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:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
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.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 //#define BONDING_DEBUG 1
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <asm/system.h>
59 #include <asm/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
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>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV 0
88 #define BOND_LINK_ARP_INTERV 0
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int num_grat_arp = 1;
92 static int num_unsol_na = 1;
93 static int miimon = BOND_LINK_MON_INTERV;
94 static int updelay = 0;
95 static int downdelay = 0;
96 static int use_carrier = 1;
97 static char *mode = NULL;
98 static char *primary = NULL;
99 static char *lacp_rate = NULL;
100 static char *ad_select = NULL;
101 static char *xmit_hash_policy = NULL;
102 static int arp_interval = BOND_LINK_ARP_INTERV;
103 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
104 static char *arp_validate = NULL;
105 static char *fail_over_mac = NULL;
106 struct bond_params bonding_defaults;
108 module_param(max_bonds, int, 0);
109 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
110 module_param(num_grat_arp, int, 0644);
111 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
112 module_param(num_unsol_na, int, 0644);
113 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
114 module_param(miimon, int, 0);
115 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
116 module_param(updelay, int, 0);
117 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
118 module_param(downdelay, int, 0);
119 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
121 module_param(use_carrier, int, 0);
122 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
123 "0 for off, 1 for on (default)");
124 module_param(mode, charp, 0);
125 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
126 "1 for active-backup, 2 for balance-xor, "
127 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
128 "6 for balance-alb");
129 module_param(primary, charp, 0);
130 MODULE_PARM_DESC(primary, "Primary network device to use");
131 module_param(lacp_rate, charp, 0);
132 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
134 module_param(ad_select, charp, 0);
135 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
136 module_param(xmit_hash_policy, charp, 0);
137 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
138 ", 1 for layer 3+4");
139 module_param(arp_interval, int, 0);
140 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
141 module_param_array(arp_ip_target, charp, NULL, 0);
142 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
143 module_param(arp_validate, charp, 0);
144 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
145 module_param(fail_over_mac, charp, 0);
146 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
148 /*----------------------------- Global variables ----------------------------*/
150 static const char * const version =
151 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
153 LIST_HEAD(bond_dev_list);
155 #ifdef CONFIG_PROC_FS
156 static struct proc_dir_entry *bond_proc_dir = NULL;
159 extern struct rw_semaphore bonding_rwsem;
160 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
161 static int arp_ip_count = 0;
162 static int bond_mode = BOND_MODE_ROUNDROBIN;
163 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
164 static int lacp_fast = 0;
167 struct bond_parm_tbl bond_lacp_tbl[] = {
168 { "slow", AD_LACP_SLOW},
169 { "fast", AD_LACP_FAST},
173 struct bond_parm_tbl bond_mode_tbl[] = {
174 { "balance-rr", BOND_MODE_ROUNDROBIN},
175 { "active-backup", BOND_MODE_ACTIVEBACKUP},
176 { "balance-xor", BOND_MODE_XOR},
177 { "broadcast", BOND_MODE_BROADCAST},
178 { "802.3ad", BOND_MODE_8023AD},
179 { "balance-tlb", BOND_MODE_TLB},
180 { "balance-alb", BOND_MODE_ALB},
184 struct bond_parm_tbl xmit_hashtype_tbl[] = {
185 { "layer2", BOND_XMIT_POLICY_LAYER2},
186 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
187 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
191 struct bond_parm_tbl arp_validate_tbl[] = {
192 { "none", BOND_ARP_VALIDATE_NONE},
193 { "active", BOND_ARP_VALIDATE_ACTIVE},
194 { "backup", BOND_ARP_VALIDATE_BACKUP},
195 { "all", BOND_ARP_VALIDATE_ALL},
199 struct bond_parm_tbl fail_over_mac_tbl[] = {
200 { "none", BOND_FOM_NONE},
201 { "active", BOND_FOM_ACTIVE},
202 { "follow", BOND_FOM_FOLLOW},
206 struct bond_parm_tbl ad_select_tbl[] = {
207 { "stable", BOND_AD_STABLE},
208 { "bandwidth", BOND_AD_BANDWIDTH},
209 { "count", BOND_AD_COUNT},
213 /*-------------------------- Forward declarations ---------------------------*/
215 static void bond_send_gratuitous_arp(struct bonding *bond);
216 static void bond_deinit(struct net_device *bond_dev);
218 /*---------------------------- General routines -----------------------------*/
220 static const char *bond_mode_name(int mode)
222 static const char *names[] = {
223 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
224 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
225 [BOND_MODE_XOR] = "load balancing (xor)",
226 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
227 [BOND_MODE_8023AD]= "IEEE 802.3ad Dynamic link aggregation",
228 [BOND_MODE_TLB] = "transmit load balancing",
229 [BOND_MODE_ALB] = "adaptive load balancing",
232 if (mode < 0 || mode > BOND_MODE_ALB)
238 /*---------------------------------- VLAN -----------------------------------*/
241 * bond_add_vlan - add a new vlan id on bond
242 * @bond: bond that got the notification
243 * @vlan_id: the vlan id to add
245 * Returns -ENOMEM if allocation failed.
247 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
249 struct vlan_entry *vlan;
251 dprintk("bond: %s, vlan id %d\n",
252 (bond ? bond->dev->name: "None"), vlan_id);
254 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
259 INIT_LIST_HEAD(&vlan->vlan_list);
260 vlan->vlan_id = vlan_id;
262 write_lock_bh(&bond->lock);
264 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
266 write_unlock_bh(&bond->lock);
268 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
274 * bond_del_vlan - delete a vlan id from bond
275 * @bond: bond that got the notification
276 * @vlan_id: the vlan id to delete
278 * returns -ENODEV if @vlan_id was not found in @bond.
280 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
282 struct vlan_entry *vlan;
285 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
287 write_lock_bh(&bond->lock);
289 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
290 if (vlan->vlan_id == vlan_id) {
291 list_del(&vlan->vlan_list);
293 if (bond_is_lb(bond))
294 bond_alb_clear_vlan(bond, vlan_id);
296 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
301 if (list_empty(&bond->vlan_list) &&
302 (bond->slave_cnt == 0)) {
303 /* Last VLAN removed and no slaves, so
304 * restore block on adding VLANs. This will
305 * be removed once new slaves that are not
306 * VLAN challenged will be added.
308 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
316 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
320 write_unlock_bh(&bond->lock);
325 * bond_has_challenged_slaves
326 * @bond: the bond we're working on
328 * Searches the slave list. Returns 1 if a vlan challenged slave
329 * was found, 0 otherwise.
331 * Assumes bond->lock is held.
333 static int bond_has_challenged_slaves(struct bonding *bond)
338 bond_for_each_slave(bond, slave, i) {
339 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
340 dprintk("found VLAN challenged slave - %s\n",
346 dprintk("no VLAN challenged slaves found\n");
351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
358 * Caller must hold bond->lock
360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
362 struct vlan_entry *next, *last;
364 if (list_empty(&bond->vlan_list)) {
369 next = list_entry(bond->vlan_list.next,
370 struct vlan_entry, vlan_list);
372 last = list_entry(bond->vlan_list.prev,
373 struct vlan_entry, vlan_list);
375 next = list_entry(bond->vlan_list.next,
376 struct vlan_entry, vlan_list);
378 next = list_entry(curr->vlan_list.next,
379 struct vlan_entry, vlan_list);
387 * bond_dev_queue_xmit - Prepare skb for xmit.
389 * @bond: bond device that got this skb for tx.
390 * @skb: hw accel VLAN tagged skb to transmit
391 * @slave_dev: slave that is supposed to xmit this skbuff
393 * When the bond gets an skb to transmit that is
394 * already hardware accelerated VLAN tagged, and it
395 * needs to relay this skb to a slave that is not
396 * hw accel capable, the skb needs to be "unaccelerated",
397 * i.e. strip the hwaccel tag and re-insert it as part
400 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
402 unsigned short uninitialized_var(vlan_id);
404 if (!list_empty(&bond->vlan_list) &&
405 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
406 vlan_get_tag(skb, &vlan_id) == 0) {
407 skb->dev = slave_dev;
408 skb = vlan_put_tag(skb, vlan_id);
410 /* vlan_put_tag() frees the skb in case of error,
411 * so return success here so the calling functions
412 * won't attempt to free is again.
417 skb->dev = slave_dev;
427 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
428 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
430 * a. This operation is performed in IOCTL context,
431 * b. The operation is protected by the RTNL semaphore in the 8021q code,
432 * c. Holding a lock with BH disabled while directly calling a base driver
433 * entry point is generally a BAD idea.
435 * The design of synchronization/protection for this operation in the 8021q
436 * module is good for one or more VLAN devices over a single physical device
437 * and cannot be extended for a teaming solution like bonding, so there is a
438 * potential race condition here where a net device from the vlan group might
439 * be referenced (either by a base driver or the 8021q code) while it is being
440 * removed from the system. However, it turns out we're not making matters
441 * worse, and if it works for regular VLAN usage it will work here too.
445 * bond_vlan_rx_register - Propagates registration to slaves
446 * @bond_dev: bonding net device that got called
447 * @grp: vlan group being registered
449 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
451 struct bonding *bond = netdev_priv(bond_dev);
457 bond_for_each_slave(bond, slave, i) {
458 struct net_device *slave_dev = slave->dev;
459 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
461 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
462 slave_ops->ndo_vlan_rx_register) {
463 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
469 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
470 * @bond_dev: bonding net device that got called
471 * @vid: vlan id being added
473 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
475 struct bonding *bond = netdev_priv(bond_dev);
479 bond_for_each_slave(bond, slave, i) {
480 struct net_device *slave_dev = slave->dev;
481 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
483 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
484 slave_ops->ndo_vlan_rx_add_vid) {
485 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
489 res = bond_add_vlan(bond, vid);
491 printk(KERN_ERR DRV_NAME
492 ": %s: Error: Failed to add vlan id %d\n",
493 bond_dev->name, vid);
498 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
499 * @bond_dev: bonding net device that got called
500 * @vid: vlan id being removed
502 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
504 struct bonding *bond = netdev_priv(bond_dev);
506 struct net_device *vlan_dev;
509 bond_for_each_slave(bond, slave, i) {
510 struct net_device *slave_dev = slave->dev;
511 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
513 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
514 slave_ops->ndo_vlan_rx_kill_vid) {
515 /* Save and then restore vlan_dev in the grp array,
516 * since the slave's driver might clear it.
518 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
519 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
520 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
524 res = bond_del_vlan(bond, vid);
526 printk(KERN_ERR DRV_NAME
527 ": %s: Error: Failed to remove vlan id %d\n",
528 bond_dev->name, vid);
532 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
534 struct vlan_entry *vlan;
535 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
537 write_lock_bh(&bond->lock);
539 if (list_empty(&bond->vlan_list))
542 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
543 slave_ops->ndo_vlan_rx_register)
544 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
546 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
547 !(slave_ops->ndo_vlan_rx_add_vid))
550 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
551 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
554 write_unlock_bh(&bond->lock);
557 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
559 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
560 struct vlan_entry *vlan;
561 struct net_device *vlan_dev;
563 write_lock_bh(&bond->lock);
565 if (list_empty(&bond->vlan_list))
568 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
569 !(slave_ops->ndo_vlan_rx_kill_vid))
572 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
573 /* Save and then restore vlan_dev in the grp array,
574 * since the slave's driver might clear it.
576 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
577 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
578 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
582 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
583 slave_ops->ndo_vlan_rx_register)
584 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
587 write_unlock_bh(&bond->lock);
590 /*------------------------------- Link status -------------------------------*/
593 * Set the carrier state for the master according to the state of its
594 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
595 * do special 802.3ad magic.
597 * Returns zero if carrier state does not change, nonzero if it does.
599 static int bond_set_carrier(struct bonding *bond)
604 if (bond->slave_cnt == 0)
607 if (bond->params.mode == BOND_MODE_8023AD)
608 return bond_3ad_set_carrier(bond);
610 bond_for_each_slave(bond, slave, i) {
611 if (slave->link == BOND_LINK_UP) {
612 if (!netif_carrier_ok(bond->dev)) {
613 netif_carrier_on(bond->dev);
621 if (netif_carrier_ok(bond->dev)) {
622 netif_carrier_off(bond->dev);
629 * Get link speed and duplex from the slave's base driver
630 * using ethtool. If for some reason the call fails or the
631 * values are invalid, fake speed and duplex to 100/Full
634 static int bond_update_speed_duplex(struct slave *slave)
636 struct net_device *slave_dev = slave->dev;
637 struct ethtool_cmd etool;
640 /* Fake speed and duplex */
641 slave->speed = SPEED_100;
642 slave->duplex = DUPLEX_FULL;
644 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
647 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
651 switch (etool.speed) {
661 switch (etool.duplex) {
669 slave->speed = etool.speed;
670 slave->duplex = etool.duplex;
676 * if <dev> supports MII link status reporting, check its link status.
678 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
679 * depening upon the setting of the use_carrier parameter.
681 * Return either BMSR_LSTATUS, meaning that the link is up (or we
682 * can't tell and just pretend it is), or 0, meaning that the link is
685 * If reporting is non-zero, instead of faking link up, return -1 if
686 * both ETHTOOL and MII ioctls fail (meaning the device does not
687 * support them). If use_carrier is set, return whatever it says.
688 * It'd be nice if there was a good way to tell if a driver supports
689 * netif_carrier, but there really isn't.
691 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
693 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
694 static int (* ioctl)(struct net_device *, struct ifreq *, int);
696 struct mii_ioctl_data *mii;
698 if (bond->params.use_carrier)
699 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
701 ioctl = slave_ops->ndo_do_ioctl;
703 /* TODO: set pointer to correct ioctl on a per team member */
704 /* bases to make this more efficient. that is, once */
705 /* we determine the correct ioctl, we will always */
706 /* call it and not the others for that team */
710 * We cannot assume that SIOCGMIIPHY will also read a
711 * register; not all network drivers (e.g., e100)
715 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
716 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
718 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
719 mii->reg_num = MII_BMSR;
720 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
721 return (mii->val_out & BMSR_LSTATUS);
727 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
728 * attempt to get link status from it if the above MII ioctls fail.
730 if (slave_dev->ethtool_ops) {
731 if (slave_dev->ethtool_ops->get_link) {
734 link = slave_dev->ethtool_ops->get_link(slave_dev);
736 return link ? BMSR_LSTATUS : 0;
741 * If reporting, report that either there's no dev->do_ioctl,
742 * or both SIOCGMIIREG and get_link failed (meaning that we
743 * cannot report link status). If not reporting, pretend
746 return (reporting ? -1 : BMSR_LSTATUS);
749 /*----------------------------- Multicast list ------------------------------*/
752 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
754 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
756 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
757 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
761 * returns dmi entry if found, NULL otherwise
763 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
765 struct dev_mc_list *idmi;
767 for (idmi = mc_list; idmi; idmi = idmi->next) {
768 if (bond_is_dmi_same(dmi, idmi)) {
777 * Push the promiscuity flag down to appropriate slaves
779 static int bond_set_promiscuity(struct bonding *bond, int inc)
782 if (USES_PRIMARY(bond->params.mode)) {
783 /* write lock already acquired */
784 if (bond->curr_active_slave) {
785 err = dev_set_promiscuity(bond->curr_active_slave->dev,
791 bond_for_each_slave(bond, slave, i) {
792 err = dev_set_promiscuity(slave->dev, inc);
801 * Push the allmulti flag down to all slaves
803 static int bond_set_allmulti(struct bonding *bond, int inc)
806 if (USES_PRIMARY(bond->params.mode)) {
807 /* write lock already acquired */
808 if (bond->curr_active_slave) {
809 err = dev_set_allmulti(bond->curr_active_slave->dev,
815 bond_for_each_slave(bond, slave, i) {
816 err = dev_set_allmulti(slave->dev, inc);
825 * Add a Multicast address to slaves
828 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
830 if (USES_PRIMARY(bond->params.mode)) {
831 /* write lock already acquired */
832 if (bond->curr_active_slave) {
833 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
838 bond_for_each_slave(bond, slave, i) {
839 dev_mc_add(slave->dev, addr, alen, 0);
845 * Remove a multicast address from slave
848 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
850 if (USES_PRIMARY(bond->params.mode)) {
851 /* write lock already acquired */
852 if (bond->curr_active_slave) {
853 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
858 bond_for_each_slave(bond, slave, i) {
859 dev_mc_delete(slave->dev, addr, alen, 0);
866 * Retrieve the list of registered multicast addresses for the bonding
867 * device and retransmit an IGMP JOIN request to the current active
870 static void bond_resend_igmp_join_requests(struct bonding *bond)
872 struct in_device *in_dev;
873 struct ip_mc_list *im;
876 in_dev = __in_dev_get_rcu(bond->dev);
878 for (im = in_dev->mc_list; im; im = im->next) {
879 ip_mc_rejoin_group(im);
887 * Totally destroys the mc_list in bond
889 static void bond_mc_list_destroy(struct bonding *bond)
891 struct dev_mc_list *dmi;
895 bond->mc_list = dmi->next;
899 bond->mc_list = NULL;
903 * Copy all the Multicast addresses from src to the bonding device dst
905 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
908 struct dev_mc_list *dmi, *new_dmi;
910 for (dmi = mc_list; dmi; dmi = dmi->next) {
911 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
914 /* FIXME: Potential memory leak !!! */
918 new_dmi->next = bond->mc_list;
919 bond->mc_list = new_dmi;
920 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
921 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
922 new_dmi->dmi_users = dmi->dmi_users;
923 new_dmi->dmi_gusers = dmi->dmi_gusers;
930 * flush all members of flush->mc_list from device dev->mc_list
932 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
934 struct bonding *bond = netdev_priv(bond_dev);
935 struct dev_mc_list *dmi;
937 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
938 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
941 if (bond->params.mode == BOND_MODE_8023AD) {
942 /* del lacpdu mc addr from mc list */
943 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
945 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
949 /*--------------------------- Active slave change ---------------------------*/
952 * Update the mc list and multicast-related flags for the new and
953 * old active slaves (if any) according to the multicast mode, and
954 * promiscuous flags unconditionally.
956 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
958 struct dev_mc_list *dmi;
960 if (!USES_PRIMARY(bond->params.mode)) {
961 /* nothing to do - mc list is already up-to-date on
968 if (bond->dev->flags & IFF_PROMISC) {
969 dev_set_promiscuity(old_active->dev, -1);
972 if (bond->dev->flags & IFF_ALLMULTI) {
973 dev_set_allmulti(old_active->dev, -1);
976 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
977 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
982 /* FIXME: Signal errors upstream. */
983 if (bond->dev->flags & IFF_PROMISC) {
984 dev_set_promiscuity(new_active->dev, 1);
987 if (bond->dev->flags & IFF_ALLMULTI) {
988 dev_set_allmulti(new_active->dev, 1);
991 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
992 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
994 bond_resend_igmp_join_requests(bond);
999 * bond_do_fail_over_mac
1001 * Perform special MAC address swapping for fail_over_mac settings
1003 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1005 static void bond_do_fail_over_mac(struct bonding *bond,
1006 struct slave *new_active,
1007 struct slave *old_active)
1009 u8 tmp_mac[ETH_ALEN];
1010 struct sockaddr saddr;
1013 switch (bond->params.fail_over_mac) {
1014 case BOND_FOM_ACTIVE:
1016 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1017 new_active->dev->addr_len);
1019 case BOND_FOM_FOLLOW:
1021 * if new_active && old_active, swap them
1022 * if just old_active, do nothing (going to no active slave)
1023 * if just new_active, set new_active to bond's MAC
1028 write_unlock_bh(&bond->curr_slave_lock);
1029 read_unlock(&bond->lock);
1032 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1033 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1035 saddr.sa_family = new_active->dev->type;
1037 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1038 saddr.sa_family = bond->dev->type;
1041 rv = dev_set_mac_address(new_active->dev, &saddr);
1043 printk(KERN_ERR DRV_NAME
1044 ": %s: Error %d setting MAC of slave %s\n",
1045 bond->dev->name, -rv, new_active->dev->name);
1052 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1053 saddr.sa_family = old_active->dev->type;
1055 rv = dev_set_mac_address(old_active->dev, &saddr);
1057 printk(KERN_ERR DRV_NAME
1058 ": %s: Error %d setting MAC of slave %s\n",
1059 bond->dev->name, -rv, new_active->dev->name);
1061 read_lock(&bond->lock);
1062 write_lock_bh(&bond->curr_slave_lock);
1065 printk(KERN_ERR DRV_NAME
1066 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1067 bond->dev->name, bond->params.fail_over_mac);
1075 * find_best_interface - select the best available slave to be the active one
1076 * @bond: our bonding struct
1078 * Warning: Caller must hold curr_slave_lock for writing.
1080 static struct slave *bond_find_best_slave(struct bonding *bond)
1082 struct slave *new_active, *old_active;
1083 struct slave *bestslave = NULL;
1084 int mintime = bond->params.updelay;
1087 new_active = old_active = bond->curr_active_slave;
1089 if (!new_active) { /* there were no active slaves left */
1090 if (bond->slave_cnt > 0) { /* found one slave */
1091 new_active = bond->first_slave;
1093 return NULL; /* still no slave, return NULL */
1097 /* first try the primary link; if arping, a link must tx/rx traffic
1098 * before it can be considered the curr_active_slave - also, we would skip
1099 * slaves between the curr_active_slave and primary_slave that may be up
1102 if ((bond->primary_slave) &&
1103 (!bond->params.arp_interval) &&
1104 (IS_UP(bond->primary_slave->dev))) {
1105 new_active = bond->primary_slave;
1108 /* remember where to stop iterating over the slaves */
1109 old_active = new_active;
1111 bond_for_each_slave_from(bond, new_active, i, old_active) {
1112 if (IS_UP(new_active->dev)) {
1113 if (new_active->link == BOND_LINK_UP) {
1115 } else if (new_active->link == BOND_LINK_BACK) {
1116 /* link up, but waiting for stabilization */
1117 if (new_active->delay < mintime) {
1118 mintime = new_active->delay;
1119 bestslave = new_active;
1129 * change_active_interface - change the active slave into the specified one
1130 * @bond: our bonding struct
1131 * @new: the new slave to make the active one
1133 * Set the new slave to the bond's settings and unset them on the old
1134 * curr_active_slave.
1135 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1137 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1138 * because it is apparently the best available slave we have, even though its
1139 * updelay hasn't timed out yet.
1141 * If new_active is not NULL, caller must hold bond->lock for read and
1142 * curr_slave_lock for write_bh.
1144 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1146 struct slave *old_active = bond->curr_active_slave;
1148 if (old_active == new_active) {
1153 new_active->jiffies = jiffies;
1155 if (new_active->link == BOND_LINK_BACK) {
1156 if (USES_PRIMARY(bond->params.mode)) {
1157 printk(KERN_INFO DRV_NAME
1158 ": %s: making interface %s the new "
1159 "active one %d ms earlier.\n",
1160 bond->dev->name, new_active->dev->name,
1161 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1164 new_active->delay = 0;
1165 new_active->link = BOND_LINK_UP;
1167 if (bond->params.mode == BOND_MODE_8023AD) {
1168 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1171 if (bond_is_lb(bond))
1172 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1174 if (USES_PRIMARY(bond->params.mode)) {
1175 printk(KERN_INFO DRV_NAME
1176 ": %s: making interface %s the new "
1178 bond->dev->name, new_active->dev->name);
1183 if (USES_PRIMARY(bond->params.mode)) {
1184 bond_mc_swap(bond, new_active, old_active);
1187 if (bond_is_lb(bond)) {
1188 bond_alb_handle_active_change(bond, new_active);
1190 bond_set_slave_inactive_flags(old_active);
1192 bond_set_slave_active_flags(new_active);
1194 bond->curr_active_slave = new_active;
1197 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1199 bond_set_slave_inactive_flags(old_active);
1203 bond_set_slave_active_flags(new_active);
1205 if (bond->params.fail_over_mac)
1206 bond_do_fail_over_mac(bond, new_active,
1209 bond->send_grat_arp = bond->params.num_grat_arp;
1210 bond_send_gratuitous_arp(bond);
1212 bond->send_unsol_na = bond->params.num_unsol_na;
1213 bond_send_unsolicited_na(bond);
1215 write_unlock_bh(&bond->curr_slave_lock);
1216 read_unlock(&bond->lock);
1218 netdev_bonding_change(bond->dev);
1220 read_lock(&bond->lock);
1221 write_lock_bh(&bond->curr_slave_lock);
1227 * bond_select_active_slave - select a new active slave, if needed
1228 * @bond: our bonding struct
1230 * This functions shoud be called when one of the following occurs:
1231 * - The old curr_active_slave has been released or lost its link.
1232 * - The primary_slave has got its link back.
1233 * - A slave has got its link back and there's no old curr_active_slave.
1235 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1237 void bond_select_active_slave(struct bonding *bond)
1239 struct slave *best_slave;
1242 best_slave = bond_find_best_slave(bond);
1243 if (best_slave != bond->curr_active_slave) {
1244 bond_change_active_slave(bond, best_slave);
1245 rv = bond_set_carrier(bond);
1249 if (netif_carrier_ok(bond->dev)) {
1250 printk(KERN_INFO DRV_NAME
1251 ": %s: first active interface up!\n",
1254 printk(KERN_INFO DRV_NAME ": %s: "
1255 "now running without any active interface !\n",
1261 /*--------------------------- slave list handling ---------------------------*/
1264 * This function attaches the slave to the end of list.
1266 * bond->lock held for writing by caller.
1268 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1270 if (bond->first_slave == NULL) { /* attaching the first slave */
1271 new_slave->next = new_slave;
1272 new_slave->prev = new_slave;
1273 bond->first_slave = new_slave;
1275 new_slave->next = bond->first_slave;
1276 new_slave->prev = bond->first_slave->prev;
1277 new_slave->next->prev = new_slave;
1278 new_slave->prev->next = new_slave;
1285 * This function detaches the slave from the list.
1286 * WARNING: no check is made to verify if the slave effectively
1287 * belongs to <bond>.
1288 * Nothing is freed on return, structures are just unchained.
1289 * If any slave pointer in bond was pointing to <slave>,
1290 * it should be changed by the calling function.
1292 * bond->lock held for writing by caller.
1294 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1297 slave->next->prev = slave->prev;
1301 slave->prev->next = slave->next;
1304 if (bond->first_slave == slave) { /* slave is the first slave */
1305 if (bond->slave_cnt > 1) { /* there are more slave */
1306 bond->first_slave = slave->next;
1308 bond->first_slave = NULL; /* slave was the last one */
1317 /*---------------------------------- IOCTL ----------------------------------*/
1319 static int bond_sethwaddr(struct net_device *bond_dev,
1320 struct net_device *slave_dev)
1322 dprintk("bond_dev=%p\n", bond_dev);
1323 dprintk("slave_dev=%p\n", slave_dev);
1324 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1325 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1329 #define BOND_VLAN_FEATURES \
1330 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1331 NETIF_F_HW_VLAN_FILTER)
1334 * Compute the common dev->feature set available to all slaves. Some
1335 * feature bits are managed elsewhere, so preserve those feature bits
1336 * on the master device.
1338 static int bond_compute_features(struct bonding *bond)
1340 struct slave *slave;
1341 struct net_device *bond_dev = bond->dev;
1342 unsigned long features = bond_dev->features;
1343 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1344 bond_dev->hard_header_len);
1347 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1348 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1350 if (!bond->first_slave)
1353 features &= ~NETIF_F_ONE_FOR_ALL;
1355 bond_for_each_slave(bond, slave, i) {
1356 features = netdev_increment_features(features,
1357 slave->dev->features,
1358 NETIF_F_ONE_FOR_ALL);
1359 if (slave->dev->hard_header_len > max_hard_header_len)
1360 max_hard_header_len = slave->dev->hard_header_len;
1364 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1365 bond_dev->features = netdev_fix_features(features, NULL);
1366 bond_dev->hard_header_len = max_hard_header_len;
1371 static void bond_setup_by_slave(struct net_device *bond_dev,
1372 struct net_device *slave_dev)
1374 struct bonding *bond = netdev_priv(bond_dev);
1376 bond_dev->header_ops = slave_dev->header_ops;
1378 bond_dev->type = slave_dev->type;
1379 bond_dev->hard_header_len = slave_dev->hard_header_len;
1380 bond_dev->addr_len = slave_dev->addr_len;
1382 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1383 slave_dev->addr_len);
1384 bond->setup_by_slave = 1;
1387 /* enslave device <slave> to bond device <master> */
1388 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1390 struct bonding *bond = netdev_priv(bond_dev);
1391 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1392 struct slave *new_slave = NULL;
1393 struct dev_mc_list *dmi;
1394 struct sockaddr addr;
1396 int old_features = bond_dev->features;
1399 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1400 slave_ops->ndo_do_ioctl == NULL) {
1401 printk(KERN_WARNING DRV_NAME
1402 ": %s: Warning: no link monitoring support for %s\n",
1403 bond_dev->name, slave_dev->name);
1406 /* bond must be initialized by bond_open() before enslaving */
1407 if (!(bond_dev->flags & IFF_UP)) {
1408 printk(KERN_WARNING DRV_NAME
1409 " %s: master_dev is not up in bond_enslave\n",
1413 /* already enslaved */
1414 if (slave_dev->flags & IFF_SLAVE) {
1415 dprintk("Error, Device was already enslaved\n");
1419 /* vlan challenged mutual exclusion */
1420 /* no need to lock since we're protected by rtnl_lock */
1421 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1422 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1423 if (!list_empty(&bond->vlan_list)) {
1424 printk(KERN_ERR DRV_NAME
1425 ": %s: Error: cannot enslave VLAN "
1426 "challenged slave %s on VLAN enabled "
1427 "bond %s\n", bond_dev->name, slave_dev->name,
1431 printk(KERN_WARNING DRV_NAME
1432 ": %s: Warning: enslaved VLAN challenged "
1433 "slave %s. Adding VLANs will be blocked as "
1434 "long as %s is part of bond %s\n",
1435 bond_dev->name, slave_dev->name, slave_dev->name,
1437 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1440 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1441 if (bond->slave_cnt == 0) {
1442 /* First slave, and it is not VLAN challenged,
1443 * so remove the block of adding VLANs over the bond.
1445 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1450 * Old ifenslave binaries are no longer supported. These can
1451 * be identified with moderate accurary by the state of the slave:
1452 * the current ifenslave will set the interface down prior to
1453 * enslaving it; the old ifenslave will not.
1455 if ((slave_dev->flags & IFF_UP)) {
1456 printk(KERN_ERR DRV_NAME ": %s is up. "
1457 "This may be due to an out of date ifenslave.\n",
1460 goto err_undo_flags;
1463 /* set bonding device ether type by slave - bonding netdevices are
1464 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1465 * there is a need to override some of the type dependent attribs/funcs.
1467 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1468 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1470 if (bond->slave_cnt == 0) {
1471 if (slave_dev->type != ARPHRD_ETHER)
1472 bond_setup_by_slave(bond_dev, slave_dev);
1473 } else if (bond_dev->type != slave_dev->type) {
1474 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1475 "from other slaves (%d), can not enslave it.\n",
1477 slave_dev->type, bond_dev->type);
1479 goto err_undo_flags;
1482 if (slave_ops->ndo_set_mac_address == NULL) {
1483 if (bond->slave_cnt == 0) {
1484 printk(KERN_WARNING DRV_NAME
1485 ": %s: Warning: The first slave device "
1486 "specified does not support setting the MAC "
1487 "address. Setting fail_over_mac to active.",
1489 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1490 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1491 printk(KERN_ERR DRV_NAME
1492 ": %s: Error: The slave device specified "
1493 "does not support setting the MAC address, "
1494 "but fail_over_mac is not set to active.\n"
1497 goto err_undo_flags;
1501 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1504 goto err_undo_flags;
1507 /* save slave's original flags before calling
1508 * netdev_set_master and dev_open
1510 new_slave->original_flags = slave_dev->flags;
1513 * Save slave's original ("permanent") mac address for modes
1514 * that need it, and for restoring it upon release, and then
1515 * set it to the master's address
1517 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1519 if (!bond->params.fail_over_mac) {
1521 * Set slave to master's mac address. The application already
1522 * set the master's mac address to that of the first slave
1524 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1525 addr.sa_family = slave_dev->type;
1526 res = dev_set_mac_address(slave_dev, &addr);
1528 dprintk("Error %d calling set_mac_address\n", res);
1533 res = netdev_set_master(slave_dev, bond_dev);
1535 dprintk("Error %d calling netdev_set_master\n", res);
1536 goto err_restore_mac;
1538 /* open the slave since the application closed it */
1539 res = dev_open(slave_dev);
1541 dprintk("Openning slave %s failed\n", slave_dev->name);
1542 goto err_unset_master;
1545 new_slave->dev = slave_dev;
1546 slave_dev->priv_flags |= IFF_BONDING;
1548 if (bond_is_lb(bond)) {
1549 /* bond_alb_init_slave() must be called before all other stages since
1550 * it might fail and we do not want to have to undo everything
1552 res = bond_alb_init_slave(bond, new_slave);
1558 /* If the mode USES_PRIMARY, then the new slave gets the
1559 * master's promisc (and mc) settings only if it becomes the
1560 * curr_active_slave, and that is taken care of later when calling
1561 * bond_change_active()
1563 if (!USES_PRIMARY(bond->params.mode)) {
1564 /* set promiscuity level to new slave */
1565 if (bond_dev->flags & IFF_PROMISC) {
1566 res = dev_set_promiscuity(slave_dev, 1);
1571 /* set allmulti level to new slave */
1572 if (bond_dev->flags & IFF_ALLMULTI) {
1573 res = dev_set_allmulti(slave_dev, 1);
1578 netif_addr_lock_bh(bond_dev);
1579 /* upload master's mc_list to new slave */
1580 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1581 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1583 netif_addr_unlock_bh(bond_dev);
1586 if (bond->params.mode == BOND_MODE_8023AD) {
1587 /* add lacpdu mc addr to mc list */
1588 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1590 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1593 bond_add_vlans_on_slave(bond, slave_dev);
1595 write_lock_bh(&bond->lock);
1597 bond_attach_slave(bond, new_slave);
1599 new_slave->delay = 0;
1600 new_slave->link_failure_count = 0;
1602 bond_compute_features(bond);
1604 write_unlock_bh(&bond->lock);
1606 read_lock(&bond->lock);
1608 new_slave->last_arp_rx = jiffies;
1610 if (bond->params.miimon && !bond->params.use_carrier) {
1611 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1613 if ((link_reporting == -1) && !bond->params.arp_interval) {
1615 * miimon is set but a bonded network driver
1616 * does not support ETHTOOL/MII and
1617 * arp_interval is not set. Note: if
1618 * use_carrier is enabled, we will never go
1619 * here (because netif_carrier is always
1620 * supported); thus, we don't need to change
1621 * the messages for netif_carrier.
1623 printk(KERN_WARNING DRV_NAME
1624 ": %s: Warning: MII and ETHTOOL support not "
1625 "available for interface %s, and "
1626 "arp_interval/arp_ip_target module parameters "
1627 "not specified, thus bonding will not detect "
1628 "link failures! see bonding.txt for details.\n",
1629 bond_dev->name, slave_dev->name);
1630 } else if (link_reporting == -1) {
1631 /* unable get link status using mii/ethtool */
1632 printk(KERN_WARNING DRV_NAME
1633 ": %s: Warning: can't get link status from "
1634 "interface %s; the network driver associated "
1635 "with this interface does not support MII or "
1636 "ETHTOOL link status reporting, thus miimon "
1637 "has no effect on this interface.\n",
1638 bond_dev->name, slave_dev->name);
1642 /* check for initial state */
1643 if (!bond->params.miimon ||
1644 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1645 if (bond->params.updelay) {
1646 dprintk("Initial state of slave_dev is "
1647 "BOND_LINK_BACK\n");
1648 new_slave->link = BOND_LINK_BACK;
1649 new_slave->delay = bond->params.updelay;
1651 dprintk("Initial state of slave_dev is "
1653 new_slave->link = BOND_LINK_UP;
1655 new_slave->jiffies = jiffies;
1657 dprintk("Initial state of slave_dev is "
1658 "BOND_LINK_DOWN\n");
1659 new_slave->link = BOND_LINK_DOWN;
1662 if (bond_update_speed_duplex(new_slave) &&
1663 (new_slave->link != BOND_LINK_DOWN)) {
1664 printk(KERN_WARNING DRV_NAME
1665 ": %s: Warning: failed to get speed and duplex from %s, "
1666 "assumed to be 100Mb/sec and Full.\n",
1667 bond_dev->name, new_slave->dev->name);
1669 if (bond->params.mode == BOND_MODE_8023AD) {
1670 printk(KERN_WARNING DRV_NAME
1671 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1672 "support in base driver for proper aggregator "
1673 "selection.\n", bond_dev->name);
1677 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1678 /* if there is a primary slave, remember it */
1679 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1680 bond->primary_slave = new_slave;
1684 write_lock_bh(&bond->curr_slave_lock);
1686 switch (bond->params.mode) {
1687 case BOND_MODE_ACTIVEBACKUP:
1688 bond_set_slave_inactive_flags(new_slave);
1689 bond_select_active_slave(bond);
1691 case BOND_MODE_8023AD:
1692 /* in 802.3ad mode, the internal mechanism
1693 * will activate the slaves in the selected
1696 bond_set_slave_inactive_flags(new_slave);
1697 /* if this is the first slave */
1698 if (bond->slave_cnt == 1) {
1699 SLAVE_AD_INFO(new_slave).id = 1;
1700 /* Initialize AD with the number of times that the AD timer is called in 1 second
1701 * can be called only after the mac address of the bond is set
1703 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1704 bond->params.lacp_fast);
1706 SLAVE_AD_INFO(new_slave).id =
1707 SLAVE_AD_INFO(new_slave->prev).id + 1;
1710 bond_3ad_bind_slave(new_slave);
1714 new_slave->state = BOND_STATE_ACTIVE;
1715 bond_set_slave_inactive_flags(new_slave);
1718 dprintk("This slave is always active in trunk mode\n");
1720 /* always active in trunk mode */
1721 new_slave->state = BOND_STATE_ACTIVE;
1723 /* In trunking mode there is little meaning to curr_active_slave
1724 * anyway (it holds no special properties of the bond device),
1725 * so we can change it without calling change_active_interface()
1727 if (!bond->curr_active_slave) {
1728 bond->curr_active_slave = new_slave;
1731 } /* switch(bond_mode) */
1733 write_unlock_bh(&bond->curr_slave_lock);
1735 bond_set_carrier(bond);
1737 read_unlock(&bond->lock);
1739 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1743 printk(KERN_INFO DRV_NAME
1744 ": %s: enslaving %s as a%s interface with a%s link.\n",
1745 bond_dev->name, slave_dev->name,
1746 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1747 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1749 /* enslave is successful */
1752 /* Undo stages on error */
1754 dev_close(slave_dev);
1757 netdev_set_master(slave_dev, NULL);
1760 if (!bond->params.fail_over_mac) {
1761 /* XXX TODO - fom follow mode needs to change master's
1762 * MAC if this slave's MAC is in use by the bond, or at
1763 * least print a warning.
1765 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1766 addr.sa_family = slave_dev->type;
1767 dev_set_mac_address(slave_dev, &addr);
1774 bond_dev->features = old_features;
1780 * Try to release the slave device <slave> from the bond device <master>
1781 * It is legal to access curr_active_slave without a lock because all the function
1784 * The rules for slave state should be:
1785 * for Active/Backup:
1786 * Active stays on all backups go down
1787 * for Bonded connections:
1788 * The first up interface should be left on and all others downed.
1790 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1792 struct bonding *bond = netdev_priv(bond_dev);
1793 struct slave *slave, *oldcurrent;
1794 struct sockaddr addr;
1795 int mac_addr_differ;
1797 /* slave is not a slave or master is not master of this slave */
1798 if (!(slave_dev->flags & IFF_SLAVE) ||
1799 (slave_dev->master != bond_dev)) {
1800 printk(KERN_ERR DRV_NAME
1801 ": %s: Error: cannot release %s.\n",
1802 bond_dev->name, slave_dev->name);
1806 write_lock_bh(&bond->lock);
1808 slave = bond_get_slave_by_dev(bond, slave_dev);
1810 /* not a slave of this bond */
1811 printk(KERN_INFO DRV_NAME
1812 ": %s: %s not enslaved\n",
1813 bond_dev->name, slave_dev->name);
1814 write_unlock_bh(&bond->lock);
1818 if (!bond->params.fail_over_mac) {
1819 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1821 if (!mac_addr_differ && (bond->slave_cnt > 1))
1822 printk(KERN_WARNING DRV_NAME
1823 ": %s: Warning: the permanent HWaddr of %s - "
1824 "%pM - is still in use by %s. "
1825 "Set the HWaddr of %s to a different address "
1826 "to avoid conflicts.\n",
1827 bond_dev->name, slave_dev->name,
1829 bond_dev->name, slave_dev->name);
1832 /* Inform AD package of unbinding of slave. */
1833 if (bond->params.mode == BOND_MODE_8023AD) {
1834 /* must be called before the slave is
1835 * detached from the list
1837 bond_3ad_unbind_slave(slave);
1840 printk(KERN_INFO DRV_NAME
1841 ": %s: releasing %s interface %s\n",
1843 (slave->state == BOND_STATE_ACTIVE)
1844 ? "active" : "backup",
1847 oldcurrent = bond->curr_active_slave;
1849 bond->current_arp_slave = NULL;
1851 /* release the slave from its bond */
1852 bond_detach_slave(bond, slave);
1854 bond_compute_features(bond);
1856 if (bond->primary_slave == slave) {
1857 bond->primary_slave = NULL;
1860 if (oldcurrent == slave) {
1861 bond_change_active_slave(bond, NULL);
1864 if (bond_is_lb(bond)) {
1865 /* Must be called only after the slave has been
1866 * detached from the list and the curr_active_slave
1867 * has been cleared (if our_slave == old_current),
1868 * but before a new active slave is selected.
1870 write_unlock_bh(&bond->lock);
1871 bond_alb_deinit_slave(bond, slave);
1872 write_lock_bh(&bond->lock);
1875 if (oldcurrent == slave) {
1877 * Note that we hold RTNL over this sequence, so there
1878 * is no concern that another slave add/remove event
1881 write_unlock_bh(&bond->lock);
1882 read_lock(&bond->lock);
1883 write_lock_bh(&bond->curr_slave_lock);
1885 bond_select_active_slave(bond);
1887 write_unlock_bh(&bond->curr_slave_lock);
1888 read_unlock(&bond->lock);
1889 write_lock_bh(&bond->lock);
1892 if (bond->slave_cnt == 0) {
1893 bond_set_carrier(bond);
1895 /* if the last slave was removed, zero the mac address
1896 * of the master so it will be set by the application
1897 * to the mac address of the first slave
1899 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1901 if (list_empty(&bond->vlan_list)) {
1902 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1904 printk(KERN_WARNING DRV_NAME
1905 ": %s: Warning: clearing HW address of %s while it "
1906 "still has VLANs.\n",
1907 bond_dev->name, bond_dev->name);
1908 printk(KERN_WARNING DRV_NAME
1909 ": %s: When re-adding slaves, make sure the bond's "
1910 "HW address matches its VLANs'.\n",
1913 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1914 !bond_has_challenged_slaves(bond)) {
1915 printk(KERN_INFO DRV_NAME
1916 ": %s: last VLAN challenged slave %s "
1917 "left bond %s. VLAN blocking is removed\n",
1918 bond_dev->name, slave_dev->name, bond_dev->name);
1919 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1922 write_unlock_bh(&bond->lock);
1924 /* must do this from outside any spinlocks */
1925 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1927 bond_del_vlans_from_slave(bond, slave_dev);
1929 /* If the mode USES_PRIMARY, then we should only remove its
1930 * promisc and mc settings if it was the curr_active_slave, but that was
1931 * already taken care of above when we detached the slave
1933 if (!USES_PRIMARY(bond->params.mode)) {
1934 /* unset promiscuity level from slave */
1935 if (bond_dev->flags & IFF_PROMISC) {
1936 dev_set_promiscuity(slave_dev, -1);
1939 /* unset allmulti level from slave */
1940 if (bond_dev->flags & IFF_ALLMULTI) {
1941 dev_set_allmulti(slave_dev, -1);
1944 /* flush master's mc_list from slave */
1945 netif_addr_lock_bh(bond_dev);
1946 bond_mc_list_flush(bond_dev, slave_dev);
1947 netif_addr_unlock_bh(bond_dev);
1950 netdev_set_master(slave_dev, NULL);
1952 /* close slave before restoring its mac address */
1953 dev_close(slave_dev);
1955 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1956 /* restore original ("permanent") mac address */
1957 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1958 addr.sa_family = slave_dev->type;
1959 dev_set_mac_address(slave_dev, &addr);
1962 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1963 IFF_SLAVE_INACTIVE | IFF_BONDING |
1968 return 0; /* deletion OK */
1972 * Destroy a bonding device.
1973 * Must be under rtnl_lock when this function is called.
1975 void bond_destroy(struct bonding *bond)
1977 bond_deinit(bond->dev);
1978 bond_destroy_sysfs_entry(bond);
1979 unregister_netdevice(bond->dev);
1982 static void bond_destructor(struct net_device *bond_dev)
1984 struct bonding *bond = netdev_priv(bond_dev);
1987 destroy_workqueue(bond->wq);
1989 netif_addr_lock_bh(bond_dev);
1990 bond_mc_list_destroy(bond);
1991 netif_addr_unlock_bh(bond_dev);
1993 free_netdev(bond_dev);
1997 * First release a slave and than destroy the bond if no more slaves iare left.
1998 * Must be under rtnl_lock when this function is called.
2000 int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2002 struct bonding *bond = netdev_priv(bond_dev);
2005 ret = bond_release(bond_dev, slave_dev);
2006 if ((ret == 0) && (bond->slave_cnt == 0)) {
2007 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2008 bond_dev->name, bond_dev->name);
2015 * This function releases all slaves.
2017 static int bond_release_all(struct net_device *bond_dev)
2019 struct bonding *bond = netdev_priv(bond_dev);
2020 struct slave *slave;
2021 struct net_device *slave_dev;
2022 struct sockaddr addr;
2024 write_lock_bh(&bond->lock);
2026 netif_carrier_off(bond_dev);
2028 if (bond->slave_cnt == 0) {
2032 bond->current_arp_slave = NULL;
2033 bond->primary_slave = NULL;
2034 bond_change_active_slave(bond, NULL);
2036 while ((slave = bond->first_slave) != NULL) {
2037 /* Inform AD package of unbinding of slave
2038 * before slave is detached from the list.
2040 if (bond->params.mode == BOND_MODE_8023AD) {
2041 bond_3ad_unbind_slave(slave);
2044 slave_dev = slave->dev;
2045 bond_detach_slave(bond, slave);
2047 /* now that the slave is detached, unlock and perform
2048 * all the undo steps that should not be called from
2051 write_unlock_bh(&bond->lock);
2053 if (bond_is_lb(bond)) {
2054 /* must be called only after the slave
2055 * has been detached from the list
2057 bond_alb_deinit_slave(bond, slave);
2060 bond_compute_features(bond);
2062 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2063 bond_del_vlans_from_slave(bond, slave_dev);
2065 /* If the mode USES_PRIMARY, then we should only remove its
2066 * promisc and mc settings if it was the curr_active_slave, but that was
2067 * already taken care of above when we detached the slave
2069 if (!USES_PRIMARY(bond->params.mode)) {
2070 /* unset promiscuity level from slave */
2071 if (bond_dev->flags & IFF_PROMISC) {
2072 dev_set_promiscuity(slave_dev, -1);
2075 /* unset allmulti level from slave */
2076 if (bond_dev->flags & IFF_ALLMULTI) {
2077 dev_set_allmulti(slave_dev, -1);
2080 /* flush master's mc_list from slave */
2081 netif_addr_lock_bh(bond_dev);
2082 bond_mc_list_flush(bond_dev, slave_dev);
2083 netif_addr_unlock_bh(bond_dev);
2086 netdev_set_master(slave_dev, NULL);
2088 /* close slave before restoring its mac address */
2089 dev_close(slave_dev);
2091 if (!bond->params.fail_over_mac) {
2092 /* restore original ("permanent") mac address*/
2093 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2094 addr.sa_family = slave_dev->type;
2095 dev_set_mac_address(slave_dev, &addr);
2098 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2099 IFF_SLAVE_INACTIVE);
2103 /* re-acquire the lock before getting the next slave */
2104 write_lock_bh(&bond->lock);
2107 /* zero the mac address of the master so it will be
2108 * set by the application to the mac address of the
2111 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2113 if (list_empty(&bond->vlan_list)) {
2114 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2116 printk(KERN_WARNING DRV_NAME
2117 ": %s: Warning: clearing HW address of %s while it "
2118 "still has VLANs.\n",
2119 bond_dev->name, bond_dev->name);
2120 printk(KERN_WARNING DRV_NAME
2121 ": %s: When re-adding slaves, make sure the bond's "
2122 "HW address matches its VLANs'.\n",
2126 printk(KERN_INFO DRV_NAME
2127 ": %s: released all slaves\n",
2131 write_unlock_bh(&bond->lock);
2137 * This function changes the active slave to slave <slave_dev>.
2138 * It returns -EINVAL in the following cases.
2139 * - <slave_dev> is not found in the list.
2140 * - There is not active slave now.
2141 * - <slave_dev> is already active.
2142 * - The link state of <slave_dev> is not BOND_LINK_UP.
2143 * - <slave_dev> is not running.
2144 * In these cases, this fuction does nothing.
2145 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2147 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2149 struct bonding *bond = netdev_priv(bond_dev);
2150 struct slave *old_active = NULL;
2151 struct slave *new_active = NULL;
2154 if (!USES_PRIMARY(bond->params.mode)) {
2158 /* Verify that master_dev is indeed the master of slave_dev */
2159 if (!(slave_dev->flags & IFF_SLAVE) ||
2160 (slave_dev->master != bond_dev)) {
2164 read_lock(&bond->lock);
2166 read_lock(&bond->curr_slave_lock);
2167 old_active = bond->curr_active_slave;
2168 read_unlock(&bond->curr_slave_lock);
2170 new_active = bond_get_slave_by_dev(bond, slave_dev);
2173 * Changing to the current active: do nothing; return success.
2175 if (new_active && (new_active == old_active)) {
2176 read_unlock(&bond->lock);
2182 (new_active->link == BOND_LINK_UP) &&
2183 IS_UP(new_active->dev)) {
2184 write_lock_bh(&bond->curr_slave_lock);
2185 bond_change_active_slave(bond, new_active);
2186 write_unlock_bh(&bond->curr_slave_lock);
2191 read_unlock(&bond->lock);
2196 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2198 struct bonding *bond = netdev_priv(bond_dev);
2200 info->bond_mode = bond->params.mode;
2201 info->miimon = bond->params.miimon;
2203 read_lock(&bond->lock);
2204 info->num_slaves = bond->slave_cnt;
2205 read_unlock(&bond->lock);
2210 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2212 struct bonding *bond = netdev_priv(bond_dev);
2213 struct slave *slave;
2216 if (info->slave_id < 0) {
2220 read_lock(&bond->lock);
2222 bond_for_each_slave(bond, slave, i) {
2223 if (i == (int)info->slave_id) {
2229 read_unlock(&bond->lock);
2232 strcpy(info->slave_name, slave->dev->name);
2233 info->link = slave->link;
2234 info->state = slave->state;
2235 info->link_failure_count = slave->link_failure_count;
2243 /*-------------------------------- Monitoring -------------------------------*/
2246 static int bond_miimon_inspect(struct bonding *bond)
2248 struct slave *slave;
2249 int i, link_state, commit = 0;
2251 bond_for_each_slave(bond, slave, i) {
2252 slave->new_link = BOND_LINK_NOCHANGE;
2254 link_state = bond_check_dev_link(bond, slave->dev, 0);
2256 switch (slave->link) {
2261 slave->link = BOND_LINK_FAIL;
2262 slave->delay = bond->params.downdelay;
2264 printk(KERN_INFO DRV_NAME
2265 ": %s: link status down for %s"
2266 "interface %s, disabling it in %d ms.\n",
2268 (bond->params.mode ==
2269 BOND_MODE_ACTIVEBACKUP) ?
2270 ((slave->state == BOND_STATE_ACTIVE) ?
2271 "active " : "backup ") : "",
2273 bond->params.downdelay * bond->params.miimon);
2276 case BOND_LINK_FAIL:
2279 * recovered before downdelay expired
2281 slave->link = BOND_LINK_UP;
2282 slave->jiffies = jiffies;
2283 printk(KERN_INFO DRV_NAME
2284 ": %s: link status up again after %d "
2285 "ms for interface %s.\n",
2287 (bond->params.downdelay - slave->delay) *
2288 bond->params.miimon,
2293 if (slave->delay <= 0) {
2294 slave->new_link = BOND_LINK_DOWN;
2302 case BOND_LINK_DOWN:
2306 slave->link = BOND_LINK_BACK;
2307 slave->delay = bond->params.updelay;
2310 printk(KERN_INFO DRV_NAME
2311 ": %s: link status up for "
2312 "interface %s, enabling it in %d ms.\n",
2313 bond->dev->name, slave->dev->name,
2314 bond->params.updelay *
2315 bond->params.miimon);
2318 case BOND_LINK_BACK:
2320 slave->link = BOND_LINK_DOWN;
2321 printk(KERN_INFO DRV_NAME
2322 ": %s: link status down again after %d "
2323 "ms for interface %s.\n",
2325 (bond->params.updelay - slave->delay) *
2326 bond->params.miimon,
2332 if (slave->delay <= 0) {
2333 slave->new_link = BOND_LINK_UP;
2346 static void bond_miimon_commit(struct bonding *bond)
2348 struct slave *slave;
2351 bond_for_each_slave(bond, slave, i) {
2352 switch (slave->new_link) {
2353 case BOND_LINK_NOCHANGE:
2357 slave->link = BOND_LINK_UP;
2358 slave->jiffies = jiffies;
2360 if (bond->params.mode == BOND_MODE_8023AD) {
2361 /* prevent it from being the active one */
2362 slave->state = BOND_STATE_BACKUP;
2363 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2364 /* make it immediately active */
2365 slave->state = BOND_STATE_ACTIVE;
2366 } else if (slave != bond->primary_slave) {
2367 /* prevent it from being the active one */
2368 slave->state = BOND_STATE_BACKUP;
2371 printk(KERN_INFO DRV_NAME
2372 ": %s: link status definitely "
2373 "up for interface %s.\n",
2374 bond->dev->name, slave->dev->name);
2376 /* notify ad that the link status has changed */
2377 if (bond->params.mode == BOND_MODE_8023AD)
2378 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2380 if (bond_is_lb(bond))
2381 bond_alb_handle_link_change(bond, slave,
2384 if (!bond->curr_active_slave ||
2385 (slave == bond->primary_slave))
2390 case BOND_LINK_DOWN:
2391 if (slave->link_failure_count < UINT_MAX)
2392 slave->link_failure_count++;
2394 slave->link = BOND_LINK_DOWN;
2396 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2397 bond->params.mode == BOND_MODE_8023AD)
2398 bond_set_slave_inactive_flags(slave);
2400 printk(KERN_INFO DRV_NAME
2401 ": %s: link status definitely down for "
2402 "interface %s, disabling it\n",
2403 bond->dev->name, slave->dev->name);
2405 if (bond->params.mode == BOND_MODE_8023AD)
2406 bond_3ad_handle_link_change(slave,
2409 if (bond->params.mode == BOND_MODE_TLB ||
2410 bond->params.mode == BOND_MODE_ALB)
2411 bond_alb_handle_link_change(bond, slave,
2414 if (slave == bond->curr_active_slave)
2420 printk(KERN_ERR DRV_NAME
2421 ": %s: invalid new link %d on slave %s\n",
2422 bond->dev->name, slave->new_link,
2424 slave->new_link = BOND_LINK_NOCHANGE;
2431 write_lock_bh(&bond->curr_slave_lock);
2432 bond_select_active_slave(bond);
2433 write_unlock_bh(&bond->curr_slave_lock);
2436 bond_set_carrier(bond);
2442 * Really a wrapper that splits the mii monitor into two phases: an
2443 * inspection, then (if inspection indicates something needs to be done)
2444 * an acquisition of appropriate locks followed by a commit phase to
2445 * implement whatever link state changes are indicated.
2447 void bond_mii_monitor(struct work_struct *work)
2449 struct bonding *bond = container_of(work, struct bonding,
2452 read_lock(&bond->lock);
2453 if (bond->kill_timers)
2456 if (bond->slave_cnt == 0)
2459 if (bond->send_grat_arp) {
2460 read_lock(&bond->curr_slave_lock);
2461 bond_send_gratuitous_arp(bond);
2462 read_unlock(&bond->curr_slave_lock);
2465 if (bond->send_unsol_na) {
2466 read_lock(&bond->curr_slave_lock);
2467 bond_send_unsolicited_na(bond);
2468 read_unlock(&bond->curr_slave_lock);
2471 if (bond_miimon_inspect(bond)) {
2472 read_unlock(&bond->lock);
2474 read_lock(&bond->lock);
2476 bond_miimon_commit(bond);
2478 read_unlock(&bond->lock);
2479 rtnl_unlock(); /* might sleep, hold no other locks */
2480 read_lock(&bond->lock);
2484 if (bond->params.miimon)
2485 queue_delayed_work(bond->wq, &bond->mii_work,
2486 msecs_to_jiffies(bond->params.miimon));
2488 read_unlock(&bond->lock);
2491 static __be32 bond_glean_dev_ip(struct net_device *dev)
2493 struct in_device *idev;
2494 struct in_ifaddr *ifa;
2501 idev = __in_dev_get_rcu(dev);
2505 ifa = idev->ifa_list;
2509 addr = ifa->ifa_local;
2515 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2517 struct vlan_entry *vlan;
2519 if (ip == bond->master_ip)
2522 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2523 if (ip == vlan->vlan_ip)
2531 * We go to the (large) trouble of VLAN tagging ARP frames because
2532 * switches in VLAN mode (especially if ports are configured as
2533 * "native" to a VLAN) might not pass non-tagged frames.
2535 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2537 struct sk_buff *skb;
2539 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2540 slave_dev->name, dest_ip, src_ip, vlan_id);
2542 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2543 NULL, slave_dev->dev_addr, NULL);
2546 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2550 skb = vlan_put_tag(skb, vlan_id);
2552 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2560 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2563 __be32 *targets = bond->params.arp_targets;
2564 struct vlan_entry *vlan;
2565 struct net_device *vlan_dev;
2569 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2572 dprintk("basa: target %x\n", targets[i]);
2573 if (list_empty(&bond->vlan_list)) {
2574 dprintk("basa: empty vlan: arp_send\n");
2575 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2576 bond->master_ip, 0);
2581 * If VLANs are configured, we do a route lookup to
2582 * determine which VLAN interface would be used, so we
2583 * can tag the ARP with the proper VLAN tag.
2585 memset(&fl, 0, sizeof(fl));
2586 fl.fl4_dst = targets[i];
2587 fl.fl4_tos = RTO_ONLINK;
2589 rv = ip_route_output_key(&init_net, &rt, &fl);
2591 if (net_ratelimit()) {
2592 printk(KERN_WARNING DRV_NAME
2593 ": %s: no route to arp_ip_target %pI4\n",
2594 bond->dev->name, &fl.fl4_dst);
2600 * This target is not on a VLAN
2602 if (rt->u.dst.dev == bond->dev) {
2604 dprintk("basa: rtdev == bond->dev: arp_send\n");
2605 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2606 bond->master_ip, 0);
2611 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2612 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2613 if (vlan_dev == rt->u.dst.dev) {
2614 vlan_id = vlan->vlan_id;
2615 dprintk("basa: vlan match on %s %d\n",
2616 vlan_dev->name, vlan_id);
2623 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2624 vlan->vlan_ip, vlan_id);
2628 if (net_ratelimit()) {
2629 printk(KERN_WARNING DRV_NAME
2630 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2631 bond->dev->name, &fl.fl4_dst,
2632 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2639 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2640 * for each VLAN above us.
2642 * Caller must hold curr_slave_lock for read or better
2644 static void bond_send_gratuitous_arp(struct bonding *bond)
2646 struct slave *slave = bond->curr_active_slave;
2647 struct vlan_entry *vlan;
2648 struct net_device *vlan_dev;
2650 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2651 slave ? slave->dev->name : "NULL");
2653 if (!slave || !bond->send_grat_arp ||
2654 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2657 bond->send_grat_arp--;
2659 if (bond->master_ip) {
2660 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2661 bond->master_ip, 0);
2664 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2665 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2666 if (vlan->vlan_ip) {
2667 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2668 vlan->vlan_ip, vlan->vlan_id);
2673 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2676 __be32 *targets = bond->params.arp_targets;
2678 targets = bond->params.arp_targets;
2679 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2680 dprintk("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2681 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2682 if (sip == targets[i]) {
2683 if (bond_has_this_ip(bond, tip))
2684 slave->last_arp_rx = jiffies;
2690 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2693 struct slave *slave;
2694 struct bonding *bond;
2695 unsigned char *arp_ptr;
2698 if (dev_net(dev) != &init_net)
2701 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2704 bond = netdev_priv(dev);
2705 read_lock(&bond->lock);
2707 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2708 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2709 orig_dev ? orig_dev->name : "NULL");
2711 slave = bond_get_slave_by_dev(bond, orig_dev);
2712 if (!slave || !slave_do_arp_validate(bond, slave))
2715 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2719 if (arp->ar_hln != dev->addr_len ||
2720 skb->pkt_type == PACKET_OTHERHOST ||
2721 skb->pkt_type == PACKET_LOOPBACK ||
2722 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2723 arp->ar_pro != htons(ETH_P_IP) ||
2727 arp_ptr = (unsigned char *)(arp + 1);
2728 arp_ptr += dev->addr_len;
2729 memcpy(&sip, arp_ptr, 4);
2730 arp_ptr += 4 + dev->addr_len;
2731 memcpy(&tip, arp_ptr, 4);
2733 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2734 bond->dev->name, slave->dev->name, slave->state,
2735 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2739 * Backup slaves won't see the ARP reply, but do come through
2740 * here for each ARP probe (so we swap the sip/tip to validate
2741 * the probe). In a "redundant switch, common router" type of
2742 * configuration, the ARP probe will (hopefully) travel from
2743 * the active, through one switch, the router, then the other
2744 * switch before reaching the backup.
2746 if (slave->state == BOND_STATE_ACTIVE)
2747 bond_validate_arp(bond, slave, sip, tip);
2749 bond_validate_arp(bond, slave, tip, sip);
2752 read_unlock(&bond->lock);
2755 return NET_RX_SUCCESS;
2759 * this function is called regularly to monitor each slave's link
2760 * ensuring that traffic is being sent and received when arp monitoring
2761 * is used in load-balancing mode. if the adapter has been dormant, then an
2762 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2763 * arp monitoring in active backup mode.
2765 void bond_loadbalance_arp_mon(struct work_struct *work)
2767 struct bonding *bond = container_of(work, struct bonding,
2769 struct slave *slave, *oldcurrent;
2770 int do_failover = 0;
2774 read_lock(&bond->lock);
2776 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2778 if (bond->kill_timers) {
2782 if (bond->slave_cnt == 0) {
2786 read_lock(&bond->curr_slave_lock);
2787 oldcurrent = bond->curr_active_slave;
2788 read_unlock(&bond->curr_slave_lock);
2790 /* see if any of the previous devices are up now (i.e. they have
2791 * xmt and rcv traffic). the curr_active_slave does not come into
2792 * the picture unless it is null. also, slave->jiffies is not needed
2793 * here because we send an arp on each slave and give a slave as
2794 * long as it needs to get the tx/rx within the delta.
2795 * TODO: what about up/down delay in arp mode? it wasn't here before
2798 bond_for_each_slave(bond, slave, i) {
2799 if (slave->link != BOND_LINK_UP) {
2800 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2801 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2803 slave->link = BOND_LINK_UP;
2804 slave->state = BOND_STATE_ACTIVE;
2806 /* primary_slave has no meaning in round-robin
2807 * mode. the window of a slave being up and
2808 * curr_active_slave being null after enslaving
2812 printk(KERN_INFO DRV_NAME
2813 ": %s: link status definitely "
2814 "up for interface %s, ",
2819 printk(KERN_INFO DRV_NAME
2820 ": %s: interface %s is now up\n",
2826 /* slave->link == BOND_LINK_UP */
2828 /* not all switches will respond to an arp request
2829 * when the source ip is 0, so don't take the link down
2830 * if we don't know our ip yet
2832 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2833 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2835 slave->link = BOND_LINK_DOWN;
2836 slave->state = BOND_STATE_BACKUP;
2838 if (slave->link_failure_count < UINT_MAX) {
2839 slave->link_failure_count++;
2842 printk(KERN_INFO DRV_NAME
2843 ": %s: interface %s is now down.\n",
2847 if (slave == oldcurrent) {
2853 /* note: if switch is in round-robin mode, all links
2854 * must tx arp to ensure all links rx an arp - otherwise
2855 * links may oscillate or not come up at all; if switch is
2856 * in something like xor mode, there is nothing we can
2857 * do - all replies will be rx'ed on same link causing slaves
2858 * to be unstable during low/no traffic periods
2860 if (IS_UP(slave->dev)) {
2861 bond_arp_send_all(bond, slave);
2866 write_lock_bh(&bond->curr_slave_lock);
2868 bond_select_active_slave(bond);
2870 write_unlock_bh(&bond->curr_slave_lock);
2874 if (bond->params.arp_interval)
2875 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2877 read_unlock(&bond->lock);
2881 * Called to inspect slaves for active-backup mode ARP monitor link state
2882 * changes. Sets new_link in slaves to specify what action should take
2883 * place for the slave. Returns 0 if no changes are found, >0 if changes
2884 * to link states must be committed.
2886 * Called with bond->lock held for read.
2888 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2890 struct slave *slave;
2893 bond_for_each_slave(bond, slave, i) {
2894 slave->new_link = BOND_LINK_NOCHANGE;
2896 if (slave->link != BOND_LINK_UP) {
2897 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2899 slave->new_link = BOND_LINK_UP;
2907 * Give slaves 2*delta after being enslaved or made
2908 * active. This avoids bouncing, as the last receive
2909 * times need a full ARP monitor cycle to be updated.
2911 if (!time_after_eq(jiffies, slave->jiffies +
2912 2 * delta_in_ticks))
2916 * Backup slave is down if:
2917 * - No current_arp_slave AND
2918 * - more than 3*delta since last receive AND
2919 * - the bond has an IP address
2921 * Note: a non-null current_arp_slave indicates
2922 * the curr_active_slave went down and we are
2923 * searching for a new one; under this condition
2924 * we only take the curr_active_slave down - this
2925 * gives each slave a chance to tx/rx traffic
2926 * before being taken out
2928 if (slave->state == BOND_STATE_BACKUP &&
2929 !bond->current_arp_slave &&
2930 time_after(jiffies, slave_last_rx(bond, slave) +
2931 3 * delta_in_ticks)) {
2932 slave->new_link = BOND_LINK_DOWN;
2937 * Active slave is down if:
2938 * - more than 2*delta since transmitting OR
2939 * - (more than 2*delta since receive AND
2940 * the bond has an IP address)
2942 if ((slave->state == BOND_STATE_ACTIVE) &&
2943 (time_after_eq(jiffies, slave->dev->trans_start +
2944 2 * delta_in_ticks) ||
2945 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2946 + 2 * delta_in_ticks)))) {
2947 slave->new_link = BOND_LINK_DOWN;
2952 read_lock(&bond->curr_slave_lock);
2955 * Trigger a commit if the primary option setting has changed.
2957 if (bond->primary_slave &&
2958 (bond->primary_slave != bond->curr_active_slave) &&
2959 (bond->primary_slave->link == BOND_LINK_UP))
2962 read_unlock(&bond->curr_slave_lock);
2968 * Called to commit link state changes noted by inspection step of
2969 * active-backup mode ARP monitor.
2971 * Called with RTNL and bond->lock for read.
2973 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2975 struct slave *slave;
2978 bond_for_each_slave(bond, slave, i) {
2979 switch (slave->new_link) {
2980 case BOND_LINK_NOCHANGE:
2984 write_lock_bh(&bond->curr_slave_lock);
2986 if (!bond->curr_active_slave &&
2987 time_before_eq(jiffies, slave->dev->trans_start +
2989 slave->link = BOND_LINK_UP;
2990 bond_change_active_slave(bond, slave);
2991 bond->current_arp_slave = NULL;
2993 printk(KERN_INFO DRV_NAME
2994 ": %s: %s is up and now the "
2995 "active interface\n",
2996 bond->dev->name, slave->dev->name);
2998 } else if (bond->curr_active_slave != slave) {
2999 /* this slave has just come up but we
3000 * already have a current slave; this can
3001 * also happen if bond_enslave adds a new
3002 * slave that is up while we are searching
3005 slave->link = BOND_LINK_UP;
3006 bond_set_slave_inactive_flags(slave);
3007 bond->current_arp_slave = NULL;
3009 printk(KERN_INFO DRV_NAME
3010 ": %s: backup interface %s is now up\n",
3011 bond->dev->name, slave->dev->name);
3014 write_unlock_bh(&bond->curr_slave_lock);
3018 case BOND_LINK_DOWN:
3019 if (slave->link_failure_count < UINT_MAX)
3020 slave->link_failure_count++;
3022 slave->link = BOND_LINK_DOWN;
3024 if (slave == bond->curr_active_slave) {
3025 printk(KERN_INFO DRV_NAME
3026 ": %s: link status down for active "
3027 "interface %s, disabling it\n",
3028 bond->dev->name, slave->dev->name);
3030 bond_set_slave_inactive_flags(slave);
3032 write_lock_bh(&bond->curr_slave_lock);
3034 bond_select_active_slave(bond);
3035 if (bond->curr_active_slave)
3036 bond->curr_active_slave->jiffies =
3039 write_unlock_bh(&bond->curr_slave_lock);
3041 bond->current_arp_slave = NULL;
3043 } else if (slave->state == BOND_STATE_BACKUP) {
3044 printk(KERN_INFO DRV_NAME
3045 ": %s: backup interface %s is now down\n",
3046 bond->dev->name, slave->dev->name);
3048 bond_set_slave_inactive_flags(slave);
3053 printk(KERN_ERR DRV_NAME
3054 ": %s: impossible: new_link %d on slave %s\n",
3055 bond->dev->name, slave->new_link,
3061 * No race with changes to primary via sysfs, as we hold rtnl.
3063 if (bond->primary_slave &&
3064 (bond->primary_slave != bond->curr_active_slave) &&
3065 (bond->primary_slave->link == BOND_LINK_UP)) {
3066 write_lock_bh(&bond->curr_slave_lock);
3067 bond_change_active_slave(bond, bond->primary_slave);
3068 write_unlock_bh(&bond->curr_slave_lock);
3071 bond_set_carrier(bond);
3075 * Send ARP probes for active-backup mode ARP monitor.
3077 * Called with bond->lock held for read.
3079 static void bond_ab_arp_probe(struct bonding *bond)
3081 struct slave *slave;
3084 read_lock(&bond->curr_slave_lock);
3086 if (bond->current_arp_slave && bond->curr_active_slave)
3087 printk("PROBE: c_arp %s && cas %s BAD\n",
3088 bond->current_arp_slave->dev->name,
3089 bond->curr_active_slave->dev->name);
3091 if (bond->curr_active_slave) {
3092 bond_arp_send_all(bond, bond->curr_active_slave);
3093 read_unlock(&bond->curr_slave_lock);
3097 read_unlock(&bond->curr_slave_lock);
3099 /* if we don't have a curr_active_slave, search for the next available
3100 * backup slave from the current_arp_slave and make it the candidate
3101 * for becoming the curr_active_slave
3104 if (!bond->current_arp_slave) {
3105 bond->current_arp_slave = bond->first_slave;
3106 if (!bond->current_arp_slave)
3110 bond_set_slave_inactive_flags(bond->current_arp_slave);
3112 /* search for next candidate */
3113 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3114 if (IS_UP(slave->dev)) {
3115 slave->link = BOND_LINK_BACK;
3116 bond_set_slave_active_flags(slave);
3117 bond_arp_send_all(bond, slave);
3118 slave->jiffies = jiffies;
3119 bond->current_arp_slave = slave;
3123 /* if the link state is up at this point, we
3124 * mark it down - this can happen if we have
3125 * simultaneous link failures and
3126 * reselect_active_interface doesn't make this
3127 * one the current slave so it is still marked
3128 * up when it is actually down
3130 if (slave->link == BOND_LINK_UP) {
3131 slave->link = BOND_LINK_DOWN;
3132 if (slave->link_failure_count < UINT_MAX)
3133 slave->link_failure_count++;
3135 bond_set_slave_inactive_flags(slave);
3137 printk(KERN_INFO DRV_NAME
3138 ": %s: backup interface %s is now down.\n",
3139 bond->dev->name, slave->dev->name);
3144 void bond_activebackup_arp_mon(struct work_struct *work)
3146 struct bonding *bond = container_of(work, struct bonding,
3150 read_lock(&bond->lock);
3152 if (bond->kill_timers)
3155 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3157 if (bond->slave_cnt == 0)
3160 if (bond->send_grat_arp) {
3161 read_lock(&bond->curr_slave_lock);
3162 bond_send_gratuitous_arp(bond);
3163 read_unlock(&bond->curr_slave_lock);
3166 if (bond->send_unsol_na) {
3167 read_lock(&bond->curr_slave_lock);
3168 bond_send_unsolicited_na(bond);
3169 read_unlock(&bond->curr_slave_lock);
3172 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3173 read_unlock(&bond->lock);
3175 read_lock(&bond->lock);
3177 bond_ab_arp_commit(bond, delta_in_ticks);
3179 read_unlock(&bond->lock);
3181 read_lock(&bond->lock);
3184 bond_ab_arp_probe(bond);
3187 if (bond->params.arp_interval) {
3188 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3191 read_unlock(&bond->lock);
3194 /*------------------------------ proc/seq_file-------------------------------*/
3196 #ifdef CONFIG_PROC_FS
3198 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3200 struct bonding *bond = seq->private;
3202 struct slave *slave;
3205 /* make sure the bond won't be taken away */
3206 read_lock(&dev_base_lock);
3207 read_lock(&bond->lock);
3210 return SEQ_START_TOKEN;
3213 bond_for_each_slave(bond, slave, i) {
3214 if (++off == *pos) {
3222 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3224 struct bonding *bond = seq->private;
3225 struct slave *slave = v;
3228 if (v == SEQ_START_TOKEN) {
3229 return bond->first_slave;
3232 slave = slave->next;
3234 return (slave == bond->first_slave) ? NULL : slave;
3237 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3239 struct bonding *bond = seq->private;
3241 read_unlock(&bond->lock);
3242 read_unlock(&dev_base_lock);
3245 static void bond_info_show_master(struct seq_file *seq)
3247 struct bonding *bond = seq->private;
3251 read_lock(&bond->curr_slave_lock);
3252 curr = bond->curr_active_slave;
3253 read_unlock(&bond->curr_slave_lock);
3255 seq_printf(seq, "Bonding Mode: %s",
3256 bond_mode_name(bond->params.mode));
3258 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3259 bond->params.fail_over_mac)
3260 seq_printf(seq, " (fail_over_mac %s)",
3261 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3263 seq_printf(seq, "\n");
3265 if (bond->params.mode == BOND_MODE_XOR ||
3266 bond->params.mode == BOND_MODE_8023AD) {
3267 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3268 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3269 bond->params.xmit_policy);
3272 if (USES_PRIMARY(bond->params.mode)) {
3273 seq_printf(seq, "Primary Slave: %s\n",
3274 (bond->primary_slave) ?
3275 bond->primary_slave->dev->name : "None");
3277 seq_printf(seq, "Currently Active Slave: %s\n",
3278 (curr) ? curr->dev->name : "None");
3281 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3283 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3284 seq_printf(seq, "Up Delay (ms): %d\n",
3285 bond->params.updelay * bond->params.miimon);
3286 seq_printf(seq, "Down Delay (ms): %d\n",
3287 bond->params.downdelay * bond->params.miimon);
3290 /* ARP information */
3291 if(bond->params.arp_interval > 0) {
3293 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3294 bond->params.arp_interval);
3296 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3298 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3299 if (!bond->params.arp_targets[i])
3302 seq_printf(seq, ",");
3303 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3306 seq_printf(seq, "\n");
3309 if (bond->params.mode == BOND_MODE_8023AD) {
3310 struct ad_info ad_info;
3312 seq_puts(seq, "\n802.3ad info\n");
3313 seq_printf(seq, "LACP rate: %s\n",
3314 (bond->params.lacp_fast) ? "fast" : "slow");
3315 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3316 ad_select_tbl[bond->params.ad_select].modename);
3318 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3319 seq_printf(seq, "bond %s has no active aggregator\n",
3322 seq_printf(seq, "Active Aggregator Info:\n");
3324 seq_printf(seq, "\tAggregator ID: %d\n",
3325 ad_info.aggregator_id);
3326 seq_printf(seq, "\tNumber of ports: %d\n",
3328 seq_printf(seq, "\tActor Key: %d\n",
3330 seq_printf(seq, "\tPartner Key: %d\n",
3331 ad_info.partner_key);
3332 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3333 ad_info.partner_system);
3338 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3340 struct bonding *bond = seq->private;
3342 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3343 seq_printf(seq, "MII Status: %s\n",
3344 (slave->link == BOND_LINK_UP) ? "up" : "down");
3345 seq_printf(seq, "Link Failure Count: %u\n",
3346 slave->link_failure_count);
3348 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3350 if (bond->params.mode == BOND_MODE_8023AD) {
3351 const struct aggregator *agg
3352 = SLAVE_AD_INFO(slave).port.aggregator;
3355 seq_printf(seq, "Aggregator ID: %d\n",
3356 agg->aggregator_identifier);
3358 seq_puts(seq, "Aggregator ID: N/A\n");
3363 static int bond_info_seq_show(struct seq_file *seq, void *v)
3365 if (v == SEQ_START_TOKEN) {
3366 seq_printf(seq, "%s\n", version);
3367 bond_info_show_master(seq);
3369 bond_info_show_slave(seq, v);
3375 static struct seq_operations bond_info_seq_ops = {
3376 .start = bond_info_seq_start,
3377 .next = bond_info_seq_next,
3378 .stop = bond_info_seq_stop,
3379 .show = bond_info_seq_show,
3382 static int bond_info_open(struct inode *inode, struct file *file)
3384 struct seq_file *seq;
3385 struct proc_dir_entry *proc;
3388 res = seq_open(file, &bond_info_seq_ops);
3390 /* recover the pointer buried in proc_dir_entry data */
3391 seq = file->private_data;
3393 seq->private = proc->data;
3399 static const struct file_operations bond_info_fops = {
3400 .owner = THIS_MODULE,
3401 .open = bond_info_open,
3403 .llseek = seq_lseek,
3404 .release = seq_release,
3407 static int bond_create_proc_entry(struct bonding *bond)
3409 struct net_device *bond_dev = bond->dev;
3411 if (bond_proc_dir) {
3412 bond->proc_entry = proc_create_data(bond_dev->name,
3413 S_IRUGO, bond_proc_dir,
3414 &bond_info_fops, bond);
3415 if (bond->proc_entry == NULL) {
3416 printk(KERN_WARNING DRV_NAME
3417 ": Warning: Cannot create /proc/net/%s/%s\n",
3418 DRV_NAME, bond_dev->name);
3420 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3427 static void bond_remove_proc_entry(struct bonding *bond)
3429 if (bond_proc_dir && bond->proc_entry) {
3430 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3431 memset(bond->proc_file_name, 0, IFNAMSIZ);
3432 bond->proc_entry = NULL;
3436 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3437 * Caller must hold rtnl_lock.
3439 static void bond_create_proc_dir(void)
3441 int len = strlen(DRV_NAME);
3443 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3444 bond_proc_dir = bond_proc_dir->next) {
3445 if ((bond_proc_dir->namelen == len) &&
3446 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3451 if (!bond_proc_dir) {
3452 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3453 if (bond_proc_dir) {
3454 bond_proc_dir->owner = THIS_MODULE;
3456 printk(KERN_WARNING DRV_NAME
3457 ": Warning: cannot create /proc/net/%s\n",
3463 /* Destroy the bonding directory under /proc/net, if empty.
3464 * Caller must hold rtnl_lock.
3466 static void bond_destroy_proc_dir(void)
3468 struct proc_dir_entry *de;
3470 if (!bond_proc_dir) {
3474 /* verify that the /proc dir is empty */
3475 for (de = bond_proc_dir->subdir; de; de = de->next) {
3476 /* ignore . and .. */
3477 if (*(de->name) != '.') {
3483 if (bond_proc_dir->owner == THIS_MODULE) {
3484 bond_proc_dir->owner = NULL;
3487 remove_proc_entry(DRV_NAME, init_net.proc_net);
3488 bond_proc_dir = NULL;
3491 #endif /* CONFIG_PROC_FS */
3493 /*-------------------------- netdev event handling --------------------------*/
3496 * Change device name
3498 static int bond_event_changename(struct bonding *bond)
3500 #ifdef CONFIG_PROC_FS
3501 bond_remove_proc_entry(bond);
3502 bond_create_proc_entry(bond);
3504 down_write(&(bonding_rwsem));
3505 bond_destroy_sysfs_entry(bond);
3506 bond_create_sysfs_entry(bond);
3507 up_write(&(bonding_rwsem));
3511 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3513 struct bonding *event_bond = netdev_priv(bond_dev);
3516 case NETDEV_CHANGENAME:
3517 return bond_event_changename(event_bond);
3518 case NETDEV_UNREGISTER:
3519 bond_release_all(event_bond->dev);
3528 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3530 struct net_device *bond_dev = slave_dev->master;
3531 struct bonding *bond = netdev_priv(bond_dev);
3534 case NETDEV_UNREGISTER:
3536 if (bond->setup_by_slave)
3537 bond_release_and_destroy(bond_dev, slave_dev);
3539 bond_release(bond_dev, slave_dev);
3544 * TODO: is this what we get if somebody
3545 * sets up a hierarchical bond, then rmmod's
3546 * one of the slave bonding devices?
3551 * ... Or is it this?
3554 case NETDEV_CHANGEMTU:
3556 * TODO: Should slaves be allowed to
3557 * independently alter their MTU? For
3558 * an active-backup bond, slaves need
3559 * not be the same type of device, so
3560 * MTUs may vary. For other modes,
3561 * slaves arguably should have the
3562 * same MTUs. To do this, we'd need to
3563 * take over the slave's change_mtu
3564 * function for the duration of their
3568 case NETDEV_CHANGENAME:
3570 * TODO: handle changing the primary's name
3573 case NETDEV_FEAT_CHANGE:
3574 bond_compute_features(bond);
3584 * bond_netdev_event: handle netdev notifier chain events.
3586 * This function receives events for the netdev chain. The caller (an
3587 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3588 * locks for us to safely manipulate the slave devices (RTNL lock,
3591 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3593 struct net_device *event_dev = (struct net_device *)ptr;
3595 if (dev_net(event_dev) != &init_net)
3598 dprintk("event_dev: %s, event: %lx\n",
3599 (event_dev ? event_dev->name : "None"),
3602 if (!(event_dev->priv_flags & IFF_BONDING))
3605 if (event_dev->flags & IFF_MASTER) {
3606 dprintk("IFF_MASTER\n");
3607 return bond_master_netdev_event(event, event_dev);
3610 if (event_dev->flags & IFF_SLAVE) {
3611 dprintk("IFF_SLAVE\n");
3612 return bond_slave_netdev_event(event, event_dev);
3619 * bond_inetaddr_event: handle inetaddr notifier chain events.
3621 * We keep track of device IPs primarily to use as source addresses in
3622 * ARP monitor probes (rather than spewing out broadcasts all the time).
3624 * We track one IP for the main device (if it has one), plus one per VLAN.
3626 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3628 struct in_ifaddr *ifa = ptr;
3629 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3630 struct bonding *bond;
3631 struct vlan_entry *vlan;
3633 if (dev_net(ifa->ifa_dev->dev) != &init_net)
3636 list_for_each_entry(bond, &bond_dev_list, bond_list) {
3637 if (bond->dev == event_dev) {
3640 bond->master_ip = ifa->ifa_local;
3643 bond->master_ip = bond_glean_dev_ip(bond->dev);
3650 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3651 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3652 if (vlan_dev == event_dev) {
3655 vlan->vlan_ip = ifa->ifa_local;
3659 bond_glean_dev_ip(vlan_dev);
3670 static struct notifier_block bond_netdev_notifier = {
3671 .notifier_call = bond_netdev_event,
3674 static struct notifier_block bond_inetaddr_notifier = {
3675 .notifier_call = bond_inetaddr_event,
3678 /*-------------------------- Packet type handling ---------------------------*/
3680 /* register to receive lacpdus on a bond */
3681 static void bond_register_lacpdu(struct bonding *bond)
3683 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3685 /* initialize packet type */
3686 pk_type->type = PKT_TYPE_LACPDU;
3687 pk_type->dev = bond->dev;
3688 pk_type->func = bond_3ad_lacpdu_recv;
3690 dev_add_pack(pk_type);
3693 /* unregister to receive lacpdus on a bond */
3694 static void bond_unregister_lacpdu(struct bonding *bond)
3696 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3699 void bond_register_arp(struct bonding *bond)
3701 struct packet_type *pt = &bond->arp_mon_pt;
3706 pt->type = htons(ETH_P_ARP);
3707 pt->dev = bond->dev;
3708 pt->func = bond_arp_rcv;
3712 void bond_unregister_arp(struct bonding *bond)
3714 struct packet_type *pt = &bond->arp_mon_pt;
3716 dev_remove_pack(pt);
3720 /*---------------------------- Hashing Policies -----------------------------*/
3723 * Hash for the output device based upon layer 2 and layer 3 data. If
3724 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3726 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3727 struct net_device *bond_dev, int count)
3729 struct ethhdr *data = (struct ethhdr *)skb->data;
3730 struct iphdr *iph = ip_hdr(skb);
3732 if (skb->protocol == htons(ETH_P_IP)) {
3733 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3734 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3737 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3741 * Hash for the output device based upon layer 3 and layer 4 data. If
3742 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3743 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3745 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3746 struct net_device *bond_dev, int count)
3748 struct ethhdr *data = (struct ethhdr *)skb->data;
3749 struct iphdr *iph = ip_hdr(skb);
3750 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3753 if (skb->protocol == htons(ETH_P_IP)) {
3754 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3755 (iph->protocol == IPPROTO_TCP ||
3756 iph->protocol == IPPROTO_UDP)) {
3757 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3759 return (layer4_xor ^
3760 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3764 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3768 * Hash for the output device based upon layer 2 data
3770 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3771 struct net_device *bond_dev, int count)
3773 struct ethhdr *data = (struct ethhdr *)skb->data;
3775 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3778 /*-------------------------- Device entry points ----------------------------*/
3780 static int bond_open(struct net_device *bond_dev)
3782 struct bonding *bond = netdev_priv(bond_dev);
3784 bond->kill_timers = 0;
3786 if (bond_is_lb(bond)) {
3787 /* bond_alb_initialize must be called before the timer
3790 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3791 /* something went wrong - fail the open operation */
3795 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3796 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3799 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3800 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3801 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3804 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3805 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3806 INIT_DELAYED_WORK(&bond->arp_work,
3807 bond_activebackup_arp_mon);
3809 INIT_DELAYED_WORK(&bond->arp_work,
3810 bond_loadbalance_arp_mon);
3812 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3813 if (bond->params.arp_validate)
3814 bond_register_arp(bond);
3817 if (bond->params.mode == BOND_MODE_8023AD) {
3818 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3819 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3820 /* register to receive LACPDUs */
3821 bond_register_lacpdu(bond);
3822 bond_3ad_initiate_agg_selection(bond, 1);
3828 static int bond_close(struct net_device *bond_dev)
3830 struct bonding *bond = netdev_priv(bond_dev);
3832 if (bond->params.mode == BOND_MODE_8023AD) {
3833 /* Unregister the receive of LACPDUs */
3834 bond_unregister_lacpdu(bond);
3837 if (bond->params.arp_validate)
3838 bond_unregister_arp(bond);
3840 write_lock_bh(&bond->lock);
3842 bond->send_grat_arp = 0;
3843 bond->send_unsol_na = 0;
3845 /* signal timers not to re-arm */
3846 bond->kill_timers = 1;
3848 write_unlock_bh(&bond->lock);
3850 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3851 cancel_delayed_work(&bond->mii_work);
3854 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3855 cancel_delayed_work(&bond->arp_work);
3858 switch (bond->params.mode) {
3859 case BOND_MODE_8023AD:
3860 cancel_delayed_work(&bond->ad_work);
3864 cancel_delayed_work(&bond->alb_work);
3871 if (bond_is_lb(bond)) {
3872 /* Must be called only after all
3873 * slaves have been released
3875 bond_alb_deinitialize(bond);
3881 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3883 struct bonding *bond = netdev_priv(bond_dev);
3884 struct net_device_stats *stats = &bond->stats;
3885 struct net_device_stats local_stats;
3886 struct slave *slave;
3889 memset(&local_stats, 0, sizeof(struct net_device_stats));
3891 read_lock_bh(&bond->lock);
3893 bond_for_each_slave(bond, slave, i) {
3894 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3896 local_stats.rx_packets += sstats->rx_packets;
3897 local_stats.rx_bytes += sstats->rx_bytes;
3898 local_stats.rx_errors += sstats->rx_errors;
3899 local_stats.rx_dropped += sstats->rx_dropped;
3901 local_stats.tx_packets += sstats->tx_packets;
3902 local_stats.tx_bytes += sstats->tx_bytes;
3903 local_stats.tx_errors += sstats->tx_errors;
3904 local_stats.tx_dropped += sstats->tx_dropped;
3906 local_stats.multicast += sstats->multicast;
3907 local_stats.collisions += sstats->collisions;
3909 local_stats.rx_length_errors += sstats->rx_length_errors;
3910 local_stats.rx_over_errors += sstats->rx_over_errors;
3911 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3912 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3913 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3914 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3916 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3917 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3918 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3919 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3920 local_stats.tx_window_errors += sstats->tx_window_errors;
3923 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3925 read_unlock_bh(&bond->lock);
3930 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3932 struct net_device *slave_dev = NULL;
3933 struct ifbond k_binfo;
3934 struct ifbond __user *u_binfo = NULL;
3935 struct ifslave k_sinfo;
3936 struct ifslave __user *u_sinfo = NULL;
3937 struct mii_ioctl_data *mii = NULL;
3940 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3941 bond_dev->name, cmd);
3953 * We do this again just in case we were called by SIOCGMIIREG
3954 * instead of SIOCGMIIPHY.
3961 if (mii->reg_num == 1) {
3962 struct bonding *bond = netdev_priv(bond_dev);
3964 read_lock(&bond->lock);
3965 read_lock(&bond->curr_slave_lock);
3966 if (netif_carrier_ok(bond->dev)) {
3967 mii->val_out = BMSR_LSTATUS;
3969 read_unlock(&bond->curr_slave_lock);
3970 read_unlock(&bond->lock);
3974 case BOND_INFO_QUERY_OLD:
3975 case SIOCBONDINFOQUERY:
3976 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3978 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3982 res = bond_info_query(bond_dev, &k_binfo);
3984 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3990 case BOND_SLAVE_INFO_QUERY_OLD:
3991 case SIOCBONDSLAVEINFOQUERY:
3992 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3994 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3998 res = bond_slave_info_query(bond_dev, &k_sinfo);
4000 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4011 if (!capable(CAP_NET_ADMIN)) {
4015 down_write(&(bonding_rwsem));
4016 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
4018 dprintk("slave_dev=%p: \n", slave_dev);
4023 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4025 case BOND_ENSLAVE_OLD:
4026 case SIOCBONDENSLAVE:
4027 res = bond_enslave(bond_dev, slave_dev);
4029 case BOND_RELEASE_OLD:
4030 case SIOCBONDRELEASE:
4031 res = bond_release(bond_dev, slave_dev);
4033 case BOND_SETHWADDR_OLD:
4034 case SIOCBONDSETHWADDR:
4035 res = bond_sethwaddr(bond_dev, slave_dev);
4037 case BOND_CHANGE_ACTIVE_OLD:
4038 case SIOCBONDCHANGEACTIVE:
4039 res = bond_ioctl_change_active(bond_dev, slave_dev);
4048 up_write(&(bonding_rwsem));
4052 static void bond_set_multicast_list(struct net_device *bond_dev)
4054 struct bonding *bond = netdev_priv(bond_dev);
4055 struct dev_mc_list *dmi;
4058 * Do promisc before checking multicast_mode
4060 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4062 * FIXME: Need to handle the error when one of the multi-slaves
4065 bond_set_promiscuity(bond, 1);
4068 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4069 bond_set_promiscuity(bond, -1);
4072 /* set allmulti flag to slaves */
4073 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4075 * FIXME: Need to handle the error when one of the multi-slaves
4078 bond_set_allmulti(bond, 1);
4081 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4082 bond_set_allmulti(bond, -1);
4085 read_lock(&bond->lock);
4087 bond->flags = bond_dev->flags;
4089 /* looking for addresses to add to slaves' mc list */
4090 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4091 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4092 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4096 /* looking for addresses to delete from slaves' list */
4097 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4098 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4099 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4103 /* save master's multicast list */
4104 bond_mc_list_destroy(bond);
4105 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4107 read_unlock(&bond->lock);
4110 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4112 struct bonding *bond = netdev_priv(dev);
4113 struct slave *slave = bond->first_slave;
4116 const struct net_device_ops *slave_ops
4117 = slave->dev->netdev_ops;
4118 if (slave_ops->ndo_neigh_setup)
4119 return slave_ops->ndo_neigh_setup(dev, parms);
4125 * Change the MTU of all of a master's slaves to match the master
4127 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4129 struct bonding *bond = netdev_priv(bond_dev);
4130 struct slave *slave, *stop_at;
4134 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4135 (bond_dev ? bond_dev->name : "None"), new_mtu);
4137 /* Can't hold bond->lock with bh disabled here since
4138 * some base drivers panic. On the other hand we can't
4139 * hold bond->lock without bh disabled because we'll
4140 * deadlock. The only solution is to rely on the fact
4141 * that we're under rtnl_lock here, and the slaves
4142 * list won't change. This doesn't solve the problem
4143 * of setting the slave's MTU while it is
4144 * transmitting, but the assumption is that the base
4145 * driver can handle that.
4147 * TODO: figure out a way to safely iterate the slaves
4148 * list, but without holding a lock around the actual
4149 * call to the base driver.
4152 bond_for_each_slave(bond, slave, i) {
4153 dprintk("s %p s->p %p c_m %p\n", slave,
4154 slave->prev, slave->dev->change_mtu);
4156 res = dev_set_mtu(slave->dev, new_mtu);
4159 /* If we failed to set the slave's mtu to the new value
4160 * we must abort the operation even in ACTIVE_BACKUP
4161 * mode, because if we allow the backup slaves to have
4162 * different mtu values than the active slave we'll
4163 * need to change their mtu when doing a failover. That
4164 * means changing their mtu from timer context, which
4165 * is probably not a good idea.
4167 dprintk("err %d %s\n", res, slave->dev->name);
4172 bond_dev->mtu = new_mtu;
4177 /* unwind from head to the slave that failed */
4179 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4182 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4184 dprintk("unwind err %d dev %s\n", tmp_res,
4195 * Note that many devices must be down to change the HW address, and
4196 * downing the master releases all slaves. We can make bonds full of
4197 * bonding devices to test this, however.
4199 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4201 struct bonding *bond = netdev_priv(bond_dev);
4202 struct sockaddr *sa = addr, tmp_sa;
4203 struct slave *slave, *stop_at;
4207 if (bond->params.mode == BOND_MODE_ALB)
4208 return bond_alb_set_mac_address(bond_dev, addr);
4211 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4214 * If fail_over_mac is set to active, do nothing and return
4215 * success. Returning an error causes ifenslave to fail.
4217 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4220 if (!is_valid_ether_addr(sa->sa_data)) {
4221 return -EADDRNOTAVAIL;
4224 /* Can't hold bond->lock with bh disabled here since
4225 * some base drivers panic. On the other hand we can't
4226 * hold bond->lock without bh disabled because we'll
4227 * deadlock. The only solution is to rely on the fact
4228 * that we're under rtnl_lock here, and the slaves
4229 * list won't change. This doesn't solve the problem
4230 * of setting the slave's hw address while it is
4231 * transmitting, but the assumption is that the base
4232 * driver can handle that.
4234 * TODO: figure out a way to safely iterate the slaves
4235 * list, but without holding a lock around the actual
4236 * call to the base driver.
4239 bond_for_each_slave(bond, slave, i) {
4240 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4241 dprintk("slave %p %s\n", slave, slave->dev->name);
4243 if (slave_ops->ndo_set_mac_address == NULL) {
4245 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4249 res = dev_set_mac_address(slave->dev, addr);
4251 /* TODO: consider downing the slave
4253 * User should expect communications
4254 * breakage anyway until ARP finish
4257 dprintk("err %d %s\n", res, slave->dev->name);
4263 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4267 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4268 tmp_sa.sa_family = bond_dev->type;
4270 /* unwind from head to the slave that failed */
4272 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4275 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4277 dprintk("unwind err %d dev %s\n", tmp_res,
4285 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4287 struct bonding *bond = netdev_priv(bond_dev);
4288 struct slave *slave, *start_at;
4289 int i, slave_no, res = 1;
4291 read_lock(&bond->lock);
4293 if (!BOND_IS_OK(bond)) {
4298 * Concurrent TX may collide on rr_tx_counter; we accept that
4299 * as being rare enough not to justify using an atomic op here
4301 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4303 bond_for_each_slave(bond, slave, i) {
4311 bond_for_each_slave_from(bond, slave, i, start_at) {
4312 if (IS_UP(slave->dev) &&
4313 (slave->link == BOND_LINK_UP) &&
4314 (slave->state == BOND_STATE_ACTIVE)) {
4315 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4322 /* no suitable interface, frame not sent */
4325 read_unlock(&bond->lock);
4331 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4332 * the bond has a usable interface.
4334 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4336 struct bonding *bond = netdev_priv(bond_dev);
4339 read_lock(&bond->lock);
4340 read_lock(&bond->curr_slave_lock);
4342 if (!BOND_IS_OK(bond)) {
4346 if (!bond->curr_active_slave)
4349 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4353 /* no suitable interface, frame not sent */
4356 read_unlock(&bond->curr_slave_lock);
4357 read_unlock(&bond->lock);
4362 * In bond_xmit_xor() , we determine the output device by using a pre-
4363 * determined xmit_hash_policy(), If the selected device is not enabled,
4364 * find the next active slave.
4366 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4368 struct bonding *bond = netdev_priv(bond_dev);
4369 struct slave *slave, *start_at;
4374 read_lock(&bond->lock);
4376 if (!BOND_IS_OK(bond)) {
4380 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4382 bond_for_each_slave(bond, slave, i) {
4391 bond_for_each_slave_from(bond, slave, i, start_at) {
4392 if (IS_UP(slave->dev) &&
4393 (slave->link == BOND_LINK_UP) &&
4394 (slave->state == BOND_STATE_ACTIVE)) {
4395 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4402 /* no suitable interface, frame not sent */
4405 read_unlock(&bond->lock);
4410 * in broadcast mode, we send everything to all usable interfaces.
4412 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4414 struct bonding *bond = netdev_priv(bond_dev);
4415 struct slave *slave, *start_at;
4416 struct net_device *tx_dev = NULL;
4420 read_lock(&bond->lock);
4422 if (!BOND_IS_OK(bond)) {
4426 read_lock(&bond->curr_slave_lock);
4427 start_at = bond->curr_active_slave;
4428 read_unlock(&bond->curr_slave_lock);
4434 bond_for_each_slave_from(bond, slave, i, start_at) {
4435 if (IS_UP(slave->dev) &&
4436 (slave->link == BOND_LINK_UP) &&
4437 (slave->state == BOND_STATE_ACTIVE)) {
4439 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4441 printk(KERN_ERR DRV_NAME
4442 ": %s: Error: bond_xmit_broadcast(): "
4443 "skb_clone() failed\n",
4448 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4450 dev_kfree_skb(skb2);
4454 tx_dev = slave->dev;
4459 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4464 /* no suitable interface, frame not sent */
4467 /* frame sent to all suitable interfaces */
4468 read_unlock(&bond->lock);
4472 /*------------------------- Device initialization ---------------------------*/
4474 static void bond_set_xmit_hash_policy(struct bonding *bond)
4476 switch (bond->params.xmit_policy) {
4477 case BOND_XMIT_POLICY_LAYER23:
4478 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4480 case BOND_XMIT_POLICY_LAYER34:
4481 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4483 case BOND_XMIT_POLICY_LAYER2:
4485 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4490 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4492 const struct bonding *bond = netdev_priv(dev);
4494 switch (bond->params.mode) {
4495 case BOND_MODE_ROUNDROBIN:
4496 return bond_xmit_roundrobin(skb, dev);
4497 case BOND_MODE_ACTIVEBACKUP:
4498 return bond_xmit_activebackup(skb, dev);
4500 return bond_xmit_xor(skb, dev);
4501 case BOND_MODE_BROADCAST:
4502 return bond_xmit_broadcast(skb, dev);
4503 case BOND_MODE_8023AD:
4504 return bond_3ad_xmit_xor(skb, dev);
4507 return bond_alb_xmit(skb, dev);
4509 /* Should never happen, mode already checked */
4510 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4511 dev->name, bond->params.mode);
4514 return NETDEV_TX_OK;
4520 * set bond mode specific net device operations
4522 void bond_set_mode_ops(struct bonding *bond, int mode)
4524 struct net_device *bond_dev = bond->dev;
4527 case BOND_MODE_ROUNDROBIN:
4529 case BOND_MODE_ACTIVEBACKUP:
4532 bond_set_xmit_hash_policy(bond);
4534 case BOND_MODE_BROADCAST:
4536 case BOND_MODE_8023AD:
4537 bond_set_master_3ad_flags(bond);
4538 bond_set_xmit_hash_policy(bond);
4541 bond_set_master_alb_flags(bond);
4546 /* Should never happen, mode already checked */
4547 printk(KERN_ERR DRV_NAME
4548 ": %s: Error: Unknown bonding mode %d\n",
4555 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4556 struct ethtool_drvinfo *drvinfo)
4558 strncpy(drvinfo->driver, DRV_NAME, 32);
4559 strncpy(drvinfo->version, DRV_VERSION, 32);
4560 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4563 static const struct ethtool_ops bond_ethtool_ops = {
4564 .get_drvinfo = bond_ethtool_get_drvinfo,
4565 .get_link = ethtool_op_get_link,
4566 .get_tx_csum = ethtool_op_get_tx_csum,
4567 .get_sg = ethtool_op_get_sg,
4568 .get_tso = ethtool_op_get_tso,
4569 .get_ufo = ethtool_op_get_ufo,
4570 .get_flags = ethtool_op_get_flags,
4573 static const struct net_device_ops bond_netdev_ops = {
4574 .ndo_open = bond_open,
4575 .ndo_stop = bond_close,
4576 .ndo_start_xmit = bond_start_xmit,
4577 .ndo_get_stats = bond_get_stats,
4578 .ndo_do_ioctl = bond_do_ioctl,
4579 .ndo_set_multicast_list = bond_set_multicast_list,
4580 .ndo_change_mtu = bond_change_mtu,
4581 .ndo_set_mac_address = bond_set_mac_address,
4582 .ndo_neigh_setup = bond_neigh_setup,
4583 .ndo_vlan_rx_register = bond_vlan_rx_register,
4584 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4585 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4589 * Does not allocate but creates a /proc entry.
4592 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4594 struct bonding *bond = netdev_priv(bond_dev);
4596 dprintk("Begin bond_init for %s\n", bond_dev->name);
4598 /* initialize rwlocks */
4599 rwlock_init(&bond->lock);
4600 rwlock_init(&bond->curr_slave_lock);
4602 bond->params = *params; /* copy params struct */
4604 bond->wq = create_singlethread_workqueue(bond_dev->name);
4608 /* Initialize pointers */
4609 bond->first_slave = NULL;
4610 bond->curr_active_slave = NULL;
4611 bond->current_arp_slave = NULL;
4612 bond->primary_slave = NULL;
4613 bond->dev = bond_dev;
4614 bond->send_grat_arp = 0;
4615 bond->send_unsol_na = 0;
4616 bond->setup_by_slave = 0;
4617 INIT_LIST_HEAD(&bond->vlan_list);
4619 /* Initialize the device entry points */
4620 bond_dev->netdev_ops = &bond_netdev_ops;
4621 bond_dev->ethtool_ops = &bond_ethtool_ops;
4622 bond_set_mode_ops(bond, bond->params.mode);
4624 bond_dev->destructor = bond_destructor;
4626 /* Initialize the device options */
4627 bond_dev->tx_queue_len = 0;
4628 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4629 bond_dev->priv_flags |= IFF_BONDING;
4630 if (bond->params.arp_interval)
4631 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4633 /* At first, we block adding VLANs. That's the only way to
4634 * prevent problems that occur when adding VLANs over an
4635 * empty bond. The block will be removed once non-challenged
4636 * slaves are enslaved.
4638 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4640 /* don't acquire bond device's netif_tx_lock when
4642 bond_dev->features |= NETIF_F_LLTX;
4644 /* By default, we declare the bond to be fully
4645 * VLAN hardware accelerated capable. Special
4646 * care is taken in the various xmit functions
4647 * when there are slaves that are not hw accel
4650 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4651 NETIF_F_HW_VLAN_RX |
4652 NETIF_F_HW_VLAN_FILTER);
4654 #ifdef CONFIG_PROC_FS
4655 bond_create_proc_entry(bond);
4657 list_add_tail(&bond->bond_list, &bond_dev_list);
4662 static void bond_work_cancel_all(struct bonding *bond)
4664 write_lock_bh(&bond->lock);
4665 bond->kill_timers = 1;
4666 write_unlock_bh(&bond->lock);
4668 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4669 cancel_delayed_work(&bond->mii_work);
4671 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4672 cancel_delayed_work(&bond->arp_work);
4674 if (bond->params.mode == BOND_MODE_ALB &&
4675 delayed_work_pending(&bond->alb_work))
4676 cancel_delayed_work(&bond->alb_work);
4678 if (bond->params.mode == BOND_MODE_8023AD &&
4679 delayed_work_pending(&bond->ad_work))
4680 cancel_delayed_work(&bond->ad_work);
4683 /* De-initialize device specific data.
4684 * Caller must hold rtnl_lock.
4686 static void bond_deinit(struct net_device *bond_dev)
4688 struct bonding *bond = netdev_priv(bond_dev);
4690 list_del(&bond->bond_list);
4692 bond_work_cancel_all(bond);
4694 #ifdef CONFIG_PROC_FS
4695 bond_remove_proc_entry(bond);
4699 /* Unregister and free all bond devices.
4700 * Caller must hold rtnl_lock.
4702 static void bond_free_all(void)
4704 struct bonding *bond, *nxt;
4706 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4707 struct net_device *bond_dev = bond->dev;
4709 bond_work_cancel_all(bond);
4710 /* Release the bonded slaves */
4711 bond_release_all(bond_dev);
4715 #ifdef CONFIG_PROC_FS
4716 bond_destroy_proc_dir();
4720 /*------------------------- Module initialization ---------------------------*/
4723 * Convert string input module parms. Accept either the
4724 * number of the mode or its string name. A bit complicated because
4725 * some mode names are substrings of other names, and calls from sysfs
4726 * may have whitespace in the name (trailing newlines, for example).
4728 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4730 int mode = -1, i, rv;
4731 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4733 for (p = (char *)buf; *p; p++)
4734 if (!(isdigit(*p) || isspace(*p)))
4738 rv = sscanf(buf, "%20s", modestr);
4740 rv = sscanf(buf, "%d", &mode);
4745 for (i = 0; tbl[i].modename; i++) {
4746 if (mode == tbl[i].mode)
4748 if (strcmp(modestr, tbl[i].modename) == 0)
4755 static int bond_check_params(struct bond_params *params)
4757 int arp_validate_value, fail_over_mac_value;
4760 * Convert string parameters.
4763 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4764 if (bond_mode == -1) {
4765 printk(KERN_ERR DRV_NAME
4766 ": Error: Invalid bonding mode \"%s\"\n",
4767 mode == NULL ? "NULL" : mode);
4772 if (xmit_hash_policy) {
4773 if ((bond_mode != BOND_MODE_XOR) &&
4774 (bond_mode != BOND_MODE_8023AD)) {
4775 printk(KERN_INFO DRV_NAME
4776 ": xor_mode param is irrelevant in mode %s\n",
4777 bond_mode_name(bond_mode));
4779 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4781 if (xmit_hashtype == -1) {
4782 printk(KERN_ERR DRV_NAME
4783 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4784 xmit_hash_policy == NULL ? "NULL" :
4792 if (bond_mode != BOND_MODE_8023AD) {
4793 printk(KERN_INFO DRV_NAME
4794 ": lacp_rate param is irrelevant in mode %s\n",
4795 bond_mode_name(bond_mode));
4797 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4798 if (lacp_fast == -1) {
4799 printk(KERN_ERR DRV_NAME
4800 ": Error: Invalid lacp rate \"%s\"\n",
4801 lacp_rate == NULL ? "NULL" : lacp_rate);
4808 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4809 if (params->ad_select == -1) {
4810 printk(KERN_ERR DRV_NAME
4811 ": Error: Invalid ad_select \"%s\"\n",
4812 ad_select == NULL ? "NULL" : ad_select);
4816 if (bond_mode != BOND_MODE_8023AD) {
4817 printk(KERN_WARNING DRV_NAME
4818 ": ad_select param only affects 802.3ad mode\n");
4821 params->ad_select = BOND_AD_STABLE;
4824 if (max_bonds < 0 || max_bonds > INT_MAX) {
4825 printk(KERN_WARNING DRV_NAME
4826 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4827 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4828 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4829 max_bonds = BOND_DEFAULT_MAX_BONDS;
4833 printk(KERN_WARNING DRV_NAME
4834 ": Warning: miimon module parameter (%d), "
4835 "not in range 0-%d, so it was reset to %d\n",
4836 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4837 miimon = BOND_LINK_MON_INTERV;
4841 printk(KERN_WARNING DRV_NAME
4842 ": Warning: updelay module parameter (%d), "
4843 "not in range 0-%d, so it was reset to 0\n",
4848 if (downdelay < 0) {
4849 printk(KERN_WARNING DRV_NAME
4850 ": Warning: downdelay module parameter (%d), "
4851 "not in range 0-%d, so it was reset to 0\n",
4852 downdelay, INT_MAX);
4856 if ((use_carrier != 0) && (use_carrier != 1)) {
4857 printk(KERN_WARNING DRV_NAME
4858 ": Warning: use_carrier module parameter (%d), "
4859 "not of valid value (0/1), so it was set to 1\n",
4864 if (num_grat_arp < 0 || num_grat_arp > 255) {
4865 printk(KERN_WARNING DRV_NAME
4866 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4867 "was reset to 1 \n", num_grat_arp);
4871 if (num_unsol_na < 0 || num_unsol_na > 255) {
4872 printk(KERN_WARNING DRV_NAME
4873 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4874 "was reset to 1 \n", num_unsol_na);
4878 /* reset values for 802.3ad */
4879 if (bond_mode == BOND_MODE_8023AD) {
4881 printk(KERN_WARNING DRV_NAME
4882 ": Warning: miimon must be specified, "
4883 "otherwise bonding will not detect link "
4884 "failure, speed and duplex which are "
4885 "essential for 802.3ad operation\n");
4886 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4891 /* reset values for TLB/ALB */
4892 if ((bond_mode == BOND_MODE_TLB) ||
4893 (bond_mode == BOND_MODE_ALB)) {
4895 printk(KERN_WARNING DRV_NAME
4896 ": Warning: miimon must be specified, "
4897 "otherwise bonding will not detect link "
4898 "failure and link speed which are essential "
4899 "for TLB/ALB load balancing\n");
4900 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4905 if (bond_mode == BOND_MODE_ALB) {
4906 printk(KERN_NOTICE DRV_NAME
4907 ": In ALB mode you might experience client "
4908 "disconnections upon reconnection of a link if the "
4909 "bonding module updelay parameter (%d msec) is "
4910 "incompatible with the forwarding delay time of the "
4916 if (updelay || downdelay) {
4917 /* just warn the user the up/down delay will have
4918 * no effect since miimon is zero...
4920 printk(KERN_WARNING DRV_NAME
4921 ": Warning: miimon module parameter not set "
4922 "and updelay (%d) or downdelay (%d) module "
4923 "parameter is set; updelay and downdelay have "
4924 "no effect unless miimon is set\n",
4925 updelay, downdelay);
4928 /* don't allow arp monitoring */
4930 printk(KERN_WARNING DRV_NAME
4931 ": Warning: miimon (%d) and arp_interval (%d) "
4932 "can't be used simultaneously, disabling ARP "
4934 miimon, arp_interval);
4938 if ((updelay % miimon) != 0) {
4939 printk(KERN_WARNING DRV_NAME
4940 ": Warning: updelay (%d) is not a multiple "
4941 "of miimon (%d), updelay rounded to %d ms\n",
4942 updelay, miimon, (updelay / miimon) * miimon);
4947 if ((downdelay % miimon) != 0) {
4948 printk(KERN_WARNING DRV_NAME
4949 ": Warning: downdelay (%d) is not a multiple "
4950 "of miimon (%d), downdelay rounded to %d ms\n",
4952 (downdelay / miimon) * miimon);
4955 downdelay /= miimon;
4958 if (arp_interval < 0) {
4959 printk(KERN_WARNING DRV_NAME
4960 ": Warning: arp_interval module parameter (%d) "
4961 ", not in range 0-%d, so it was reset to %d\n",
4962 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4963 arp_interval = BOND_LINK_ARP_INTERV;
4966 for (arp_ip_count = 0;
4967 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4969 /* not complete check, but should be good enough to
4971 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4972 printk(KERN_WARNING DRV_NAME
4973 ": Warning: bad arp_ip_target module parameter "
4974 "(%s), ARP monitoring will not be performed\n",
4975 arp_ip_target[arp_ip_count]);
4978 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4979 arp_target[arp_ip_count] = ip;
4983 if (arp_interval && !arp_ip_count) {
4984 /* don't allow arping if no arp_ip_target given... */
4985 printk(KERN_WARNING DRV_NAME
4986 ": Warning: arp_interval module parameter (%d) "
4987 "specified without providing an arp_ip_target "
4988 "parameter, arp_interval was reset to 0\n",
4994 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4995 printk(KERN_ERR DRV_NAME
4996 ": arp_validate only supported in active-backup mode\n");
4999 if (!arp_interval) {
5000 printk(KERN_ERR DRV_NAME
5001 ": arp_validate requires arp_interval\n");
5005 arp_validate_value = bond_parse_parm(arp_validate,
5007 if (arp_validate_value == -1) {
5008 printk(KERN_ERR DRV_NAME
5009 ": Error: invalid arp_validate \"%s\"\n",
5010 arp_validate == NULL ? "NULL" : arp_validate);
5014 arp_validate_value = 0;
5017 printk(KERN_INFO DRV_NAME
5018 ": MII link monitoring set to %d ms\n",
5020 } else if (arp_interval) {
5023 printk(KERN_INFO DRV_NAME
5024 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5026 arp_validate_tbl[arp_validate_value].modename,
5029 for (i = 0; i < arp_ip_count; i++)
5030 printk (" %s", arp_ip_target[i]);
5034 } else if (max_bonds) {
5035 /* miimon and arp_interval not set, we need one so things
5036 * work as expected, see bonding.txt for details
5038 printk(KERN_WARNING DRV_NAME
5039 ": Warning: either miimon or arp_interval and "
5040 "arp_ip_target module parameters must be specified, "
5041 "otherwise bonding will not detect link failures! see "
5042 "bonding.txt for details.\n");
5045 if (primary && !USES_PRIMARY(bond_mode)) {
5046 /* currently, using a primary only makes sense
5047 * in active backup, TLB or ALB modes
5049 printk(KERN_WARNING DRV_NAME
5050 ": Warning: %s primary device specified but has no "
5051 "effect in %s mode\n",
5052 primary, bond_mode_name(bond_mode));
5056 if (fail_over_mac) {
5057 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5059 if (fail_over_mac_value == -1) {
5060 printk(KERN_ERR DRV_NAME
5061 ": Error: invalid fail_over_mac \"%s\"\n",
5062 arp_validate == NULL ? "NULL" : arp_validate);
5066 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5067 printk(KERN_WARNING DRV_NAME
5068 ": Warning: fail_over_mac only affects "
5069 "active-backup mode.\n");
5071 fail_over_mac_value = BOND_FOM_NONE;
5074 /* fill params struct with the proper values */
5075 params->mode = bond_mode;
5076 params->xmit_policy = xmit_hashtype;
5077 params->miimon = miimon;
5078 params->num_grat_arp = num_grat_arp;
5079 params->num_unsol_na = num_unsol_na;
5080 params->arp_interval = arp_interval;
5081 params->arp_validate = arp_validate_value;
5082 params->updelay = updelay;
5083 params->downdelay = downdelay;
5084 params->use_carrier = use_carrier;
5085 params->lacp_fast = lacp_fast;
5086 params->primary[0] = 0;
5087 params->fail_over_mac = fail_over_mac_value;
5090 strncpy(params->primary, primary, IFNAMSIZ);
5091 params->primary[IFNAMSIZ - 1] = 0;
5094 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5099 static struct lock_class_key bonding_netdev_xmit_lock_key;
5100 static struct lock_class_key bonding_netdev_addr_lock_key;
5102 static void bond_set_lockdep_class_one(struct net_device *dev,
5103 struct netdev_queue *txq,
5106 lockdep_set_class(&txq->_xmit_lock,
5107 &bonding_netdev_xmit_lock_key);
5110 static void bond_set_lockdep_class(struct net_device *dev)
5112 lockdep_set_class(&dev->addr_list_lock,
5113 &bonding_netdev_addr_lock_key);
5114 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5117 /* Create a new bond based on the specified name and bonding parameters.
5118 * If name is NULL, obtain a suitable "bond%d" name for us.
5119 * Caller must NOT hold rtnl_lock; we need to release it here before we
5120 * set up our sysfs entries.
5122 int bond_create(char *name, struct bond_params *params)
5124 struct net_device *bond_dev;
5125 struct bonding *bond;
5129 down_write(&bonding_rwsem);
5131 /* Check to see if the bond already exists. */
5133 list_for_each_entry(bond, &bond_dev_list, bond_list)
5134 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5135 printk(KERN_ERR DRV_NAME
5136 ": cannot add bond %s; it already exists\n",
5143 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5146 printk(KERN_ERR DRV_NAME
5147 ": %s: eek! can't alloc netdev!\n",
5154 res = dev_alloc_name(bond_dev, "bond%d");
5159 /* bond_init() must be called after dev_alloc_name() (for the
5160 * /proc files), but before register_netdevice(), because we
5161 * need to set function pointers.
5164 res = bond_init(bond_dev, params);
5169 res = register_netdevice(bond_dev);
5174 bond_set_lockdep_class(bond_dev);
5176 netif_carrier_off(bond_dev);
5178 up_write(&bonding_rwsem);
5179 rtnl_unlock(); /* allows sysfs registration of net device */
5180 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5183 down_write(&bonding_rwsem);
5184 bond_deinit(bond_dev);
5185 unregister_netdevice(bond_dev);
5192 bond_deinit(bond_dev);
5194 free_netdev(bond_dev);
5196 up_write(&bonding_rwsem);
5201 static int __init bonding_init(void)
5205 struct bonding *bond;
5207 printk(KERN_INFO "%s", version);
5209 res = bond_check_params(&bonding_defaults);
5214 #ifdef CONFIG_PROC_FS
5215 bond_create_proc_dir();
5218 init_rwsem(&bonding_rwsem);
5220 for (i = 0; i < max_bonds; i++) {
5221 res = bond_create(NULL, &bonding_defaults);
5226 res = bond_create_sysfs();
5230 register_netdevice_notifier(&bond_netdev_notifier);
5231 register_inetaddr_notifier(&bond_inetaddr_notifier);
5232 bond_register_ipv6_notifier();
5236 list_for_each_entry(bond, &bond_dev_list, bond_list) {
5237 bond_work_cancel_all(bond);
5238 destroy_workqueue(bond->wq);
5241 bond_destroy_sysfs();
5251 static void __exit bonding_exit(void)
5253 unregister_netdevice_notifier(&bond_netdev_notifier);
5254 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5255 bond_unregister_ipv6_notifier();
5257 bond_destroy_sysfs();
5264 module_init(bonding_init);
5265 module_exit(bonding_exit);
5266 MODULE_LICENSE("GPL");
5267 MODULE_VERSION(DRV_VERSION);
5268 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5269 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5270 MODULE_SUPPORTED_DEVICE("most ethernet devices");