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 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
54 #include <asm/system.h>
57 #include <asm/uaccess.h>
58 #include <linux/errno.h>
59 #include <linux/netdevice.h>
60 #include <linux/inetdevice.h>
61 #include <linux/igmp.h>
62 #include <linux/etherdevice.h>
63 #include <linux/skbuff.h>
65 #include <linux/rtnetlink.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
82 /*---------------------------- Module parameters ----------------------------*/
84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
85 #define BOND_LINK_MON_INTERV 0
86 #define BOND_LINK_ARP_INTERV 0
88 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
89 static int num_grat_arp = 1;
90 static int num_unsol_na = 1;
91 static int miimon = BOND_LINK_MON_INTERV;
92 static int updelay = 0;
93 static int downdelay = 0;
94 static int use_carrier = 1;
95 static char *mode = NULL;
96 static char *primary = NULL;
97 static char *lacp_rate = NULL;
98 static char *ad_select = NULL;
99 static char *xmit_hash_policy = NULL;
100 static int arp_interval = BOND_LINK_ARP_INTERV;
101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
102 static char *arp_validate = NULL;
103 static char *fail_over_mac = NULL;
104 struct bond_params bonding_defaults;
106 module_param(max_bonds, int, 0);
107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
108 module_param(num_grat_arp, int, 0644);
109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
110 module_param(num_unsol_na, int, 0644);
111 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
112 module_param(miimon, int, 0);
113 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
114 module_param(updelay, int, 0);
115 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
116 module_param(downdelay, int, 0);
117 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
119 module_param(use_carrier, int, 0);
120 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
121 "0 for off, 1 for on (default)");
122 module_param(mode, charp, 0);
123 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
124 "1 for active-backup, 2 for balance-xor, "
125 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
126 "6 for balance-alb");
127 module_param(primary, charp, 0);
128 MODULE_PARM_DESC(primary, "Primary network device to use");
129 module_param(lacp_rate, charp, 0);
130 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
132 module_param(ad_select, charp, 0);
133 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
134 module_param(xmit_hash_policy, charp, 0);
135 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
136 ", 1 for layer 3+4");
137 module_param(arp_interval, int, 0);
138 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
139 module_param_array(arp_ip_target, charp, NULL, 0);
140 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
141 module_param(arp_validate, charp, 0);
142 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
143 module_param(fail_over_mac, charp, 0);
144 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
146 /*----------------------------- Global variables ----------------------------*/
148 static const char * const version =
149 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
151 LIST_HEAD(bond_dev_list);
153 #ifdef CONFIG_PROC_FS
154 static struct proc_dir_entry *bond_proc_dir = NULL;
157 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
158 static int arp_ip_count = 0;
159 static int bond_mode = BOND_MODE_ROUNDROBIN;
160 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
161 static int lacp_fast = 0;
164 const struct bond_parm_tbl bond_lacp_tbl[] = {
165 { "slow", AD_LACP_SLOW},
166 { "fast", AD_LACP_FAST},
170 const struct bond_parm_tbl bond_mode_tbl[] = {
171 { "balance-rr", BOND_MODE_ROUNDROBIN},
172 { "active-backup", BOND_MODE_ACTIVEBACKUP},
173 { "balance-xor", BOND_MODE_XOR},
174 { "broadcast", BOND_MODE_BROADCAST},
175 { "802.3ad", BOND_MODE_8023AD},
176 { "balance-tlb", BOND_MODE_TLB},
177 { "balance-alb", BOND_MODE_ALB},
181 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
182 { "layer2", BOND_XMIT_POLICY_LAYER2},
183 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
184 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
188 const struct bond_parm_tbl arp_validate_tbl[] = {
189 { "none", BOND_ARP_VALIDATE_NONE},
190 { "active", BOND_ARP_VALIDATE_ACTIVE},
191 { "backup", BOND_ARP_VALIDATE_BACKUP},
192 { "all", BOND_ARP_VALIDATE_ALL},
196 const struct bond_parm_tbl fail_over_mac_tbl[] = {
197 { "none", BOND_FOM_NONE},
198 { "active", BOND_FOM_ACTIVE},
199 { "follow", BOND_FOM_FOLLOW},
203 struct bond_parm_tbl ad_select_tbl[] = {
204 { "stable", BOND_AD_STABLE},
205 { "bandwidth", BOND_AD_BANDWIDTH},
206 { "count", BOND_AD_COUNT},
210 /*-------------------------- Forward declarations ---------------------------*/
212 static void bond_send_gratuitous_arp(struct bonding *bond);
213 static void bond_deinit(struct net_device *bond_dev);
215 /*---------------------------- General routines -----------------------------*/
217 static const char *bond_mode_name(int mode)
219 static const char *names[] = {
220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
224 [BOND_MODE_8023AD]= "IEEE 802.3ad Dynamic link aggregation",
225 [BOND_MODE_TLB] = "transmit load balancing",
226 [BOND_MODE_ALB] = "adaptive load balancing",
229 if (mode < 0 || mode > BOND_MODE_ALB)
235 /*---------------------------------- VLAN -----------------------------------*/
238 * bond_add_vlan - add a new vlan id on bond
239 * @bond: bond that got the notification
240 * @vlan_id: the vlan id to add
242 * Returns -ENOMEM if allocation failed.
244 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
246 struct vlan_entry *vlan;
248 pr_debug("bond: %s, vlan id %d\n",
249 (bond ? bond->dev->name: "None"), vlan_id);
251 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
256 INIT_LIST_HEAD(&vlan->vlan_list);
257 vlan->vlan_id = vlan_id;
259 write_lock_bh(&bond->lock);
261 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
263 write_unlock_bh(&bond->lock);
265 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
271 * bond_del_vlan - delete a vlan id from bond
272 * @bond: bond that got the notification
273 * @vlan_id: the vlan id to delete
275 * returns -ENODEV if @vlan_id was not found in @bond.
277 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
279 struct vlan_entry *vlan;
282 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
284 write_lock_bh(&bond->lock);
286 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
287 if (vlan->vlan_id == vlan_id) {
288 list_del(&vlan->vlan_list);
290 if (bond_is_lb(bond))
291 bond_alb_clear_vlan(bond, vlan_id);
293 pr_debug("removed VLAN ID %d from bond %s\n", vlan_id,
298 if (list_empty(&bond->vlan_list) &&
299 (bond->slave_cnt == 0)) {
300 /* Last VLAN removed and no slaves, so
301 * restore block on adding VLANs. This will
302 * be removed once new slaves that are not
303 * VLAN challenged will be added.
305 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
313 pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id,
317 write_unlock_bh(&bond->lock);
322 * bond_has_challenged_slaves
323 * @bond: the bond we're working on
325 * Searches the slave list. Returns 1 if a vlan challenged slave
326 * was found, 0 otherwise.
328 * Assumes bond->lock is held.
330 static int bond_has_challenged_slaves(struct bonding *bond)
335 bond_for_each_slave(bond, slave, i) {
336 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
337 pr_debug("found VLAN challenged slave - %s\n",
343 pr_debug("no VLAN challenged slaves found\n");
348 * bond_next_vlan - safely skip to the next item in the vlans list.
349 * @bond: the bond we're working on
350 * @curr: item we're advancing from
352 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
353 * or @curr->next otherwise (even if it is @curr itself again).
355 * Caller must hold bond->lock
357 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
359 struct vlan_entry *next, *last;
361 if (list_empty(&bond->vlan_list)) {
366 next = list_entry(bond->vlan_list.next,
367 struct vlan_entry, vlan_list);
369 last = list_entry(bond->vlan_list.prev,
370 struct vlan_entry, vlan_list);
372 next = list_entry(bond->vlan_list.next,
373 struct vlan_entry, vlan_list);
375 next = list_entry(curr->vlan_list.next,
376 struct vlan_entry, vlan_list);
384 * bond_dev_queue_xmit - Prepare skb for xmit.
386 * @bond: bond device that got this skb for tx.
387 * @skb: hw accel VLAN tagged skb to transmit
388 * @slave_dev: slave that is supposed to xmit this skbuff
390 * When the bond gets an skb to transmit that is
391 * already hardware accelerated VLAN tagged, and it
392 * needs to relay this skb to a slave that is not
393 * hw accel capable, the skb needs to be "unaccelerated",
394 * i.e. strip the hwaccel tag and re-insert it as part
397 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
399 unsigned short uninitialized_var(vlan_id);
401 if (!list_empty(&bond->vlan_list) &&
402 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
403 vlan_get_tag(skb, &vlan_id) == 0) {
404 skb->dev = slave_dev;
405 skb = vlan_put_tag(skb, vlan_id);
407 /* vlan_put_tag() frees the skb in case of error,
408 * so return success here so the calling functions
409 * won't attempt to free is again.
414 skb->dev = slave_dev;
424 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
425 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
427 * a. This operation is performed in IOCTL context,
428 * b. The operation is protected by the RTNL semaphore in the 8021q code,
429 * c. Holding a lock with BH disabled while directly calling a base driver
430 * entry point is generally a BAD idea.
432 * The design of synchronization/protection for this operation in the 8021q
433 * module is good for one or more VLAN devices over a single physical device
434 * and cannot be extended for a teaming solution like bonding, so there is a
435 * potential race condition here where a net device from the vlan group might
436 * be referenced (either by a base driver or the 8021q code) while it is being
437 * removed from the system. However, it turns out we're not making matters
438 * worse, and if it works for regular VLAN usage it will work here too.
442 * bond_vlan_rx_register - Propagates registration to slaves
443 * @bond_dev: bonding net device that got called
444 * @grp: vlan group being registered
446 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
448 struct bonding *bond = netdev_priv(bond_dev);
454 bond_for_each_slave(bond, slave, i) {
455 struct net_device *slave_dev = slave->dev;
456 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
458 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
459 slave_ops->ndo_vlan_rx_register) {
460 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
466 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
467 * @bond_dev: bonding net device that got called
468 * @vid: vlan id being added
470 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
472 struct bonding *bond = netdev_priv(bond_dev);
476 bond_for_each_slave(bond, slave, i) {
477 struct net_device *slave_dev = slave->dev;
478 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
480 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
481 slave_ops->ndo_vlan_rx_add_vid) {
482 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
486 res = bond_add_vlan(bond, vid);
488 printk(KERN_ERR DRV_NAME
489 ": %s: Error: Failed to add vlan id %d\n",
490 bond_dev->name, vid);
495 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
496 * @bond_dev: bonding net device that got called
497 * @vid: vlan id being removed
499 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
501 struct bonding *bond = netdev_priv(bond_dev);
503 struct net_device *vlan_dev;
506 bond_for_each_slave(bond, slave, i) {
507 struct net_device *slave_dev = slave->dev;
508 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
510 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
511 slave_ops->ndo_vlan_rx_kill_vid) {
512 /* Save and then restore vlan_dev in the grp array,
513 * since the slave's driver might clear it.
515 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
516 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
517 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
521 res = bond_del_vlan(bond, vid);
523 printk(KERN_ERR DRV_NAME
524 ": %s: Error: Failed to remove vlan id %d\n",
525 bond_dev->name, vid);
529 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
531 struct vlan_entry *vlan;
532 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
534 write_lock_bh(&bond->lock);
536 if (list_empty(&bond->vlan_list))
539 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
540 slave_ops->ndo_vlan_rx_register)
541 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
543 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
544 !(slave_ops->ndo_vlan_rx_add_vid))
547 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
548 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
551 write_unlock_bh(&bond->lock);
554 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
556 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
557 struct vlan_entry *vlan;
558 struct net_device *vlan_dev;
560 write_lock_bh(&bond->lock);
562 if (list_empty(&bond->vlan_list))
565 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
566 !(slave_ops->ndo_vlan_rx_kill_vid))
569 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
570 /* Save and then restore vlan_dev in the grp array,
571 * since the slave's driver might clear it.
573 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
574 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
575 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
579 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
580 slave_ops->ndo_vlan_rx_register)
581 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
584 write_unlock_bh(&bond->lock);
587 /*------------------------------- Link status -------------------------------*/
590 * Set the carrier state for the master according to the state of its
591 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
592 * do special 802.3ad magic.
594 * Returns zero if carrier state does not change, nonzero if it does.
596 static int bond_set_carrier(struct bonding *bond)
601 if (bond->slave_cnt == 0)
604 if (bond->params.mode == BOND_MODE_8023AD)
605 return bond_3ad_set_carrier(bond);
607 bond_for_each_slave(bond, slave, i) {
608 if (slave->link == BOND_LINK_UP) {
609 if (!netif_carrier_ok(bond->dev)) {
610 netif_carrier_on(bond->dev);
618 if (netif_carrier_ok(bond->dev)) {
619 netif_carrier_off(bond->dev);
626 * Get link speed and duplex from the slave's base driver
627 * using ethtool. If for some reason the call fails or the
628 * values are invalid, fake speed and duplex to 100/Full
631 static int bond_update_speed_duplex(struct slave *slave)
633 struct net_device *slave_dev = slave->dev;
634 struct ethtool_cmd etool;
637 /* Fake speed and duplex */
638 slave->speed = SPEED_100;
639 slave->duplex = DUPLEX_FULL;
641 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
644 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
648 switch (etool.speed) {
658 switch (etool.duplex) {
666 slave->speed = etool.speed;
667 slave->duplex = etool.duplex;
673 * if <dev> supports MII link status reporting, check its link status.
675 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
676 * depening upon the setting of the use_carrier parameter.
678 * Return either BMSR_LSTATUS, meaning that the link is up (or we
679 * can't tell and just pretend it is), or 0, meaning that the link is
682 * If reporting is non-zero, instead of faking link up, return -1 if
683 * both ETHTOOL and MII ioctls fail (meaning the device does not
684 * support them). If use_carrier is set, return whatever it says.
685 * It'd be nice if there was a good way to tell if a driver supports
686 * netif_carrier, but there really isn't.
688 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
690 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
691 static int (* ioctl)(struct net_device *, struct ifreq *, int);
693 struct mii_ioctl_data *mii;
695 if (bond->params.use_carrier)
696 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
698 ioctl = slave_ops->ndo_do_ioctl;
700 /* TODO: set pointer to correct ioctl on a per team member */
701 /* bases to make this more efficient. that is, once */
702 /* we determine the correct ioctl, we will always */
703 /* call it and not the others for that team */
707 * We cannot assume that SIOCGMIIPHY will also read a
708 * register; not all network drivers (e.g., e100)
712 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
713 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
715 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
716 mii->reg_num = MII_BMSR;
717 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
718 return (mii->val_out & BMSR_LSTATUS);
724 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
725 * attempt to get link status from it if the above MII ioctls fail.
727 if (slave_dev->ethtool_ops) {
728 if (slave_dev->ethtool_ops->get_link) {
731 link = slave_dev->ethtool_ops->get_link(slave_dev);
733 return link ? BMSR_LSTATUS : 0;
738 * If reporting, report that either there's no dev->do_ioctl,
739 * or both SIOCGMIIREG and get_link failed (meaning that we
740 * cannot report link status). If not reporting, pretend
743 return (reporting ? -1 : BMSR_LSTATUS);
746 /*----------------------------- Multicast list ------------------------------*/
749 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
751 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
753 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
754 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
758 * returns dmi entry if found, NULL otherwise
760 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
762 struct dev_mc_list *idmi;
764 for (idmi = mc_list; idmi; idmi = idmi->next) {
765 if (bond_is_dmi_same(dmi, idmi)) {
774 * Push the promiscuity flag down to appropriate slaves
776 static int bond_set_promiscuity(struct bonding *bond, int inc)
779 if (USES_PRIMARY(bond->params.mode)) {
780 /* write lock already acquired */
781 if (bond->curr_active_slave) {
782 err = dev_set_promiscuity(bond->curr_active_slave->dev,
788 bond_for_each_slave(bond, slave, i) {
789 err = dev_set_promiscuity(slave->dev, inc);
798 * Push the allmulti flag down to all slaves
800 static int bond_set_allmulti(struct bonding *bond, int inc)
803 if (USES_PRIMARY(bond->params.mode)) {
804 /* write lock already acquired */
805 if (bond->curr_active_slave) {
806 err = dev_set_allmulti(bond->curr_active_slave->dev,
812 bond_for_each_slave(bond, slave, i) {
813 err = dev_set_allmulti(slave->dev, inc);
822 * Add a Multicast address to slaves
825 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
827 if (USES_PRIMARY(bond->params.mode)) {
828 /* write lock already acquired */
829 if (bond->curr_active_slave) {
830 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
835 bond_for_each_slave(bond, slave, i) {
836 dev_mc_add(slave->dev, addr, alen, 0);
842 * Remove a multicast address from slave
845 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
847 if (USES_PRIMARY(bond->params.mode)) {
848 /* write lock already acquired */
849 if (bond->curr_active_slave) {
850 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
855 bond_for_each_slave(bond, slave, i) {
856 dev_mc_delete(slave->dev, addr, alen, 0);
863 * Retrieve the list of registered multicast addresses for the bonding
864 * device and retransmit an IGMP JOIN request to the current active
867 static void bond_resend_igmp_join_requests(struct bonding *bond)
869 struct in_device *in_dev;
870 struct ip_mc_list *im;
873 in_dev = __in_dev_get_rcu(bond->dev);
875 for (im = in_dev->mc_list; im; im = im->next) {
876 ip_mc_rejoin_group(im);
884 * Totally destroys the mc_list in bond
886 static void bond_mc_list_destroy(struct bonding *bond)
888 struct dev_mc_list *dmi;
892 bond->mc_list = dmi->next;
896 bond->mc_list = NULL;
900 * Copy all the Multicast addresses from src to the bonding device dst
902 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
905 struct dev_mc_list *dmi, *new_dmi;
907 for (dmi = mc_list; dmi; dmi = dmi->next) {
908 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
911 /* FIXME: Potential memory leak !!! */
915 new_dmi->next = bond->mc_list;
916 bond->mc_list = new_dmi;
917 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
918 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
919 new_dmi->dmi_users = dmi->dmi_users;
920 new_dmi->dmi_gusers = dmi->dmi_gusers;
927 * flush all members of flush->mc_list from device dev->mc_list
929 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
931 struct bonding *bond = netdev_priv(bond_dev);
932 struct dev_mc_list *dmi;
934 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
935 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
938 if (bond->params.mode == BOND_MODE_8023AD) {
939 /* del lacpdu mc addr from mc list */
940 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
942 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
946 /*--------------------------- Active slave change ---------------------------*/
949 * Update the mc list and multicast-related flags for the new and
950 * old active slaves (if any) according to the multicast mode, and
951 * promiscuous flags unconditionally.
953 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
955 struct dev_mc_list *dmi;
957 if (!USES_PRIMARY(bond->params.mode)) {
958 /* nothing to do - mc list is already up-to-date on
965 if (bond->dev->flags & IFF_PROMISC) {
966 dev_set_promiscuity(old_active->dev, -1);
969 if (bond->dev->flags & IFF_ALLMULTI) {
970 dev_set_allmulti(old_active->dev, -1);
973 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
974 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
979 /* FIXME: Signal errors upstream. */
980 if (bond->dev->flags & IFF_PROMISC) {
981 dev_set_promiscuity(new_active->dev, 1);
984 if (bond->dev->flags & IFF_ALLMULTI) {
985 dev_set_allmulti(new_active->dev, 1);
988 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
989 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
991 bond_resend_igmp_join_requests(bond);
996 * bond_do_fail_over_mac
998 * Perform special MAC address swapping for fail_over_mac settings
1000 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1002 static void bond_do_fail_over_mac(struct bonding *bond,
1003 struct slave *new_active,
1004 struct slave *old_active)
1005 __releases(&bond->curr_slave_lock)
1006 __releases(&bond->lock)
1007 __acquires(&bond->lock)
1008 __acquires(&bond->curr_slave_lock)
1010 u8 tmp_mac[ETH_ALEN];
1011 struct sockaddr saddr;
1014 switch (bond->params.fail_over_mac) {
1015 case BOND_FOM_ACTIVE:
1017 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1018 new_active->dev->addr_len);
1020 case BOND_FOM_FOLLOW:
1022 * if new_active && old_active, swap them
1023 * if just old_active, do nothing (going to no active slave)
1024 * if just new_active, set new_active to bond's MAC
1029 write_unlock_bh(&bond->curr_slave_lock);
1030 read_unlock(&bond->lock);
1033 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1034 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1036 saddr.sa_family = new_active->dev->type;
1038 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1039 saddr.sa_family = bond->dev->type;
1042 rv = dev_set_mac_address(new_active->dev, &saddr);
1044 printk(KERN_ERR DRV_NAME
1045 ": %s: Error %d setting MAC of slave %s\n",
1046 bond->dev->name, -rv, new_active->dev->name);
1053 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1054 saddr.sa_family = old_active->dev->type;
1056 rv = dev_set_mac_address(old_active->dev, &saddr);
1058 printk(KERN_ERR DRV_NAME
1059 ": %s: Error %d setting MAC of slave %s\n",
1060 bond->dev->name, -rv, new_active->dev->name);
1062 read_lock(&bond->lock);
1063 write_lock_bh(&bond->curr_slave_lock);
1066 printk(KERN_ERR DRV_NAME
1067 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1068 bond->dev->name, bond->params.fail_over_mac);
1076 * find_best_interface - select the best available slave to be the active one
1077 * @bond: our bonding struct
1079 * Warning: Caller must hold curr_slave_lock for writing.
1081 static struct slave *bond_find_best_slave(struct bonding *bond)
1083 struct slave *new_active, *old_active;
1084 struct slave *bestslave = NULL;
1085 int mintime = bond->params.updelay;
1088 new_active = old_active = bond->curr_active_slave;
1090 if (!new_active) { /* there were no active slaves left */
1091 if (bond->slave_cnt > 0) { /* found one slave */
1092 new_active = bond->first_slave;
1094 return NULL; /* still no slave, return NULL */
1098 /* first try the primary link; if arping, a link must tx/rx traffic
1099 * before it can be considered the curr_active_slave - also, we would skip
1100 * slaves between the curr_active_slave and primary_slave that may be up
1103 if ((bond->primary_slave) &&
1104 (!bond->params.arp_interval) &&
1105 (IS_UP(bond->primary_slave->dev))) {
1106 new_active = bond->primary_slave;
1109 /* remember where to stop iterating over the slaves */
1110 old_active = new_active;
1112 bond_for_each_slave_from(bond, new_active, i, old_active) {
1113 if (IS_UP(new_active->dev)) {
1114 if (new_active->link == BOND_LINK_UP) {
1116 } else if (new_active->link == BOND_LINK_BACK) {
1117 /* link up, but waiting for stabilization */
1118 if (new_active->delay < mintime) {
1119 mintime = new_active->delay;
1120 bestslave = new_active;
1130 * change_active_interface - change the active slave into the specified one
1131 * @bond: our bonding struct
1132 * @new: the new slave to make the active one
1134 * Set the new slave to the bond's settings and unset them on the old
1135 * curr_active_slave.
1136 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1138 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1139 * because it is apparently the best available slave we have, even though its
1140 * updelay hasn't timed out yet.
1142 * If new_active is not NULL, caller must hold bond->lock for read and
1143 * curr_slave_lock for write_bh.
1145 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1147 struct slave *old_active = bond->curr_active_slave;
1149 if (old_active == new_active) {
1154 new_active->jiffies = jiffies;
1156 if (new_active->link == BOND_LINK_BACK) {
1157 if (USES_PRIMARY(bond->params.mode)) {
1158 printk(KERN_INFO DRV_NAME
1159 ": %s: making interface %s the new "
1160 "active one %d ms earlier.\n",
1161 bond->dev->name, new_active->dev->name,
1162 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1165 new_active->delay = 0;
1166 new_active->link = BOND_LINK_UP;
1168 if (bond->params.mode == BOND_MODE_8023AD) {
1169 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1172 if (bond_is_lb(bond))
1173 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1175 if (USES_PRIMARY(bond->params.mode)) {
1176 printk(KERN_INFO DRV_NAME
1177 ": %s: making interface %s the new "
1179 bond->dev->name, new_active->dev->name);
1184 if (USES_PRIMARY(bond->params.mode)) {
1185 bond_mc_swap(bond, new_active, old_active);
1188 if (bond_is_lb(bond)) {
1189 bond_alb_handle_active_change(bond, new_active);
1191 bond_set_slave_inactive_flags(old_active);
1193 bond_set_slave_active_flags(new_active);
1195 bond->curr_active_slave = new_active;
1198 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1200 bond_set_slave_inactive_flags(old_active);
1204 bond_set_slave_active_flags(new_active);
1206 if (bond->params.fail_over_mac)
1207 bond_do_fail_over_mac(bond, new_active,
1210 bond->send_grat_arp = bond->params.num_grat_arp;
1211 bond_send_gratuitous_arp(bond);
1213 bond->send_unsol_na = bond->params.num_unsol_na;
1214 bond_send_unsolicited_na(bond);
1216 write_unlock_bh(&bond->curr_slave_lock);
1217 read_unlock(&bond->lock);
1219 netdev_bonding_change(bond->dev);
1221 read_lock(&bond->lock);
1222 write_lock_bh(&bond->curr_slave_lock);
1228 * bond_select_active_slave - select a new active slave, if needed
1229 * @bond: our bonding struct
1231 * This functions shoud be called when one of the following occurs:
1232 * - The old curr_active_slave has been released or lost its link.
1233 * - The primary_slave has got its link back.
1234 * - A slave has got its link back and there's no old curr_active_slave.
1236 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1238 void bond_select_active_slave(struct bonding *bond)
1240 struct slave *best_slave;
1243 best_slave = bond_find_best_slave(bond);
1244 if (best_slave != bond->curr_active_slave) {
1245 bond_change_active_slave(bond, best_slave);
1246 rv = bond_set_carrier(bond);
1250 if (netif_carrier_ok(bond->dev)) {
1251 printk(KERN_INFO DRV_NAME
1252 ": %s: first active interface up!\n",
1255 printk(KERN_INFO DRV_NAME ": %s: "
1256 "now running without any active interface !\n",
1262 /*--------------------------- slave list handling ---------------------------*/
1265 * This function attaches the slave to the end of list.
1267 * bond->lock held for writing by caller.
1269 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1271 if (bond->first_slave == NULL) { /* attaching the first slave */
1272 new_slave->next = new_slave;
1273 new_slave->prev = new_slave;
1274 bond->first_slave = new_slave;
1276 new_slave->next = bond->first_slave;
1277 new_slave->prev = bond->first_slave->prev;
1278 new_slave->next->prev = new_slave;
1279 new_slave->prev->next = new_slave;
1286 * This function detaches the slave from the list.
1287 * WARNING: no check is made to verify if the slave effectively
1288 * belongs to <bond>.
1289 * Nothing is freed on return, structures are just unchained.
1290 * If any slave pointer in bond was pointing to <slave>,
1291 * it should be changed by the calling function.
1293 * bond->lock held for writing by caller.
1295 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1298 slave->next->prev = slave->prev;
1302 slave->prev->next = slave->next;
1305 if (bond->first_slave == slave) { /* slave is the first slave */
1306 if (bond->slave_cnt > 1) { /* there are more slave */
1307 bond->first_slave = slave->next;
1309 bond->first_slave = NULL; /* slave was the last one */
1318 /*---------------------------------- IOCTL ----------------------------------*/
1320 static int bond_sethwaddr(struct net_device *bond_dev,
1321 struct net_device *slave_dev)
1323 pr_debug("bond_dev=%p\n", bond_dev);
1324 pr_debug("slave_dev=%p\n", slave_dev);
1325 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1326 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1330 #define BOND_VLAN_FEATURES \
1331 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1332 NETIF_F_HW_VLAN_FILTER)
1335 * Compute the common dev->feature set available to all slaves. Some
1336 * feature bits are managed elsewhere, so preserve those feature bits
1337 * on the master device.
1339 static int bond_compute_features(struct bonding *bond)
1341 struct slave *slave;
1342 struct net_device *bond_dev = bond->dev;
1343 unsigned long features = bond_dev->features;
1344 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1345 bond_dev->hard_header_len);
1348 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1349 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1351 if (!bond->first_slave)
1354 features &= ~NETIF_F_ONE_FOR_ALL;
1356 bond_for_each_slave(bond, slave, i) {
1357 features = netdev_increment_features(features,
1358 slave->dev->features,
1359 NETIF_F_ONE_FOR_ALL);
1360 if (slave->dev->hard_header_len > max_hard_header_len)
1361 max_hard_header_len = slave->dev->hard_header_len;
1365 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1366 bond_dev->features = netdev_fix_features(features, NULL);
1367 bond_dev->hard_header_len = max_hard_header_len;
1372 static void bond_setup_by_slave(struct net_device *bond_dev,
1373 struct net_device *slave_dev)
1375 struct bonding *bond = netdev_priv(bond_dev);
1377 bond_dev->header_ops = slave_dev->header_ops;
1379 bond_dev->type = slave_dev->type;
1380 bond_dev->hard_header_len = slave_dev->hard_header_len;
1381 bond_dev->addr_len = slave_dev->addr_len;
1383 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1384 slave_dev->addr_len);
1385 bond->setup_by_slave = 1;
1388 /* enslave device <slave> to bond device <master> */
1389 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1391 struct bonding *bond = netdev_priv(bond_dev);
1392 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1393 struct slave *new_slave = NULL;
1394 struct dev_mc_list *dmi;
1395 struct sockaddr addr;
1397 int old_features = bond_dev->features;
1400 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1401 slave_ops->ndo_do_ioctl == NULL) {
1402 printk(KERN_WARNING DRV_NAME
1403 ": %s: Warning: no link monitoring support for %s\n",
1404 bond_dev->name, slave_dev->name);
1407 /* bond must be initialized by bond_open() before enslaving */
1408 if (!(bond_dev->flags & IFF_UP)) {
1409 printk(KERN_WARNING DRV_NAME
1410 " %s: master_dev is not up in bond_enslave\n",
1414 /* already enslaved */
1415 if (slave_dev->flags & IFF_SLAVE) {
1416 pr_debug("Error, Device was already enslaved\n");
1420 /* vlan challenged mutual exclusion */
1421 /* no need to lock since we're protected by rtnl_lock */
1422 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1423 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1424 if (!list_empty(&bond->vlan_list)) {
1425 printk(KERN_ERR DRV_NAME
1426 ": %s: Error: cannot enslave VLAN "
1427 "challenged slave %s on VLAN enabled "
1428 "bond %s\n", bond_dev->name, slave_dev->name,
1432 printk(KERN_WARNING DRV_NAME
1433 ": %s: Warning: enslaved VLAN challenged "
1434 "slave %s. Adding VLANs will be blocked as "
1435 "long as %s is part of bond %s\n",
1436 bond_dev->name, slave_dev->name, slave_dev->name,
1438 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1441 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1442 if (bond->slave_cnt == 0) {
1443 /* First slave, and it is not VLAN challenged,
1444 * so remove the block of adding VLANs over the bond.
1446 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1451 * Old ifenslave binaries are no longer supported. These can
1452 * be identified with moderate accurary by the state of the slave:
1453 * the current ifenslave will set the interface down prior to
1454 * enslaving it; the old ifenslave will not.
1456 if ((slave_dev->flags & IFF_UP)) {
1457 printk(KERN_ERR DRV_NAME ": %s is up. "
1458 "This may be due to an out of date ifenslave.\n",
1461 goto err_undo_flags;
1464 /* set bonding device ether type by slave - bonding netdevices are
1465 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1466 * there is a need to override some of the type dependent attribs/funcs.
1468 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1469 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1471 if (bond->slave_cnt == 0) {
1472 if (slave_dev->type != ARPHRD_ETHER)
1473 bond_setup_by_slave(bond_dev, slave_dev);
1474 } else if (bond_dev->type != slave_dev->type) {
1475 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1476 "from other slaves (%d), can not enslave it.\n",
1478 slave_dev->type, bond_dev->type);
1480 goto err_undo_flags;
1483 if (slave_ops->ndo_set_mac_address == NULL) {
1484 if (bond->slave_cnt == 0) {
1485 printk(KERN_WARNING DRV_NAME
1486 ": %s: Warning: The first slave device "
1487 "specified does not support setting the MAC "
1488 "address. Setting fail_over_mac to active.",
1490 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1491 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1492 printk(KERN_ERR DRV_NAME
1493 ": %s: Error: The slave device specified "
1494 "does not support setting the MAC address, "
1495 "but fail_over_mac is not set to active.\n"
1498 goto err_undo_flags;
1502 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1505 goto err_undo_flags;
1508 /* save slave's original flags before calling
1509 * netdev_set_master and dev_open
1511 new_slave->original_flags = slave_dev->flags;
1514 * Save slave's original ("permanent") mac address for modes
1515 * that need it, and for restoring it upon release, and then
1516 * set it to the master's address
1518 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1520 if (!bond->params.fail_over_mac) {
1522 * Set slave to master's mac address. The application already
1523 * set the master's mac address to that of the first slave
1525 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1526 addr.sa_family = slave_dev->type;
1527 res = dev_set_mac_address(slave_dev, &addr);
1529 pr_debug("Error %d calling set_mac_address\n", res);
1534 res = netdev_set_master(slave_dev, bond_dev);
1536 pr_debug("Error %d calling netdev_set_master\n", res);
1537 goto err_restore_mac;
1539 /* open the slave since the application closed it */
1540 res = dev_open(slave_dev);
1542 pr_debug("Openning slave %s failed\n", slave_dev->name);
1543 goto err_unset_master;
1546 new_slave->dev = slave_dev;
1547 slave_dev->priv_flags |= IFF_BONDING;
1549 if (bond_is_lb(bond)) {
1550 /* bond_alb_init_slave() must be called before all other stages since
1551 * it might fail and we do not want to have to undo everything
1553 res = bond_alb_init_slave(bond, new_slave);
1559 /* If the mode USES_PRIMARY, then the new slave gets the
1560 * master's promisc (and mc) settings only if it becomes the
1561 * curr_active_slave, and that is taken care of later when calling
1562 * bond_change_active()
1564 if (!USES_PRIMARY(bond->params.mode)) {
1565 /* set promiscuity level to new slave */
1566 if (bond_dev->flags & IFF_PROMISC) {
1567 res = dev_set_promiscuity(slave_dev, 1);
1572 /* set allmulti level to new slave */
1573 if (bond_dev->flags & IFF_ALLMULTI) {
1574 res = dev_set_allmulti(slave_dev, 1);
1579 netif_addr_lock_bh(bond_dev);
1580 /* upload master's mc_list to new slave */
1581 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1582 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1584 netif_addr_unlock_bh(bond_dev);
1587 if (bond->params.mode == BOND_MODE_8023AD) {
1588 /* add lacpdu mc addr to mc list */
1589 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1591 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1594 bond_add_vlans_on_slave(bond, slave_dev);
1596 write_lock_bh(&bond->lock);
1598 bond_attach_slave(bond, new_slave);
1600 new_slave->delay = 0;
1601 new_slave->link_failure_count = 0;
1603 bond_compute_features(bond);
1605 write_unlock_bh(&bond->lock);
1607 read_lock(&bond->lock);
1609 new_slave->last_arp_rx = jiffies;
1611 if (bond->params.miimon && !bond->params.use_carrier) {
1612 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1614 if ((link_reporting == -1) && !bond->params.arp_interval) {
1616 * miimon is set but a bonded network driver
1617 * does not support ETHTOOL/MII and
1618 * arp_interval is not set. Note: if
1619 * use_carrier is enabled, we will never go
1620 * here (because netif_carrier is always
1621 * supported); thus, we don't need to change
1622 * the messages for netif_carrier.
1624 printk(KERN_WARNING DRV_NAME
1625 ": %s: Warning: MII and ETHTOOL support not "
1626 "available for interface %s, and "
1627 "arp_interval/arp_ip_target module parameters "
1628 "not specified, thus bonding will not detect "
1629 "link failures! see bonding.txt for details.\n",
1630 bond_dev->name, slave_dev->name);
1631 } else if (link_reporting == -1) {
1632 /* unable get link status using mii/ethtool */
1633 printk(KERN_WARNING DRV_NAME
1634 ": %s: Warning: can't get link status from "
1635 "interface %s; the network driver associated "
1636 "with this interface does not support MII or "
1637 "ETHTOOL link status reporting, thus miimon "
1638 "has no effect on this interface.\n",
1639 bond_dev->name, slave_dev->name);
1643 /* check for initial state */
1644 if (!bond->params.miimon ||
1645 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1646 if (bond->params.updelay) {
1647 pr_debug("Initial state of slave_dev is "
1648 "BOND_LINK_BACK\n");
1649 new_slave->link = BOND_LINK_BACK;
1650 new_slave->delay = bond->params.updelay;
1652 pr_debug("Initial state of slave_dev is "
1654 new_slave->link = BOND_LINK_UP;
1656 new_slave->jiffies = jiffies;
1658 pr_debug("Initial state of slave_dev is "
1659 "BOND_LINK_DOWN\n");
1660 new_slave->link = BOND_LINK_DOWN;
1663 if (bond_update_speed_duplex(new_slave) &&
1664 (new_slave->link != BOND_LINK_DOWN)) {
1665 printk(KERN_WARNING DRV_NAME
1666 ": %s: Warning: failed to get speed and duplex from %s, "
1667 "assumed to be 100Mb/sec and Full.\n",
1668 bond_dev->name, new_slave->dev->name);
1670 if (bond->params.mode == BOND_MODE_8023AD) {
1671 printk(KERN_WARNING DRV_NAME
1672 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1673 "support in base driver for proper aggregator "
1674 "selection.\n", bond_dev->name);
1678 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1679 /* if there is a primary slave, remember it */
1680 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1681 bond->primary_slave = new_slave;
1685 write_lock_bh(&bond->curr_slave_lock);
1687 switch (bond->params.mode) {
1688 case BOND_MODE_ACTIVEBACKUP:
1689 bond_set_slave_inactive_flags(new_slave);
1690 bond_select_active_slave(bond);
1692 case BOND_MODE_8023AD:
1693 /* in 802.3ad mode, the internal mechanism
1694 * will activate the slaves in the selected
1697 bond_set_slave_inactive_flags(new_slave);
1698 /* if this is the first slave */
1699 if (bond->slave_cnt == 1) {
1700 SLAVE_AD_INFO(new_slave).id = 1;
1701 /* Initialize AD with the number of times that the AD timer is called in 1 second
1702 * can be called only after the mac address of the bond is set
1704 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1705 bond->params.lacp_fast);
1707 SLAVE_AD_INFO(new_slave).id =
1708 SLAVE_AD_INFO(new_slave->prev).id + 1;
1711 bond_3ad_bind_slave(new_slave);
1715 new_slave->state = BOND_STATE_ACTIVE;
1716 bond_set_slave_inactive_flags(new_slave);
1719 pr_debug("This slave is always active in trunk mode\n");
1721 /* always active in trunk mode */
1722 new_slave->state = BOND_STATE_ACTIVE;
1724 /* In trunking mode there is little meaning to curr_active_slave
1725 * anyway (it holds no special properties of the bond device),
1726 * so we can change it without calling change_active_interface()
1728 if (!bond->curr_active_slave) {
1729 bond->curr_active_slave = new_slave;
1732 } /* switch(bond_mode) */
1734 write_unlock_bh(&bond->curr_slave_lock);
1736 bond_set_carrier(bond);
1738 read_unlock(&bond->lock);
1740 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1744 printk(KERN_INFO DRV_NAME
1745 ": %s: enslaving %s as a%s interface with a%s link.\n",
1746 bond_dev->name, slave_dev->name,
1747 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1748 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1750 /* enslave is successful */
1753 /* Undo stages on error */
1755 dev_close(slave_dev);
1758 netdev_set_master(slave_dev, NULL);
1761 if (!bond->params.fail_over_mac) {
1762 /* XXX TODO - fom follow mode needs to change master's
1763 * MAC if this slave's MAC is in use by the bond, or at
1764 * least print a warning.
1766 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1767 addr.sa_family = slave_dev->type;
1768 dev_set_mac_address(slave_dev, &addr);
1775 bond_dev->features = old_features;
1781 * Try to release the slave device <slave> from the bond device <master>
1782 * It is legal to access curr_active_slave without a lock because all the function
1785 * The rules for slave state should be:
1786 * for Active/Backup:
1787 * Active stays on all backups go down
1788 * for Bonded connections:
1789 * The first up interface should be left on and all others downed.
1791 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1793 struct bonding *bond = netdev_priv(bond_dev);
1794 struct slave *slave, *oldcurrent;
1795 struct sockaddr addr;
1796 int mac_addr_differ;
1798 /* slave is not a slave or master is not master of this slave */
1799 if (!(slave_dev->flags & IFF_SLAVE) ||
1800 (slave_dev->master != bond_dev)) {
1801 printk(KERN_ERR DRV_NAME
1802 ": %s: Error: cannot release %s.\n",
1803 bond_dev->name, slave_dev->name);
1807 write_lock_bh(&bond->lock);
1809 slave = bond_get_slave_by_dev(bond, slave_dev);
1811 /* not a slave of this bond */
1812 printk(KERN_INFO DRV_NAME
1813 ": %s: %s not enslaved\n",
1814 bond_dev->name, slave_dev->name);
1815 write_unlock_bh(&bond->lock);
1819 if (!bond->params.fail_over_mac) {
1820 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1822 if (!mac_addr_differ && (bond->slave_cnt > 1))
1823 printk(KERN_WARNING DRV_NAME
1824 ": %s: Warning: the permanent HWaddr of %s - "
1825 "%pM - is still in use by %s. "
1826 "Set the HWaddr of %s to a different address "
1827 "to avoid conflicts.\n",
1828 bond_dev->name, slave_dev->name,
1830 bond_dev->name, slave_dev->name);
1833 /* Inform AD package of unbinding of slave. */
1834 if (bond->params.mode == BOND_MODE_8023AD) {
1835 /* must be called before the slave is
1836 * detached from the list
1838 bond_3ad_unbind_slave(slave);
1841 printk(KERN_INFO DRV_NAME
1842 ": %s: releasing %s interface %s\n",
1844 (slave->state == BOND_STATE_ACTIVE)
1845 ? "active" : "backup",
1848 oldcurrent = bond->curr_active_slave;
1850 bond->current_arp_slave = NULL;
1852 /* release the slave from its bond */
1853 bond_detach_slave(bond, slave);
1855 bond_compute_features(bond);
1857 if (bond->primary_slave == slave) {
1858 bond->primary_slave = NULL;
1861 if (oldcurrent == slave) {
1862 bond_change_active_slave(bond, NULL);
1865 if (bond_is_lb(bond)) {
1866 /* Must be called only after the slave has been
1867 * detached from the list and the curr_active_slave
1868 * has been cleared (if our_slave == old_current),
1869 * but before a new active slave is selected.
1871 write_unlock_bh(&bond->lock);
1872 bond_alb_deinit_slave(bond, slave);
1873 write_lock_bh(&bond->lock);
1876 if (oldcurrent == slave) {
1878 * Note that we hold RTNL over this sequence, so there
1879 * is no concern that another slave add/remove event
1882 write_unlock_bh(&bond->lock);
1883 read_lock(&bond->lock);
1884 write_lock_bh(&bond->curr_slave_lock);
1886 bond_select_active_slave(bond);
1888 write_unlock_bh(&bond->curr_slave_lock);
1889 read_unlock(&bond->lock);
1890 write_lock_bh(&bond->lock);
1893 if (bond->slave_cnt == 0) {
1894 bond_set_carrier(bond);
1896 /* if the last slave was removed, zero the mac address
1897 * of the master so it will be set by the application
1898 * to the mac address of the first slave
1900 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1902 if (list_empty(&bond->vlan_list)) {
1903 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1905 printk(KERN_WARNING DRV_NAME
1906 ": %s: Warning: clearing HW address of %s while it "
1907 "still has VLANs.\n",
1908 bond_dev->name, bond_dev->name);
1909 printk(KERN_WARNING DRV_NAME
1910 ": %s: When re-adding slaves, make sure the bond's "
1911 "HW address matches its VLANs'.\n",
1914 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1915 !bond_has_challenged_slaves(bond)) {
1916 printk(KERN_INFO DRV_NAME
1917 ": %s: last VLAN challenged slave %s "
1918 "left bond %s. VLAN blocking is removed\n",
1919 bond_dev->name, slave_dev->name, bond_dev->name);
1920 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1923 write_unlock_bh(&bond->lock);
1925 /* must do this from outside any spinlocks */
1926 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1928 bond_del_vlans_from_slave(bond, slave_dev);
1930 /* If the mode USES_PRIMARY, then we should only remove its
1931 * promisc and mc settings if it was the curr_active_slave, but that was
1932 * already taken care of above when we detached the slave
1934 if (!USES_PRIMARY(bond->params.mode)) {
1935 /* unset promiscuity level from slave */
1936 if (bond_dev->flags & IFF_PROMISC) {
1937 dev_set_promiscuity(slave_dev, -1);
1940 /* unset allmulti level from slave */
1941 if (bond_dev->flags & IFF_ALLMULTI) {
1942 dev_set_allmulti(slave_dev, -1);
1945 /* flush master's mc_list from slave */
1946 netif_addr_lock_bh(bond_dev);
1947 bond_mc_list_flush(bond_dev, slave_dev);
1948 netif_addr_unlock_bh(bond_dev);
1951 netdev_set_master(slave_dev, NULL);
1953 /* close slave before restoring its mac address */
1954 dev_close(slave_dev);
1956 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1957 /* restore original ("permanent") mac address */
1958 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1959 addr.sa_family = slave_dev->type;
1960 dev_set_mac_address(slave_dev, &addr);
1963 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1964 IFF_SLAVE_INACTIVE | IFF_BONDING |
1969 return 0; /* deletion OK */
1973 * Destroy a bonding device.
1974 * Must be under rtnl_lock when this function is called.
1976 void bond_destroy(struct bonding *bond)
1978 bond_deinit(bond->dev);
1979 bond_destroy_sysfs_entry(bond);
1980 unregister_netdevice(bond->dev);
1983 static void bond_destructor(struct net_device *bond_dev)
1985 struct bonding *bond = netdev_priv(bond_dev);
1988 destroy_workqueue(bond->wq);
1990 netif_addr_lock_bh(bond_dev);
1991 bond_mc_list_destroy(bond);
1992 netif_addr_unlock_bh(bond_dev);
1994 free_netdev(bond_dev);
1998 * First release a slave and than destroy the bond if no more slaves iare left.
1999 * Must be under rtnl_lock when this function is called.
2001 int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2003 struct bonding *bond = netdev_priv(bond_dev);
2006 ret = bond_release(bond_dev, slave_dev);
2007 if ((ret == 0) && (bond->slave_cnt == 0)) {
2008 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2009 bond_dev->name, bond_dev->name);
2016 * This function releases all slaves.
2018 static int bond_release_all(struct net_device *bond_dev)
2020 struct bonding *bond = netdev_priv(bond_dev);
2021 struct slave *slave;
2022 struct net_device *slave_dev;
2023 struct sockaddr addr;
2025 write_lock_bh(&bond->lock);
2027 netif_carrier_off(bond_dev);
2029 if (bond->slave_cnt == 0) {
2033 bond->current_arp_slave = NULL;
2034 bond->primary_slave = NULL;
2035 bond_change_active_slave(bond, NULL);
2037 while ((slave = bond->first_slave) != NULL) {
2038 /* Inform AD package of unbinding of slave
2039 * before slave is detached from the list.
2041 if (bond->params.mode == BOND_MODE_8023AD) {
2042 bond_3ad_unbind_slave(slave);
2045 slave_dev = slave->dev;
2046 bond_detach_slave(bond, slave);
2048 /* now that the slave is detached, unlock and perform
2049 * all the undo steps that should not be called from
2052 write_unlock_bh(&bond->lock);
2054 if (bond_is_lb(bond)) {
2055 /* must be called only after the slave
2056 * has been detached from the list
2058 bond_alb_deinit_slave(bond, slave);
2061 bond_compute_features(bond);
2063 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2064 bond_del_vlans_from_slave(bond, slave_dev);
2066 /* If the mode USES_PRIMARY, then we should only remove its
2067 * promisc and mc settings if it was the curr_active_slave, but that was
2068 * already taken care of above when we detached the slave
2070 if (!USES_PRIMARY(bond->params.mode)) {
2071 /* unset promiscuity level from slave */
2072 if (bond_dev->flags & IFF_PROMISC) {
2073 dev_set_promiscuity(slave_dev, -1);
2076 /* unset allmulti level from slave */
2077 if (bond_dev->flags & IFF_ALLMULTI) {
2078 dev_set_allmulti(slave_dev, -1);
2081 /* flush master's mc_list from slave */
2082 netif_addr_lock_bh(bond_dev);
2083 bond_mc_list_flush(bond_dev, slave_dev);
2084 netif_addr_unlock_bh(bond_dev);
2087 netdev_set_master(slave_dev, NULL);
2089 /* close slave before restoring its mac address */
2090 dev_close(slave_dev);
2092 if (!bond->params.fail_over_mac) {
2093 /* restore original ("permanent") mac address*/
2094 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2095 addr.sa_family = slave_dev->type;
2096 dev_set_mac_address(slave_dev, &addr);
2099 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2100 IFF_SLAVE_INACTIVE);
2104 /* re-acquire the lock before getting the next slave */
2105 write_lock_bh(&bond->lock);
2108 /* zero the mac address of the master so it will be
2109 * set by the application to the mac address of the
2112 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2114 if (list_empty(&bond->vlan_list)) {
2115 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2117 printk(KERN_WARNING DRV_NAME
2118 ": %s: Warning: clearing HW address of %s while it "
2119 "still has VLANs.\n",
2120 bond_dev->name, bond_dev->name);
2121 printk(KERN_WARNING DRV_NAME
2122 ": %s: When re-adding slaves, make sure the bond's "
2123 "HW address matches its VLANs'.\n",
2127 printk(KERN_INFO DRV_NAME
2128 ": %s: released all slaves\n",
2132 write_unlock_bh(&bond->lock);
2138 * This function changes the active slave to slave <slave_dev>.
2139 * It returns -EINVAL in the following cases.
2140 * - <slave_dev> is not found in the list.
2141 * - There is not active slave now.
2142 * - <slave_dev> is already active.
2143 * - The link state of <slave_dev> is not BOND_LINK_UP.
2144 * - <slave_dev> is not running.
2145 * In these cases, this fuction does nothing.
2146 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2148 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2150 struct bonding *bond = netdev_priv(bond_dev);
2151 struct slave *old_active = NULL;
2152 struct slave *new_active = NULL;
2155 if (!USES_PRIMARY(bond->params.mode)) {
2159 /* Verify that master_dev is indeed the master of slave_dev */
2160 if (!(slave_dev->flags & IFF_SLAVE) ||
2161 (slave_dev->master != bond_dev)) {
2165 read_lock(&bond->lock);
2167 read_lock(&bond->curr_slave_lock);
2168 old_active = bond->curr_active_slave;
2169 read_unlock(&bond->curr_slave_lock);
2171 new_active = bond_get_slave_by_dev(bond, slave_dev);
2174 * Changing to the current active: do nothing; return success.
2176 if (new_active && (new_active == old_active)) {
2177 read_unlock(&bond->lock);
2183 (new_active->link == BOND_LINK_UP) &&
2184 IS_UP(new_active->dev)) {
2185 write_lock_bh(&bond->curr_slave_lock);
2186 bond_change_active_slave(bond, new_active);
2187 write_unlock_bh(&bond->curr_slave_lock);
2192 read_unlock(&bond->lock);
2197 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2199 struct bonding *bond = netdev_priv(bond_dev);
2201 info->bond_mode = bond->params.mode;
2202 info->miimon = bond->params.miimon;
2204 read_lock(&bond->lock);
2205 info->num_slaves = bond->slave_cnt;
2206 read_unlock(&bond->lock);
2211 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2213 struct bonding *bond = netdev_priv(bond_dev);
2214 struct slave *slave;
2217 if (info->slave_id < 0) {
2221 read_lock(&bond->lock);
2223 bond_for_each_slave(bond, slave, i) {
2224 if (i == (int)info->slave_id) {
2230 read_unlock(&bond->lock);
2233 strcpy(info->slave_name, slave->dev->name);
2234 info->link = slave->link;
2235 info->state = slave->state;
2236 info->link_failure_count = slave->link_failure_count;
2244 /*-------------------------------- Monitoring -------------------------------*/
2247 static int bond_miimon_inspect(struct bonding *bond)
2249 struct slave *slave;
2250 int i, link_state, commit = 0;
2252 bond_for_each_slave(bond, slave, i) {
2253 slave->new_link = BOND_LINK_NOCHANGE;
2255 link_state = bond_check_dev_link(bond, slave->dev, 0);
2257 switch (slave->link) {
2262 slave->link = BOND_LINK_FAIL;
2263 slave->delay = bond->params.downdelay;
2265 printk(KERN_INFO DRV_NAME
2266 ": %s: link status down for %s"
2267 "interface %s, disabling it in %d ms.\n",
2269 (bond->params.mode ==
2270 BOND_MODE_ACTIVEBACKUP) ?
2271 ((slave->state == BOND_STATE_ACTIVE) ?
2272 "active " : "backup ") : "",
2274 bond->params.downdelay * bond->params.miimon);
2277 case BOND_LINK_FAIL:
2280 * recovered before downdelay expired
2282 slave->link = BOND_LINK_UP;
2283 slave->jiffies = jiffies;
2284 printk(KERN_INFO DRV_NAME
2285 ": %s: link status up again after %d "
2286 "ms for interface %s.\n",
2288 (bond->params.downdelay - slave->delay) *
2289 bond->params.miimon,
2294 if (slave->delay <= 0) {
2295 slave->new_link = BOND_LINK_DOWN;
2303 case BOND_LINK_DOWN:
2307 slave->link = BOND_LINK_BACK;
2308 slave->delay = bond->params.updelay;
2311 printk(KERN_INFO DRV_NAME
2312 ": %s: link status up for "
2313 "interface %s, enabling it in %d ms.\n",
2314 bond->dev->name, slave->dev->name,
2315 bond->params.updelay *
2316 bond->params.miimon);
2319 case BOND_LINK_BACK:
2321 slave->link = BOND_LINK_DOWN;
2322 printk(KERN_INFO DRV_NAME
2323 ": %s: link status down again after %d "
2324 "ms for interface %s.\n",
2326 (bond->params.updelay - slave->delay) *
2327 bond->params.miimon,
2333 if (slave->delay <= 0) {
2334 slave->new_link = BOND_LINK_UP;
2347 static void bond_miimon_commit(struct bonding *bond)
2349 struct slave *slave;
2352 bond_for_each_slave(bond, slave, i) {
2353 switch (slave->new_link) {
2354 case BOND_LINK_NOCHANGE:
2358 slave->link = BOND_LINK_UP;
2359 slave->jiffies = jiffies;
2361 if (bond->params.mode == BOND_MODE_8023AD) {
2362 /* prevent it from being the active one */
2363 slave->state = BOND_STATE_BACKUP;
2364 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2365 /* make it immediately active */
2366 slave->state = BOND_STATE_ACTIVE;
2367 } else if (slave != bond->primary_slave) {
2368 /* prevent it from being the active one */
2369 slave->state = BOND_STATE_BACKUP;
2372 printk(KERN_INFO DRV_NAME
2373 ": %s: link status definitely "
2374 "up for interface %s.\n",
2375 bond->dev->name, slave->dev->name);
2377 /* notify ad that the link status has changed */
2378 if (bond->params.mode == BOND_MODE_8023AD)
2379 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2381 if (bond_is_lb(bond))
2382 bond_alb_handle_link_change(bond, slave,
2385 if (!bond->curr_active_slave ||
2386 (slave == bond->primary_slave))
2391 case BOND_LINK_DOWN:
2392 if (slave->link_failure_count < UINT_MAX)
2393 slave->link_failure_count++;
2395 slave->link = BOND_LINK_DOWN;
2397 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2398 bond->params.mode == BOND_MODE_8023AD)
2399 bond_set_slave_inactive_flags(slave);
2401 printk(KERN_INFO DRV_NAME
2402 ": %s: link status definitely down for "
2403 "interface %s, disabling it\n",
2404 bond->dev->name, slave->dev->name);
2406 if (bond->params.mode == BOND_MODE_8023AD)
2407 bond_3ad_handle_link_change(slave,
2410 if (bond->params.mode == BOND_MODE_TLB ||
2411 bond->params.mode == BOND_MODE_ALB)
2412 bond_alb_handle_link_change(bond, slave,
2415 if (slave == bond->curr_active_slave)
2421 printk(KERN_ERR DRV_NAME
2422 ": %s: invalid new link %d on slave %s\n",
2423 bond->dev->name, slave->new_link,
2425 slave->new_link = BOND_LINK_NOCHANGE;
2432 write_lock_bh(&bond->curr_slave_lock);
2433 bond_select_active_slave(bond);
2434 write_unlock_bh(&bond->curr_slave_lock);
2437 bond_set_carrier(bond);
2443 * Really a wrapper that splits the mii monitor into two phases: an
2444 * inspection, then (if inspection indicates something needs to be done)
2445 * an acquisition of appropriate locks followed by a commit phase to
2446 * implement whatever link state changes are indicated.
2448 void bond_mii_monitor(struct work_struct *work)
2450 struct bonding *bond = container_of(work, struct bonding,
2453 read_lock(&bond->lock);
2454 if (bond->kill_timers)
2457 if (bond->slave_cnt == 0)
2460 if (bond->send_grat_arp) {
2461 read_lock(&bond->curr_slave_lock);
2462 bond_send_gratuitous_arp(bond);
2463 read_unlock(&bond->curr_slave_lock);
2466 if (bond->send_unsol_na) {
2467 read_lock(&bond->curr_slave_lock);
2468 bond_send_unsolicited_na(bond);
2469 read_unlock(&bond->curr_slave_lock);
2472 if (bond_miimon_inspect(bond)) {
2473 read_unlock(&bond->lock);
2475 read_lock(&bond->lock);
2477 bond_miimon_commit(bond);
2479 read_unlock(&bond->lock);
2480 rtnl_unlock(); /* might sleep, hold no other locks */
2481 read_lock(&bond->lock);
2485 if (bond->params.miimon)
2486 queue_delayed_work(bond->wq, &bond->mii_work,
2487 msecs_to_jiffies(bond->params.miimon));
2489 read_unlock(&bond->lock);
2492 static __be32 bond_glean_dev_ip(struct net_device *dev)
2494 struct in_device *idev;
2495 struct in_ifaddr *ifa;
2502 idev = __in_dev_get_rcu(dev);
2506 ifa = idev->ifa_list;
2510 addr = ifa->ifa_local;
2516 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2518 struct vlan_entry *vlan;
2520 if (ip == bond->master_ip)
2523 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2524 if (ip == vlan->vlan_ip)
2532 * We go to the (large) trouble of VLAN tagging ARP frames because
2533 * switches in VLAN mode (especially if ports are configured as
2534 * "native" to a VLAN) might not pass non-tagged frames.
2536 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2538 struct sk_buff *skb;
2540 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2541 slave_dev->name, dest_ip, src_ip, vlan_id);
2543 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2544 NULL, slave_dev->dev_addr, NULL);
2547 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2551 skb = vlan_put_tag(skb, vlan_id);
2553 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2561 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2564 __be32 *targets = bond->params.arp_targets;
2565 struct vlan_entry *vlan;
2566 struct net_device *vlan_dev;
2570 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2573 pr_debug("basa: target %x\n", targets[i]);
2574 if (list_empty(&bond->vlan_list)) {
2575 pr_debug("basa: empty vlan: arp_send\n");
2576 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2577 bond->master_ip, 0);
2582 * If VLANs are configured, we do a route lookup to
2583 * determine which VLAN interface would be used, so we
2584 * can tag the ARP with the proper VLAN tag.
2586 memset(&fl, 0, sizeof(fl));
2587 fl.fl4_dst = targets[i];
2588 fl.fl4_tos = RTO_ONLINK;
2590 rv = ip_route_output_key(&init_net, &rt, &fl);
2592 if (net_ratelimit()) {
2593 printk(KERN_WARNING DRV_NAME
2594 ": %s: no route to arp_ip_target %pI4\n",
2595 bond->dev->name, &fl.fl4_dst);
2601 * This target is not on a VLAN
2603 if (rt->u.dst.dev == bond->dev) {
2605 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2606 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2607 bond->master_ip, 0);
2612 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2613 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2614 if (vlan_dev == rt->u.dst.dev) {
2615 vlan_id = vlan->vlan_id;
2616 pr_debug("basa: vlan match on %s %d\n",
2617 vlan_dev->name, vlan_id);
2624 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2625 vlan->vlan_ip, vlan_id);
2629 if (net_ratelimit()) {
2630 printk(KERN_WARNING DRV_NAME
2631 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2632 bond->dev->name, &fl.fl4_dst,
2633 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2640 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2641 * for each VLAN above us.
2643 * Caller must hold curr_slave_lock for read or better
2645 static void bond_send_gratuitous_arp(struct bonding *bond)
2647 struct slave *slave = bond->curr_active_slave;
2648 struct vlan_entry *vlan;
2649 struct net_device *vlan_dev;
2651 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2652 slave ? slave->dev->name : "NULL");
2654 if (!slave || !bond->send_grat_arp ||
2655 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2658 bond->send_grat_arp--;
2660 if (bond->master_ip) {
2661 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2662 bond->master_ip, 0);
2665 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2666 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2667 if (vlan->vlan_ip) {
2668 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2669 vlan->vlan_ip, vlan->vlan_id);
2674 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2677 __be32 *targets = bond->params.arp_targets;
2679 targets = bond->params.arp_targets;
2680 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2681 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2682 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2683 if (sip == targets[i]) {
2684 if (bond_has_this_ip(bond, tip))
2685 slave->last_arp_rx = jiffies;
2691 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2694 struct slave *slave;
2695 struct bonding *bond;
2696 unsigned char *arp_ptr;
2699 if (dev_net(dev) != &init_net)
2702 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2705 bond = netdev_priv(dev);
2706 read_lock(&bond->lock);
2708 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2709 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2710 orig_dev ? orig_dev->name : "NULL");
2712 slave = bond_get_slave_by_dev(bond, orig_dev);
2713 if (!slave || !slave_do_arp_validate(bond, slave))
2716 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2720 if (arp->ar_hln != dev->addr_len ||
2721 skb->pkt_type == PACKET_OTHERHOST ||
2722 skb->pkt_type == PACKET_LOOPBACK ||
2723 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2724 arp->ar_pro != htons(ETH_P_IP) ||
2728 arp_ptr = (unsigned char *)(arp + 1);
2729 arp_ptr += dev->addr_len;
2730 memcpy(&sip, arp_ptr, 4);
2731 arp_ptr += 4 + dev->addr_len;
2732 memcpy(&tip, arp_ptr, 4);
2734 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2735 bond->dev->name, slave->dev->name, slave->state,
2736 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2740 * Backup slaves won't see the ARP reply, but do come through
2741 * here for each ARP probe (so we swap the sip/tip to validate
2742 * the probe). In a "redundant switch, common router" type of
2743 * configuration, the ARP probe will (hopefully) travel from
2744 * the active, through one switch, the router, then the other
2745 * switch before reaching the backup.
2747 if (slave->state == BOND_STATE_ACTIVE)
2748 bond_validate_arp(bond, slave, sip, tip);
2750 bond_validate_arp(bond, slave, tip, sip);
2753 read_unlock(&bond->lock);
2756 return NET_RX_SUCCESS;
2760 * this function is called regularly to monitor each slave's link
2761 * ensuring that traffic is being sent and received when arp monitoring
2762 * is used in load-balancing mode. if the adapter has been dormant, then an
2763 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2764 * arp monitoring in active backup mode.
2766 void bond_loadbalance_arp_mon(struct work_struct *work)
2768 struct bonding *bond = container_of(work, struct bonding,
2770 struct slave *slave, *oldcurrent;
2771 int do_failover = 0;
2775 read_lock(&bond->lock);
2777 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2779 if (bond->kill_timers) {
2783 if (bond->slave_cnt == 0) {
2787 read_lock(&bond->curr_slave_lock);
2788 oldcurrent = bond->curr_active_slave;
2789 read_unlock(&bond->curr_slave_lock);
2791 /* see if any of the previous devices are up now (i.e. they have
2792 * xmt and rcv traffic). the curr_active_slave does not come into
2793 * the picture unless it is null. also, slave->jiffies is not needed
2794 * here because we send an arp on each slave and give a slave as
2795 * long as it needs to get the tx/rx within the delta.
2796 * TODO: what about up/down delay in arp mode? it wasn't here before
2799 bond_for_each_slave(bond, slave, i) {
2800 if (slave->link != BOND_LINK_UP) {
2801 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2802 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2804 slave->link = BOND_LINK_UP;
2805 slave->state = BOND_STATE_ACTIVE;
2807 /* primary_slave has no meaning in round-robin
2808 * mode. the window of a slave being up and
2809 * curr_active_slave being null after enslaving
2813 printk(KERN_INFO DRV_NAME
2814 ": %s: link status definitely "
2815 "up for interface %s, ",
2820 printk(KERN_INFO DRV_NAME
2821 ": %s: interface %s is now up\n",
2827 /* slave->link == BOND_LINK_UP */
2829 /* not all switches will respond to an arp request
2830 * when the source ip is 0, so don't take the link down
2831 * if we don't know our ip yet
2833 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2834 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2836 slave->link = BOND_LINK_DOWN;
2837 slave->state = BOND_STATE_BACKUP;
2839 if (slave->link_failure_count < UINT_MAX) {
2840 slave->link_failure_count++;
2843 printk(KERN_INFO DRV_NAME
2844 ": %s: interface %s is now down.\n",
2848 if (slave == oldcurrent) {
2854 /* note: if switch is in round-robin mode, all links
2855 * must tx arp to ensure all links rx an arp - otherwise
2856 * links may oscillate or not come up at all; if switch is
2857 * in something like xor mode, there is nothing we can
2858 * do - all replies will be rx'ed on same link causing slaves
2859 * to be unstable during low/no traffic periods
2861 if (IS_UP(slave->dev)) {
2862 bond_arp_send_all(bond, slave);
2867 write_lock_bh(&bond->curr_slave_lock);
2869 bond_select_active_slave(bond);
2871 write_unlock_bh(&bond->curr_slave_lock);
2875 if (bond->params.arp_interval)
2876 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2878 read_unlock(&bond->lock);
2882 * Called to inspect slaves for active-backup mode ARP monitor link state
2883 * changes. Sets new_link in slaves to specify what action should take
2884 * place for the slave. Returns 0 if no changes are found, >0 if changes
2885 * to link states must be committed.
2887 * Called with bond->lock held for read.
2889 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2891 struct slave *slave;
2894 bond_for_each_slave(bond, slave, i) {
2895 slave->new_link = BOND_LINK_NOCHANGE;
2897 if (slave->link != BOND_LINK_UP) {
2898 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2900 slave->new_link = BOND_LINK_UP;
2908 * Give slaves 2*delta after being enslaved or made
2909 * active. This avoids bouncing, as the last receive
2910 * times need a full ARP monitor cycle to be updated.
2912 if (!time_after_eq(jiffies, slave->jiffies +
2913 2 * delta_in_ticks))
2917 * Backup slave is down if:
2918 * - No current_arp_slave AND
2919 * - more than 3*delta since last receive AND
2920 * - the bond has an IP address
2922 * Note: a non-null current_arp_slave indicates
2923 * the curr_active_slave went down and we are
2924 * searching for a new one; under this condition
2925 * we only take the curr_active_slave down - this
2926 * gives each slave a chance to tx/rx traffic
2927 * before being taken out
2929 if (slave->state == BOND_STATE_BACKUP &&
2930 !bond->current_arp_slave &&
2931 time_after(jiffies, slave_last_rx(bond, slave) +
2932 3 * delta_in_ticks)) {
2933 slave->new_link = BOND_LINK_DOWN;
2938 * Active slave is down if:
2939 * - more than 2*delta since transmitting OR
2940 * - (more than 2*delta since receive AND
2941 * the bond has an IP address)
2943 if ((slave->state == BOND_STATE_ACTIVE) &&
2944 (time_after_eq(jiffies, slave->dev->trans_start +
2945 2 * delta_in_ticks) ||
2946 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2947 + 2 * delta_in_ticks)))) {
2948 slave->new_link = BOND_LINK_DOWN;
2953 read_lock(&bond->curr_slave_lock);
2956 * Trigger a commit if the primary option setting has changed.
2958 if (bond->primary_slave &&
2959 (bond->primary_slave != bond->curr_active_slave) &&
2960 (bond->primary_slave->link == BOND_LINK_UP))
2963 read_unlock(&bond->curr_slave_lock);
2969 * Called to commit link state changes noted by inspection step of
2970 * active-backup mode ARP monitor.
2972 * Called with RTNL and bond->lock for read.
2974 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2976 struct slave *slave;
2979 bond_for_each_slave(bond, slave, i) {
2980 switch (slave->new_link) {
2981 case BOND_LINK_NOCHANGE:
2985 write_lock_bh(&bond->curr_slave_lock);
2987 if (!bond->curr_active_slave &&
2988 time_before_eq(jiffies, slave->dev->trans_start +
2990 slave->link = BOND_LINK_UP;
2991 bond_change_active_slave(bond, slave);
2992 bond->current_arp_slave = NULL;
2994 printk(KERN_INFO DRV_NAME
2995 ": %s: %s is up and now the "
2996 "active interface\n",
2997 bond->dev->name, slave->dev->name);
2999 } else if (bond->curr_active_slave != slave) {
3000 /* this slave has just come up but we
3001 * already have a current slave; this can
3002 * also happen if bond_enslave adds a new
3003 * slave that is up while we are searching
3006 slave->link = BOND_LINK_UP;
3007 bond_set_slave_inactive_flags(slave);
3008 bond->current_arp_slave = NULL;
3010 printk(KERN_INFO DRV_NAME
3011 ": %s: backup interface %s is now up\n",
3012 bond->dev->name, slave->dev->name);
3015 write_unlock_bh(&bond->curr_slave_lock);
3019 case BOND_LINK_DOWN:
3020 if (slave->link_failure_count < UINT_MAX)
3021 slave->link_failure_count++;
3023 slave->link = BOND_LINK_DOWN;
3025 if (slave == bond->curr_active_slave) {
3026 printk(KERN_INFO DRV_NAME
3027 ": %s: link status down for active "
3028 "interface %s, disabling it\n",
3029 bond->dev->name, slave->dev->name);
3031 bond_set_slave_inactive_flags(slave);
3033 write_lock_bh(&bond->curr_slave_lock);
3035 bond_select_active_slave(bond);
3036 if (bond->curr_active_slave)
3037 bond->curr_active_slave->jiffies =
3040 write_unlock_bh(&bond->curr_slave_lock);
3042 bond->current_arp_slave = NULL;
3044 } else if (slave->state == BOND_STATE_BACKUP) {
3045 printk(KERN_INFO DRV_NAME
3046 ": %s: backup interface %s is now down\n",
3047 bond->dev->name, slave->dev->name);
3049 bond_set_slave_inactive_flags(slave);
3054 printk(KERN_ERR DRV_NAME
3055 ": %s: impossible: new_link %d on slave %s\n",
3056 bond->dev->name, slave->new_link,
3062 * No race with changes to primary via sysfs, as we hold rtnl.
3064 if (bond->primary_slave &&
3065 (bond->primary_slave != bond->curr_active_slave) &&
3066 (bond->primary_slave->link == BOND_LINK_UP)) {
3067 write_lock_bh(&bond->curr_slave_lock);
3068 bond_change_active_slave(bond, bond->primary_slave);
3069 write_unlock_bh(&bond->curr_slave_lock);
3072 bond_set_carrier(bond);
3076 * Send ARP probes for active-backup mode ARP monitor.
3078 * Called with bond->lock held for read.
3080 static void bond_ab_arp_probe(struct bonding *bond)
3082 struct slave *slave;
3085 read_lock(&bond->curr_slave_lock);
3087 if (bond->current_arp_slave && bond->curr_active_slave)
3088 printk("PROBE: c_arp %s && cas %s BAD\n",
3089 bond->current_arp_slave->dev->name,
3090 bond->curr_active_slave->dev->name);
3092 if (bond->curr_active_slave) {
3093 bond_arp_send_all(bond, bond->curr_active_slave);
3094 read_unlock(&bond->curr_slave_lock);
3098 read_unlock(&bond->curr_slave_lock);
3100 /* if we don't have a curr_active_slave, search for the next available
3101 * backup slave from the current_arp_slave and make it the candidate
3102 * for becoming the curr_active_slave
3105 if (!bond->current_arp_slave) {
3106 bond->current_arp_slave = bond->first_slave;
3107 if (!bond->current_arp_slave)
3111 bond_set_slave_inactive_flags(bond->current_arp_slave);
3113 /* search for next candidate */
3114 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3115 if (IS_UP(slave->dev)) {
3116 slave->link = BOND_LINK_BACK;
3117 bond_set_slave_active_flags(slave);
3118 bond_arp_send_all(bond, slave);
3119 slave->jiffies = jiffies;
3120 bond->current_arp_slave = slave;
3124 /* if the link state is up at this point, we
3125 * mark it down - this can happen if we have
3126 * simultaneous link failures and
3127 * reselect_active_interface doesn't make this
3128 * one the current slave so it is still marked
3129 * up when it is actually down
3131 if (slave->link == BOND_LINK_UP) {
3132 slave->link = BOND_LINK_DOWN;
3133 if (slave->link_failure_count < UINT_MAX)
3134 slave->link_failure_count++;
3136 bond_set_slave_inactive_flags(slave);
3138 printk(KERN_INFO DRV_NAME
3139 ": %s: backup interface %s is now down.\n",
3140 bond->dev->name, slave->dev->name);
3145 void bond_activebackup_arp_mon(struct work_struct *work)
3147 struct bonding *bond = container_of(work, struct bonding,
3151 read_lock(&bond->lock);
3153 if (bond->kill_timers)
3156 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3158 if (bond->slave_cnt == 0)
3161 if (bond->send_grat_arp) {
3162 read_lock(&bond->curr_slave_lock);
3163 bond_send_gratuitous_arp(bond);
3164 read_unlock(&bond->curr_slave_lock);
3167 if (bond->send_unsol_na) {
3168 read_lock(&bond->curr_slave_lock);
3169 bond_send_unsolicited_na(bond);
3170 read_unlock(&bond->curr_slave_lock);
3173 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3174 read_unlock(&bond->lock);
3176 read_lock(&bond->lock);
3178 bond_ab_arp_commit(bond, delta_in_ticks);
3180 read_unlock(&bond->lock);
3182 read_lock(&bond->lock);
3185 bond_ab_arp_probe(bond);
3188 if (bond->params.arp_interval) {
3189 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3192 read_unlock(&bond->lock);
3195 /*------------------------------ proc/seq_file-------------------------------*/
3197 #ifdef CONFIG_PROC_FS
3199 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3200 __acquires(&dev_base_lock)
3201 __acquires(&bond->lock)
3203 struct bonding *bond = seq->private;
3205 struct slave *slave;
3208 /* make sure the bond won't be taken away */
3209 read_lock(&dev_base_lock);
3210 read_lock(&bond->lock);
3213 return SEQ_START_TOKEN;
3216 bond_for_each_slave(bond, slave, i) {
3217 if (++off == *pos) {
3225 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3227 struct bonding *bond = seq->private;
3228 struct slave *slave = v;
3231 if (v == SEQ_START_TOKEN) {
3232 return bond->first_slave;
3235 slave = slave->next;
3237 return (slave == bond->first_slave) ? NULL : slave;
3240 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3241 __releases(&bond->lock)
3242 __releases(&dev_base_lock)
3244 struct bonding *bond = seq->private;
3246 read_unlock(&bond->lock);
3247 read_unlock(&dev_base_lock);
3250 static void bond_info_show_master(struct seq_file *seq)
3252 struct bonding *bond = seq->private;
3256 read_lock(&bond->curr_slave_lock);
3257 curr = bond->curr_active_slave;
3258 read_unlock(&bond->curr_slave_lock);
3260 seq_printf(seq, "Bonding Mode: %s",
3261 bond_mode_name(bond->params.mode));
3263 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3264 bond->params.fail_over_mac)
3265 seq_printf(seq, " (fail_over_mac %s)",
3266 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3268 seq_printf(seq, "\n");
3270 if (bond->params.mode == BOND_MODE_XOR ||
3271 bond->params.mode == BOND_MODE_8023AD) {
3272 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3273 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3274 bond->params.xmit_policy);
3277 if (USES_PRIMARY(bond->params.mode)) {
3278 seq_printf(seq, "Primary Slave: %s\n",
3279 (bond->primary_slave) ?
3280 bond->primary_slave->dev->name : "None");
3282 seq_printf(seq, "Currently Active Slave: %s\n",
3283 (curr) ? curr->dev->name : "None");
3286 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3288 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3289 seq_printf(seq, "Up Delay (ms): %d\n",
3290 bond->params.updelay * bond->params.miimon);
3291 seq_printf(seq, "Down Delay (ms): %d\n",
3292 bond->params.downdelay * bond->params.miimon);
3295 /* ARP information */
3296 if(bond->params.arp_interval > 0) {
3298 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3299 bond->params.arp_interval);
3301 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3303 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3304 if (!bond->params.arp_targets[i])
3307 seq_printf(seq, ",");
3308 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3311 seq_printf(seq, "\n");
3314 if (bond->params.mode == BOND_MODE_8023AD) {
3315 struct ad_info ad_info;
3317 seq_puts(seq, "\n802.3ad info\n");
3318 seq_printf(seq, "LACP rate: %s\n",
3319 (bond->params.lacp_fast) ? "fast" : "slow");
3320 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3321 ad_select_tbl[bond->params.ad_select].modename);
3323 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3324 seq_printf(seq, "bond %s has no active aggregator\n",
3327 seq_printf(seq, "Active Aggregator Info:\n");
3329 seq_printf(seq, "\tAggregator ID: %d\n",
3330 ad_info.aggregator_id);
3331 seq_printf(seq, "\tNumber of ports: %d\n",
3333 seq_printf(seq, "\tActor Key: %d\n",
3335 seq_printf(seq, "\tPartner Key: %d\n",
3336 ad_info.partner_key);
3337 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3338 ad_info.partner_system);
3343 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3345 struct bonding *bond = seq->private;
3347 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3348 seq_printf(seq, "MII Status: %s\n",
3349 (slave->link == BOND_LINK_UP) ? "up" : "down");
3350 seq_printf(seq, "Link Failure Count: %u\n",
3351 slave->link_failure_count);
3353 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3355 if (bond->params.mode == BOND_MODE_8023AD) {
3356 const struct aggregator *agg
3357 = SLAVE_AD_INFO(slave).port.aggregator;
3360 seq_printf(seq, "Aggregator ID: %d\n",
3361 agg->aggregator_identifier);
3363 seq_puts(seq, "Aggregator ID: N/A\n");
3368 static int bond_info_seq_show(struct seq_file *seq, void *v)
3370 if (v == SEQ_START_TOKEN) {
3371 seq_printf(seq, "%s\n", version);
3372 bond_info_show_master(seq);
3374 bond_info_show_slave(seq, v);
3380 static const struct seq_operations bond_info_seq_ops = {
3381 .start = bond_info_seq_start,
3382 .next = bond_info_seq_next,
3383 .stop = bond_info_seq_stop,
3384 .show = bond_info_seq_show,
3387 static int bond_info_open(struct inode *inode, struct file *file)
3389 struct seq_file *seq;
3390 struct proc_dir_entry *proc;
3393 res = seq_open(file, &bond_info_seq_ops);
3395 /* recover the pointer buried in proc_dir_entry data */
3396 seq = file->private_data;
3398 seq->private = proc->data;
3404 static const struct file_operations bond_info_fops = {
3405 .owner = THIS_MODULE,
3406 .open = bond_info_open,
3408 .llseek = seq_lseek,
3409 .release = seq_release,
3412 static int bond_create_proc_entry(struct bonding *bond)
3414 struct net_device *bond_dev = bond->dev;
3416 if (bond_proc_dir) {
3417 bond->proc_entry = proc_create_data(bond_dev->name,
3418 S_IRUGO, bond_proc_dir,
3419 &bond_info_fops, bond);
3420 if (bond->proc_entry == NULL) {
3421 printk(KERN_WARNING DRV_NAME
3422 ": Warning: Cannot create /proc/net/%s/%s\n",
3423 DRV_NAME, bond_dev->name);
3425 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3432 static void bond_remove_proc_entry(struct bonding *bond)
3434 if (bond_proc_dir && bond->proc_entry) {
3435 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3436 memset(bond->proc_file_name, 0, IFNAMSIZ);
3437 bond->proc_entry = NULL;
3441 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3442 * Caller must hold rtnl_lock.
3444 static void bond_create_proc_dir(void)
3446 int len = strlen(DRV_NAME);
3448 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3449 bond_proc_dir = bond_proc_dir->next) {
3450 if ((bond_proc_dir->namelen == len) &&
3451 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3456 if (!bond_proc_dir) {
3457 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3458 if (bond_proc_dir) {
3459 bond_proc_dir->owner = THIS_MODULE;
3461 printk(KERN_WARNING DRV_NAME
3462 ": Warning: cannot create /proc/net/%s\n",
3468 /* Destroy the bonding directory under /proc/net, if empty.
3469 * Caller must hold rtnl_lock.
3471 static void bond_destroy_proc_dir(void)
3473 struct proc_dir_entry *de;
3475 if (!bond_proc_dir) {
3479 /* verify that the /proc dir is empty */
3480 for (de = bond_proc_dir->subdir; de; de = de->next) {
3481 /* ignore . and .. */
3482 if (*(de->name) != '.') {
3488 if (bond_proc_dir->owner == THIS_MODULE) {
3489 bond_proc_dir->owner = NULL;
3492 remove_proc_entry(DRV_NAME, init_net.proc_net);
3493 bond_proc_dir = NULL;
3496 #endif /* CONFIG_PROC_FS */
3498 /*-------------------------- netdev event handling --------------------------*/
3501 * Change device name
3503 static int bond_event_changename(struct bonding *bond)
3505 #ifdef CONFIG_PROC_FS
3506 bond_remove_proc_entry(bond);
3507 bond_create_proc_entry(bond);
3509 down_write(&(bonding_rwsem));
3510 bond_destroy_sysfs_entry(bond);
3511 bond_create_sysfs_entry(bond);
3512 up_write(&(bonding_rwsem));
3516 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3518 struct bonding *event_bond = netdev_priv(bond_dev);
3521 case NETDEV_CHANGENAME:
3522 return bond_event_changename(event_bond);
3523 case NETDEV_UNREGISTER:
3524 bond_release_all(event_bond->dev);
3533 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3535 struct net_device *bond_dev = slave_dev->master;
3536 struct bonding *bond = netdev_priv(bond_dev);
3539 case NETDEV_UNREGISTER:
3541 if (bond->setup_by_slave)
3542 bond_release_and_destroy(bond_dev, slave_dev);
3544 bond_release(bond_dev, slave_dev);
3548 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3549 struct slave *slave;
3551 slave = bond_get_slave_by_dev(bond, slave_dev);
3553 u16 old_speed = slave->speed;
3554 u16 old_duplex = slave->duplex;
3556 bond_update_speed_duplex(slave);
3558 if (bond_is_lb(bond))
3561 if (old_speed != slave->speed)
3562 bond_3ad_adapter_speed_changed(slave);
3563 if (old_duplex != slave->duplex)
3564 bond_3ad_adapter_duplex_changed(slave);
3571 * ... Or is it this?
3574 case NETDEV_CHANGEMTU:
3576 * TODO: Should slaves be allowed to
3577 * independently alter their MTU? For
3578 * an active-backup bond, slaves need
3579 * not be the same type of device, so
3580 * MTUs may vary. For other modes,
3581 * slaves arguably should have the
3582 * same MTUs. To do this, we'd need to
3583 * take over the slave's change_mtu
3584 * function for the duration of their
3588 case NETDEV_CHANGENAME:
3590 * TODO: handle changing the primary's name
3593 case NETDEV_FEAT_CHANGE:
3594 bond_compute_features(bond);
3604 * bond_netdev_event: handle netdev notifier chain events.
3606 * This function receives events for the netdev chain. The caller (an
3607 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3608 * locks for us to safely manipulate the slave devices (RTNL lock,
3611 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3613 struct net_device *event_dev = (struct net_device *)ptr;
3615 if (dev_net(event_dev) != &init_net)
3618 pr_debug("event_dev: %s, event: %lx\n",
3619 (event_dev ? event_dev->name : "None"),
3622 if (!(event_dev->priv_flags & IFF_BONDING))
3625 if (event_dev->flags & IFF_MASTER) {
3626 pr_debug("IFF_MASTER\n");
3627 return bond_master_netdev_event(event, event_dev);
3630 if (event_dev->flags & IFF_SLAVE) {
3631 pr_debug("IFF_SLAVE\n");
3632 return bond_slave_netdev_event(event, event_dev);
3639 * bond_inetaddr_event: handle inetaddr notifier chain events.
3641 * We keep track of device IPs primarily to use as source addresses in
3642 * ARP monitor probes (rather than spewing out broadcasts all the time).
3644 * We track one IP for the main device (if it has one), plus one per VLAN.
3646 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3648 struct in_ifaddr *ifa = ptr;
3649 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3650 struct bonding *bond;
3651 struct vlan_entry *vlan;
3653 if (dev_net(ifa->ifa_dev->dev) != &init_net)
3656 list_for_each_entry(bond, &bond_dev_list, bond_list) {
3657 if (bond->dev == event_dev) {
3660 bond->master_ip = ifa->ifa_local;
3663 bond->master_ip = bond_glean_dev_ip(bond->dev);
3670 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3671 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3672 if (vlan_dev == event_dev) {
3675 vlan->vlan_ip = ifa->ifa_local;
3679 bond_glean_dev_ip(vlan_dev);
3690 static struct notifier_block bond_netdev_notifier = {
3691 .notifier_call = bond_netdev_event,
3694 static struct notifier_block bond_inetaddr_notifier = {
3695 .notifier_call = bond_inetaddr_event,
3698 /*-------------------------- Packet type handling ---------------------------*/
3700 /* register to receive lacpdus on a bond */
3701 static void bond_register_lacpdu(struct bonding *bond)
3703 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3705 /* initialize packet type */
3706 pk_type->type = PKT_TYPE_LACPDU;
3707 pk_type->dev = bond->dev;
3708 pk_type->func = bond_3ad_lacpdu_recv;
3710 dev_add_pack(pk_type);
3713 /* unregister to receive lacpdus on a bond */
3714 static void bond_unregister_lacpdu(struct bonding *bond)
3716 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3719 void bond_register_arp(struct bonding *bond)
3721 struct packet_type *pt = &bond->arp_mon_pt;
3726 pt->type = htons(ETH_P_ARP);
3727 pt->dev = bond->dev;
3728 pt->func = bond_arp_rcv;
3732 void bond_unregister_arp(struct bonding *bond)
3734 struct packet_type *pt = &bond->arp_mon_pt;
3736 dev_remove_pack(pt);
3740 /*---------------------------- Hashing Policies -----------------------------*/
3743 * Hash for the output device based upon layer 2 and layer 3 data. If
3744 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3746 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3747 struct net_device *bond_dev, int count)
3749 struct ethhdr *data = (struct ethhdr *)skb->data;
3750 struct iphdr *iph = ip_hdr(skb);
3752 if (skb->protocol == htons(ETH_P_IP)) {
3753 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3754 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3757 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3761 * Hash for the output device based upon layer 3 and layer 4 data. If
3762 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3763 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3765 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3766 struct net_device *bond_dev, int count)
3768 struct ethhdr *data = (struct ethhdr *)skb->data;
3769 struct iphdr *iph = ip_hdr(skb);
3770 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3773 if (skb->protocol == htons(ETH_P_IP)) {
3774 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3775 (iph->protocol == IPPROTO_TCP ||
3776 iph->protocol == IPPROTO_UDP)) {
3777 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3779 return (layer4_xor ^
3780 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3784 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3788 * Hash for the output device based upon layer 2 data
3790 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3791 struct net_device *bond_dev, int count)
3793 struct ethhdr *data = (struct ethhdr *)skb->data;
3795 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3798 /*-------------------------- Device entry points ----------------------------*/
3800 static int bond_open(struct net_device *bond_dev)
3802 struct bonding *bond = netdev_priv(bond_dev);
3804 bond->kill_timers = 0;
3806 if (bond_is_lb(bond)) {
3807 /* bond_alb_initialize must be called before the timer
3810 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3811 /* something went wrong - fail the open operation */
3815 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3816 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3819 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3820 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3821 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3824 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3825 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3826 INIT_DELAYED_WORK(&bond->arp_work,
3827 bond_activebackup_arp_mon);
3829 INIT_DELAYED_WORK(&bond->arp_work,
3830 bond_loadbalance_arp_mon);
3832 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3833 if (bond->params.arp_validate)
3834 bond_register_arp(bond);
3837 if (bond->params.mode == BOND_MODE_8023AD) {
3838 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3839 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3840 /* register to receive LACPDUs */
3841 bond_register_lacpdu(bond);
3842 bond_3ad_initiate_agg_selection(bond, 1);
3848 static int bond_close(struct net_device *bond_dev)
3850 struct bonding *bond = netdev_priv(bond_dev);
3852 if (bond->params.mode == BOND_MODE_8023AD) {
3853 /* Unregister the receive of LACPDUs */
3854 bond_unregister_lacpdu(bond);
3857 if (bond->params.arp_validate)
3858 bond_unregister_arp(bond);
3860 write_lock_bh(&bond->lock);
3862 bond->send_grat_arp = 0;
3863 bond->send_unsol_na = 0;
3865 /* signal timers not to re-arm */
3866 bond->kill_timers = 1;
3868 write_unlock_bh(&bond->lock);
3870 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3871 cancel_delayed_work(&bond->mii_work);
3874 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3875 cancel_delayed_work(&bond->arp_work);
3878 switch (bond->params.mode) {
3879 case BOND_MODE_8023AD:
3880 cancel_delayed_work(&bond->ad_work);
3884 cancel_delayed_work(&bond->alb_work);
3891 if (bond_is_lb(bond)) {
3892 /* Must be called only after all
3893 * slaves have been released
3895 bond_alb_deinitialize(bond);
3901 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3903 struct bonding *bond = netdev_priv(bond_dev);
3904 struct net_device_stats *stats = &bond->stats;
3905 struct net_device_stats local_stats;
3906 struct slave *slave;
3909 memset(&local_stats, 0, sizeof(struct net_device_stats));
3911 read_lock_bh(&bond->lock);
3913 bond_for_each_slave(bond, slave, i) {
3914 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3916 local_stats.rx_packets += sstats->rx_packets;
3917 local_stats.rx_bytes += sstats->rx_bytes;
3918 local_stats.rx_errors += sstats->rx_errors;
3919 local_stats.rx_dropped += sstats->rx_dropped;
3921 local_stats.tx_packets += sstats->tx_packets;
3922 local_stats.tx_bytes += sstats->tx_bytes;
3923 local_stats.tx_errors += sstats->tx_errors;
3924 local_stats.tx_dropped += sstats->tx_dropped;
3926 local_stats.multicast += sstats->multicast;
3927 local_stats.collisions += sstats->collisions;
3929 local_stats.rx_length_errors += sstats->rx_length_errors;
3930 local_stats.rx_over_errors += sstats->rx_over_errors;
3931 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3932 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3933 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3934 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3936 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3937 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3938 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3939 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3940 local_stats.tx_window_errors += sstats->tx_window_errors;
3943 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3945 read_unlock_bh(&bond->lock);
3950 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3952 struct net_device *slave_dev = NULL;
3953 struct ifbond k_binfo;
3954 struct ifbond __user *u_binfo = NULL;
3955 struct ifslave k_sinfo;
3956 struct ifslave __user *u_sinfo = NULL;
3957 struct mii_ioctl_data *mii = NULL;
3960 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3961 bond_dev->name, cmd);
3973 * We do this again just in case we were called by SIOCGMIIREG
3974 * instead of SIOCGMIIPHY.
3981 if (mii->reg_num == 1) {
3982 struct bonding *bond = netdev_priv(bond_dev);
3984 read_lock(&bond->lock);
3985 read_lock(&bond->curr_slave_lock);
3986 if (netif_carrier_ok(bond->dev)) {
3987 mii->val_out = BMSR_LSTATUS;
3989 read_unlock(&bond->curr_slave_lock);
3990 read_unlock(&bond->lock);
3994 case BOND_INFO_QUERY_OLD:
3995 case SIOCBONDINFOQUERY:
3996 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3998 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
4002 res = bond_info_query(bond_dev, &k_binfo);
4004 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
4010 case BOND_SLAVE_INFO_QUERY_OLD:
4011 case SIOCBONDSLAVEINFOQUERY:
4012 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4014 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4018 res = bond_slave_info_query(bond_dev, &k_sinfo);
4020 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4031 if (!capable(CAP_NET_ADMIN)) {
4035 down_write(&(bonding_rwsem));
4036 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
4038 pr_debug("slave_dev=%p: \n", slave_dev);
4043 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
4045 case BOND_ENSLAVE_OLD:
4046 case SIOCBONDENSLAVE:
4047 res = bond_enslave(bond_dev, slave_dev);
4049 case BOND_RELEASE_OLD:
4050 case SIOCBONDRELEASE:
4051 res = bond_release(bond_dev, slave_dev);
4053 case BOND_SETHWADDR_OLD:
4054 case SIOCBONDSETHWADDR:
4055 res = bond_sethwaddr(bond_dev, slave_dev);
4057 case BOND_CHANGE_ACTIVE_OLD:
4058 case SIOCBONDCHANGEACTIVE:
4059 res = bond_ioctl_change_active(bond_dev, slave_dev);
4068 up_write(&(bonding_rwsem));
4072 static void bond_set_multicast_list(struct net_device *bond_dev)
4074 struct bonding *bond = netdev_priv(bond_dev);
4075 struct dev_mc_list *dmi;
4078 * Do promisc before checking multicast_mode
4080 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4082 * FIXME: Need to handle the error when one of the multi-slaves
4085 bond_set_promiscuity(bond, 1);
4088 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4089 bond_set_promiscuity(bond, -1);
4092 /* set allmulti flag to slaves */
4093 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4095 * FIXME: Need to handle the error when one of the multi-slaves
4098 bond_set_allmulti(bond, 1);
4101 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4102 bond_set_allmulti(bond, -1);
4105 read_lock(&bond->lock);
4107 bond->flags = bond_dev->flags;
4109 /* looking for addresses to add to slaves' mc list */
4110 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4111 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4112 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4116 /* looking for addresses to delete from slaves' list */
4117 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4118 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4119 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4123 /* save master's multicast list */
4124 bond_mc_list_destroy(bond);
4125 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4127 read_unlock(&bond->lock);
4130 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4132 struct bonding *bond = netdev_priv(dev);
4133 struct slave *slave = bond->first_slave;
4136 const struct net_device_ops *slave_ops
4137 = slave->dev->netdev_ops;
4138 if (slave_ops->ndo_neigh_setup)
4139 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4145 * Change the MTU of all of a master's slaves to match the master
4147 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4149 struct bonding *bond = netdev_priv(bond_dev);
4150 struct slave *slave, *stop_at;
4154 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4155 (bond_dev ? bond_dev->name : "None"), new_mtu);
4157 /* Can't hold bond->lock with bh disabled here since
4158 * some base drivers panic. On the other hand we can't
4159 * hold bond->lock without bh disabled because we'll
4160 * deadlock. The only solution is to rely on the fact
4161 * that we're under rtnl_lock here, and the slaves
4162 * list won't change. This doesn't solve the problem
4163 * of setting the slave's MTU while it is
4164 * transmitting, but the assumption is that the base
4165 * driver can handle that.
4167 * TODO: figure out a way to safely iterate the slaves
4168 * list, but without holding a lock around the actual
4169 * call to the base driver.
4172 bond_for_each_slave(bond, slave, i) {
4173 pr_debug("s %p s->p %p c_m %p\n", slave,
4174 slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
4176 res = dev_set_mtu(slave->dev, new_mtu);
4179 /* If we failed to set the slave's mtu to the new value
4180 * we must abort the operation even in ACTIVE_BACKUP
4181 * mode, because if we allow the backup slaves to have
4182 * different mtu values than the active slave we'll
4183 * need to change their mtu when doing a failover. That
4184 * means changing their mtu from timer context, which
4185 * is probably not a good idea.
4187 pr_debug("err %d %s\n", res, slave->dev->name);
4192 bond_dev->mtu = new_mtu;
4197 /* unwind from head to the slave that failed */
4199 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4202 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4204 pr_debug("unwind err %d dev %s\n", tmp_res,
4215 * Note that many devices must be down to change the HW address, and
4216 * downing the master releases all slaves. We can make bonds full of
4217 * bonding devices to test this, however.
4219 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4221 struct bonding *bond = netdev_priv(bond_dev);
4222 struct sockaddr *sa = addr, tmp_sa;
4223 struct slave *slave, *stop_at;
4227 if (bond->params.mode == BOND_MODE_ALB)
4228 return bond_alb_set_mac_address(bond_dev, addr);
4231 pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4234 * If fail_over_mac is set to active, do nothing and return
4235 * success. Returning an error causes ifenslave to fail.
4237 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4240 if (!is_valid_ether_addr(sa->sa_data)) {
4241 return -EADDRNOTAVAIL;
4244 /* Can't hold bond->lock with bh disabled here since
4245 * some base drivers panic. On the other hand we can't
4246 * hold bond->lock without bh disabled because we'll
4247 * deadlock. The only solution is to rely on the fact
4248 * that we're under rtnl_lock here, and the slaves
4249 * list won't change. This doesn't solve the problem
4250 * of setting the slave's hw address while it is
4251 * transmitting, but the assumption is that the base
4252 * driver can handle that.
4254 * TODO: figure out a way to safely iterate the slaves
4255 * list, but without holding a lock around the actual
4256 * call to the base driver.
4259 bond_for_each_slave(bond, slave, i) {
4260 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4261 pr_debug("slave %p %s\n", slave, slave->dev->name);
4263 if (slave_ops->ndo_set_mac_address == NULL) {
4265 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4269 res = dev_set_mac_address(slave->dev, addr);
4271 /* TODO: consider downing the slave
4273 * User should expect communications
4274 * breakage anyway until ARP finish
4277 pr_debug("err %d %s\n", res, slave->dev->name);
4283 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4287 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4288 tmp_sa.sa_family = bond_dev->type;
4290 /* unwind from head to the slave that failed */
4292 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4295 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4297 pr_debug("unwind err %d dev %s\n", tmp_res,
4305 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4307 struct bonding *bond = netdev_priv(bond_dev);
4308 struct slave *slave, *start_at;
4309 int i, slave_no, res = 1;
4311 read_lock(&bond->lock);
4313 if (!BOND_IS_OK(bond)) {
4318 * Concurrent TX may collide on rr_tx_counter; we accept that
4319 * as being rare enough not to justify using an atomic op here
4321 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4323 bond_for_each_slave(bond, slave, i) {
4331 bond_for_each_slave_from(bond, slave, i, start_at) {
4332 if (IS_UP(slave->dev) &&
4333 (slave->link == BOND_LINK_UP) &&
4334 (slave->state == BOND_STATE_ACTIVE)) {
4335 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4342 /* no suitable interface, frame not sent */
4345 read_unlock(&bond->lock);
4351 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4352 * the bond has a usable interface.
4354 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4356 struct bonding *bond = netdev_priv(bond_dev);
4359 read_lock(&bond->lock);
4360 read_lock(&bond->curr_slave_lock);
4362 if (!BOND_IS_OK(bond)) {
4366 if (!bond->curr_active_slave)
4369 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4373 /* no suitable interface, frame not sent */
4376 read_unlock(&bond->curr_slave_lock);
4377 read_unlock(&bond->lock);
4382 * In bond_xmit_xor() , we determine the output device by using a pre-
4383 * determined xmit_hash_policy(), If the selected device is not enabled,
4384 * find the next active slave.
4386 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4388 struct bonding *bond = netdev_priv(bond_dev);
4389 struct slave *slave, *start_at;
4394 read_lock(&bond->lock);
4396 if (!BOND_IS_OK(bond)) {
4400 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4402 bond_for_each_slave(bond, slave, i) {
4411 bond_for_each_slave_from(bond, slave, i, start_at) {
4412 if (IS_UP(slave->dev) &&
4413 (slave->link == BOND_LINK_UP) &&
4414 (slave->state == BOND_STATE_ACTIVE)) {
4415 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4422 /* no suitable interface, frame not sent */
4425 read_unlock(&bond->lock);
4430 * in broadcast mode, we send everything to all usable interfaces.
4432 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4434 struct bonding *bond = netdev_priv(bond_dev);
4435 struct slave *slave, *start_at;
4436 struct net_device *tx_dev = NULL;
4440 read_lock(&bond->lock);
4442 if (!BOND_IS_OK(bond)) {
4446 read_lock(&bond->curr_slave_lock);
4447 start_at = bond->curr_active_slave;
4448 read_unlock(&bond->curr_slave_lock);
4454 bond_for_each_slave_from(bond, slave, i, start_at) {
4455 if (IS_UP(slave->dev) &&
4456 (slave->link == BOND_LINK_UP) &&
4457 (slave->state == BOND_STATE_ACTIVE)) {
4459 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4461 printk(KERN_ERR DRV_NAME
4462 ": %s: Error: bond_xmit_broadcast(): "
4463 "skb_clone() failed\n",
4468 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4470 dev_kfree_skb(skb2);
4474 tx_dev = slave->dev;
4479 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4484 /* no suitable interface, frame not sent */
4487 /* frame sent to all suitable interfaces */
4488 read_unlock(&bond->lock);
4492 /*------------------------- Device initialization ---------------------------*/
4494 static void bond_set_xmit_hash_policy(struct bonding *bond)
4496 switch (bond->params.xmit_policy) {
4497 case BOND_XMIT_POLICY_LAYER23:
4498 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4500 case BOND_XMIT_POLICY_LAYER34:
4501 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4503 case BOND_XMIT_POLICY_LAYER2:
4505 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4510 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4512 const struct bonding *bond = netdev_priv(dev);
4514 switch (bond->params.mode) {
4515 case BOND_MODE_ROUNDROBIN:
4516 return bond_xmit_roundrobin(skb, dev);
4517 case BOND_MODE_ACTIVEBACKUP:
4518 return bond_xmit_activebackup(skb, dev);
4520 return bond_xmit_xor(skb, dev);
4521 case BOND_MODE_BROADCAST:
4522 return bond_xmit_broadcast(skb, dev);
4523 case BOND_MODE_8023AD:
4524 return bond_3ad_xmit_xor(skb, dev);
4527 return bond_alb_xmit(skb, dev);
4529 /* Should never happen, mode already checked */
4530 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4531 dev->name, bond->params.mode);
4534 return NETDEV_TX_OK;
4540 * set bond mode specific net device operations
4542 void bond_set_mode_ops(struct bonding *bond, int mode)
4544 struct net_device *bond_dev = bond->dev;
4547 case BOND_MODE_ROUNDROBIN:
4549 case BOND_MODE_ACTIVEBACKUP:
4552 bond_set_xmit_hash_policy(bond);
4554 case BOND_MODE_BROADCAST:
4556 case BOND_MODE_8023AD:
4557 bond_set_master_3ad_flags(bond);
4558 bond_set_xmit_hash_policy(bond);
4561 bond_set_master_alb_flags(bond);
4566 /* Should never happen, mode already checked */
4567 printk(KERN_ERR DRV_NAME
4568 ": %s: Error: Unknown bonding mode %d\n",
4575 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4576 struct ethtool_drvinfo *drvinfo)
4578 strncpy(drvinfo->driver, DRV_NAME, 32);
4579 strncpy(drvinfo->version, DRV_VERSION, 32);
4580 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4583 static const struct ethtool_ops bond_ethtool_ops = {
4584 .get_drvinfo = bond_ethtool_get_drvinfo,
4585 .get_link = ethtool_op_get_link,
4586 .get_tx_csum = ethtool_op_get_tx_csum,
4587 .get_sg = ethtool_op_get_sg,
4588 .get_tso = ethtool_op_get_tso,
4589 .get_ufo = ethtool_op_get_ufo,
4590 .get_flags = ethtool_op_get_flags,
4593 static const struct net_device_ops bond_netdev_ops = {
4594 .ndo_open = bond_open,
4595 .ndo_stop = bond_close,
4596 .ndo_start_xmit = bond_start_xmit,
4597 .ndo_get_stats = bond_get_stats,
4598 .ndo_do_ioctl = bond_do_ioctl,
4599 .ndo_set_multicast_list = bond_set_multicast_list,
4600 .ndo_change_mtu = bond_change_mtu,
4601 .ndo_set_mac_address = bond_set_mac_address,
4602 .ndo_neigh_setup = bond_neigh_setup,
4603 .ndo_vlan_rx_register = bond_vlan_rx_register,
4604 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4605 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4609 * Does not allocate but creates a /proc entry.
4612 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4614 struct bonding *bond = netdev_priv(bond_dev);
4616 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4618 /* initialize rwlocks */
4619 rwlock_init(&bond->lock);
4620 rwlock_init(&bond->curr_slave_lock);
4622 bond->params = *params; /* copy params struct */
4624 bond->wq = create_singlethread_workqueue(bond_dev->name);
4628 /* Initialize pointers */
4629 bond->first_slave = NULL;
4630 bond->curr_active_slave = NULL;
4631 bond->current_arp_slave = NULL;
4632 bond->primary_slave = NULL;
4633 bond->dev = bond_dev;
4634 bond->send_grat_arp = 0;
4635 bond->send_unsol_na = 0;
4636 bond->setup_by_slave = 0;
4637 INIT_LIST_HEAD(&bond->vlan_list);
4639 /* Initialize the device entry points */
4640 bond_dev->netdev_ops = &bond_netdev_ops;
4641 bond_dev->ethtool_ops = &bond_ethtool_ops;
4642 bond_set_mode_ops(bond, bond->params.mode);
4644 bond_dev->destructor = bond_destructor;
4646 /* Initialize the device options */
4647 bond_dev->tx_queue_len = 0;
4648 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4649 bond_dev->priv_flags |= IFF_BONDING;
4650 if (bond->params.arp_interval)
4651 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4653 /* At first, we block adding VLANs. That's the only way to
4654 * prevent problems that occur when adding VLANs over an
4655 * empty bond. The block will be removed once non-challenged
4656 * slaves are enslaved.
4658 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4660 /* don't acquire bond device's netif_tx_lock when
4662 bond_dev->features |= NETIF_F_LLTX;
4664 /* By default, we declare the bond to be fully
4665 * VLAN hardware accelerated capable. Special
4666 * care is taken in the various xmit functions
4667 * when there are slaves that are not hw accel
4670 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4671 NETIF_F_HW_VLAN_RX |
4672 NETIF_F_HW_VLAN_FILTER);
4674 #ifdef CONFIG_PROC_FS
4675 bond_create_proc_entry(bond);
4677 list_add_tail(&bond->bond_list, &bond_dev_list);
4682 static void bond_work_cancel_all(struct bonding *bond)
4684 write_lock_bh(&bond->lock);
4685 bond->kill_timers = 1;
4686 write_unlock_bh(&bond->lock);
4688 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4689 cancel_delayed_work(&bond->mii_work);
4691 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4692 cancel_delayed_work(&bond->arp_work);
4694 if (bond->params.mode == BOND_MODE_ALB &&
4695 delayed_work_pending(&bond->alb_work))
4696 cancel_delayed_work(&bond->alb_work);
4698 if (bond->params.mode == BOND_MODE_8023AD &&
4699 delayed_work_pending(&bond->ad_work))
4700 cancel_delayed_work(&bond->ad_work);
4703 /* De-initialize device specific data.
4704 * Caller must hold rtnl_lock.
4706 static void bond_deinit(struct net_device *bond_dev)
4708 struct bonding *bond = netdev_priv(bond_dev);
4710 list_del(&bond->bond_list);
4712 bond_work_cancel_all(bond);
4714 #ifdef CONFIG_PROC_FS
4715 bond_remove_proc_entry(bond);
4719 /* Unregister and free all bond devices.
4720 * Caller must hold rtnl_lock.
4722 static void bond_free_all(void)
4724 struct bonding *bond, *nxt;
4726 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4727 struct net_device *bond_dev = bond->dev;
4729 bond_work_cancel_all(bond);
4730 /* Release the bonded slaves */
4731 bond_release_all(bond_dev);
4735 #ifdef CONFIG_PROC_FS
4736 bond_destroy_proc_dir();
4740 /*------------------------- Module initialization ---------------------------*/
4743 * Convert string input module parms. Accept either the
4744 * number of the mode or its string name. A bit complicated because
4745 * some mode names are substrings of other names, and calls from sysfs
4746 * may have whitespace in the name (trailing newlines, for example).
4748 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4750 int modeint = -1, i, rv;
4751 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4753 for (p = (char *)buf; *p; p++)
4754 if (!(isdigit(*p) || isspace(*p)))
4758 rv = sscanf(buf, "%20s", modestr);
4760 rv = sscanf(buf, "%d", &modeint);
4765 for (i = 0; tbl[i].modename; i++) {
4766 if (modeint == tbl[i].mode)
4768 if (strcmp(modestr, tbl[i].modename) == 0)
4775 static int bond_check_params(struct bond_params *params)
4777 int arp_validate_value, fail_over_mac_value;
4780 * Convert string parameters.
4783 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4784 if (bond_mode == -1) {
4785 printk(KERN_ERR DRV_NAME
4786 ": Error: Invalid bonding mode \"%s\"\n",
4787 mode == NULL ? "NULL" : mode);
4792 if (xmit_hash_policy) {
4793 if ((bond_mode != BOND_MODE_XOR) &&
4794 (bond_mode != BOND_MODE_8023AD)) {
4795 printk(KERN_INFO DRV_NAME
4796 ": xor_mode param is irrelevant in mode %s\n",
4797 bond_mode_name(bond_mode));
4799 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4801 if (xmit_hashtype == -1) {
4802 printk(KERN_ERR DRV_NAME
4803 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4804 xmit_hash_policy == NULL ? "NULL" :
4812 if (bond_mode != BOND_MODE_8023AD) {
4813 printk(KERN_INFO DRV_NAME
4814 ": lacp_rate param is irrelevant in mode %s\n",
4815 bond_mode_name(bond_mode));
4817 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4818 if (lacp_fast == -1) {
4819 printk(KERN_ERR DRV_NAME
4820 ": Error: Invalid lacp rate \"%s\"\n",
4821 lacp_rate == NULL ? "NULL" : lacp_rate);
4828 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4829 if (params->ad_select == -1) {
4830 printk(KERN_ERR DRV_NAME
4831 ": Error: Invalid ad_select \"%s\"\n",
4832 ad_select == NULL ? "NULL" : ad_select);
4836 if (bond_mode != BOND_MODE_8023AD) {
4837 printk(KERN_WARNING DRV_NAME
4838 ": ad_select param only affects 802.3ad mode\n");
4841 params->ad_select = BOND_AD_STABLE;
4844 if (max_bonds < 0 || max_bonds > INT_MAX) {
4845 printk(KERN_WARNING DRV_NAME
4846 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4847 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4848 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4849 max_bonds = BOND_DEFAULT_MAX_BONDS;
4853 printk(KERN_WARNING DRV_NAME
4854 ": Warning: miimon module parameter (%d), "
4855 "not in range 0-%d, so it was reset to %d\n",
4856 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4857 miimon = BOND_LINK_MON_INTERV;
4861 printk(KERN_WARNING DRV_NAME
4862 ": Warning: updelay module parameter (%d), "
4863 "not in range 0-%d, so it was reset to 0\n",
4868 if (downdelay < 0) {
4869 printk(KERN_WARNING DRV_NAME
4870 ": Warning: downdelay module parameter (%d), "
4871 "not in range 0-%d, so it was reset to 0\n",
4872 downdelay, INT_MAX);
4876 if ((use_carrier != 0) && (use_carrier != 1)) {
4877 printk(KERN_WARNING DRV_NAME
4878 ": Warning: use_carrier module parameter (%d), "
4879 "not of valid value (0/1), so it was set to 1\n",
4884 if (num_grat_arp < 0 || num_grat_arp > 255) {
4885 printk(KERN_WARNING DRV_NAME
4886 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4887 "was reset to 1 \n", num_grat_arp);
4891 if (num_unsol_na < 0 || num_unsol_na > 255) {
4892 printk(KERN_WARNING DRV_NAME
4893 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4894 "was reset to 1 \n", num_unsol_na);
4898 /* reset values for 802.3ad */
4899 if (bond_mode == BOND_MODE_8023AD) {
4901 printk(KERN_WARNING DRV_NAME
4902 ": Warning: miimon must be specified, "
4903 "otherwise bonding will not detect link "
4904 "failure, speed and duplex which are "
4905 "essential for 802.3ad operation\n");
4906 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4911 /* reset values for TLB/ALB */
4912 if ((bond_mode == BOND_MODE_TLB) ||
4913 (bond_mode == BOND_MODE_ALB)) {
4915 printk(KERN_WARNING DRV_NAME
4916 ": Warning: miimon must be specified, "
4917 "otherwise bonding will not detect link "
4918 "failure and link speed which are essential "
4919 "for TLB/ALB load balancing\n");
4920 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4925 if (bond_mode == BOND_MODE_ALB) {
4926 printk(KERN_NOTICE DRV_NAME
4927 ": In ALB mode you might experience client "
4928 "disconnections upon reconnection of a link if the "
4929 "bonding module updelay parameter (%d msec) is "
4930 "incompatible with the forwarding delay time of the "
4936 if (updelay || downdelay) {
4937 /* just warn the user the up/down delay will have
4938 * no effect since miimon is zero...
4940 printk(KERN_WARNING DRV_NAME
4941 ": Warning: miimon module parameter not set "
4942 "and updelay (%d) or downdelay (%d) module "
4943 "parameter is set; updelay and downdelay have "
4944 "no effect unless miimon is set\n",
4945 updelay, downdelay);
4948 /* don't allow arp monitoring */
4950 printk(KERN_WARNING DRV_NAME
4951 ": Warning: miimon (%d) and arp_interval (%d) "
4952 "can't be used simultaneously, disabling ARP "
4954 miimon, arp_interval);
4958 if ((updelay % miimon) != 0) {
4959 printk(KERN_WARNING DRV_NAME
4960 ": Warning: updelay (%d) is not a multiple "
4961 "of miimon (%d), updelay rounded to %d ms\n",
4962 updelay, miimon, (updelay / miimon) * miimon);
4967 if ((downdelay % miimon) != 0) {
4968 printk(KERN_WARNING DRV_NAME
4969 ": Warning: downdelay (%d) is not a multiple "
4970 "of miimon (%d), downdelay rounded to %d ms\n",
4972 (downdelay / miimon) * miimon);
4975 downdelay /= miimon;
4978 if (arp_interval < 0) {
4979 printk(KERN_WARNING DRV_NAME
4980 ": Warning: arp_interval module parameter (%d) "
4981 ", not in range 0-%d, so it was reset to %d\n",
4982 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4983 arp_interval = BOND_LINK_ARP_INTERV;
4986 for (arp_ip_count = 0;
4987 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4989 /* not complete check, but should be good enough to
4991 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4992 printk(KERN_WARNING DRV_NAME
4993 ": Warning: bad arp_ip_target module parameter "
4994 "(%s), ARP monitoring will not be performed\n",
4995 arp_ip_target[arp_ip_count]);
4998 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4999 arp_target[arp_ip_count] = ip;
5003 if (arp_interval && !arp_ip_count) {
5004 /* don't allow arping if no arp_ip_target given... */
5005 printk(KERN_WARNING DRV_NAME
5006 ": Warning: arp_interval module parameter (%d) "
5007 "specified without providing an arp_ip_target "
5008 "parameter, arp_interval was reset to 0\n",
5014 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5015 printk(KERN_ERR DRV_NAME
5016 ": arp_validate only supported in active-backup mode\n");
5019 if (!arp_interval) {
5020 printk(KERN_ERR DRV_NAME
5021 ": arp_validate requires arp_interval\n");
5025 arp_validate_value = bond_parse_parm(arp_validate,
5027 if (arp_validate_value == -1) {
5028 printk(KERN_ERR DRV_NAME
5029 ": Error: invalid arp_validate \"%s\"\n",
5030 arp_validate == NULL ? "NULL" : arp_validate);
5034 arp_validate_value = 0;
5037 printk(KERN_INFO DRV_NAME
5038 ": MII link monitoring set to %d ms\n",
5040 } else if (arp_interval) {
5043 printk(KERN_INFO DRV_NAME
5044 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5046 arp_validate_tbl[arp_validate_value].modename,
5049 for (i = 0; i < arp_ip_count; i++)
5050 printk (" %s", arp_ip_target[i]);
5054 } else if (max_bonds) {
5055 /* miimon and arp_interval not set, we need one so things
5056 * work as expected, see bonding.txt for details
5058 printk(KERN_WARNING DRV_NAME
5059 ": Warning: either miimon or arp_interval and "
5060 "arp_ip_target module parameters must be specified, "
5061 "otherwise bonding will not detect link failures! see "
5062 "bonding.txt for details.\n");
5065 if (primary && !USES_PRIMARY(bond_mode)) {
5066 /* currently, using a primary only makes sense
5067 * in active backup, TLB or ALB modes
5069 printk(KERN_WARNING DRV_NAME
5070 ": Warning: %s primary device specified but has no "
5071 "effect in %s mode\n",
5072 primary, bond_mode_name(bond_mode));
5076 if (fail_over_mac) {
5077 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5079 if (fail_over_mac_value == -1) {
5080 printk(KERN_ERR DRV_NAME
5081 ": Error: invalid fail_over_mac \"%s\"\n",
5082 arp_validate == NULL ? "NULL" : arp_validate);
5086 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5087 printk(KERN_WARNING DRV_NAME
5088 ": Warning: fail_over_mac only affects "
5089 "active-backup mode.\n");
5091 fail_over_mac_value = BOND_FOM_NONE;
5094 /* fill params struct with the proper values */
5095 params->mode = bond_mode;
5096 params->xmit_policy = xmit_hashtype;
5097 params->miimon = miimon;
5098 params->num_grat_arp = num_grat_arp;
5099 params->num_unsol_na = num_unsol_na;
5100 params->arp_interval = arp_interval;
5101 params->arp_validate = arp_validate_value;
5102 params->updelay = updelay;
5103 params->downdelay = downdelay;
5104 params->use_carrier = use_carrier;
5105 params->lacp_fast = lacp_fast;
5106 params->primary[0] = 0;
5107 params->fail_over_mac = fail_over_mac_value;
5110 strncpy(params->primary, primary, IFNAMSIZ);
5111 params->primary[IFNAMSIZ - 1] = 0;
5114 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5119 static struct lock_class_key bonding_netdev_xmit_lock_key;
5120 static struct lock_class_key bonding_netdev_addr_lock_key;
5122 static void bond_set_lockdep_class_one(struct net_device *dev,
5123 struct netdev_queue *txq,
5126 lockdep_set_class(&txq->_xmit_lock,
5127 &bonding_netdev_xmit_lock_key);
5130 static void bond_set_lockdep_class(struct net_device *dev)
5132 lockdep_set_class(&dev->addr_list_lock,
5133 &bonding_netdev_addr_lock_key);
5134 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5137 /* Create a new bond based on the specified name and bonding parameters.
5138 * If name is NULL, obtain a suitable "bond%d" name for us.
5139 * Caller must NOT hold rtnl_lock; we need to release it here before we
5140 * set up our sysfs entries.
5142 int bond_create(char *name, struct bond_params *params)
5144 struct net_device *bond_dev;
5145 struct bonding *bond;
5149 down_write(&bonding_rwsem);
5151 /* Check to see if the bond already exists. */
5153 list_for_each_entry(bond, &bond_dev_list, bond_list)
5154 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5155 printk(KERN_ERR DRV_NAME
5156 ": cannot add bond %s; it already exists\n",
5163 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5166 printk(KERN_ERR DRV_NAME
5167 ": %s: eek! can't alloc netdev!\n",
5174 res = dev_alloc_name(bond_dev, "bond%d");
5179 /* bond_init() must be called after dev_alloc_name() (for the
5180 * /proc files), but before register_netdevice(), because we
5181 * need to set function pointers.
5184 res = bond_init(bond_dev, params);
5189 res = register_netdevice(bond_dev);
5194 bond_set_lockdep_class(bond_dev);
5196 netif_carrier_off(bond_dev);
5198 up_write(&bonding_rwsem);
5199 rtnl_unlock(); /* allows sysfs registration of net device */
5200 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5203 down_write(&bonding_rwsem);
5204 bond_deinit(bond_dev);
5205 unregister_netdevice(bond_dev);
5212 bond_deinit(bond_dev);
5214 free_netdev(bond_dev);
5216 up_write(&bonding_rwsem);
5221 static int __init bonding_init(void)
5225 struct bonding *bond;
5227 printk(KERN_INFO "%s", version);
5229 res = bond_check_params(&bonding_defaults);
5234 #ifdef CONFIG_PROC_FS
5235 bond_create_proc_dir();
5238 init_rwsem(&bonding_rwsem);
5240 for (i = 0; i < max_bonds; i++) {
5241 res = bond_create(NULL, &bonding_defaults);
5246 res = bond_create_sysfs();
5250 register_netdevice_notifier(&bond_netdev_notifier);
5251 register_inetaddr_notifier(&bond_inetaddr_notifier);
5252 bond_register_ipv6_notifier();
5256 list_for_each_entry(bond, &bond_dev_list, bond_list) {
5257 bond_work_cancel_all(bond);
5258 destroy_workqueue(bond->wq);
5261 bond_destroy_sysfs();
5271 static void __exit bonding_exit(void)
5273 unregister_netdevice_notifier(&bond_netdev_notifier);
5274 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5275 bond_unregister_ipv6_notifier();
5277 bond_destroy_sysfs();
5284 module_init(bonding_init);
5285 module_exit(bonding_exit);
5286 MODULE_LICENSE("GPL");
5287 MODULE_VERSION(DRV_VERSION);
5288 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5289 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5290 MODULE_SUPPORTED_DEVICE("most ethernet devices");