslave(bond_dev, slave_dev);
3752 break;
3753 case BOND_RELEASE_OLD:
3754 case SIOCBONDRELEASE:
3755 res = bond_release(bond_dev, slave_dev);
3756 break;
3757 case BOND_SETHWADDR_OLD:
3758 case SIOCBONDSETHWADDR:
3759 res = bond_sethwaddr(bond_dev, slave_dev);
3760 break;
3761 case BOND_CHANGE_ACTIVE_OLD:
3762 case SIOCBONDCHANGEACTIVE:
3763 res = bond_ioctl_change_active(bond_dev, slave_dev);
3764 break;
3765 default:
3766 res = -EOPNOTSUPP;
3769 dev_put(slave_dev);
3772 up_write(&(bonding_rwsem));
3773 return res;
3776 static void bond_set_multicast_list(struct net_device *bond_dev)
3778 struct bonding *bond = bond_dev->priv;
3779 struct dev_mc_list *dmi;
3781 write_lock_bh(&bond->lock);
3784 * Do promisc before checking multicast_mode
3786 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3787 bond_set_promiscuity(bond, 1);
3790 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3791 bond_set_promiscuity(bond, -1);
3794 /* set allmulti flag to slaves */
3795 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3796 bond_set_allmulti(bond, 1);
3799 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3800 bond_set_allmulti(bond, -1);
3803 bond->flags = bond_dev->flags;
3805 /* looking for addresses to add to slaves' mc list */
3806 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3807 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3808 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3812 /* looking for addresses to delete from slaves' list */
3813 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3814 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3815 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3819 /* save master's multicast list */
3820 bond_mc_list_destroy(bond);
3821 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3823 write_unlock_bh(&bond->lock);
3827 * Change the MTU of all of a master's slaves to match the master
3829 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3831 struct bonding *bond = bond_dev->priv;
3832 struct slave *slave, *stop_at;
3833 int res = 0;
3834 int i;
3836 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3837 (bond_dev ? bond_dev->name : "None"), new_mtu);
3839 /* Can't hold bond->lock with bh disabled here since
3840 * some base drivers panic. On the other hand we can't
3841 * hold bond->lock without bh disabled because we'll
3842 * deadlock. The only solution is to rely on the fact
3843 * that we're under rtnl_lock here, and the slaves
3844 * list won't change. This doesn't solve the problem
3845 * of setting the slave's MTU while it is
3846 * transmitting, but the assumption is that the base
3847 * driver can handle that.
3849 * TODO: figure out a way to safely iterate the slaves
3850 * list, but without holding a lock around the actual
3851 * call to the base driver.
3854 bond_for_each_slave(bond, slave, i) {
3855 dprintk("s %p s->p %p c_m %p\n", slave,
3856 slave->prev, slave->dev->change_mtu);
3858 res = dev_set_mtu(slave->dev, new_mtu);
3860 if (res) {
3861 /* If we failed to set the slave's mtu to the new value
3862 * we must abort the operation even in ACTIVE_BACKUP
3863 * mode, because if we allow the backup slaves to have
3864 * different mtu values than the active slave we'll
3865 * need to change their mtu when doing a failover. That
3866 * means changing their mtu from timer context, which
3867 * is probably not a good idea.
3869 dprintk("err %d %s\n", res, slave->dev->name);
3870 goto unwind;
3874 bond_dev->mtu = new_mtu;
3876 return 0;
3878 unwind:
3879 /* unwind from head to the slave that failed */
3880 stop_at = slave;
3881 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3882 int tmp_res;
3884 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3885 if (tmp_res) {
3886 dprintk("unwind err %d dev %s\n", tmp_res,
3887 slave->dev->name);
3891 return res;
3895 * Change HW address
3897 * Note that many devices must be down to change the HW address, and
3898 * downing the master releases all slaves. We can make bonds full of
3899 * bonding devices to test this, however.
3901 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3903 struct bonding *bond = bond_dev->priv;
3904 struct sockaddr *sa = addr, tmp_sa;
3905 struct slave *slave, *stop_at;
3906 int res = 0;
3907 int i;
3909 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3911 if (!is_valid_ether_addr(sa->sa_data)) {
3912 return -EADDRNOTAVAIL;
3915 /* Can't hold bond->lock with bh disabled here since
3916 * some base drivers panic. On the other hand we can't
3917 * hold bond->lock without bh disabled because we'll
3918 * deadlock. The only solution is to rely on the fact
3919 * that we're under rtnl_lock here, and the slaves
3920 * list won't change. This doesn't solve the problem
3921 * of setting the slave's hw address while it is
3922 * transmitting, but the assumption is that the base
3923 * driver can handle that.
3925 * TODO: figure out a way to safely iterate the slaves
3926 * list, but without holding a lock around the actual
3927 * call to the base driver.
3930 bond_for_each_slave(bond, slave, i) {
3931 dprintk("slave %p %s\n", slave, slave->dev->name);
3933 if (slave->dev->set_mac_address == NULL) {
3934 res = -EOPNOTSUPP;
3935 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3936 goto unwind;
3939 res = dev_set_mac_address(slave->dev, addr);
3940 if (res) {
3941 /* TODO: consider downing the slave
3942 * and retry ?
3943 * User should expect communications
3944 * breakage anyway until ARP finish
3945 * updating, so...
3947 dprintk("err %d %s\n", res, slave->dev->name);
3948 goto unwind;
3952 /* success */
3953 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3954 return 0;
3956 unwind:
3957 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3958 tmp_sa.sa_family = bond_dev->type;
3960 /* unwind from head to the slave that failed */
3961 stop_at = slave;
3962 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3963 int tmp_res;
3965 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3966 if (tmp_res) {
3967 dprintk("unwind err %d dev %s\n", tmp_res,
3968 slave->dev->name);
3972 return res;
3975 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3977 struct bonding *bond = bond_dev->priv;
3978 struct slave *slave, *start_at;
3979 int i;
3980 int res = 1;
3982 read_lock(&bond->lock);
3984 if (!BOND_IS_OK(bond)) {
3985 goto out;
3988 read_lock(&bond->curr_slave_lock);
3989 slave = start_at = bond->curr_active_slave;
3990 read_unlock(&bond->curr_slave_lock);
3992 if (!slave) {
3993 goto out;
3996 bond_for_each_slave_from(bond, slave, i, start_at) {
3997 if (IS_UP(slave->dev) &&
3998 (slave->link == BOND_LINK_UP) &&
3999 (slave->state == BOND_STATE_ACTIVE)) {
4000 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4002 write_lock(&bond->curr_slave_lock);
4003 bond->curr_active_slave = slave->next;
4004 write_unlock(&bond->curr_slave_lock);
4006 break;
4011 out:
4012 if (res) {
4013 /* no suitable interface, frame not sent */
4014 dev_kfree_skb(skb);
4016 read_unlock(&bond->lock);
4017 return 0;
4020 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
4021 struct bonding *bond,
4022 struct slave *slave)
4024 struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
4025 struct ethhdr *eth_data;
4026 u8 *hwaddr;
4027 int res;
4029 if (!skb2) {
4030 printk(KERN_ERR DRV_NAME ": Error: "
4031 "bond_activebackup_xmit_copy(): skb_copy() failed\n");
4032 return;
4035 skb2->mac.raw = (unsigned char *)skb2->data;
4036 eth_data = eth_hdr(skb2);
4038 /* Pick an appropriate source MAC address
4039 * -- use slave's perm MAC addr, unless used by bond
4040 * -- otherwise, borrow active slave's perm MAC addr
4041 * since that will not be used
4043 hwaddr = slave->perm_hwaddr;
4044 if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
4045 hwaddr = bond->curr_active_slave->perm_hwaddr;
4047 /* Set source MAC address appropriately */
4048 memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
4050 res = bond_dev_queue_xmit(bond, skb2, slave->dev);
4051 if (res)
4052 dev_kfree_skb(skb2);
4054 return;
4058 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4059 * the bond has a usable interface.
4061 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4063 struct bonding *bond = bond_dev->priv;
4064 int res = 1;
4066 read_lock(&bond->lock);
4067 read_lock(&bond->curr_slave_lock);
4069 if (!BOND_IS_OK(bond)) {
4070 goto out;
4073 if (!bond->curr_active_slave)
4074 goto out;
4076 /* Xmit IGMP frames on all slaves to ensure rapid fail-over
4077 for multicast traffic on snooping switches */
4078 if (skb->protocol == __constant_htons(ETH_P_IP) &&
4079 skb->nh.iph->protocol == IPPROTO_IGMP) {
4080 struct slave *slave, *active_slave;
4081 int i;
4083 active_slave = bond->curr_active_slave;
4084 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
4085 active_slave->prev)
4086 if (IS_UP(slave->dev) &&
4087 (slave->link == BOND_LINK_UP))
4088 bond_activebackup_xmit_copy(skb, bond, slave);
4091 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4093 out:
4094 if (res) {
4095 /* no suitable interface, frame not sent */
4096 dev_kfree_skb(skb);
4098 read_unlock(&bond->curr_slave_lock);
4099 read_unlock(&bond->lock);
4100 return 0;
4104 * In bond_xmit_xor() , we determine the output device by using a pre-
4105 * determined xmit_hash_policy(), If the selected device is not enabled,
4106 * find the next active slave.
4108 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4110 struct bonding *bond = bond_dev->priv;
4111 struct slave *slave, *start_at;
4112 int slave_no;
4113 int i;
4114 int res = 1;
4116 read_lock(&bond->lock);
4118 if (!BOND_IS_OK(bond)) {
4119 goto out;
4122 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4124 bond_for_each_slave(bond, slave, i) {
4125 slave_no--;
4126 if (slave_no < 0) {
4127 break;
4131 start_at = slave;
4133 bond_for_each_slave_from(bond, slave, i, start_at) {
4134 if (IS_UP(slave->dev) &&
4135 (slave->link == BOND_LINK_UP) &&
4136 (slave->state == BOND_STATE_ACTIVE)) {
4137 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4138 break;
4142 out:
4143 if (res) {
4144 /* no suitable interface, frame not sent */
4145 dev_kfree_skb(skb);
4147 read_unlock(&bond->lock);
4148 return 0;
4152 * in broadcast mode, we send everything to all usable interfaces.
4154 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4156 struct bonding *bond = bond_dev->priv;
4157 struct slave *slave, *start_at;
4158 struct net_device *tx_dev = NULL;
4159 int i;
4160 int res = 1;
4162 read_lock(&bond->lock);
4164 if (!BOND_IS_OK(bond)) {
4165 goto out;
4168 read_lock(&bond->curr_slave_lock);
4169 start_at = bond->curr_active_slave;
4170 read_unlock(&bond->curr_slave_lock);
4172 if (!start_at) {
4173 goto out;
4176 bond_for_each_slave_from(bond, slave, i, start_at) {
4177 if (IS_UP(slave->dev) &&
4178 (slave->link == BOND_LINK_UP) &&
4179 (slave->state == BOND_STATE_ACTIVE)) {
4180 if (tx_dev) {
4181 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4182 if (!skb2) {
4183 printk(KERN_ERR DRV_NAME
4184 ": %s: Error: bond_xmit_broadcast(): "
4185 "skb_clone() failed\n",
4186 bond_dev->name);
4187 continue;
4190 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4191 if (res) {
4192 dev_kfree_skb(skb2);
4193 continue;
4196 tx_dev = slave->dev;
4200 if (tx_dev) {
4201 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4204 out:
4205 if (res) {
4206 /* no suitable interface, frame not sent */
4207 dev_kfree_skb(skb);
4209 /* frame sent to all suitable interfaces */
4210 read_unlock(&bond->lock);
4211 return 0;
4214 /*------------------------- Device initialization ---------------------------*/
4217 * set bond mode specific net device operations
4219 void bond_set_mode_ops(struct bonding *bond, int mode)
4221 struct net_device *bond_dev = bond->dev;
4223 switch (mode) {
4224 case BOND_MODE_ROUNDROBIN:
4225 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4226 break;
4227 case BOND_MODE_ACTIVEBACKUP:
4228 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4229 break;
4230 case BOND_MODE_XOR:
4231 bond_dev->hard_start_xmit = bond_xmit_xor;
4232 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4233 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4234 else
4235 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4236 break;
4237 case BOND_MODE_BROADCAST:
4238 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4239 break;
4240 case BOND_MODE_8023AD:
4241 bond_set_master_3ad_flags(bond);
4242 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4243 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4244 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4245 else
4246 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4247 break;
4248 case BOND_MODE_ALB:
4249 bond_set_master_alb_flags(bond);
4250 /* FALLTHRU */
4251 case BOND_MODE_TLB:
4252 bond_dev->hard_start_xmit = bond_alb_xmit;
4253 bond_dev->set_mac_address = bond_alb_set_mac_address;
4254 break;
4255 default:
4256 /* Should never happen, mode already checked */
4257 printk(KERN_ERR DRV_NAME
4258 ": %s: Error: Unknown bonding mode %d\n",
4259 bond_dev->name,
4260 mode);
4261 break;
4265 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4266 struct ethtool_drvinfo *drvinfo)
4268 strncpy(drvinfo->driver, DRV_NAME, 32);
4269 strncpy(drvinfo->version, DRV_VERSION, 32);
4270 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4273 static const struct ethtool_ops bond_ethtool_ops = {
4274 .get_tx_csum = ethtool_op_get_tx_csum,
4275 .get_tso = ethtool_op_get_tso,
4276 .get_ufo = ethtool_op_get_ufo,
4277 .get_sg = ethtool_op_get_sg,
4278 .get_drvinfo = bond_ethtool_get_drvinfo,
4282 * Does not allocate but creates a /proc entry.
4283 * Allowed to fail.
4285 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4287 struct bonding *bond = bond_dev->priv;
4289 dprintk("Begin bond_init for %s\n", bond_dev->name);
4291 /* initialize rwlocks */
4292 rwlock_init(&bond->lock);
4293 rwlock_init(&bond->curr_slave_lock);
4295 bond->params = *params; /* copy params struct */
4297 /* Initialize pointers */
4298 bond->first_slave = NULL;
4299 bond->curr_active_slave = NULL;
4300 bond->current_arp_slave = NULL;
4301 bond->primary_slave = NULL;
4302 bond->dev = bond_dev;
4303 INIT_LIST_HEAD(&bond->vlan_list);
4305 /* Initialize the device entry points */
4306 bond_dev->open = bond_open;
4307 bond_dev->stop = bond_close;
4308 bond_dev->get_stats = bond_get_stats;
4309 bond_dev->do_ioctl = bond_do_ioctl;
4310 bond_dev->ethtool_ops = &bond_ethtool_ops;
4311 bond_dev->set_multicast_list = bond_set_multicast_list;
4312 bond_dev->change_mtu = bond_change_mtu;
4313 bond_dev->set_mac_address = bond_set_mac_address;
4315 bond_set_mode_ops(bond, bond->params.mode);
4317 bond_dev->destructor = free_netdev;
4319 /* Initialize the device options */
4320 bond_dev->tx_queue_len = 0;
4321 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4322 bond_dev->priv_flags |= IFF_BONDING;
4324 /* At first, we block adding VLANs. That's the only way to
4325 * prevent problems that occur when adding VLANs over an
4326 * empty bond. The block will be removed once non-challenged
4327 * slaves are enslaved.
4329 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4331 /* don't acquire bond device's netif_tx_lock when
4332 * transmitting */
4333 bond_dev->features |= NETIF_F_LLTX;
4335 /* By default, we declare the bond to be fully
4336 * VLAN hardware accelerated capable. Special
4337 * care is taken in the various xmit functions
4338 * when there are slaves that are not hw accel
4339 * capable
4341 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4342 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4343 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4344 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4345 NETIF_F_HW_VLAN_RX |
4346 NETIF_F_HW_VLAN_FILTER);
4348 #ifdef CONFIG_PROC_FS
4349 bond_create_proc_entry(bond);
4350 #endif
4352 list_add_tail(&bond->bond_list, &bond_dev_list);
4354 return 0;
4357 /* De-initialize device specific data.
4358 * Caller must hold rtnl_lock.
4360 void bond_deinit(struct net_device *bond_dev)
4362 struct bonding *bond = bond_dev->priv;
4364 list_del(&bond->bond_list);
4366 #ifdef CONFIG_PROC_FS
4367 bond_remove_proc_entry(bond);
4368 #endif
4371 /* Unregister and free all bond devices.
4372 * Caller must hold rtnl_lock.
4374 static void bond_free_all(void)
4376 struct bonding *bond, *nxt;
4378 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4379 struct net_device *bond_dev = bond->dev;
4381 bond_mc_list_destroy(bond);
4382 /* Release the bonded slaves */
4383 bond_release_all(bond_dev);
4384 unregister_netdevice(bond_dev);
4385 bond_deinit(bond_dev);
4388 #ifdef CONFIG_PROC_FS
4389 bond_destroy_proc_dir();
4390 #endif
4393 /*------------------------- Module initialization ---------------------------*/
4396 * Convert string input module parms. Accept either the
4397 * number of the mode or its string name.
4399 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4401 int i;
4403 for (i = 0; tbl[i].modename; i++) {
4404 if ((isdigit(*mode_arg) &&
4405 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4406 (strncmp(mode_arg, tbl[i].modename,
4407 strlen(tbl[i].modename)) == 0)) {
4408 return tbl[i].mode;
4412 return -1;
4415 static int bond_check_params(struct bond_params *params)
4417 int arp_validate_value;
4420 * Convert string parameters.
4422 if (mode) {
4423 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4424 if (bond_mode == -1) {
4425 printk(KERN_ERR DRV_NAME
4426 ": Error: Invalid bonding mode \"%s\"\n",
4427 mode == NULL ? "NULL" : mode);
4428 return -EINVAL;
4432 if (xmit_hash_policy) {
4433 if ((bond_mode != BOND_MODE_XOR) &&
4434 (bond_mode != BOND_MODE_8023AD)) {
4435 printk(KERN_INFO DRV_NAME
4436 ": xor_mode param is irrelevant in mode %s\n",
4437 bond_mode_name(bond_mode));
4438 } else {
4439 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4440 xmit_hashtype_tbl);
4441 if (xmit_hashtype == -1) {
4442 printk(KERN_ERR DRV_NAME
4443 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4444 xmit_hash_policy == NULL ? "NULL" :
4445 xmit_hash_policy);
4446 return -EINVAL;
4451 if (lacp_rate) {
4452 if (bond_mode != BOND_MODE_8023AD) {
4453 printk(KERN_INFO DRV_NAME
4454 ": lacp_rate param is irrelevant in mode %s\n",
4455 bond_mode_name(bond_mode));
4456 } else {
4457 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4458 if (lacp_fast == -1) {
4459 printk(KERN_ERR DRV_NAME
4460 ": Error: Invalid lacp rate \"%s\"\n",
4461 lacp_rate == NULL ? "NULL" : lacp_rate);
4462 return -EINVAL;
4467 if (max_bonds < 1 || max_bonds > INT_MAX) {
4468 printk(KERN_WARNING DRV_NAME
4469 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4470 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4471 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4472 max_bonds = BOND_DEFAULT_MAX_BONDS;
4475 if (miimon < 0) {
4476 printk(KERN_WARNING DRV_NAME
4477 ": Warning: miimon module parameter (%d), "
4478 "not in range 0-%d, so it was reset to %d\n",
4479 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4480 miimon = BOND_LINK_MON_INTERV;
4483 if (updelay < 0) {
4484 printk(KERN_WARNING DRV_NAME
4485 ": Warning: updelay module parameter (%d), "
4486 "not in range 0-%d, so it was reset to 0\n",
4487 updelay, INT_MAX);
4488 updelay = 0;
4491 if (downdelay < 0) {
4492 printk(KERN_WARNING DRV_NAME
4493 ": Warning: downdelay module parameter (%d), "
4494 "not in range 0-%d, so it was reset to 0\n",
4495 downdelay, INT_MAX);
4496 downdelay = 0;
4499 if ((use_carrier != 0) && (use_carrier != 1)) {
4500 printk(KERN_WARNING DRV_NAME
4501 ": Warning: use_carrier module parameter (%d), "
4502 "not of valid value (0/1), so it was set to 1\n",
4503 use_carrier);
4504 use_carrier = 1;
4507 /* reset values for 802.3ad */
4508 if (bond_mode == BOND_MODE_8023AD) {
4509 if (!miimon) {
4510 printk(KERN_WARNING DRV_NAME
4511 ": Warning: miimon must be specified, "
4512 "otherwise bonding will not detect link "
4513 "failure, speed and duplex which are "
4514 "essential for 802.3ad operation\n");
4515 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4516 miimon = 100;
4520 /* reset values for TLB/ALB */
4521 if ((bond_mode == BOND_MODE_TLB) ||
4522 (bond_mode == BOND_MODE_ALB)) {
4523 if (!miimon) {
4524 printk(KERN_WARNING DRV_NAME
4525 ": Warning: miimon must be specified, "
4526 "otherwise bonding will not detect link "
4527 "failure and link speed which are essential "
4528 "for TLB/ALB load balancing\n");
4529 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4530 miimon = 100;
4534 if (bond_mode == BOND_MODE_ALB) {
4535 printk(KERN_NOTICE DRV_NAME
4536 ": In ALB mode you might experience client "
4537 "disconnections upon reconnection of a link if the "
4538 "bonding module updelay parameter (%d msec) is "
4539 "incompatible with the forwarding delay time of the "
4540 "switch\n",
4541 updelay);
4544 if (!miimon) {
4545 if (updelay || downdelay) {
4546 /* just warn the user the up/down delay will have
4547 * no effect since miimon is zero...
4549 printk(KERN_WARNING DRV_NAME
4550 ": Warning: miimon module parameter not set "
4551 "and updelay (%d) or downdelay (%d) module "
4552 "parameter is set; updelay and downdelay have "
4553 "no effect unless miimon is set\n",
4554 updelay, downdelay);
4556 } else {
4557 /* don't allow arp monitoring */
4558 if (arp_interval) {
4559 printk(KERN_WARNING DRV_NAME
4560 ": Warning: miimon (%d) and arp_interval (%d) "
4561 "can't be used simultaneously, disabling ARP "
4562 "monitoring\n",
4563 miimon, arp_interval);
4564 arp_interval = 0;
4567 if ((updelay % miimon) != 0) {
4568 printk(KERN_WARNING DRV_NAME
4569 ": Warning: updelay (%d) is not a multiple "
4570 "of miimon (%d), updelay rounded to %d ms\n",
4571 updelay, miimon, (updelay / miimon) * miimon);
4574 updelay /= miimon;
4576 if ((downdelay % miimon) != 0) {
4577 printk(KERN_WARNING DRV_NAME
4578 ": Warning: downdelay (%d) is not a multiple "
4579 "of miimon (%d), downdelay rounded to %d ms\n",
4580 downdelay, miimon,
4581 (downdelay / miimon) * miimon);
4584 downdelay /= miimon;
4587 if (arp_interval < 0) {
4588 printk(KERN_WARNING DRV_NAME
4589 ": Warning: arp_interval module parameter (%d) "
4590 ", not in range 0-%d, so it was reset to %d\n",
4591 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4592 arp_interval = BOND_LINK_ARP_INTERV;
4595 for (arp_ip_count = 0;
4596 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4597 arp_ip_count++) {
4598 /* not complete check, but should be good enough to
4599 catch mistakes */
4600 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4601 printk(KERN_WARNING DRV_NAME
4602 ": Warning: bad arp_ip_target module parameter "
4603 "(%s), ARP monitoring will not be performed\n",
4604 arp_ip_target[arp_ip_count]);
4605 arp_interval = 0;
4606 } else {
4607 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4608 arp_target[arp_ip_count] = ip;
4612 if (arp_interval && !arp_ip_count) {
4613 /* don't allow arping if no arp_ip_target given... */
4614 printk(KERN_WARNING DRV_NAME
4615 ": Warning: arp_interval module parameter (%d) "
4616 "specified without providing an arp_ip_target "
4617 "parameter, arp_interval was reset to 0\n",
4618 arp_interval);
4619 arp_interval = 0;
4622 if (arp_validate) {
4623 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4624 printk(KERN_ERR DRV_NAME
4625 ": arp_validate only supported in active-backup mode\n");
4626 return -EINVAL;
4628 if (!arp_interval) {
4629 printk(KERN_ERR DRV_NAME
4630 ": arp_validate requires arp_interval\n");
4631 return -EINVAL;
4634 arp_validate_value = bond_parse_parm(arp_validate,
4635 arp_validate_tbl);
4636 if (arp_validate_value == -1) {
4637 printk(KERN_ERR DRV_NAME
4638 ": Error: invalid arp_validate \"%s\"\n",
4639 arp_validate == NULL ? "NULL" : arp_validate);
4640 return -EINVAL;
4642 } else
4643 arp_validate_value = 0;
4645 if (miimon) {
4646 printk(KERN_INFO DRV_NAME
4647 ": MII link monitoring set to %d ms\n",
4648 miimon);
4649 } else if (arp_interval) {
4650 int i;
4652 printk(KERN_INFO DRV_NAME
4653 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4654 arp_interval,
4655 arp_validate_tbl[arp_validate_value].modename,
4656 arp_ip_count);
4658 for (i = 0; i < arp_ip_count; i++)
4659 printk (" %s", arp_ip_target[i]);
4661 printk("\n");
4663 } else {
4664 /* miimon and arp_interval not set, we need one so things
4665 * work as expected, see bonding.txt for details
4667 printk(KERN_WARNING DRV_NAME
4668 ": Warning: either miimon or arp_interval and "
4669 "arp_ip_target module parameters must be specified, "
4670 "otherwise bonding will not detect link failures! see "
4671 "bonding.txt for details.\n");
4674 if (primary && !USES_PRIMARY(bond_mode)) {
4675 /* currently, using a primary only makes sense
4676 * in active backup, TLB or ALB modes
4678 printk(KERN_WARNING DRV_NAME
4679 ": Warning: %s primary device specified but has no "
4680 "effect in %s mode\n",
4681 primary, bond_mode_name(bond_mode));
4682 primary = NULL;
4685 /* fill params struct with the proper values */
4686 params->mode = bond_mode;
4687 params->xmit_policy = xmit_hashtype;
4688 params->miimon = miimon;
4689 params->arp_interval = arp_interval;
4690 params->arp_validate = arp_validate_value;
4691 params->updelay = updelay;
4692 params->downdelay = downdelay;
4693 params->use_carrier = use_carrier;
4694 params->lacp_fast = lacp_fast;
4695 params->primary[0] = 0;
4697 if (primary) {
4698 strncpy(params->primary, primary, IFNAMSIZ);
4699 params->primary[IFNAMSIZ - 1] = 0;
4702 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4704 return 0;
4707 static struct lock_class_key bonding_netdev_xmit_lock_key;
4709 /* Create a new bond based on the specified name and bonding parameters.
4710 * If name is NULL, obtain a suitable "bond%d" name for us.
4711 * Caller must NOT hold rtnl_lock; we need to release it here before we
4712 * set up our sysfs entries.
4714 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4716 struct net_device *bond_dev;
4717 int res;
4719 rtnl_lock();
4720 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4721 ether_setup);
4722 if (!bond_dev) {
4723 printk(KERN_ERR DRV_NAME
4724 ": %s: eek! can't alloc netdev!\n",
4725 name);
4726 res = -ENOMEM;
4727 goto out_rtnl;
4730 if (!name) {
4731 res = dev_alloc_name(bond_dev, "bond%d");
4732 if (res < 0)
4733 goto out_netdev;
4736 /* bond_init() must be called after dev_alloc_name() (for the
4737 * /proc files), but before register_netdevice(), because we
4738 * need to set function pointers.
4741 res = bond_init(bond_dev, params);
4742 if (res < 0) {
4743 goto out_netdev;
4746 SET_MODULE_OWNER(bond_dev);
4748 res = register_netdevice(bond_dev);
4749 if (res < 0) {
4750 goto out_bond;
4753 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4755 if (newbond)
4756 *newbond = bond_dev->priv;
4758 netif_carrier_off(bond_dev);
4760 rtnl_unlock(); /* allows sysfs registration of net device */
4761 res = bond_create_sysfs_entry(bond_dev->priv);
4762 if (res < 0) {
4763 rtnl_lock();
4764 goto out_bond;
4767 return 0;
4769 out_bond:
4770 bond_deinit(bond_dev);
4771 out_netdev:
4772 free_netdev(bond_dev);
4773 out_rtnl:
4774 rtnl_unlock();
4775 return res;
4778 static int __init bonding_init(void)
4780 int i;
4781 int res;
4783 printk(KERN_INFO "%s", version);
4785 res = bond_check_params(&bonding_defaults);
4786 if (res) {
4787 goto out;
4790 #ifdef CONFIG_PROC_FS
4791 bond_create_proc_dir();
4792 #endif
4793 for (i = 0; i < max_bonds; i++) {
4794 res = bond_create(NULL, &bonding_defaults, NULL);
4795 if (res)
4796 goto err;
4799 res = bond_create_sysfs();
4800 if (res)
4801 goto err;
4803 register_netdevice_notifier(&bond_netdev_notifier);
4804 register_inetaddr_notifier(&bond_inetaddr_notifier);
4806 goto out;
4807 err:
4808 rtnl_lock();
4809 bond_free_all();
4810 bond_destroy_sysfs();
4811 rtnl_unlock();
4812 out:
4813 return res;
4817 static void __exit bonding_exit(void)
4819 unregister_netdevice_notifier(&bond_netdev_notifier);
4820 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4822 rtnl_lock();
4823 bond_free_all();
4824 bond_destroy_sysfs();
4825 rtnl_unlock();
4828 module_init(bonding_init);
4829 module_exit(bonding_exit);
4830 MODULE_LICENSE("GPL");
4831 MODULE_VERSION(DRV_VERSION);
4832 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4833 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4834 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4837 * Local variables:
4838 * c-indent-level: 8
4839 * c-basic-offset: 8
4840 * tab-width: 8
4841 * End: