Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / net / bonding / bond_main.c
blobc0bbddae4ec41ef998d38f17a0a0c4ec8bd28a54
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/sched.h>
39 #include <linux/types.h>
40 #include <linux/fcntl.h>
41 #include <linux/interrupt.h>
42 #include <linux/ptrace.h>
43 #include <linux/ioport.h>
44 #include <linux/in.h>
45 #include <net/ip.h>
46 #include <linux/ip.h>
47 #include <linux/tcp.h>
48 #include <linux/udp.h>
49 #include <linux/slab.h>
50 #include <linux/string.h>
51 #include <linux/init.h>
52 #include <linux/timer.h>
53 #include <linux/socket.h>
54 #include <linux/ctype.h>
55 #include <linux/inet.h>
56 #include <linux/bitops.h>
57 #include <asm/system.h>
58 #include <asm/io.h>
59 #include <asm/dma.h>
60 #include <asm/uaccess.h>
61 #include <linux/errno.h>
62 #include <linux/netdevice.h>
63 #include <linux/inetdevice.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 = bond->vlgrp->vlan_devices[vid];
493 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
494 bond->vlgrp->vlan_devices[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 = bond->vlgrp->vlan_devices[vlan->vlan_id];
555 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
556 bond->vlgrp->vlan_devices[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);
866 * Totally destroys the mc_list in bond
868 static void bond_mc_list_destroy(struct bonding *bond)
870 struct dev_mc_list *dmi;
872 dmi = bond->mc_list;
873 while (dmi) {
874 bond->mc_list = dmi->next;
875 kfree(dmi);
876 dmi = bond->mc_list;
881 * Copy all the Multicast addresses from src to the bonding device dst
883 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
884 gfp_t gfp_flag)
886 struct dev_mc_list *dmi, *new_dmi;
888 for (dmi = mc_list; dmi; dmi = dmi->next) {
889 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
891 if (!new_dmi) {
892 /* FIXME: Potential memory leak !!! */
893 return -ENOMEM;
896 new_dmi->next = bond->mc_list;
897 bond->mc_list = new_dmi;
898 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
899 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
900 new_dmi->dmi_users = dmi->dmi_users;
901 new_dmi->dmi_gusers = dmi->dmi_gusers;
904 return 0;
908 * flush all members of flush->mc_list from device dev->mc_list
910 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
912 struct bonding *bond = bond_dev->priv;
913 struct dev_mc_list *dmi;
915 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
916 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
919 if (bond->params.mode == BOND_MODE_8023AD) {
920 /* del lacpdu mc addr from mc list */
921 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
923 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
927 /*--------------------------- Active slave change ---------------------------*/
930 * Update the mc list and multicast-related flags for the new and
931 * old active slaves (if any) according to the multicast mode, and
932 * promiscuous flags unconditionally.
934 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
936 struct dev_mc_list *dmi;
938 if (!USES_PRIMARY(bond->params.mode)) {
939 /* nothing to do - mc list is already up-to-date on
940 * all slaves
942 return;
945 if (old_active) {
946 if (bond->dev->flags & IFF_PROMISC) {
947 dev_set_promiscuity(old_active->dev, -1);
950 if (bond->dev->flags & IFF_ALLMULTI) {
951 dev_set_allmulti(old_active->dev, -1);
954 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
955 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
959 if (new_active) {
960 if (bond->dev->flags & IFF_PROMISC) {
961 dev_set_promiscuity(new_active->dev, 1);
964 if (bond->dev->flags & IFF_ALLMULTI) {
965 dev_set_allmulti(new_active->dev, 1);
968 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
969 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
975 * find_best_interface - select the best available slave to be the active one
976 * @bond: our bonding struct
978 * Warning: Caller must hold curr_slave_lock for writing.
980 static struct slave *bond_find_best_slave(struct bonding *bond)
982 struct slave *new_active, *old_active;
983 struct slave *bestslave = NULL;
984 int mintime = bond->params.updelay;
985 int i;
987 new_active = old_active = bond->curr_active_slave;
989 if (!new_active) { /* there were no active slaves left */
990 if (bond->slave_cnt > 0) { /* found one slave */
991 new_active = bond->first_slave;
992 } else {
993 return NULL; /* still no slave, return NULL */
997 /* first try the primary link; if arping, a link must tx/rx traffic
998 * before it can be considered the curr_active_slave - also, we would skip
999 * slaves between the curr_active_slave and primary_slave that may be up
1000 * and able to arp
1002 if ((bond->primary_slave) &&
1003 (!bond->params.arp_interval) &&
1004 (IS_UP(bond->primary_slave->dev))) {
1005 new_active = bond->primary_slave;
1008 /* remember where to stop iterating over the slaves */
1009 old_active = new_active;
1011 bond_for_each_slave_from(bond, new_active, i, old_active) {
1012 if (IS_UP(new_active->dev)) {
1013 if (new_active->link == BOND_LINK_UP) {
1014 return new_active;
1015 } else if (new_active->link == BOND_LINK_BACK) {
1016 /* link up, but waiting for stabilization */
1017 if (new_active->delay < mintime) {
1018 mintime = new_active->delay;
1019 bestslave = new_active;
1025 return bestslave;
1029 * change_active_interface - change the active slave into the specified one
1030 * @bond: our bonding struct
1031 * @new: the new slave to make the active one
1033 * Set the new slave to the bond's settings and unset them on the old
1034 * curr_active_slave.
1035 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1037 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1038 * because it is apparently the best available slave we have, even though its
1039 * updelay hasn't timed out yet.
1041 * Warning: Caller must hold curr_slave_lock for writing.
1043 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1045 struct slave *old_active = bond->curr_active_slave;
1047 if (old_active == new_active) {
1048 return;
1051 if (new_active) {
1052 if (new_active->link == BOND_LINK_BACK) {
1053 if (USES_PRIMARY(bond->params.mode)) {
1054 printk(KERN_INFO DRV_NAME
1055 ": %s: making interface %s the new "
1056 "active one %d ms earlier.\n",
1057 bond->dev->name, new_active->dev->name,
1058 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1061 new_active->delay = 0;
1062 new_active->link = BOND_LINK_UP;
1063 new_active->jiffies = jiffies;
1065 if (bond->params.mode == BOND_MODE_8023AD) {
1066 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1069 if ((bond->params.mode == BOND_MODE_TLB) ||
1070 (bond->params.mode == BOND_MODE_ALB)) {
1071 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1073 } else {
1074 if (USES_PRIMARY(bond->params.mode)) {
1075 printk(KERN_INFO DRV_NAME
1076 ": %s: making interface %s the new "
1077 "active one.\n",
1078 bond->dev->name, new_active->dev->name);
1083 if (USES_PRIMARY(bond->params.mode)) {
1084 bond_mc_swap(bond, new_active, old_active);
1087 if ((bond->params.mode == BOND_MODE_TLB) ||
1088 (bond->params.mode == BOND_MODE_ALB)) {
1089 bond_alb_handle_active_change(bond, new_active);
1090 if (old_active)
1091 bond_set_slave_inactive_flags(old_active);
1092 if (new_active)
1093 bond_set_slave_active_flags(new_active);
1094 } else {
1095 bond->curr_active_slave = new_active;
1098 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1099 if (old_active) {
1100 bond_set_slave_inactive_flags(old_active);
1103 if (new_active) {
1104 bond_set_slave_active_flags(new_active);
1106 bond_send_gratuitous_arp(bond);
1111 * bond_select_active_slave - select a new active slave, if needed
1112 * @bond: our bonding struct
1114 * This functions shoud be called when one of the following occurs:
1115 * - The old curr_active_slave has been released or lost its link.
1116 * - The primary_slave has got its link back.
1117 * - A slave has got its link back and there's no old curr_active_slave.
1119 * Warning: Caller must hold curr_slave_lock for writing.
1121 void bond_select_active_slave(struct bonding *bond)
1123 struct slave *best_slave;
1124 int rv;
1126 best_slave = bond_find_best_slave(bond);
1127 if (best_slave != bond->curr_active_slave) {
1128 bond_change_active_slave(bond, best_slave);
1129 rv = bond_set_carrier(bond);
1130 if (!rv)
1131 return;
1133 if (netif_carrier_ok(bond->dev)) {
1134 printk(KERN_INFO DRV_NAME
1135 ": %s: first active interface up!\n",
1136 bond->dev->name);
1137 } else {
1138 printk(KERN_INFO DRV_NAME ": %s: "
1139 "now running without any active interface !\n",
1140 bond->dev->name);
1145 /*--------------------------- slave list handling ---------------------------*/
1148 * This function attaches the slave to the end of list.
1150 * bond->lock held for writing by caller.
1152 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1154 if (bond->first_slave == NULL) { /* attaching the first slave */
1155 new_slave->next = new_slave;
1156 new_slave->prev = new_slave;
1157 bond->first_slave = new_slave;
1158 } else {
1159 new_slave->next = bond->first_slave;
1160 new_slave->prev = bond->first_slave->prev;
1161 new_slave->next->prev = new_slave;
1162 new_slave->prev->next = new_slave;
1165 bond->slave_cnt++;
1169 * This function detaches the slave from the list.
1170 * WARNING: no check is made to verify if the slave effectively
1171 * belongs to <bond>.
1172 * Nothing is freed on return, structures are just unchained.
1173 * If any slave pointer in bond was pointing to <slave>,
1174 * it should be changed by the calling function.
1176 * bond->lock held for writing by caller.
1178 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1180 if (slave->next) {
1181 slave->next->prev = slave->prev;
1184 if (slave->prev) {
1185 slave->prev->next = slave->next;
1188 if (bond->first_slave == slave) { /* slave is the first slave */
1189 if (bond->slave_cnt > 1) { /* there are more slave */
1190 bond->first_slave = slave->next;
1191 } else {
1192 bond->first_slave = NULL; /* slave was the last one */
1196 slave->next = NULL;
1197 slave->prev = NULL;
1198 bond->slave_cnt--;
1201 /*---------------------------------- IOCTL ----------------------------------*/
1203 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1205 dprintk("bond_dev=%p\n", bond_dev);
1206 dprintk("slave_dev=%p\n", slave_dev);
1207 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1208 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1209 return 0;
1212 #define BOND_INTERSECT_FEATURES \
1213 (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_TSO | NETIF_F_UFO)
1216 * Compute the common dev->feature set available to all slaves. Some
1217 * feature bits are managed elsewhere, so preserve feature bits set on
1218 * master device that are not part of the examined set.
1220 static int bond_compute_features(struct bonding *bond)
1222 unsigned long features = BOND_INTERSECT_FEATURES;
1223 struct slave *slave;
1224 struct net_device *bond_dev = bond->dev;
1225 unsigned short max_hard_header_len = ETH_HLEN;
1226 int i;
1228 bond_for_each_slave(bond, slave, i) {
1229 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1230 if (slave->dev->hard_header_len > max_hard_header_len)
1231 max_hard_header_len = slave->dev->hard_header_len;
1234 if ((features & NETIF_F_SG) &&
1235 !(features & NETIF_F_ALL_CSUM))
1236 features &= ~NETIF_F_SG;
1239 * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1240 * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1241 * implies that all slaves also support scatter-gather
1242 * (NETIF_F_SG), which implies that features also includes
1243 * NETIF_F_SG. So no need to check whether we have an
1244 * illegal combination of NETIF_F_{TSO,UFO} and
1245 * !NETIF_F_SG
1248 features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1249 bond_dev->features = features;
1250 bond_dev->hard_header_len = max_hard_header_len;
1252 return 0;
1255 /* enslave device <slave> to bond device <master> */
1256 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1258 struct bonding *bond = bond_dev->priv;
1259 struct slave *new_slave = NULL;
1260 struct dev_mc_list *dmi;
1261 struct sockaddr addr;
1262 int link_reporting;
1263 int old_features = bond_dev->features;
1264 int res = 0;
1266 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1267 slave_dev->do_ioctl == NULL) {
1268 printk(KERN_WARNING DRV_NAME
1269 ": %s: Warning: no link monitoring support for %s\n",
1270 bond_dev->name, slave_dev->name);
1273 /* bond must be initialized by bond_open() before enslaving */
1274 if (!(bond_dev->flags & IFF_UP)) {
1275 dprintk("Error, master_dev is not up\n");
1276 return -EPERM;
1279 /* already enslaved */
1280 if (slave_dev->flags & IFF_SLAVE) {
1281 dprintk("Error, Device was already enslaved\n");
1282 return -EBUSY;
1285 /* vlan challenged mutual exclusion */
1286 /* no need to lock since we're protected by rtnl_lock */
1287 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1288 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1289 if (!list_empty(&bond->vlan_list)) {
1290 printk(KERN_ERR DRV_NAME
1291 ": %s: Error: cannot enslave VLAN "
1292 "challenged slave %s on VLAN enabled "
1293 "bond %s\n", bond_dev->name, slave_dev->name,
1294 bond_dev->name);
1295 return -EPERM;
1296 } else {
1297 printk(KERN_WARNING DRV_NAME
1298 ": %s: Warning: enslaved VLAN challenged "
1299 "slave %s. Adding VLANs will be blocked as "
1300 "long as %s is part of bond %s\n",
1301 bond_dev->name, slave_dev->name, slave_dev->name,
1302 bond_dev->name);
1303 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1305 } else {
1306 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1307 if (bond->slave_cnt == 0) {
1308 /* First slave, and it is not VLAN challenged,
1309 * so remove the block of adding VLANs over the bond.
1311 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1316 * Old ifenslave binaries are no longer supported. These can
1317 * be identified with moderate accurary by the state of the slave:
1318 * the current ifenslave will set the interface down prior to
1319 * enslaving it; the old ifenslave will not.
1321 if ((slave_dev->flags & IFF_UP)) {
1322 printk(KERN_ERR DRV_NAME ": %s is up. "
1323 "This may be due to an out of date ifenslave.\n",
1324 slave_dev->name);
1325 res = -EPERM;
1326 goto err_undo_flags;
1329 if (slave_dev->set_mac_address == NULL) {
1330 printk(KERN_ERR DRV_NAME
1331 ": %s: Error: The slave device you specified does "
1332 "not support setting the MAC address. "
1333 "Your kernel likely does not support slave "
1334 "devices.\n", bond_dev->name);
1335 res = -EOPNOTSUPP;
1336 goto err_undo_flags;
1339 new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1340 if (!new_slave) {
1341 res = -ENOMEM;
1342 goto err_undo_flags;
1345 memset(new_slave, 0, sizeof(struct slave));
1347 /* save slave's original flags before calling
1348 * netdev_set_master and dev_open
1350 new_slave->original_flags = slave_dev->flags;
1353 * Save slave's original ("permanent") mac address for modes
1354 * that need it, and for restoring it upon release, and then
1355 * set it to the master's address
1357 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1360 * Set slave to master's mac address. The application already
1361 * set the master's mac address to that of the first slave
1363 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1364 addr.sa_family = slave_dev->type;
1365 res = dev_set_mac_address(slave_dev, &addr);
1366 if (res) {
1367 dprintk("Error %d calling set_mac_address\n", res);
1368 goto err_free;
1371 /* open the slave since the application closed it */
1372 res = dev_open(slave_dev);
1373 if (res) {
1374 dprintk("Openning slave %s failed\n", slave_dev->name);
1375 goto err_restore_mac;
1378 res = netdev_set_master(slave_dev, bond_dev);
1379 if (res) {
1380 dprintk("Error %d calling netdev_set_master\n", res);
1381 goto err_close;
1384 new_slave->dev = slave_dev;
1385 slave_dev->priv_flags |= IFF_BONDING;
1387 if ((bond->params.mode == BOND_MODE_TLB) ||
1388 (bond->params.mode == BOND_MODE_ALB)) {
1389 /* bond_alb_init_slave() must be called before all other stages since
1390 * it might fail and we do not want to have to undo everything
1392 res = bond_alb_init_slave(bond, new_slave);
1393 if (res) {
1394 goto err_unset_master;
1398 /* If the mode USES_PRIMARY, then the new slave gets the
1399 * master's promisc (and mc) settings only if it becomes the
1400 * curr_active_slave, and that is taken care of later when calling
1401 * bond_change_active()
1403 if (!USES_PRIMARY(bond->params.mode)) {
1404 /* set promiscuity level to new slave */
1405 if (bond_dev->flags & IFF_PROMISC) {
1406 dev_set_promiscuity(slave_dev, 1);
1409 /* set allmulti level to new slave */
1410 if (bond_dev->flags & IFF_ALLMULTI) {
1411 dev_set_allmulti(slave_dev, 1);
1414 /* upload master's mc_list to new slave */
1415 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1416 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1420 if (bond->params.mode == BOND_MODE_8023AD) {
1421 /* add lacpdu mc addr to mc list */
1422 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1424 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1427 bond_add_vlans_on_slave(bond, slave_dev);
1429 write_lock_bh(&bond->lock);
1431 bond_attach_slave(bond, new_slave);
1433 new_slave->delay = 0;
1434 new_slave->link_failure_count = 0;
1436 bond_compute_features(bond);
1438 new_slave->last_arp_rx = jiffies;
1440 if (bond->params.miimon && !bond->params.use_carrier) {
1441 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1443 if ((link_reporting == -1) && !bond->params.arp_interval) {
1445 * miimon is set but a bonded network driver
1446 * does not support ETHTOOL/MII and
1447 * arp_interval is not set. Note: if
1448 * use_carrier is enabled, we will never go
1449 * here (because netif_carrier is always
1450 * supported); thus, we don't need to change
1451 * the messages for netif_carrier.
1453 printk(KERN_WARNING DRV_NAME
1454 ": %s: Warning: MII and ETHTOOL support not "
1455 "available for interface %s, and "
1456 "arp_interval/arp_ip_target module parameters "
1457 "not specified, thus bonding will not detect "
1458 "link failures! see bonding.txt for details.\n",
1459 bond_dev->name, slave_dev->name);
1460 } else if (link_reporting == -1) {
1461 /* unable get link status using mii/ethtool */
1462 printk(KERN_WARNING DRV_NAME
1463 ": %s: Warning: can't get link status from "
1464 "interface %s; the network driver associated "
1465 "with this interface does not support MII or "
1466 "ETHTOOL link status reporting, thus miimon "
1467 "has no effect on this interface.\n",
1468 bond_dev->name, slave_dev->name);
1472 /* check for initial state */
1473 if (!bond->params.miimon ||
1474 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1475 if (bond->params.updelay) {
1476 dprintk("Initial state of slave_dev is "
1477 "BOND_LINK_BACK\n");
1478 new_slave->link = BOND_LINK_BACK;
1479 new_slave->delay = bond->params.updelay;
1480 } else {
1481 dprintk("Initial state of slave_dev is "
1482 "BOND_LINK_UP\n");
1483 new_slave->link = BOND_LINK_UP;
1485 new_slave->jiffies = jiffies;
1486 } else {
1487 dprintk("Initial state of slave_dev is "
1488 "BOND_LINK_DOWN\n");
1489 new_slave->link = BOND_LINK_DOWN;
1492 if (bond_update_speed_duplex(new_slave) &&
1493 (new_slave->link != BOND_LINK_DOWN)) {
1494 printk(KERN_WARNING DRV_NAME
1495 ": %s: Warning: failed to get speed and duplex from %s, "
1496 "assumed to be 100Mb/sec and Full.\n",
1497 bond_dev->name, new_slave->dev->name);
1499 if (bond->params.mode == BOND_MODE_8023AD) {
1500 printk(KERN_WARNING DRV_NAME
1501 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1502 "support in base driver for proper aggregator "
1503 "selection.\n", bond_dev->name);
1507 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1508 /* if there is a primary slave, remember it */
1509 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1510 bond->primary_slave = new_slave;
1514 switch (bond->params.mode) {
1515 case BOND_MODE_ACTIVEBACKUP:
1516 bond_set_slave_inactive_flags(new_slave);
1517 bond_select_active_slave(bond);
1518 break;
1519 case BOND_MODE_8023AD:
1520 /* in 802.3ad mode, the internal mechanism
1521 * will activate the slaves in the selected
1522 * aggregator
1524 bond_set_slave_inactive_flags(new_slave);
1525 /* if this is the first slave */
1526 if (bond->slave_cnt == 1) {
1527 SLAVE_AD_INFO(new_slave).id = 1;
1528 /* Initialize AD with the number of times that the AD timer is called in 1 second
1529 * can be called only after the mac address of the bond is set
1531 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1532 bond->params.lacp_fast);
1533 } else {
1534 SLAVE_AD_INFO(new_slave).id =
1535 SLAVE_AD_INFO(new_slave->prev).id + 1;
1538 bond_3ad_bind_slave(new_slave);
1539 break;
1540 case BOND_MODE_TLB:
1541 case BOND_MODE_ALB:
1542 new_slave->state = BOND_STATE_ACTIVE;
1543 if ((!bond->curr_active_slave) &&
1544 (new_slave->link != BOND_LINK_DOWN)) {
1545 /* first slave or no active slave yet, and this link
1546 * is OK, so make this interface the active one
1548 bond_change_active_slave(bond, new_slave);
1549 } else {
1550 bond_set_slave_inactive_flags(new_slave);
1552 break;
1553 default:
1554 dprintk("This slave is always active in trunk mode\n");
1556 /* always active in trunk mode */
1557 new_slave->state = BOND_STATE_ACTIVE;
1559 /* In trunking mode there is little meaning to curr_active_slave
1560 * anyway (it holds no special properties of the bond device),
1561 * so we can change it without calling change_active_interface()
1563 if (!bond->curr_active_slave) {
1564 bond->curr_active_slave = new_slave;
1566 break;
1567 } /* switch(bond_mode) */
1569 bond_set_carrier(bond);
1571 write_unlock_bh(&bond->lock);
1573 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1574 if (res)
1575 goto err_unset_master;
1577 printk(KERN_INFO DRV_NAME
1578 ": %s: enslaving %s as a%s interface with a%s link.\n",
1579 bond_dev->name, slave_dev->name,
1580 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1581 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1583 /* enslave is successful */
1584 return 0;
1586 /* Undo stages on error */
1587 err_unset_master:
1588 netdev_set_master(slave_dev, NULL);
1590 err_close:
1591 dev_close(slave_dev);
1593 err_restore_mac:
1594 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1595 addr.sa_family = slave_dev->type;
1596 dev_set_mac_address(slave_dev, &addr);
1598 err_free:
1599 kfree(new_slave);
1601 err_undo_flags:
1602 bond_dev->features = old_features;
1604 return res;
1608 * Try to release the slave device <slave> from the bond device <master>
1609 * It is legal to access curr_active_slave without a lock because all the function
1610 * is write-locked.
1612 * The rules for slave state should be:
1613 * for Active/Backup:
1614 * Active stays on all backups go down
1615 * for Bonded connections:
1616 * The first up interface should be left on and all others downed.
1618 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1620 struct bonding *bond = bond_dev->priv;
1621 struct slave *slave, *oldcurrent;
1622 struct sockaddr addr;
1623 int mac_addr_differ;
1625 /* slave is not a slave or master is not master of this slave */
1626 if (!(slave_dev->flags & IFF_SLAVE) ||
1627 (slave_dev->master != bond_dev)) {
1628 printk(KERN_ERR DRV_NAME
1629 ": %s: Error: cannot release %s.\n",
1630 bond_dev->name, slave_dev->name);
1631 return -EINVAL;
1634 write_lock_bh(&bond->lock);
1636 slave = bond_get_slave_by_dev(bond, slave_dev);
1637 if (!slave) {
1638 /* not a slave of this bond */
1639 printk(KERN_INFO DRV_NAME
1640 ": %s: %s not enslaved\n",
1641 bond_dev->name, slave_dev->name);
1642 write_unlock_bh(&bond->lock);
1643 return -EINVAL;
1646 mac_addr_differ = memcmp(bond_dev->dev_addr,
1647 slave->perm_hwaddr,
1648 ETH_ALEN);
1649 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1650 printk(KERN_WARNING DRV_NAME
1651 ": %s: Warning: the permanent HWaddr of %s "
1652 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1653 "still in use by %s. Set the HWaddr of "
1654 "%s to a different address to avoid "
1655 "conflicts.\n",
1656 bond_dev->name,
1657 slave_dev->name,
1658 slave->perm_hwaddr[0],
1659 slave->perm_hwaddr[1],
1660 slave->perm_hwaddr[2],
1661 slave->perm_hwaddr[3],
1662 slave->perm_hwaddr[4],
1663 slave->perm_hwaddr[5],
1664 bond_dev->name,
1665 slave_dev->name);
1668 /* Inform AD package of unbinding of slave. */
1669 if (bond->params.mode == BOND_MODE_8023AD) {
1670 /* must be called before the slave is
1671 * detached from the list
1673 bond_3ad_unbind_slave(slave);
1676 printk(KERN_INFO DRV_NAME
1677 ": %s: releasing %s interface %s\n",
1678 bond_dev->name,
1679 (slave->state == BOND_STATE_ACTIVE)
1680 ? "active" : "backup",
1681 slave_dev->name);
1683 oldcurrent = bond->curr_active_slave;
1685 bond->current_arp_slave = NULL;
1687 /* release the slave from its bond */
1688 bond_detach_slave(bond, slave);
1690 bond_compute_features(bond);
1692 if (bond->primary_slave == slave) {
1693 bond->primary_slave = NULL;
1696 if (oldcurrent == slave) {
1697 bond_change_active_slave(bond, NULL);
1700 if ((bond->params.mode == BOND_MODE_TLB) ||
1701 (bond->params.mode == BOND_MODE_ALB)) {
1702 /* Must be called only after the slave has been
1703 * detached from the list and the curr_active_slave
1704 * has been cleared (if our_slave == old_current),
1705 * but before a new active slave is selected.
1707 bond_alb_deinit_slave(bond, slave);
1710 if (oldcurrent == slave)
1711 bond_select_active_slave(bond);
1713 if (bond->slave_cnt == 0) {
1714 bond_set_carrier(bond);
1716 /* if the last slave was removed, zero the mac address
1717 * of the master so it will be set by the application
1718 * to the mac address of the first slave
1720 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1722 if (list_empty(&bond->vlan_list)) {
1723 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1724 } else {
1725 printk(KERN_WARNING DRV_NAME
1726 ": %s: Warning: clearing HW address of %s while it "
1727 "still has VLANs.\n",
1728 bond_dev->name, bond_dev->name);
1729 printk(KERN_WARNING DRV_NAME
1730 ": %s: When re-adding slaves, make sure the bond's "
1731 "HW address matches its VLANs'.\n",
1732 bond_dev->name);
1734 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1735 !bond_has_challenged_slaves(bond)) {
1736 printk(KERN_INFO DRV_NAME
1737 ": %s: last VLAN challenged slave %s "
1738 "left bond %s. VLAN blocking is removed\n",
1739 bond_dev->name, slave_dev->name, bond_dev->name);
1740 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1743 write_unlock_bh(&bond->lock);
1745 /* must do this from outside any spinlocks */
1746 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1748 bond_del_vlans_from_slave(bond, slave_dev);
1750 /* If the mode USES_PRIMARY, then we should only remove its
1751 * promisc and mc settings if it was the curr_active_slave, but that was
1752 * already taken care of above when we detached the slave
1754 if (!USES_PRIMARY(bond->params.mode)) {
1755 /* unset promiscuity level from slave */
1756 if (bond_dev->flags & IFF_PROMISC) {
1757 dev_set_promiscuity(slave_dev, -1);
1760 /* unset allmulti level from slave */
1761 if (bond_dev->flags & IFF_ALLMULTI) {
1762 dev_set_allmulti(slave_dev, -1);
1765 /* flush master's mc_list from slave */
1766 bond_mc_list_flush(bond_dev, slave_dev);
1769 netdev_set_master(slave_dev, NULL);
1771 /* close slave before restoring its mac address */
1772 dev_close(slave_dev);
1774 /* restore original ("permanent") mac address */
1775 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1776 addr.sa_family = slave_dev->type;
1777 dev_set_mac_address(slave_dev, &addr);
1779 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1780 IFF_SLAVE_INACTIVE | IFF_BONDING |
1781 IFF_SLAVE_NEEDARP);
1783 kfree(slave);
1785 return 0; /* deletion OK */
1789 * This function releases all slaves.
1791 static int bond_release_all(struct net_device *bond_dev)
1793 struct bonding *bond = bond_dev->priv;
1794 struct slave *slave;
1795 struct net_device *slave_dev;
1796 struct sockaddr addr;
1798 write_lock_bh(&bond->lock);
1800 netif_carrier_off(bond_dev);
1802 if (bond->slave_cnt == 0) {
1803 goto out;
1806 bond->current_arp_slave = NULL;
1807 bond->primary_slave = NULL;
1808 bond_change_active_slave(bond, NULL);
1810 while ((slave = bond->first_slave) != NULL) {
1811 /* Inform AD package of unbinding of slave
1812 * before slave is detached from the list.
1814 if (bond->params.mode == BOND_MODE_8023AD) {
1815 bond_3ad_unbind_slave(slave);
1818 slave_dev = slave->dev;
1819 bond_detach_slave(bond, slave);
1821 if ((bond->params.mode == BOND_MODE_TLB) ||
1822 (bond->params.mode == BOND_MODE_ALB)) {
1823 /* must be called only after the slave
1824 * has been detached from the list
1826 bond_alb_deinit_slave(bond, slave);
1829 bond_compute_features(bond);
1831 /* now that the slave is detached, unlock and perform
1832 * all the undo steps that should not be called from
1833 * within a lock.
1835 write_unlock_bh(&bond->lock);
1837 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1838 bond_del_vlans_from_slave(bond, slave_dev);
1840 /* If the mode USES_PRIMARY, then we should only remove its
1841 * promisc and mc settings if it was the curr_active_slave, but that was
1842 * already taken care of above when we detached the slave
1844 if (!USES_PRIMARY(bond->params.mode)) {
1845 /* unset promiscuity level from slave */
1846 if (bond_dev->flags & IFF_PROMISC) {
1847 dev_set_promiscuity(slave_dev, -1);
1850 /* unset allmulti level from slave */
1851 if (bond_dev->flags & IFF_ALLMULTI) {
1852 dev_set_allmulti(slave_dev, -1);
1855 /* flush master's mc_list from slave */
1856 bond_mc_list_flush(bond_dev, slave_dev);
1859 netdev_set_master(slave_dev, NULL);
1861 /* close slave before restoring its mac address */
1862 dev_close(slave_dev);
1864 /* restore original ("permanent") mac address*/
1865 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1866 addr.sa_family = slave_dev->type;
1867 dev_set_mac_address(slave_dev, &addr);
1869 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1870 IFF_SLAVE_INACTIVE);
1872 kfree(slave);
1874 /* re-acquire the lock before getting the next slave */
1875 write_lock_bh(&bond->lock);
1878 /* zero the mac address of the master so it will be
1879 * set by the application to the mac address of the
1880 * first slave
1882 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1884 if (list_empty(&bond->vlan_list)) {
1885 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1886 } else {
1887 printk(KERN_WARNING DRV_NAME
1888 ": %s: Warning: clearing HW address of %s while it "
1889 "still has VLANs.\n",
1890 bond_dev->name, bond_dev->name);
1891 printk(KERN_WARNING DRV_NAME
1892 ": %s: When re-adding slaves, make sure the bond's "
1893 "HW address matches its VLANs'.\n",
1894 bond_dev->name);
1897 printk(KERN_INFO DRV_NAME
1898 ": %s: released all slaves\n",
1899 bond_dev->name);
1901 out:
1902 write_unlock_bh(&bond->lock);
1904 return 0;
1908 * This function changes the active slave to slave <slave_dev>.
1909 * It returns -EINVAL in the following cases.
1910 * - <slave_dev> is not found in the list.
1911 * - There is not active slave now.
1912 * - <slave_dev> is already active.
1913 * - The link state of <slave_dev> is not BOND_LINK_UP.
1914 * - <slave_dev> is not running.
1915 * In these cases, this fuction does nothing.
1916 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1918 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1920 struct bonding *bond = bond_dev->priv;
1921 struct slave *old_active = NULL;
1922 struct slave *new_active = NULL;
1923 int res = 0;
1925 if (!USES_PRIMARY(bond->params.mode)) {
1926 return -EINVAL;
1929 /* Verify that master_dev is indeed the master of slave_dev */
1930 if (!(slave_dev->flags & IFF_SLAVE) ||
1931 (slave_dev->master != bond_dev)) {
1932 return -EINVAL;
1935 write_lock_bh(&bond->lock);
1937 old_active = bond->curr_active_slave;
1938 new_active = bond_get_slave_by_dev(bond, slave_dev);
1941 * Changing to the current active: do nothing; return success.
1943 if (new_active && (new_active == old_active)) {
1944 write_unlock_bh(&bond->lock);
1945 return 0;
1948 if ((new_active) &&
1949 (old_active) &&
1950 (new_active->link == BOND_LINK_UP) &&
1951 IS_UP(new_active->dev)) {
1952 bond_change_active_slave(bond, new_active);
1953 } else {
1954 res = -EINVAL;
1957 write_unlock_bh(&bond->lock);
1959 return res;
1962 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1964 struct bonding *bond = bond_dev->priv;
1966 info->bond_mode = bond->params.mode;
1967 info->miimon = bond->params.miimon;
1969 read_lock_bh(&bond->lock);
1970 info->num_slaves = bond->slave_cnt;
1971 read_unlock_bh(&bond->lock);
1973 return 0;
1976 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1978 struct bonding *bond = bond_dev->priv;
1979 struct slave *slave;
1980 int i, found = 0;
1982 if (info->slave_id < 0) {
1983 return -ENODEV;
1986 read_lock_bh(&bond->lock);
1988 bond_for_each_slave(bond, slave, i) {
1989 if (i == (int)info->slave_id) {
1990 found = 1;
1991 break;
1995 read_unlock_bh(&bond->lock);
1997 if (found) {
1998 strcpy(info->slave_name, slave->dev->name);
1999 info->link = slave->link;
2000 info->state = slave->state;
2001 info->link_failure_count = slave->link_failure_count;
2002 } else {
2003 return -ENODEV;
2006 return 0;
2009 /*-------------------------------- Monitoring -------------------------------*/
2011 /* this function is called regularly to monitor each slave's link. */
2012 void bond_mii_monitor(struct net_device *bond_dev)
2014 struct bonding *bond = bond_dev->priv;
2015 struct slave *slave, *oldcurrent;
2016 int do_failover = 0;
2017 int delta_in_ticks;
2018 int i;
2020 read_lock(&bond->lock);
2022 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2024 if (bond->kill_timers) {
2025 goto out;
2028 if (bond->slave_cnt == 0) {
2029 goto re_arm;
2032 /* we will try to read the link status of each of our slaves, and
2033 * set their IFF_RUNNING flag appropriately. For each slave not
2034 * supporting MII status, we won't do anything so that a user-space
2035 * program could monitor the link itself if needed.
2038 read_lock(&bond->curr_slave_lock);
2039 oldcurrent = bond->curr_active_slave;
2040 read_unlock(&bond->curr_slave_lock);
2042 bond_for_each_slave(bond, slave, i) {
2043 struct net_device *slave_dev = slave->dev;
2044 int link_state;
2045 u16 old_speed = slave->speed;
2046 u8 old_duplex = slave->duplex;
2048 link_state = bond_check_dev_link(bond, slave_dev, 0);
2050 switch (slave->link) {
2051 case BOND_LINK_UP: /* the link was up */
2052 if (link_state == BMSR_LSTATUS) {
2053 /* link stays up, nothing more to do */
2054 break;
2055 } else { /* link going down */
2056 slave->link = BOND_LINK_FAIL;
2057 slave->delay = bond->params.downdelay;
2059 if (slave->link_failure_count < UINT_MAX) {
2060 slave->link_failure_count++;
2063 if (bond->params.downdelay) {
2064 printk(KERN_INFO DRV_NAME
2065 ": %s: link status down for %s "
2066 "interface %s, disabling it in "
2067 "%d ms.\n",
2068 bond_dev->name,
2069 IS_UP(slave_dev)
2070 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2071 ? ((slave == oldcurrent)
2072 ? "active " : "backup ")
2073 : "")
2074 : "idle ",
2075 slave_dev->name,
2076 bond->params.downdelay * bond->params.miimon);
2079 /* no break ! fall through the BOND_LINK_FAIL test to
2080 ensure proper action to be taken
2082 case BOND_LINK_FAIL: /* the link has just gone down */
2083 if (link_state != BMSR_LSTATUS) {
2084 /* link stays down */
2085 if (slave->delay <= 0) {
2086 /* link down for too long time */
2087 slave->link = BOND_LINK_DOWN;
2089 /* in active/backup mode, we must
2090 * completely disable this interface
2092 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2093 (bond->params.mode == BOND_MODE_8023AD)) {
2094 bond_set_slave_inactive_flags(slave);
2097 printk(KERN_INFO DRV_NAME
2098 ": %s: link status definitely "
2099 "down for interface %s, "
2100 "disabling it\n",
2101 bond_dev->name,
2102 slave_dev->name);
2104 /* notify ad that the link status has changed */
2105 if (bond->params.mode == BOND_MODE_8023AD) {
2106 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2109 if ((bond->params.mode == BOND_MODE_TLB) ||
2110 (bond->params.mode == BOND_MODE_ALB)) {
2111 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2114 if (slave == oldcurrent) {
2115 do_failover = 1;
2117 } else {
2118 slave->delay--;
2120 } else {
2121 /* link up again */
2122 slave->link = BOND_LINK_UP;
2123 slave->jiffies = jiffies;
2124 printk(KERN_INFO DRV_NAME
2125 ": %s: link status up again after %d "
2126 "ms for interface %s.\n",
2127 bond_dev->name,
2128 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2129 slave_dev->name);
2131 break;
2132 case BOND_LINK_DOWN: /* the link was down */
2133 if (link_state != BMSR_LSTATUS) {
2134 /* the link stays down, nothing more to do */
2135 break;
2136 } else { /* link going up */
2137 slave->link = BOND_LINK_BACK;
2138 slave->delay = bond->params.updelay;
2140 if (bond->params.updelay) {
2141 /* if updelay == 0, no need to
2142 advertise about a 0 ms delay */
2143 printk(KERN_INFO DRV_NAME
2144 ": %s: link status up for "
2145 "interface %s, enabling it "
2146 "in %d ms.\n",
2147 bond_dev->name,
2148 slave_dev->name,
2149 bond->params.updelay * bond->params.miimon);
2152 /* no break ! fall through the BOND_LINK_BACK state in
2153 case there's something to do.
2155 case BOND_LINK_BACK: /* the link has just come back */
2156 if (link_state != BMSR_LSTATUS) {
2157 /* link down again */
2158 slave->link = BOND_LINK_DOWN;
2160 printk(KERN_INFO DRV_NAME
2161 ": %s: link status down again after %d "
2162 "ms for interface %s.\n",
2163 bond_dev->name,
2164 (bond->params.updelay - slave->delay) * bond->params.miimon,
2165 slave_dev->name);
2166 } else {
2167 /* link stays up */
2168 if (slave->delay == 0) {
2169 /* now the link has been up for long time enough */
2170 slave->link = BOND_LINK_UP;
2171 slave->jiffies = jiffies;
2173 if (bond->params.mode == BOND_MODE_8023AD) {
2174 /* prevent it from being the active one */
2175 slave->state = BOND_STATE_BACKUP;
2176 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2177 /* make it immediately active */
2178 slave->state = BOND_STATE_ACTIVE;
2179 } else if (slave != bond->primary_slave) {
2180 /* prevent it from being the active one */
2181 slave->state = BOND_STATE_BACKUP;
2184 printk(KERN_INFO DRV_NAME
2185 ": %s: link status definitely "
2186 "up for interface %s.\n",
2187 bond_dev->name,
2188 slave_dev->name);
2190 /* notify ad that the link status has changed */
2191 if (bond->params.mode == BOND_MODE_8023AD) {
2192 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2195 if ((bond->params.mode == BOND_MODE_TLB) ||
2196 (bond->params.mode == BOND_MODE_ALB)) {
2197 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2200 if ((!oldcurrent) ||
2201 (slave == bond->primary_slave)) {
2202 do_failover = 1;
2204 } else {
2205 slave->delay--;
2208 break;
2209 default:
2210 /* Should not happen */
2211 printk(KERN_ERR DRV_NAME
2212 ": %s: Error: %s Illegal value (link=%d)\n",
2213 bond_dev->name,
2214 slave->dev->name,
2215 slave->link);
2216 goto out;
2217 } /* end of switch (slave->link) */
2219 bond_update_speed_duplex(slave);
2221 if (bond->params.mode == BOND_MODE_8023AD) {
2222 if (old_speed != slave->speed) {
2223 bond_3ad_adapter_speed_changed(slave);
2226 if (old_duplex != slave->duplex) {
2227 bond_3ad_adapter_duplex_changed(slave);
2231 } /* end of for */
2233 if (do_failover) {
2234 write_lock(&bond->curr_slave_lock);
2236 bond_select_active_slave(bond);
2238 write_unlock(&bond->curr_slave_lock);
2239 } else
2240 bond_set_carrier(bond);
2242 re_arm:
2243 if (bond->params.miimon) {
2244 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2246 out:
2247 read_unlock(&bond->lock);
2251 static u32 bond_glean_dev_ip(struct net_device *dev)
2253 struct in_device *idev;
2254 struct in_ifaddr *ifa;
2255 __be32 addr = 0;
2257 if (!dev)
2258 return 0;
2260 rcu_read_lock();
2261 idev = __in_dev_get_rcu(dev);
2262 if (!idev)
2263 goto out;
2265 ifa = idev->ifa_list;
2266 if (!ifa)
2267 goto out;
2269 addr = ifa->ifa_local;
2270 out:
2271 rcu_read_unlock();
2272 return addr;
2275 static int bond_has_ip(struct bonding *bond)
2277 struct vlan_entry *vlan, *vlan_next;
2279 if (bond->master_ip)
2280 return 1;
2282 if (list_empty(&bond->vlan_list))
2283 return 0;
2285 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2286 vlan_list) {
2287 if (vlan->vlan_ip)
2288 return 1;
2291 return 0;
2294 static int bond_has_this_ip(struct bonding *bond, u32 ip)
2296 struct vlan_entry *vlan, *vlan_next;
2298 if (ip == bond->master_ip)
2299 return 1;
2301 if (list_empty(&bond->vlan_list))
2302 return 0;
2304 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2305 vlan_list) {
2306 if (ip == vlan->vlan_ip)
2307 return 1;
2310 return 0;
2314 * We go to the (large) trouble of VLAN tagging ARP frames because
2315 * switches in VLAN mode (especially if ports are configured as
2316 * "native" to a VLAN) might not pass non-tagged frames.
2318 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2320 struct sk_buff *skb;
2322 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2323 slave_dev->name, dest_ip, src_ip, vlan_id);
2325 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2326 NULL, slave_dev->dev_addr, NULL);
2328 if (!skb) {
2329 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2330 return;
2332 if (vlan_id) {
2333 skb = vlan_put_tag(skb, vlan_id);
2334 if (!skb) {
2335 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2336 return;
2339 arp_xmit(skb);
2343 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2345 int i, vlan_id, rv;
2346 u32 *targets = bond->params.arp_targets;
2347 struct vlan_entry *vlan, *vlan_next;
2348 struct net_device *vlan_dev;
2349 struct flowi fl;
2350 struct rtable *rt;
2352 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2353 if (!targets[i])
2354 continue;
2355 dprintk("basa: target %x\n", targets[i]);
2356 if (list_empty(&bond->vlan_list)) {
2357 dprintk("basa: empty vlan: arp_send\n");
2358 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2359 bond->master_ip, 0);
2360 continue;
2364 * If VLANs are configured, we do a route lookup to
2365 * determine which VLAN interface would be used, so we
2366 * can tag the ARP with the proper VLAN tag.
2368 memset(&fl, 0, sizeof(fl));
2369 fl.fl4_dst = targets[i];
2370 fl.fl4_tos = RTO_ONLINK;
2372 rv = ip_route_output_key(&rt, &fl);
2373 if (rv) {
2374 if (net_ratelimit()) {
2375 printk(KERN_WARNING DRV_NAME
2376 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2377 bond->dev->name, NIPQUAD(fl.fl4_dst));
2379 continue;
2383 * This target is not on a VLAN
2385 if (rt->u.dst.dev == bond->dev) {
2386 ip_rt_put(rt);
2387 dprintk("basa: rtdev == bond->dev: arp_send\n");
2388 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2389 bond->master_ip, 0);
2390 continue;
2393 vlan_id = 0;
2394 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2395 vlan_list) {
2396 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2397 if (vlan_dev == rt->u.dst.dev) {
2398 vlan_id = vlan->vlan_id;
2399 dprintk("basa: vlan match on %s %d\n",
2400 vlan_dev->name, vlan_id);
2401 break;
2405 if (vlan_id) {
2406 ip_rt_put(rt);
2407 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2408 vlan->vlan_ip, vlan_id);
2409 continue;
2412 if (net_ratelimit()) {
2413 printk(KERN_WARNING DRV_NAME
2414 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2415 bond->dev->name, NIPQUAD(fl.fl4_dst),
2416 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2418 ip_rt_put(rt);
2423 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2424 * for each VLAN above us.
2426 static void bond_send_gratuitous_arp(struct bonding *bond)
2428 struct slave *slave = bond->curr_active_slave;
2429 struct vlan_entry *vlan;
2430 struct net_device *vlan_dev;
2432 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2433 slave ? slave->dev->name : "NULL");
2434 if (!slave)
2435 return;
2437 if (bond->master_ip) {
2438 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2439 bond->master_ip, 0);
2442 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2443 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2444 if (vlan->vlan_ip) {
2445 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2446 vlan->vlan_ip, vlan->vlan_id);
2451 static void bond_validate_arp(struct bonding *bond, struct slave *slave, u32 sip, u32 tip)
2453 int i;
2454 u32 *targets = bond->params.arp_targets;
2456 targets = bond->params.arp_targets;
2457 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2458 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2459 "%u.%u.%u.%u bhti(tip) %d\n",
2460 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2461 bond_has_this_ip(bond, tip));
2462 if (sip == targets[i]) {
2463 if (bond_has_this_ip(bond, tip))
2464 slave->last_arp_rx = jiffies;
2465 return;
2470 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2472 struct arphdr *arp;
2473 struct slave *slave;
2474 struct bonding *bond;
2475 unsigned char *arp_ptr;
2476 u32 sip, tip;
2478 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2479 goto out;
2481 bond = dev->priv;
2482 read_lock(&bond->lock);
2484 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2485 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2486 orig_dev ? orig_dev->name : "NULL");
2488 slave = bond_get_slave_by_dev(bond, orig_dev);
2489 if (!slave || !slave_do_arp_validate(bond, slave))
2490 goto out_unlock;
2492 /* ARP header, plus 2 device addresses, plus 2 IP addresses. */
2493 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2494 (2 * dev->addr_len) +
2495 (2 * sizeof(u32)))))
2496 goto out_unlock;
2498 arp = skb->nh.arph;
2499 if (arp->ar_hln != dev->addr_len ||
2500 skb->pkt_type == PACKET_OTHERHOST ||
2501 skb->pkt_type == PACKET_LOOPBACK ||
2502 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2503 arp->ar_pro != htons(ETH_P_IP) ||
2504 arp->ar_pln != 4)
2505 goto out_unlock;
2507 arp_ptr = (unsigned char *)(arp + 1);
2508 arp_ptr += dev->addr_len;
2509 memcpy(&sip, arp_ptr, 4);
2510 arp_ptr += 4 + dev->addr_len;
2511 memcpy(&tip, arp_ptr, 4);
2513 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2514 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2515 slave->state, bond->params.arp_validate,
2516 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2519 * Backup slaves won't see the ARP reply, but do come through
2520 * here for each ARP probe (so we swap the sip/tip to validate
2521 * the probe). In a "redundant switch, common router" type of
2522 * configuration, the ARP probe will (hopefully) travel from
2523 * the active, through one switch, the router, then the other
2524 * switch before reaching the backup.
2526 if (slave->state == BOND_STATE_ACTIVE)
2527 bond_validate_arp(bond, slave, sip, tip);
2528 else
2529 bond_validate_arp(bond, slave, tip, sip);
2531 out_unlock:
2532 read_unlock(&bond->lock);
2533 out:
2534 dev_kfree_skb(skb);
2535 return NET_RX_SUCCESS;
2539 * this function is called regularly to monitor each slave's link
2540 * ensuring that traffic is being sent and received when arp monitoring
2541 * is used in load-balancing mode. if the adapter has been dormant, then an
2542 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2543 * arp monitoring in active backup mode.
2545 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2547 struct bonding *bond = bond_dev->priv;
2548 struct slave *slave, *oldcurrent;
2549 int do_failover = 0;
2550 int delta_in_ticks;
2551 int i;
2553 read_lock(&bond->lock);
2555 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2557 if (bond->kill_timers) {
2558 goto out;
2561 if (bond->slave_cnt == 0) {
2562 goto re_arm;
2565 read_lock(&bond->curr_slave_lock);
2566 oldcurrent = bond->curr_active_slave;
2567 read_unlock(&bond->curr_slave_lock);
2569 /* see if any of the previous devices are up now (i.e. they have
2570 * xmt and rcv traffic). the curr_active_slave does not come into
2571 * the picture unless it is null. also, slave->jiffies is not needed
2572 * here because we send an arp on each slave and give a slave as
2573 * long as it needs to get the tx/rx within the delta.
2574 * TODO: what about up/down delay in arp mode? it wasn't here before
2575 * so it can wait
2577 bond_for_each_slave(bond, slave, i) {
2578 if (slave->link != BOND_LINK_UP) {
2579 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2580 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2582 slave->link = BOND_LINK_UP;
2583 slave->state = BOND_STATE_ACTIVE;
2585 /* primary_slave has no meaning in round-robin
2586 * mode. the window of a slave being up and
2587 * curr_active_slave being null after enslaving
2588 * is closed.
2590 if (!oldcurrent) {
2591 printk(KERN_INFO DRV_NAME
2592 ": %s: link status definitely "
2593 "up for interface %s, ",
2594 bond_dev->name,
2595 slave->dev->name);
2596 do_failover = 1;
2597 } else {
2598 printk(KERN_INFO DRV_NAME
2599 ": %s: interface %s is now up\n",
2600 bond_dev->name,
2601 slave->dev->name);
2604 } else {
2605 /* slave->link == BOND_LINK_UP */
2607 /* not all switches will respond to an arp request
2608 * when the source ip is 0, so don't take the link down
2609 * if we don't know our ip yet
2611 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2612 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2613 bond_has_ip(bond))) {
2615 slave->link = BOND_LINK_DOWN;
2616 slave->state = BOND_STATE_BACKUP;
2618 if (slave->link_failure_count < UINT_MAX) {
2619 slave->link_failure_count++;
2622 printk(KERN_INFO DRV_NAME
2623 ": %s: interface %s is now down.\n",
2624 bond_dev->name,
2625 slave->dev->name);
2627 if (slave == oldcurrent) {
2628 do_failover = 1;
2633 /* note: if switch is in round-robin mode, all links
2634 * must tx arp to ensure all links rx an arp - otherwise
2635 * links may oscillate or not come up at all; if switch is
2636 * in something like xor mode, there is nothing we can
2637 * do - all replies will be rx'ed on same link causing slaves
2638 * to be unstable during low/no traffic periods
2640 if (IS_UP(slave->dev)) {
2641 bond_arp_send_all(bond, slave);
2645 if (do_failover) {
2646 write_lock(&bond->curr_slave_lock);
2648 bond_select_active_slave(bond);
2650 write_unlock(&bond->curr_slave_lock);
2653 re_arm:
2654 if (bond->params.arp_interval) {
2655 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2657 out:
2658 read_unlock(&bond->lock);
2662 * When using arp monitoring in active-backup mode, this function is
2663 * called to determine if any backup slaves have went down or a new
2664 * current slave needs to be found.
2665 * The backup slaves never generate traffic, they are considered up by merely
2666 * receiving traffic. If the current slave goes down, each backup slave will
2667 * be given the opportunity to tx/rx an arp before being taken down - this
2668 * prevents all slaves from being taken down due to the current slave not
2669 * sending any traffic for the backups to receive. The arps are not necessarily
2670 * necessary, any tx and rx traffic will keep the current slave up. While any
2671 * rx traffic will keep the backup slaves up, the current slave is responsible
2672 * for generating traffic to keep them up regardless of any other traffic they
2673 * may have received.
2674 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2676 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2678 struct bonding *bond = bond_dev->priv;
2679 struct slave *slave;
2680 int delta_in_ticks;
2681 int i;
2683 read_lock(&bond->lock);
2685 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2687 if (bond->kill_timers) {
2688 goto out;
2691 if (bond->slave_cnt == 0) {
2692 goto re_arm;
2695 /* determine if any slave has come up or any backup slave has
2696 * gone down
2697 * TODO: what about up/down delay in arp mode? it wasn't here before
2698 * so it can wait
2700 bond_for_each_slave(bond, slave, i) {
2701 if (slave->link != BOND_LINK_UP) {
2702 if ((jiffies - slave_last_rx(bond, slave)) <=
2703 delta_in_ticks) {
2705 slave->link = BOND_LINK_UP;
2707 write_lock(&bond->curr_slave_lock);
2709 if ((!bond->curr_active_slave) &&
2710 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2711 bond_change_active_slave(bond, slave);
2712 bond->current_arp_slave = NULL;
2713 } else if (bond->curr_active_slave != slave) {
2714 /* this slave has just come up but we
2715 * already have a current slave; this
2716 * can also happen if bond_enslave adds
2717 * a new slave that is up while we are
2718 * searching for a new slave
2720 bond_set_slave_inactive_flags(slave);
2721 bond->current_arp_slave = NULL;
2724 bond_set_carrier(bond);
2726 if (slave == bond->curr_active_slave) {
2727 printk(KERN_INFO DRV_NAME
2728 ": %s: %s is up and now the "
2729 "active interface\n",
2730 bond_dev->name,
2731 slave->dev->name);
2732 netif_carrier_on(bond->dev);
2733 } else {
2734 printk(KERN_INFO DRV_NAME
2735 ": %s: backup interface %s is "
2736 "now up\n",
2737 bond_dev->name,
2738 slave->dev->name);
2741 write_unlock(&bond->curr_slave_lock);
2743 } else {
2744 read_lock(&bond->curr_slave_lock);
2746 if ((slave != bond->curr_active_slave) &&
2747 (!bond->current_arp_slave) &&
2748 (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2749 bond_has_ip(bond))) {
2750 /* a backup slave has gone down; three times
2751 * the delta allows the current slave to be
2752 * taken out before the backup slave.
2753 * note: a non-null current_arp_slave indicates
2754 * the curr_active_slave went down and we are
2755 * searching for a new one; under this
2756 * condition we only take the curr_active_slave
2757 * down - this gives each slave a chance to
2758 * tx/rx traffic before being taken out
2761 read_unlock(&bond->curr_slave_lock);
2763 slave->link = BOND_LINK_DOWN;
2765 if (slave->link_failure_count < UINT_MAX) {
2766 slave->link_failure_count++;
2769 bond_set_slave_inactive_flags(slave);
2771 printk(KERN_INFO DRV_NAME
2772 ": %s: backup interface %s is now down\n",
2773 bond_dev->name,
2774 slave->dev->name);
2775 } else {
2776 read_unlock(&bond->curr_slave_lock);
2781 read_lock(&bond->curr_slave_lock);
2782 slave = bond->curr_active_slave;
2783 read_unlock(&bond->curr_slave_lock);
2785 if (slave) {
2786 /* if we have sent traffic in the past 2*arp_intervals but
2787 * haven't xmit and rx traffic in that time interval, select
2788 * a different slave. slave->jiffies is only updated when
2789 * a slave first becomes the curr_active_slave - not necessarily
2790 * after every arp; this ensures the slave has a full 2*delta
2791 * before being taken out. if a primary is being used, check
2792 * if it is up and needs to take over as the curr_active_slave
2794 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2795 (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2796 bond_has_ip(bond))) &&
2797 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2799 slave->link = BOND_LINK_DOWN;
2801 if (slave->link_failure_count < UINT_MAX) {
2802 slave->link_failure_count++;
2805 printk(KERN_INFO DRV_NAME
2806 ": %s: link status down for active interface "
2807 "%s, disabling it\n",
2808 bond_dev->name,
2809 slave->dev->name);
2811 write_lock(&bond->curr_slave_lock);
2813 bond_select_active_slave(bond);
2814 slave = bond->curr_active_slave;
2816 write_unlock(&bond->curr_slave_lock);
2818 bond->current_arp_slave = slave;
2820 if (slave) {
2821 slave->jiffies = jiffies;
2823 } else if ((bond->primary_slave) &&
2824 (bond->primary_slave != slave) &&
2825 (bond->primary_slave->link == BOND_LINK_UP)) {
2826 /* at this point, slave is the curr_active_slave */
2827 printk(KERN_INFO DRV_NAME
2828 ": %s: changing from interface %s to primary "
2829 "interface %s\n",
2830 bond_dev->name,
2831 slave->dev->name,
2832 bond->primary_slave->dev->name);
2834 /* primary is up so switch to it */
2835 write_lock(&bond->curr_slave_lock);
2836 bond_change_active_slave(bond, bond->primary_slave);
2837 write_unlock(&bond->curr_slave_lock);
2839 slave = bond->primary_slave;
2840 slave->jiffies = jiffies;
2841 } else {
2842 bond->current_arp_slave = NULL;
2845 /* the current slave must tx an arp to ensure backup slaves
2846 * rx traffic
2848 if (slave && bond_has_ip(bond)) {
2849 bond_arp_send_all(bond, slave);
2853 /* if we don't have a curr_active_slave, search for the next available
2854 * backup slave from the current_arp_slave and make it the candidate
2855 * for becoming the curr_active_slave
2857 if (!slave) {
2858 if (!bond->current_arp_slave) {
2859 bond->current_arp_slave = bond->first_slave;
2862 if (bond->current_arp_slave) {
2863 bond_set_slave_inactive_flags(bond->current_arp_slave);
2865 /* search for next candidate */
2866 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2867 if (IS_UP(slave->dev)) {
2868 slave->link = BOND_LINK_BACK;
2869 bond_set_slave_active_flags(slave);
2870 bond_arp_send_all(bond, slave);
2871 slave->jiffies = jiffies;
2872 bond->current_arp_slave = slave;
2873 break;
2876 /* if the link state is up at this point, we
2877 * mark it down - this can happen if we have
2878 * simultaneous link failures and
2879 * reselect_active_interface doesn't make this
2880 * one the current slave so it is still marked
2881 * up when it is actually down
2883 if (slave->link == BOND_LINK_UP) {
2884 slave->link = BOND_LINK_DOWN;
2885 if (slave->link_failure_count < UINT_MAX) {
2886 slave->link_failure_count++;
2889 bond_set_slave_inactive_flags(slave);
2891 printk(KERN_INFO DRV_NAME
2892 ": %s: backup interface %s is "
2893 "now down.\n",
2894 bond_dev->name,
2895 slave->dev->name);
2901 re_arm:
2902 if (bond->params.arp_interval) {
2903 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2905 out:
2906 read_unlock(&bond->lock);
2909 /*------------------------------ proc/seq_file-------------------------------*/
2911 #ifdef CONFIG_PROC_FS
2913 #define SEQ_START_TOKEN ((void *)1)
2915 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2917 struct bonding *bond = seq->private;
2918 loff_t off = 0;
2919 struct slave *slave;
2920 int i;
2922 /* make sure the bond won't be taken away */
2923 read_lock(&dev_base_lock);
2924 read_lock_bh(&bond->lock);
2926 if (*pos == 0) {
2927 return SEQ_START_TOKEN;
2930 bond_for_each_slave(bond, slave, i) {
2931 if (++off == *pos) {
2932 return slave;
2936 return NULL;
2939 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2941 struct bonding *bond = seq->private;
2942 struct slave *slave = v;
2944 ++*pos;
2945 if (v == SEQ_START_TOKEN) {
2946 return bond->first_slave;
2949 slave = slave->next;
2951 return (slave == bond->first_slave) ? NULL : slave;
2954 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2956 struct bonding *bond = seq->private;
2958 read_unlock_bh(&bond->lock);
2959 read_unlock(&dev_base_lock);
2962 static void bond_info_show_master(struct seq_file *seq)
2964 struct bonding *bond = seq->private;
2965 struct slave *curr;
2966 int i;
2967 u32 target;
2969 read_lock(&bond->curr_slave_lock);
2970 curr = bond->curr_active_slave;
2971 read_unlock(&bond->curr_slave_lock);
2973 seq_printf(seq, "Bonding Mode: %s\n",
2974 bond_mode_name(bond->params.mode));
2976 if (bond->params.mode == BOND_MODE_XOR ||
2977 bond->params.mode == BOND_MODE_8023AD) {
2978 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2979 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2980 bond->params.xmit_policy);
2983 if (USES_PRIMARY(bond->params.mode)) {
2984 seq_printf(seq, "Primary Slave: %s\n",
2985 (bond->primary_slave) ?
2986 bond->primary_slave->dev->name : "None");
2988 seq_printf(seq, "Currently Active Slave: %s\n",
2989 (curr) ? curr->dev->name : "None");
2992 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
2993 "up" : "down");
2994 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
2995 seq_printf(seq, "Up Delay (ms): %d\n",
2996 bond->params.updelay * bond->params.miimon);
2997 seq_printf(seq, "Down Delay (ms): %d\n",
2998 bond->params.downdelay * bond->params.miimon);
3001 /* ARP information */
3002 if(bond->params.arp_interval > 0) {
3003 int printed=0;
3004 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3005 bond->params.arp_interval);
3007 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3009 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3010 if (!bond->params.arp_targets[i])
3011 continue;
3012 if (printed)
3013 seq_printf(seq, ",");
3014 target = ntohl(bond->params.arp_targets[i]);
3015 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3016 printed = 1;
3018 seq_printf(seq, "\n");
3021 if (bond->params.mode == BOND_MODE_8023AD) {
3022 struct ad_info ad_info;
3024 seq_puts(seq, "\n802.3ad info\n");
3025 seq_printf(seq, "LACP rate: %s\n",
3026 (bond->params.lacp_fast) ? "fast" : "slow");
3028 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3029 seq_printf(seq, "bond %s has no active aggregator\n",
3030 bond->dev->name);
3031 } else {
3032 seq_printf(seq, "Active Aggregator Info:\n");
3034 seq_printf(seq, "\tAggregator ID: %d\n",
3035 ad_info.aggregator_id);
3036 seq_printf(seq, "\tNumber of ports: %d\n",
3037 ad_info.ports);
3038 seq_printf(seq, "\tActor Key: %d\n",
3039 ad_info.actor_key);
3040 seq_printf(seq, "\tPartner Key: %d\n",
3041 ad_info.partner_key);
3042 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3043 ad_info.partner_system[0],
3044 ad_info.partner_system[1],
3045 ad_info.partner_system[2],
3046 ad_info.partner_system[3],
3047 ad_info.partner_system[4],
3048 ad_info.partner_system[5]);
3053 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3055 struct bonding *bond = seq->private;
3057 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3058 seq_printf(seq, "MII Status: %s\n",
3059 (slave->link == BOND_LINK_UP) ? "up" : "down");
3060 seq_printf(seq, "Link Failure Count: %u\n",
3061 slave->link_failure_count);
3063 seq_printf(seq,
3064 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3065 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3066 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3067 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
3069 if (bond->params.mode == BOND_MODE_8023AD) {
3070 const struct aggregator *agg
3071 = SLAVE_AD_INFO(slave).port.aggregator;
3073 if (agg) {
3074 seq_printf(seq, "Aggregator ID: %d\n",
3075 agg->aggregator_identifier);
3076 } else {
3077 seq_puts(seq, "Aggregator ID: N/A\n");
3082 static int bond_info_seq_show(struct seq_file *seq, void *v)
3084 if (v == SEQ_START_TOKEN) {
3085 seq_printf(seq, "%s\n", version);
3086 bond_info_show_master(seq);
3087 } else {
3088 bond_info_show_slave(seq, v);
3091 return 0;
3094 static struct seq_operations bond_info_seq_ops = {
3095 .start = bond_info_seq_start,
3096 .next = bond_info_seq_next,
3097 .stop = bond_info_seq_stop,
3098 .show = bond_info_seq_show,
3101 static int bond_info_open(struct inode *inode, struct file *file)
3103 struct seq_file *seq;
3104 struct proc_dir_entry *proc;
3105 int res;
3107 res = seq_open(file, &bond_info_seq_ops);
3108 if (!res) {
3109 /* recover the pointer buried in proc_dir_entry data */
3110 seq = file->private_data;
3111 proc = PDE(inode);
3112 seq->private = proc->data;
3115 return res;
3118 static struct file_operations bond_info_fops = {
3119 .owner = THIS_MODULE,
3120 .open = bond_info_open,
3121 .read = seq_read,
3122 .llseek = seq_lseek,
3123 .release = seq_release,
3126 static int bond_create_proc_entry(struct bonding *bond)
3128 struct net_device *bond_dev = bond->dev;
3130 if (bond_proc_dir) {
3131 bond->proc_entry = create_proc_entry(bond_dev->name,
3132 S_IRUGO,
3133 bond_proc_dir);
3134 if (bond->proc_entry == NULL) {
3135 printk(KERN_WARNING DRV_NAME
3136 ": Warning: Cannot create /proc/net/%s/%s\n",
3137 DRV_NAME, bond_dev->name);
3138 } else {
3139 bond->proc_entry->data = bond;
3140 bond->proc_entry->proc_fops = &bond_info_fops;
3141 bond->proc_entry->owner = THIS_MODULE;
3142 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3146 return 0;
3149 static void bond_remove_proc_entry(struct bonding *bond)
3151 if (bond_proc_dir && bond->proc_entry) {
3152 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3153 memset(bond->proc_file_name, 0, IFNAMSIZ);
3154 bond->proc_entry = NULL;
3158 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3159 * Caller must hold rtnl_lock.
3161 static void bond_create_proc_dir(void)
3163 int len = strlen(DRV_NAME);
3165 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3166 bond_proc_dir = bond_proc_dir->next) {
3167 if ((bond_proc_dir->namelen == len) &&
3168 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3169 break;
3173 if (!bond_proc_dir) {
3174 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3175 if (bond_proc_dir) {
3176 bond_proc_dir->owner = THIS_MODULE;
3177 } else {
3178 printk(KERN_WARNING DRV_NAME
3179 ": Warning: cannot create /proc/net/%s\n",
3180 DRV_NAME);
3185 /* Destroy the bonding directory under /proc/net, if empty.
3186 * Caller must hold rtnl_lock.
3188 static void bond_destroy_proc_dir(void)
3190 struct proc_dir_entry *de;
3192 if (!bond_proc_dir) {
3193 return;
3196 /* verify that the /proc dir is empty */
3197 for (de = bond_proc_dir->subdir; de; de = de->next) {
3198 /* ignore . and .. */
3199 if (*(de->name) != '.') {
3200 break;
3204 if (de) {
3205 if (bond_proc_dir->owner == THIS_MODULE) {
3206 bond_proc_dir->owner = NULL;
3208 } else {
3209 remove_proc_entry(DRV_NAME, proc_net);
3210 bond_proc_dir = NULL;
3213 #endif /* CONFIG_PROC_FS */
3215 /*-------------------------- netdev event handling --------------------------*/
3218 * Change device name
3220 static int bond_event_changename(struct bonding *bond)
3222 #ifdef CONFIG_PROC_FS
3223 bond_remove_proc_entry(bond);
3224 bond_create_proc_entry(bond);
3225 #endif
3226 down_write(&(bonding_rwsem));
3227 bond_destroy_sysfs_entry(bond);
3228 bond_create_sysfs_entry(bond);
3229 up_write(&(bonding_rwsem));
3230 return NOTIFY_DONE;
3233 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3235 struct bonding *event_bond = bond_dev->priv;
3237 switch (event) {
3238 case NETDEV_CHANGENAME:
3239 return bond_event_changename(event_bond);
3240 case NETDEV_UNREGISTER:
3242 * TODO: remove a bond from the list?
3244 break;
3245 default:
3246 break;
3249 return NOTIFY_DONE;
3252 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3254 struct net_device *bond_dev = slave_dev->master;
3255 struct bonding *bond = bond_dev->priv;
3257 switch (event) {
3258 case NETDEV_UNREGISTER:
3259 if (bond_dev) {
3260 bond_release(bond_dev, slave_dev);
3262 break;
3263 case NETDEV_CHANGE:
3265 * TODO: is this what we get if somebody
3266 * sets up a hierarchical bond, then rmmod's
3267 * one of the slave bonding devices?
3269 break;
3270 case NETDEV_DOWN:
3272 * ... Or is it this?
3274 break;
3275 case NETDEV_CHANGEMTU:
3277 * TODO: Should slaves be allowed to
3278 * independently alter their MTU? For
3279 * an active-backup bond, slaves need
3280 * not be the same type of device, so
3281 * MTUs may vary. For other modes,
3282 * slaves arguably should have the
3283 * same MTUs. To do this, we'd need to
3284 * take over the slave's change_mtu
3285 * function for the duration of their
3286 * servitude.
3288 break;
3289 case NETDEV_CHANGENAME:
3291 * TODO: handle changing the primary's name
3293 break;
3294 case NETDEV_FEAT_CHANGE:
3295 bond_compute_features(bond);
3296 break;
3297 default:
3298 break;
3301 return NOTIFY_DONE;
3305 * bond_netdev_event: handle netdev notifier chain events.
3307 * This function receives events for the netdev chain. The caller (an
3308 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3309 * locks for us to safely manipulate the slave devices (RTNL lock,
3310 * dev_probe_lock).
3312 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3314 struct net_device *event_dev = (struct net_device *)ptr;
3316 dprintk("event_dev: %s, event: %lx\n",
3317 (event_dev ? event_dev->name : "None"),
3318 event);
3320 if (!(event_dev->priv_flags & IFF_BONDING))
3321 return NOTIFY_DONE;
3323 if (event_dev->flags & IFF_MASTER) {
3324 dprintk("IFF_MASTER\n");
3325 return bond_master_netdev_event(event, event_dev);
3328 if (event_dev->flags & IFF_SLAVE) {
3329 dprintk("IFF_SLAVE\n");
3330 return bond_slave_netdev_event(event, event_dev);
3333 return NOTIFY_DONE;
3337 * bond_inetaddr_event: handle inetaddr notifier chain events.
3339 * We keep track of device IPs primarily to use as source addresses in
3340 * ARP monitor probes (rather than spewing out broadcasts all the time).
3342 * We track one IP for the main device (if it has one), plus one per VLAN.
3344 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3346 struct in_ifaddr *ifa = ptr;
3347 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3348 struct bonding *bond, *bond_next;
3349 struct vlan_entry *vlan, *vlan_next;
3351 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3352 if (bond->dev == event_dev) {
3353 switch (event) {
3354 case NETDEV_UP:
3355 bond->master_ip = ifa->ifa_local;
3356 return NOTIFY_OK;
3357 case NETDEV_DOWN:
3358 bond->master_ip = bond_glean_dev_ip(bond->dev);
3359 return NOTIFY_OK;
3360 default:
3361 return NOTIFY_DONE;
3365 if (list_empty(&bond->vlan_list))
3366 continue;
3368 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3369 vlan_list) {
3370 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3371 if (vlan_dev == event_dev) {
3372 switch (event) {
3373 case NETDEV_UP:
3374 vlan->vlan_ip = ifa->ifa_local;
3375 return NOTIFY_OK;
3376 case NETDEV_DOWN:
3377 vlan->vlan_ip =
3378 bond_glean_dev_ip(vlan_dev);
3379 return NOTIFY_OK;
3380 default:
3381 return NOTIFY_DONE;
3386 return NOTIFY_DONE;
3389 static struct notifier_block bond_netdev_notifier = {
3390 .notifier_call = bond_netdev_event,
3393 static struct notifier_block bond_inetaddr_notifier = {
3394 .notifier_call = bond_inetaddr_event,
3397 /*-------------------------- Packet type handling ---------------------------*/
3399 /* register to receive lacpdus on a bond */
3400 static void bond_register_lacpdu(struct bonding *bond)
3402 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3404 /* initialize packet type */
3405 pk_type->type = PKT_TYPE_LACPDU;
3406 pk_type->dev = bond->dev;
3407 pk_type->func = bond_3ad_lacpdu_recv;
3409 dev_add_pack(pk_type);
3412 /* unregister to receive lacpdus on a bond */
3413 static void bond_unregister_lacpdu(struct bonding *bond)
3415 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3418 void bond_register_arp(struct bonding *bond)
3420 struct packet_type *pt = &bond->arp_mon_pt;
3422 pt->type = htons(ETH_P_ARP);
3423 pt->dev = NULL; /*bond->dev;XXX*/
3424 pt->func = bond_arp_rcv;
3425 dev_add_pack(pt);
3428 void bond_unregister_arp(struct bonding *bond)
3430 dev_remove_pack(&bond->arp_mon_pt);
3433 /*---------------------------- Hashing Policies -----------------------------*/
3436 * Hash for the the output device based upon layer 3 and layer 4 data. If
3437 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3438 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3440 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3441 struct net_device *bond_dev, int count)
3443 struct ethhdr *data = (struct ethhdr *)skb->data;
3444 struct iphdr *iph = skb->nh.iph;
3445 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3446 int layer4_xor = 0;
3448 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3449 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3450 (iph->protocol == IPPROTO_TCP ||
3451 iph->protocol == IPPROTO_UDP)) {
3452 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3454 return (layer4_xor ^
3455 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3459 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3463 * Hash for the output device based upon layer 2 data
3465 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3466 struct net_device *bond_dev, int count)
3468 struct ethhdr *data = (struct ethhdr *)skb->data;
3470 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3473 /*-------------------------- Device entry points ----------------------------*/
3475 static int bond_open(struct net_device *bond_dev)
3477 struct bonding *bond = bond_dev->priv;
3478 struct timer_list *mii_timer = &bond->mii_timer;
3479 struct timer_list *arp_timer = &bond->arp_timer;
3481 bond->kill_timers = 0;
3483 if ((bond->params.mode == BOND_MODE_TLB) ||
3484 (bond->params.mode == BOND_MODE_ALB)) {
3485 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3487 /* bond_alb_initialize must be called before the timer
3488 * is started.
3490 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3491 /* something went wrong - fail the open operation */
3492 return -1;
3495 init_timer(alb_timer);
3496 alb_timer->expires = jiffies + 1;
3497 alb_timer->data = (unsigned long)bond;
3498 alb_timer->function = (void *)&bond_alb_monitor;
3499 add_timer(alb_timer);
3502 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3503 init_timer(mii_timer);
3504 mii_timer->expires = jiffies + 1;
3505 mii_timer->data = (unsigned long)bond_dev;
3506 mii_timer->function = (void *)&bond_mii_monitor;
3507 add_timer(mii_timer);
3510 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3511 init_timer(arp_timer);
3512 arp_timer->expires = jiffies + 1;
3513 arp_timer->data = (unsigned long)bond_dev;
3514 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3515 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3516 } else {
3517 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3519 if (bond->params.arp_validate)
3520 bond_register_arp(bond);
3522 add_timer(arp_timer);
3525 if (bond->params.mode == BOND_MODE_8023AD) {
3526 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3527 init_timer(ad_timer);
3528 ad_timer->expires = jiffies + 1;
3529 ad_timer->data = (unsigned long)bond;
3530 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3531 add_timer(ad_timer);
3533 /* register to receive LACPDUs */
3534 bond_register_lacpdu(bond);
3537 return 0;
3540 static int bond_close(struct net_device *bond_dev)
3542 struct bonding *bond = bond_dev->priv;
3544 if (bond->params.mode == BOND_MODE_8023AD) {
3545 /* Unregister the receive of LACPDUs */
3546 bond_unregister_lacpdu(bond);
3549 if (bond->params.arp_validate)
3550 bond_unregister_arp(bond);
3552 write_lock_bh(&bond->lock);
3555 /* signal timers not to re-arm */
3556 bond->kill_timers = 1;
3558 write_unlock_bh(&bond->lock);
3560 /* del_timer_sync must run without holding the bond->lock
3561 * because a running timer might be trying to hold it too
3564 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3565 del_timer_sync(&bond->mii_timer);
3568 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3569 del_timer_sync(&bond->arp_timer);
3572 switch (bond->params.mode) {
3573 case BOND_MODE_8023AD:
3574 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3575 break;
3576 case BOND_MODE_TLB:
3577 case BOND_MODE_ALB:
3578 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3579 break;
3580 default:
3581 break;
3585 if ((bond->params.mode == BOND_MODE_TLB) ||
3586 (bond->params.mode == BOND_MODE_ALB)) {
3587 /* Must be called only after all
3588 * slaves have been released
3590 bond_alb_deinitialize(bond);
3593 return 0;
3596 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3598 struct bonding *bond = bond_dev->priv;
3599 struct net_device_stats *stats = &(bond->stats), *sstats;
3600 struct slave *slave;
3601 int i;
3603 memset(stats, 0, sizeof(struct net_device_stats));
3605 read_lock_bh(&bond->lock);
3607 bond_for_each_slave(bond, slave, i) {
3608 sstats = slave->dev->get_stats(slave->dev);
3610 stats->rx_packets += sstats->rx_packets;
3611 stats->rx_bytes += sstats->rx_bytes;
3612 stats->rx_errors += sstats->rx_errors;
3613 stats->rx_dropped += sstats->rx_dropped;
3615 stats->tx_packets += sstats->tx_packets;
3616 stats->tx_bytes += sstats->tx_bytes;
3617 stats->tx_errors += sstats->tx_errors;
3618 stats->tx_dropped += sstats->tx_dropped;
3620 stats->multicast += sstats->multicast;
3621 stats->collisions += sstats->collisions;
3623 stats->rx_length_errors += sstats->rx_length_errors;
3624 stats->rx_over_errors += sstats->rx_over_errors;
3625 stats->rx_crc_errors += sstats->rx_crc_errors;
3626 stats->rx_frame_errors += sstats->rx_frame_errors;
3627 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3628 stats->rx_missed_errors += sstats->rx_missed_errors;
3630 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3631 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3632 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3633 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3634 stats->tx_window_errors += sstats->tx_window_errors;
3637 read_unlock_bh(&bond->lock);
3639 return stats;
3642 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3644 struct net_device *slave_dev = NULL;
3645 struct ifbond k_binfo;
3646 struct ifbond __user *u_binfo = NULL;
3647 struct ifslave k_sinfo;
3648 struct ifslave __user *u_sinfo = NULL;
3649 struct mii_ioctl_data *mii = NULL;
3650 int res = 0;
3652 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3653 bond_dev->name, cmd);
3655 switch (cmd) {
3656 case SIOCGMIIPHY:
3657 mii = if_mii(ifr);
3658 if (!mii) {
3659 return -EINVAL;
3661 mii->phy_id = 0;
3662 /* Fall Through */
3663 case SIOCGMIIREG:
3665 * We do this again just in case we were called by SIOCGMIIREG
3666 * instead of SIOCGMIIPHY.
3668 mii = if_mii(ifr);
3669 if (!mii) {
3670 return -EINVAL;
3673 if (mii->reg_num == 1) {
3674 struct bonding *bond = bond_dev->priv;
3675 mii->val_out = 0;
3676 read_lock_bh(&bond->lock);
3677 read_lock(&bond->curr_slave_lock);
3678 if (bond->curr_active_slave) {
3679 mii->val_out = BMSR_LSTATUS;
3681 read_unlock(&bond->curr_slave_lock);
3682 read_unlock_bh(&bond->lock);
3685 return 0;
3686 case BOND_INFO_QUERY_OLD:
3687 case SIOCBONDINFOQUERY:
3688 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3690 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3691 return -EFAULT;
3694 res = bond_info_query(bond_dev, &k_binfo);
3695 if (res == 0) {
3696 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3697 return -EFAULT;
3701 return res;
3702 case BOND_SLAVE_INFO_QUERY_OLD:
3703 case SIOCBONDSLAVEINFOQUERY:
3704 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3706 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3707 return -EFAULT;
3710 res = bond_slave_info_query(bond_dev, &k_sinfo);
3711 if (res == 0) {
3712 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3713 return -EFAULT;
3717 return res;
3718 default:
3719 /* Go on */
3720 break;
3723 if (!capable(CAP_NET_ADMIN)) {
3724 return -EPERM;
3727 down_write(&(bonding_rwsem));
3728 slave_dev = dev_get_by_name(ifr->ifr_slave);
3730 dprintk("slave_dev=%p: \n", slave_dev);
3732 if (!slave_dev) {
3733 res = -ENODEV;
3734 } else {
3735 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3736 switch (cmd) {
3737 case BOND_ENSLAVE_OLD:
3738 case SIOCBONDENSLAVE:
3739 res = bond_enslave(bond_dev, slave_dev);
3740 break;
3741 case BOND_RELEASE_OLD:
3742 case SIOCBONDRELEASE:
3743 res = bond_release(bond_dev, slave_dev);
3744 break;
3745 case BOND_SETHWADDR_OLD:
3746 case SIOCBONDSETHWADDR:
3747 res = bond_sethwaddr(bond_dev, slave_dev);
3748 break;
3749 case BOND_CHANGE_ACTIVE_OLD:
3750 case SIOCBONDCHANGEACTIVE:
3751 res = bond_ioctl_change_active(bond_dev, slave_dev);
3752 break;
3753 default:
3754 res = -EOPNOTSUPP;
3757 dev_put(slave_dev);
3760 up_write(&(bonding_rwsem));
3761 return res;
3764 static void bond_set_multicast_list(struct net_device *bond_dev)
3766 struct bonding *bond = bond_dev->priv;
3767 struct dev_mc_list *dmi;
3769 write_lock_bh(&bond->lock);
3772 * Do promisc before checking multicast_mode
3774 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3775 bond_set_promiscuity(bond, 1);
3778 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3779 bond_set_promiscuity(bond, -1);
3782 /* set allmulti flag to slaves */
3783 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3784 bond_set_allmulti(bond, 1);
3787 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3788 bond_set_allmulti(bond, -1);
3791 bond->flags = bond_dev->flags;
3793 /* looking for addresses to add to slaves' mc list */
3794 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3795 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3796 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3800 /* looking for addresses to delete from slaves' list */
3801 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3802 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3803 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3807 /* save master's multicast list */
3808 bond_mc_list_destroy(bond);
3809 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3811 write_unlock_bh(&bond->lock);
3815 * Change the MTU of all of a master's slaves to match the master
3817 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3819 struct bonding *bond = bond_dev->priv;
3820 struct slave *slave, *stop_at;
3821 int res = 0;
3822 int i;
3824 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3825 (bond_dev ? bond_dev->name : "None"), new_mtu);
3827 /* Can't hold bond->lock with bh disabled here since
3828 * some base drivers panic. On the other hand we can't
3829 * hold bond->lock without bh disabled because we'll
3830 * deadlock. The only solution is to rely on the fact
3831 * that we're under rtnl_lock here, and the slaves
3832 * list won't change. This doesn't solve the problem
3833 * of setting the slave's MTU while it is
3834 * transmitting, but the assumption is that the base
3835 * driver can handle that.
3837 * TODO: figure out a way to safely iterate the slaves
3838 * list, but without holding a lock around the actual
3839 * call to the base driver.
3842 bond_for_each_slave(bond, slave, i) {
3843 dprintk("s %p s->p %p c_m %p\n", slave,
3844 slave->prev, slave->dev->change_mtu);
3846 res = dev_set_mtu(slave->dev, new_mtu);
3848 if (res) {
3849 /* If we failed to set the slave's mtu to the new value
3850 * we must abort the operation even in ACTIVE_BACKUP
3851 * mode, because if we allow the backup slaves to have
3852 * different mtu values than the active slave we'll
3853 * need to change their mtu when doing a failover. That
3854 * means changing their mtu from timer context, which
3855 * is probably not a good idea.
3857 dprintk("err %d %s\n", res, slave->dev->name);
3858 goto unwind;
3862 bond_dev->mtu = new_mtu;
3864 return 0;
3866 unwind:
3867 /* unwind from head to the slave that failed */
3868 stop_at = slave;
3869 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3870 int tmp_res;
3872 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3873 if (tmp_res) {
3874 dprintk("unwind err %d dev %s\n", tmp_res,
3875 slave->dev->name);
3879 return res;
3883 * Change HW address
3885 * Note that many devices must be down to change the HW address, and
3886 * downing the master releases all slaves. We can make bonds full of
3887 * bonding devices to test this, however.
3889 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3891 struct bonding *bond = bond_dev->priv;
3892 struct sockaddr *sa = addr, tmp_sa;
3893 struct slave *slave, *stop_at;
3894 int res = 0;
3895 int i;
3897 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3899 if (!is_valid_ether_addr(sa->sa_data)) {
3900 return -EADDRNOTAVAIL;
3903 /* Can't hold bond->lock with bh disabled here since
3904 * some base drivers panic. On the other hand we can't
3905 * hold bond->lock without bh disabled because we'll
3906 * deadlock. The only solution is to rely on the fact
3907 * that we're under rtnl_lock here, and the slaves
3908 * list won't change. This doesn't solve the problem
3909 * of setting the slave's hw address while it is
3910 * transmitting, but the assumption is that the base
3911 * driver can handle that.
3913 * TODO: figure out a way to safely iterate the slaves
3914 * list, but without holding a lock around the actual
3915 * call to the base driver.
3918 bond_for_each_slave(bond, slave, i) {
3919 dprintk("slave %p %s\n", slave, slave->dev->name);
3921 if (slave->dev->set_mac_address == NULL) {
3922 res = -EOPNOTSUPP;
3923 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3924 goto unwind;
3927 res = dev_set_mac_address(slave->dev, addr);
3928 if (res) {
3929 /* TODO: consider downing the slave
3930 * and retry ?
3931 * User should expect communications
3932 * breakage anyway until ARP finish
3933 * updating, so...
3935 dprintk("err %d %s\n", res, slave->dev->name);
3936 goto unwind;
3940 /* success */
3941 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3942 return 0;
3944 unwind:
3945 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3946 tmp_sa.sa_family = bond_dev->type;
3948 /* unwind from head to the slave that failed */
3949 stop_at = slave;
3950 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3951 int tmp_res;
3953 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3954 if (tmp_res) {
3955 dprintk("unwind err %d dev %s\n", tmp_res,
3956 slave->dev->name);
3960 return res;
3963 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3965 struct bonding *bond = bond_dev->priv;
3966 struct slave *slave, *start_at;
3967 int i;
3968 int res = 1;
3970 read_lock(&bond->lock);
3972 if (!BOND_IS_OK(bond)) {
3973 goto out;
3976 read_lock(&bond->curr_slave_lock);
3977 slave = start_at = bond->curr_active_slave;
3978 read_unlock(&bond->curr_slave_lock);
3980 if (!slave) {
3981 goto out;
3984 bond_for_each_slave_from(bond, slave, i, start_at) {
3985 if (IS_UP(slave->dev) &&
3986 (slave->link == BOND_LINK_UP) &&
3987 (slave->state == BOND_STATE_ACTIVE)) {
3988 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3990 write_lock(&bond->curr_slave_lock);
3991 bond->curr_active_slave = slave->next;
3992 write_unlock(&bond->curr_slave_lock);
3994 break;
3999 out:
4000 if (res) {
4001 /* no suitable interface, frame not sent */
4002 dev_kfree_skb(skb);
4004 read_unlock(&bond->lock);
4005 return 0;
4008 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
4009 struct bonding *bond,
4010 struct slave *slave)
4012 struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
4013 struct ethhdr *eth_data;
4014 u8 *hwaddr;
4015 int res;
4017 if (!skb2) {
4018 printk(KERN_ERR DRV_NAME ": Error: "
4019 "bond_activebackup_xmit_copy(): skb_copy() failed\n");
4020 return;
4023 skb2->mac.raw = (unsigned char *)skb2->data;
4024 eth_data = eth_hdr(skb2);
4026 /* Pick an appropriate source MAC address
4027 * -- use slave's perm MAC addr, unless used by bond
4028 * -- otherwise, borrow active slave's perm MAC addr
4029 * since that will not be used
4031 hwaddr = slave->perm_hwaddr;
4032 if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
4033 hwaddr = bond->curr_active_slave->perm_hwaddr;
4035 /* Set source MAC address appropriately */
4036 memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
4038 res = bond_dev_queue_xmit(bond, skb2, slave->dev);
4039 if (res)
4040 dev_kfree_skb(skb2);
4042 return;
4046 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4047 * the bond has a usable interface.
4049 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4051 struct bonding *bond = bond_dev->priv;
4052 int res = 1;
4054 read_lock(&bond->lock);
4055 read_lock(&bond->curr_slave_lock);
4057 if (!BOND_IS_OK(bond)) {
4058 goto out;
4061 if (!bond->curr_active_slave)
4062 goto out;
4064 /* Xmit IGMP frames on all slaves to ensure rapid fail-over
4065 for multicast traffic on snooping switches */
4066 if (skb->protocol == __constant_htons(ETH_P_IP) &&
4067 skb->nh.iph->protocol == IPPROTO_IGMP) {
4068 struct slave *slave, *active_slave;
4069 int i;
4071 active_slave = bond->curr_active_slave;
4072 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
4073 active_slave->prev)
4074 if (IS_UP(slave->dev) &&
4075 (slave->link == BOND_LINK_UP))
4076 bond_activebackup_xmit_copy(skb, bond, slave);
4079 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4081 out:
4082 if (res) {
4083 /* no suitable interface, frame not sent */
4084 dev_kfree_skb(skb);
4086 read_unlock(&bond->curr_slave_lock);
4087 read_unlock(&bond->lock);
4088 return 0;
4092 * In bond_xmit_xor() , we determine the output device by using a pre-
4093 * determined xmit_hash_policy(), If the selected device is not enabled,
4094 * find the next active slave.
4096 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4098 struct bonding *bond = bond_dev->priv;
4099 struct slave *slave, *start_at;
4100 int slave_no;
4101 int i;
4102 int res = 1;
4104 read_lock(&bond->lock);
4106 if (!BOND_IS_OK(bond)) {
4107 goto out;
4110 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4112 bond_for_each_slave(bond, slave, i) {
4113 slave_no--;
4114 if (slave_no < 0) {
4115 break;
4119 start_at = slave;
4121 bond_for_each_slave_from(bond, slave, i, start_at) {
4122 if (IS_UP(slave->dev) &&
4123 (slave->link == BOND_LINK_UP) &&
4124 (slave->state == BOND_STATE_ACTIVE)) {
4125 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4126 break;
4130 out:
4131 if (res) {
4132 /* no suitable interface, frame not sent */
4133 dev_kfree_skb(skb);
4135 read_unlock(&bond->lock);
4136 return 0;
4140 * in broadcast mode, we send everything to all usable interfaces.
4142 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4144 struct bonding *bond = bond_dev->priv;
4145 struct slave *slave, *start_at;
4146 struct net_device *tx_dev = NULL;
4147 int i;
4148 int res = 1;
4150 read_lock(&bond->lock);
4152 if (!BOND_IS_OK(bond)) {
4153 goto out;
4156 read_lock(&bond->curr_slave_lock);
4157 start_at = bond->curr_active_slave;
4158 read_unlock(&bond->curr_slave_lock);
4160 if (!start_at) {
4161 goto out;
4164 bond_for_each_slave_from(bond, slave, i, start_at) {
4165 if (IS_UP(slave->dev) &&
4166 (slave->link == BOND_LINK_UP) &&
4167 (slave->state == BOND_STATE_ACTIVE)) {
4168 if (tx_dev) {
4169 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4170 if (!skb2) {
4171 printk(KERN_ERR DRV_NAME
4172 ": %s: Error: bond_xmit_broadcast(): "
4173 "skb_clone() failed\n",
4174 bond_dev->name);
4175 continue;
4178 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4179 if (res) {
4180 dev_kfree_skb(skb2);
4181 continue;
4184 tx_dev = slave->dev;
4188 if (tx_dev) {
4189 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4192 out:
4193 if (res) {
4194 /* no suitable interface, frame not sent */
4195 dev_kfree_skb(skb);
4197 /* frame sent to all suitable interfaces */
4198 read_unlock(&bond->lock);
4199 return 0;
4202 /*------------------------- Device initialization ---------------------------*/
4205 * set bond mode specific net device operations
4207 void bond_set_mode_ops(struct bonding *bond, int mode)
4209 struct net_device *bond_dev = bond->dev;
4211 switch (mode) {
4212 case BOND_MODE_ROUNDROBIN:
4213 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4214 break;
4215 case BOND_MODE_ACTIVEBACKUP:
4216 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4217 break;
4218 case BOND_MODE_XOR:
4219 bond_dev->hard_start_xmit = bond_xmit_xor;
4220 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4221 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4222 else
4223 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4224 break;
4225 case BOND_MODE_BROADCAST:
4226 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4227 break;
4228 case BOND_MODE_8023AD:
4229 bond_set_master_3ad_flags(bond);
4230 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4231 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4232 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4233 else
4234 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4235 break;
4236 case BOND_MODE_ALB:
4237 bond_set_master_alb_flags(bond);
4238 /* FALLTHRU */
4239 case BOND_MODE_TLB:
4240 bond_dev->hard_start_xmit = bond_alb_xmit;
4241 bond_dev->set_mac_address = bond_alb_set_mac_address;
4242 break;
4243 default:
4244 /* Should never happen, mode already checked */
4245 printk(KERN_ERR DRV_NAME
4246 ": %s: Error: Unknown bonding mode %d\n",
4247 bond_dev->name,
4248 mode);
4249 break;
4253 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4254 struct ethtool_drvinfo *drvinfo)
4256 strncpy(drvinfo->driver, DRV_NAME, 32);
4257 strncpy(drvinfo->version, DRV_VERSION, 32);
4258 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4261 static const struct ethtool_ops bond_ethtool_ops = {
4262 .get_tx_csum = ethtool_op_get_tx_csum,
4263 .get_tso = ethtool_op_get_tso,
4264 .get_ufo = ethtool_op_get_ufo,
4265 .get_sg = ethtool_op_get_sg,
4266 .get_drvinfo = bond_ethtool_get_drvinfo,
4270 * Does not allocate but creates a /proc entry.
4271 * Allowed to fail.
4273 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4275 struct bonding *bond = bond_dev->priv;
4277 dprintk("Begin bond_init for %s\n", bond_dev->name);
4279 /* initialize rwlocks */
4280 rwlock_init(&bond->lock);
4281 rwlock_init(&bond->curr_slave_lock);
4283 bond->params = *params; /* copy params struct */
4285 /* Initialize pointers */
4286 bond->first_slave = NULL;
4287 bond->curr_active_slave = NULL;
4288 bond->current_arp_slave = NULL;
4289 bond->primary_slave = NULL;
4290 bond->dev = bond_dev;
4291 INIT_LIST_HEAD(&bond->vlan_list);
4293 /* Initialize the device entry points */
4294 bond_dev->open = bond_open;
4295 bond_dev->stop = bond_close;
4296 bond_dev->get_stats = bond_get_stats;
4297 bond_dev->do_ioctl = bond_do_ioctl;
4298 bond_dev->ethtool_ops = &bond_ethtool_ops;
4299 bond_dev->set_multicast_list = bond_set_multicast_list;
4300 bond_dev->change_mtu = bond_change_mtu;
4301 bond_dev->set_mac_address = bond_set_mac_address;
4303 bond_set_mode_ops(bond, bond->params.mode);
4305 bond_dev->destructor = free_netdev;
4307 /* Initialize the device options */
4308 bond_dev->tx_queue_len = 0;
4309 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4310 bond_dev->priv_flags |= IFF_BONDING;
4312 /* At first, we block adding VLANs. That's the only way to
4313 * prevent problems that occur when adding VLANs over an
4314 * empty bond. The block will be removed once non-challenged
4315 * slaves are enslaved.
4317 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4319 /* don't acquire bond device's netif_tx_lock when
4320 * transmitting */
4321 bond_dev->features |= NETIF_F_LLTX;
4323 /* By default, we declare the bond to be fully
4324 * VLAN hardware accelerated capable. Special
4325 * care is taken in the various xmit functions
4326 * when there are slaves that are not hw accel
4327 * capable
4329 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4330 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4331 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4332 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4333 NETIF_F_HW_VLAN_RX |
4334 NETIF_F_HW_VLAN_FILTER);
4336 #ifdef CONFIG_PROC_FS
4337 bond_create_proc_entry(bond);
4338 #endif
4340 list_add_tail(&bond->bond_list, &bond_dev_list);
4342 return 0;
4345 /* De-initialize device specific data.
4346 * Caller must hold rtnl_lock.
4348 void bond_deinit(struct net_device *bond_dev)
4350 struct bonding *bond = bond_dev->priv;
4352 list_del(&bond->bond_list);
4354 #ifdef CONFIG_PROC_FS
4355 bond_remove_proc_entry(bond);
4356 #endif
4359 /* Unregister and free all bond devices.
4360 * Caller must hold rtnl_lock.
4362 static void bond_free_all(void)
4364 struct bonding *bond, *nxt;
4366 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4367 struct net_device *bond_dev = bond->dev;
4369 bond_mc_list_destroy(bond);
4370 /* Release the bonded slaves */
4371 bond_release_all(bond_dev);
4372 unregister_netdevice(bond_dev);
4373 bond_deinit(bond_dev);
4376 #ifdef CONFIG_PROC_FS
4377 bond_destroy_proc_dir();
4378 #endif
4381 /*------------------------- Module initialization ---------------------------*/
4384 * Convert string input module parms. Accept either the
4385 * number of the mode or its string name.
4387 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4389 int i;
4391 for (i = 0; tbl[i].modename; i++) {
4392 if ((isdigit(*mode_arg) &&
4393 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4394 (strncmp(mode_arg, tbl[i].modename,
4395 strlen(tbl[i].modename)) == 0)) {
4396 return tbl[i].mode;
4400 return -1;
4403 static int bond_check_params(struct bond_params *params)
4405 int arp_validate_value;
4408 * Convert string parameters.
4410 if (mode) {
4411 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4412 if (bond_mode == -1) {
4413 printk(KERN_ERR DRV_NAME
4414 ": Error: Invalid bonding mode \"%s\"\n",
4415 mode == NULL ? "NULL" : mode);
4416 return -EINVAL;
4420 if (xmit_hash_policy) {
4421 if ((bond_mode != BOND_MODE_XOR) &&
4422 (bond_mode != BOND_MODE_8023AD)) {
4423 printk(KERN_INFO DRV_NAME
4424 ": xor_mode param is irrelevant in mode %s\n",
4425 bond_mode_name(bond_mode));
4426 } else {
4427 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4428 xmit_hashtype_tbl);
4429 if (xmit_hashtype == -1) {
4430 printk(KERN_ERR DRV_NAME
4431 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4432 xmit_hash_policy == NULL ? "NULL" :
4433 xmit_hash_policy);
4434 return -EINVAL;
4439 if (lacp_rate) {
4440 if (bond_mode != BOND_MODE_8023AD) {
4441 printk(KERN_INFO DRV_NAME
4442 ": lacp_rate param is irrelevant in mode %s\n",
4443 bond_mode_name(bond_mode));
4444 } else {
4445 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4446 if (lacp_fast == -1) {
4447 printk(KERN_ERR DRV_NAME
4448 ": Error: Invalid lacp rate \"%s\"\n",
4449 lacp_rate == NULL ? "NULL" : lacp_rate);
4450 return -EINVAL;
4455 if (max_bonds < 1 || max_bonds > INT_MAX) {
4456 printk(KERN_WARNING DRV_NAME
4457 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4458 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4459 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4460 max_bonds = BOND_DEFAULT_MAX_BONDS;
4463 if (miimon < 0) {
4464 printk(KERN_WARNING DRV_NAME
4465 ": Warning: miimon module parameter (%d), "
4466 "not in range 0-%d, so it was reset to %d\n",
4467 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4468 miimon = BOND_LINK_MON_INTERV;
4471 if (updelay < 0) {
4472 printk(KERN_WARNING DRV_NAME
4473 ": Warning: updelay module parameter (%d), "
4474 "not in range 0-%d, so it was reset to 0\n",
4475 updelay, INT_MAX);
4476 updelay = 0;
4479 if (downdelay < 0) {
4480 printk(KERN_WARNING DRV_NAME
4481 ": Warning: downdelay module parameter (%d), "
4482 "not in range 0-%d, so it was reset to 0\n",
4483 downdelay, INT_MAX);
4484 downdelay = 0;
4487 if ((use_carrier != 0) && (use_carrier != 1)) {
4488 printk(KERN_WARNING DRV_NAME
4489 ": Warning: use_carrier module parameter (%d), "
4490 "not of valid value (0/1), so it was set to 1\n",
4491 use_carrier);
4492 use_carrier = 1;
4495 /* reset values for 802.3ad */
4496 if (bond_mode == BOND_MODE_8023AD) {
4497 if (!miimon) {
4498 printk(KERN_WARNING DRV_NAME
4499 ": Warning: miimon must be specified, "
4500 "otherwise bonding will not detect link "
4501 "failure, speed and duplex which are "
4502 "essential for 802.3ad operation\n");
4503 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4504 miimon = 100;
4508 /* reset values for TLB/ALB */
4509 if ((bond_mode == BOND_MODE_TLB) ||
4510 (bond_mode == BOND_MODE_ALB)) {
4511 if (!miimon) {
4512 printk(KERN_WARNING DRV_NAME
4513 ": Warning: miimon must be specified, "
4514 "otherwise bonding will not detect link "
4515 "failure and link speed which are essential "
4516 "for TLB/ALB load balancing\n");
4517 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4518 miimon = 100;
4522 if (bond_mode == BOND_MODE_ALB) {
4523 printk(KERN_NOTICE DRV_NAME
4524 ": In ALB mode you might experience client "
4525 "disconnections upon reconnection of a link if the "
4526 "bonding module updelay parameter (%d msec) is "
4527 "incompatible with the forwarding delay time of the "
4528 "switch\n",
4529 updelay);
4532 if (!miimon) {
4533 if (updelay || downdelay) {
4534 /* just warn the user the up/down delay will have
4535 * no effect since miimon is zero...
4537 printk(KERN_WARNING DRV_NAME
4538 ": Warning: miimon module parameter not set "
4539 "and updelay (%d) or downdelay (%d) module "
4540 "parameter is set; updelay and downdelay have "
4541 "no effect unless miimon is set\n",
4542 updelay, downdelay);
4544 } else {
4545 /* don't allow arp monitoring */
4546 if (arp_interval) {
4547 printk(KERN_WARNING DRV_NAME
4548 ": Warning: miimon (%d) and arp_interval (%d) "
4549 "can't be used simultaneously, disabling ARP "
4550 "monitoring\n",
4551 miimon, arp_interval);
4552 arp_interval = 0;
4555 if ((updelay % miimon) != 0) {
4556 printk(KERN_WARNING DRV_NAME
4557 ": Warning: updelay (%d) is not a multiple "
4558 "of miimon (%d), updelay rounded to %d ms\n",
4559 updelay, miimon, (updelay / miimon) * miimon);
4562 updelay /= miimon;
4564 if ((downdelay % miimon) != 0) {
4565 printk(KERN_WARNING DRV_NAME
4566 ": Warning: downdelay (%d) is not a multiple "
4567 "of miimon (%d), downdelay rounded to %d ms\n",
4568 downdelay, miimon,
4569 (downdelay / miimon) * miimon);
4572 downdelay /= miimon;
4575 if (arp_interval < 0) {
4576 printk(KERN_WARNING DRV_NAME
4577 ": Warning: arp_interval module parameter (%d) "
4578 ", not in range 0-%d, so it was reset to %d\n",
4579 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4580 arp_interval = BOND_LINK_ARP_INTERV;
4583 for (arp_ip_count = 0;
4584 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4585 arp_ip_count++) {
4586 /* not complete check, but should be good enough to
4587 catch mistakes */
4588 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4589 printk(KERN_WARNING DRV_NAME
4590 ": Warning: bad arp_ip_target module parameter "
4591 "(%s), ARP monitoring will not be performed\n",
4592 arp_ip_target[arp_ip_count]);
4593 arp_interval = 0;
4594 } else {
4595 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4596 arp_target[arp_ip_count] = ip;
4600 if (arp_interval && !arp_ip_count) {
4601 /* don't allow arping if no arp_ip_target given... */
4602 printk(KERN_WARNING DRV_NAME
4603 ": Warning: arp_interval module parameter (%d) "
4604 "specified without providing an arp_ip_target "
4605 "parameter, arp_interval was reset to 0\n",
4606 arp_interval);
4607 arp_interval = 0;
4610 if (arp_validate) {
4611 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4612 printk(KERN_ERR DRV_NAME
4613 ": arp_validate only supported in active-backup mode\n");
4614 return -EINVAL;
4616 if (!arp_interval) {
4617 printk(KERN_ERR DRV_NAME
4618 ": arp_validate requires arp_interval\n");
4619 return -EINVAL;
4622 arp_validate_value = bond_parse_parm(arp_validate,
4623 arp_validate_tbl);
4624 if (arp_validate_value == -1) {
4625 printk(KERN_ERR DRV_NAME
4626 ": Error: invalid arp_validate \"%s\"\n",
4627 arp_validate == NULL ? "NULL" : arp_validate);
4628 return -EINVAL;
4630 } else
4631 arp_validate_value = 0;
4633 if (miimon) {
4634 printk(KERN_INFO DRV_NAME
4635 ": MII link monitoring set to %d ms\n",
4636 miimon);
4637 } else if (arp_interval) {
4638 int i;
4640 printk(KERN_INFO DRV_NAME
4641 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4642 arp_interval,
4643 arp_validate_tbl[arp_validate_value].modename,
4644 arp_ip_count);
4646 for (i = 0; i < arp_ip_count; i++)
4647 printk (" %s", arp_ip_target[i]);
4649 printk("\n");
4651 } else {
4652 /* miimon and arp_interval not set, we need one so things
4653 * work as expected, see bonding.txt for details
4655 printk(KERN_WARNING DRV_NAME
4656 ": Warning: either miimon or arp_interval and "
4657 "arp_ip_target module parameters must be specified, "
4658 "otherwise bonding will not detect link failures! see "
4659 "bonding.txt for details.\n");
4662 if (primary && !USES_PRIMARY(bond_mode)) {
4663 /* currently, using a primary only makes sense
4664 * in active backup, TLB or ALB modes
4666 printk(KERN_WARNING DRV_NAME
4667 ": Warning: %s primary device specified but has no "
4668 "effect in %s mode\n",
4669 primary, bond_mode_name(bond_mode));
4670 primary = NULL;
4673 /* fill params struct with the proper values */
4674 params->mode = bond_mode;
4675 params->xmit_policy = xmit_hashtype;
4676 params->miimon = miimon;
4677 params->arp_interval = arp_interval;
4678 params->arp_validate = arp_validate_value;
4679 params->updelay = updelay;
4680 params->downdelay = downdelay;
4681 params->use_carrier = use_carrier;
4682 params->lacp_fast = lacp_fast;
4683 params->primary[0] = 0;
4685 if (primary) {
4686 strncpy(params->primary, primary, IFNAMSIZ);
4687 params->primary[IFNAMSIZ - 1] = 0;
4690 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4692 return 0;
4695 /* Create a new bond based on the specified name and bonding parameters.
4696 * Caller must NOT hold rtnl_lock; we need to release it here before we
4697 * set up our sysfs entries.
4699 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4701 struct net_device *bond_dev;
4702 int res;
4704 rtnl_lock();
4705 bond_dev = alloc_netdev(sizeof(struct bonding), name, ether_setup);
4706 if (!bond_dev) {
4707 printk(KERN_ERR DRV_NAME
4708 ": %s: eek! can't alloc netdev!\n",
4709 name);
4710 res = -ENOMEM;
4711 goto out_rtnl;
4714 /* bond_init() must be called after dev_alloc_name() (for the
4715 * /proc files), but before register_netdevice(), because we
4716 * need to set function pointers.
4719 res = bond_init(bond_dev, params);
4720 if (res < 0) {
4721 goto out_netdev;
4724 SET_MODULE_OWNER(bond_dev);
4726 res = register_netdevice(bond_dev);
4727 if (res < 0) {
4728 goto out_bond;
4730 if (newbond)
4731 *newbond = bond_dev->priv;
4733 netif_carrier_off(bond_dev);
4735 rtnl_unlock(); /* allows sysfs registration of net device */
4736 res = bond_create_sysfs_entry(bond_dev->priv);
4737 goto done;
4738 out_bond:
4739 bond_deinit(bond_dev);
4740 out_netdev:
4741 free_netdev(bond_dev);
4742 out_rtnl:
4743 rtnl_unlock();
4744 done:
4745 return res;
4748 static int __init bonding_init(void)
4750 int i;
4751 int res;
4752 char new_bond_name[8]; /* Enough room for 999 bonds at init. */
4754 printk(KERN_INFO "%s", version);
4756 res = bond_check_params(&bonding_defaults);
4757 if (res) {
4758 goto out;
4761 #ifdef CONFIG_PROC_FS
4762 bond_create_proc_dir();
4763 #endif
4764 for (i = 0; i < max_bonds; i++) {
4765 sprintf(new_bond_name, "bond%d",i);
4766 res = bond_create(new_bond_name,&bonding_defaults, NULL);
4767 if (res)
4768 goto err;
4771 res = bond_create_sysfs();
4772 if (res)
4773 goto err;
4775 register_netdevice_notifier(&bond_netdev_notifier);
4776 register_inetaddr_notifier(&bond_inetaddr_notifier);
4778 goto out;
4779 err:
4780 rtnl_lock();
4781 bond_free_all();
4782 bond_destroy_sysfs();
4783 rtnl_unlock();
4784 out:
4785 return res;
4789 static void __exit bonding_exit(void)
4791 unregister_netdevice_notifier(&bond_netdev_notifier);
4792 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4794 rtnl_lock();
4795 bond_free_all();
4796 bond_destroy_sysfs();
4797 rtnl_unlock();
4800 module_init(bonding_init);
4801 module_exit(bonding_exit);
4802 MODULE_LICENSE("GPL");
4803 MODULE_VERSION(DRV_VERSION);
4804 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4805 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4806 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4809 * Local variables:
4810 * c-indent-level: 8
4811 * c-basic-offset: 8
4812 * tab-width: 8
4813 * End: