bonding / ipv6: no addrconf for slaves separately from master.
[tomato.git] / release / src-rt / linux / linux-2.6 / drivers / net / bonding / bond_main.c
blob99cd0818f33a37a309cab94308e7b70ae7c48ac1
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/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
66 #include <net/sock.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
72 #include <net/arp.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <net/route.h>
78 #include "bonding.h"
79 #include "bond_3ad.h"
80 #include "bond_alb.h"
82 /*---------------------------- Module parameters ----------------------------*/
84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
85 #define BOND_LINK_MON_INTERV 0
86 #define BOND_LINK_ARP_INTERV 0
88 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
89 static int miimon = BOND_LINK_MON_INTERV;
90 static int updelay = 0;
91 static int downdelay = 0;
92 static int use_carrier = 1;
93 static char *mode = NULL;
94 static char *primary = NULL;
95 static char *lacp_rate = NULL;
96 static char *xmit_hash_policy = NULL;
97 static int arp_interval = BOND_LINK_ARP_INTERV;
98 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
99 static char *arp_validate = NULL;
100 struct bond_params bonding_defaults;
102 module_param(max_bonds, int, 0);
103 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
104 module_param(miimon, int, 0);
105 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
106 module_param(updelay, int, 0);
107 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
108 module_param(downdelay, int, 0);
109 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
110 "in milliseconds");
111 module_param(use_carrier, int, 0);
112 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
113 "0 for off, 1 for on (default)");
114 module_param(mode, charp, 0);
115 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
116 "1 for active-backup, 2 for balance-xor, "
117 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
118 "6 for balance-alb");
119 module_param(primary, charp, 0);
120 MODULE_PARM_DESC(primary, "Primary network device to use");
121 module_param(lacp_rate, charp, 0);
122 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
123 "(slow/fast)");
124 module_param(xmit_hash_policy, charp, 0);
125 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
126 ", 1 for layer 3+4");
127 module_param(arp_interval, int, 0);
128 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
129 module_param_array(arp_ip_target, charp, NULL, 0);
130 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
131 module_param(arp_validate, charp, 0);
132 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
134 /*----------------------------- Global variables ----------------------------*/
136 static const char * const version =
137 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
139 LIST_HEAD(bond_dev_list);
141 #ifdef CONFIG_PROC_FS
142 static struct proc_dir_entry *bond_proc_dir = NULL;
143 #endif
145 extern struct rw_semaphore bonding_rwsem;
146 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
147 static int arp_ip_count = 0;
148 static int bond_mode = BOND_MODE_ROUNDROBIN;
149 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
150 static int lacp_fast = 0;
153 struct bond_parm_tbl bond_lacp_tbl[] = {
154 { "slow", AD_LACP_SLOW},
155 { "fast", AD_LACP_FAST},
156 { NULL, -1},
159 struct bond_parm_tbl bond_mode_tbl[] = {
160 { "balance-rr", BOND_MODE_ROUNDROBIN},
161 { "active-backup", BOND_MODE_ACTIVEBACKUP},
162 { "balance-xor", BOND_MODE_XOR},
163 { "broadcast", BOND_MODE_BROADCAST},
164 { "802.3ad", BOND_MODE_8023AD},
165 { "balance-tlb", BOND_MODE_TLB},
166 { "balance-alb", BOND_MODE_ALB},
167 { NULL, -1},
170 struct bond_parm_tbl xmit_hashtype_tbl[] = {
171 { "layer2", BOND_XMIT_POLICY_LAYER2},
172 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
173 { NULL, -1},
176 struct bond_parm_tbl arp_validate_tbl[] = {
177 { "none", BOND_ARP_VALIDATE_NONE},
178 { "active", BOND_ARP_VALIDATE_ACTIVE},
179 { "backup", BOND_ARP_VALIDATE_BACKUP},
180 { "all", BOND_ARP_VALIDATE_ALL},
181 { NULL, -1},
184 /*-------------------------- Forward declarations ---------------------------*/
186 static void bond_send_gratuitous_arp(struct bonding *bond);
188 /*---------------------------- General routines -----------------------------*/
190 const char *bond_mode_name(int mode)
192 switch (mode) {
193 case BOND_MODE_ROUNDROBIN :
194 return "load balancing (round-robin)";
195 case BOND_MODE_ACTIVEBACKUP :
196 return "fault-tolerance (active-backup)";
197 case BOND_MODE_XOR :
198 return "load balancing (xor)";
199 case BOND_MODE_BROADCAST :
200 return "fault-tolerance (broadcast)";
201 case BOND_MODE_8023AD:
202 return "IEEE 802.3ad Dynamic link aggregation";
203 case BOND_MODE_TLB:
204 return "transmit load balancing";
205 case BOND_MODE_ALB:
206 return "adaptive load balancing";
207 default:
208 return "unknown";
212 /*---------------------------------- VLAN -----------------------------------*/
215 * bond_add_vlan - add a new vlan id on bond
216 * @bond: bond that got the notification
217 * @vlan_id: the vlan id to add
219 * Returns -ENOMEM if allocation failed.
221 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
223 struct vlan_entry *vlan;
225 dprintk("bond: %s, vlan id %d\n",
226 (bond ? bond->dev->name: "None"), vlan_id);
228 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
229 if (!vlan) {
230 return -ENOMEM;
233 INIT_LIST_HEAD(&vlan->vlan_list);
234 vlan->vlan_id = vlan_id;
235 vlan->vlan_ip = 0;
237 write_lock_bh(&bond->lock);
239 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
241 write_unlock_bh(&bond->lock);
243 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
245 return 0;
249 * bond_del_vlan - delete a vlan id from bond
250 * @bond: bond that got the notification
251 * @vlan_id: the vlan id to delete
253 * returns -ENODEV if @vlan_id was not found in @bond.
255 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
257 struct vlan_entry *vlan, *next;
258 int res = -ENODEV;
260 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
262 write_lock_bh(&bond->lock);
264 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
265 if (vlan->vlan_id == vlan_id) {
266 list_del(&vlan->vlan_list);
268 if ((bond->params.mode == BOND_MODE_TLB) ||
269 (bond->params.mode == BOND_MODE_ALB)) {
270 bond_alb_clear_vlan(bond, vlan_id);
273 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
274 bond->dev->name);
276 kfree(vlan);
278 if (list_empty(&bond->vlan_list) &&
279 (bond->slave_cnt == 0)) {
280 /* Last VLAN removed and no slaves, so
281 * restore block on adding VLANs. This will
282 * be removed once new slaves that are not
283 * VLAN challenged will be added.
285 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
288 res = 0;
289 goto out;
293 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
294 bond->dev->name);
296 out:
297 write_unlock_bh(&bond->lock);
298 return res;
302 * bond_has_challenged_slaves
303 * @bond: the bond we're working on
305 * Searches the slave list. Returns 1 if a vlan challenged slave
306 * was found, 0 otherwise.
308 * Assumes bond->lock is held.
310 static int bond_has_challenged_slaves(struct bonding *bond)
312 struct slave *slave;
313 int i;
315 bond_for_each_slave(bond, slave, i) {
316 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
317 dprintk("found VLAN challenged slave - %s\n",
318 slave->dev->name);
319 return 1;
323 dprintk("no VLAN challenged slaves found\n");
324 return 0;
328 * bond_next_vlan - safely skip to the next item in the vlans list.
329 * @bond: the bond we're working on
330 * @curr: item we're advancing from
332 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
333 * or @curr->next otherwise (even if it is @curr itself again).
335 * Caller must hold bond->lock
337 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
339 struct vlan_entry *next, *last;
341 if (list_empty(&bond->vlan_list)) {
342 return NULL;
345 if (!curr) {
346 next = list_entry(bond->vlan_list.next,
347 struct vlan_entry, vlan_list);
348 } else {
349 last = list_entry(bond->vlan_list.prev,
350 struct vlan_entry, vlan_list);
351 if (last == curr) {
352 next = list_entry(bond->vlan_list.next,
353 struct vlan_entry, vlan_list);
354 } else {
355 next = list_entry(curr->vlan_list.next,
356 struct vlan_entry, vlan_list);
360 return next;
364 * bond_dev_queue_xmit - Prepare skb for xmit.
366 * @bond: bond device that got this skb for tx.
367 * @skb: hw accel VLAN tagged skb to transmit
368 * @slave_dev: slave that is supposed to xmit this skbuff
370 * When the bond gets an skb to transmit that is
371 * already hardware accelerated VLAN tagged, and it
372 * needs to relay this skb to a slave that is not
373 * hw accel capable, the skb needs to be "unaccelerated",
374 * i.e. strip the hwaccel tag and re-insert it as part
375 * of the payload.
377 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
379 unsigned short vlan_id;
381 if (!list_empty(&bond->vlan_list) &&
382 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
383 vlan_get_tag(skb, &vlan_id) == 0) {
384 skb->dev = slave_dev;
385 skb = vlan_put_tag(skb, vlan_id);
386 if (!skb) {
387 /* vlan_put_tag() frees the skb in case of error,
388 * so return success here so the calling functions
389 * won't attempt to free is again.
391 return 0;
393 } else {
394 skb->dev = slave_dev;
397 skb->priority = 1;
398 dev_queue_xmit(skb);
400 return 0;
404 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
405 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
406 * lock because:
407 * a. This operation is performed in IOCTL context,
408 * b. The operation is protected by the RTNL semaphore in the 8021q code,
409 * c. Holding a lock with BH disabled while directly calling a base driver
410 * entry point is generally a BAD idea.
412 * The design of synchronization/protection for this operation in the 8021q
413 * module is good for one or more VLAN devices over a single physical device
414 * and cannot be extended for a teaming solution like bonding, so there is a
415 * potential race condition here where a net device from the vlan group might
416 * be referenced (either by a base driver or the 8021q code) while it is being
417 * removed from the system. However, it turns out we're not making matters
418 * worse, and if it works for regular VLAN usage it will work here too.
422 * bond_vlan_rx_register - Propagates registration to slaves
423 * @bond_dev: bonding net device that got called
424 * @grp: vlan group being registered
426 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
428 struct bonding *bond = bond_dev->priv;
429 struct slave *slave;
430 int i;
432 bond->vlgrp = grp;
434 bond_for_each_slave(bond, slave, i) {
435 struct net_device *slave_dev = slave->dev;
437 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
438 slave_dev->vlan_rx_register) {
439 slave_dev->vlan_rx_register(slave_dev, grp);
445 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
446 * @bond_dev: bonding net device that got called
447 * @vid: vlan id being added
449 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
451 struct bonding *bond = bond_dev->priv;
452 struct slave *slave;
453 int i, res;
455 bond_for_each_slave(bond, slave, i) {
456 struct net_device *slave_dev = slave->dev;
458 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
459 slave_dev->vlan_rx_add_vid) {
460 slave_dev->vlan_rx_add_vid(slave_dev, vid);
464 res = bond_add_vlan(bond, vid);
465 if (res) {
466 printk(KERN_ERR DRV_NAME
467 ": %s: Error: Failed to add vlan id %d\n",
468 bond_dev->name, vid);
473 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
474 * @bond_dev: bonding net device that got called
475 * @vid: vlan id being removed
477 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
479 struct bonding *bond = bond_dev->priv;
480 struct slave *slave;
481 struct net_device *vlan_dev;
482 int i, res;
484 bond_for_each_slave(bond, slave, i) {
485 struct net_device *slave_dev = slave->dev;
487 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
488 slave_dev->vlan_rx_kill_vid) {
489 /* Save and then restore vlan_dev in the grp array,
490 * since the slave's driver might clear it.
492 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
493 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
494 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
498 res = bond_del_vlan(bond, vid);
499 if (res) {
500 printk(KERN_ERR DRV_NAME
501 ": %s: Error: Failed to remove vlan id %d\n",
502 bond_dev->name, vid);
506 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
508 struct vlan_entry *vlan;
510 write_lock_bh(&bond->lock);
512 if (list_empty(&bond->vlan_list)) {
513 goto out;
516 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
517 slave_dev->vlan_rx_register) {
518 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
521 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
522 !(slave_dev->vlan_rx_add_vid)) {
523 goto out;
526 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
527 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
530 out:
531 write_unlock_bh(&bond->lock);
534 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
536 struct vlan_entry *vlan;
537 struct net_device *vlan_dev;
539 write_lock_bh(&bond->lock);
541 if (list_empty(&bond->vlan_list)) {
542 goto out;
545 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
546 !(slave_dev->vlan_rx_kill_vid)) {
547 goto unreg;
550 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
551 /* Save and then restore vlan_dev in the grp array,
552 * since the slave's driver might clear it.
554 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
555 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
556 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
559 unreg:
560 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
561 slave_dev->vlan_rx_register) {
562 slave_dev->vlan_rx_register(slave_dev, NULL);
565 out:
566 write_unlock_bh(&bond->lock);
569 /*------------------------------- Link status -------------------------------*/
572 * Set the carrier state for the master according to the state of its
573 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
574 * do special 802.3ad magic.
576 * Returns zero if carrier state does not change, nonzero if it does.
578 static int bond_set_carrier(struct bonding *bond)
580 struct slave *slave;
581 int i;
583 if (bond->slave_cnt == 0)
584 goto down;
586 if (bond->params.mode == BOND_MODE_8023AD)
587 return bond_3ad_set_carrier(bond);
589 bond_for_each_slave(bond, slave, i) {
590 if (slave->link == BOND_LINK_UP) {
591 if (!netif_carrier_ok(bond->dev)) {
592 netif_carrier_on(bond->dev);
593 return 1;
595 return 0;
599 down:
600 if (netif_carrier_ok(bond->dev)) {
601 netif_carrier_off(bond->dev);
602 return 1;
604 return 0;
608 * Get link speed and duplex from the slave's base driver
609 * using ethtool. If for some reason the call fails or the
610 * values are invalid, fake speed and duplex to 100/Full
611 * and return error.
613 static int bond_update_speed_duplex(struct slave *slave)
615 struct net_device *slave_dev = slave->dev;
616 static int (* ioctl)(struct net_device *, struct ifreq *, int);
617 struct ifreq ifr;
618 struct ethtool_cmd etool;
620 /* Fake speed and duplex */
621 slave->speed = SPEED_100;
622 slave->duplex = DUPLEX_FULL;
624 if (slave_dev->ethtool_ops) {
625 int res;
627 if (!slave_dev->ethtool_ops->get_settings) {
628 return -1;
631 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
632 if (res < 0) {
633 return -1;
636 goto verify;
639 ioctl = slave_dev->do_ioctl;
640 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
641 etool.cmd = ETHTOOL_GSET;
642 ifr.ifr_data = (char*)&etool;
643 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
644 return -1;
647 verify:
648 switch (etool.speed) {
649 case SPEED_10:
650 case SPEED_100:
651 case SPEED_1000:
652 case SPEED_10000:
653 break;
654 default:
655 return -1;
658 switch (etool.duplex) {
659 case DUPLEX_FULL:
660 case DUPLEX_HALF:
661 break;
662 default:
663 return -1;
666 slave->speed = etool.speed;
667 slave->duplex = etool.duplex;
669 return 0;
673 * if <dev> supports MII link status reporting, check its link status.
675 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
676 * depening upon the setting of the use_carrier parameter.
678 * Return either BMSR_LSTATUS, meaning that the link is up (or we
679 * can't tell and just pretend it is), or 0, meaning that the link is
680 * down.
682 * If reporting is non-zero, instead of faking link up, return -1 if
683 * both ETHTOOL and MII ioctls fail (meaning the device does not
684 * support them). If use_carrier is set, return whatever it says.
685 * It'd be nice if there was a good way to tell if a driver supports
686 * netif_carrier, but there really isn't.
688 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
690 static int (* ioctl)(struct net_device *, struct ifreq *, int);
691 struct ifreq ifr;
692 struct mii_ioctl_data *mii;
693 struct ethtool_value etool;
695 if (bond->params.use_carrier) {
696 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
699 ioctl = slave_dev->do_ioctl;
700 if (ioctl) {
701 /* TODO: set pointer to correct ioctl on a per team member */
702 /* bases to make this more efficient. that is, once */
703 /* we determine the correct ioctl, we will always */
704 /* call it and not the others for that team */
705 /* member. */
708 * We cannot assume that SIOCGMIIPHY will also read a
709 * register; not all network drivers (e.g., e100)
710 * support that.
713 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
714 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
715 mii = if_mii(&ifr);
716 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
717 mii->reg_num = MII_BMSR;
718 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
719 return (mii->val_out & BMSR_LSTATUS);
724 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
725 /* for a period of time so we attempt to get link status */
726 /* from it last if the above MII ioctls fail... */
727 if (slave_dev->ethtool_ops) {
728 if (slave_dev->ethtool_ops->get_link) {
729 u32 link;
731 link = slave_dev->ethtool_ops->get_link(slave_dev);
733 return link ? BMSR_LSTATUS : 0;
737 if (ioctl) {
738 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
739 etool.cmd = ETHTOOL_GLINK;
740 ifr.ifr_data = (char*)&etool;
741 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
742 if (etool.data == 1) {
743 return BMSR_LSTATUS;
744 } else {
745 dprintk("SIOCETHTOOL shows link down\n");
746 return 0;
752 * If reporting, report that either there's no dev->do_ioctl,
753 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
754 * cannot report link status). If not reporting, pretend
755 * we're ok.
757 return (reporting ? -1 : BMSR_LSTATUS);
760 /*----------------------------- Multicast list ------------------------------*/
763 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
765 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
767 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
768 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
772 * returns dmi entry if found, NULL otherwise
774 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
776 struct dev_mc_list *idmi;
778 for (idmi = mc_list; idmi; idmi = idmi->next) {
779 if (bond_is_dmi_same(dmi, idmi)) {
780 return idmi;
784 return NULL;
788 * Push the promiscuity flag down to appropriate slaves
790 static void bond_set_promiscuity(struct bonding *bond, int inc)
792 if (USES_PRIMARY(bond->params.mode)) {
793 /* write lock already acquired */
794 if (bond->curr_active_slave) {
795 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
797 } else {
798 struct slave *slave;
799 int i;
800 bond_for_each_slave(bond, slave, i) {
801 dev_set_promiscuity(slave->dev, inc);
807 * Push the allmulti flag down to all slaves
809 static void bond_set_allmulti(struct bonding *bond, int inc)
811 if (USES_PRIMARY(bond->params.mode)) {
812 /* write lock already acquired */
813 if (bond->curr_active_slave) {
814 dev_set_allmulti(bond->curr_active_slave->dev, inc);
816 } else {
817 struct slave *slave;
818 int i;
819 bond_for_each_slave(bond, slave, i) {
820 dev_set_allmulti(slave->dev, inc);
826 * Add a Multicast address to slaves
827 * according to mode
829 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
831 if (USES_PRIMARY(bond->params.mode)) {
832 /* write lock already acquired */
833 if (bond->curr_active_slave) {
834 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
836 } else {
837 struct slave *slave;
838 int i;
839 bond_for_each_slave(bond, slave, i) {
840 dev_mc_add(slave->dev, addr, alen, 0);
846 * Remove a multicast address from slave
847 * according to mode
849 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
851 if (USES_PRIMARY(bond->params.mode)) {
852 /* write lock already acquired */
853 if (bond->curr_active_slave) {
854 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
856 } else {
857 struct slave *slave;
858 int i;
859 bond_for_each_slave(bond, slave, i) {
860 dev_mc_delete(slave->dev, addr, alen, 0);
867 * Retrieve the list of registered multicast addresses for the bonding
868 * device and retransmit an IGMP JOIN request to the current active
869 * slave.
871 static void bond_resend_igmp_join_requests(struct bonding *bond)
873 struct in_device *in_dev;
874 struct ip_mc_list *im;
876 rcu_read_lock();
877 in_dev = __in_dev_get_rcu(bond->dev);
878 if (in_dev) {
879 for (im = in_dev->mc_list; im; im = im->next) {
880 ip_mc_rejoin_group(im);
884 rcu_read_unlock();
888 * Totally destroys the mc_list in bond
890 static void bond_mc_list_destroy(struct bonding *bond)
892 struct dev_mc_list *dmi;
894 dmi = bond->mc_list;
895 while (dmi) {
896 bond->mc_list = dmi->next;
897 kfree(dmi);
898 dmi = bond->mc_list;
900 bond->mc_list = NULL;
904 * Copy all the Multicast addresses from src to the bonding device dst
906 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
907 gfp_t gfp_flag)
909 struct dev_mc_list *dmi, *new_dmi;
911 for (dmi = mc_list; dmi; dmi = dmi->next) {
912 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
914 if (!new_dmi) {
915 /* FIXME: Potential memory leak !!! */
916 return -ENOMEM;
919 new_dmi->next = bond->mc_list;
920 bond->mc_list = new_dmi;
921 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
922 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
923 new_dmi->dmi_users = dmi->dmi_users;
924 new_dmi->dmi_gusers = dmi->dmi_gusers;
927 return 0;
931 * flush all members of flush->mc_list from device dev->mc_list
933 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
935 struct bonding *bond = bond_dev->priv;
936 struct dev_mc_list *dmi;
938 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
939 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
942 if (bond->params.mode == BOND_MODE_8023AD) {
943 /* del lacpdu mc addr from mc list */
944 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
946 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
950 /*--------------------------- Active slave change ---------------------------*/
953 * Update the mc list and multicast-related flags for the new and
954 * old active slaves (if any) according to the multicast mode, and
955 * promiscuous flags unconditionally.
957 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
959 struct dev_mc_list *dmi;
961 if (!USES_PRIMARY(bond->params.mode)) {
962 /* nothing to do - mc list is already up-to-date on
963 * all slaves
965 return;
968 if (old_active) {
969 if (bond->dev->flags & IFF_PROMISC) {
970 dev_set_promiscuity(old_active->dev, -1);
973 if (bond->dev->flags & IFF_ALLMULTI) {
974 dev_set_allmulti(old_active->dev, -1);
977 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
978 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
982 if (new_active) {
983 if (bond->dev->flags & IFF_PROMISC) {
984 dev_set_promiscuity(new_active->dev, 1);
987 if (bond->dev->flags & IFF_ALLMULTI) {
988 dev_set_allmulti(new_active->dev, 1);
991 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
992 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
994 bond_resend_igmp_join_requests(bond);
999 * find_best_interface - select the best available slave to be the active one
1000 * @bond: our bonding struct
1002 * Warning: Caller must hold curr_slave_lock for writing.
1004 static struct slave *bond_find_best_slave(struct bonding *bond)
1006 struct slave *new_active, *old_active;
1007 struct slave *bestslave = NULL;
1008 int mintime = bond->params.updelay;
1009 int i;
1011 new_active = old_active = bond->curr_active_slave;
1013 if (!new_active) { /* there were no active slaves left */
1014 if (bond->slave_cnt > 0) { /* found one slave */
1015 new_active = bond->first_slave;
1016 } else {
1017 return NULL; /* still no slave, return NULL */
1021 /* first try the primary link; if arping, a link must tx/rx traffic
1022 * before it can be considered the curr_active_slave - also, we would skip
1023 * slaves between the curr_active_slave and primary_slave that may be up
1024 * and able to arp
1026 if ((bond->primary_slave) &&
1027 (!bond->params.arp_interval) &&
1028 (IS_UP(bond->primary_slave->dev))) {
1029 new_active = bond->primary_slave;
1032 /* remember where to stop iterating over the slaves */
1033 old_active = new_active;
1035 bond_for_each_slave_from(bond, new_active, i, old_active) {
1036 if (IS_UP(new_active->dev)) {
1037 if (new_active->link == BOND_LINK_UP) {
1038 return new_active;
1039 } else if (new_active->link == BOND_LINK_BACK) {
1040 /* link up, but waiting for stabilization */
1041 if (new_active->delay < mintime) {
1042 mintime = new_active->delay;
1043 bestslave = new_active;
1049 return bestslave;
1053 * change_active_interface - change the active slave into the specified one
1054 * @bond: our bonding struct
1055 * @new: the new slave to make the active one
1057 * Set the new slave to the bond's settings and unset them on the old
1058 * curr_active_slave.
1059 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1061 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1062 * because it is apparently the best available slave we have, even though its
1063 * updelay hasn't timed out yet.
1065 * Warning: Caller must hold curr_slave_lock for writing.
1067 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1069 struct slave *old_active = bond->curr_active_slave;
1071 if (old_active == new_active) {
1072 return;
1075 if (new_active) {
1076 if (new_active->link == BOND_LINK_BACK) {
1077 if (USES_PRIMARY(bond->params.mode)) {
1078 printk(KERN_INFO DRV_NAME
1079 ": %s: making interface %s the new "
1080 "active one %d ms earlier.\n",
1081 bond->dev->name, new_active->dev->name,
1082 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1085 new_active->delay = 0;
1086 new_active->link = BOND_LINK_UP;
1087 new_active->jiffies = jiffies;
1089 if (bond->params.mode == BOND_MODE_8023AD) {
1090 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1093 if ((bond->params.mode == BOND_MODE_TLB) ||
1094 (bond->params.mode == BOND_MODE_ALB)) {
1095 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1097 } else {
1098 if (USES_PRIMARY(bond->params.mode)) {
1099 printk(KERN_INFO DRV_NAME
1100 ": %s: making interface %s the new "
1101 "active one.\n",
1102 bond->dev->name, new_active->dev->name);
1107 if (USES_PRIMARY(bond->params.mode)) {
1108 bond_mc_swap(bond, new_active, old_active);
1111 if ((bond->params.mode == BOND_MODE_TLB) ||
1112 (bond->params.mode == BOND_MODE_ALB)) {
1113 bond_alb_handle_active_change(bond, new_active);
1114 if (old_active)
1115 bond_set_slave_inactive_flags(old_active);
1116 if (new_active)
1117 bond_set_slave_active_flags(new_active);
1118 } else {
1119 bond->curr_active_slave = new_active;
1122 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1123 if (old_active) {
1124 bond_set_slave_inactive_flags(old_active);
1127 if (new_active) {
1128 bond_set_slave_active_flags(new_active);
1130 bond_send_gratuitous_arp(bond);
1135 * bond_select_active_slave - select a new active slave, if needed
1136 * @bond: our bonding struct
1138 * This functions shoud be called when one of the following occurs:
1139 * - The old curr_active_slave has been released or lost its link.
1140 * - The primary_slave has got its link back.
1141 * - A slave has got its link back and there's no old curr_active_slave.
1143 * Warning: Caller must hold curr_slave_lock for writing.
1145 void bond_select_active_slave(struct bonding *bond)
1147 struct slave *best_slave;
1148 int rv;
1150 best_slave = bond_find_best_slave(bond);
1151 if (best_slave != bond->curr_active_slave) {
1152 bond_change_active_slave(bond, best_slave);
1153 rv = bond_set_carrier(bond);
1154 if (!rv)
1155 return;
1157 if (netif_carrier_ok(bond->dev)) {
1158 printk(KERN_INFO DRV_NAME
1159 ": %s: first active interface up!\n",
1160 bond->dev->name);
1161 } else {
1162 printk(KERN_INFO DRV_NAME ": %s: "
1163 "now running without any active interface !\n",
1164 bond->dev->name);
1169 /*--------------------------- slave list handling ---------------------------*/
1172 * This function attaches the slave to the end of list.
1174 * bond->lock held for writing by caller.
1176 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1178 if (bond->first_slave == NULL) { /* attaching the first slave */
1179 new_slave->next = new_slave;
1180 new_slave->prev = new_slave;
1181 bond->first_slave = new_slave;
1182 } else {
1183 new_slave->next = bond->first_slave;
1184 new_slave->prev = bond->first_slave->prev;
1185 new_slave->next->prev = new_slave;
1186 new_slave->prev->next = new_slave;
1189 bond->slave_cnt++;
1193 * This function detaches the slave from the list.
1194 * WARNING: no check is made to verify if the slave effectively
1195 * belongs to <bond>.
1196 * Nothing is freed on return, structures are just unchained.
1197 * If any slave pointer in bond was pointing to <slave>,
1198 * it should be changed by the calling function.
1200 * bond->lock held for writing by caller.
1202 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1204 if (slave->next) {
1205 slave->next->prev = slave->prev;
1208 if (slave->prev) {
1209 slave->prev->next = slave->next;
1212 if (bond->first_slave == slave) { /* slave is the first slave */
1213 if (bond->slave_cnt > 1) { /* there are more slave */
1214 bond->first_slave = slave->next;
1215 } else {
1216 bond->first_slave = NULL; /* slave was the last one */
1220 slave->next = NULL;
1221 slave->prev = NULL;
1222 bond->slave_cnt--;
1225 /*---------------------------------- IOCTL ----------------------------------*/
1227 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1229 dprintk("bond_dev=%p\n", bond_dev);
1230 dprintk("slave_dev=%p\n", slave_dev);
1231 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1232 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1233 return 0;
1236 #define BOND_VLAN_FEATURES \
1237 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1238 NETIF_F_HW_VLAN_FILTER)
1241 * Compute the common dev->feature set available to all slaves. Some
1242 * feature bits are managed elsewhere, so preserve those feature bits
1243 * on the master device.
1245 static int bond_compute_features(struct bonding *bond)
1247 struct slave *slave;
1248 struct net_device *bond_dev = bond->dev;
1249 unsigned long features = bond_dev->features & ~BOND_VLAN_FEATURES;
1250 unsigned short max_hard_header_len = ETH_HLEN;
1251 int i;
1253 bond_for_each_slave(bond, slave, i) {
1254 features = netdev_compute_features(features,
1255 slave->dev->features);
1256 if (slave->dev->hard_header_len > max_hard_header_len)
1257 max_hard_header_len = slave->dev->hard_header_len;
1260 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1261 bond_dev->features = features;
1262 bond_dev->hard_header_len = max_hard_header_len;
1264 return 0;
1267 /* enslave device <slave> to bond device <master> */
1268 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1270 struct bonding *bond = bond_dev->priv;
1271 struct slave *new_slave = NULL;
1272 struct dev_mc_list *dmi;
1273 struct sockaddr addr;
1274 int link_reporting;
1275 int old_features = bond_dev->features;
1276 int res = 0;
1278 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1279 slave_dev->do_ioctl == NULL) {
1280 printk(KERN_WARNING DRV_NAME
1281 ": %s: Warning: no link monitoring support for %s\n",
1282 bond_dev->name, slave_dev->name);
1285 /* bond must be initialized by bond_open() before enslaving */
1286 if (!(bond_dev->flags & IFF_UP)) {
1287 dprintk("Error, master_dev is not up\n");
1288 return -EPERM;
1291 /* already enslaved */
1292 if (slave_dev->flags & IFF_SLAVE) {
1293 dprintk("Error, Device was already enslaved\n");
1294 return -EBUSY;
1297 /* vlan challenged mutual exclusion */
1298 /* no need to lock since we're protected by rtnl_lock */
1299 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1300 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1301 if (!list_empty(&bond->vlan_list)) {
1302 printk(KERN_ERR DRV_NAME
1303 ": %s: Error: cannot enslave VLAN "
1304 "challenged slave %s on VLAN enabled "
1305 "bond %s\n", bond_dev->name, slave_dev->name,
1306 bond_dev->name);
1307 return -EPERM;
1308 } else {
1309 printk(KERN_WARNING DRV_NAME
1310 ": %s: Warning: enslaved VLAN challenged "
1311 "slave %s. Adding VLANs will be blocked as "
1312 "long as %s is part of bond %s\n",
1313 bond_dev->name, slave_dev->name, slave_dev->name,
1314 bond_dev->name);
1315 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1317 } else {
1318 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1319 if (bond->slave_cnt == 0) {
1320 /* First slave, and it is not VLAN challenged,
1321 * so remove the block of adding VLANs over the bond.
1323 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1328 * Old ifenslave binaries are no longer supported. These can
1329 * be identified with moderate accurary by the state of the slave:
1330 * the current ifenslave will set the interface down prior to
1331 * enslaving it; the old ifenslave will not.
1333 if ((slave_dev->flags & IFF_UP)) {
1334 printk(KERN_ERR DRV_NAME ": %s is up. "
1335 "This may be due to an out of date ifenslave.\n",
1336 slave_dev->name);
1337 res = -EPERM;
1338 goto err_undo_flags;
1341 if (slave_dev->set_mac_address == NULL) {
1342 printk(KERN_ERR DRV_NAME
1343 ": %s: Error: The slave device you specified does "
1344 "not support setting the MAC address. "
1345 "Your kernel likely does not support slave "
1346 "devices.\n", bond_dev->name);
1347 res = -EOPNOTSUPP;
1348 goto err_undo_flags;
1351 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1352 if (!new_slave) {
1353 res = -ENOMEM;
1354 goto err_undo_flags;
1357 /* save slave's original flags before calling
1358 * netdev_set_master and dev_open
1360 new_slave->original_flags = slave_dev->flags;
1363 * Save slave's original ("permanent") mac address for modes
1364 * that need it, and for restoring it upon release, and then
1365 * set it to the master's address
1367 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1370 * Set slave to master's mac address. The application already
1371 * set the master's mac address to that of the first slave
1373 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1374 addr.sa_family = slave_dev->type;
1375 res = dev_set_mac_address(slave_dev, &addr);
1376 if (res) {
1377 dprintk("Error %d calling set_mac_address\n", res);
1378 goto err_free;
1381 res = netdev_set_master(slave_dev, bond_dev);
1382 if (res) {
1383 dprintk("Error %d calling netdev_set_master\n", res);
1384 goto err_restore_mac;
1387 /* open the slave since the application closed it */
1388 res = dev_open(slave_dev);
1389 if (res) {
1390 dprintk("Openning slave %s failed\n", slave_dev->name);
1391 goto err_unset_master;
1394 new_slave->dev = slave_dev;
1395 slave_dev->priv_flags |= IFF_BONDING;
1397 if ((bond->params.mode == BOND_MODE_TLB) ||
1398 (bond->params.mode == BOND_MODE_ALB)) {
1399 /* bond_alb_init_slave() must be called before all other stages since
1400 * it might fail and we do not want to have to undo everything
1402 res = bond_alb_init_slave(bond, new_slave);
1403 if (res) {
1404 goto err_close;
1408 /* If the mode USES_PRIMARY, then the new slave gets the
1409 * master's promisc (and mc) settings only if it becomes the
1410 * curr_active_slave, and that is taken care of later when calling
1411 * bond_change_active()
1413 if (!USES_PRIMARY(bond->params.mode)) {
1414 /* set promiscuity level to new slave */
1415 if (bond_dev->flags & IFF_PROMISC) {
1416 dev_set_promiscuity(slave_dev, 1);
1419 /* set allmulti level to new slave */
1420 if (bond_dev->flags & IFF_ALLMULTI) {
1421 dev_set_allmulti(slave_dev, 1);
1424 /* upload master's mc_list to new slave */
1425 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1426 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1430 if (bond->params.mode == BOND_MODE_8023AD) {
1431 /* add lacpdu mc addr to mc list */
1432 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1434 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1437 bond_add_vlans_on_slave(bond, slave_dev);
1439 write_lock_bh(&bond->lock);
1441 bond_attach_slave(bond, new_slave);
1443 new_slave->delay = 0;
1444 new_slave->link_failure_count = 0;
1446 bond_compute_features(bond);
1448 new_slave->last_arp_rx = jiffies;
1450 if (bond->params.miimon && !bond->params.use_carrier) {
1451 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1453 if ((link_reporting == -1) && !bond->params.arp_interval) {
1455 * miimon is set but a bonded network driver
1456 * does not support ETHTOOL/MII and
1457 * arp_interval is not set. Note: if
1458 * use_carrier is enabled, we will never go
1459 * here (because netif_carrier is always
1460 * supported); thus, we don't need to change
1461 * the messages for netif_carrier.
1463 printk(KERN_WARNING DRV_NAME
1464 ": %s: Warning: MII and ETHTOOL support not "
1465 "available for interface %s, and "
1466 "arp_interval/arp_ip_target module parameters "
1467 "not specified, thus bonding will not detect "
1468 "link failures! see bonding.txt for details.\n",
1469 bond_dev->name, slave_dev->name);
1470 } else if (link_reporting == -1) {
1471 /* unable get link status using mii/ethtool */
1472 printk(KERN_WARNING DRV_NAME
1473 ": %s: Warning: can't get link status from "
1474 "interface %s; the network driver associated "
1475 "with this interface does not support MII or "
1476 "ETHTOOL link status reporting, thus miimon "
1477 "has no effect on this interface.\n",
1478 bond_dev->name, slave_dev->name);
1482 /* check for initial state */
1483 if (!bond->params.miimon ||
1484 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1485 if (bond->params.updelay) {
1486 dprintk("Initial state of slave_dev is "
1487 "BOND_LINK_BACK\n");
1488 new_slave->link = BOND_LINK_BACK;
1489 new_slave->delay = bond->params.updelay;
1490 } else {
1491 dprintk("Initial state of slave_dev is "
1492 "BOND_LINK_UP\n");
1493 new_slave->link = BOND_LINK_UP;
1495 new_slave->jiffies = jiffies;
1496 } else {
1497 dprintk("Initial state of slave_dev is "
1498 "BOND_LINK_DOWN\n");
1499 new_slave->link = BOND_LINK_DOWN;
1502 if (bond_update_speed_duplex(new_slave) &&
1503 (new_slave->link != BOND_LINK_DOWN)) {
1504 printk(KERN_WARNING DRV_NAME
1505 ": %s: Warning: failed to get speed and duplex from %s, "
1506 "assumed to be 100Mb/sec and Full.\n",
1507 bond_dev->name, new_slave->dev->name);
1509 if (bond->params.mode == BOND_MODE_8023AD) {
1510 printk(KERN_WARNING DRV_NAME
1511 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1512 "support in base driver for proper aggregator "
1513 "selection.\n", bond_dev->name);
1517 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1518 /* if there is a primary slave, remember it */
1519 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1520 bond->primary_slave = new_slave;
1524 switch (bond->params.mode) {
1525 case BOND_MODE_ACTIVEBACKUP:
1526 bond_set_slave_inactive_flags(new_slave);
1527 bond_select_active_slave(bond);
1528 break;
1529 case BOND_MODE_8023AD:
1530 /* in 802.3ad mode, the internal mechanism
1531 * will activate the slaves in the selected
1532 * aggregator
1534 bond_set_slave_inactive_flags(new_slave);
1535 /* if this is the first slave */
1536 if (bond->slave_cnt == 1) {
1537 SLAVE_AD_INFO(new_slave).id = 1;
1538 /* Initialize AD with the number of times that the AD timer is called in 1 second
1539 * can be called only after the mac address of the bond is set
1541 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1542 bond->params.lacp_fast);
1543 } else {
1544 SLAVE_AD_INFO(new_slave).id =
1545 SLAVE_AD_INFO(new_slave->prev).id + 1;
1548 bond_3ad_bind_slave(new_slave);
1549 break;
1550 case BOND_MODE_TLB:
1551 case BOND_MODE_ALB:
1552 new_slave->state = BOND_STATE_ACTIVE;
1553 if ((!bond->curr_active_slave) &&
1554 (new_slave->link != BOND_LINK_DOWN)) {
1555 /* first slave or no active slave yet, and this link
1556 * is OK, so make this interface the active one
1558 bond_change_active_slave(bond, new_slave);
1559 } else {
1560 bond_set_slave_inactive_flags(new_slave);
1562 break;
1563 default:
1564 dprintk("This slave is always active in trunk mode\n");
1566 /* always active in trunk mode */
1567 new_slave->state = BOND_STATE_ACTIVE;
1569 /* In trunking mode there is little meaning to curr_active_slave
1570 * anyway (it holds no special properties of the bond device),
1571 * so we can change it without calling change_active_interface()
1573 if (!bond->curr_active_slave) {
1574 bond->curr_active_slave = new_slave;
1576 break;
1577 } /* switch(bond_mode) */
1579 bond_set_carrier(bond);
1581 write_unlock_bh(&bond->lock);
1583 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1584 if (res)
1585 goto err_close;
1587 printk(KERN_INFO DRV_NAME
1588 ": %s: enslaving %s as a%s interface with a%s link.\n",
1589 bond_dev->name, slave_dev->name,
1590 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1591 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1593 /* enslave is successful */
1594 return 0;
1596 /* Undo stages on error */
1597 err_close:
1598 dev_close(slave_dev);
1600 err_unset_master:
1601 netdev_set_master(slave_dev, NULL);
1603 err_restore_mac:
1604 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1605 addr.sa_family = slave_dev->type;
1606 dev_set_mac_address(slave_dev, &addr);
1608 err_free:
1609 kfree(new_slave);
1611 err_undo_flags:
1612 bond_dev->features = old_features;
1614 return res;
1618 * Try to release the slave device <slave> from the bond device <master>
1619 * It is legal to access curr_active_slave without a lock because all the function
1620 * is write-locked.
1622 * The rules for slave state should be:
1623 * for Active/Backup:
1624 * Active stays on all backups go down
1625 * for Bonded connections:
1626 * The first up interface should be left on and all others downed.
1628 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1630 struct bonding *bond = bond_dev->priv;
1631 struct slave *slave, *oldcurrent;
1632 struct sockaddr addr;
1633 int mac_addr_differ;
1635 /* slave is not a slave or master is not master of this slave */
1636 if (!(slave_dev->flags & IFF_SLAVE) ||
1637 (slave_dev->master != bond_dev)) {
1638 printk(KERN_ERR DRV_NAME
1639 ": %s: Error: cannot release %s.\n",
1640 bond_dev->name, slave_dev->name);
1641 return -EINVAL;
1644 write_lock_bh(&bond->lock);
1646 slave = bond_get_slave_by_dev(bond, slave_dev);
1647 if (!slave) {
1648 /* not a slave of this bond */
1649 printk(KERN_INFO DRV_NAME
1650 ": %s: %s not enslaved\n",
1651 bond_dev->name, slave_dev->name);
1652 write_unlock_bh(&bond->lock);
1653 return -EINVAL;
1656 mac_addr_differ = memcmp(bond_dev->dev_addr,
1657 slave->perm_hwaddr,
1658 ETH_ALEN);
1659 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1660 printk(KERN_WARNING DRV_NAME
1661 ": %s: Warning: the permanent HWaddr of %s "
1662 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1663 "still in use by %s. Set the HWaddr of "
1664 "%s to a different address to avoid "
1665 "conflicts.\n",
1666 bond_dev->name,
1667 slave_dev->name,
1668 slave->perm_hwaddr[0],
1669 slave->perm_hwaddr[1],
1670 slave->perm_hwaddr[2],
1671 slave->perm_hwaddr[3],
1672 slave->perm_hwaddr[4],
1673 slave->perm_hwaddr[5],
1674 bond_dev->name,
1675 slave_dev->name);
1678 /* Inform AD package of unbinding of slave. */
1679 if (bond->params.mode == BOND_MODE_8023AD) {
1680 /* must be called before the slave is
1681 * detached from the list
1683 bond_3ad_unbind_slave(slave);
1686 printk(KERN_INFO DRV_NAME
1687 ": %s: releasing %s interface %s\n",
1688 bond_dev->name,
1689 (slave->state == BOND_STATE_ACTIVE)
1690 ? "active" : "backup",
1691 slave_dev->name);
1693 oldcurrent = bond->curr_active_slave;
1695 bond->current_arp_slave = NULL;
1697 /* release the slave from its bond */
1698 bond_detach_slave(bond, slave);
1700 bond_compute_features(bond);
1702 if (bond->primary_slave == slave) {
1703 bond->primary_slave = NULL;
1706 if (oldcurrent == slave) {
1707 bond_change_active_slave(bond, NULL);
1710 if ((bond->params.mode == BOND_MODE_TLB) ||
1711 (bond->params.mode == BOND_MODE_ALB)) {
1712 /* Must be called only after the slave has been
1713 * detached from the list and the curr_active_slave
1714 * has been cleared (if our_slave == old_current),
1715 * but before a new active slave is selected.
1717 bond_alb_deinit_slave(bond, slave);
1720 if (oldcurrent == slave)
1721 bond_select_active_slave(bond);
1723 if (bond->slave_cnt == 0) {
1724 bond_set_carrier(bond);
1726 /* if the last slave was removed, zero the mac address
1727 * of the master so it will be set by the application
1728 * to the mac address of the first slave
1730 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1732 if (list_empty(&bond->vlan_list)) {
1733 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1734 } else {
1735 printk(KERN_WARNING DRV_NAME
1736 ": %s: Warning: clearing HW address of %s while it "
1737 "still has VLANs.\n",
1738 bond_dev->name, bond_dev->name);
1739 printk(KERN_WARNING DRV_NAME
1740 ": %s: When re-adding slaves, make sure the bond's "
1741 "HW address matches its VLANs'.\n",
1742 bond_dev->name);
1744 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1745 !bond_has_challenged_slaves(bond)) {
1746 printk(KERN_INFO DRV_NAME
1747 ": %s: last VLAN challenged slave %s "
1748 "left bond %s. VLAN blocking is removed\n",
1749 bond_dev->name, slave_dev->name, bond_dev->name);
1750 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1753 write_unlock_bh(&bond->lock);
1755 /* must do this from outside any spinlocks */
1756 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1758 bond_del_vlans_from_slave(bond, slave_dev);
1760 /* If the mode USES_PRIMARY, then we should only remove its
1761 * promisc and mc settings if it was the curr_active_slave, but that was
1762 * already taken care of above when we detached the slave
1764 if (!USES_PRIMARY(bond->params.mode)) {
1765 /* unset promiscuity level from slave */
1766 if (bond_dev->flags & IFF_PROMISC) {
1767 dev_set_promiscuity(slave_dev, -1);
1770 /* unset allmulti level from slave */
1771 if (bond_dev->flags & IFF_ALLMULTI) {
1772 dev_set_allmulti(slave_dev, -1);
1775 /* flush master's mc_list from slave */
1776 bond_mc_list_flush(bond_dev, slave_dev);
1779 netdev_set_master(slave_dev, NULL);
1781 /* close slave before restoring its mac address */
1782 dev_close(slave_dev);
1784 /* restore original ("permanent") mac address */
1785 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1786 addr.sa_family = slave_dev->type;
1787 dev_set_mac_address(slave_dev, &addr);
1789 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1790 IFF_SLAVE_INACTIVE | IFF_BONDING |
1791 IFF_SLAVE_NEEDARP);
1793 kfree(slave);
1795 return 0; /* deletion OK */
1799 * This function releases all slaves.
1801 static int bond_release_all(struct net_device *bond_dev)
1803 struct bonding *bond = bond_dev->priv;
1804 struct slave *slave;
1805 struct net_device *slave_dev;
1806 struct sockaddr addr;
1808 write_lock_bh(&bond->lock);
1810 netif_carrier_off(bond_dev);
1812 if (bond->slave_cnt == 0) {
1813 goto out;
1816 bond->current_arp_slave = NULL;
1817 bond->primary_slave = NULL;
1818 bond_change_active_slave(bond, NULL);
1820 while ((slave = bond->first_slave) != NULL) {
1821 /* Inform AD package of unbinding of slave
1822 * before slave is detached from the list.
1824 if (bond->params.mode == BOND_MODE_8023AD) {
1825 bond_3ad_unbind_slave(slave);
1828 slave_dev = slave->dev;
1829 bond_detach_slave(bond, slave);
1831 if ((bond->params.mode == BOND_MODE_TLB) ||
1832 (bond->params.mode == BOND_MODE_ALB)) {
1833 /* must be called only after the slave
1834 * has been detached from the list
1836 bond_alb_deinit_slave(bond, slave);
1839 bond_compute_features(bond);
1841 /* now that the slave is detached, unlock and perform
1842 * all the undo steps that should not be called from
1843 * within a lock.
1845 write_unlock_bh(&bond->lock);
1847 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1848 bond_del_vlans_from_slave(bond, slave_dev);
1850 /* If the mode USES_PRIMARY, then we should only remove its
1851 * promisc and mc settings if it was the curr_active_slave, but that was
1852 * already taken care of above when we detached the slave
1854 if (!USES_PRIMARY(bond->params.mode)) {
1855 /* unset promiscuity level from slave */
1856 if (bond_dev->flags & IFF_PROMISC) {
1857 dev_set_promiscuity(slave_dev, -1);
1860 /* unset allmulti level from slave */
1861 if (bond_dev->flags & IFF_ALLMULTI) {
1862 dev_set_allmulti(slave_dev, -1);
1865 /* flush master's mc_list from slave */
1866 bond_mc_list_flush(bond_dev, slave_dev);
1869 netdev_set_master(slave_dev, NULL);
1871 /* close slave before restoring its mac address */
1872 dev_close(slave_dev);
1874 /* restore original ("permanent") mac address*/
1875 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1876 addr.sa_family = slave_dev->type;
1877 dev_set_mac_address(slave_dev, &addr);
1879 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1880 IFF_SLAVE_INACTIVE);
1882 kfree(slave);
1884 /* re-acquire the lock before getting the next slave */
1885 write_lock_bh(&bond->lock);
1888 /* zero the mac address of the master so it will be
1889 * set by the application to the mac address of the
1890 * first slave
1892 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1894 if (list_empty(&bond->vlan_list)) {
1895 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1896 } else {
1897 printk(KERN_WARNING DRV_NAME
1898 ": %s: Warning: clearing HW address of %s while it "
1899 "still has VLANs.\n",
1900 bond_dev->name, bond_dev->name);
1901 printk(KERN_WARNING DRV_NAME
1902 ": %s: When re-adding slaves, make sure the bond's "
1903 "HW address matches its VLANs'.\n",
1904 bond_dev->name);
1907 printk(KERN_INFO DRV_NAME
1908 ": %s: released all slaves\n",
1909 bond_dev->name);
1911 out:
1912 write_unlock_bh(&bond->lock);
1914 return 0;
1918 * This function changes the active slave to slave <slave_dev>.
1919 * It returns -EINVAL in the following cases.
1920 * - <slave_dev> is not found in the list.
1921 * - There is not active slave now.
1922 * - <slave_dev> is already active.
1923 * - The link state of <slave_dev> is not BOND_LINK_UP.
1924 * - <slave_dev> is not running.
1925 * In these cases, this fuction does nothing.
1926 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1928 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1930 struct bonding *bond = bond_dev->priv;
1931 struct slave *old_active = NULL;
1932 struct slave *new_active = NULL;
1933 int res = 0;
1935 if (!USES_PRIMARY(bond->params.mode)) {
1936 return -EINVAL;
1939 /* Verify that master_dev is indeed the master of slave_dev */
1940 if (!(slave_dev->flags & IFF_SLAVE) ||
1941 (slave_dev->master != bond_dev)) {
1942 return -EINVAL;
1945 write_lock_bh(&bond->lock);
1947 old_active = bond->curr_active_slave;
1948 new_active = bond_get_slave_by_dev(bond, slave_dev);
1951 * Changing to the current active: do nothing; return success.
1953 if (new_active && (new_active == old_active)) {
1954 write_unlock_bh(&bond->lock);
1955 return 0;
1958 if ((new_active) &&
1959 (old_active) &&
1960 (new_active->link == BOND_LINK_UP) &&
1961 IS_UP(new_active->dev)) {
1962 bond_change_active_slave(bond, new_active);
1963 } else {
1964 res = -EINVAL;
1967 write_unlock_bh(&bond->lock);
1969 return res;
1972 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1974 struct bonding *bond = bond_dev->priv;
1976 info->bond_mode = bond->params.mode;
1977 info->miimon = bond->params.miimon;
1979 read_lock_bh(&bond->lock);
1980 info->num_slaves = bond->slave_cnt;
1981 read_unlock_bh(&bond->lock);
1983 return 0;
1986 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1988 struct bonding *bond = bond_dev->priv;
1989 struct slave *slave;
1990 int i, found = 0;
1992 if (info->slave_id < 0) {
1993 return -ENODEV;
1996 read_lock_bh(&bond->lock);
1998 bond_for_each_slave(bond, slave, i) {
1999 if (i == (int)info->slave_id) {
2000 found = 1;
2001 break;
2005 read_unlock_bh(&bond->lock);
2007 if (found) {
2008 strcpy(info->slave_name, slave->dev->name);
2009 info->link = slave->link;
2010 info->state = slave->state;
2011 info->link_failure_count = slave->link_failure_count;
2012 } else {
2013 return -ENODEV;
2016 return 0;
2019 /*-------------------------------- Monitoring -------------------------------*/
2021 /* this function is called regularly to monitor each slave's link. */
2022 void bond_mii_monitor(struct net_device *bond_dev)
2024 struct bonding *bond = bond_dev->priv;
2025 struct slave *slave, *oldcurrent;
2026 int do_failover = 0;
2027 int delta_in_ticks;
2028 int i;
2030 read_lock(&bond->lock);
2032 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2034 if (bond->kill_timers) {
2035 goto out;
2038 if (bond->slave_cnt == 0) {
2039 goto re_arm;
2042 /* we will try to read the link status of each of our slaves, and
2043 * set their IFF_RUNNING flag appropriately. For each slave not
2044 * supporting MII status, we won't do anything so that a user-space
2045 * program could monitor the link itself if needed.
2048 read_lock(&bond->curr_slave_lock);
2049 oldcurrent = bond->curr_active_slave;
2050 read_unlock(&bond->curr_slave_lock);
2052 bond_for_each_slave(bond, slave, i) {
2053 struct net_device *slave_dev = slave->dev;
2054 int link_state;
2055 u16 old_speed = slave->speed;
2056 u8 old_duplex = slave->duplex;
2058 link_state = bond_check_dev_link(bond, slave_dev, 0);
2060 switch (slave->link) {
2061 case BOND_LINK_UP: /* the link was up */
2062 if (link_state == BMSR_LSTATUS) {
2063 /* link stays up, nothing more to do */
2064 break;
2065 } else { /* link going down */
2066 slave->link = BOND_LINK_FAIL;
2067 slave->delay = bond->params.downdelay;
2069 if (slave->link_failure_count < UINT_MAX) {
2070 slave->link_failure_count++;
2073 if (bond->params.downdelay) {
2074 printk(KERN_INFO DRV_NAME
2075 ": %s: link status down for %s "
2076 "interface %s, disabling it in "
2077 "%d ms.\n",
2078 bond_dev->name,
2079 IS_UP(slave_dev)
2080 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2081 ? ((slave == oldcurrent)
2082 ? "active " : "backup ")
2083 : "")
2084 : "idle ",
2085 slave_dev->name,
2086 bond->params.downdelay * bond->params.miimon);
2089 /* no break ! fall through the BOND_LINK_FAIL test to
2090 ensure proper action to be taken
2092 case BOND_LINK_FAIL: /* the link has just gone down */
2093 if (link_state != BMSR_LSTATUS) {
2094 /* link stays down */
2095 if (slave->delay <= 0) {
2096 /* link down for too long time */
2097 slave->link = BOND_LINK_DOWN;
2099 /* in active/backup mode, we must
2100 * completely disable this interface
2102 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2103 (bond->params.mode == BOND_MODE_8023AD)) {
2104 bond_set_slave_inactive_flags(slave);
2107 printk(KERN_INFO DRV_NAME
2108 ": %s: link status definitely "
2109 "down for interface %s, "
2110 "disabling it\n",
2111 bond_dev->name,
2112 slave_dev->name);
2114 /* notify ad that the link status has changed */
2115 if (bond->params.mode == BOND_MODE_8023AD) {
2116 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2119 if ((bond->params.mode == BOND_MODE_TLB) ||
2120 (bond->params.mode == BOND_MODE_ALB)) {
2121 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2124 if (slave == oldcurrent) {
2125 do_failover = 1;
2127 } else {
2128 slave->delay--;
2130 } else {
2131 /* link up again */
2132 slave->link = BOND_LINK_UP;
2133 slave->jiffies = jiffies;
2134 printk(KERN_INFO DRV_NAME
2135 ": %s: link status up again after %d "
2136 "ms for interface %s.\n",
2137 bond_dev->name,
2138 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2139 slave_dev->name);
2141 break;
2142 case BOND_LINK_DOWN: /* the link was down */
2143 if (link_state != BMSR_LSTATUS) {
2144 /* the link stays down, nothing more to do */
2145 break;
2146 } else { /* link going up */
2147 slave->link = BOND_LINK_BACK;
2148 slave->delay = bond->params.updelay;
2150 if (bond->params.updelay) {
2151 /* if updelay == 0, no need to
2152 advertise about a 0 ms delay */
2153 printk(KERN_INFO DRV_NAME
2154 ": %s: link status up for "
2155 "interface %s, enabling it "
2156 "in %d ms.\n",
2157 bond_dev->name,
2158 slave_dev->name,
2159 bond->params.updelay * bond->params.miimon);
2162 /* no break ! fall through the BOND_LINK_BACK state in
2163 case there's something to do.
2165 case BOND_LINK_BACK: /* the link has just come back */
2166 if (link_state != BMSR_LSTATUS) {
2167 /* link down again */
2168 slave->link = BOND_LINK_DOWN;
2170 printk(KERN_INFO DRV_NAME
2171 ": %s: link status down again after %d "
2172 "ms for interface %s.\n",
2173 bond_dev->name,
2174 (bond->params.updelay - slave->delay) * bond->params.miimon,
2175 slave_dev->name);
2176 } else {
2177 /* link stays up */
2178 if (slave->delay == 0) {
2179 /* now the link has been up for long time enough */
2180 slave->link = BOND_LINK_UP;
2181 slave->jiffies = jiffies;
2183 if (bond->params.mode == BOND_MODE_8023AD) {
2184 /* prevent it from being the active one */
2185 slave->state = BOND_STATE_BACKUP;
2186 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2187 /* make it immediately active */
2188 slave->state = BOND_STATE_ACTIVE;
2189 } else if (slave != bond->primary_slave) {
2190 /* prevent it from being the active one */
2191 slave->state = BOND_STATE_BACKUP;
2194 printk(KERN_INFO DRV_NAME
2195 ": %s: link status definitely "
2196 "up for interface %s.\n",
2197 bond_dev->name,
2198 slave_dev->name);
2200 /* notify ad that the link status has changed */
2201 if (bond->params.mode == BOND_MODE_8023AD) {
2202 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2205 if ((bond->params.mode == BOND_MODE_TLB) ||
2206 (bond->params.mode == BOND_MODE_ALB)) {
2207 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2210 if ((!oldcurrent) ||
2211 (slave == bond->primary_slave)) {
2212 do_failover = 1;
2214 } else {
2215 slave->delay--;
2218 break;
2219 default:
2220 /* Should not happen */
2221 printk(KERN_ERR DRV_NAME
2222 ": %s: Error: %s Illegal value (link=%d)\n",
2223 bond_dev->name,
2224 slave->dev->name,
2225 slave->link);
2226 goto out;
2227 } /* end of switch (slave->link) */
2229 bond_update_speed_duplex(slave);
2231 if (bond->params.mode == BOND_MODE_8023AD) {
2232 if (old_speed != slave->speed) {
2233 bond_3ad_adapter_speed_changed(slave);
2236 if (old_duplex != slave->duplex) {
2237 bond_3ad_adapter_duplex_changed(slave);
2241 } /* end of for */
2243 if (do_failover) {
2244 write_lock(&bond->curr_slave_lock);
2246 bond_select_active_slave(bond);
2248 write_unlock(&bond->curr_slave_lock);
2249 } else
2250 bond_set_carrier(bond);
2252 re_arm:
2253 if (bond->params.miimon) {
2254 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2256 out:
2257 read_unlock(&bond->lock);
2261 static u32 bond_glean_dev_ip(struct net_device *dev)
2263 struct in_device *idev;
2264 struct in_ifaddr *ifa;
2265 __be32 addr = 0;
2267 if (!dev)
2268 return 0;
2270 rcu_read_lock();
2271 idev = __in_dev_get_rcu(dev);
2272 if (!idev)
2273 goto out;
2275 ifa = idev->ifa_list;
2276 if (!ifa)
2277 goto out;
2279 addr = ifa->ifa_local;
2280 out:
2281 rcu_read_unlock();
2282 return addr;
2285 static int bond_has_ip(struct bonding *bond)
2287 struct vlan_entry *vlan, *vlan_next;
2289 if (bond->master_ip)
2290 return 1;
2292 if (list_empty(&bond->vlan_list))
2293 return 0;
2295 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2296 vlan_list) {
2297 if (vlan->vlan_ip)
2298 return 1;
2301 return 0;
2304 static int bond_has_this_ip(struct bonding *bond, u32 ip)
2306 struct vlan_entry *vlan, *vlan_next;
2308 if (ip == bond->master_ip)
2309 return 1;
2311 if (list_empty(&bond->vlan_list))
2312 return 0;
2314 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2315 vlan_list) {
2316 if (ip == vlan->vlan_ip)
2317 return 1;
2320 return 0;
2324 * We go to the (large) trouble of VLAN tagging ARP frames because
2325 * switches in VLAN mode (especially if ports are configured as
2326 * "native" to a VLAN) might not pass non-tagged frames.
2328 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2330 struct sk_buff *skb;
2332 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2333 slave_dev->name, dest_ip, src_ip, vlan_id);
2335 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2336 NULL, slave_dev->dev_addr, NULL);
2338 if (!skb) {
2339 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2340 return;
2342 if (vlan_id) {
2343 skb = vlan_put_tag(skb, vlan_id);
2344 if (!skb) {
2345 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2346 return;
2349 arp_xmit(skb);
2353 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2355 int i, vlan_id, rv;
2356 u32 *targets = bond->params.arp_targets;
2357 struct vlan_entry *vlan, *vlan_next;
2358 struct net_device *vlan_dev;
2359 struct flowi fl;
2360 struct rtable *rt;
2362 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2363 if (!targets[i])
2364 continue;
2365 dprintk("basa: target %x\n", targets[i]);
2366 if (list_empty(&bond->vlan_list)) {
2367 dprintk("basa: empty vlan: arp_send\n");
2368 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2369 bond->master_ip, 0);
2370 continue;
2374 * If VLANs are configured, we do a route lookup to
2375 * determine which VLAN interface would be used, so we
2376 * can tag the ARP with the proper VLAN tag.
2378 memset(&fl, 0, sizeof(fl));
2379 fl.fl4_dst = targets[i];
2380 fl.fl4_tos = RTO_ONLINK;
2382 rv = ip_route_output_key(&rt, &fl);
2383 if (rv) {
2384 if (net_ratelimit()) {
2385 printk(KERN_WARNING DRV_NAME
2386 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2387 bond->dev->name, NIPQUAD(fl.fl4_dst));
2389 continue;
2393 * This target is not on a VLAN
2395 if (rt->u.dst.dev == bond->dev) {
2396 ip_rt_put(rt);
2397 dprintk("basa: rtdev == bond->dev: arp_send\n");
2398 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2399 bond->master_ip, 0);
2400 continue;
2403 vlan_id = 0;
2404 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2405 vlan_list) {
2406 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2407 if (vlan_dev == rt->u.dst.dev) {
2408 vlan_id = vlan->vlan_id;
2409 dprintk("basa: vlan match on %s %d\n",
2410 vlan_dev->name, vlan_id);
2411 break;
2415 if (vlan_id) {
2416 ip_rt_put(rt);
2417 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2418 vlan->vlan_ip, vlan_id);
2419 continue;
2422 if (net_ratelimit()) {
2423 printk(KERN_WARNING DRV_NAME
2424 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2425 bond->dev->name, NIPQUAD(fl.fl4_dst),
2426 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2428 ip_rt_put(rt);
2433 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2434 * for each VLAN above us.
2436 static void bond_send_gratuitous_arp(struct bonding *bond)
2438 struct slave *slave = bond->curr_active_slave;
2439 struct vlan_entry *vlan;
2440 struct net_device *vlan_dev;
2442 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2443 slave ? slave->dev->name : "NULL");
2444 if (!slave)
2445 return;
2447 if (bond->master_ip) {
2448 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2449 bond->master_ip, 0);
2452 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2453 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2454 if (vlan->vlan_ip) {
2455 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2456 vlan->vlan_ip, vlan->vlan_id);
2461 static void bond_validate_arp(struct bonding *bond, struct slave *slave, u32 sip, u32 tip)
2463 int i;
2464 u32 *targets = bond->params.arp_targets;
2466 targets = bond->params.arp_targets;
2467 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2468 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2469 "%u.%u.%u.%u bhti(tip) %d\n",
2470 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2471 bond_has_this_ip(bond, tip));
2472 if (sip == targets[i]) {
2473 if (bond_has_this_ip(bond, tip))
2474 slave->last_arp_rx = jiffies;
2475 return;
2480 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2482 struct arphdr *arp;
2483 struct slave *slave;
2484 struct bonding *bond;
2485 unsigned char *arp_ptr;
2486 u32 sip, tip;
2488 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2489 goto out;
2491 bond = dev->priv;
2492 read_lock(&bond->lock);
2494 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2495 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2496 orig_dev ? orig_dev->name : "NULL");
2498 slave = bond_get_slave_by_dev(bond, orig_dev);
2499 if (!slave || !slave_do_arp_validate(bond, slave))
2500 goto out_unlock;
2502 /* ARP header, plus 2 device addresses, plus 2 IP addresses. */
2503 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2504 (2 * dev->addr_len) +
2505 (2 * sizeof(u32)))))
2506 goto out_unlock;
2508 arp = arp_hdr(skb);
2509 if (arp->ar_hln != dev->addr_len ||
2510 skb->pkt_type == PACKET_OTHERHOST ||
2511 skb->pkt_type == PACKET_LOOPBACK ||
2512 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2513 arp->ar_pro != htons(ETH_P_IP) ||
2514 arp->ar_pln != 4)
2515 goto out_unlock;
2517 arp_ptr = (unsigned char *)(arp + 1);
2518 arp_ptr += dev->addr_len;
2519 memcpy(&sip, arp_ptr, 4);
2520 arp_ptr += 4 + dev->addr_len;
2521 memcpy(&tip, arp_ptr, 4);
2523 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2524 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2525 slave->state, bond->params.arp_validate,
2526 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2529 * Backup slaves won't see the ARP reply, but do come through
2530 * here for each ARP probe (so we swap the sip/tip to validate
2531 * the probe). In a "redundant switch, common router" type of
2532 * configuration, the ARP probe will (hopefully) travel from
2533 * the active, through one switch, the router, then the other
2534 * switch before reaching the backup.
2536 if (slave->state == BOND_STATE_ACTIVE)
2537 bond_validate_arp(bond, slave, sip, tip);
2538 else
2539 bond_validate_arp(bond, slave, tip, sip);
2541 out_unlock:
2542 read_unlock(&bond->lock);
2543 out:
2544 dev_kfree_skb(skb);
2545 return NET_RX_SUCCESS;
2549 * this function is called regularly to monitor each slave's link
2550 * ensuring that traffic is being sent and received when arp monitoring
2551 * is used in load-balancing mode. if the adapter has been dormant, then an
2552 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2553 * arp monitoring in active backup mode.
2555 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2557 struct bonding *bond = bond_dev->priv;
2558 struct slave *slave, *oldcurrent;
2559 int do_failover = 0;
2560 int delta_in_ticks;
2561 int i;
2563 read_lock(&bond->lock);
2565 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2567 if (bond->kill_timers) {
2568 goto out;
2571 if (bond->slave_cnt == 0) {
2572 goto re_arm;
2575 read_lock(&bond->curr_slave_lock);
2576 oldcurrent = bond->curr_active_slave;
2577 read_unlock(&bond->curr_slave_lock);
2579 /* see if any of the previous devices are up now (i.e. they have
2580 * xmt and rcv traffic). the curr_active_slave does not come into
2581 * the picture unless it is null. also, slave->jiffies is not needed
2582 * here because we send an arp on each slave and give a slave as
2583 * long as it needs to get the tx/rx within the delta.
2584 * TODO: what about up/down delay in arp mode? it wasn't here before
2585 * so it can wait
2587 bond_for_each_slave(bond, slave, i) {
2588 if (slave->link != BOND_LINK_UP) {
2589 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2590 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2592 slave->link = BOND_LINK_UP;
2593 slave->state = BOND_STATE_ACTIVE;
2595 /* primary_slave has no meaning in round-robin
2596 * mode. the window of a slave being up and
2597 * curr_active_slave being null after enslaving
2598 * is closed.
2600 if (!oldcurrent) {
2601 printk(KERN_INFO DRV_NAME
2602 ": %s: link status definitely "
2603 "up for interface %s, ",
2604 bond_dev->name,
2605 slave->dev->name);
2606 do_failover = 1;
2607 } else {
2608 printk(KERN_INFO DRV_NAME
2609 ": %s: interface %s is now up\n",
2610 bond_dev->name,
2611 slave->dev->name);
2614 } else {
2615 /* slave->link == BOND_LINK_UP */
2617 /* not all switches will respond to an arp request
2618 * when the source ip is 0, so don't take the link down
2619 * if we don't know our ip yet
2621 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2622 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2623 bond_has_ip(bond))) {
2625 slave->link = BOND_LINK_DOWN;
2626 slave->state = BOND_STATE_BACKUP;
2628 if (slave->link_failure_count < UINT_MAX) {
2629 slave->link_failure_count++;
2632 printk(KERN_INFO DRV_NAME
2633 ": %s: interface %s is now down.\n",
2634 bond_dev->name,
2635 slave->dev->name);
2637 if (slave == oldcurrent) {
2638 do_failover = 1;
2643 /* note: if switch is in round-robin mode, all links
2644 * must tx arp to ensure all links rx an arp - otherwise
2645 * links may oscillate or not come up at all; if switch is
2646 * in something like xor mode, there is nothing we can
2647 * do - all replies will be rx'ed on same link causing slaves
2648 * to be unstable during low/no traffic periods
2650 if (IS_UP(slave->dev)) {
2651 bond_arp_send_all(bond, slave);
2655 if (do_failover) {
2656 write_lock(&bond->curr_slave_lock);
2658 bond_select_active_slave(bond);
2660 write_unlock(&bond->curr_slave_lock);
2663 re_arm:
2664 if (bond->params.arp_interval) {
2665 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2667 out:
2668 read_unlock(&bond->lock);
2672 * When using arp monitoring in active-backup mode, this function is
2673 * called to determine if any backup slaves have went down or a new
2674 * current slave needs to be found.
2675 * The backup slaves never generate traffic, they are considered up by merely
2676 * receiving traffic. If the current slave goes down, each backup slave will
2677 * be given the opportunity to tx/rx an arp before being taken down - this
2678 * prevents all slaves from being taken down due to the current slave not
2679 * sending any traffic for the backups to receive. The arps are not necessarily
2680 * necessary, any tx and rx traffic will keep the current slave up. While any
2681 * rx traffic will keep the backup slaves up, the current slave is responsible
2682 * for generating traffic to keep them up regardless of any other traffic they
2683 * may have received.
2684 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2686 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2688 struct bonding *bond = bond_dev->priv;
2689 struct slave *slave;
2690 int delta_in_ticks;
2691 int i;
2693 read_lock(&bond->lock);
2695 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2697 if (bond->kill_timers) {
2698 goto out;
2701 if (bond->slave_cnt == 0) {
2702 goto re_arm;
2705 /* determine if any slave has come up or any backup slave has
2706 * gone down
2707 * TODO: what about up/down delay in arp mode? it wasn't here before
2708 * so it can wait
2710 bond_for_each_slave(bond, slave, i) {
2711 if (slave->link != BOND_LINK_UP) {
2712 if ((jiffies - slave_last_rx(bond, slave)) <=
2713 delta_in_ticks) {
2715 slave->link = BOND_LINK_UP;
2717 write_lock(&bond->curr_slave_lock);
2719 if ((!bond->curr_active_slave) &&
2720 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2721 bond_change_active_slave(bond, slave);
2722 bond->current_arp_slave = NULL;
2723 } else if (bond->curr_active_slave != slave) {
2724 /* this slave has just come up but we
2725 * already have a current slave; this
2726 * can also happen if bond_enslave adds
2727 * a new slave that is up while we are
2728 * searching for a new slave
2730 bond_set_slave_inactive_flags(slave);
2731 bond->current_arp_slave = NULL;
2734 bond_set_carrier(bond);
2736 if (slave == bond->curr_active_slave) {
2737 printk(KERN_INFO DRV_NAME
2738 ": %s: %s is up and now the "
2739 "active interface\n",
2740 bond_dev->name,
2741 slave->dev->name);
2742 netif_carrier_on(bond->dev);
2743 } else {
2744 printk(KERN_INFO DRV_NAME
2745 ": %s: backup interface %s is "
2746 "now up\n",
2747 bond_dev->name,
2748 slave->dev->name);
2751 write_unlock(&bond->curr_slave_lock);
2753 } else {
2754 read_lock(&bond->curr_slave_lock);
2756 if ((slave != bond->curr_active_slave) &&
2757 (!bond->current_arp_slave) &&
2758 (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2759 bond_has_ip(bond))) {
2760 /* a backup slave has gone down; three times
2761 * the delta allows the current slave to be
2762 * taken out before the backup slave.
2763 * note: a non-null current_arp_slave indicates
2764 * the curr_active_slave went down and we are
2765 * searching for a new one; under this
2766 * condition we only take the curr_active_slave
2767 * down - this gives each slave a chance to
2768 * tx/rx traffic before being taken out
2771 read_unlock(&bond->curr_slave_lock);
2773 slave->link = BOND_LINK_DOWN;
2775 if (slave->link_failure_count < UINT_MAX) {
2776 slave->link_failure_count++;
2779 bond_set_slave_inactive_flags(slave);
2781 printk(KERN_INFO DRV_NAME
2782 ": %s: backup interface %s is now down\n",
2783 bond_dev->name,
2784 slave->dev->name);
2785 } else {
2786 read_unlock(&bond->curr_slave_lock);
2791 read_lock(&bond->curr_slave_lock);
2792 slave = bond->curr_active_slave;
2793 read_unlock(&bond->curr_slave_lock);
2795 if (slave) {
2796 /* if we have sent traffic in the past 2*arp_intervals but
2797 * haven't xmit and rx traffic in that time interval, select
2798 * a different slave. slave->jiffies is only updated when
2799 * a slave first becomes the curr_active_slave - not necessarily
2800 * after every arp; this ensures the slave has a full 2*delta
2801 * before being taken out. if a primary is being used, check
2802 * if it is up and needs to take over as the curr_active_slave
2804 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2805 (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2806 bond_has_ip(bond))) &&
2807 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2809 slave->link = BOND_LINK_DOWN;
2811 if (slave->link_failure_count < UINT_MAX) {
2812 slave->link_failure_count++;
2815 printk(KERN_INFO DRV_NAME
2816 ": %s: link status down for active interface "
2817 "%s, disabling it\n",
2818 bond_dev->name,
2819 slave->dev->name);
2821 write_lock(&bond->curr_slave_lock);
2823 bond_select_active_slave(bond);
2824 slave = bond->curr_active_slave;
2826 write_unlock(&bond->curr_slave_lock);
2828 bond->current_arp_slave = slave;
2830 if (slave) {
2831 slave->jiffies = jiffies;
2833 } else if ((bond->primary_slave) &&
2834 (bond->primary_slave != slave) &&
2835 (bond->primary_slave->link == BOND_LINK_UP)) {
2836 /* at this point, slave is the curr_active_slave */
2837 printk(KERN_INFO DRV_NAME
2838 ": %s: changing from interface %s to primary "
2839 "interface %s\n",
2840 bond_dev->name,
2841 slave->dev->name,
2842 bond->primary_slave->dev->name);
2844 /* primary is up so switch to it */
2845 write_lock(&bond->curr_slave_lock);
2846 bond_change_active_slave(bond, bond->primary_slave);
2847 write_unlock(&bond->curr_slave_lock);
2849 slave = bond->primary_slave;
2850 slave->jiffies = jiffies;
2851 } else {
2852 bond->current_arp_slave = NULL;
2855 /* the current slave must tx an arp to ensure backup slaves
2856 * rx traffic
2858 if (slave && bond_has_ip(bond)) {
2859 bond_arp_send_all(bond, slave);
2863 /* if we don't have a curr_active_slave, search for the next available
2864 * backup slave from the current_arp_slave and make it the candidate
2865 * for becoming the curr_active_slave
2867 if (!slave) {
2868 if (!bond->current_arp_slave) {
2869 bond->current_arp_slave = bond->first_slave;
2872 if (bond->current_arp_slave) {
2873 bond_set_slave_inactive_flags(bond->current_arp_slave);
2875 /* search for next candidate */
2876 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2877 if (IS_UP(slave->dev)) {
2878 slave->link = BOND_LINK_BACK;
2879 bond_set_slave_active_flags(slave);
2880 bond_arp_send_all(bond, slave);
2881 slave->jiffies = jiffies;
2882 bond->current_arp_slave = slave;
2883 break;
2886 /* if the link state is up at this point, we
2887 * mark it down - this can happen if we have
2888 * simultaneous link failures and
2889 * reselect_active_interface doesn't make this
2890 * one the current slave so it is still marked
2891 * up when it is actually down
2893 if (slave->link == BOND_LINK_UP) {
2894 slave->link = BOND_LINK_DOWN;
2895 if (slave->link_failure_count < UINT_MAX) {
2896 slave->link_failure_count++;
2899 bond_set_slave_inactive_flags(slave);
2901 printk(KERN_INFO DRV_NAME
2902 ": %s: backup interface %s is "
2903 "now down.\n",
2904 bond_dev->name,
2905 slave->dev->name);
2911 re_arm:
2912 if (bond->params.arp_interval) {
2913 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2915 out:
2916 read_unlock(&bond->lock);
2919 /*------------------------------ proc/seq_file-------------------------------*/
2921 #ifdef CONFIG_PROC_FS
2923 #define SEQ_START_TOKEN ((void *)1)
2925 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2927 struct bonding *bond = seq->private;
2928 loff_t off = 0;
2929 struct slave *slave;
2930 int i;
2932 /* make sure the bond won't be taken away */
2933 read_lock(&dev_base_lock);
2934 read_lock_bh(&bond->lock);
2936 if (*pos == 0) {
2937 return SEQ_START_TOKEN;
2940 bond_for_each_slave(bond, slave, i) {
2941 if (++off == *pos) {
2942 return slave;
2946 return NULL;
2949 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2951 struct bonding *bond = seq->private;
2952 struct slave *slave = v;
2954 ++*pos;
2955 if (v == SEQ_START_TOKEN) {
2956 return bond->first_slave;
2959 slave = slave->next;
2961 return (slave == bond->first_slave) ? NULL : slave;
2964 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2966 struct bonding *bond = seq->private;
2968 read_unlock_bh(&bond->lock);
2969 read_unlock(&dev_base_lock);
2972 static void bond_info_show_master(struct seq_file *seq)
2974 struct bonding *bond = seq->private;
2975 struct slave *curr;
2976 int i;
2977 u32 target;
2979 read_lock(&bond->curr_slave_lock);
2980 curr = bond->curr_active_slave;
2981 read_unlock(&bond->curr_slave_lock);
2983 seq_printf(seq, "Bonding Mode: %s\n",
2984 bond_mode_name(bond->params.mode));
2986 if (bond->params.mode == BOND_MODE_XOR ||
2987 bond->params.mode == BOND_MODE_8023AD) {
2988 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2989 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2990 bond->params.xmit_policy);
2993 if (USES_PRIMARY(bond->params.mode)) {
2994 seq_printf(seq, "Primary Slave: %s\n",
2995 (bond->primary_slave) ?
2996 bond->primary_slave->dev->name : "None");
2998 seq_printf(seq, "Currently Active Slave: %s\n",
2999 (curr) ? curr->dev->name : "None");
3002 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3003 "up" : "down");
3004 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3005 seq_printf(seq, "Up Delay (ms): %d\n",
3006 bond->params.updelay * bond->params.miimon);
3007 seq_printf(seq, "Down Delay (ms): %d\n",
3008 bond->params.downdelay * bond->params.miimon);
3011 /* ARP information */
3012 if(bond->params.arp_interval > 0) {
3013 int printed=0;
3014 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3015 bond->params.arp_interval);
3017 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3019 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3020 if (!bond->params.arp_targets[i])
3021 continue;
3022 if (printed)
3023 seq_printf(seq, ",");
3024 target = ntohl(bond->params.arp_targets[i]);
3025 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3026 printed = 1;
3028 seq_printf(seq, "\n");
3031 if (bond->params.mode == BOND_MODE_8023AD) {
3032 struct ad_info ad_info;
3034 seq_puts(seq, "\n802.3ad info\n");
3035 seq_printf(seq, "LACP rate: %s\n",
3036 (bond->params.lacp_fast) ? "fast" : "slow");
3038 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3039 seq_printf(seq, "bond %s has no active aggregator\n",
3040 bond->dev->name);
3041 } else {
3042 seq_printf(seq, "Active Aggregator Info:\n");
3044 seq_printf(seq, "\tAggregator ID: %d\n",
3045 ad_info.aggregator_id);
3046 seq_printf(seq, "\tNumber of ports: %d\n",
3047 ad_info.ports);
3048 seq_printf(seq, "\tActor Key: %d\n",
3049 ad_info.actor_key);
3050 seq_printf(seq, "\tPartner Key: %d\n",
3051 ad_info.partner_key);
3052 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3053 ad_info.partner_system[0],
3054 ad_info.partner_system[1],
3055 ad_info.partner_system[2],
3056 ad_info.partner_system[3],
3057 ad_info.partner_system[4],
3058 ad_info.partner_system[5]);
3063 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3065 struct bonding *bond = seq->private;
3067 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3068 seq_printf(seq, "MII Status: %s\n",
3069 (slave->link == BOND_LINK_UP) ? "up" : "down");
3070 seq_printf(seq, "Link Failure Count: %u\n",
3071 slave->link_failure_count);
3073 seq_printf(seq,
3074 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3075 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3076 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3077 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
3079 if (bond->params.mode == BOND_MODE_8023AD) {
3080 const struct aggregator *agg
3081 = SLAVE_AD_INFO(slave).port.aggregator;
3083 if (agg) {
3084 seq_printf(seq, "Aggregator ID: %d\n",
3085 agg->aggregator_identifier);
3086 } else {
3087 seq_puts(seq, "Aggregator ID: N/A\n");
3092 static int bond_info_seq_show(struct seq_file *seq, void *v)
3094 if (v == SEQ_START_TOKEN) {
3095 seq_printf(seq, "%s\n", version);
3096 bond_info_show_master(seq);
3097 } else {
3098 bond_info_show_slave(seq, v);
3101 return 0;
3104 static struct seq_operations bond_info_seq_ops = {
3105 .start = bond_info_seq_start,
3106 .next = bond_info_seq_next,
3107 .stop = bond_info_seq_stop,
3108 .show = bond_info_seq_show,
3111 static int bond_info_open(struct inode *inode, struct file *file)
3113 struct seq_file *seq;
3114 struct proc_dir_entry *proc;
3115 int res;
3117 res = seq_open(file, &bond_info_seq_ops);
3118 if (!res) {
3119 /* recover the pointer buried in proc_dir_entry data */
3120 seq = file->private_data;
3121 proc = PDE(inode);
3122 seq->private = proc->data;
3125 return res;
3128 static const struct file_operations bond_info_fops = {
3129 .owner = THIS_MODULE,
3130 .open = bond_info_open,
3131 .read = seq_read,
3132 .llseek = seq_lseek,
3133 .release = seq_release,
3136 static int bond_create_proc_entry(struct bonding *bond)
3138 struct net_device *bond_dev = bond->dev;
3140 if (bond_proc_dir) {
3141 bond->proc_entry = create_proc_entry(bond_dev->name,
3142 S_IRUGO,
3143 bond_proc_dir);
3144 if (bond->proc_entry == NULL) {
3145 printk(KERN_WARNING DRV_NAME
3146 ": Warning: Cannot create /proc/net/%s/%s\n",
3147 DRV_NAME, bond_dev->name);
3148 } else {
3149 bond->proc_entry->data = bond;
3150 bond->proc_entry->proc_fops = &bond_info_fops;
3151 bond->proc_entry->owner = THIS_MODULE;
3152 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3156 return 0;
3159 static void bond_remove_proc_entry(struct bonding *bond)
3161 if (bond_proc_dir && bond->proc_entry) {
3162 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3163 memset(bond->proc_file_name, 0, IFNAMSIZ);
3164 bond->proc_entry = NULL;
3168 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3169 * Caller must hold rtnl_lock.
3171 static void bond_create_proc_dir(void)
3173 int len = strlen(DRV_NAME);
3175 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3176 bond_proc_dir = bond_proc_dir->next) {
3177 if ((bond_proc_dir->namelen == len) &&
3178 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3179 break;
3183 if (!bond_proc_dir) {
3184 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3185 if (bond_proc_dir) {
3186 bond_proc_dir->owner = THIS_MODULE;
3187 } else {
3188 printk(KERN_WARNING DRV_NAME
3189 ": Warning: cannot create /proc/net/%s\n",
3190 DRV_NAME);
3195 /* Destroy the bonding directory under /proc/net, if empty.
3196 * Caller must hold rtnl_lock.
3198 static void bond_destroy_proc_dir(void)
3200 struct proc_dir_entry *de;
3202 if (!bond_proc_dir) {
3203 return;
3206 /* verify that the /proc dir is empty */
3207 for (de = bond_proc_dir->subdir; de; de = de->next) {
3208 /* ignore . and .. */
3209 if (*(de->name) != '.') {
3210 break;
3214 if (de) {
3215 if (bond_proc_dir->owner == THIS_MODULE) {
3216 bond_proc_dir->owner = NULL;
3218 } else {
3219 remove_proc_entry(DRV_NAME, proc_net);
3220 bond_proc_dir = NULL;
3223 #endif /* CONFIG_PROC_FS */
3225 /*-------------------------- netdev event handling --------------------------*/
3228 * Change device name
3230 static int bond_event_changename(struct bonding *bond)
3232 #ifdef CONFIG_PROC_FS
3233 bond_remove_proc_entry(bond);
3234 bond_create_proc_entry(bond);
3235 #endif
3236 down_write(&(bonding_rwsem));
3237 bond_destroy_sysfs_entry(bond);
3238 bond_create_sysfs_entry(bond);
3239 up_write(&(bonding_rwsem));
3240 return NOTIFY_DONE;
3243 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3245 struct bonding *event_bond = bond_dev->priv;
3247 switch (event) {
3248 case NETDEV_CHANGENAME:
3249 return bond_event_changename(event_bond);
3250 case NETDEV_UNREGISTER:
3252 * TODO: remove a bond from the list?
3254 break;
3255 default:
3256 break;
3259 return NOTIFY_DONE;
3262 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3264 struct net_device *bond_dev = slave_dev->master;
3265 struct bonding *bond = bond_dev->priv;
3267 switch (event) {
3268 case NETDEV_UNREGISTER:
3269 if (bond_dev) {
3270 bond_release(bond_dev, slave_dev);
3272 break;
3273 case NETDEV_CHANGE:
3275 * TODO: is this what we get if somebody
3276 * sets up a hierarchical bond, then rmmod's
3277 * one of the slave bonding devices?
3279 break;
3280 case NETDEV_DOWN:
3282 * ... Or is it this?
3284 break;
3285 case NETDEV_CHANGEMTU:
3287 * TODO: Should slaves be allowed to
3288 * independently alter their MTU? For
3289 * an active-backup bond, slaves need
3290 * not be the same type of device, so
3291 * MTUs may vary. For other modes,
3292 * slaves arguably should have the
3293 * same MTUs. To do this, we'd need to
3294 * take over the slave's change_mtu
3295 * function for the duration of their
3296 * servitude.
3298 break;
3299 case NETDEV_CHANGENAME:
3301 * TODO: handle changing the primary's name
3303 break;
3304 case NETDEV_FEAT_CHANGE:
3305 bond_compute_features(bond);
3306 break;
3307 default:
3308 break;
3311 return NOTIFY_DONE;
3315 * bond_netdev_event: handle netdev notifier chain events.
3317 * This function receives events for the netdev chain. The caller (an
3318 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3319 * locks for us to safely manipulate the slave devices (RTNL lock,
3320 * dev_probe_lock).
3322 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3324 struct net_device *event_dev = (struct net_device *)ptr;
3326 dprintk("event_dev: %s, event: %lx\n",
3327 (event_dev ? event_dev->name : "None"),
3328 event);
3330 if (!(event_dev->priv_flags & IFF_BONDING))
3331 return NOTIFY_DONE;
3333 if (event_dev->flags & IFF_MASTER) {
3334 dprintk("IFF_MASTER\n");
3335 return bond_master_netdev_event(event, event_dev);
3338 if (event_dev->flags & IFF_SLAVE) {
3339 dprintk("IFF_SLAVE\n");
3340 return bond_slave_netdev_event(event, event_dev);
3343 return NOTIFY_DONE;
3347 * bond_inetaddr_event: handle inetaddr notifier chain events.
3349 * We keep track of device IPs primarily to use as source addresses in
3350 * ARP monitor probes (rather than spewing out broadcasts all the time).
3352 * We track one IP for the main device (if it has one), plus one per VLAN.
3354 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3356 struct in_ifaddr *ifa = ptr;
3357 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3358 struct bonding *bond, *bond_next;
3359 struct vlan_entry *vlan, *vlan_next;
3361 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3362 if (bond->dev == event_dev) {
3363 switch (event) {
3364 case NETDEV_UP:
3365 bond->master_ip = ifa->ifa_local;
3366 return NOTIFY_OK;
3367 case NETDEV_DOWN:
3368 bond->master_ip = bond_glean_dev_ip(bond->dev);
3369 return NOTIFY_OK;
3370 default:
3371 return NOTIFY_DONE;
3375 if (list_empty(&bond->vlan_list))
3376 continue;
3378 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3379 vlan_list) {
3380 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3381 if (vlan_dev == event_dev) {
3382 switch (event) {
3383 case NETDEV_UP:
3384 vlan->vlan_ip = ifa->ifa_local;
3385 return NOTIFY_OK;
3386 case NETDEV_DOWN:
3387 vlan->vlan_ip =
3388 bond_glean_dev_ip(vlan_dev);
3389 return NOTIFY_OK;
3390 default:
3391 return NOTIFY_DONE;
3396 return NOTIFY_DONE;
3399 static struct notifier_block bond_netdev_notifier = {
3400 .notifier_call = bond_netdev_event,
3403 static struct notifier_block bond_inetaddr_notifier = {
3404 .notifier_call = bond_inetaddr_event,
3407 /*-------------------------- Packet type handling ---------------------------*/
3409 /* register to receive lacpdus on a bond */
3410 static void bond_register_lacpdu(struct bonding *bond)
3412 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3414 /* initialize packet type */
3415 pk_type->type = PKT_TYPE_LACPDU;
3416 pk_type->dev = bond->dev;
3417 pk_type->func = bond_3ad_lacpdu_recv;
3419 dev_add_pack(pk_type);
3422 /* unregister to receive lacpdus on a bond */
3423 static void bond_unregister_lacpdu(struct bonding *bond)
3425 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3428 void bond_register_arp(struct bonding *bond)
3430 struct packet_type *pt = &bond->arp_mon_pt;
3432 if (pt->type)
3433 return;
3435 pt->type = htons(ETH_P_ARP);
3436 pt->dev = bond->dev;
3437 pt->func = bond_arp_rcv;
3438 dev_add_pack(pt);
3441 void bond_unregister_arp(struct bonding *bond)
3443 struct packet_type *pt = &bond->arp_mon_pt;
3445 dev_remove_pack(pt);
3446 pt->type = 0;
3449 /*---------------------------- Hashing Policies -----------------------------*/
3452 * Hash for the output device based upon layer 3 and layer 4 data. If
3453 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3454 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3456 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3457 struct net_device *bond_dev, int count)
3459 struct ethhdr *data = (struct ethhdr *)skb->data;
3460 struct iphdr *iph = ip_hdr(skb);
3461 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3462 int layer4_xor = 0;
3464 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3465 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3466 (iph->protocol == IPPROTO_TCP ||
3467 iph->protocol == IPPROTO_UDP)) {
3468 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3470 return (layer4_xor ^
3471 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3475 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3479 * Hash for the output device based upon layer 2 data
3481 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3482 struct net_device *bond_dev, int count)
3484 struct ethhdr *data = (struct ethhdr *)skb->data;
3486 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3489 /*-------------------------- Device entry points ----------------------------*/
3491 static int bond_open(struct net_device *bond_dev)
3493 struct bonding *bond = bond_dev->priv;
3494 struct timer_list *mii_timer = &bond->mii_timer;
3495 struct timer_list *arp_timer = &bond->arp_timer;
3497 bond->kill_timers = 0;
3499 if ((bond->params.mode == BOND_MODE_TLB) ||
3500 (bond->params.mode == BOND_MODE_ALB)) {
3501 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3503 /* bond_alb_initialize must be called before the timer
3504 * is started.
3506 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3507 /* something went wrong - fail the open operation */
3508 return -1;
3511 init_timer(alb_timer);
3512 alb_timer->expires = jiffies + 1;
3513 alb_timer->data = (unsigned long)bond;
3514 alb_timer->function = (void *)&bond_alb_monitor;
3515 add_timer(alb_timer);
3518 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3519 init_timer(mii_timer);
3520 mii_timer->expires = jiffies + 1;
3521 mii_timer->data = (unsigned long)bond_dev;
3522 mii_timer->function = (void *)&bond_mii_monitor;
3523 add_timer(mii_timer);
3526 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3527 init_timer(arp_timer);
3528 arp_timer->expires = jiffies + 1;
3529 arp_timer->data = (unsigned long)bond_dev;
3530 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3531 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3532 } else {
3533 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3535 if (bond->params.arp_validate)
3536 bond_register_arp(bond);
3538 add_timer(arp_timer);
3541 if (bond->params.mode == BOND_MODE_8023AD) {
3542 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3543 init_timer(ad_timer);
3544 ad_timer->expires = jiffies + 1;
3545 ad_timer->data = (unsigned long)bond;
3546 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3547 add_timer(ad_timer);
3549 /* register to receive LACPDUs */
3550 bond_register_lacpdu(bond);
3553 return 0;
3556 static int bond_close(struct net_device *bond_dev)
3558 struct bonding *bond = bond_dev->priv;
3560 if (bond->params.mode == BOND_MODE_8023AD) {
3561 /* Unregister the receive of LACPDUs */
3562 bond_unregister_lacpdu(bond);
3565 if (bond->params.arp_validate)
3566 bond_unregister_arp(bond);
3568 write_lock_bh(&bond->lock);
3571 /* signal timers not to re-arm */
3572 bond->kill_timers = 1;
3574 write_unlock_bh(&bond->lock);
3576 /* del_timer_sync must run without holding the bond->lock
3577 * because a running timer might be trying to hold it too
3580 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3581 del_timer_sync(&bond->mii_timer);
3584 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3585 del_timer_sync(&bond->arp_timer);
3588 switch (bond->params.mode) {
3589 case BOND_MODE_8023AD:
3590 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3591 break;
3592 case BOND_MODE_TLB:
3593 case BOND_MODE_ALB:
3594 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3595 break;
3596 default:
3597 break;
3601 if ((bond->params.mode == BOND_MODE_TLB) ||
3602 (bond->params.mode == BOND_MODE_ALB)) {
3603 /* Must be called only after all
3604 * slaves have been released
3606 bond_alb_deinitialize(bond);
3609 return 0;
3612 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3614 struct bonding *bond = bond_dev->priv;
3615 struct net_device_stats *stats = &(bond->stats), *sstats;
3616 struct slave *slave;
3617 int i;
3619 memset(stats, 0, sizeof(struct net_device_stats));
3621 read_lock_bh(&bond->lock);
3623 bond_for_each_slave(bond, slave, i) {
3624 sstats = slave->dev->get_stats(slave->dev);
3625 stats->rx_packets += sstats->rx_packets;
3626 stats->rx_bytes += sstats->rx_bytes;
3627 stats->rx_errors += sstats->rx_errors;
3628 stats->rx_dropped += sstats->rx_dropped;
3630 stats->tx_packets += sstats->tx_packets;
3631 stats->tx_bytes += sstats->tx_bytes;
3632 stats->tx_errors += sstats->tx_errors;
3633 stats->tx_dropped += sstats->tx_dropped;
3635 stats->multicast += sstats->multicast;
3636 stats->collisions += sstats->collisions;
3638 stats->rx_length_errors += sstats->rx_length_errors;
3639 stats->rx_over_errors += sstats->rx_over_errors;
3640 stats->rx_crc_errors += sstats->rx_crc_errors;
3641 stats->rx_frame_errors += sstats->rx_frame_errors;
3642 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3643 stats->rx_missed_errors += sstats->rx_missed_errors;
3645 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3646 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3647 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3648 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3649 stats->tx_window_errors += sstats->tx_window_errors;
3652 read_unlock_bh(&bond->lock);
3654 return stats;
3657 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3659 struct net_device *slave_dev = NULL;
3660 struct ifbond k_binfo;
3661 struct ifbond __user *u_binfo = NULL;
3662 struct ifslave k_sinfo;
3663 struct ifslave __user *u_sinfo = NULL;
3664 struct mii_ioctl_data *mii = NULL;
3665 int res = 0;
3667 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3668 bond_dev->name, cmd);
3670 switch (cmd) {
3671 case SIOCGMIIPHY:
3672 mii = if_mii(ifr);
3673 if (!mii) {
3674 return -EINVAL;
3676 mii->phy_id = 0;
3677 /* Fall Through */
3678 case SIOCGMIIREG:
3680 * We do this again just in case we were called by SIOCGMIIREG
3681 * instead of SIOCGMIIPHY.
3683 mii = if_mii(ifr);
3684 if (!mii) {
3685 return -EINVAL;
3688 if (mii->reg_num == 1) {
3689 struct bonding *bond = bond_dev->priv;
3690 mii->val_out = 0;
3691 read_lock_bh(&bond->lock);
3692 read_lock(&bond->curr_slave_lock);
3693 if (netif_carrier_ok(bond->dev)) {
3694 mii->val_out = BMSR_LSTATUS;
3696 read_unlock(&bond->curr_slave_lock);
3697 read_unlock_bh(&bond->lock);
3700 return 0;
3701 case BOND_INFO_QUERY_OLD:
3702 case SIOCBONDINFOQUERY:
3703 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3705 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3706 return -EFAULT;
3709 res = bond_info_query(bond_dev, &k_binfo);
3710 if (res == 0) {
3711 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3712 return -EFAULT;
3716 return res;
3717 case BOND_SLAVE_INFO_QUERY_OLD:
3718 case SIOCBONDSLAVEINFOQUERY:
3719 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3721 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3722 return -EFAULT;
3725 res = bond_slave_info_query(bond_dev, &k_sinfo);
3726 if (res == 0) {
3727 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3728 return -EFAULT;
3732 return res;
3733 default:
3734 /* Go on */
3735 break;
3738 if (!capable(CAP_NET_ADMIN)) {
3739 return -EPERM;
3742 down_write(&(bonding_rwsem));
3743 slave_dev = dev_get_by_name(ifr->ifr_slave);
3745 dprintk("slave_dev=%p: \n", slave_dev);
3747 if (!slave_dev) {
3748 res = -ENODEV;
3749 } else {
3750 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3751 switch (cmd) {
3752 case BOND_ENSLAVE_OLD:
3753 case SIOCBONDENSLAVE:
3754 res = bond_enslave(bond_dev, slave_dev);
3755 break;
3756 case BOND_RELEASE_OLD:
3757 case SIOCBONDRELEASE:
3758 res = bond_release(bond_dev, slave_dev);
3759 break;
3760 case BOND_SETHWADDR_OLD:
3761 case SIOCBONDSETHWADDR:
3762 res = bond_sethwaddr(bond_dev, slave_dev);
3763 break;
3764 case BOND_CHANGE_ACTIVE_OLD:
3765 case SIOCBONDCHANGEACTIVE:
3766 res = bond_ioctl_change_active(bond_dev, slave_dev);
3767 break;
3768 default:
3769 res = -EOPNOTSUPP;
3772 dev_put(slave_dev);
3775 up_write(&(bonding_rwsem));
3776 return res;
3779 static void bond_set_multicast_list(struct net_device *bond_dev)
3781 struct bonding *bond = bond_dev->priv;
3782 struct dev_mc_list *dmi;
3784 write_lock_bh(&bond->lock);
3787 * Do promisc before checking multicast_mode
3789 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3790 bond_set_promiscuity(bond, 1);
3793 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3794 bond_set_promiscuity(bond, -1);
3797 /* set allmulti flag to slaves */
3798 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3799 bond_set_allmulti(bond, 1);
3802 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3803 bond_set_allmulti(bond, -1);
3806 bond->flags = bond_dev->flags;
3808 /* looking for addresses to add to slaves' mc list */
3809 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3810 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3811 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3815 /* looking for addresses to delete from slaves' list */
3816 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3817 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3818 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3822 /* save master's multicast list */
3823 bond_mc_list_destroy(bond);
3824 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3826 write_unlock_bh(&bond->lock);
3830 * Change the MTU of all of a master's slaves to match the master
3832 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3834 struct bonding *bond = bond_dev->priv;
3835 struct slave *slave, *stop_at;
3836 int res = 0;
3837 int i;
3839 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3840 (bond_dev ? bond_dev->name : "None"), new_mtu);
3842 /* Can't hold bond->lock with bh disabled here since
3843 * some base drivers panic. On the other hand we can't
3844 * hold bond->lock without bh disabled because we'll
3845 * deadlock. The only solution is to rely on the fact
3846 * that we're under rtnl_lock here, and the slaves
3847 * list won't change. This doesn't solve the problem
3848 * of setting the slave's MTU while it is
3849 * transmitting, but the assumption is that the base
3850 * driver can handle that.
3852 * TODO: figure out a way to safely iterate the slaves
3853 * list, but without holding a lock around the actual
3854 * call to the base driver.
3857 bond_for_each_slave(bond, slave, i) {
3858 dprintk("s %p s->p %p c_m %p\n", slave,
3859 slave->prev, slave->dev->change_mtu);
3861 res = dev_set_mtu(slave->dev, new_mtu);
3863 if (res) {
3864 /* If we failed to set the slave's mtu to the new value
3865 * we must abort the operation even in ACTIVE_BACKUP
3866 * mode, because if we allow the backup slaves to have
3867 * different mtu values than the active slave we'll
3868 * need to change their mtu when doing a failover. That
3869 * means changing their mtu from timer context, which
3870 * is probably not a good idea.
3872 dprintk("err %d %s\n", res, slave->dev->name);
3873 goto unwind;
3877 bond_dev->mtu = new_mtu;
3879 return 0;
3881 unwind:
3882 /* unwind from head to the slave that failed */
3883 stop_at = slave;
3884 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3885 int tmp_res;
3887 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3888 if (tmp_res) {
3889 dprintk("unwind err %d dev %s\n", tmp_res,
3890 slave->dev->name);
3894 return res;
3898 * Change HW address
3900 * Note that many devices must be down to change the HW address, and
3901 * downing the master releases all slaves. We can make bonds full of
3902 * bonding devices to test this, however.
3904 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3906 struct bonding *bond = bond_dev->priv;
3907 struct sockaddr *sa = addr, tmp_sa;
3908 struct slave *slave, *stop_at;
3909 int res = 0;
3910 int i;
3912 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3914 if (!is_valid_ether_addr(sa->sa_data)) {
3915 return -EADDRNOTAVAIL;
3918 /* Can't hold bond->lock with bh disabled here since
3919 * some base drivers panic. On the other hand we can't
3920 * hold bond->lock without bh disabled because we'll
3921 * deadlock. The only solution is to rely on the fact
3922 * that we're under rtnl_lock here, and the slaves
3923 * list won't change. This doesn't solve the problem
3924 * of setting the slave's hw address while it is
3925 * transmitting, but the assumption is that the base
3926 * driver can handle that.
3928 * TODO: figure out a way to safely iterate the slaves
3929 * list, but without holding a lock around the actual
3930 * call to the base driver.
3933 bond_for_each_slave(bond, slave, i) {
3934 dprintk("slave %p %s\n", slave, slave->dev->name);
3936 if (slave->dev->set_mac_address == NULL) {
3937 res = -EOPNOTSUPP;
3938 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3939 goto unwind;
3942 res = dev_set_mac_address(slave->dev, addr);
3943 if (res) {
3944 /* TODO: consider downing the slave
3945 * and retry ?
3946 * User should expect communications
3947 * breakage anyway until ARP finish
3948 * updating, so...
3950 dprintk("err %d %s\n", res, slave->dev->name);
3951 goto unwind;
3955 /* success */
3956 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3957 return 0;
3959 unwind:
3960 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3961 tmp_sa.sa_family = bond_dev->type;
3963 /* unwind from head to the slave that failed */
3964 stop_at = slave;
3965 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3966 int tmp_res;
3968 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3969 if (tmp_res) {
3970 dprintk("unwind err %d dev %s\n", tmp_res,
3971 slave->dev->name);
3975 return res;
3978 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3980 struct bonding *bond = bond_dev->priv;
3981 struct slave *slave, *start_at;
3982 int i;
3983 int res = 1;
3985 read_lock(&bond->lock);
3987 if (!BOND_IS_OK(bond)) {
3988 goto out;
3991 read_lock(&bond->curr_slave_lock);
3992 slave = start_at = bond->curr_active_slave;
3993 read_unlock(&bond->curr_slave_lock);
3995 if (!slave) {
3996 goto out;
3999 bond_for_each_slave_from(bond, slave, i, start_at) {
4000 if (IS_UP(slave->dev) &&
4001 (slave->link == BOND_LINK_UP) &&
4002 (slave->state == BOND_STATE_ACTIVE)) {
4003 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4005 write_lock(&bond->curr_slave_lock);
4006 bond->curr_active_slave = slave->next;
4007 write_unlock(&bond->curr_slave_lock);
4009 break;
4014 out:
4015 if (res) {
4016 /* no suitable interface, frame not sent */
4017 dev_kfree_skb(skb);
4019 read_unlock(&bond->lock);
4020 return 0;
4025 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4026 * the bond has a usable interface.
4028 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4030 struct bonding *bond = bond_dev->priv;
4031 int res = 1;
4033 read_lock(&bond->lock);
4034 read_lock(&bond->curr_slave_lock);
4036 if (!BOND_IS_OK(bond)) {
4037 goto out;
4040 if (!bond->curr_active_slave)
4041 goto out;
4043 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4045 out:
4046 if (res) {
4047 /* no suitable interface, frame not sent */
4048 dev_kfree_skb(skb);
4050 read_unlock(&bond->curr_slave_lock);
4051 read_unlock(&bond->lock);
4052 return 0;
4056 * In bond_xmit_xor() , we determine the output device by using a pre-
4057 * determined xmit_hash_policy(), If the selected device is not enabled,
4058 * find the next active slave.
4060 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4062 struct bonding *bond = bond_dev->priv;
4063 struct slave *slave, *start_at;
4064 int slave_no;
4065 int i;
4066 int res = 1;
4068 read_lock(&bond->lock);
4070 if (!BOND_IS_OK(bond)) {
4071 goto out;
4074 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4076 bond_for_each_slave(bond, slave, i) {
4077 slave_no--;
4078 if (slave_no < 0) {
4079 break;
4083 start_at = slave;
4085 bond_for_each_slave_from(bond, slave, i, start_at) {
4086 if (IS_UP(slave->dev) &&
4087 (slave->link == BOND_LINK_UP) &&
4088 (slave->state == BOND_STATE_ACTIVE)) {
4089 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4090 break;
4094 out:
4095 if (res) {
4096 /* no suitable interface, frame not sent */
4097 dev_kfree_skb(skb);
4099 read_unlock(&bond->lock);
4100 return 0;
4104 * in broadcast mode, we send everything to all usable interfaces.
4106 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4108 struct bonding *bond = bond_dev->priv;
4109 struct slave *slave, *start_at;
4110 struct net_device *tx_dev = NULL;
4111 int i;
4112 int res = 1;
4114 read_lock(&bond->lock);
4116 if (!BOND_IS_OK(bond)) {
4117 goto out;
4120 read_lock(&bond->curr_slave_lock);
4121 start_at = bond->curr_active_slave;
4122 read_unlock(&bond->curr_slave_lock);
4124 if (!start_at) {
4125 goto out;
4128 bond_for_each_slave_from(bond, slave, i, start_at) {
4129 if (IS_UP(slave->dev) &&
4130 (slave->link == BOND_LINK_UP) &&
4131 (slave->state == BOND_STATE_ACTIVE)) {
4132 if (tx_dev) {
4133 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4134 if (!skb2) {
4135 printk(KERN_ERR DRV_NAME
4136 ": %s: Error: bond_xmit_broadcast(): "
4137 "skb_clone() failed\n",
4138 bond_dev->name);
4139 continue;
4142 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4143 if (res) {
4144 dev_kfree_skb(skb2);
4145 continue;
4148 tx_dev = slave->dev;
4152 if (tx_dev) {
4153 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4156 out:
4157 if (res) {
4158 /* no suitable interface, frame not sent */
4159 dev_kfree_skb(skb);
4161 /* frame sent to all suitable interfaces */
4162 read_unlock(&bond->lock);
4163 return 0;
4166 /*------------------------- Device initialization ---------------------------*/
4169 * set bond mode specific net device operations
4171 void bond_set_mode_ops(struct bonding *bond, int mode)
4173 struct net_device *bond_dev = bond->dev;
4175 switch (mode) {
4176 case BOND_MODE_ROUNDROBIN:
4177 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4178 break;
4179 case BOND_MODE_ACTIVEBACKUP:
4180 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4181 break;
4182 case BOND_MODE_XOR:
4183 bond_dev->hard_start_xmit = bond_xmit_xor;
4184 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4185 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4186 else
4187 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4188 break;
4189 case BOND_MODE_BROADCAST:
4190 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4191 break;
4192 case BOND_MODE_8023AD:
4193 bond_set_master_3ad_flags(bond);
4194 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4195 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4196 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4197 else
4198 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4199 break;
4200 case BOND_MODE_ALB:
4201 bond_set_master_alb_flags(bond);
4202 /* FALLTHRU */
4203 case BOND_MODE_TLB:
4204 bond_dev->hard_start_xmit = bond_alb_xmit;
4205 bond_dev->set_mac_address = bond_alb_set_mac_address;
4206 break;
4207 default:
4208 /* Should never happen, mode already checked */
4209 printk(KERN_ERR DRV_NAME
4210 ": %s: Error: Unknown bonding mode %d\n",
4211 bond_dev->name,
4212 mode);
4213 break;
4217 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4218 struct ethtool_drvinfo *drvinfo)
4220 strncpy(drvinfo->driver, DRV_NAME, 32);
4221 strncpy(drvinfo->version, DRV_VERSION, 32);
4222 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4225 static const struct ethtool_ops bond_ethtool_ops = {
4226 .get_tx_csum = ethtool_op_get_tx_csum,
4227 .get_tso = ethtool_op_get_tso,
4228 .get_ufo = ethtool_op_get_ufo,
4229 .get_sg = ethtool_op_get_sg,
4230 .get_drvinfo = bond_ethtool_get_drvinfo,
4234 * Does not allocate but creates a /proc entry.
4235 * Allowed to fail.
4237 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4239 struct bonding *bond = bond_dev->priv;
4241 dprintk("Begin bond_init for %s\n", bond_dev->name);
4243 /* initialize rwlocks */
4244 rwlock_init(&bond->lock);
4245 rwlock_init(&bond->curr_slave_lock);
4247 bond->params = *params; /* copy params struct */
4249 /* Initialize pointers */
4250 bond->first_slave = NULL;
4251 bond->curr_active_slave = NULL;
4252 bond->current_arp_slave = NULL;
4253 bond->primary_slave = NULL;
4254 bond->dev = bond_dev;
4255 INIT_LIST_HEAD(&bond->vlan_list);
4257 /* Initialize the device entry points */
4258 bond_dev->open = bond_open;
4259 bond_dev->stop = bond_close;
4260 bond_dev->get_stats = bond_get_stats;
4261 bond_dev->do_ioctl = bond_do_ioctl;
4262 bond_dev->ethtool_ops = &bond_ethtool_ops;
4263 bond_dev->set_multicast_list = bond_set_multicast_list;
4264 bond_dev->change_mtu = bond_change_mtu;
4265 bond_dev->set_mac_address = bond_set_mac_address;
4267 bond_set_mode_ops(bond, bond->params.mode);
4269 bond_dev->destructor = free_netdev;
4271 /* Initialize the device options */
4272 bond_dev->tx_queue_len = 0;
4273 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4274 bond_dev->priv_flags |= IFF_BONDING;
4276 /* At first, we block adding VLANs. That's the only way to
4277 * prevent problems that occur when adding VLANs over an
4278 * empty bond. The block will be removed once non-challenged
4279 * slaves are enslaved.
4281 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4283 /* don't acquire bond device's netif_tx_lock when
4284 * transmitting */
4285 bond_dev->features |= NETIF_F_LLTX;
4287 /* By default, we declare the bond to be fully
4288 * VLAN hardware accelerated capable. Special
4289 * care is taken in the various xmit functions
4290 * when there are slaves that are not hw accel
4291 * capable
4293 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4294 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4295 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4296 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4297 NETIF_F_HW_VLAN_RX |
4298 NETIF_F_HW_VLAN_FILTER);
4300 #ifdef CONFIG_PROC_FS
4301 bond_create_proc_entry(bond);
4302 #endif
4304 list_add_tail(&bond->bond_list, &bond_dev_list);
4306 return 0;
4309 /* De-initialize device specific data.
4310 * Caller must hold rtnl_lock.
4312 void bond_deinit(struct net_device *bond_dev)
4314 struct bonding *bond = bond_dev->priv;
4316 list_del(&bond->bond_list);
4318 #ifdef CONFIG_PROC_FS
4319 bond_remove_proc_entry(bond);
4320 #endif
4323 /* Unregister and free all bond devices.
4324 * Caller must hold rtnl_lock.
4326 static void bond_free_all(void)
4328 struct bonding *bond, *nxt;
4330 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4331 struct net_device *bond_dev = bond->dev;
4333 bond_mc_list_destroy(bond);
4334 /* Release the bonded slaves */
4335 bond_release_all(bond_dev);
4336 bond_deinit(bond_dev);
4337 unregister_netdevice(bond_dev);
4340 #ifdef CONFIG_PROC_FS
4341 bond_destroy_proc_dir();
4342 #endif
4345 /*------------------------- Module initialization ---------------------------*/
4348 * Convert string input module parms. Accept either the
4349 * number of the mode or its string name.
4351 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4353 int i;
4355 for (i = 0; tbl[i].modename; i++) {
4356 if ((isdigit(*mode_arg) &&
4357 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4358 (strncmp(mode_arg, tbl[i].modename,
4359 strlen(tbl[i].modename)) == 0)) {
4360 return tbl[i].mode;
4364 return -1;
4367 static int bond_check_params(struct bond_params *params)
4369 int arp_validate_value;
4372 * Convert string parameters.
4374 if (mode) {
4375 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4376 if (bond_mode == -1) {
4377 printk(KERN_ERR DRV_NAME
4378 ": Error: Invalid bonding mode \"%s\"\n",
4379 mode == NULL ? "NULL" : mode);
4380 return -EINVAL;
4384 if (xmit_hash_policy) {
4385 if ((bond_mode != BOND_MODE_XOR) &&
4386 (bond_mode != BOND_MODE_8023AD)) {
4387 printk(KERN_INFO DRV_NAME
4388 ": xor_mode param is irrelevant in mode %s\n",
4389 bond_mode_name(bond_mode));
4390 } else {
4391 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4392 xmit_hashtype_tbl);
4393 if (xmit_hashtype == -1) {
4394 printk(KERN_ERR DRV_NAME
4395 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4396 xmit_hash_policy == NULL ? "NULL" :
4397 xmit_hash_policy);
4398 return -EINVAL;
4403 if (lacp_rate) {
4404 if (bond_mode != BOND_MODE_8023AD) {
4405 printk(KERN_INFO DRV_NAME
4406 ": lacp_rate param is irrelevant in mode %s\n",
4407 bond_mode_name(bond_mode));
4408 } else {
4409 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4410 if (lacp_fast == -1) {
4411 printk(KERN_ERR DRV_NAME
4412 ": Error: Invalid lacp rate \"%s\"\n",
4413 lacp_rate == NULL ? "NULL" : lacp_rate);
4414 return -EINVAL;
4419 if (max_bonds < 1 || max_bonds > INT_MAX) {
4420 printk(KERN_WARNING DRV_NAME
4421 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4422 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4423 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4424 max_bonds = BOND_DEFAULT_MAX_BONDS;
4427 if (miimon < 0) {
4428 printk(KERN_WARNING DRV_NAME
4429 ": Warning: miimon module parameter (%d), "
4430 "not in range 0-%d, so it was reset to %d\n",
4431 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4432 miimon = BOND_LINK_MON_INTERV;
4435 if (updelay < 0) {
4436 printk(KERN_WARNING DRV_NAME
4437 ": Warning: updelay module parameter (%d), "
4438 "not in range 0-%d, so it was reset to 0\n",
4439 updelay, INT_MAX);
4440 updelay = 0;
4443 if (downdelay < 0) {
4444 printk(KERN_WARNING DRV_NAME
4445 ": Warning: downdelay module parameter (%d), "
4446 "not in range 0-%d, so it was reset to 0\n",
4447 downdelay, INT_MAX);
4448 downdelay = 0;
4451 if ((use_carrier != 0) && (use_carrier != 1)) {
4452 printk(KERN_WARNING DRV_NAME
4453 ": Warning: use_carrier module parameter (%d), "
4454 "not of valid value (0/1), so it was set to 1\n",
4455 use_carrier);
4456 use_carrier = 1;
4459 /* reset values for 802.3ad */
4460 if (bond_mode == BOND_MODE_8023AD) {
4461 if (!miimon) {
4462 printk(KERN_WARNING DRV_NAME
4463 ": Warning: miimon must be specified, "
4464 "otherwise bonding will not detect link "
4465 "failure, speed and duplex which are "
4466 "essential for 802.3ad operation\n");
4467 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4468 miimon = 100;
4472 /* reset values for TLB/ALB */
4473 if ((bond_mode == BOND_MODE_TLB) ||
4474 (bond_mode == BOND_MODE_ALB)) {
4475 if (!miimon) {
4476 printk(KERN_WARNING DRV_NAME
4477 ": Warning: miimon must be specified, "
4478 "otherwise bonding will not detect link "
4479 "failure and link speed which are essential "
4480 "for TLB/ALB load balancing\n");
4481 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4482 miimon = 100;
4486 if (bond_mode == BOND_MODE_ALB) {
4487 printk(KERN_NOTICE DRV_NAME
4488 ": In ALB mode you might experience client "
4489 "disconnections upon reconnection of a link if the "
4490 "bonding module updelay parameter (%d msec) is "
4491 "incompatible with the forwarding delay time of the "
4492 "switch\n",
4493 updelay);
4496 if (!miimon) {
4497 if (updelay || downdelay) {
4498 /* just warn the user the up/down delay will have
4499 * no effect since miimon is zero...
4501 printk(KERN_WARNING DRV_NAME
4502 ": Warning: miimon module parameter not set "
4503 "and updelay (%d) or downdelay (%d) module "
4504 "parameter is set; updelay and downdelay have "
4505 "no effect unless miimon is set\n",
4506 updelay, downdelay);
4508 } else {
4509 /* don't allow arp monitoring */
4510 if (arp_interval) {
4511 printk(KERN_WARNING DRV_NAME
4512 ": Warning: miimon (%d) and arp_interval (%d) "
4513 "can't be used simultaneously, disabling ARP "
4514 "monitoring\n",
4515 miimon, arp_interval);
4516 arp_interval = 0;
4519 if ((updelay % miimon) != 0) {
4520 printk(KERN_WARNING DRV_NAME
4521 ": Warning: updelay (%d) is not a multiple "
4522 "of miimon (%d), updelay rounded to %d ms\n",
4523 updelay, miimon, (updelay / miimon) * miimon);
4526 updelay /= miimon;
4528 if ((downdelay % miimon) != 0) {
4529 printk(KERN_WARNING DRV_NAME
4530 ": Warning: downdelay (%d) is not a multiple "
4531 "of miimon (%d), downdelay rounded to %d ms\n",
4532 downdelay, miimon,
4533 (downdelay / miimon) * miimon);
4536 downdelay /= miimon;
4539 if (arp_interval < 0) {
4540 printk(KERN_WARNING DRV_NAME
4541 ": Warning: arp_interval module parameter (%d) "
4542 ", not in range 0-%d, so it was reset to %d\n",
4543 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4544 arp_interval = BOND_LINK_ARP_INTERV;
4547 for (arp_ip_count = 0;
4548 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4549 arp_ip_count++) {
4550 /* not complete check, but should be good enough to
4551 catch mistakes */
4552 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4553 printk(KERN_WARNING DRV_NAME
4554 ": Warning: bad arp_ip_target module parameter "
4555 "(%s), ARP monitoring will not be performed\n",
4556 arp_ip_target[arp_ip_count]);
4557 arp_interval = 0;
4558 } else {
4559 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4560 arp_target[arp_ip_count] = ip;
4564 if (arp_interval && !arp_ip_count) {
4565 /* don't allow arping if no arp_ip_target given... */
4566 printk(KERN_WARNING DRV_NAME
4567 ": Warning: arp_interval module parameter (%d) "
4568 "specified without providing an arp_ip_target "
4569 "parameter, arp_interval was reset to 0\n",
4570 arp_interval);
4571 arp_interval = 0;
4574 if (arp_validate) {
4575 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4576 printk(KERN_ERR DRV_NAME
4577 ": arp_validate only supported in active-backup mode\n");
4578 return -EINVAL;
4580 if (!arp_interval) {
4581 printk(KERN_ERR DRV_NAME
4582 ": arp_validate requires arp_interval\n");
4583 return -EINVAL;
4586 arp_validate_value = bond_parse_parm(arp_validate,
4587 arp_validate_tbl);
4588 if (arp_validate_value == -1) {
4589 printk(KERN_ERR DRV_NAME
4590 ": Error: invalid arp_validate \"%s\"\n",
4591 arp_validate == NULL ? "NULL" : arp_validate);
4592 return -EINVAL;
4594 } else
4595 arp_validate_value = 0;
4597 if (miimon) {
4598 printk(KERN_INFO DRV_NAME
4599 ": MII link monitoring set to %d ms\n",
4600 miimon);
4601 } else if (arp_interval) {
4602 int i;
4604 printk(KERN_INFO DRV_NAME
4605 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4606 arp_interval,
4607 arp_validate_tbl[arp_validate_value].modename,
4608 arp_ip_count);
4610 for (i = 0; i < arp_ip_count; i++)
4611 printk (" %s", arp_ip_target[i]);
4613 printk("\n");
4615 } else {
4616 /* miimon and arp_interval not set, we need one so things
4617 * work as expected, see bonding.txt for details
4619 printk(KERN_WARNING DRV_NAME
4620 ": Warning: either miimon or arp_interval and "
4621 "arp_ip_target module parameters must be specified, "
4622 "otherwise bonding will not detect link failures! see "
4623 "bonding.txt for details.\n");
4626 if (primary && !USES_PRIMARY(bond_mode)) {
4627 /* currently, using a primary only makes sense
4628 * in active backup, TLB or ALB modes
4630 printk(KERN_WARNING DRV_NAME
4631 ": Warning: %s primary device specified but has no "
4632 "effect in %s mode\n",
4633 primary, bond_mode_name(bond_mode));
4634 primary = NULL;
4637 /* fill params struct with the proper values */
4638 params->mode = bond_mode;
4639 params->xmit_policy = xmit_hashtype;
4640 params->miimon = miimon;
4641 params->arp_interval = arp_interval;
4642 params->arp_validate = arp_validate_value;
4643 params->updelay = updelay;
4644 params->downdelay = downdelay;
4645 params->use_carrier = use_carrier;
4646 params->lacp_fast = lacp_fast;
4647 params->primary[0] = 0;
4649 if (primary) {
4650 strncpy(params->primary, primary, IFNAMSIZ);
4651 params->primary[IFNAMSIZ - 1] = 0;
4654 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4656 return 0;
4659 static struct lock_class_key bonding_netdev_xmit_lock_key;
4661 /* Create a new bond based on the specified name and bonding parameters.
4662 * If name is NULL, obtain a suitable "bond%d" name for us.
4663 * Caller must NOT hold rtnl_lock; we need to release it here before we
4664 * set up our sysfs entries.
4666 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4668 struct net_device *bond_dev;
4669 int res;
4671 rtnl_lock();
4672 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4673 ether_setup);
4674 if (!bond_dev) {
4675 printk(KERN_ERR DRV_NAME
4676 ": %s: eek! can't alloc netdev!\n",
4677 name);
4678 res = -ENOMEM;
4679 goto out_rtnl;
4682 if (!name) {
4683 res = dev_alloc_name(bond_dev, "bond%d");
4684 if (res < 0)
4685 goto out_netdev;
4688 /* bond_init() must be called after dev_alloc_name() (for the
4689 * /proc files), but before register_netdevice(), because we
4690 * need to set function pointers.
4693 res = bond_init(bond_dev, params);
4694 if (res < 0) {
4695 goto out_netdev;
4698 SET_MODULE_OWNER(bond_dev);
4700 res = register_netdevice(bond_dev);
4701 if (res < 0) {
4702 goto out_bond;
4705 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4707 if (newbond)
4708 *newbond = bond_dev->priv;
4710 netif_carrier_off(bond_dev);
4712 rtnl_unlock(); /* allows sysfs registration of net device */
4713 res = bond_create_sysfs_entry(bond_dev->priv);
4714 if (res < 0) {
4715 rtnl_lock();
4716 goto out_bond;
4719 return 0;
4721 out_bond:
4722 bond_deinit(bond_dev);
4723 out_netdev:
4724 free_netdev(bond_dev);
4725 out_rtnl:
4726 rtnl_unlock();
4727 return res;
4730 static int __init bonding_init(void)
4732 int i;
4733 int res;
4735 printk(KERN_INFO "%s", version);
4737 res = bond_check_params(&bonding_defaults);
4738 if (res) {
4739 goto out;
4742 #ifdef CONFIG_PROC_FS
4743 bond_create_proc_dir();
4744 #endif
4745 for (i = 0; i < max_bonds; i++) {
4746 res = bond_create(NULL, &bonding_defaults, NULL);
4747 if (res)
4748 goto err;
4751 res = bond_create_sysfs();
4752 if (res)
4753 goto err;
4755 register_netdevice_notifier(&bond_netdev_notifier);
4756 register_inetaddr_notifier(&bond_inetaddr_notifier);
4758 goto out;
4759 err:
4760 rtnl_lock();
4761 bond_free_all();
4762 bond_destroy_sysfs();
4763 rtnl_unlock();
4764 out:
4765 return res;
4769 static void __exit bonding_exit(void)
4771 unregister_netdevice_notifier(&bond_netdev_notifier);
4772 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4774 rtnl_lock();
4775 bond_free_all();
4776 bond_destroy_sysfs();
4777 rtnl_unlock();
4780 module_init(bonding_init);
4781 module_exit(bonding_exit);
4782 MODULE_LICENSE("GPL");
4783 MODULE_VERSION(DRV_VERSION);
4784 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4785 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4786 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4789 * Local variables:
4790 * c-indent-level: 8
4791 * c-basic-offset: 8
4792 * tab-width: 8
4793 * End: