[PATCH] mark struct file_operations const 5
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / net / bonding / bond_main.c
blob61a6fa465d716fba007d839638cf136f9ef5a7ad
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 if (slave_dev->get_stats == NULL) {
1340 printk(KERN_NOTICE DRV_NAME
1341 ": %s: the driver for slave device %s does not provide "
1342 "get_stats function, network statistics will be "
1343 "inaccurate.\n", bond_dev->name, slave_dev->name);
1346 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1347 if (!new_slave) {
1348 res = -ENOMEM;
1349 goto err_undo_flags;
1352 /* save slave's original flags before calling
1353 * netdev_set_master and dev_open
1355 new_slave->original_flags = slave_dev->flags;
1358 * Save slave's original ("permanent") mac address for modes
1359 * that need it, and for restoring it upon release, and then
1360 * set it to the master's address
1362 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1365 * Set slave to master's mac address. The application already
1366 * set the master's mac address to that of the first slave
1368 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1369 addr.sa_family = slave_dev->type;
1370 res = dev_set_mac_address(slave_dev, &addr);
1371 if (res) {
1372 dprintk("Error %d calling set_mac_address\n", res);
1373 goto err_free;
1376 /* open the slave since the application closed it */
1377 res = dev_open(slave_dev);
1378 if (res) {
1379 dprintk("Openning slave %s failed\n", slave_dev->name);
1380 goto err_restore_mac;
1383 res = netdev_set_master(slave_dev, bond_dev);
1384 if (res) {
1385 dprintk("Error %d calling netdev_set_master\n", res);
1386 goto err_close;
1389 new_slave->dev = slave_dev;
1390 slave_dev->priv_flags |= IFF_BONDING;
1392 if ((bond->params.mode == BOND_MODE_TLB) ||
1393 (bond->params.mode == BOND_MODE_ALB)) {
1394 /* bond_alb_init_slave() must be called before all other stages since
1395 * it might fail and we do not want to have to undo everything
1397 res = bond_alb_init_slave(bond, new_slave);
1398 if (res) {
1399 goto err_unset_master;
1403 /* If the mode USES_PRIMARY, then the new slave gets the
1404 * master's promisc (and mc) settings only if it becomes the
1405 * curr_active_slave, and that is taken care of later when calling
1406 * bond_change_active()
1408 if (!USES_PRIMARY(bond->params.mode)) {
1409 /* set promiscuity level to new slave */
1410 if (bond_dev->flags & IFF_PROMISC) {
1411 dev_set_promiscuity(slave_dev, 1);
1414 /* set allmulti level to new slave */
1415 if (bond_dev->flags & IFF_ALLMULTI) {
1416 dev_set_allmulti(slave_dev, 1);
1419 /* upload master's mc_list to new slave */
1420 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1421 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1425 if (bond->params.mode == BOND_MODE_8023AD) {
1426 /* add lacpdu mc addr to mc list */
1427 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1429 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1432 bond_add_vlans_on_slave(bond, slave_dev);
1434 write_lock_bh(&bond->lock);
1436 bond_attach_slave(bond, new_slave);
1438 new_slave->delay = 0;
1439 new_slave->link_failure_count = 0;
1441 bond_compute_features(bond);
1443 new_slave->last_arp_rx = jiffies;
1445 if (bond->params.miimon && !bond->params.use_carrier) {
1446 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1448 if ((link_reporting == -1) && !bond->params.arp_interval) {
1450 * miimon is set but a bonded network driver
1451 * does not support ETHTOOL/MII and
1452 * arp_interval is not set. Note: if
1453 * use_carrier is enabled, we will never go
1454 * here (because netif_carrier is always
1455 * supported); thus, we don't need to change
1456 * the messages for netif_carrier.
1458 printk(KERN_WARNING DRV_NAME
1459 ": %s: Warning: MII and ETHTOOL support not "
1460 "available for interface %s, and "
1461 "arp_interval/arp_ip_target module parameters "
1462 "not specified, thus bonding will not detect "
1463 "link failures! see bonding.txt for details.\n",
1464 bond_dev->name, slave_dev->name);
1465 } else if (link_reporting == -1) {
1466 /* unable get link status using mii/ethtool */
1467 printk(KERN_WARNING DRV_NAME
1468 ": %s: Warning: can't get link status from "
1469 "interface %s; the network driver associated "
1470 "with this interface does not support MII or "
1471 "ETHTOOL link status reporting, thus miimon "
1472 "has no effect on this interface.\n",
1473 bond_dev->name, slave_dev->name);
1477 /* check for initial state */
1478 if (!bond->params.miimon ||
1479 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1480 if (bond->params.updelay) {
1481 dprintk("Initial state of slave_dev is "
1482 "BOND_LINK_BACK\n");
1483 new_slave->link = BOND_LINK_BACK;
1484 new_slave->delay = bond->params.updelay;
1485 } else {
1486 dprintk("Initial state of slave_dev is "
1487 "BOND_LINK_UP\n");
1488 new_slave->link = BOND_LINK_UP;
1490 new_slave->jiffies = jiffies;
1491 } else {
1492 dprintk("Initial state of slave_dev is "
1493 "BOND_LINK_DOWN\n");
1494 new_slave->link = BOND_LINK_DOWN;
1497 if (bond_update_speed_duplex(new_slave) &&
1498 (new_slave->link != BOND_LINK_DOWN)) {
1499 printk(KERN_WARNING DRV_NAME
1500 ": %s: Warning: failed to get speed and duplex from %s, "
1501 "assumed to be 100Mb/sec and Full.\n",
1502 bond_dev->name, new_slave->dev->name);
1504 if (bond->params.mode == BOND_MODE_8023AD) {
1505 printk(KERN_WARNING DRV_NAME
1506 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1507 "support in base driver for proper aggregator "
1508 "selection.\n", bond_dev->name);
1512 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1513 /* if there is a primary slave, remember it */
1514 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1515 bond->primary_slave = new_slave;
1519 switch (bond->params.mode) {
1520 case BOND_MODE_ACTIVEBACKUP:
1521 bond_set_slave_inactive_flags(new_slave);
1522 bond_select_active_slave(bond);
1523 break;
1524 case BOND_MODE_8023AD:
1525 /* in 802.3ad mode, the internal mechanism
1526 * will activate the slaves in the selected
1527 * aggregator
1529 bond_set_slave_inactive_flags(new_slave);
1530 /* if this is the first slave */
1531 if (bond->slave_cnt == 1) {
1532 SLAVE_AD_INFO(new_slave).id = 1;
1533 /* Initialize AD with the number of times that the AD timer is called in 1 second
1534 * can be called only after the mac address of the bond is set
1536 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1537 bond->params.lacp_fast);
1538 } else {
1539 SLAVE_AD_INFO(new_slave).id =
1540 SLAVE_AD_INFO(new_slave->prev).id + 1;
1543 bond_3ad_bind_slave(new_slave);
1544 break;
1545 case BOND_MODE_TLB:
1546 case BOND_MODE_ALB:
1547 new_slave->state = BOND_STATE_ACTIVE;
1548 if ((!bond->curr_active_slave) &&
1549 (new_slave->link != BOND_LINK_DOWN)) {
1550 /* first slave or no active slave yet, and this link
1551 * is OK, so make this interface the active one
1553 bond_change_active_slave(bond, new_slave);
1554 } else {
1555 bond_set_slave_inactive_flags(new_slave);
1557 break;
1558 default:
1559 dprintk("This slave is always active in trunk mode\n");
1561 /* always active in trunk mode */
1562 new_slave->state = BOND_STATE_ACTIVE;
1564 /* In trunking mode there is little meaning to curr_active_slave
1565 * anyway (it holds no special properties of the bond device),
1566 * so we can change it without calling change_active_interface()
1568 if (!bond->curr_active_slave) {
1569 bond->curr_active_slave = new_slave;
1571 break;
1572 } /* switch(bond_mode) */
1574 bond_set_carrier(bond);
1576 write_unlock_bh(&bond->lock);
1578 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1579 if (res)
1580 goto err_unset_master;
1582 printk(KERN_INFO DRV_NAME
1583 ": %s: enslaving %s as a%s interface with a%s link.\n",
1584 bond_dev->name, slave_dev->name,
1585 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1586 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1588 /* enslave is successful */
1589 return 0;
1591 /* Undo stages on error */
1592 err_unset_master:
1593 netdev_set_master(slave_dev, NULL);
1595 err_close:
1596 dev_close(slave_dev);
1598 err_restore_mac:
1599 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1600 addr.sa_family = slave_dev->type;
1601 dev_set_mac_address(slave_dev, &addr);
1603 err_free:
1604 kfree(new_slave);
1606 err_undo_flags:
1607 bond_dev->features = old_features;
1609 return res;
1613 * Try to release the slave device <slave> from the bond device <master>
1614 * It is legal to access curr_active_slave without a lock because all the function
1615 * is write-locked.
1617 * The rules for slave state should be:
1618 * for Active/Backup:
1619 * Active stays on all backups go down
1620 * for Bonded connections:
1621 * The first up interface should be left on and all others downed.
1623 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1625 struct bonding *bond = bond_dev->priv;
1626 struct slave *slave, *oldcurrent;
1627 struct sockaddr addr;
1628 int mac_addr_differ;
1630 /* slave is not a slave or master is not master of this slave */
1631 if (!(slave_dev->flags & IFF_SLAVE) ||
1632 (slave_dev->master != bond_dev)) {
1633 printk(KERN_ERR DRV_NAME
1634 ": %s: Error: cannot release %s.\n",
1635 bond_dev->name, slave_dev->name);
1636 return -EINVAL;
1639 write_lock_bh(&bond->lock);
1641 slave = bond_get_slave_by_dev(bond, slave_dev);
1642 if (!slave) {
1643 /* not a slave of this bond */
1644 printk(KERN_INFO DRV_NAME
1645 ": %s: %s not enslaved\n",
1646 bond_dev->name, slave_dev->name);
1647 write_unlock_bh(&bond->lock);
1648 return -EINVAL;
1651 mac_addr_differ = memcmp(bond_dev->dev_addr,
1652 slave->perm_hwaddr,
1653 ETH_ALEN);
1654 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1655 printk(KERN_WARNING DRV_NAME
1656 ": %s: Warning: the permanent HWaddr of %s "
1657 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1658 "still in use by %s. Set the HWaddr of "
1659 "%s to a different address to avoid "
1660 "conflicts.\n",
1661 bond_dev->name,
1662 slave_dev->name,
1663 slave->perm_hwaddr[0],
1664 slave->perm_hwaddr[1],
1665 slave->perm_hwaddr[2],
1666 slave->perm_hwaddr[3],
1667 slave->perm_hwaddr[4],
1668 slave->perm_hwaddr[5],
1669 bond_dev->name,
1670 slave_dev->name);
1673 /* Inform AD package of unbinding of slave. */
1674 if (bond->params.mode == BOND_MODE_8023AD) {
1675 /* must be called before the slave is
1676 * detached from the list
1678 bond_3ad_unbind_slave(slave);
1681 printk(KERN_INFO DRV_NAME
1682 ": %s: releasing %s interface %s\n",
1683 bond_dev->name,
1684 (slave->state == BOND_STATE_ACTIVE)
1685 ? "active" : "backup",
1686 slave_dev->name);
1688 oldcurrent = bond->curr_active_slave;
1690 bond->current_arp_slave = NULL;
1692 /* release the slave from its bond */
1693 bond_detach_slave(bond, slave);
1695 bond_compute_features(bond);
1697 if (bond->primary_slave == slave) {
1698 bond->primary_slave = NULL;
1701 if (oldcurrent == slave) {
1702 bond_change_active_slave(bond, NULL);
1705 if ((bond->params.mode == BOND_MODE_TLB) ||
1706 (bond->params.mode == BOND_MODE_ALB)) {
1707 /* Must be called only after the slave has been
1708 * detached from the list and the curr_active_slave
1709 * has been cleared (if our_slave == old_current),
1710 * but before a new active slave is selected.
1712 bond_alb_deinit_slave(bond, slave);
1715 if (oldcurrent == slave)
1716 bond_select_active_slave(bond);
1718 if (bond->slave_cnt == 0) {
1719 bond_set_carrier(bond);
1721 /* if the last slave was removed, zero the mac address
1722 * of the master so it will be set by the application
1723 * to the mac address of the first slave
1725 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1727 if (list_empty(&bond->vlan_list)) {
1728 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1729 } else {
1730 printk(KERN_WARNING DRV_NAME
1731 ": %s: Warning: clearing HW address of %s while it "
1732 "still has VLANs.\n",
1733 bond_dev->name, bond_dev->name);
1734 printk(KERN_WARNING DRV_NAME
1735 ": %s: When re-adding slaves, make sure the bond's "
1736 "HW address matches its VLANs'.\n",
1737 bond_dev->name);
1739 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1740 !bond_has_challenged_slaves(bond)) {
1741 printk(KERN_INFO DRV_NAME
1742 ": %s: last VLAN challenged slave %s "
1743 "left bond %s. VLAN blocking is removed\n",
1744 bond_dev->name, slave_dev->name, bond_dev->name);
1745 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1748 write_unlock_bh(&bond->lock);
1750 /* must do this from outside any spinlocks */
1751 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1753 bond_del_vlans_from_slave(bond, slave_dev);
1755 /* If the mode USES_PRIMARY, then we should only remove its
1756 * promisc and mc settings if it was the curr_active_slave, but that was
1757 * already taken care of above when we detached the slave
1759 if (!USES_PRIMARY(bond->params.mode)) {
1760 /* unset promiscuity level from slave */
1761 if (bond_dev->flags & IFF_PROMISC) {
1762 dev_set_promiscuity(slave_dev, -1);
1765 /* unset allmulti level from slave */
1766 if (bond_dev->flags & IFF_ALLMULTI) {
1767 dev_set_allmulti(slave_dev, -1);
1770 /* flush master's mc_list from slave */
1771 bond_mc_list_flush(bond_dev, slave_dev);
1774 netdev_set_master(slave_dev, NULL);
1776 /* close slave before restoring its mac address */
1777 dev_close(slave_dev);
1779 /* restore original ("permanent") mac address */
1780 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1781 addr.sa_family = slave_dev->type;
1782 dev_set_mac_address(slave_dev, &addr);
1784 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1785 IFF_SLAVE_INACTIVE | IFF_BONDING |
1786 IFF_SLAVE_NEEDARP);
1788 kfree(slave);
1790 return 0; /* deletion OK */
1794 * This function releases all slaves.
1796 static int bond_release_all(struct net_device *bond_dev)
1798 struct bonding *bond = bond_dev->priv;
1799 struct slave *slave;
1800 struct net_device *slave_dev;
1801 struct sockaddr addr;
1803 write_lock_bh(&bond->lock);
1805 netif_carrier_off(bond_dev);
1807 if (bond->slave_cnt == 0) {
1808 goto out;
1811 bond->current_arp_slave = NULL;
1812 bond->primary_slave = NULL;
1813 bond_change_active_slave(bond, NULL);
1815 while ((slave = bond->first_slave) != NULL) {
1816 /* Inform AD package of unbinding of slave
1817 * before slave is detached from the list.
1819 if (bond->params.mode == BOND_MODE_8023AD) {
1820 bond_3ad_unbind_slave(slave);
1823 slave_dev = slave->dev;
1824 bond_detach_slave(bond, slave);
1826 if ((bond->params.mode == BOND_MODE_TLB) ||
1827 (bond->params.mode == BOND_MODE_ALB)) {
1828 /* must be called only after the slave
1829 * has been detached from the list
1831 bond_alb_deinit_slave(bond, slave);
1834 bond_compute_features(bond);
1836 /* now that the slave is detached, unlock and perform
1837 * all the undo steps that should not be called from
1838 * within a lock.
1840 write_unlock_bh(&bond->lock);
1842 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1843 bond_del_vlans_from_slave(bond, slave_dev);
1845 /* If the mode USES_PRIMARY, then we should only remove its
1846 * promisc and mc settings if it was the curr_active_slave, but that was
1847 * already taken care of above when we detached the slave
1849 if (!USES_PRIMARY(bond->params.mode)) {
1850 /* unset promiscuity level from slave */
1851 if (bond_dev->flags & IFF_PROMISC) {
1852 dev_set_promiscuity(slave_dev, -1);
1855 /* unset allmulti level from slave */
1856 if (bond_dev->flags & IFF_ALLMULTI) {
1857 dev_set_allmulti(slave_dev, -1);
1860 /* flush master's mc_list from slave */
1861 bond_mc_list_flush(bond_dev, slave_dev);
1864 netdev_set_master(slave_dev, NULL);
1866 /* close slave before restoring its mac address */
1867 dev_close(slave_dev);
1869 /* restore original ("permanent") mac address*/
1870 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1871 addr.sa_family = slave_dev->type;
1872 dev_set_mac_address(slave_dev, &addr);
1874 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1875 IFF_SLAVE_INACTIVE);
1877 kfree(slave);
1879 /* re-acquire the lock before getting the next slave */
1880 write_lock_bh(&bond->lock);
1883 /* zero the mac address of the master so it will be
1884 * set by the application to the mac address of the
1885 * first slave
1887 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1889 if (list_empty(&bond->vlan_list)) {
1890 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1891 } else {
1892 printk(KERN_WARNING DRV_NAME
1893 ": %s: Warning: clearing HW address of %s while it "
1894 "still has VLANs.\n",
1895 bond_dev->name, bond_dev->name);
1896 printk(KERN_WARNING DRV_NAME
1897 ": %s: When re-adding slaves, make sure the bond's "
1898 "HW address matches its VLANs'.\n",
1899 bond_dev->name);
1902 printk(KERN_INFO DRV_NAME
1903 ": %s: released all slaves\n",
1904 bond_dev->name);
1906 out:
1907 write_unlock_bh(&bond->lock);
1909 return 0;
1913 * This function changes the active slave to slave <slave_dev>.
1914 * It returns -EINVAL in the following cases.
1915 * - <slave_dev> is not found in the list.
1916 * - There is not active slave now.
1917 * - <slave_dev> is already active.
1918 * - The link state of <slave_dev> is not BOND_LINK_UP.
1919 * - <slave_dev> is not running.
1920 * In these cases, this fuction does nothing.
1921 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1923 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1925 struct bonding *bond = bond_dev->priv;
1926 struct slave *old_active = NULL;
1927 struct slave *new_active = NULL;
1928 int res = 0;
1930 if (!USES_PRIMARY(bond->params.mode)) {
1931 return -EINVAL;
1934 /* Verify that master_dev is indeed the master of slave_dev */
1935 if (!(slave_dev->flags & IFF_SLAVE) ||
1936 (slave_dev->master != bond_dev)) {
1937 return -EINVAL;
1940 write_lock_bh(&bond->lock);
1942 old_active = bond->curr_active_slave;
1943 new_active = bond_get_slave_by_dev(bond, slave_dev);
1946 * Changing to the current active: do nothing; return success.
1948 if (new_active && (new_active == old_active)) {
1949 write_unlock_bh(&bond->lock);
1950 return 0;
1953 if ((new_active) &&
1954 (old_active) &&
1955 (new_active->link == BOND_LINK_UP) &&
1956 IS_UP(new_active->dev)) {
1957 bond_change_active_slave(bond, new_active);
1958 } else {
1959 res = -EINVAL;
1962 write_unlock_bh(&bond->lock);
1964 return res;
1967 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1969 struct bonding *bond = bond_dev->priv;
1971 info->bond_mode = bond->params.mode;
1972 info->miimon = bond->params.miimon;
1974 read_lock_bh(&bond->lock);
1975 info->num_slaves = bond->slave_cnt;
1976 read_unlock_bh(&bond->lock);
1978 return 0;
1981 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1983 struct bonding *bond = bond_dev->priv;
1984 struct slave *slave;
1985 int i, found = 0;
1987 if (info->slave_id < 0) {
1988 return -ENODEV;
1991 read_lock_bh(&bond->lock);
1993 bond_for_each_slave(bond, slave, i) {
1994 if (i == (int)info->slave_id) {
1995 found = 1;
1996 break;
2000 read_unlock_bh(&bond->lock);
2002 if (found) {
2003 strcpy(info->slave_name, slave->dev->name);
2004 info->link = slave->link;
2005 info->state = slave->state;
2006 info->link_failure_count = slave->link_failure_count;
2007 } else {
2008 return -ENODEV;
2011 return 0;
2014 /*-------------------------------- Monitoring -------------------------------*/
2016 /* this function is called regularly to monitor each slave's link. */
2017 void bond_mii_monitor(struct net_device *bond_dev)
2019 struct bonding *bond = bond_dev->priv;
2020 struct slave *slave, *oldcurrent;
2021 int do_failover = 0;
2022 int delta_in_ticks;
2023 int i;
2025 read_lock(&bond->lock);
2027 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2029 if (bond->kill_timers) {
2030 goto out;
2033 if (bond->slave_cnt == 0) {
2034 goto re_arm;
2037 /* we will try to read the link status of each of our slaves, and
2038 * set their IFF_RUNNING flag appropriately. For each slave not
2039 * supporting MII status, we won't do anything so that a user-space
2040 * program could monitor the link itself if needed.
2043 read_lock(&bond->curr_slave_lock);
2044 oldcurrent = bond->curr_active_slave;
2045 read_unlock(&bond->curr_slave_lock);
2047 bond_for_each_slave(bond, slave, i) {
2048 struct net_device *slave_dev = slave->dev;
2049 int link_state;
2050 u16 old_speed = slave->speed;
2051 u8 old_duplex = slave->duplex;
2053 link_state = bond_check_dev_link(bond, slave_dev, 0);
2055 switch (slave->link) {
2056 case BOND_LINK_UP: /* the link was up */
2057 if (link_state == BMSR_LSTATUS) {
2058 /* link stays up, nothing more to do */
2059 break;
2060 } else { /* link going down */
2061 slave->link = BOND_LINK_FAIL;
2062 slave->delay = bond->params.downdelay;
2064 if (slave->link_failure_count < UINT_MAX) {
2065 slave->link_failure_count++;
2068 if (bond->params.downdelay) {
2069 printk(KERN_INFO DRV_NAME
2070 ": %s: link status down for %s "
2071 "interface %s, disabling it in "
2072 "%d ms.\n",
2073 bond_dev->name,
2074 IS_UP(slave_dev)
2075 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2076 ? ((slave == oldcurrent)
2077 ? "active " : "backup ")
2078 : "")
2079 : "idle ",
2080 slave_dev->name,
2081 bond->params.downdelay * bond->params.miimon);
2084 /* no break ! fall through the BOND_LINK_FAIL test to
2085 ensure proper action to be taken
2087 case BOND_LINK_FAIL: /* the link has just gone down */
2088 if (link_state != BMSR_LSTATUS) {
2089 /* link stays down */
2090 if (slave->delay <= 0) {
2091 /* link down for too long time */
2092 slave->link = BOND_LINK_DOWN;
2094 /* in active/backup mode, we must
2095 * completely disable this interface
2097 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2098 (bond->params.mode == BOND_MODE_8023AD)) {
2099 bond_set_slave_inactive_flags(slave);
2102 printk(KERN_INFO DRV_NAME
2103 ": %s: link status definitely "
2104 "down for interface %s, "
2105 "disabling it\n",
2106 bond_dev->name,
2107 slave_dev->name);
2109 /* notify ad that the link status has changed */
2110 if (bond->params.mode == BOND_MODE_8023AD) {
2111 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2114 if ((bond->params.mode == BOND_MODE_TLB) ||
2115 (bond->params.mode == BOND_MODE_ALB)) {
2116 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2119 if (slave == oldcurrent) {
2120 do_failover = 1;
2122 } else {
2123 slave->delay--;
2125 } else {
2126 /* link up again */
2127 slave->link = BOND_LINK_UP;
2128 slave->jiffies = jiffies;
2129 printk(KERN_INFO DRV_NAME
2130 ": %s: link status up again after %d "
2131 "ms for interface %s.\n",
2132 bond_dev->name,
2133 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2134 slave_dev->name);
2136 break;
2137 case BOND_LINK_DOWN: /* the link was down */
2138 if (link_state != BMSR_LSTATUS) {
2139 /* the link stays down, nothing more to do */
2140 break;
2141 } else { /* link going up */
2142 slave->link = BOND_LINK_BACK;
2143 slave->delay = bond->params.updelay;
2145 if (bond->params.updelay) {
2146 /* if updelay == 0, no need to
2147 advertise about a 0 ms delay */
2148 printk(KERN_INFO DRV_NAME
2149 ": %s: link status up for "
2150 "interface %s, enabling it "
2151 "in %d ms.\n",
2152 bond_dev->name,
2153 slave_dev->name,
2154 bond->params.updelay * bond->params.miimon);
2157 /* no break ! fall through the BOND_LINK_BACK state in
2158 case there's something to do.
2160 case BOND_LINK_BACK: /* the link has just come back */
2161 if (link_state != BMSR_LSTATUS) {
2162 /* link down again */
2163 slave->link = BOND_LINK_DOWN;
2165 printk(KERN_INFO DRV_NAME
2166 ": %s: link status down again after %d "
2167 "ms for interface %s.\n",
2168 bond_dev->name,
2169 (bond->params.updelay - slave->delay) * bond->params.miimon,
2170 slave_dev->name);
2171 } else {
2172 /* link stays up */
2173 if (slave->delay == 0) {
2174 /* now the link has been up for long time enough */
2175 slave->link = BOND_LINK_UP;
2176 slave->jiffies = jiffies;
2178 if (bond->params.mode == BOND_MODE_8023AD) {
2179 /* prevent it from being the active one */
2180 slave->state = BOND_STATE_BACKUP;
2181 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2182 /* make it immediately active */
2183 slave->state = BOND_STATE_ACTIVE;
2184 } else if (slave != bond->primary_slave) {
2185 /* prevent it from being the active one */
2186 slave->state = BOND_STATE_BACKUP;
2189 printk(KERN_INFO DRV_NAME
2190 ": %s: link status definitely "
2191 "up for interface %s.\n",
2192 bond_dev->name,
2193 slave_dev->name);
2195 /* notify ad that the link status has changed */
2196 if (bond->params.mode == BOND_MODE_8023AD) {
2197 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2200 if ((bond->params.mode == BOND_MODE_TLB) ||
2201 (bond->params.mode == BOND_MODE_ALB)) {
2202 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2205 if ((!oldcurrent) ||
2206 (slave == bond->primary_slave)) {
2207 do_failover = 1;
2209 } else {
2210 slave->delay--;
2213 break;
2214 default:
2215 /* Should not happen */
2216 printk(KERN_ERR DRV_NAME
2217 ": %s: Error: %s Illegal value (link=%d)\n",
2218 bond_dev->name,
2219 slave->dev->name,
2220 slave->link);
2221 goto out;
2222 } /* end of switch (slave->link) */
2224 bond_update_speed_duplex(slave);
2226 if (bond->params.mode == BOND_MODE_8023AD) {
2227 if (old_speed != slave->speed) {
2228 bond_3ad_adapter_speed_changed(slave);
2231 if (old_duplex != slave->duplex) {
2232 bond_3ad_adapter_duplex_changed(slave);
2236 } /* end of for */
2238 if (do_failover) {
2239 write_lock(&bond->curr_slave_lock);
2241 bond_select_active_slave(bond);
2243 write_unlock(&bond->curr_slave_lock);
2244 } else
2245 bond_set_carrier(bond);
2247 re_arm:
2248 if (bond->params.miimon) {
2249 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2251 out:
2252 read_unlock(&bond->lock);
2256 static u32 bond_glean_dev_ip(struct net_device *dev)
2258 struct in_device *idev;
2259 struct in_ifaddr *ifa;
2260 __be32 addr = 0;
2262 if (!dev)
2263 return 0;
2265 rcu_read_lock();
2266 idev = __in_dev_get_rcu(dev);
2267 if (!idev)
2268 goto out;
2270 ifa = idev->ifa_list;
2271 if (!ifa)
2272 goto out;
2274 addr = ifa->ifa_local;
2275 out:
2276 rcu_read_unlock();
2277 return addr;
2280 static int bond_has_ip(struct bonding *bond)
2282 struct vlan_entry *vlan, *vlan_next;
2284 if (bond->master_ip)
2285 return 1;
2287 if (list_empty(&bond->vlan_list))
2288 return 0;
2290 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2291 vlan_list) {
2292 if (vlan->vlan_ip)
2293 return 1;
2296 return 0;
2299 static int bond_has_this_ip(struct bonding *bond, u32 ip)
2301 struct vlan_entry *vlan, *vlan_next;
2303 if (ip == bond->master_ip)
2304 return 1;
2306 if (list_empty(&bond->vlan_list))
2307 return 0;
2309 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2310 vlan_list) {
2311 if (ip == vlan->vlan_ip)
2312 return 1;
2315 return 0;
2319 * We go to the (large) trouble of VLAN tagging ARP frames because
2320 * switches in VLAN mode (especially if ports are configured as
2321 * "native" to a VLAN) might not pass non-tagged frames.
2323 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2325 struct sk_buff *skb;
2327 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2328 slave_dev->name, dest_ip, src_ip, vlan_id);
2330 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2331 NULL, slave_dev->dev_addr, NULL);
2333 if (!skb) {
2334 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2335 return;
2337 if (vlan_id) {
2338 skb = vlan_put_tag(skb, vlan_id);
2339 if (!skb) {
2340 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2341 return;
2344 arp_xmit(skb);
2348 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2350 int i, vlan_id, rv;
2351 u32 *targets = bond->params.arp_targets;
2352 struct vlan_entry *vlan, *vlan_next;
2353 struct net_device *vlan_dev;
2354 struct flowi fl;
2355 struct rtable *rt;
2357 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2358 if (!targets[i])
2359 continue;
2360 dprintk("basa: target %x\n", targets[i]);
2361 if (list_empty(&bond->vlan_list)) {
2362 dprintk("basa: empty vlan: arp_send\n");
2363 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2364 bond->master_ip, 0);
2365 continue;
2369 * If VLANs are configured, we do a route lookup to
2370 * determine which VLAN interface would be used, so we
2371 * can tag the ARP with the proper VLAN tag.
2373 memset(&fl, 0, sizeof(fl));
2374 fl.fl4_dst = targets[i];
2375 fl.fl4_tos = RTO_ONLINK;
2377 rv = ip_route_output_key(&rt, &fl);
2378 if (rv) {
2379 if (net_ratelimit()) {
2380 printk(KERN_WARNING DRV_NAME
2381 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2382 bond->dev->name, NIPQUAD(fl.fl4_dst));
2384 continue;
2388 * This target is not on a VLAN
2390 if (rt->u.dst.dev == bond->dev) {
2391 ip_rt_put(rt);
2392 dprintk("basa: rtdev == bond->dev: arp_send\n");
2393 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2394 bond->master_ip, 0);
2395 continue;
2398 vlan_id = 0;
2399 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2400 vlan_list) {
2401 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2402 if (vlan_dev == rt->u.dst.dev) {
2403 vlan_id = vlan->vlan_id;
2404 dprintk("basa: vlan match on %s %d\n",
2405 vlan_dev->name, vlan_id);
2406 break;
2410 if (vlan_id) {
2411 ip_rt_put(rt);
2412 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2413 vlan->vlan_ip, vlan_id);
2414 continue;
2417 if (net_ratelimit()) {
2418 printk(KERN_WARNING DRV_NAME
2419 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2420 bond->dev->name, NIPQUAD(fl.fl4_dst),
2421 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2423 ip_rt_put(rt);
2428 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2429 * for each VLAN above us.
2431 static void bond_send_gratuitous_arp(struct bonding *bond)
2433 struct slave *slave = bond->curr_active_slave;
2434 struct vlan_entry *vlan;
2435 struct net_device *vlan_dev;
2437 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2438 slave ? slave->dev->name : "NULL");
2439 if (!slave)
2440 return;
2442 if (bond->master_ip) {
2443 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2444 bond->master_ip, 0);
2447 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2448 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2449 if (vlan->vlan_ip) {
2450 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2451 vlan->vlan_ip, vlan->vlan_id);
2456 static void bond_validate_arp(struct bonding *bond, struct slave *slave, u32 sip, u32 tip)
2458 int i;
2459 u32 *targets = bond->params.arp_targets;
2461 targets = bond->params.arp_targets;
2462 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2463 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2464 "%u.%u.%u.%u bhti(tip) %d\n",
2465 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2466 bond_has_this_ip(bond, tip));
2467 if (sip == targets[i]) {
2468 if (bond_has_this_ip(bond, tip))
2469 slave->last_arp_rx = jiffies;
2470 return;
2475 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2477 struct arphdr *arp;
2478 struct slave *slave;
2479 struct bonding *bond;
2480 unsigned char *arp_ptr;
2481 u32 sip, tip;
2483 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2484 goto out;
2486 bond = dev->priv;
2487 read_lock(&bond->lock);
2489 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2490 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2491 orig_dev ? orig_dev->name : "NULL");
2493 slave = bond_get_slave_by_dev(bond, orig_dev);
2494 if (!slave || !slave_do_arp_validate(bond, slave))
2495 goto out_unlock;
2497 /* ARP header, plus 2 device addresses, plus 2 IP addresses. */
2498 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2499 (2 * dev->addr_len) +
2500 (2 * sizeof(u32)))))
2501 goto out_unlock;
2503 arp = skb->nh.arph;
2504 if (arp->ar_hln != dev->addr_len ||
2505 skb->pkt_type == PACKET_OTHERHOST ||
2506 skb->pkt_type == PACKET_LOOPBACK ||
2507 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2508 arp->ar_pro != htons(ETH_P_IP) ||
2509 arp->ar_pln != 4)
2510 goto out_unlock;
2512 arp_ptr = (unsigned char *)(arp + 1);
2513 arp_ptr += dev->addr_len;
2514 memcpy(&sip, arp_ptr, 4);
2515 arp_ptr += 4 + dev->addr_len;
2516 memcpy(&tip, arp_ptr, 4);
2518 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2519 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2520 slave->state, bond->params.arp_validate,
2521 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2524 * Backup slaves won't see the ARP reply, but do come through
2525 * here for each ARP probe (so we swap the sip/tip to validate
2526 * the probe). In a "redundant switch, common router" type of
2527 * configuration, the ARP probe will (hopefully) travel from
2528 * the active, through one switch, the router, then the other
2529 * switch before reaching the backup.
2531 if (slave->state == BOND_STATE_ACTIVE)
2532 bond_validate_arp(bond, slave, sip, tip);
2533 else
2534 bond_validate_arp(bond, slave, tip, sip);
2536 out_unlock:
2537 read_unlock(&bond->lock);
2538 out:
2539 dev_kfree_skb(skb);
2540 return NET_RX_SUCCESS;
2544 * this function is called regularly to monitor each slave's link
2545 * ensuring that traffic is being sent and received when arp monitoring
2546 * is used in load-balancing mode. if the adapter has been dormant, then an
2547 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2548 * arp monitoring in active backup mode.
2550 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2552 struct bonding *bond = bond_dev->priv;
2553 struct slave *slave, *oldcurrent;
2554 int do_failover = 0;
2555 int delta_in_ticks;
2556 int i;
2558 read_lock(&bond->lock);
2560 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2562 if (bond->kill_timers) {
2563 goto out;
2566 if (bond->slave_cnt == 0) {
2567 goto re_arm;
2570 read_lock(&bond->curr_slave_lock);
2571 oldcurrent = bond->curr_active_slave;
2572 read_unlock(&bond->curr_slave_lock);
2574 /* see if any of the previous devices are up now (i.e. they have
2575 * xmt and rcv traffic). the curr_active_slave does not come into
2576 * the picture unless it is null. also, slave->jiffies is not needed
2577 * here because we send an arp on each slave and give a slave as
2578 * long as it needs to get the tx/rx within the delta.
2579 * TODO: what about up/down delay in arp mode? it wasn't here before
2580 * so it can wait
2582 bond_for_each_slave(bond, slave, i) {
2583 if (slave->link != BOND_LINK_UP) {
2584 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2585 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2587 slave->link = BOND_LINK_UP;
2588 slave->state = BOND_STATE_ACTIVE;
2590 /* primary_slave has no meaning in round-robin
2591 * mode. the window of a slave being up and
2592 * curr_active_slave being null after enslaving
2593 * is closed.
2595 if (!oldcurrent) {
2596 printk(KERN_INFO DRV_NAME
2597 ": %s: link status definitely "
2598 "up for interface %s, ",
2599 bond_dev->name,
2600 slave->dev->name);
2601 do_failover = 1;
2602 } else {
2603 printk(KERN_INFO DRV_NAME
2604 ": %s: interface %s is now up\n",
2605 bond_dev->name,
2606 slave->dev->name);
2609 } else {
2610 /* slave->link == BOND_LINK_UP */
2612 /* not all switches will respond to an arp request
2613 * when the source ip is 0, so don't take the link down
2614 * if we don't know our ip yet
2616 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2617 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2618 bond_has_ip(bond))) {
2620 slave->link = BOND_LINK_DOWN;
2621 slave->state = BOND_STATE_BACKUP;
2623 if (slave->link_failure_count < UINT_MAX) {
2624 slave->link_failure_count++;
2627 printk(KERN_INFO DRV_NAME
2628 ": %s: interface %s is now down.\n",
2629 bond_dev->name,
2630 slave->dev->name);
2632 if (slave == oldcurrent) {
2633 do_failover = 1;
2638 /* note: if switch is in round-robin mode, all links
2639 * must tx arp to ensure all links rx an arp - otherwise
2640 * links may oscillate or not come up at all; if switch is
2641 * in something like xor mode, there is nothing we can
2642 * do - all replies will be rx'ed on same link causing slaves
2643 * to be unstable during low/no traffic periods
2645 if (IS_UP(slave->dev)) {
2646 bond_arp_send_all(bond, slave);
2650 if (do_failover) {
2651 write_lock(&bond->curr_slave_lock);
2653 bond_select_active_slave(bond);
2655 write_unlock(&bond->curr_slave_lock);
2658 re_arm:
2659 if (bond->params.arp_interval) {
2660 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2662 out:
2663 read_unlock(&bond->lock);
2667 * When using arp monitoring in active-backup mode, this function is
2668 * called to determine if any backup slaves have went down or a new
2669 * current slave needs to be found.
2670 * The backup slaves never generate traffic, they are considered up by merely
2671 * receiving traffic. If the current slave goes down, each backup slave will
2672 * be given the opportunity to tx/rx an arp before being taken down - this
2673 * prevents all slaves from being taken down due to the current slave not
2674 * sending any traffic for the backups to receive. The arps are not necessarily
2675 * necessary, any tx and rx traffic will keep the current slave up. While any
2676 * rx traffic will keep the backup slaves up, the current slave is responsible
2677 * for generating traffic to keep them up regardless of any other traffic they
2678 * may have received.
2679 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2681 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2683 struct bonding *bond = bond_dev->priv;
2684 struct slave *slave;
2685 int delta_in_ticks;
2686 int i;
2688 read_lock(&bond->lock);
2690 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2692 if (bond->kill_timers) {
2693 goto out;
2696 if (bond->slave_cnt == 0) {
2697 goto re_arm;
2700 /* determine if any slave has come up or any backup slave has
2701 * gone down
2702 * TODO: what about up/down delay in arp mode? it wasn't here before
2703 * so it can wait
2705 bond_for_each_slave(bond, slave, i) {
2706 if (slave->link != BOND_LINK_UP) {
2707 if ((jiffies - slave_last_rx(bond, slave)) <=
2708 delta_in_ticks) {
2710 slave->link = BOND_LINK_UP;
2712 write_lock(&bond->curr_slave_lock);
2714 if ((!bond->curr_active_slave) &&
2715 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2716 bond_change_active_slave(bond, slave);
2717 bond->current_arp_slave = NULL;
2718 } else if (bond->curr_active_slave != slave) {
2719 /* this slave has just come up but we
2720 * already have a current slave; this
2721 * can also happen if bond_enslave adds
2722 * a new slave that is up while we are
2723 * searching for a new slave
2725 bond_set_slave_inactive_flags(slave);
2726 bond->current_arp_slave = NULL;
2729 bond_set_carrier(bond);
2731 if (slave == bond->curr_active_slave) {
2732 printk(KERN_INFO DRV_NAME
2733 ": %s: %s is up and now the "
2734 "active interface\n",
2735 bond_dev->name,
2736 slave->dev->name);
2737 netif_carrier_on(bond->dev);
2738 } else {
2739 printk(KERN_INFO DRV_NAME
2740 ": %s: backup interface %s is "
2741 "now up\n",
2742 bond_dev->name,
2743 slave->dev->name);
2746 write_unlock(&bond->curr_slave_lock);
2748 } else {
2749 read_lock(&bond->curr_slave_lock);
2751 if ((slave != bond->curr_active_slave) &&
2752 (!bond->current_arp_slave) &&
2753 (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2754 bond_has_ip(bond))) {
2755 /* a backup slave has gone down; three times
2756 * the delta allows the current slave to be
2757 * taken out before the backup slave.
2758 * note: a non-null current_arp_slave indicates
2759 * the curr_active_slave went down and we are
2760 * searching for a new one; under this
2761 * condition we only take the curr_active_slave
2762 * down - this gives each slave a chance to
2763 * tx/rx traffic before being taken out
2766 read_unlock(&bond->curr_slave_lock);
2768 slave->link = BOND_LINK_DOWN;
2770 if (slave->link_failure_count < UINT_MAX) {
2771 slave->link_failure_count++;
2774 bond_set_slave_inactive_flags(slave);
2776 printk(KERN_INFO DRV_NAME
2777 ": %s: backup interface %s is now down\n",
2778 bond_dev->name,
2779 slave->dev->name);
2780 } else {
2781 read_unlock(&bond->curr_slave_lock);
2786 read_lock(&bond->curr_slave_lock);
2787 slave = bond->curr_active_slave;
2788 read_unlock(&bond->curr_slave_lock);
2790 if (slave) {
2791 /* if we have sent traffic in the past 2*arp_intervals but
2792 * haven't xmit and rx traffic in that time interval, select
2793 * a different slave. slave->jiffies is only updated when
2794 * a slave first becomes the curr_active_slave - not necessarily
2795 * after every arp; this ensures the slave has a full 2*delta
2796 * before being taken out. if a primary is being used, check
2797 * if it is up and needs to take over as the curr_active_slave
2799 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2800 (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2801 bond_has_ip(bond))) &&
2802 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2804 slave->link = BOND_LINK_DOWN;
2806 if (slave->link_failure_count < UINT_MAX) {
2807 slave->link_failure_count++;
2810 printk(KERN_INFO DRV_NAME
2811 ": %s: link status down for active interface "
2812 "%s, disabling it\n",
2813 bond_dev->name,
2814 slave->dev->name);
2816 write_lock(&bond->curr_slave_lock);
2818 bond_select_active_slave(bond);
2819 slave = bond->curr_active_slave;
2821 write_unlock(&bond->curr_slave_lock);
2823 bond->current_arp_slave = slave;
2825 if (slave) {
2826 slave->jiffies = jiffies;
2828 } else if ((bond->primary_slave) &&
2829 (bond->primary_slave != slave) &&
2830 (bond->primary_slave->link == BOND_LINK_UP)) {
2831 /* at this point, slave is the curr_active_slave */
2832 printk(KERN_INFO DRV_NAME
2833 ": %s: changing from interface %s to primary "
2834 "interface %s\n",
2835 bond_dev->name,
2836 slave->dev->name,
2837 bond->primary_slave->dev->name);
2839 /* primary is up so switch to it */
2840 write_lock(&bond->curr_slave_lock);
2841 bond_change_active_slave(bond, bond->primary_slave);
2842 write_unlock(&bond->curr_slave_lock);
2844 slave = bond->primary_slave;
2845 slave->jiffies = jiffies;
2846 } else {
2847 bond->current_arp_slave = NULL;
2850 /* the current slave must tx an arp to ensure backup slaves
2851 * rx traffic
2853 if (slave && bond_has_ip(bond)) {
2854 bond_arp_send_all(bond, slave);
2858 /* if we don't have a curr_active_slave, search for the next available
2859 * backup slave from the current_arp_slave and make it the candidate
2860 * for becoming the curr_active_slave
2862 if (!slave) {
2863 if (!bond->current_arp_slave) {
2864 bond->current_arp_slave = bond->first_slave;
2867 if (bond->current_arp_slave) {
2868 bond_set_slave_inactive_flags(bond->current_arp_slave);
2870 /* search for next candidate */
2871 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2872 if (IS_UP(slave->dev)) {
2873 slave->link = BOND_LINK_BACK;
2874 bond_set_slave_active_flags(slave);
2875 bond_arp_send_all(bond, slave);
2876 slave->jiffies = jiffies;
2877 bond->current_arp_slave = slave;
2878 break;
2881 /* if the link state is up at this point, we
2882 * mark it down - this can happen if we have
2883 * simultaneous link failures and
2884 * reselect_active_interface doesn't make this
2885 * one the current slave so it is still marked
2886 * up when it is actually down
2888 if (slave->link == BOND_LINK_UP) {
2889 slave->link = BOND_LINK_DOWN;
2890 if (slave->link_failure_count < UINT_MAX) {
2891 slave->link_failure_count++;
2894 bond_set_slave_inactive_flags(slave);
2896 printk(KERN_INFO DRV_NAME
2897 ": %s: backup interface %s is "
2898 "now down.\n",
2899 bond_dev->name,
2900 slave->dev->name);
2906 re_arm:
2907 if (bond->params.arp_interval) {
2908 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2910 out:
2911 read_unlock(&bond->lock);
2914 /*------------------------------ proc/seq_file-------------------------------*/
2916 #ifdef CONFIG_PROC_FS
2918 #define SEQ_START_TOKEN ((void *)1)
2920 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2922 struct bonding *bond = seq->private;
2923 loff_t off = 0;
2924 struct slave *slave;
2925 int i;
2927 /* make sure the bond won't be taken away */
2928 read_lock(&dev_base_lock);
2929 read_lock_bh(&bond->lock);
2931 if (*pos == 0) {
2932 return SEQ_START_TOKEN;
2935 bond_for_each_slave(bond, slave, i) {
2936 if (++off == *pos) {
2937 return slave;
2941 return NULL;
2944 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2946 struct bonding *bond = seq->private;
2947 struct slave *slave = v;
2949 ++*pos;
2950 if (v == SEQ_START_TOKEN) {
2951 return bond->first_slave;
2954 slave = slave->next;
2956 return (slave == bond->first_slave) ? NULL : slave;
2959 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2961 struct bonding *bond = seq->private;
2963 read_unlock_bh(&bond->lock);
2964 read_unlock(&dev_base_lock);
2967 static void bond_info_show_master(struct seq_file *seq)
2969 struct bonding *bond = seq->private;
2970 struct slave *curr;
2971 int i;
2972 u32 target;
2974 read_lock(&bond->curr_slave_lock);
2975 curr = bond->curr_active_slave;
2976 read_unlock(&bond->curr_slave_lock);
2978 seq_printf(seq, "Bonding Mode: %s\n",
2979 bond_mode_name(bond->params.mode));
2981 if (bond->params.mode == BOND_MODE_XOR ||
2982 bond->params.mode == BOND_MODE_8023AD) {
2983 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2984 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2985 bond->params.xmit_policy);
2988 if (USES_PRIMARY(bond->params.mode)) {
2989 seq_printf(seq, "Primary Slave: %s\n",
2990 (bond->primary_slave) ?
2991 bond->primary_slave->dev->name : "None");
2993 seq_printf(seq, "Currently Active Slave: %s\n",
2994 (curr) ? curr->dev->name : "None");
2997 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
2998 "up" : "down");
2999 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3000 seq_printf(seq, "Up Delay (ms): %d\n",
3001 bond->params.updelay * bond->params.miimon);
3002 seq_printf(seq, "Down Delay (ms): %d\n",
3003 bond->params.downdelay * bond->params.miimon);
3006 /* ARP information */
3007 if(bond->params.arp_interval > 0) {
3008 int printed=0;
3009 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3010 bond->params.arp_interval);
3012 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3014 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3015 if (!bond->params.arp_targets[i])
3016 continue;
3017 if (printed)
3018 seq_printf(seq, ",");
3019 target = ntohl(bond->params.arp_targets[i]);
3020 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3021 printed = 1;
3023 seq_printf(seq, "\n");
3026 if (bond->params.mode == BOND_MODE_8023AD) {
3027 struct ad_info ad_info;
3029 seq_puts(seq, "\n802.3ad info\n");
3030 seq_printf(seq, "LACP rate: %s\n",
3031 (bond->params.lacp_fast) ? "fast" : "slow");
3033 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3034 seq_printf(seq, "bond %s has no active aggregator\n",
3035 bond->dev->name);
3036 } else {
3037 seq_printf(seq, "Active Aggregator Info:\n");
3039 seq_printf(seq, "\tAggregator ID: %d\n",
3040 ad_info.aggregator_id);
3041 seq_printf(seq, "\tNumber of ports: %d\n",
3042 ad_info.ports);
3043 seq_printf(seq, "\tActor Key: %d\n",
3044 ad_info.actor_key);
3045 seq_printf(seq, "\tPartner Key: %d\n",
3046 ad_info.partner_key);
3047 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3048 ad_info.partner_system[0],
3049 ad_info.partner_system[1],
3050 ad_info.partner_system[2],
3051 ad_info.partner_system[3],
3052 ad_info.partner_system[4],
3053 ad_info.partner_system[5]);
3058 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3060 struct bonding *bond = seq->private;
3062 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3063 seq_printf(seq, "MII Status: %s\n",
3064 (slave->link == BOND_LINK_UP) ? "up" : "down");
3065 seq_printf(seq, "Link Failure Count: %u\n",
3066 slave->link_failure_count);
3068 seq_printf(seq,
3069 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3070 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3071 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3072 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
3074 if (bond->params.mode == BOND_MODE_8023AD) {
3075 const struct aggregator *agg
3076 = SLAVE_AD_INFO(slave).port.aggregator;
3078 if (agg) {
3079 seq_printf(seq, "Aggregator ID: %d\n",
3080 agg->aggregator_identifier);
3081 } else {
3082 seq_puts(seq, "Aggregator ID: N/A\n");
3087 static int bond_info_seq_show(struct seq_file *seq, void *v)
3089 if (v == SEQ_START_TOKEN) {
3090 seq_printf(seq, "%s\n", version);
3091 bond_info_show_master(seq);
3092 } else {
3093 bond_info_show_slave(seq, v);
3096 return 0;
3099 static struct seq_operations bond_info_seq_ops = {
3100 .start = bond_info_seq_start,
3101 .next = bond_info_seq_next,
3102 .stop = bond_info_seq_stop,
3103 .show = bond_info_seq_show,
3106 static int bond_info_open(struct inode *inode, struct file *file)
3108 struct seq_file *seq;
3109 struct proc_dir_entry *proc;
3110 int res;
3112 res = seq_open(file, &bond_info_seq_ops);
3113 if (!res) {
3114 /* recover the pointer buried in proc_dir_entry data */
3115 seq = file->private_data;
3116 proc = PDE(inode);
3117 seq->private = proc->data;
3120 return res;
3123 static const struct file_operations bond_info_fops = {
3124 .owner = THIS_MODULE,
3125 .open = bond_info_open,
3126 .read = seq_read,
3127 .llseek = seq_lseek,
3128 .release = seq_release,
3131 static int bond_create_proc_entry(struct bonding *bond)
3133 struct net_device *bond_dev = bond->dev;
3135 if (bond_proc_dir) {
3136 bond->proc_entry = create_proc_entry(bond_dev->name,
3137 S_IRUGO,
3138 bond_proc_dir);
3139 if (bond->proc_entry == NULL) {
3140 printk(KERN_WARNING DRV_NAME
3141 ": Warning: Cannot create /proc/net/%s/%s\n",
3142 DRV_NAME, bond_dev->name);
3143 } else {
3144 bond->proc_entry->data = bond;
3145 bond->proc_entry->proc_fops = &bond_info_fops;
3146 bond->proc_entry->owner = THIS_MODULE;
3147 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3151 return 0;
3154 static void bond_remove_proc_entry(struct bonding *bond)
3156 if (bond_proc_dir && bond->proc_entry) {
3157 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3158 memset(bond->proc_file_name, 0, IFNAMSIZ);
3159 bond->proc_entry = NULL;
3163 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3164 * Caller must hold rtnl_lock.
3166 static void bond_create_proc_dir(void)
3168 int len = strlen(DRV_NAME);
3170 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3171 bond_proc_dir = bond_proc_dir->next) {
3172 if ((bond_proc_dir->namelen == len) &&
3173 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3174 break;
3178 if (!bond_proc_dir) {
3179 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3180 if (bond_proc_dir) {
3181 bond_proc_dir->owner = THIS_MODULE;
3182 } else {
3183 printk(KERN_WARNING DRV_NAME
3184 ": Warning: cannot create /proc/net/%s\n",
3185 DRV_NAME);
3190 /* Destroy the bonding directory under /proc/net, if empty.
3191 * Caller must hold rtnl_lock.
3193 static void bond_destroy_proc_dir(void)
3195 struct proc_dir_entry *de;
3197 if (!bond_proc_dir) {
3198 return;
3201 /* verify that the /proc dir is empty */
3202 for (de = bond_proc_dir->subdir; de; de = de->next) {
3203 /* ignore . and .. */
3204 if (*(de->name) != '.') {
3205 break;
3209 if (de) {
3210 if (bond_proc_dir->owner == THIS_MODULE) {
3211 bond_proc_dir->owner = NULL;
3213 } else {
3214 remove_proc_entry(DRV_NAME, proc_net);
3215 bond_proc_dir = NULL;
3218 #endif /* CONFIG_PROC_FS */
3220 /*-------------------------- netdev event handling --------------------------*/
3223 * Change device name
3225 static int bond_event_changename(struct bonding *bond)
3227 #ifdef CONFIG_PROC_FS
3228 bond_remove_proc_entry(bond);
3229 bond_create_proc_entry(bond);
3230 #endif
3231 down_write(&(bonding_rwsem));
3232 bond_destroy_sysfs_entry(bond);
3233 bond_create_sysfs_entry(bond);
3234 up_write(&(bonding_rwsem));
3235 return NOTIFY_DONE;
3238 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3240 struct bonding *event_bond = bond_dev->priv;
3242 switch (event) {
3243 case NETDEV_CHANGENAME:
3244 return bond_event_changename(event_bond);
3245 case NETDEV_UNREGISTER:
3247 * TODO: remove a bond from the list?
3249 break;
3250 default:
3251 break;
3254 return NOTIFY_DONE;
3257 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3259 struct net_device *bond_dev = slave_dev->master;
3260 struct bonding *bond = bond_dev->priv;
3262 switch (event) {
3263 case NETDEV_UNREGISTER:
3264 if (bond_dev) {
3265 bond_release(bond_dev, slave_dev);
3267 break;
3268 case NETDEV_CHANGE:
3270 * TODO: is this what we get if somebody
3271 * sets up a hierarchical bond, then rmmod's
3272 * one of the slave bonding devices?
3274 break;
3275 case NETDEV_DOWN:
3277 * ... Or is it this?
3279 break;
3280 case NETDEV_CHANGEMTU:
3282 * TODO: Should slaves be allowed to
3283 * independently alter their MTU? For
3284 * an active-backup bond, slaves need
3285 * not be the same type of device, so
3286 * MTUs may vary. For other modes,
3287 * slaves arguably should have the
3288 * same MTUs. To do this, we'd need to
3289 * take over the slave's change_mtu
3290 * function for the duration of their
3291 * servitude.
3293 break;
3294 case NETDEV_CHANGENAME:
3296 * TODO: handle changing the primary's name
3298 break;
3299 case NETDEV_FEAT_CHANGE:
3300 bond_compute_features(bond);
3301 break;
3302 default:
3303 break;
3306 return NOTIFY_DONE;
3310 * bond_netdev_event: handle netdev notifier chain events.
3312 * This function receives events for the netdev chain. The caller (an
3313 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3314 * locks for us to safely manipulate the slave devices (RTNL lock,
3315 * dev_probe_lock).
3317 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3319 struct net_device *event_dev = (struct net_device *)ptr;
3321 dprintk("event_dev: %s, event: %lx\n",
3322 (event_dev ? event_dev->name : "None"),
3323 event);
3325 if (!(event_dev->priv_flags & IFF_BONDING))
3326 return NOTIFY_DONE;
3328 if (event_dev->flags & IFF_MASTER) {
3329 dprintk("IFF_MASTER\n");
3330 return bond_master_netdev_event(event, event_dev);
3333 if (event_dev->flags & IFF_SLAVE) {
3334 dprintk("IFF_SLAVE\n");
3335 return bond_slave_netdev_event(event, event_dev);
3338 return NOTIFY_DONE;
3342 * bond_inetaddr_event: handle inetaddr notifier chain events.
3344 * We keep track of device IPs primarily to use as source addresses in
3345 * ARP monitor probes (rather than spewing out broadcasts all the time).
3347 * We track one IP for the main device (if it has one), plus one per VLAN.
3349 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3351 struct in_ifaddr *ifa = ptr;
3352 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3353 struct bonding *bond, *bond_next;
3354 struct vlan_entry *vlan, *vlan_next;
3356 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3357 if (bond->dev == event_dev) {
3358 switch (event) {
3359 case NETDEV_UP:
3360 bond->master_ip = ifa->ifa_local;
3361 return NOTIFY_OK;
3362 case NETDEV_DOWN:
3363 bond->master_ip = bond_glean_dev_ip(bond->dev);
3364 return NOTIFY_OK;
3365 default:
3366 return NOTIFY_DONE;
3370 if (list_empty(&bond->vlan_list))
3371 continue;
3373 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3374 vlan_list) {
3375 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3376 if (vlan_dev == event_dev) {
3377 switch (event) {
3378 case NETDEV_UP:
3379 vlan->vlan_ip = ifa->ifa_local;
3380 return NOTIFY_OK;
3381 case NETDEV_DOWN:
3382 vlan->vlan_ip =
3383 bond_glean_dev_ip(vlan_dev);
3384 return NOTIFY_OK;
3385 default:
3386 return NOTIFY_DONE;
3391 return NOTIFY_DONE;
3394 static struct notifier_block bond_netdev_notifier = {
3395 .notifier_call = bond_netdev_event,
3398 static struct notifier_block bond_inetaddr_notifier = {
3399 .notifier_call = bond_inetaddr_event,
3402 /*-------------------------- Packet type handling ---------------------------*/
3404 /* register to receive lacpdus on a bond */
3405 static void bond_register_lacpdu(struct bonding *bond)
3407 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3409 /* initialize packet type */
3410 pk_type->type = PKT_TYPE_LACPDU;
3411 pk_type->dev = bond->dev;
3412 pk_type->func = bond_3ad_lacpdu_recv;
3414 dev_add_pack(pk_type);
3417 /* unregister to receive lacpdus on a bond */
3418 static void bond_unregister_lacpdu(struct bonding *bond)
3420 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3423 void bond_register_arp(struct bonding *bond)
3425 struct packet_type *pt = &bond->arp_mon_pt;
3427 pt->type = htons(ETH_P_ARP);
3428 pt->dev = NULL; /*bond->dev;XXX*/
3429 pt->func = bond_arp_rcv;
3430 dev_add_pack(pt);
3433 void bond_unregister_arp(struct bonding *bond)
3435 dev_remove_pack(&bond->arp_mon_pt);
3438 /*---------------------------- Hashing Policies -----------------------------*/
3441 * Hash for the the output device based upon layer 3 and layer 4 data. If
3442 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3443 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3445 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3446 struct net_device *bond_dev, int count)
3448 struct ethhdr *data = (struct ethhdr *)skb->data;
3449 struct iphdr *iph = skb->nh.iph;
3450 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3451 int layer4_xor = 0;
3453 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3454 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3455 (iph->protocol == IPPROTO_TCP ||
3456 iph->protocol == IPPROTO_UDP)) {
3457 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3459 return (layer4_xor ^
3460 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3464 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3468 * Hash for the output device based upon layer 2 data
3470 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3471 struct net_device *bond_dev, int count)
3473 struct ethhdr *data = (struct ethhdr *)skb->data;
3475 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3478 /*-------------------------- Device entry points ----------------------------*/
3480 static int bond_open(struct net_device *bond_dev)
3482 struct bonding *bond = bond_dev->priv;
3483 struct timer_list *mii_timer = &bond->mii_timer;
3484 struct timer_list *arp_timer = &bond->arp_timer;
3486 bond->kill_timers = 0;
3488 if ((bond->params.mode == BOND_MODE_TLB) ||
3489 (bond->params.mode == BOND_MODE_ALB)) {
3490 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3492 /* bond_alb_initialize must be called before the timer
3493 * is started.
3495 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3496 /* something went wrong - fail the open operation */
3497 return -1;
3500 init_timer(alb_timer);
3501 alb_timer->expires = jiffies + 1;
3502 alb_timer->data = (unsigned long)bond;
3503 alb_timer->function = (void *)&bond_alb_monitor;
3504 add_timer(alb_timer);
3507 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3508 init_timer(mii_timer);
3509 mii_timer->expires = jiffies + 1;
3510 mii_timer->data = (unsigned long)bond_dev;
3511 mii_timer->function = (void *)&bond_mii_monitor;
3512 add_timer(mii_timer);
3515 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3516 init_timer(arp_timer);
3517 arp_timer->expires = jiffies + 1;
3518 arp_timer->data = (unsigned long)bond_dev;
3519 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3520 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3521 } else {
3522 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3524 if (bond->params.arp_validate)
3525 bond_register_arp(bond);
3527 add_timer(arp_timer);
3530 if (bond->params.mode == BOND_MODE_8023AD) {
3531 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3532 init_timer(ad_timer);
3533 ad_timer->expires = jiffies + 1;
3534 ad_timer->data = (unsigned long)bond;
3535 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3536 add_timer(ad_timer);
3538 /* register to receive LACPDUs */
3539 bond_register_lacpdu(bond);
3542 return 0;
3545 static int bond_close(struct net_device *bond_dev)
3547 struct bonding *bond = bond_dev->priv;
3549 if (bond->params.mode == BOND_MODE_8023AD) {
3550 /* Unregister the receive of LACPDUs */
3551 bond_unregister_lacpdu(bond);
3554 if (bond->params.arp_validate)
3555 bond_unregister_arp(bond);
3557 write_lock_bh(&bond->lock);
3560 /* signal timers not to re-arm */
3561 bond->kill_timers = 1;
3563 write_unlock_bh(&bond->lock);
3565 /* del_timer_sync must run without holding the bond->lock
3566 * because a running timer might be trying to hold it too
3569 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3570 del_timer_sync(&bond->mii_timer);
3573 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3574 del_timer_sync(&bond->arp_timer);
3577 switch (bond->params.mode) {
3578 case BOND_MODE_8023AD:
3579 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3580 break;
3581 case BOND_MODE_TLB:
3582 case BOND_MODE_ALB:
3583 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3584 break;
3585 default:
3586 break;
3590 if ((bond->params.mode == BOND_MODE_TLB) ||
3591 (bond->params.mode == BOND_MODE_ALB)) {
3592 /* Must be called only after all
3593 * slaves have been released
3595 bond_alb_deinitialize(bond);
3598 return 0;
3601 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3603 struct bonding *bond = bond_dev->priv;
3604 struct net_device_stats *stats = &(bond->stats), *sstats;
3605 struct slave *slave;
3606 int i;
3608 memset(stats, 0, sizeof(struct net_device_stats));
3610 read_lock_bh(&bond->lock);
3612 bond_for_each_slave(bond, slave, i) {
3613 if (slave->dev->get_stats) {
3614 sstats = slave->dev->get_stats(slave->dev);
3616 stats->rx_packets += sstats->rx_packets;
3617 stats->rx_bytes += sstats->rx_bytes;
3618 stats->rx_errors += sstats->rx_errors;
3619 stats->rx_dropped += sstats->rx_dropped;
3621 stats->tx_packets += sstats->tx_packets;
3622 stats->tx_bytes += sstats->tx_bytes;
3623 stats->tx_errors += sstats->tx_errors;
3624 stats->tx_dropped += sstats->tx_dropped;
3626 stats->multicast += sstats->multicast;
3627 stats->collisions += sstats->collisions;
3629 stats->rx_length_errors += sstats->rx_length_errors;
3630 stats->rx_over_errors += sstats->rx_over_errors;
3631 stats->rx_crc_errors += sstats->rx_crc_errors;
3632 stats->rx_frame_errors += sstats->rx_frame_errors;
3633 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3634 stats->rx_missed_errors += sstats->rx_missed_errors;
3636 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3637 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3638 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3639 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3640 stats->tx_window_errors += sstats->tx_window_errors;
3644 read_unlock_bh(&bond->lock);
3646 return stats;
3649 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3651 struct net_device *slave_dev = NULL;
3652 struct ifbond k_binfo;
3653 struct ifbond __user *u_binfo = NULL;
3654 struct ifslave k_sinfo;
3655 struct ifslave __user *u_sinfo = NULL;
3656 struct mii_ioctl_data *mii = NULL;
3657 int res = 0;
3659 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3660 bond_dev->name, cmd);
3662 switch (cmd) {
3663 case SIOCGMIIPHY:
3664 mii = if_mii(ifr);
3665 if (!mii) {
3666 return -EINVAL;
3668 mii->phy_id = 0;
3669 /* Fall Through */
3670 case SIOCGMIIREG:
3672 * We do this again just in case we were called by SIOCGMIIREG
3673 * instead of SIOCGMIIPHY.
3675 mii = if_mii(ifr);
3676 if (!mii) {
3677 return -EINVAL;
3680 if (mii->reg_num == 1) {
3681 struct bonding *bond = bond_dev->priv;
3682 mii->val_out = 0;
3683 read_lock_bh(&bond->lock);
3684 read_lock(&bond->curr_slave_lock);
3685 if (netif_carrier_ok(bond->dev)) {
3686 mii->val_out = BMSR_LSTATUS;
3688 read_unlock(&bond->curr_slave_lock);
3689 read_unlock_bh(&bond->lock);
3692 return 0;
3693 case BOND_INFO_QUERY_OLD:
3694 case SIOCBONDINFOQUERY:
3695 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3697 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3698 return -EFAULT;
3701 res = bond_info_query(bond_dev, &k_binfo);
3702 if (res == 0) {
3703 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3704 return -EFAULT;
3708 return res;
3709 case BOND_SLAVE_INFO_QUERY_OLD:
3710 case SIOCBONDSLAVEINFOQUERY:
3711 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3713 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3714 return -EFAULT;
3717 res = bond_slave_info_query(bond_dev, &k_sinfo);
3718 if (res == 0) {
3719 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3720 return -EFAULT;
3724 return res;
3725 default:
3726 /* Go on */
3727 break;
3730 if (!capable(CAP_NET_ADMIN)) {
3731 return -EPERM;
3734 down_write(&(bonding_rwsem));
3735 slave_dev = dev_get_by_name(ifr->ifr_slave);
3737 dprintk("slave_dev=%p: \n", slave_dev);
3739 if (!slave_dev) {
3740 res = -ENODEV;
3741 } else {
3742 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3743 switch (cmd) {
3744 case BOND_ENSLAVE_OLD:
3745 case SIOCBONDENSLAVE:
3746 res = bond_enslave(bond_dev, slave_dev);
3747 break;
3748 case BOND_RELEASE_OLD:
3749 case SIOCBONDRELEASE:
3750 res = bond_release(bond_dev, slave_dev);
3751 break;
3752 case BOND_SETHWADDR_OLD:
3753 case SIOCBONDSETHWADDR:
3754 res = bond_sethwaddr(bond_dev, slave_dev);
3755 break;
3756 case BOND_CHANGE_ACTIVE_OLD:
3757 case SIOCBONDCHANGEACTIVE:
3758 res = bond_ioctl_change_active(bond_dev, slave_dev);
3759 break;
3760 default:
3761 res = -EOPNOTSUPP;
3764 dev_put(slave_dev);
3767 up_write(&(bonding_rwsem));
3768 return res;
3771 static void bond_set_multicast_list(struct net_device *bond_dev)
3773 struct bonding *bond = bond_dev->priv;
3774 struct dev_mc_list *dmi;
3776 write_lock_bh(&bond->lock);
3779 * Do promisc before checking multicast_mode
3781 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3782 bond_set_promiscuity(bond, 1);
3785 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3786 bond_set_promiscuity(bond, -1);
3789 /* set allmulti flag to slaves */
3790 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3791 bond_set_allmulti(bond, 1);
3794 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3795 bond_set_allmulti(bond, -1);
3798 bond->flags = bond_dev->flags;
3800 /* looking for addresses to add to slaves' mc list */
3801 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3802 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3803 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3807 /* looking for addresses to delete from slaves' list */
3808 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3809 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3810 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3814 /* save master's multicast list */
3815 bond_mc_list_destroy(bond);
3816 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3818 write_unlock_bh(&bond->lock);
3822 * Change the MTU of all of a master's slaves to match the master
3824 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3826 struct bonding *bond = bond_dev->priv;
3827 struct slave *slave, *stop_at;
3828 int res = 0;
3829 int i;
3831 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3832 (bond_dev ? bond_dev->name : "None"), new_mtu);
3834 /* Can't hold bond->lock with bh disabled here since
3835 * some base drivers panic. On the other hand we can't
3836 * hold bond->lock without bh disabled because we'll
3837 * deadlock. The only solution is to rely on the fact
3838 * that we're under rtnl_lock here, and the slaves
3839 * list won't change. This doesn't solve the problem
3840 * of setting the slave's MTU while it is
3841 * transmitting, but the assumption is that the base
3842 * driver can handle that.
3844 * TODO: figure out a way to safely iterate the slaves
3845 * list, but without holding a lock around the actual
3846 * call to the base driver.
3849 bond_for_each_slave(bond, slave, i) {
3850 dprintk("s %p s->p %p c_m %p\n", slave,
3851 slave->prev, slave->dev->change_mtu);
3853 res = dev_set_mtu(slave->dev, new_mtu);
3855 if (res) {
3856 /* If we failed to set the slave's mtu to the new value
3857 * we must abort the operation even in ACTIVE_BACKUP
3858 * mode, because if we allow the backup slaves to have
3859 * different mtu values than the active slave we'll
3860 * need to change their mtu when doing a failover. That
3861 * means changing their mtu from timer context, which
3862 * is probably not a good idea.
3864 dprintk("err %d %s\n", res, slave->dev->name);
3865 goto unwind;
3869 bond_dev->mtu = new_mtu;
3871 return 0;
3873 unwind:
3874 /* unwind from head to the slave that failed */
3875 stop_at = slave;
3876 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3877 int tmp_res;
3879 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3880 if (tmp_res) {
3881 dprintk("unwind err %d dev %s\n", tmp_res,
3882 slave->dev->name);
3886 return res;
3890 * Change HW address
3892 * Note that many devices must be down to change the HW address, and
3893 * downing the master releases all slaves. We can make bonds full of
3894 * bonding devices to test this, however.
3896 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3898 struct bonding *bond = bond_dev->priv;
3899 struct sockaddr *sa = addr, tmp_sa;
3900 struct slave *slave, *stop_at;
3901 int res = 0;
3902 int i;
3904 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3906 if (!is_valid_ether_addr(sa->sa_data)) {
3907 return -EADDRNOTAVAIL;
3910 /* Can't hold bond->lock with bh disabled here since
3911 * some base drivers panic. On the other hand we can't
3912 * hold bond->lock without bh disabled because we'll
3913 * deadlock. The only solution is to rely on the fact
3914 * that we're under rtnl_lock here, and the slaves
3915 * list won't change. This doesn't solve the problem
3916 * of setting the slave's hw address while it is
3917 * transmitting, but the assumption is that the base
3918 * driver can handle that.
3920 * TODO: figure out a way to safely iterate the slaves
3921 * list, but without holding a lock around the actual
3922 * call to the base driver.
3925 bond_for_each_slave(bond, slave, i) {
3926 dprintk("slave %p %s\n", slave, slave->dev->name);
3928 if (slave->dev->set_mac_address == NULL) {
3929 res = -EOPNOTSUPP;
3930 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3931 goto unwind;
3934 res = dev_set_mac_address(slave->dev, addr);
3935 if (res) {
3936 /* TODO: consider downing the slave
3937 * and retry ?
3938 * User should expect communications
3939 * breakage anyway until ARP finish
3940 * updating, so...
3942 dprintk("err %d %s\n", res, slave->dev->name);
3943 goto unwind;
3947 /* success */
3948 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3949 return 0;
3951 unwind:
3952 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3953 tmp_sa.sa_family = bond_dev->type;
3955 /* unwind from head to the slave that failed */
3956 stop_at = slave;
3957 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3958 int tmp_res;
3960 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3961 if (tmp_res) {
3962 dprintk("unwind err %d dev %s\n", tmp_res,
3963 slave->dev->name);
3967 return res;
3970 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3972 struct bonding *bond = bond_dev->priv;
3973 struct slave *slave, *start_at;
3974 int i;
3975 int res = 1;
3977 read_lock(&bond->lock);
3979 if (!BOND_IS_OK(bond)) {
3980 goto out;
3983 read_lock(&bond->curr_slave_lock);
3984 slave = start_at = bond->curr_active_slave;
3985 read_unlock(&bond->curr_slave_lock);
3987 if (!slave) {
3988 goto out;
3991 bond_for_each_slave_from(bond, slave, i, start_at) {
3992 if (IS_UP(slave->dev) &&
3993 (slave->link == BOND_LINK_UP) &&
3994 (slave->state == BOND_STATE_ACTIVE)) {
3995 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3997 write_lock(&bond->curr_slave_lock);
3998 bond->curr_active_slave = slave->next;
3999 write_unlock(&bond->curr_slave_lock);
4001 break;
4006 out:
4007 if (res) {
4008 /* no suitable interface, frame not sent */
4009 dev_kfree_skb(skb);
4011 read_unlock(&bond->lock);
4012 return 0;
4015 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
4016 struct bonding *bond,
4017 struct slave *slave)
4019 struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
4020 struct ethhdr *eth_data;
4021 u8 *hwaddr;
4022 int res;
4024 if (!skb2) {
4025 printk(KERN_ERR DRV_NAME ": Error: "
4026 "bond_activebackup_xmit_copy(): skb_copy() failed\n");
4027 return;
4030 skb2->mac.raw = (unsigned char *)skb2->data;
4031 eth_data = eth_hdr(skb2);
4033 /* Pick an appropriate source MAC address
4034 * -- use slave's perm MAC addr, unless used by bond
4035 * -- otherwise, borrow active slave's perm MAC addr
4036 * since that will not be used
4038 hwaddr = slave->perm_hwaddr;
4039 if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
4040 hwaddr = bond->curr_active_slave->perm_hwaddr;
4042 /* Set source MAC address appropriately */
4043 memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
4045 res = bond_dev_queue_xmit(bond, skb2, slave->dev);
4046 if (res)
4047 dev_kfree_skb(skb2);
4049 return;
4053 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4054 * the bond has a usable interface.
4056 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4058 struct bonding *bond = bond_dev->priv;
4059 int res = 1;
4061 read_lock(&bond->lock);
4062 read_lock(&bond->curr_slave_lock);
4064 if (!BOND_IS_OK(bond)) {
4065 goto out;
4068 if (!bond->curr_active_slave)
4069 goto out;
4071 /* Xmit IGMP frames on all slaves to ensure rapid fail-over
4072 for multicast traffic on snooping switches */
4073 if (skb->protocol == __constant_htons(ETH_P_IP) &&
4074 skb->nh.iph->protocol == IPPROTO_IGMP) {
4075 struct slave *slave, *active_slave;
4076 int i;
4078 active_slave = bond->curr_active_slave;
4079 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
4080 active_slave->prev)
4081 if (IS_UP(slave->dev) &&
4082 (slave->link == BOND_LINK_UP))
4083 bond_activebackup_xmit_copy(skb, bond, slave);
4086 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4088 out:
4089 if (res) {
4090 /* no suitable interface, frame not sent */
4091 dev_kfree_skb(skb);
4093 read_unlock(&bond->curr_slave_lock);
4094 read_unlock(&bond->lock);
4095 return 0;
4099 * In bond_xmit_xor() , we determine the output device by using a pre-
4100 * determined xmit_hash_policy(), If the selected device is not enabled,
4101 * find the next active slave.
4103 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4105 struct bonding *bond = bond_dev->priv;
4106 struct slave *slave, *start_at;
4107 int slave_no;
4108 int i;
4109 int res = 1;
4111 read_lock(&bond->lock);
4113 if (!BOND_IS_OK(bond)) {
4114 goto out;
4117 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4119 bond_for_each_slave(bond, slave, i) {
4120 slave_no--;
4121 if (slave_no < 0) {
4122 break;
4126 start_at = slave;
4128 bond_for_each_slave_from(bond, slave, i, start_at) {
4129 if (IS_UP(slave->dev) &&
4130 (slave->link == BOND_LINK_UP) &&
4131 (slave->state == BOND_STATE_ACTIVE)) {
4132 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4133 break;
4137 out:
4138 if (res) {
4139 /* no suitable interface, frame not sent */
4140 dev_kfree_skb(skb);
4142 read_unlock(&bond->lock);
4143 return 0;
4147 * in broadcast mode, we send everything to all usable interfaces.
4149 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4151 struct bonding *bond = bond_dev->priv;
4152 struct slave *slave, *start_at;
4153 struct net_device *tx_dev = NULL;
4154 int i;
4155 int res = 1;
4157 read_lock(&bond->lock);
4159 if (!BOND_IS_OK(bond)) {
4160 goto out;
4163 read_lock(&bond->curr_slave_lock);
4164 start_at = bond->curr_active_slave;
4165 read_unlock(&bond->curr_slave_lock);
4167 if (!start_at) {
4168 goto out;
4171 bond_for_each_slave_from(bond, slave, i, start_at) {
4172 if (IS_UP(slave->dev) &&
4173 (slave->link == BOND_LINK_UP) &&
4174 (slave->state == BOND_STATE_ACTIVE)) {
4175 if (tx_dev) {
4176 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4177 if (!skb2) {
4178 printk(KERN_ERR DRV_NAME
4179 ": %s: Error: bond_xmit_broadcast(): "
4180 "skb_clone() failed\n",
4181 bond_dev->name);
4182 continue;
4185 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4186 if (res) {
4187 dev_kfree_skb(skb2);
4188 continue;
4191 tx_dev = slave->dev;
4195 if (tx_dev) {
4196 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4199 out:
4200 if (res) {
4201 /* no suitable interface, frame not sent */
4202 dev_kfree_skb(skb);
4204 /* frame sent to all suitable interfaces */
4205 read_unlock(&bond->lock);
4206 return 0;
4209 /*------------------------- Device initialization ---------------------------*/
4212 * set bond mode specific net device operations
4214 void bond_set_mode_ops(struct bonding *bond, int mode)
4216 struct net_device *bond_dev = bond->dev;
4218 switch (mode) {
4219 case BOND_MODE_ROUNDROBIN:
4220 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4221 break;
4222 case BOND_MODE_ACTIVEBACKUP:
4223 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4224 break;
4225 case BOND_MODE_XOR:
4226 bond_dev->hard_start_xmit = bond_xmit_xor;
4227 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4228 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4229 else
4230 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4231 break;
4232 case BOND_MODE_BROADCAST:
4233 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4234 break;
4235 case BOND_MODE_8023AD:
4236 bond_set_master_3ad_flags(bond);
4237 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4238 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4239 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4240 else
4241 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4242 break;
4243 case BOND_MODE_ALB:
4244 bond_set_master_alb_flags(bond);
4245 /* FALLTHRU */
4246 case BOND_MODE_TLB:
4247 bond_dev->hard_start_xmit = bond_alb_xmit;
4248 bond_dev->set_mac_address = bond_alb_set_mac_address;
4249 break;
4250 default:
4251 /* Should never happen, mode already checked */
4252 printk(KERN_ERR DRV_NAME
4253 ": %s: Error: Unknown bonding mode %d\n",
4254 bond_dev->name,
4255 mode);
4256 break;
4260 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4261 struct ethtool_drvinfo *drvinfo)
4263 strncpy(drvinfo->driver, DRV_NAME, 32);
4264 strncpy(drvinfo->version, DRV_VERSION, 32);
4265 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4268 static const struct ethtool_ops bond_ethtool_ops = {
4269 .get_tx_csum = ethtool_op_get_tx_csum,
4270 .get_tso = ethtool_op_get_tso,
4271 .get_ufo = ethtool_op_get_ufo,
4272 .get_sg = ethtool_op_get_sg,
4273 .get_drvinfo = bond_ethtool_get_drvinfo,
4277 * Does not allocate but creates a /proc entry.
4278 * Allowed to fail.
4280 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4282 struct bonding *bond = bond_dev->priv;
4284 dprintk("Begin bond_init for %s\n", bond_dev->name);
4286 /* initialize rwlocks */
4287 rwlock_init(&bond->lock);
4288 rwlock_init(&bond->curr_slave_lock);
4290 bond->params = *params; /* copy params struct */
4292 /* Initialize pointers */
4293 bond->first_slave = NULL;
4294 bond->curr_active_slave = NULL;
4295 bond->current_arp_slave = NULL;
4296 bond->primary_slave = NULL;
4297 bond->dev = bond_dev;
4298 INIT_LIST_HEAD(&bond->vlan_list);
4300 /* Initialize the device entry points */
4301 bond_dev->open = bond_open;
4302 bond_dev->stop = bond_close;
4303 bond_dev->get_stats = bond_get_stats;
4304 bond_dev->do_ioctl = bond_do_ioctl;
4305 bond_dev->ethtool_ops = &bond_ethtool_ops;
4306 bond_dev->set_multicast_list = bond_set_multicast_list;
4307 bond_dev->change_mtu = bond_change_mtu;
4308 bond_dev->set_mac_address = bond_set_mac_address;
4310 bond_set_mode_ops(bond, bond->params.mode);
4312 bond_dev->destructor = free_netdev;
4314 /* Initialize the device options */
4315 bond_dev->tx_queue_len = 0;
4316 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4317 bond_dev->priv_flags |= IFF_BONDING;
4319 /* At first, we block adding VLANs. That's the only way to
4320 * prevent problems that occur when adding VLANs over an
4321 * empty bond. The block will be removed once non-challenged
4322 * slaves are enslaved.
4324 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4326 /* don't acquire bond device's netif_tx_lock when
4327 * transmitting */
4328 bond_dev->features |= NETIF_F_LLTX;
4330 /* By default, we declare the bond to be fully
4331 * VLAN hardware accelerated capable. Special
4332 * care is taken in the various xmit functions
4333 * when there are slaves that are not hw accel
4334 * capable
4336 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4337 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4338 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4339 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4340 NETIF_F_HW_VLAN_RX |
4341 NETIF_F_HW_VLAN_FILTER);
4343 #ifdef CONFIG_PROC_FS
4344 bond_create_proc_entry(bond);
4345 #endif
4347 list_add_tail(&bond->bond_list, &bond_dev_list);
4349 return 0;
4352 /* De-initialize device specific data.
4353 * Caller must hold rtnl_lock.
4355 void bond_deinit(struct net_device *bond_dev)
4357 struct bonding *bond = bond_dev->priv;
4359 list_del(&bond->bond_list);
4361 #ifdef CONFIG_PROC_FS
4362 bond_remove_proc_entry(bond);
4363 #endif
4366 /* Unregister and free all bond devices.
4367 * Caller must hold rtnl_lock.
4369 static void bond_free_all(void)
4371 struct bonding *bond, *nxt;
4373 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4374 struct net_device *bond_dev = bond->dev;
4376 bond_mc_list_destroy(bond);
4377 /* Release the bonded slaves */
4378 bond_release_all(bond_dev);
4379 unregister_netdevice(bond_dev);
4380 bond_deinit(bond_dev);
4383 #ifdef CONFIG_PROC_FS
4384 bond_destroy_proc_dir();
4385 #endif
4388 /*------------------------- Module initialization ---------------------------*/
4391 * Convert string input module parms. Accept either the
4392 * number of the mode or its string name.
4394 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4396 int i;
4398 for (i = 0; tbl[i].modename; i++) {
4399 if ((isdigit(*mode_arg) &&
4400 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4401 (strncmp(mode_arg, tbl[i].modename,
4402 strlen(tbl[i].modename)) == 0)) {
4403 return tbl[i].mode;
4407 return -1;
4410 static int bond_check_params(struct bond_params *params)
4412 int arp_validate_value;
4415 * Convert string parameters.
4417 if (mode) {
4418 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4419 if (bond_mode == -1) {
4420 printk(KERN_ERR DRV_NAME
4421 ": Error: Invalid bonding mode \"%s\"\n",
4422 mode == NULL ? "NULL" : mode);
4423 return -EINVAL;
4427 if (xmit_hash_policy) {
4428 if ((bond_mode != BOND_MODE_XOR) &&
4429 (bond_mode != BOND_MODE_8023AD)) {
4430 printk(KERN_INFO DRV_NAME
4431 ": xor_mode param is irrelevant in mode %s\n",
4432 bond_mode_name(bond_mode));
4433 } else {
4434 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4435 xmit_hashtype_tbl);
4436 if (xmit_hashtype == -1) {
4437 printk(KERN_ERR DRV_NAME
4438 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4439 xmit_hash_policy == NULL ? "NULL" :
4440 xmit_hash_policy);
4441 return -EINVAL;
4446 if (lacp_rate) {
4447 if (bond_mode != BOND_MODE_8023AD) {
4448 printk(KERN_INFO DRV_NAME
4449 ": lacp_rate param is irrelevant in mode %s\n",
4450 bond_mode_name(bond_mode));
4451 } else {
4452 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4453 if (lacp_fast == -1) {
4454 printk(KERN_ERR DRV_NAME
4455 ": Error: Invalid lacp rate \"%s\"\n",
4456 lacp_rate == NULL ? "NULL" : lacp_rate);
4457 return -EINVAL;
4462 if (max_bonds < 1 || max_bonds > INT_MAX) {
4463 printk(KERN_WARNING DRV_NAME
4464 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4465 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4466 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4467 max_bonds = BOND_DEFAULT_MAX_BONDS;
4470 if (miimon < 0) {
4471 printk(KERN_WARNING DRV_NAME
4472 ": Warning: miimon module parameter (%d), "
4473 "not in range 0-%d, so it was reset to %d\n",
4474 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4475 miimon = BOND_LINK_MON_INTERV;
4478 if (updelay < 0) {
4479 printk(KERN_WARNING DRV_NAME
4480 ": Warning: updelay module parameter (%d), "
4481 "not in range 0-%d, so it was reset to 0\n",
4482 updelay, INT_MAX);
4483 updelay = 0;
4486 if (downdelay < 0) {
4487 printk(KERN_WARNING DRV_NAME
4488 ": Warning: downdelay module parameter (%d), "
4489 "not in range 0-%d, so it was reset to 0\n",
4490 downdelay, INT_MAX);
4491 downdelay = 0;
4494 if ((use_carrier != 0) && (use_carrier != 1)) {
4495 printk(KERN_WARNING DRV_NAME
4496 ": Warning: use_carrier module parameter (%d), "
4497 "not of valid value (0/1), so it was set to 1\n",
4498 use_carrier);
4499 use_carrier = 1;
4502 /* reset values for 802.3ad */
4503 if (bond_mode == BOND_MODE_8023AD) {
4504 if (!miimon) {
4505 printk(KERN_WARNING DRV_NAME
4506 ": Warning: miimon must be specified, "
4507 "otherwise bonding will not detect link "
4508 "failure, speed and duplex which are "
4509 "essential for 802.3ad operation\n");
4510 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4511 miimon = 100;
4515 /* reset values for TLB/ALB */
4516 if ((bond_mode == BOND_MODE_TLB) ||
4517 (bond_mode == BOND_MODE_ALB)) {
4518 if (!miimon) {
4519 printk(KERN_WARNING DRV_NAME
4520 ": Warning: miimon must be specified, "
4521 "otherwise bonding will not detect link "
4522 "failure and link speed which are essential "
4523 "for TLB/ALB load balancing\n");
4524 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4525 miimon = 100;
4529 if (bond_mode == BOND_MODE_ALB) {
4530 printk(KERN_NOTICE DRV_NAME
4531 ": In ALB mode you might experience client "
4532 "disconnections upon reconnection of a link if the "
4533 "bonding module updelay parameter (%d msec) is "
4534 "incompatible with the forwarding delay time of the "
4535 "switch\n",
4536 updelay);
4539 if (!miimon) {
4540 if (updelay || downdelay) {
4541 /* just warn the user the up/down delay will have
4542 * no effect since miimon is zero...
4544 printk(KERN_WARNING DRV_NAME
4545 ": Warning: miimon module parameter not set "
4546 "and updelay (%d) or downdelay (%d) module "
4547 "parameter is set; updelay and downdelay have "
4548 "no effect unless miimon is set\n",
4549 updelay, downdelay);
4551 } else {
4552 /* don't allow arp monitoring */
4553 if (arp_interval) {
4554 printk(KERN_WARNING DRV_NAME
4555 ": Warning: miimon (%d) and arp_interval (%d) "
4556 "can't be used simultaneously, disabling ARP "
4557 "monitoring\n",
4558 miimon, arp_interval);
4559 arp_interval = 0;
4562 if ((updelay % miimon) != 0) {
4563 printk(KERN_WARNING DRV_NAME
4564 ": Warning: updelay (%d) is not a multiple "
4565 "of miimon (%d), updelay rounded to %d ms\n",
4566 updelay, miimon, (updelay / miimon) * miimon);
4569 updelay /= miimon;
4571 if ((downdelay % miimon) != 0) {
4572 printk(KERN_WARNING DRV_NAME
4573 ": Warning: downdelay (%d) is not a multiple "
4574 "of miimon (%d), downdelay rounded to %d ms\n",
4575 downdelay, miimon,
4576 (downdelay / miimon) * miimon);
4579 downdelay /= miimon;
4582 if (arp_interval < 0) {
4583 printk(KERN_WARNING DRV_NAME
4584 ": Warning: arp_interval module parameter (%d) "
4585 ", not in range 0-%d, so it was reset to %d\n",
4586 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4587 arp_interval = BOND_LINK_ARP_INTERV;
4590 for (arp_ip_count = 0;
4591 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4592 arp_ip_count++) {
4593 /* not complete check, but should be good enough to
4594 catch mistakes */
4595 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4596 printk(KERN_WARNING DRV_NAME
4597 ": Warning: bad arp_ip_target module parameter "
4598 "(%s), ARP monitoring will not be performed\n",
4599 arp_ip_target[arp_ip_count]);
4600 arp_interval = 0;
4601 } else {
4602 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4603 arp_target[arp_ip_count] = ip;
4607 if (arp_interval && !arp_ip_count) {
4608 /* don't allow arping if no arp_ip_target given... */
4609 printk(KERN_WARNING DRV_NAME
4610 ": Warning: arp_interval module parameter (%d) "
4611 "specified without providing an arp_ip_target "
4612 "parameter, arp_interval was reset to 0\n",
4613 arp_interval);
4614 arp_interval = 0;
4617 if (arp_validate) {
4618 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4619 printk(KERN_ERR DRV_NAME
4620 ": arp_validate only supported in active-backup mode\n");
4621 return -EINVAL;
4623 if (!arp_interval) {
4624 printk(KERN_ERR DRV_NAME
4625 ": arp_validate requires arp_interval\n");
4626 return -EINVAL;
4629 arp_validate_value = bond_parse_parm(arp_validate,
4630 arp_validate_tbl);
4631 if (arp_validate_value == -1) {
4632 printk(KERN_ERR DRV_NAME
4633 ": Error: invalid arp_validate \"%s\"\n",
4634 arp_validate == NULL ? "NULL" : arp_validate);
4635 return -EINVAL;
4637 } else
4638 arp_validate_value = 0;
4640 if (miimon) {
4641 printk(KERN_INFO DRV_NAME
4642 ": MII link monitoring set to %d ms\n",
4643 miimon);
4644 } else if (arp_interval) {
4645 int i;
4647 printk(KERN_INFO DRV_NAME
4648 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4649 arp_interval,
4650 arp_validate_tbl[arp_validate_value].modename,
4651 arp_ip_count);
4653 for (i = 0; i < arp_ip_count; i++)
4654 printk (" %s", arp_ip_target[i]);
4656 printk("\n");
4658 } else {
4659 /* miimon and arp_interval not set, we need one so things
4660 * work as expected, see bonding.txt for details
4662 printk(KERN_WARNING DRV_NAME
4663 ": Warning: either miimon or arp_interval and "
4664 "arp_ip_target module parameters must be specified, "
4665 "otherwise bonding will not detect link failures! see "
4666 "bonding.txt for details.\n");
4669 if (primary && !USES_PRIMARY(bond_mode)) {
4670 /* currently, using a primary only makes sense
4671 * in active backup, TLB or ALB modes
4673 printk(KERN_WARNING DRV_NAME
4674 ": Warning: %s primary device specified but has no "
4675 "effect in %s mode\n",
4676 primary, bond_mode_name(bond_mode));
4677 primary = NULL;
4680 /* fill params struct with the proper values */
4681 params->mode = bond_mode;
4682 params->xmit_policy = xmit_hashtype;
4683 params->miimon = miimon;
4684 params->arp_interval = arp_interval;
4685 params->arp_validate = arp_validate_value;
4686 params->updelay = updelay;
4687 params->downdelay = downdelay;
4688 params->use_carrier = use_carrier;
4689 params->lacp_fast = lacp_fast;
4690 params->primary[0] = 0;
4692 if (primary) {
4693 strncpy(params->primary, primary, IFNAMSIZ);
4694 params->primary[IFNAMSIZ - 1] = 0;
4697 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4699 return 0;
4702 static struct lock_class_key bonding_netdev_xmit_lock_key;
4704 /* Create a new bond based on the specified name and bonding parameters.
4705 * If name is NULL, obtain a suitable "bond%d" name for us.
4706 * Caller must NOT hold rtnl_lock; we need to release it here before we
4707 * set up our sysfs entries.
4709 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4711 struct net_device *bond_dev;
4712 int res;
4714 rtnl_lock();
4715 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4716 ether_setup);
4717 if (!bond_dev) {
4718 printk(KERN_ERR DRV_NAME
4719 ": %s: eek! can't alloc netdev!\n",
4720 name);
4721 res = -ENOMEM;
4722 goto out_rtnl;
4725 if (!name) {
4726 res = dev_alloc_name(bond_dev, "bond%d");
4727 if (res < 0)
4728 goto out_netdev;
4731 /* bond_init() must be called after dev_alloc_name() (for the
4732 * /proc files), but before register_netdevice(), because we
4733 * need to set function pointers.
4736 res = bond_init(bond_dev, params);
4737 if (res < 0) {
4738 goto out_netdev;
4741 SET_MODULE_OWNER(bond_dev);
4743 res = register_netdevice(bond_dev);
4744 if (res < 0) {
4745 goto out_bond;
4748 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4750 if (newbond)
4751 *newbond = bond_dev->priv;
4753 netif_carrier_off(bond_dev);
4755 rtnl_unlock(); /* allows sysfs registration of net device */
4756 res = bond_create_sysfs_entry(bond_dev->priv);
4757 if (res < 0) {
4758 rtnl_lock();
4759 goto out_bond;
4762 return 0;
4764 out_bond:
4765 bond_deinit(bond_dev);
4766 out_netdev:
4767 free_netdev(bond_dev);
4768 out_rtnl:
4769 rtnl_unlock();
4770 return res;
4773 static int __init bonding_init(void)
4775 int i;
4776 int res;
4778 printk(KERN_INFO "%s", version);
4780 res = bond_check_params(&bonding_defaults);
4781 if (res) {
4782 goto out;
4785 #ifdef CONFIG_PROC_FS
4786 bond_create_proc_dir();
4787 #endif
4788 for (i = 0; i < max_bonds; i++) {
4789 res = bond_create(NULL, &bonding_defaults, NULL);
4790 if (res)
4791 goto err;
4794 res = bond_create_sysfs();
4795 if (res)
4796 goto err;
4798 register_netdevice_notifier(&bond_netdev_notifier);
4799 register_inetaddr_notifier(&bond_inetaddr_notifier);
4801 goto out;
4802 err:
4803 rtnl_lock();
4804 bond_free_all();
4805 bond_destroy_sysfs();
4806 rtnl_unlock();
4807 out:
4808 return res;
4812 static void __exit bonding_exit(void)
4814 unregister_netdevice_notifier(&bond_netdev_notifier);
4815 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4817 rtnl_lock();
4818 bond_free_all();
4819 bond_destroy_sysfs();
4820 rtnl_unlock();
4823 module_init(bonding_init);
4824 module_exit(bonding_exit);
4825 MODULE_LICENSE("GPL");
4826 MODULE_VERSION(DRV_VERSION);
4827 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4828 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4829 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4832 * Local variables:
4833 * c-indent-level: 8
4834 * c-basic-offset: 8
4835 * tab-width: 8
4836 * End: