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>
55 #include <asm/system.h>
57 #include <linux/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;
94 static int use_carrier = 1;
97 static char *lacp_rate;
98 static char *ad_select;
99 static char *xmit_hash_policy;
100 static int arp_interval = BOND_LINK_ARP_INTERV;
101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
102 static char *arp_validate;
103 static char *fail_over_mac;
104 static 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;
157 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
158 static int arp_ip_count;
159 static int bond_mode = BOND_MODE_ROUNDROBIN;
160 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
161 static int lacp_fast;
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 int bond_init(struct net_device *bond_dev);
214 static void bond_deinit(struct net_device *bond_dev);
216 /*---------------------------- General routines -----------------------------*/
218 static const char *bond_mode_name(int mode)
220 static const char *names[] = {
221 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
222 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
223 [BOND_MODE_XOR] = "load balancing (xor)",
224 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
225 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
226 [BOND_MODE_TLB] = "transmit load balancing",
227 [BOND_MODE_ALB] = "adaptive load balancing",
230 if (mode < 0 || mode > BOND_MODE_ALB)
236 /*---------------------------------- VLAN -----------------------------------*/
239 * bond_add_vlan - add a new vlan id on bond
240 * @bond: bond that got the notification
241 * @vlan_id: the vlan id to add
243 * Returns -ENOMEM if allocation failed.
245 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
247 struct vlan_entry *vlan;
249 pr_debug("bond: %s, vlan id %d\n",
250 (bond ? bond->dev->name : "None"), vlan_id);
252 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))
365 next = list_entry(bond->vlan_list.next,
366 struct vlan_entry, vlan_list);
368 last = list_entry(bond->vlan_list.prev,
369 struct vlan_entry, vlan_list);
371 next = list_entry(bond->vlan_list.next,
372 struct vlan_entry, vlan_list);
374 next = list_entry(curr->vlan_list.next,
375 struct vlan_entry, vlan_list);
383 * bond_dev_queue_xmit - Prepare skb for xmit.
385 * @bond: bond device that got this skb for tx.
386 * @skb: hw accel VLAN tagged skb to transmit
387 * @slave_dev: slave that is supposed to xmit this skbuff
389 * When the bond gets an skb to transmit that is
390 * already hardware accelerated VLAN tagged, and it
391 * needs to relay this skb to a slave that is not
392 * hw accel capable, the skb needs to be "unaccelerated",
393 * i.e. strip the hwaccel tag and re-insert it as part
396 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
397 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,
447 struct vlan_group *grp)
449 struct bonding *bond = netdev_priv(bond_dev);
455 bond_for_each_slave(bond, slave, i) {
456 struct net_device *slave_dev = slave->dev;
457 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
459 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
460 slave_ops->ndo_vlan_rx_register) {
461 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
467 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
468 * @bond_dev: bonding net device that got called
469 * @vid: vlan id being added
471 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
473 struct bonding *bond = netdev_priv(bond_dev);
477 bond_for_each_slave(bond, slave, i) {
478 struct net_device *slave_dev = slave->dev;
479 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
481 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
482 slave_ops->ndo_vlan_rx_add_vid) {
483 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
487 res = bond_add_vlan(bond, vid);
490 ": %s: Error: Failed to add vlan id %d\n",
491 bond_dev->name, vid);
496 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
497 * @bond_dev: bonding net device that got called
498 * @vid: vlan id being removed
500 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
502 struct bonding *bond = netdev_priv(bond_dev);
504 struct net_device *vlan_dev;
507 bond_for_each_slave(bond, slave, i) {
508 struct net_device *slave_dev = slave->dev;
509 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
511 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
512 slave_ops->ndo_vlan_rx_kill_vid) {
513 /* Save and then restore vlan_dev in the grp array,
514 * since the slave's driver might clear it.
516 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
517 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
518 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
522 res = bond_del_vlan(bond, vid);
525 ": %s: Error: Failed to remove vlan id %d\n",
526 bond_dev->name, vid);
530 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
532 struct vlan_entry *vlan;
533 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
535 write_lock_bh(&bond->lock);
537 if (list_empty(&bond->vlan_list))
540 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
541 slave_ops->ndo_vlan_rx_register)
542 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
544 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
545 !(slave_ops->ndo_vlan_rx_add_vid))
548 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
549 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
552 write_unlock_bh(&bond->lock);
555 static void bond_del_vlans_from_slave(struct bonding *bond,
556 struct net_device *slave_dev)
558 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
559 struct vlan_entry *vlan;
560 struct net_device *vlan_dev;
562 write_lock_bh(&bond->lock);
564 if (list_empty(&bond->vlan_list))
567 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
568 !(slave_ops->ndo_vlan_rx_kill_vid))
571 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
572 /* Save and then restore vlan_dev in the grp array,
573 * since the slave's driver might clear it.
575 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
576 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
577 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
581 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
582 slave_ops->ndo_vlan_rx_register)
583 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
586 write_unlock_bh(&bond->lock);
589 /*------------------------------- Link status -------------------------------*/
592 * Set the carrier state for the master according to the state of its
593 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
594 * do special 802.3ad magic.
596 * Returns zero if carrier state does not change, nonzero if it does.
598 static int bond_set_carrier(struct bonding *bond)
603 if (bond->slave_cnt == 0)
606 if (bond->params.mode == BOND_MODE_8023AD)
607 return bond_3ad_set_carrier(bond);
609 bond_for_each_slave(bond, slave, i) {
610 if (slave->link == BOND_LINK_UP) {
611 if (!netif_carrier_ok(bond->dev)) {
612 netif_carrier_on(bond->dev);
620 if (netif_carrier_ok(bond->dev)) {
621 netif_carrier_off(bond->dev);
628 * Get link speed and duplex from the slave's base driver
629 * using ethtool. If for some reason the call fails or the
630 * values are invalid, fake speed and duplex to 100/Full
633 static int bond_update_speed_duplex(struct slave *slave)
635 struct net_device *slave_dev = slave->dev;
636 struct ethtool_cmd etool;
639 /* Fake speed and duplex */
640 slave->speed = SPEED_100;
641 slave->duplex = DUPLEX_FULL;
643 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
646 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
650 switch (etool.speed) {
660 switch (etool.duplex) {
668 slave->speed = etool.speed;
669 slave->duplex = etool.duplex;
675 * if <dev> supports MII link status reporting, check its link status.
677 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
678 * depending upon the setting of the use_carrier parameter.
680 * Return either BMSR_LSTATUS, meaning that the link is up (or we
681 * can't tell and just pretend it is), or 0, meaning that the link is
684 * If reporting is non-zero, instead of faking link up, return -1 if
685 * both ETHTOOL and MII ioctls fail (meaning the device does not
686 * support them). If use_carrier is set, return whatever it says.
687 * It'd be nice if there was a good way to tell if a driver supports
688 * netif_carrier, but there really isn't.
690 static int bond_check_dev_link(struct bonding *bond,
691 struct net_device *slave_dev, int reporting)
693 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
694 static int (*ioctl)(struct net_device *, struct ifreq *, int);
696 struct mii_ioctl_data *mii;
698 if (!reporting && !netif_running(slave_dev))
701 if (bond->params.use_carrier)
702 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
704 /* Try to get link status using Ethtool first. */
705 if (slave_dev->ethtool_ops) {
706 if (slave_dev->ethtool_ops->get_link) {
709 link = slave_dev->ethtool_ops->get_link(slave_dev);
711 return link ? BMSR_LSTATUS : 0;
715 /* Ethtool can't be used, fallback to MII ioctls. */
716 ioctl = slave_ops->ndo_do_ioctl;
718 /* TODO: set pointer to correct ioctl on a per team member */
719 /* bases to make this more efficient. that is, once */
720 /* we determine the correct ioctl, we will always */
721 /* call it and not the others for that team */
725 * We cannot assume that SIOCGMIIPHY will also read a
726 * register; not all network drivers (e.g., e100)
730 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
731 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
733 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
734 mii->reg_num = MII_BMSR;
735 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
736 return mii->val_out & BMSR_LSTATUS;
741 * If reporting, report that either there's no dev->do_ioctl,
742 * or both SIOCGMIIREG and get_link failed (meaning that we
743 * cannot report link status). If not reporting, pretend
746 return reporting ? -1 : BMSR_LSTATUS;
749 /*----------------------------- Multicast list ------------------------------*/
752 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
754 static inline int bond_is_dmi_same(const struct dev_mc_list *dmi1,
755 const struct dev_mc_list *dmi2)
757 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
758 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
762 * returns dmi entry if found, NULL otherwise
764 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi,
765 struct dev_mc_list *mc_list)
767 struct dev_mc_list *idmi;
769 for (idmi = mc_list; idmi; idmi = idmi->next) {
770 if (bond_is_dmi_same(dmi, idmi))
778 * Push the promiscuity flag down to appropriate slaves
780 static int bond_set_promiscuity(struct bonding *bond, int inc)
783 if (USES_PRIMARY(bond->params.mode)) {
784 /* write lock already acquired */
785 if (bond->curr_active_slave) {
786 err = dev_set_promiscuity(bond->curr_active_slave->dev,
792 bond_for_each_slave(bond, slave, i) {
793 err = dev_set_promiscuity(slave->dev, inc);
802 * Push the allmulti flag down to all slaves
804 static int bond_set_allmulti(struct bonding *bond, int inc)
807 if (USES_PRIMARY(bond->params.mode)) {
808 /* write lock already acquired */
809 if (bond->curr_active_slave) {
810 err = dev_set_allmulti(bond->curr_active_slave->dev,
816 bond_for_each_slave(bond, slave, i) {
817 err = dev_set_allmulti(slave->dev, inc);
826 * Add a Multicast address to slaves
829 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
831 if (USES_PRIMARY(bond->params.mode)) {
832 /* write lock already acquired */
833 if (bond->curr_active_slave)
834 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
839 bond_for_each_slave(bond, slave, i)
840 dev_mc_add(slave->dev, addr, alen, 0);
845 * Remove a multicast address from slave
848 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
850 if (USES_PRIMARY(bond->params.mode)) {
851 /* write lock already acquired */
852 if (bond->curr_active_slave)
853 dev_mc_delete(bond->curr_active_slave->dev, addr,
858 bond_for_each_slave(bond, slave, i) {
859 dev_mc_delete(slave->dev, addr, alen, 0);
866 * Retrieve the list of registered multicast addresses for the bonding
867 * device and retransmit an IGMP JOIN request to the current active
870 static void bond_resend_igmp_join_requests(struct bonding *bond)
872 struct in_device *in_dev;
873 struct ip_mc_list *im;
876 in_dev = __in_dev_get_rcu(bond->dev);
878 for (im = in_dev->mc_list; im; im = im->next)
879 ip_mc_rejoin_group(im);
886 * Totally destroys the mc_list in bond
888 static void bond_mc_list_destroy(struct bonding *bond)
890 struct dev_mc_list *dmi;
894 bond->mc_list = dmi->next;
899 bond->mc_list = NULL;
903 * Copy all the Multicast addresses from src to the bonding device dst
905 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
908 struct dev_mc_list *dmi, *new_dmi;
910 for (dmi = mc_list; dmi; dmi = dmi->next) {
911 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
914 /* FIXME: Potential memory leak !!! */
918 new_dmi->next = bond->mc_list;
919 bond->mc_list = new_dmi;
920 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
921 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
922 new_dmi->dmi_users = dmi->dmi_users;
923 new_dmi->dmi_gusers = dmi->dmi_gusers;
930 * flush all members of flush->mc_list from device dev->mc_list
932 static void bond_mc_list_flush(struct net_device *bond_dev,
933 struct net_device *slave_dev)
935 struct bonding *bond = netdev_priv(bond_dev);
936 struct dev_mc_list *dmi;
938 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
939 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
941 if (bond->params.mode == BOND_MODE_8023AD) {
942 /* del lacpdu mc addr from mc list */
943 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
945 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
949 /*--------------------------- Active slave change ---------------------------*/
952 * Update the mc list and multicast-related flags for the new and
953 * old active slaves (if any) according to the multicast mode, and
954 * promiscuous flags unconditionally.
956 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
957 struct slave *old_active)
959 struct dev_mc_list *dmi;
961 if (!USES_PRIMARY(bond->params.mode))
962 /* nothing to do - mc list is already up-to-date on
968 if (bond->dev->flags & IFF_PROMISC)
969 dev_set_promiscuity(old_active->dev, -1);
971 if (bond->dev->flags & IFF_ALLMULTI)
972 dev_set_allmulti(old_active->dev, -1);
974 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
975 dev_mc_delete(old_active->dev, dmi->dmi_addr,
976 dmi->dmi_addrlen, 0);
980 /* FIXME: Signal errors upstream. */
981 if (bond->dev->flags & IFF_PROMISC)
982 dev_set_promiscuity(new_active->dev, 1);
984 if (bond->dev->flags & IFF_ALLMULTI)
985 dev_set_allmulti(new_active->dev, 1);
987 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
988 dev_mc_add(new_active->dev, dmi->dmi_addr,
989 dmi->dmi_addrlen, 0);
990 bond_resend_igmp_join_requests(bond);
995 * bond_do_fail_over_mac
997 * Perform special MAC address swapping for fail_over_mac settings
999 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1001 static void bond_do_fail_over_mac(struct bonding *bond,
1002 struct slave *new_active,
1003 struct slave *old_active)
1004 __releases(&bond->curr_slave_lock)
1005 __releases(&bond->lock)
1006 __acquires(&bond->lock)
1007 __acquires(&bond->curr_slave_lock)
1009 u8 tmp_mac[ETH_ALEN];
1010 struct sockaddr saddr;
1013 switch (bond->params.fail_over_mac) {
1014 case BOND_FOM_ACTIVE:
1016 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1017 new_active->dev->addr_len);
1019 case BOND_FOM_FOLLOW:
1021 * if new_active && old_active, swap them
1022 * if just old_active, do nothing (going to no active slave)
1023 * if just new_active, set new_active to bond's MAC
1028 write_unlock_bh(&bond->curr_slave_lock);
1029 read_unlock(&bond->lock);
1032 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1033 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1035 saddr.sa_family = new_active->dev->type;
1037 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1038 saddr.sa_family = bond->dev->type;
1041 rv = dev_set_mac_address(new_active->dev, &saddr);
1044 ": %s: Error %d setting MAC of slave %s\n",
1045 bond->dev->name, -rv, new_active->dev->name);
1052 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1053 saddr.sa_family = old_active->dev->type;
1055 rv = dev_set_mac_address(old_active->dev, &saddr);
1058 ": %s: Error %d setting MAC of slave %s\n",
1059 bond->dev->name, -rv, new_active->dev->name);
1061 read_lock(&bond->lock);
1062 write_lock_bh(&bond->curr_slave_lock);
1066 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1067 bond->dev->name, bond->params.fail_over_mac);
1075 * find_best_interface - select the best available slave to be the active one
1076 * @bond: our bonding struct
1078 * Warning: Caller must hold curr_slave_lock for writing.
1080 static struct slave *bond_find_best_slave(struct bonding *bond)
1082 struct slave *new_active, *old_active;
1083 struct slave *bestslave = NULL;
1084 int mintime = bond->params.updelay;
1087 new_active = old_active = bond->curr_active_slave;
1089 if (!new_active) { /* there were no active slaves left */
1090 if (bond->slave_cnt > 0) /* found one slave */
1091 new_active = bond->first_slave;
1093 return NULL; /* still no slave, return NULL */
1097 * first try the primary link; if arping, a link must tx/rx
1098 * traffic before it can be considered the curr_active_slave.
1099 * also, we would skip slaves between the curr_active_slave
1100 * and primary_slave that may be up and able to arp
1102 if ((bond->primary_slave) &&
1103 (!bond->params.arp_interval) &&
1104 (IS_UP(bond->primary_slave->dev))) {
1105 new_active = bond->primary_slave;
1108 /* remember where to stop iterating over the slaves */
1109 old_active = new_active;
1111 bond_for_each_slave_from(bond, new_active, i, old_active) {
1112 if (IS_UP(new_active->dev)) {
1113 if (new_active->link == BOND_LINK_UP) {
1115 } else if (new_active->link == BOND_LINK_BACK) {
1116 /* link up, but waiting for stabilization */
1117 if (new_active->delay < mintime) {
1118 mintime = new_active->delay;
1119 bestslave = new_active;
1129 * change_active_interface - change the active slave into the specified one
1130 * @bond: our bonding struct
1131 * @new: the new slave to make the active one
1133 * Set the new slave to the bond's settings and unset them on the old
1134 * curr_active_slave.
1135 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1137 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1138 * because it is apparently the best available slave we have, even though its
1139 * updelay hasn't timed out yet.
1141 * If new_active is not NULL, caller must hold bond->lock for read and
1142 * curr_slave_lock for write_bh.
1144 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1146 struct slave *old_active = bond->curr_active_slave;
1148 if (old_active == new_active)
1152 new_active->jiffies = jiffies;
1154 if (new_active->link == BOND_LINK_BACK) {
1155 if (USES_PRIMARY(bond->params.mode)) {
1157 ": %s: making interface %s the new "
1158 "active one %d ms earlier.\n",
1159 bond->dev->name, new_active->dev->name,
1160 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1163 new_active->delay = 0;
1164 new_active->link = BOND_LINK_UP;
1166 if (bond->params.mode == BOND_MODE_8023AD)
1167 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1169 if (bond_is_lb(bond))
1170 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1172 if (USES_PRIMARY(bond->params.mode)) {
1174 ": %s: making interface %s the new "
1176 bond->dev->name, new_active->dev->name);
1181 if (USES_PRIMARY(bond->params.mode))
1182 bond_mc_swap(bond, new_active, old_active);
1184 if (bond_is_lb(bond)) {
1185 bond_alb_handle_active_change(bond, new_active);
1187 bond_set_slave_inactive_flags(old_active);
1189 bond_set_slave_active_flags(new_active);
1191 bond->curr_active_slave = new_active;
1194 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1196 bond_set_slave_inactive_flags(old_active);
1199 bond_set_slave_active_flags(new_active);
1201 if (bond->params.fail_over_mac)
1202 bond_do_fail_over_mac(bond, new_active,
1205 bond->send_grat_arp = bond->params.num_grat_arp;
1206 bond_send_gratuitous_arp(bond);
1208 bond->send_unsol_na = bond->params.num_unsol_na;
1209 bond_send_unsolicited_na(bond);
1211 write_unlock_bh(&bond->curr_slave_lock);
1212 read_unlock(&bond->lock);
1214 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1216 read_lock(&bond->lock);
1217 write_lock_bh(&bond->curr_slave_lock);
1223 * bond_select_active_slave - select a new active slave, if needed
1224 * @bond: our bonding struct
1226 * This functions should be called when one of the following occurs:
1227 * - The old curr_active_slave has been released or lost its link.
1228 * - The primary_slave has got its link back.
1229 * - A slave has got its link back and there's no old curr_active_slave.
1231 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1233 void bond_select_active_slave(struct bonding *bond)
1235 struct slave *best_slave;
1238 best_slave = bond_find_best_slave(bond);
1239 if (best_slave != bond->curr_active_slave) {
1240 bond_change_active_slave(bond, best_slave);
1241 rv = bond_set_carrier(bond);
1245 if (netif_carrier_ok(bond->dev)) {
1247 ": %s: first active interface up!\n",
1250 pr_info(DRV_NAME ": %s: "
1251 "now running without any active interface !\n",
1257 /*--------------------------- slave list handling ---------------------------*/
1260 * This function attaches the slave to the end of list.
1262 * bond->lock held for writing by caller.
1264 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1266 if (bond->first_slave == NULL) { /* attaching the first slave */
1267 new_slave->next = new_slave;
1268 new_slave->prev = new_slave;
1269 bond->first_slave = new_slave;
1271 new_slave->next = bond->first_slave;
1272 new_slave->prev = bond->first_slave->prev;
1273 new_slave->next->prev = new_slave;
1274 new_slave->prev->next = new_slave;
1281 * This function detaches the slave from the list.
1282 * WARNING: no check is made to verify if the slave effectively
1283 * belongs to <bond>.
1284 * Nothing is freed on return, structures are just unchained.
1285 * If any slave pointer in bond was pointing to <slave>,
1286 * it should be changed by the calling function.
1288 * bond->lock held for writing by caller.
1290 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1293 slave->next->prev = slave->prev;
1296 slave->prev->next = slave->next;
1298 if (bond->first_slave == slave) { /* slave is the first slave */
1299 if (bond->slave_cnt > 1) { /* there are more slave */
1300 bond->first_slave = slave->next;
1302 bond->first_slave = NULL; /* slave was the last one */
1311 /*---------------------------------- IOCTL ----------------------------------*/
1313 static int bond_sethwaddr(struct net_device *bond_dev,
1314 struct net_device *slave_dev)
1316 pr_debug("bond_dev=%p\n", bond_dev);
1317 pr_debug("slave_dev=%p\n", slave_dev);
1318 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1319 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1323 #define BOND_VLAN_FEATURES \
1324 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1325 NETIF_F_HW_VLAN_FILTER)
1328 * Compute the common dev->feature set available to all slaves. Some
1329 * feature bits are managed elsewhere, so preserve those feature bits
1330 * on the master device.
1332 static int bond_compute_features(struct bonding *bond)
1334 struct slave *slave;
1335 struct net_device *bond_dev = bond->dev;
1336 unsigned long features = bond_dev->features;
1337 unsigned long vlan_features = 0;
1338 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1339 bond_dev->hard_header_len);
1342 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1343 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1345 if (!bond->first_slave)
1348 features &= ~NETIF_F_ONE_FOR_ALL;
1350 vlan_features = bond->first_slave->dev->vlan_features;
1351 bond_for_each_slave(bond, slave, i) {
1352 features = netdev_increment_features(features,
1353 slave->dev->features,
1354 NETIF_F_ONE_FOR_ALL);
1355 vlan_features = netdev_increment_features(vlan_features,
1356 slave->dev->vlan_features,
1357 NETIF_F_ONE_FOR_ALL);
1358 if (slave->dev->hard_header_len > max_hard_header_len)
1359 max_hard_header_len = slave->dev->hard_header_len;
1363 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1364 bond_dev->features = netdev_fix_features(features, NULL);
1365 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1366 bond_dev->hard_header_len = max_hard_header_len;
1371 static void bond_setup_by_slave(struct net_device *bond_dev,
1372 struct net_device *slave_dev)
1374 struct bonding *bond = netdev_priv(bond_dev);
1376 bond_dev->header_ops = slave_dev->header_ops;
1378 bond_dev->type = slave_dev->type;
1379 bond_dev->hard_header_len = slave_dev->hard_header_len;
1380 bond_dev->addr_len = slave_dev->addr_len;
1382 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1383 slave_dev->addr_len);
1384 bond->setup_by_slave = 1;
1387 /* enslave device <slave> to bond device <master> */
1388 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1390 struct bonding *bond = netdev_priv(bond_dev);
1391 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1392 struct slave *new_slave = NULL;
1393 struct dev_mc_list *dmi;
1394 struct sockaddr addr;
1396 int old_features = bond_dev->features;
1399 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1400 slave_ops->ndo_do_ioctl == NULL) {
1402 ": %s: Warning: no link monitoring support for %s\n",
1403 bond_dev->name, slave_dev->name);
1406 /* bond must be initialized by bond_open() before enslaving */
1407 if (!(bond_dev->flags & IFF_UP)) {
1409 " %s: master_dev is not up in bond_enslave\n",
1413 /* already enslaved */
1414 if (slave_dev->flags & IFF_SLAVE) {
1415 pr_debug("Error, Device was already enslaved\n");
1419 /* vlan challenged mutual exclusion */
1420 /* no need to lock since we're protected by rtnl_lock */
1421 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1422 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1423 if (!list_empty(&bond->vlan_list)) {
1425 ": %s: Error: cannot enslave VLAN "
1426 "challenged slave %s on VLAN enabled "
1427 "bond %s\n", bond_dev->name, slave_dev->name,
1432 ": %s: Warning: enslaved VLAN challenged "
1433 "slave %s. Adding VLANs will be blocked as "
1434 "long as %s is part of bond %s\n",
1435 bond_dev->name, slave_dev->name, slave_dev->name,
1437 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1440 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1441 if (bond->slave_cnt == 0) {
1442 /* First slave, and it is not VLAN challenged,
1443 * so remove the block of adding VLANs over the bond.
1445 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1450 * Old ifenslave binaries are no longer supported. These can
1451 * be identified with moderate accuracy by the state of the slave:
1452 * the current ifenslave will set the interface down prior to
1453 * enslaving it; the old ifenslave will not.
1455 if ((slave_dev->flags & IFF_UP)) {
1456 pr_err(DRV_NAME ": %s is up. "
1457 "This may be due to an out of date ifenslave.\n",
1460 goto err_undo_flags;
1463 /* set bonding device ether type by slave - bonding netdevices are
1464 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1465 * there is a need to override some of the type dependent attribs/funcs.
1467 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1468 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1470 if (bond->slave_cnt == 0) {
1471 if (bond_dev->type != slave_dev->type) {
1472 pr_debug("%s: change device type from %d to %d\n",
1473 bond_dev->name, bond_dev->type, slave_dev->type);
1475 netdev_bonding_change(bond_dev, NETDEV_BONDING_OLDTYPE);
1477 if (slave_dev->type != ARPHRD_ETHER)
1478 bond_setup_by_slave(bond_dev, slave_dev);
1480 ether_setup(bond_dev);
1482 netdev_bonding_change(bond_dev, NETDEV_BONDING_NEWTYPE);
1484 } else if (bond_dev->type != slave_dev->type) {
1485 pr_err(DRV_NAME ": %s ether type (%d) is different "
1486 "from other slaves (%d), can not enslave it.\n",
1488 slave_dev->type, bond_dev->type);
1490 goto err_undo_flags;
1493 if (slave_ops->ndo_set_mac_address == NULL) {
1494 if (bond->slave_cnt == 0) {
1496 ": %s: Warning: The first slave device "
1497 "specified does not support setting the MAC "
1498 "address. Setting fail_over_mac to active.",
1500 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1501 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1503 ": %s: Error: The slave device specified "
1504 "does not support setting the MAC address, "
1505 "but fail_over_mac is not set to active.\n"
1508 goto err_undo_flags;
1512 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1515 goto err_undo_flags;
1518 /* save slave's original flags before calling
1519 * netdev_set_master and dev_open
1521 new_slave->original_flags = slave_dev->flags;
1524 * Save slave's original ("permanent") mac address for modes
1525 * that need it, and for restoring it upon release, and then
1526 * set it to the master's address
1528 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1530 if (!bond->params.fail_over_mac) {
1532 * Set slave to master's mac address. The application already
1533 * set the master's mac address to that of the first slave
1535 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1536 addr.sa_family = slave_dev->type;
1537 res = dev_set_mac_address(slave_dev, &addr);
1539 pr_debug("Error %d calling set_mac_address\n", res);
1544 res = netdev_set_master(slave_dev, bond_dev);
1546 pr_debug("Error %d calling netdev_set_master\n", res);
1547 goto err_restore_mac;
1549 /* open the slave since the application closed it */
1550 res = dev_open(slave_dev);
1552 pr_debug("Opening slave %s failed\n", slave_dev->name);
1553 goto err_unset_master;
1556 new_slave->dev = slave_dev;
1557 slave_dev->priv_flags |= IFF_BONDING;
1559 if (bond_is_lb(bond)) {
1560 /* bond_alb_init_slave() must be called before all other stages since
1561 * it might fail and we do not want to have to undo everything
1563 res = bond_alb_init_slave(bond, new_slave);
1568 /* If the mode USES_PRIMARY, then the new slave gets the
1569 * master's promisc (and mc) settings only if it becomes the
1570 * curr_active_slave, and that is taken care of later when calling
1571 * bond_change_active()
1573 if (!USES_PRIMARY(bond->params.mode)) {
1574 /* set promiscuity level to new slave */
1575 if (bond_dev->flags & IFF_PROMISC) {
1576 res = dev_set_promiscuity(slave_dev, 1);
1581 /* set allmulti level to new slave */
1582 if (bond_dev->flags & IFF_ALLMULTI) {
1583 res = dev_set_allmulti(slave_dev, 1);
1588 netif_addr_lock_bh(bond_dev);
1589 /* upload master's mc_list to new slave */
1590 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1591 dev_mc_add(slave_dev, dmi->dmi_addr,
1592 dmi->dmi_addrlen, 0);
1593 netif_addr_unlock_bh(bond_dev);
1596 if (bond->params.mode == BOND_MODE_8023AD) {
1597 /* add lacpdu mc addr to mc list */
1598 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1600 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1603 bond_add_vlans_on_slave(bond, slave_dev);
1605 write_lock_bh(&bond->lock);
1607 bond_attach_slave(bond, new_slave);
1609 new_slave->delay = 0;
1610 new_slave->link_failure_count = 0;
1612 bond_compute_features(bond);
1614 write_unlock_bh(&bond->lock);
1616 read_lock(&bond->lock);
1618 new_slave->last_arp_rx = jiffies;
1620 if (bond->params.miimon && !bond->params.use_carrier) {
1621 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1623 if ((link_reporting == -1) && !bond->params.arp_interval) {
1625 * miimon is set but a bonded network driver
1626 * does not support ETHTOOL/MII and
1627 * arp_interval is not set. Note: if
1628 * use_carrier is enabled, we will never go
1629 * here (because netif_carrier is always
1630 * supported); thus, we don't need to change
1631 * the messages for netif_carrier.
1634 ": %s: Warning: MII and ETHTOOL support not "
1635 "available for interface %s, and "
1636 "arp_interval/arp_ip_target module parameters "
1637 "not specified, thus bonding will not detect "
1638 "link failures! see bonding.txt for details.\n",
1639 bond_dev->name, slave_dev->name);
1640 } else if (link_reporting == -1) {
1641 /* unable get link status using mii/ethtool */
1643 ": %s: Warning: can't get link status from "
1644 "interface %s; the network driver associated "
1645 "with this interface does not support MII or "
1646 "ETHTOOL link status reporting, thus miimon "
1647 "has no effect on this interface.\n",
1648 bond_dev->name, slave_dev->name);
1652 /* check for initial state */
1653 if (!bond->params.miimon ||
1654 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1655 if (bond->params.updelay) {
1656 pr_debug("Initial state of slave_dev is "
1657 "BOND_LINK_BACK\n");
1658 new_slave->link = BOND_LINK_BACK;
1659 new_slave->delay = bond->params.updelay;
1661 pr_debug("Initial state of slave_dev is "
1663 new_slave->link = BOND_LINK_UP;
1665 new_slave->jiffies = jiffies;
1667 pr_debug("Initial state of slave_dev is "
1668 "BOND_LINK_DOWN\n");
1669 new_slave->link = BOND_LINK_DOWN;
1672 if (bond_update_speed_duplex(new_slave) &&
1673 (new_slave->link != BOND_LINK_DOWN)) {
1675 ": %s: Warning: failed to get speed and duplex from %s, "
1676 "assumed to be 100Mb/sec and Full.\n",
1677 bond_dev->name, new_slave->dev->name);
1679 if (bond->params.mode == BOND_MODE_8023AD) {
1681 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1682 "support in base driver for proper aggregator "
1683 "selection.\n", bond_dev->name);
1687 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1688 /* if there is a primary slave, remember it */
1689 if (strcmp(bond->params.primary, new_slave->dev->name) == 0)
1690 bond->primary_slave = new_slave;
1693 write_lock_bh(&bond->curr_slave_lock);
1695 switch (bond->params.mode) {
1696 case BOND_MODE_ACTIVEBACKUP:
1697 bond_set_slave_inactive_flags(new_slave);
1698 bond_select_active_slave(bond);
1700 case BOND_MODE_8023AD:
1701 /* in 802.3ad mode, the internal mechanism
1702 * will activate the slaves in the selected
1705 bond_set_slave_inactive_flags(new_slave);
1706 /* if this is the first slave */
1707 if (bond->slave_cnt == 1) {
1708 SLAVE_AD_INFO(new_slave).id = 1;
1709 /* Initialize AD with the number of times that the AD timer is called in 1 second
1710 * can be called only after the mac address of the bond is set
1712 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1713 bond->params.lacp_fast);
1715 SLAVE_AD_INFO(new_slave).id =
1716 SLAVE_AD_INFO(new_slave->prev).id + 1;
1719 bond_3ad_bind_slave(new_slave);
1723 new_slave->state = BOND_STATE_ACTIVE;
1724 bond_set_slave_inactive_flags(new_slave);
1725 bond_select_active_slave(bond);
1728 pr_debug("This slave is always active in trunk mode\n");
1730 /* always active in trunk mode */
1731 new_slave->state = BOND_STATE_ACTIVE;
1733 /* In trunking mode there is little meaning to curr_active_slave
1734 * anyway (it holds no special properties of the bond device),
1735 * so we can change it without calling change_active_interface()
1737 if (!bond->curr_active_slave)
1738 bond->curr_active_slave = new_slave;
1741 } /* switch(bond_mode) */
1743 write_unlock_bh(&bond->curr_slave_lock);
1745 bond_set_carrier(bond);
1747 read_unlock(&bond->lock);
1749 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1754 ": %s: enslaving %s as a%s interface with a%s link.\n",
1755 bond_dev->name, slave_dev->name,
1756 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1757 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1759 /* enslave is successful */
1762 /* Undo stages on error */
1764 dev_close(slave_dev);
1767 netdev_set_master(slave_dev, NULL);
1770 if (!bond->params.fail_over_mac) {
1771 /* XXX TODO - fom follow mode needs to change master's
1772 * MAC if this slave's MAC is in use by the bond, or at
1773 * least print a warning.
1775 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1776 addr.sa_family = slave_dev->type;
1777 dev_set_mac_address(slave_dev, &addr);
1784 bond_dev->features = old_features;
1790 * Try to release the slave device <slave> from the bond device <master>
1791 * It is legal to access curr_active_slave without a lock because all the function
1794 * The rules for slave state should be:
1795 * for Active/Backup:
1796 * Active stays on all backups go down
1797 * for Bonded connections:
1798 * The first up interface should be left on and all others downed.
1800 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1802 struct bonding *bond = netdev_priv(bond_dev);
1803 struct slave *slave, *oldcurrent;
1804 struct sockaddr addr;
1806 /* slave is not a slave or master is not master of this slave */
1807 if (!(slave_dev->flags & IFF_SLAVE) ||
1808 (slave_dev->master != bond_dev)) {
1810 ": %s: Error: cannot release %s.\n",
1811 bond_dev->name, slave_dev->name);
1815 write_lock_bh(&bond->lock);
1817 slave = bond_get_slave_by_dev(bond, slave_dev);
1819 /* not a slave of this bond */
1821 ": %s: %s not enslaved\n",
1822 bond_dev->name, slave_dev->name);
1823 write_unlock_bh(&bond->lock);
1827 if (!bond->params.fail_over_mac) {
1828 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr)
1829 && bond->slave_cnt > 1)
1831 ": %s: Warning: the permanent HWaddr of %s - "
1832 "%pM - is still in use by %s. "
1833 "Set the HWaddr of %s to a different address "
1834 "to avoid conflicts.\n",
1835 bond_dev->name, slave_dev->name,
1837 bond_dev->name, slave_dev->name);
1840 /* Inform AD package of unbinding of slave. */
1841 if (bond->params.mode == BOND_MODE_8023AD) {
1842 /* must be called before the slave is
1843 * detached from the list
1845 bond_3ad_unbind_slave(slave);
1849 ": %s: releasing %s interface %s\n",
1851 (slave->state == BOND_STATE_ACTIVE)
1852 ? "active" : "backup",
1855 oldcurrent = bond->curr_active_slave;
1857 bond->current_arp_slave = NULL;
1859 /* release the slave from its bond */
1860 bond_detach_slave(bond, slave);
1862 bond_compute_features(bond);
1864 if (bond->primary_slave == slave)
1865 bond->primary_slave = NULL;
1867 if (oldcurrent == slave)
1868 bond_change_active_slave(bond, NULL);
1870 if (bond_is_lb(bond)) {
1871 /* Must be called only after the slave has been
1872 * detached from the list and the curr_active_slave
1873 * has been cleared (if our_slave == old_current),
1874 * but before a new active slave is selected.
1876 write_unlock_bh(&bond->lock);
1877 bond_alb_deinit_slave(bond, slave);
1878 write_lock_bh(&bond->lock);
1881 if (oldcurrent == slave) {
1883 * Note that we hold RTNL over this sequence, so there
1884 * is no concern that another slave add/remove event
1887 write_unlock_bh(&bond->lock);
1888 read_lock(&bond->lock);
1889 write_lock_bh(&bond->curr_slave_lock);
1891 bond_select_active_slave(bond);
1893 write_unlock_bh(&bond->curr_slave_lock);
1894 read_unlock(&bond->lock);
1895 write_lock_bh(&bond->lock);
1898 if (bond->slave_cnt == 0) {
1899 bond_set_carrier(bond);
1901 /* if the last slave was removed, zero the mac address
1902 * of the master so it will be set by the application
1903 * to the mac address of the first slave
1905 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1907 if (list_empty(&bond->vlan_list)) {
1908 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1911 ": %s: Warning: clearing HW address of %s while it "
1912 "still has VLANs.\n",
1913 bond_dev->name, bond_dev->name);
1915 ": %s: When re-adding slaves, make sure the bond's "
1916 "HW address matches its VLANs'.\n",
1919 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1920 !bond_has_challenged_slaves(bond)) {
1922 ": %s: last VLAN challenged slave %s "
1923 "left bond %s. VLAN blocking is removed\n",
1924 bond_dev->name, slave_dev->name, bond_dev->name);
1925 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1928 write_unlock_bh(&bond->lock);
1930 /* must do this from outside any spinlocks */
1931 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1933 bond_del_vlans_from_slave(bond, slave_dev);
1935 /* If the mode USES_PRIMARY, then we should only remove its
1936 * promisc and mc settings if it was the curr_active_slave, but that was
1937 * already taken care of above when we detached the slave
1939 if (!USES_PRIMARY(bond->params.mode)) {
1940 /* unset promiscuity level from slave */
1941 if (bond_dev->flags & IFF_PROMISC)
1942 dev_set_promiscuity(slave_dev, -1);
1944 /* unset allmulti level from slave */
1945 if (bond_dev->flags & IFF_ALLMULTI)
1946 dev_set_allmulti(slave_dev, -1);
1948 /* flush master's mc_list from slave */
1949 netif_addr_lock_bh(bond_dev);
1950 bond_mc_list_flush(bond_dev, slave_dev);
1951 netif_addr_unlock_bh(bond_dev);
1954 netdev_set_master(slave_dev, NULL);
1956 /* close slave before restoring its mac address */
1957 dev_close(slave_dev);
1959 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1960 /* restore original ("permanent") mac address */
1961 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1962 addr.sa_family = slave_dev->type;
1963 dev_set_mac_address(slave_dev, &addr);
1966 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1967 IFF_SLAVE_INACTIVE | IFF_BONDING |
1972 return 0; /* deletion OK */
1976 * Destroy a bonding device.
1977 * Must be under rtnl_lock when this function is called.
1979 static void bond_uninit(struct net_device *bond_dev)
1981 struct bonding *bond = netdev_priv(bond_dev);
1983 bond_deinit(bond_dev);
1984 bond_destroy_sysfs_entry(bond);
1987 destroy_workqueue(bond->wq);
1989 netif_addr_lock_bh(bond_dev);
1990 bond_mc_list_destroy(bond);
1991 netif_addr_unlock_bh(bond_dev);
1995 * First release a slave and than destroy the bond if no more slaves are left.
1996 * Must be under rtnl_lock when this function is called.
1998 int bond_release_and_destroy(struct net_device *bond_dev,
1999 struct net_device *slave_dev)
2001 struct bonding *bond = netdev_priv(bond_dev);
2004 ret = bond_release(bond_dev, slave_dev);
2005 if ((ret == 0) && (bond->slave_cnt == 0)) {
2006 pr_info(DRV_NAME ": %s: destroying bond %s.\n",
2007 bond_dev->name, bond_dev->name);
2008 unregister_netdevice(bond_dev);
2014 * This function releases all slaves.
2016 static int bond_release_all(struct net_device *bond_dev)
2018 struct bonding *bond = netdev_priv(bond_dev);
2019 struct slave *slave;
2020 struct net_device *slave_dev;
2021 struct sockaddr addr;
2023 write_lock_bh(&bond->lock);
2025 netif_carrier_off(bond_dev);
2027 if (bond->slave_cnt == 0)
2030 bond->current_arp_slave = NULL;
2031 bond->primary_slave = NULL;
2032 bond_change_active_slave(bond, NULL);
2034 while ((slave = bond->first_slave) != NULL) {
2035 /* Inform AD package of unbinding of slave
2036 * before slave is detached from the list.
2038 if (bond->params.mode == BOND_MODE_8023AD)
2039 bond_3ad_unbind_slave(slave);
2041 slave_dev = slave->dev;
2042 bond_detach_slave(bond, slave);
2044 /* now that the slave is detached, unlock and perform
2045 * all the undo steps that should not be called from
2048 write_unlock_bh(&bond->lock);
2050 if (bond_is_lb(bond)) {
2051 /* must be called only after the slave
2052 * has been detached from the list
2054 bond_alb_deinit_slave(bond, slave);
2057 bond_compute_features(bond);
2059 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2060 bond_del_vlans_from_slave(bond, slave_dev);
2062 /* If the mode USES_PRIMARY, then we should only remove its
2063 * promisc and mc settings if it was the curr_active_slave, but that was
2064 * already taken care of above when we detached the slave
2066 if (!USES_PRIMARY(bond->params.mode)) {
2067 /* unset promiscuity level from slave */
2068 if (bond_dev->flags & IFF_PROMISC)
2069 dev_set_promiscuity(slave_dev, -1);
2071 /* unset allmulti level from slave */
2072 if (bond_dev->flags & IFF_ALLMULTI)
2073 dev_set_allmulti(slave_dev, -1);
2075 /* flush master's mc_list from slave */
2076 netif_addr_lock_bh(bond_dev);
2077 bond_mc_list_flush(bond_dev, slave_dev);
2078 netif_addr_unlock_bh(bond_dev);
2081 netdev_set_master(slave_dev, NULL);
2083 /* close slave before restoring its mac address */
2084 dev_close(slave_dev);
2086 if (!bond->params.fail_over_mac) {
2087 /* restore original ("permanent") mac address*/
2088 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2089 addr.sa_family = slave_dev->type;
2090 dev_set_mac_address(slave_dev, &addr);
2093 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2094 IFF_SLAVE_INACTIVE);
2098 /* re-acquire the lock before getting the next slave */
2099 write_lock_bh(&bond->lock);
2102 /* zero the mac address of the master so it will be
2103 * set by the application to the mac address of the
2106 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2108 if (list_empty(&bond->vlan_list))
2109 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2112 ": %s: Warning: clearing HW address of %s while it "
2113 "still has VLANs.\n",
2114 bond_dev->name, bond_dev->name);
2116 ": %s: When re-adding slaves, make sure the bond's "
2117 "HW address matches its VLANs'.\n",
2122 ": %s: released all slaves\n",
2126 write_unlock_bh(&bond->lock);
2132 * This function changes the active slave to slave <slave_dev>.
2133 * It returns -EINVAL in the following cases.
2134 * - <slave_dev> is not found in the list.
2135 * - There is not active slave now.
2136 * - <slave_dev> is already active.
2137 * - The link state of <slave_dev> is not BOND_LINK_UP.
2138 * - <slave_dev> is not running.
2139 * In these cases, this function does nothing.
2140 * In the other cases, current_slave pointer is changed and 0 is returned.
2142 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2144 struct bonding *bond = netdev_priv(bond_dev);
2145 struct slave *old_active = NULL;
2146 struct slave *new_active = NULL;
2149 if (!USES_PRIMARY(bond->params.mode))
2152 /* Verify that master_dev is indeed the master of slave_dev */
2153 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2156 read_lock(&bond->lock);
2158 read_lock(&bond->curr_slave_lock);
2159 old_active = bond->curr_active_slave;
2160 read_unlock(&bond->curr_slave_lock);
2162 new_active = bond_get_slave_by_dev(bond, slave_dev);
2165 * Changing to the current active: do nothing; return success.
2167 if (new_active && (new_active == old_active)) {
2168 read_unlock(&bond->lock);
2174 (new_active->link == BOND_LINK_UP) &&
2175 IS_UP(new_active->dev)) {
2176 write_lock_bh(&bond->curr_slave_lock);
2177 bond_change_active_slave(bond, new_active);
2178 write_unlock_bh(&bond->curr_slave_lock);
2182 read_unlock(&bond->lock);
2187 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2189 struct bonding *bond = netdev_priv(bond_dev);
2191 info->bond_mode = bond->params.mode;
2192 info->miimon = bond->params.miimon;
2194 read_lock(&bond->lock);
2195 info->num_slaves = bond->slave_cnt;
2196 read_unlock(&bond->lock);
2201 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2203 struct bonding *bond = netdev_priv(bond_dev);
2204 struct slave *slave;
2205 int i, res = -ENODEV;
2207 read_lock(&bond->lock);
2209 bond_for_each_slave(bond, slave, i) {
2210 if (i == (int)info->slave_id) {
2212 strcpy(info->slave_name, slave->dev->name);
2213 info->link = slave->link;
2214 info->state = slave->state;
2215 info->link_failure_count = slave->link_failure_count;
2220 read_unlock(&bond->lock);
2225 /*-------------------------------- Monitoring -------------------------------*/
2228 static int bond_miimon_inspect(struct bonding *bond)
2230 struct slave *slave;
2231 int i, link_state, commit = 0;
2232 bool ignore_updelay;
2234 ignore_updelay = !bond->curr_active_slave ? true : false;
2236 bond_for_each_slave(bond, slave, i) {
2237 slave->new_link = BOND_LINK_NOCHANGE;
2239 link_state = bond_check_dev_link(bond, slave->dev, 0);
2241 switch (slave->link) {
2246 slave->link = BOND_LINK_FAIL;
2247 slave->delay = bond->params.downdelay;
2250 ": %s: link status down for %s"
2251 "interface %s, disabling it in %d ms.\n",
2253 (bond->params.mode ==
2254 BOND_MODE_ACTIVEBACKUP) ?
2255 ((slave->state == BOND_STATE_ACTIVE) ?
2256 "active " : "backup ") : "",
2258 bond->params.downdelay * bond->params.miimon);
2261 case BOND_LINK_FAIL:
2264 * recovered before downdelay expired
2266 slave->link = BOND_LINK_UP;
2267 slave->jiffies = jiffies;
2269 ": %s: link status up again after %d "
2270 "ms for interface %s.\n",
2272 (bond->params.downdelay - slave->delay) *
2273 bond->params.miimon,
2278 if (slave->delay <= 0) {
2279 slave->new_link = BOND_LINK_DOWN;
2287 case BOND_LINK_DOWN:
2291 slave->link = BOND_LINK_BACK;
2292 slave->delay = bond->params.updelay;
2296 ": %s: link status up for "
2297 "interface %s, enabling it in %d ms.\n",
2298 bond->dev->name, slave->dev->name,
2299 ignore_updelay ? 0 :
2300 bond->params.updelay *
2301 bond->params.miimon);
2304 case BOND_LINK_BACK:
2306 slave->link = BOND_LINK_DOWN;
2308 ": %s: link status down again after %d "
2309 "ms for interface %s.\n",
2311 (bond->params.updelay - slave->delay) *
2312 bond->params.miimon,
2321 if (slave->delay <= 0) {
2322 slave->new_link = BOND_LINK_UP;
2324 ignore_updelay = false;
2336 static void bond_miimon_commit(struct bonding *bond)
2338 struct slave *slave;
2341 bond_for_each_slave(bond, slave, i) {
2342 switch (slave->new_link) {
2343 case BOND_LINK_NOCHANGE:
2347 slave->link = BOND_LINK_UP;
2348 slave->jiffies = jiffies;
2350 if (bond->params.mode == BOND_MODE_8023AD) {
2351 /* prevent it from being the active one */
2352 slave->state = BOND_STATE_BACKUP;
2353 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2354 /* make it immediately active */
2355 slave->state = BOND_STATE_ACTIVE;
2356 } else if (slave != bond->primary_slave) {
2357 /* prevent it from being the active one */
2358 slave->state = BOND_STATE_BACKUP;
2362 ": %s: link status definitely "
2363 "up for interface %s.\n",
2364 bond->dev->name, slave->dev->name);
2366 /* notify ad that the link status has changed */
2367 if (bond->params.mode == BOND_MODE_8023AD)
2368 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2370 if (bond_is_lb(bond))
2371 bond_alb_handle_link_change(bond, slave,
2374 if (!bond->curr_active_slave ||
2375 (slave == bond->primary_slave))
2380 case BOND_LINK_DOWN:
2381 if (slave->link_failure_count < UINT_MAX)
2382 slave->link_failure_count++;
2384 slave->link = BOND_LINK_DOWN;
2386 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2387 bond->params.mode == BOND_MODE_8023AD)
2388 bond_set_slave_inactive_flags(slave);
2391 ": %s: link status definitely down for "
2392 "interface %s, disabling it\n",
2393 bond->dev->name, slave->dev->name);
2395 if (bond->params.mode == BOND_MODE_8023AD)
2396 bond_3ad_handle_link_change(slave,
2399 if (bond_is_lb(bond))
2400 bond_alb_handle_link_change(bond, slave,
2403 if (slave == bond->curr_active_slave)
2410 ": %s: invalid new link %d on slave %s\n",
2411 bond->dev->name, slave->new_link,
2413 slave->new_link = BOND_LINK_NOCHANGE;
2420 write_lock_bh(&bond->curr_slave_lock);
2421 bond_select_active_slave(bond);
2422 write_unlock_bh(&bond->curr_slave_lock);
2425 bond_set_carrier(bond);
2431 * Really a wrapper that splits the mii monitor into two phases: an
2432 * inspection, then (if inspection indicates something needs to be done)
2433 * an acquisition of appropriate locks followed by a commit phase to
2434 * implement whatever link state changes are indicated.
2436 void bond_mii_monitor(struct work_struct *work)
2438 struct bonding *bond = container_of(work, struct bonding,
2441 read_lock(&bond->lock);
2442 if (bond->kill_timers)
2445 if (bond->slave_cnt == 0)
2448 if (bond->send_grat_arp) {
2449 read_lock(&bond->curr_slave_lock);
2450 bond_send_gratuitous_arp(bond);
2451 read_unlock(&bond->curr_slave_lock);
2454 if (bond->send_unsol_na) {
2455 read_lock(&bond->curr_slave_lock);
2456 bond_send_unsolicited_na(bond);
2457 read_unlock(&bond->curr_slave_lock);
2460 if (bond_miimon_inspect(bond)) {
2461 read_unlock(&bond->lock);
2463 read_lock(&bond->lock);
2465 bond_miimon_commit(bond);
2467 read_unlock(&bond->lock);
2468 rtnl_unlock(); /* might sleep, hold no other locks */
2469 read_lock(&bond->lock);
2473 if (bond->params.miimon)
2474 queue_delayed_work(bond->wq, &bond->mii_work,
2475 msecs_to_jiffies(bond->params.miimon));
2477 read_unlock(&bond->lock);
2480 static __be32 bond_glean_dev_ip(struct net_device *dev)
2482 struct in_device *idev;
2483 struct in_ifaddr *ifa;
2490 idev = __in_dev_get_rcu(dev);
2494 ifa = idev->ifa_list;
2498 addr = ifa->ifa_local;
2504 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2506 struct vlan_entry *vlan;
2508 if (ip == bond->master_ip)
2511 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2512 if (ip == vlan->vlan_ip)
2520 * We go to the (large) trouble of VLAN tagging ARP frames because
2521 * switches in VLAN mode (especially if ports are configured as
2522 * "native" to a VLAN) might not pass non-tagged frames.
2524 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2526 struct sk_buff *skb;
2528 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2529 slave_dev->name, dest_ip, src_ip, vlan_id);
2531 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2532 NULL, slave_dev->dev_addr, NULL);
2535 pr_err(DRV_NAME ": ARP packet allocation failed\n");
2539 skb = vlan_put_tag(skb, vlan_id);
2541 pr_err(DRV_NAME ": failed to insert VLAN tag\n");
2549 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2552 __be32 *targets = bond->params.arp_targets;
2553 struct vlan_entry *vlan;
2554 struct net_device *vlan_dev;
2558 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2561 pr_debug("basa: target %x\n", targets[i]);
2562 if (list_empty(&bond->vlan_list)) {
2563 pr_debug("basa: empty vlan: arp_send\n");
2564 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2565 bond->master_ip, 0);
2570 * If VLANs are configured, we do a route lookup to
2571 * determine which VLAN interface would be used, so we
2572 * can tag the ARP with the proper VLAN tag.
2574 memset(&fl, 0, sizeof(fl));
2575 fl.fl4_dst = targets[i];
2576 fl.fl4_tos = RTO_ONLINK;
2578 rv = ip_route_output_key(&init_net, &rt, &fl);
2580 if (net_ratelimit()) {
2582 ": %s: no route to arp_ip_target %pI4\n",
2583 bond->dev->name, &fl.fl4_dst);
2589 * This target is not on a VLAN
2591 if (rt->u.dst.dev == bond->dev) {
2593 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2594 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2595 bond->master_ip, 0);
2600 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2601 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2602 if (vlan_dev == rt->u.dst.dev) {
2603 vlan_id = vlan->vlan_id;
2604 pr_debug("basa: vlan match on %s %d\n",
2605 vlan_dev->name, vlan_id);
2612 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2613 vlan->vlan_ip, vlan_id);
2617 if (net_ratelimit()) {
2619 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2620 bond->dev->name, &fl.fl4_dst,
2621 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2628 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2629 * for each VLAN above us.
2631 * Caller must hold curr_slave_lock for read or better
2633 static void bond_send_gratuitous_arp(struct bonding *bond)
2635 struct slave *slave = bond->curr_active_slave;
2636 struct vlan_entry *vlan;
2637 struct net_device *vlan_dev;
2639 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2640 slave ? slave->dev->name : "NULL");
2642 if (!slave || !bond->send_grat_arp ||
2643 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2646 bond->send_grat_arp--;
2648 if (bond->master_ip) {
2649 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2650 bond->master_ip, 0);
2653 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2654 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2655 if (vlan->vlan_ip) {
2656 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2657 vlan->vlan_ip, vlan->vlan_id);
2662 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2665 __be32 *targets = bond->params.arp_targets;
2667 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2668 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2669 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2670 if (sip == targets[i]) {
2671 if (bond_has_this_ip(bond, tip))
2672 slave->last_arp_rx = jiffies;
2678 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2681 struct slave *slave;
2682 struct bonding *bond;
2683 unsigned char *arp_ptr;
2686 if (dev_net(dev) != &init_net)
2689 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2692 bond = netdev_priv(dev);
2693 read_lock(&bond->lock);
2695 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2696 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2697 orig_dev ? orig_dev->name : "NULL");
2699 slave = bond_get_slave_by_dev(bond, orig_dev);
2700 if (!slave || !slave_do_arp_validate(bond, slave))
2703 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2707 if (arp->ar_hln != dev->addr_len ||
2708 skb->pkt_type == PACKET_OTHERHOST ||
2709 skb->pkt_type == PACKET_LOOPBACK ||
2710 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2711 arp->ar_pro != htons(ETH_P_IP) ||
2715 arp_ptr = (unsigned char *)(arp + 1);
2716 arp_ptr += dev->addr_len;
2717 memcpy(&sip, arp_ptr, 4);
2718 arp_ptr += 4 + dev->addr_len;
2719 memcpy(&tip, arp_ptr, 4);
2721 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2722 bond->dev->name, slave->dev->name, slave->state,
2723 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2727 * Backup slaves won't see the ARP reply, but do come through
2728 * here for each ARP probe (so we swap the sip/tip to validate
2729 * the probe). In a "redundant switch, common router" type of
2730 * configuration, the ARP probe will (hopefully) travel from
2731 * the active, through one switch, the router, then the other
2732 * switch before reaching the backup.
2734 if (slave->state == BOND_STATE_ACTIVE)
2735 bond_validate_arp(bond, slave, sip, tip);
2737 bond_validate_arp(bond, slave, tip, sip);
2740 read_unlock(&bond->lock);
2743 return NET_RX_SUCCESS;
2747 * this function is called regularly to monitor each slave's link
2748 * ensuring that traffic is being sent and received when arp monitoring
2749 * is used in load-balancing mode. if the adapter has been dormant, then an
2750 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2751 * arp monitoring in active backup mode.
2753 void bond_loadbalance_arp_mon(struct work_struct *work)
2755 struct bonding *bond = container_of(work, struct bonding,
2757 struct slave *slave, *oldcurrent;
2758 int do_failover = 0;
2762 read_lock(&bond->lock);
2764 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2766 if (bond->kill_timers)
2769 if (bond->slave_cnt == 0)
2772 read_lock(&bond->curr_slave_lock);
2773 oldcurrent = bond->curr_active_slave;
2774 read_unlock(&bond->curr_slave_lock);
2776 /* see if any of the previous devices are up now (i.e. they have
2777 * xmt and rcv traffic). the curr_active_slave does not come into
2778 * the picture unless it is null. also, slave->jiffies is not needed
2779 * here because we send an arp on each slave and give a slave as
2780 * long as it needs to get the tx/rx within the delta.
2781 * TODO: what about up/down delay in arp mode? it wasn't here before
2784 bond_for_each_slave(bond, slave, i) {
2785 if (slave->link != BOND_LINK_UP) {
2786 if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2787 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2789 slave->link = BOND_LINK_UP;
2790 slave->state = BOND_STATE_ACTIVE;
2792 /* primary_slave has no meaning in round-robin
2793 * mode. the window of a slave being up and
2794 * curr_active_slave being null after enslaving
2799 ": %s: link status definitely "
2800 "up for interface %s, ",
2806 ": %s: interface %s is now up\n",
2812 /* slave->link == BOND_LINK_UP */
2814 /* not all switches will respond to an arp request
2815 * when the source ip is 0, so don't take the link down
2816 * if we don't know our ip yet
2818 if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2819 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2821 slave->link = BOND_LINK_DOWN;
2822 slave->state = BOND_STATE_BACKUP;
2824 if (slave->link_failure_count < UINT_MAX)
2825 slave->link_failure_count++;
2828 ": %s: interface %s is now down.\n",
2832 if (slave == oldcurrent)
2837 /* note: if switch is in round-robin mode, all links
2838 * must tx arp to ensure all links rx an arp - otherwise
2839 * links may oscillate or not come up at all; if switch is
2840 * in something like xor mode, there is nothing we can
2841 * do - all replies will be rx'ed on same link causing slaves
2842 * to be unstable during low/no traffic periods
2844 if (IS_UP(slave->dev))
2845 bond_arp_send_all(bond, slave);
2849 write_lock_bh(&bond->curr_slave_lock);
2851 bond_select_active_slave(bond);
2853 write_unlock_bh(&bond->curr_slave_lock);
2857 if (bond->params.arp_interval)
2858 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2860 read_unlock(&bond->lock);
2864 * Called to inspect slaves for active-backup mode ARP monitor link state
2865 * changes. Sets new_link in slaves to specify what action should take
2866 * place for the slave. Returns 0 if no changes are found, >0 if changes
2867 * to link states must be committed.
2869 * Called with bond->lock held for read.
2871 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2873 struct slave *slave;
2876 bond_for_each_slave(bond, slave, i) {
2877 slave->new_link = BOND_LINK_NOCHANGE;
2879 if (slave->link != BOND_LINK_UP) {
2880 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2882 slave->new_link = BOND_LINK_UP;
2890 * Give slaves 2*delta after being enslaved or made
2891 * active. This avoids bouncing, as the last receive
2892 * times need a full ARP monitor cycle to be updated.
2894 if (!time_after_eq(jiffies, slave->jiffies +
2895 2 * delta_in_ticks))
2899 * Backup slave is down if:
2900 * - No current_arp_slave AND
2901 * - more than 3*delta since last receive AND
2902 * - the bond has an IP address
2904 * Note: a non-null current_arp_slave indicates
2905 * the curr_active_slave went down and we are
2906 * searching for a new one; under this condition
2907 * we only take the curr_active_slave down - this
2908 * gives each slave a chance to tx/rx traffic
2909 * before being taken out
2911 if (slave->state == BOND_STATE_BACKUP &&
2912 !bond->current_arp_slave &&
2913 time_after(jiffies, slave_last_rx(bond, slave) +
2914 3 * delta_in_ticks)) {
2915 slave->new_link = BOND_LINK_DOWN;
2920 * Active slave is down if:
2921 * - more than 2*delta since transmitting OR
2922 * - (more than 2*delta since receive AND
2923 * the bond has an IP address)
2925 if ((slave->state == BOND_STATE_ACTIVE) &&
2926 (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2927 2 * delta_in_ticks) ||
2928 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2929 + 2 * delta_in_ticks)))) {
2930 slave->new_link = BOND_LINK_DOWN;
2935 read_lock(&bond->curr_slave_lock);
2938 * Trigger a commit if the primary option setting has changed.
2940 if (bond->primary_slave &&
2941 (bond->primary_slave != bond->curr_active_slave) &&
2942 (bond->primary_slave->link == BOND_LINK_UP))
2945 read_unlock(&bond->curr_slave_lock);
2951 * Called to commit link state changes noted by inspection step of
2952 * active-backup mode ARP monitor.
2954 * Called with RTNL and bond->lock for read.
2956 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2958 struct slave *slave;
2961 bond_for_each_slave(bond, slave, i) {
2962 switch (slave->new_link) {
2963 case BOND_LINK_NOCHANGE:
2967 write_lock_bh(&bond->curr_slave_lock);
2969 if (!bond->curr_active_slave &&
2970 time_before_eq(jiffies, dev_trans_start(slave->dev) +
2972 slave->link = BOND_LINK_UP;
2973 bond_change_active_slave(bond, slave);
2974 bond->current_arp_slave = NULL;
2977 ": %s: %s is up and now the "
2978 "active interface\n",
2979 bond->dev->name, slave->dev->name);
2981 } else if (bond->curr_active_slave != slave) {
2982 /* this slave has just come up but we
2983 * already have a current slave; this can
2984 * also happen if bond_enslave adds a new
2985 * slave that is up while we are searching
2988 slave->link = BOND_LINK_UP;
2989 bond_set_slave_inactive_flags(slave);
2990 bond->current_arp_slave = NULL;
2993 ": %s: backup interface %s is now up\n",
2994 bond->dev->name, slave->dev->name);
2997 write_unlock_bh(&bond->curr_slave_lock);
3001 case BOND_LINK_DOWN:
3002 if (slave->link_failure_count < UINT_MAX)
3003 slave->link_failure_count++;
3005 slave->link = BOND_LINK_DOWN;
3007 if (slave == bond->curr_active_slave) {
3009 ": %s: link status down for active "
3010 "interface %s, disabling it\n",
3011 bond->dev->name, slave->dev->name);
3013 bond_set_slave_inactive_flags(slave);
3015 write_lock_bh(&bond->curr_slave_lock);
3017 bond_select_active_slave(bond);
3018 if (bond->curr_active_slave)
3019 bond->curr_active_slave->jiffies =
3022 write_unlock_bh(&bond->curr_slave_lock);
3024 bond->current_arp_slave = NULL;
3026 } else if (slave->state == BOND_STATE_BACKUP) {
3028 ": %s: backup interface %s is now down\n",
3029 bond->dev->name, slave->dev->name);
3031 bond_set_slave_inactive_flags(slave);
3037 ": %s: impossible: new_link %d on slave %s\n",
3038 bond->dev->name, slave->new_link,
3044 * No race with changes to primary via sysfs, as we hold rtnl.
3046 if (bond->primary_slave &&
3047 (bond->primary_slave != bond->curr_active_slave) &&
3048 (bond->primary_slave->link == BOND_LINK_UP)) {
3049 write_lock_bh(&bond->curr_slave_lock);
3050 bond_change_active_slave(bond, bond->primary_slave);
3051 write_unlock_bh(&bond->curr_slave_lock);
3054 bond_set_carrier(bond);
3058 * Send ARP probes for active-backup mode ARP monitor.
3060 * Called with bond->lock held for read.
3062 static void bond_ab_arp_probe(struct bonding *bond)
3064 struct slave *slave;
3067 read_lock(&bond->curr_slave_lock);
3069 if (bond->current_arp_slave && bond->curr_active_slave)
3070 pr_info(DRV_NAME "PROBE: c_arp %s && cas %s BAD\n",
3071 bond->current_arp_slave->dev->name,
3072 bond->curr_active_slave->dev->name);
3074 if (bond->curr_active_slave) {
3075 bond_arp_send_all(bond, bond->curr_active_slave);
3076 read_unlock(&bond->curr_slave_lock);
3080 read_unlock(&bond->curr_slave_lock);
3082 /* if we don't have a curr_active_slave, search for the next available
3083 * backup slave from the current_arp_slave and make it the candidate
3084 * for becoming the curr_active_slave
3087 if (!bond->current_arp_slave) {
3088 bond->current_arp_slave = bond->first_slave;
3089 if (!bond->current_arp_slave)
3093 bond_set_slave_inactive_flags(bond->current_arp_slave);
3095 /* search for next candidate */
3096 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3097 if (IS_UP(slave->dev)) {
3098 slave->link = BOND_LINK_BACK;
3099 bond_set_slave_active_flags(slave);
3100 bond_arp_send_all(bond, slave);
3101 slave->jiffies = jiffies;
3102 bond->current_arp_slave = slave;
3106 /* if the link state is up at this point, we
3107 * mark it down - this can happen if we have
3108 * simultaneous link failures and
3109 * reselect_active_interface doesn't make this
3110 * one the current slave so it is still marked
3111 * up when it is actually down
3113 if (slave->link == BOND_LINK_UP) {
3114 slave->link = BOND_LINK_DOWN;
3115 if (slave->link_failure_count < UINT_MAX)
3116 slave->link_failure_count++;
3118 bond_set_slave_inactive_flags(slave);
3121 ": %s: backup interface %s is now down.\n",
3122 bond->dev->name, slave->dev->name);
3127 void bond_activebackup_arp_mon(struct work_struct *work)
3129 struct bonding *bond = container_of(work, struct bonding,
3133 read_lock(&bond->lock);
3135 if (bond->kill_timers)
3138 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3140 if (bond->slave_cnt == 0)
3143 if (bond->send_grat_arp) {
3144 read_lock(&bond->curr_slave_lock);
3145 bond_send_gratuitous_arp(bond);
3146 read_unlock(&bond->curr_slave_lock);
3149 if (bond->send_unsol_na) {
3150 read_lock(&bond->curr_slave_lock);
3151 bond_send_unsolicited_na(bond);
3152 read_unlock(&bond->curr_slave_lock);
3155 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3156 read_unlock(&bond->lock);
3158 read_lock(&bond->lock);
3160 bond_ab_arp_commit(bond, delta_in_ticks);
3162 read_unlock(&bond->lock);
3164 read_lock(&bond->lock);
3167 bond_ab_arp_probe(bond);
3170 if (bond->params.arp_interval)
3171 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3173 read_unlock(&bond->lock);
3176 /*------------------------------ proc/seq_file-------------------------------*/
3178 #ifdef CONFIG_PROC_FS
3180 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3181 __acquires(&dev_base_lock)
3182 __acquires(&bond->lock)
3184 struct bonding *bond = seq->private;
3186 struct slave *slave;
3189 /* make sure the bond won't be taken away */
3190 read_lock(&dev_base_lock);
3191 read_lock(&bond->lock);
3194 return SEQ_START_TOKEN;
3196 bond_for_each_slave(bond, slave, i) {
3204 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3206 struct bonding *bond = seq->private;
3207 struct slave *slave = v;
3210 if (v == SEQ_START_TOKEN)
3211 return bond->first_slave;
3213 slave = slave->next;
3215 return (slave == bond->first_slave) ? NULL : slave;
3218 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3219 __releases(&bond->lock)
3220 __releases(&dev_base_lock)
3222 struct bonding *bond = seq->private;
3224 read_unlock(&bond->lock);
3225 read_unlock(&dev_base_lock);
3228 static void bond_info_show_master(struct seq_file *seq)
3230 struct bonding *bond = seq->private;
3234 read_lock(&bond->curr_slave_lock);
3235 curr = bond->curr_active_slave;
3236 read_unlock(&bond->curr_slave_lock);
3238 seq_printf(seq, "Bonding Mode: %s",
3239 bond_mode_name(bond->params.mode));
3241 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3242 bond->params.fail_over_mac)
3243 seq_printf(seq, " (fail_over_mac %s)",
3244 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3246 seq_printf(seq, "\n");
3248 if (bond->params.mode == BOND_MODE_XOR ||
3249 bond->params.mode == BOND_MODE_8023AD) {
3250 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3251 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3252 bond->params.xmit_policy);
3255 if (USES_PRIMARY(bond->params.mode)) {
3256 seq_printf(seq, "Primary Slave: %s\n",
3257 (bond->primary_slave) ?
3258 bond->primary_slave->dev->name : "None");
3260 seq_printf(seq, "Currently Active Slave: %s\n",
3261 (curr) ? curr->dev->name : "None");
3264 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3266 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3267 seq_printf(seq, "Up Delay (ms): %d\n",
3268 bond->params.updelay * bond->params.miimon);
3269 seq_printf(seq, "Down Delay (ms): %d\n",
3270 bond->params.downdelay * bond->params.miimon);
3273 /* ARP information */
3274 if (bond->params.arp_interval > 0) {
3276 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3277 bond->params.arp_interval);
3279 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3281 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3282 if (!bond->params.arp_targets[i])
3285 seq_printf(seq, ",");
3286 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3289 seq_printf(seq, "\n");
3292 if (bond->params.mode == BOND_MODE_8023AD) {
3293 struct ad_info ad_info;
3295 seq_puts(seq, "\n802.3ad info\n");
3296 seq_printf(seq, "LACP rate: %s\n",
3297 (bond->params.lacp_fast) ? "fast" : "slow");
3298 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3299 ad_select_tbl[bond->params.ad_select].modename);
3301 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3302 seq_printf(seq, "bond %s has no active aggregator\n",
3305 seq_printf(seq, "Active Aggregator Info:\n");
3307 seq_printf(seq, "\tAggregator ID: %d\n",
3308 ad_info.aggregator_id);
3309 seq_printf(seq, "\tNumber of ports: %d\n",
3311 seq_printf(seq, "\tActor Key: %d\n",
3313 seq_printf(seq, "\tPartner Key: %d\n",
3314 ad_info.partner_key);
3315 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3316 ad_info.partner_system);
3321 static void bond_info_show_slave(struct seq_file *seq,
3322 const struct slave *slave)
3324 struct bonding *bond = seq->private;
3326 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3327 seq_printf(seq, "MII Status: %s\n",
3328 (slave->link == BOND_LINK_UP) ? "up" : "down");
3329 seq_printf(seq, "Link Failure Count: %u\n",
3330 slave->link_failure_count);
3332 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3334 if (bond->params.mode == BOND_MODE_8023AD) {
3335 const struct aggregator *agg
3336 = SLAVE_AD_INFO(slave).port.aggregator;
3339 seq_printf(seq, "Aggregator ID: %d\n",
3340 agg->aggregator_identifier);
3342 seq_puts(seq, "Aggregator ID: N/A\n");
3346 static int bond_info_seq_show(struct seq_file *seq, void *v)
3348 if (v == SEQ_START_TOKEN) {
3349 seq_printf(seq, "%s\n", version);
3350 bond_info_show_master(seq);
3352 bond_info_show_slave(seq, v);
3357 static const struct seq_operations bond_info_seq_ops = {
3358 .start = bond_info_seq_start,
3359 .next = bond_info_seq_next,
3360 .stop = bond_info_seq_stop,
3361 .show = bond_info_seq_show,
3364 static int bond_info_open(struct inode *inode, struct file *file)
3366 struct seq_file *seq;
3367 struct proc_dir_entry *proc;
3370 res = seq_open(file, &bond_info_seq_ops);
3372 /* recover the pointer buried in proc_dir_entry data */
3373 seq = file->private_data;
3375 seq->private = proc->data;
3381 static const struct file_operations bond_info_fops = {
3382 .owner = THIS_MODULE,
3383 .open = bond_info_open,
3385 .llseek = seq_lseek,
3386 .release = seq_release,
3389 static int bond_create_proc_entry(struct bonding *bond)
3391 struct net_device *bond_dev = bond->dev;
3393 if (bond_proc_dir) {
3394 bond->proc_entry = proc_create_data(bond_dev->name,
3395 S_IRUGO, bond_proc_dir,
3396 &bond_info_fops, bond);
3397 if (bond->proc_entry == NULL)
3399 ": Warning: Cannot create /proc/net/%s/%s\n",
3400 DRV_NAME, bond_dev->name);
3402 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3408 static void bond_remove_proc_entry(struct bonding *bond)
3410 if (bond_proc_dir && bond->proc_entry) {
3411 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3412 memset(bond->proc_file_name, 0, IFNAMSIZ);
3413 bond->proc_entry = NULL;
3417 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3418 * Caller must hold rtnl_lock.
3420 static void bond_create_proc_dir(void)
3422 if (!bond_proc_dir) {
3423 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3426 ": Warning: cannot create /proc/net/%s\n",
3431 /* Destroy the bonding directory under /proc/net, if empty.
3432 * Caller must hold rtnl_lock.
3434 static void bond_destroy_proc_dir(void)
3436 if (bond_proc_dir) {
3437 remove_proc_entry(DRV_NAME, init_net.proc_net);
3438 bond_proc_dir = NULL;
3442 #else /* !CONFIG_PROC_FS */
3444 static int bond_create_proc_entry(struct bonding *bond)
3448 static void bond_remove_proc_entry(struct bonding *bond)
3452 static void bond_create_proc_dir(void)
3456 static void bond_destroy_proc_dir(void)
3460 #endif /* CONFIG_PROC_FS */
3463 /*-------------------------- netdev event handling --------------------------*/
3466 * Change device name
3468 static int bond_event_changename(struct bonding *bond)
3470 bond_remove_proc_entry(bond);
3471 bond_create_proc_entry(bond);
3473 bond_destroy_sysfs_entry(bond);
3474 bond_create_sysfs_entry(bond);
3479 static int bond_master_netdev_event(unsigned long event,
3480 struct net_device *bond_dev)
3482 struct bonding *event_bond = netdev_priv(bond_dev);
3485 case NETDEV_CHANGENAME:
3486 return bond_event_changename(event_bond);
3487 case NETDEV_UNREGISTER:
3488 bond_release_all(event_bond->dev);
3497 static int bond_slave_netdev_event(unsigned long event,
3498 struct net_device *slave_dev)
3500 struct net_device *bond_dev = slave_dev->master;
3501 struct bonding *bond = netdev_priv(bond_dev);
3504 case NETDEV_UNREGISTER:
3506 if (bond->setup_by_slave)
3507 bond_release_and_destroy(bond_dev, slave_dev);
3509 bond_release(bond_dev, slave_dev);
3513 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3514 struct slave *slave;
3516 slave = bond_get_slave_by_dev(bond, slave_dev);
3518 u16 old_speed = slave->speed;
3519 u16 old_duplex = slave->duplex;
3521 bond_update_speed_duplex(slave);
3523 if (bond_is_lb(bond))
3526 if (old_speed != slave->speed)
3527 bond_3ad_adapter_speed_changed(slave);
3528 if (old_duplex != slave->duplex)
3529 bond_3ad_adapter_duplex_changed(slave);
3536 * ... Or is it this?
3539 case NETDEV_CHANGEMTU:
3541 * TODO: Should slaves be allowed to
3542 * independently alter their MTU? For
3543 * an active-backup bond, slaves need
3544 * not be the same type of device, so
3545 * MTUs may vary. For other modes,
3546 * slaves arguably should have the
3547 * same MTUs. To do this, we'd need to
3548 * take over the slave's change_mtu
3549 * function for the duration of their
3553 case NETDEV_CHANGENAME:
3555 * TODO: handle changing the primary's name
3558 case NETDEV_FEAT_CHANGE:
3559 bond_compute_features(bond);
3569 * bond_netdev_event: handle netdev notifier chain events.
3571 * This function receives events for the netdev chain. The caller (an
3572 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3573 * locks for us to safely manipulate the slave devices (RTNL lock,
3576 static int bond_netdev_event(struct notifier_block *this,
3577 unsigned long event, void *ptr)
3579 struct net_device *event_dev = (struct net_device *)ptr;
3581 if (dev_net(event_dev) != &init_net)
3584 pr_debug("event_dev: %s, event: %lx\n",
3585 (event_dev ? event_dev->name : "None"),
3588 if (!(event_dev->priv_flags & IFF_BONDING))
3591 if (event_dev->flags & IFF_MASTER) {
3592 pr_debug("IFF_MASTER\n");
3593 return bond_master_netdev_event(event, event_dev);
3596 if (event_dev->flags & IFF_SLAVE) {
3597 pr_debug("IFF_SLAVE\n");
3598 return bond_slave_netdev_event(event, event_dev);
3605 * bond_inetaddr_event: handle inetaddr notifier chain events.
3607 * We keep track of device IPs primarily to use as source addresses in
3608 * ARP monitor probes (rather than spewing out broadcasts all the time).
3610 * We track one IP for the main device (if it has one), plus one per VLAN.
3612 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3614 struct in_ifaddr *ifa = ptr;
3615 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3616 struct bonding *bond;
3617 struct vlan_entry *vlan;
3619 if (dev_net(ifa->ifa_dev->dev) != &init_net)
3622 list_for_each_entry(bond, &bond_dev_list, bond_list) {
3623 if (bond->dev == event_dev) {
3626 bond->master_ip = ifa->ifa_local;
3629 bond->master_ip = bond_glean_dev_ip(bond->dev);
3636 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3637 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3638 if (vlan_dev == event_dev) {
3641 vlan->vlan_ip = ifa->ifa_local;
3645 bond_glean_dev_ip(vlan_dev);
3656 static struct notifier_block bond_netdev_notifier = {
3657 .notifier_call = bond_netdev_event,
3660 static struct notifier_block bond_inetaddr_notifier = {
3661 .notifier_call = bond_inetaddr_event,
3664 /*-------------------------- Packet type handling ---------------------------*/
3666 /* register to receive lacpdus on a bond */
3667 static void bond_register_lacpdu(struct bonding *bond)
3669 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3671 /* initialize packet type */
3672 pk_type->type = PKT_TYPE_LACPDU;
3673 pk_type->dev = bond->dev;
3674 pk_type->func = bond_3ad_lacpdu_recv;
3676 dev_add_pack(pk_type);
3679 /* unregister to receive lacpdus on a bond */
3680 static void bond_unregister_lacpdu(struct bonding *bond)
3682 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3685 void bond_register_arp(struct bonding *bond)
3687 struct packet_type *pt = &bond->arp_mon_pt;
3692 pt->type = htons(ETH_P_ARP);
3693 pt->dev = bond->dev;
3694 pt->func = bond_arp_rcv;
3698 void bond_unregister_arp(struct bonding *bond)
3700 struct packet_type *pt = &bond->arp_mon_pt;
3702 dev_remove_pack(pt);
3706 /*---------------------------- Hashing Policies -----------------------------*/
3709 * Hash for the output device based upon layer 2 and layer 3 data. If
3710 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3712 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3713 struct net_device *bond_dev, int count)
3715 struct ethhdr *data = (struct ethhdr *)skb->data;
3716 struct iphdr *iph = ip_hdr(skb);
3718 if (skb->protocol == htons(ETH_P_IP)) {
3719 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3720 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3723 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3727 * Hash for the output device based upon layer 3 and layer 4 data. If
3728 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3729 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3731 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3732 struct net_device *bond_dev, int count)
3734 struct ethhdr *data = (struct ethhdr *)skb->data;
3735 struct iphdr *iph = ip_hdr(skb);
3736 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3739 if (skb->protocol == htons(ETH_P_IP)) {
3740 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3741 (iph->protocol == IPPROTO_TCP ||
3742 iph->protocol == IPPROTO_UDP)) {
3743 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3745 return (layer4_xor ^
3746 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3750 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3754 * Hash for the output device based upon layer 2 data
3756 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3757 struct net_device *bond_dev, int count)
3759 struct ethhdr *data = (struct ethhdr *)skb->data;
3761 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3764 /*-------------------------- Device entry points ----------------------------*/
3766 static int bond_open(struct net_device *bond_dev)
3768 struct bonding *bond = netdev_priv(bond_dev);
3770 bond->kill_timers = 0;
3772 if (bond_is_lb(bond)) {
3773 /* bond_alb_initialize must be called before the timer
3776 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3777 /* something went wrong - fail the open operation */
3781 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3782 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3785 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3786 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3787 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3790 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3791 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3792 INIT_DELAYED_WORK(&bond->arp_work,
3793 bond_activebackup_arp_mon);
3795 INIT_DELAYED_WORK(&bond->arp_work,
3796 bond_loadbalance_arp_mon);
3798 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3799 if (bond->params.arp_validate)
3800 bond_register_arp(bond);
3803 if (bond->params.mode == BOND_MODE_8023AD) {
3804 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3805 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3806 /* register to receive LACPDUs */
3807 bond_register_lacpdu(bond);
3808 bond_3ad_initiate_agg_selection(bond, 1);
3814 static int bond_close(struct net_device *bond_dev)
3816 struct bonding *bond = netdev_priv(bond_dev);
3818 if (bond->params.mode == BOND_MODE_8023AD) {
3819 /* Unregister the receive of LACPDUs */
3820 bond_unregister_lacpdu(bond);
3823 if (bond->params.arp_validate)
3824 bond_unregister_arp(bond);
3826 write_lock_bh(&bond->lock);
3828 bond->send_grat_arp = 0;
3829 bond->send_unsol_na = 0;
3831 /* signal timers not to re-arm */
3832 bond->kill_timers = 1;
3834 write_unlock_bh(&bond->lock);
3836 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3837 cancel_delayed_work(&bond->mii_work);
3840 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3841 cancel_delayed_work(&bond->arp_work);
3844 switch (bond->params.mode) {
3845 case BOND_MODE_8023AD:
3846 cancel_delayed_work(&bond->ad_work);
3850 cancel_delayed_work(&bond->alb_work);
3857 if (bond_is_lb(bond)) {
3858 /* Must be called only after all
3859 * slaves have been released
3861 bond_alb_deinitialize(bond);
3867 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3869 struct bonding *bond = netdev_priv(bond_dev);
3870 struct net_device_stats *stats = &bond->stats;
3871 struct net_device_stats local_stats;
3872 struct slave *slave;
3875 memset(&local_stats, 0, sizeof(struct net_device_stats));
3877 read_lock_bh(&bond->lock);
3879 bond_for_each_slave(bond, slave, i) {
3880 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3882 local_stats.rx_packets += sstats->rx_packets;
3883 local_stats.rx_bytes += sstats->rx_bytes;
3884 local_stats.rx_errors += sstats->rx_errors;
3885 local_stats.rx_dropped += sstats->rx_dropped;
3887 local_stats.tx_packets += sstats->tx_packets;
3888 local_stats.tx_bytes += sstats->tx_bytes;
3889 local_stats.tx_errors += sstats->tx_errors;
3890 local_stats.tx_dropped += sstats->tx_dropped;
3892 local_stats.multicast += sstats->multicast;
3893 local_stats.collisions += sstats->collisions;
3895 local_stats.rx_length_errors += sstats->rx_length_errors;
3896 local_stats.rx_over_errors += sstats->rx_over_errors;
3897 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3898 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3899 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3900 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3902 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3903 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3904 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3905 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3906 local_stats.tx_window_errors += sstats->tx_window_errors;
3909 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3911 read_unlock_bh(&bond->lock);
3916 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3918 struct net_device *slave_dev = NULL;
3919 struct ifbond k_binfo;
3920 struct ifbond __user *u_binfo = NULL;
3921 struct ifslave k_sinfo;
3922 struct ifslave __user *u_sinfo = NULL;
3923 struct mii_ioctl_data *mii = NULL;
3926 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3927 bond_dev->name, cmd);
3939 * We do this again just in case we were called by SIOCGMIIREG
3940 * instead of SIOCGMIIPHY.
3947 if (mii->reg_num == 1) {
3948 struct bonding *bond = netdev_priv(bond_dev);
3950 read_lock(&bond->lock);
3951 read_lock(&bond->curr_slave_lock);
3952 if (netif_carrier_ok(bond->dev))
3953 mii->val_out = BMSR_LSTATUS;
3955 read_unlock(&bond->curr_slave_lock);
3956 read_unlock(&bond->lock);
3960 case BOND_INFO_QUERY_OLD:
3961 case SIOCBONDINFOQUERY:
3962 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3964 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3967 res = bond_info_query(bond_dev, &k_binfo);
3969 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3973 case BOND_SLAVE_INFO_QUERY_OLD:
3974 case SIOCBONDSLAVEINFOQUERY:
3975 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3977 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3980 res = bond_slave_info_query(bond_dev, &k_sinfo);
3982 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3991 if (!capable(CAP_NET_ADMIN))
3994 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3996 pr_debug("slave_dev=%p: \n", slave_dev);
4001 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
4003 case BOND_ENSLAVE_OLD:
4004 case SIOCBONDENSLAVE:
4005 res = bond_enslave(bond_dev, slave_dev);
4007 case BOND_RELEASE_OLD:
4008 case SIOCBONDRELEASE:
4009 res = bond_release(bond_dev, slave_dev);
4011 case BOND_SETHWADDR_OLD:
4012 case SIOCBONDSETHWADDR:
4013 res = bond_sethwaddr(bond_dev, slave_dev);
4015 case BOND_CHANGE_ACTIVE_OLD:
4016 case SIOCBONDCHANGEACTIVE:
4017 res = bond_ioctl_change_active(bond_dev, slave_dev);
4029 static void bond_set_multicast_list(struct net_device *bond_dev)
4031 struct bonding *bond = netdev_priv(bond_dev);
4032 struct dev_mc_list *dmi;
4035 * Do promisc before checking multicast_mode
4037 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4039 * FIXME: Need to handle the error when one of the multi-slaves
4042 bond_set_promiscuity(bond, 1);
4045 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4046 bond_set_promiscuity(bond, -1);
4049 /* set allmulti flag to slaves */
4050 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4052 * FIXME: Need to handle the error when one of the multi-slaves
4055 bond_set_allmulti(bond, 1);
4058 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4059 bond_set_allmulti(bond, -1);
4062 read_lock(&bond->lock);
4064 bond->flags = bond_dev->flags;
4066 /* looking for addresses to add to slaves' mc list */
4067 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4068 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
4069 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4072 /* looking for addresses to delete from slaves' list */
4073 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4074 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
4075 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4078 /* save master's multicast list */
4079 bond_mc_list_destroy(bond);
4080 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4082 read_unlock(&bond->lock);
4085 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4087 struct bonding *bond = netdev_priv(dev);
4088 struct slave *slave = bond->first_slave;
4091 const struct net_device_ops *slave_ops
4092 = slave->dev->netdev_ops;
4093 if (slave_ops->ndo_neigh_setup)
4094 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4100 * Change the MTU of all of a master's slaves to match the master
4102 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4104 struct bonding *bond = netdev_priv(bond_dev);
4105 struct slave *slave, *stop_at;
4109 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4110 (bond_dev ? bond_dev->name : "None"), new_mtu);
4112 /* Can't hold bond->lock with bh disabled here since
4113 * some base drivers panic. On the other hand we can't
4114 * hold bond->lock without bh disabled because we'll
4115 * deadlock. The only solution is to rely on the fact
4116 * that we're under rtnl_lock here, and the slaves
4117 * list won't change. This doesn't solve the problem
4118 * of setting the slave's MTU while it is
4119 * transmitting, but the assumption is that the base
4120 * driver can handle that.
4122 * TODO: figure out a way to safely iterate the slaves
4123 * list, but without holding a lock around the actual
4124 * call to the base driver.
4127 bond_for_each_slave(bond, slave, i) {
4128 pr_debug("s %p s->p %p c_m %p\n", slave,
4129 slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
4131 res = dev_set_mtu(slave->dev, new_mtu);
4134 /* If we failed to set the slave's mtu to the new value
4135 * we must abort the operation even in ACTIVE_BACKUP
4136 * mode, because if we allow the backup slaves to have
4137 * different mtu values than the active slave we'll
4138 * need to change their mtu when doing a failover. That
4139 * means changing their mtu from timer context, which
4140 * is probably not a good idea.
4142 pr_debug("err %d %s\n", res, slave->dev->name);
4147 bond_dev->mtu = new_mtu;
4152 /* unwind from head to the slave that failed */
4154 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4157 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4159 pr_debug("unwind err %d dev %s\n", tmp_res,
4170 * Note that many devices must be down to change the HW address, and
4171 * downing the master releases all slaves. We can make bonds full of
4172 * bonding devices to test this, however.
4174 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4176 struct bonding *bond = netdev_priv(bond_dev);
4177 struct sockaddr *sa = addr, tmp_sa;
4178 struct slave *slave, *stop_at;
4182 if (bond->params.mode == BOND_MODE_ALB)
4183 return bond_alb_set_mac_address(bond_dev, addr);
4186 pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4189 * If fail_over_mac is set to active, do nothing and return
4190 * success. Returning an error causes ifenslave to fail.
4192 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4195 if (!is_valid_ether_addr(sa->sa_data))
4196 return -EADDRNOTAVAIL;
4198 /* Can't hold bond->lock with bh disabled here since
4199 * some base drivers panic. On the other hand we can't
4200 * hold bond->lock without bh disabled because we'll
4201 * deadlock. The only solution is to rely on the fact
4202 * that we're under rtnl_lock here, and the slaves
4203 * list won't change. This doesn't solve the problem
4204 * of setting the slave's hw address while it is
4205 * transmitting, but the assumption is that the base
4206 * driver can handle that.
4208 * TODO: figure out a way to safely iterate the slaves
4209 * list, but without holding a lock around the actual
4210 * call to the base driver.
4213 bond_for_each_slave(bond, slave, i) {
4214 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4215 pr_debug("slave %p %s\n", slave, slave->dev->name);
4217 if (slave_ops->ndo_set_mac_address == NULL) {
4219 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4223 res = dev_set_mac_address(slave->dev, addr);
4225 /* TODO: consider downing the slave
4227 * User should expect communications
4228 * breakage anyway until ARP finish
4231 pr_debug("err %d %s\n", res, slave->dev->name);
4237 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4241 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4242 tmp_sa.sa_family = bond_dev->type;
4244 /* unwind from head to the slave that failed */
4246 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4249 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4251 pr_debug("unwind err %d dev %s\n", tmp_res,
4259 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4261 struct bonding *bond = netdev_priv(bond_dev);
4262 struct slave *slave, *start_at;
4263 int i, slave_no, res = 1;
4265 read_lock(&bond->lock);
4267 if (!BOND_IS_OK(bond))
4271 * Concurrent TX may collide on rr_tx_counter; we accept that
4272 * as being rare enough not to justify using an atomic op here
4274 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4276 bond_for_each_slave(bond, slave, i) {
4283 bond_for_each_slave_from(bond, slave, i, start_at) {
4284 if (IS_UP(slave->dev) &&
4285 (slave->link == BOND_LINK_UP) &&
4286 (slave->state == BOND_STATE_ACTIVE)) {
4287 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4294 /* no suitable interface, frame not sent */
4297 read_unlock(&bond->lock);
4298 return NETDEV_TX_OK;
4303 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4304 * the bond has a usable interface.
4306 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4308 struct bonding *bond = netdev_priv(bond_dev);
4311 read_lock(&bond->lock);
4312 read_lock(&bond->curr_slave_lock);
4314 if (!BOND_IS_OK(bond))
4317 if (!bond->curr_active_slave)
4320 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4324 /* no suitable interface, frame not sent */
4327 read_unlock(&bond->curr_slave_lock);
4328 read_unlock(&bond->lock);
4329 return NETDEV_TX_OK;
4333 * In bond_xmit_xor() , we determine the output device by using a pre-
4334 * determined xmit_hash_policy(), If the selected device is not enabled,
4335 * find the next active slave.
4337 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4339 struct bonding *bond = netdev_priv(bond_dev);
4340 struct slave *slave, *start_at;
4345 read_lock(&bond->lock);
4347 if (!BOND_IS_OK(bond))
4350 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4352 bond_for_each_slave(bond, slave, i) {
4360 bond_for_each_slave_from(bond, slave, i, start_at) {
4361 if (IS_UP(slave->dev) &&
4362 (slave->link == BOND_LINK_UP) &&
4363 (slave->state == BOND_STATE_ACTIVE)) {
4364 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4371 /* no suitable interface, frame not sent */
4374 read_unlock(&bond->lock);
4375 return NETDEV_TX_OK;
4379 * in broadcast mode, we send everything to all usable interfaces.
4381 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4383 struct bonding *bond = netdev_priv(bond_dev);
4384 struct slave *slave, *start_at;
4385 struct net_device *tx_dev = NULL;
4389 read_lock(&bond->lock);
4391 if (!BOND_IS_OK(bond))
4394 read_lock(&bond->curr_slave_lock);
4395 start_at = bond->curr_active_slave;
4396 read_unlock(&bond->curr_slave_lock);
4401 bond_for_each_slave_from(bond, slave, i, start_at) {
4402 if (IS_UP(slave->dev) &&
4403 (slave->link == BOND_LINK_UP) &&
4404 (slave->state == BOND_STATE_ACTIVE)) {
4406 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4409 ": %s: Error: bond_xmit_broadcast(): "
4410 "skb_clone() failed\n",
4415 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4417 dev_kfree_skb(skb2);
4421 tx_dev = slave->dev;
4426 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4430 /* no suitable interface, frame not sent */
4433 /* frame sent to all suitable interfaces */
4434 read_unlock(&bond->lock);
4435 return NETDEV_TX_OK;
4438 /*------------------------- Device initialization ---------------------------*/
4440 static void bond_set_xmit_hash_policy(struct bonding *bond)
4442 switch (bond->params.xmit_policy) {
4443 case BOND_XMIT_POLICY_LAYER23:
4444 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4446 case BOND_XMIT_POLICY_LAYER34:
4447 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4449 case BOND_XMIT_POLICY_LAYER2:
4451 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4456 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4458 const struct bonding *bond = netdev_priv(dev);
4460 switch (bond->params.mode) {
4461 case BOND_MODE_ROUNDROBIN:
4462 return bond_xmit_roundrobin(skb, dev);
4463 case BOND_MODE_ACTIVEBACKUP:
4464 return bond_xmit_activebackup(skb, dev);
4466 return bond_xmit_xor(skb, dev);
4467 case BOND_MODE_BROADCAST:
4468 return bond_xmit_broadcast(skb, dev);
4469 case BOND_MODE_8023AD:
4470 return bond_3ad_xmit_xor(skb, dev);
4473 return bond_alb_xmit(skb, dev);
4475 /* Should never happen, mode already checked */
4476 pr_err(DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4477 dev->name, bond->params.mode);
4480 return NETDEV_TX_OK;
4486 * set bond mode specific net device operations
4488 void bond_set_mode_ops(struct bonding *bond, int mode)
4490 struct net_device *bond_dev = bond->dev;
4493 case BOND_MODE_ROUNDROBIN:
4495 case BOND_MODE_ACTIVEBACKUP:
4498 bond_set_xmit_hash_policy(bond);
4500 case BOND_MODE_BROADCAST:
4502 case BOND_MODE_8023AD:
4503 bond_set_master_3ad_flags(bond);
4504 bond_set_xmit_hash_policy(bond);
4507 bond_set_master_alb_flags(bond);
4512 /* Should never happen, mode already checked */
4514 ": %s: Error: Unknown bonding mode %d\n",
4521 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4522 struct ethtool_drvinfo *drvinfo)
4524 strncpy(drvinfo->driver, DRV_NAME, 32);
4525 strncpy(drvinfo->version, DRV_VERSION, 32);
4526 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4529 static const struct ethtool_ops bond_ethtool_ops = {
4530 .get_drvinfo = bond_ethtool_get_drvinfo,
4531 .get_link = ethtool_op_get_link,
4532 .get_tx_csum = ethtool_op_get_tx_csum,
4533 .get_sg = ethtool_op_get_sg,
4534 .get_tso = ethtool_op_get_tso,
4535 .get_ufo = ethtool_op_get_ufo,
4536 .get_flags = ethtool_op_get_flags,
4539 static const struct net_device_ops bond_netdev_ops = {
4540 .ndo_init = bond_init,
4541 .ndo_uninit = bond_uninit,
4542 .ndo_open = bond_open,
4543 .ndo_stop = bond_close,
4544 .ndo_start_xmit = bond_start_xmit,
4545 .ndo_get_stats = bond_get_stats,
4546 .ndo_do_ioctl = bond_do_ioctl,
4547 .ndo_set_multicast_list = bond_set_multicast_list,
4548 .ndo_change_mtu = bond_change_mtu,
4549 .ndo_set_mac_address = bond_set_mac_address,
4550 .ndo_neigh_setup = bond_neigh_setup,
4551 .ndo_vlan_rx_register = bond_vlan_rx_register,
4552 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4553 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4556 static void bond_setup(struct net_device *bond_dev)
4558 struct bonding *bond = netdev_priv(bond_dev);
4560 /* initialize rwlocks */
4561 rwlock_init(&bond->lock);
4562 rwlock_init(&bond->curr_slave_lock);
4564 bond->params = bonding_defaults;
4566 /* Initialize pointers */
4567 bond->dev = bond_dev;
4568 INIT_LIST_HEAD(&bond->vlan_list);
4570 /* Initialize the device entry points */
4571 ether_setup(bond_dev);
4572 bond_dev->netdev_ops = &bond_netdev_ops;
4573 bond_dev->ethtool_ops = &bond_ethtool_ops;
4574 bond_set_mode_ops(bond, bond->params.mode);
4576 bond_dev->destructor = free_netdev;
4578 /* Initialize the device options */
4579 bond_dev->tx_queue_len = 0;
4580 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4581 bond_dev->priv_flags |= IFF_BONDING;
4582 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4584 if (bond->params.arp_interval)
4585 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4587 /* At first, we block adding VLANs. That's the only way to
4588 * prevent problems that occur when adding VLANs over an
4589 * empty bond. The block will be removed once non-challenged
4590 * slaves are enslaved.
4592 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4594 /* don't acquire bond device's netif_tx_lock when
4596 bond_dev->features |= NETIF_F_LLTX;
4598 /* By default, we declare the bond to be fully
4599 * VLAN hardware accelerated capable. Special
4600 * care is taken in the various xmit functions
4601 * when there are slaves that are not hw accel
4604 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4605 NETIF_F_HW_VLAN_RX |
4606 NETIF_F_HW_VLAN_FILTER);
4610 static void bond_work_cancel_all(struct bonding *bond)
4612 write_lock_bh(&bond->lock);
4613 bond->kill_timers = 1;
4614 write_unlock_bh(&bond->lock);
4616 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4617 cancel_delayed_work(&bond->mii_work);
4619 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4620 cancel_delayed_work(&bond->arp_work);
4622 if (bond->params.mode == BOND_MODE_ALB &&
4623 delayed_work_pending(&bond->alb_work))
4624 cancel_delayed_work(&bond->alb_work);
4626 if (bond->params.mode == BOND_MODE_8023AD &&
4627 delayed_work_pending(&bond->ad_work))
4628 cancel_delayed_work(&bond->ad_work);
4631 /* De-initialize device specific data.
4632 * Caller must hold rtnl_lock.
4634 static void bond_deinit(struct net_device *bond_dev)
4636 struct bonding *bond = netdev_priv(bond_dev);
4638 list_del(&bond->bond_list);
4640 bond_work_cancel_all(bond);
4642 bond_remove_proc_entry(bond);
4645 /* Unregister and free all bond devices.
4646 * Caller must hold rtnl_lock.
4648 static void bond_free_all(void)
4650 struct bonding *bond, *nxt;
4652 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4653 struct net_device *bond_dev = bond->dev;
4655 bond_work_cancel_all(bond);
4656 /* Release the bonded slaves */
4657 bond_release_all(bond_dev);
4658 unregister_netdevice(bond_dev);
4661 bond_destroy_proc_dir();
4664 /*------------------------- Module initialization ---------------------------*/
4667 * Convert string input module parms. Accept either the
4668 * number of the mode or its string name. A bit complicated because
4669 * some mode names are substrings of other names, and calls from sysfs
4670 * may have whitespace in the name (trailing newlines, for example).
4672 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4674 int modeint = -1, i, rv;
4675 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4677 for (p = (char *)buf; *p; p++)
4678 if (!(isdigit(*p) || isspace(*p)))
4682 rv = sscanf(buf, "%20s", modestr);
4684 rv = sscanf(buf, "%d", &modeint);
4689 for (i = 0; tbl[i].modename; i++) {
4690 if (modeint == tbl[i].mode)
4692 if (strcmp(modestr, tbl[i].modename) == 0)
4699 static int bond_check_params(struct bond_params *params)
4701 int arp_validate_value, fail_over_mac_value;
4704 * Convert string parameters.
4707 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4708 if (bond_mode == -1) {
4710 ": Error: Invalid bonding mode \"%s\"\n",
4711 mode == NULL ? "NULL" : mode);
4716 if (xmit_hash_policy) {
4717 if ((bond_mode != BOND_MODE_XOR) &&
4718 (bond_mode != BOND_MODE_8023AD)) {
4720 ": xor_mode param is irrelevant in mode %s\n",
4721 bond_mode_name(bond_mode));
4723 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4725 if (xmit_hashtype == -1) {
4727 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4728 xmit_hash_policy == NULL ? "NULL" :
4736 if (bond_mode != BOND_MODE_8023AD) {
4738 ": lacp_rate param is irrelevant in mode %s\n",
4739 bond_mode_name(bond_mode));
4741 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4742 if (lacp_fast == -1) {
4744 ": Error: Invalid lacp rate \"%s\"\n",
4745 lacp_rate == NULL ? "NULL" : lacp_rate);
4752 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4753 if (params->ad_select == -1) {
4755 ": Error: Invalid ad_select \"%s\"\n",
4756 ad_select == NULL ? "NULL" : ad_select);
4760 if (bond_mode != BOND_MODE_8023AD) {
4762 ": ad_select param only affects 802.3ad mode\n");
4765 params->ad_select = BOND_AD_STABLE;
4768 if (max_bonds < 0) {
4770 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4771 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4772 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4773 max_bonds = BOND_DEFAULT_MAX_BONDS;
4778 ": Warning: miimon module parameter (%d), "
4779 "not in range 0-%d, so it was reset to %d\n",
4780 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4781 miimon = BOND_LINK_MON_INTERV;
4786 ": Warning: updelay module parameter (%d), "
4787 "not in range 0-%d, so it was reset to 0\n",
4792 if (downdelay < 0) {
4794 ": Warning: downdelay module parameter (%d), "
4795 "not in range 0-%d, so it was reset to 0\n",
4796 downdelay, INT_MAX);
4800 if ((use_carrier != 0) && (use_carrier != 1)) {
4802 ": Warning: use_carrier module parameter (%d), "
4803 "not of valid value (0/1), so it was set to 1\n",
4808 if (num_grat_arp < 0 || num_grat_arp > 255) {
4810 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4811 "was reset to 1 \n", num_grat_arp);
4815 if (num_unsol_na < 0 || num_unsol_na > 255) {
4817 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4818 "was reset to 1 \n", num_unsol_na);
4822 /* reset values for 802.3ad */
4823 if (bond_mode == BOND_MODE_8023AD) {
4826 ": Warning: miimon must be specified, "
4827 "otherwise bonding will not detect link "
4828 "failure, speed and duplex which are "
4829 "essential for 802.3ad operation\n");
4830 pr_warning("Forcing miimon to 100msec\n");
4835 /* reset values for TLB/ALB */
4836 if ((bond_mode == BOND_MODE_TLB) ||
4837 (bond_mode == BOND_MODE_ALB)) {
4840 ": Warning: miimon must be specified, "
4841 "otherwise bonding will not detect link "
4842 "failure and link speed which are essential "
4843 "for TLB/ALB load balancing\n");
4844 pr_warning("Forcing miimon to 100msec\n");
4849 if (bond_mode == BOND_MODE_ALB) {
4851 ": In ALB mode you might experience client "
4852 "disconnections upon reconnection of a link if the "
4853 "bonding module updelay parameter (%d msec) is "
4854 "incompatible with the forwarding delay time of the "
4860 if (updelay || downdelay) {
4861 /* just warn the user the up/down delay will have
4862 * no effect since miimon is zero...
4865 ": Warning: miimon module parameter not set "
4866 "and updelay (%d) or downdelay (%d) module "
4867 "parameter is set; updelay and downdelay have "
4868 "no effect unless miimon is set\n",
4869 updelay, downdelay);
4872 /* don't allow arp monitoring */
4875 ": Warning: miimon (%d) and arp_interval (%d) "
4876 "can't be used simultaneously, disabling ARP "
4878 miimon, arp_interval);
4882 if ((updelay % miimon) != 0) {
4884 ": Warning: updelay (%d) is not a multiple "
4885 "of miimon (%d), updelay rounded to %d ms\n",
4886 updelay, miimon, (updelay / miimon) * miimon);
4891 if ((downdelay % miimon) != 0) {
4893 ": Warning: downdelay (%d) is not a multiple "
4894 "of miimon (%d), downdelay rounded to %d ms\n",
4896 (downdelay / miimon) * miimon);
4899 downdelay /= miimon;
4902 if (arp_interval < 0) {
4904 ": Warning: arp_interval module parameter (%d) "
4905 ", not in range 0-%d, so it was reset to %d\n",
4906 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4907 arp_interval = BOND_LINK_ARP_INTERV;
4910 for (arp_ip_count = 0;
4911 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4913 /* not complete check, but should be good enough to
4915 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4917 ": Warning: bad arp_ip_target module parameter "
4918 "(%s), ARP monitoring will not be performed\n",
4919 arp_ip_target[arp_ip_count]);
4922 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4923 arp_target[arp_ip_count] = ip;
4927 if (arp_interval && !arp_ip_count) {
4928 /* don't allow arping if no arp_ip_target given... */
4930 ": Warning: arp_interval module parameter (%d) "
4931 "specified without providing an arp_ip_target "
4932 "parameter, arp_interval was reset to 0\n",
4938 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4940 ": arp_validate only supported in active-backup mode\n");
4943 if (!arp_interval) {
4945 ": arp_validate requires arp_interval\n");
4949 arp_validate_value = bond_parse_parm(arp_validate,
4951 if (arp_validate_value == -1) {
4953 ": Error: invalid arp_validate \"%s\"\n",
4954 arp_validate == NULL ? "NULL" : arp_validate);
4958 arp_validate_value = 0;
4962 ": MII link monitoring set to %d ms\n",
4964 } else if (arp_interval) {
4967 pr_info(DRV_NAME ": ARP monitoring set to %d ms,"
4968 " validate %s, with %d target(s):",
4970 arp_validate_tbl[arp_validate_value].modename,
4973 for (i = 0; i < arp_ip_count; i++)
4974 pr_info(" %s", arp_ip_target[i]);
4978 } else if (max_bonds) {
4979 /* miimon and arp_interval not set, we need one so things
4980 * work as expected, see bonding.txt for details
4983 ": Warning: either miimon or arp_interval and "
4984 "arp_ip_target module parameters must be specified, "
4985 "otherwise bonding will not detect link failures! see "
4986 "bonding.txt for details.\n");
4989 if (primary && !USES_PRIMARY(bond_mode)) {
4990 /* currently, using a primary only makes sense
4991 * in active backup, TLB or ALB modes
4994 ": Warning: %s primary device specified but has no "
4995 "effect in %s mode\n",
4996 primary, bond_mode_name(bond_mode));
5000 if (fail_over_mac) {
5001 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5003 if (fail_over_mac_value == -1) {
5005 ": Error: invalid fail_over_mac \"%s\"\n",
5006 arp_validate == NULL ? "NULL" : arp_validate);
5010 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5012 ": Warning: fail_over_mac only affects "
5013 "active-backup mode.\n");
5015 fail_over_mac_value = BOND_FOM_NONE;
5018 /* fill params struct with the proper values */
5019 params->mode = bond_mode;
5020 params->xmit_policy = xmit_hashtype;
5021 params->miimon = miimon;
5022 params->num_grat_arp = num_grat_arp;
5023 params->num_unsol_na = num_unsol_na;
5024 params->arp_interval = arp_interval;
5025 params->arp_validate = arp_validate_value;
5026 params->updelay = updelay;
5027 params->downdelay = downdelay;
5028 params->use_carrier = use_carrier;
5029 params->lacp_fast = lacp_fast;
5030 params->primary[0] = 0;
5031 params->fail_over_mac = fail_over_mac_value;
5034 strncpy(params->primary, primary, IFNAMSIZ);
5035 params->primary[IFNAMSIZ - 1] = 0;
5038 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5043 static struct lock_class_key bonding_netdev_xmit_lock_key;
5044 static struct lock_class_key bonding_netdev_addr_lock_key;
5046 static void bond_set_lockdep_class_one(struct net_device *dev,
5047 struct netdev_queue *txq,
5050 lockdep_set_class(&txq->_xmit_lock,
5051 &bonding_netdev_xmit_lock_key);
5054 static void bond_set_lockdep_class(struct net_device *dev)
5056 lockdep_set_class(&dev->addr_list_lock,
5057 &bonding_netdev_addr_lock_key);
5058 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5062 * Called from registration process
5064 static int bond_init(struct net_device *bond_dev)
5066 struct bonding *bond = netdev_priv(bond_dev);
5068 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5070 bond->wq = create_singlethread_workqueue(bond_dev->name);
5074 bond_set_lockdep_class(bond_dev);
5076 netif_carrier_off(bond_dev);
5078 bond_create_proc_entry(bond);
5079 list_add_tail(&bond->bond_list, &bond_dev_list);
5084 /* Create a new bond based on the specified name and bonding parameters.
5085 * If name is NULL, obtain a suitable "bond%d" name for us.
5086 * Caller must NOT hold rtnl_lock; we need to release it here before we
5087 * set up our sysfs entries.
5089 int bond_create(const char *name)
5091 struct net_device *bond_dev;
5095 /* Check to see if the bond already exists. */
5096 /* FIXME: pass netns from caller */
5097 if (name && __dev_get_by_name(&init_net, name)) {
5098 pr_err(DRV_NAME ": cannot add bond %s; already exists\n",
5104 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5107 pr_err(DRV_NAME ": %s: eek! can't alloc netdev!\n",
5114 res = dev_alloc_name(bond_dev, "bond%d");
5119 res = register_netdevice(bond_dev);
5123 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5131 unregister_netdevice(bond_dev);
5133 bond_deinit(bond_dev);
5135 free_netdev(bond_dev);
5141 static int __init bonding_init(void)
5146 pr_info("%s", version);
5148 res = bond_check_params(&bonding_defaults);
5152 bond_create_proc_dir();
5154 for (i = 0; i < max_bonds; i++) {
5155 res = bond_create(NULL);
5160 res = bond_create_sysfs();
5164 register_netdevice_notifier(&bond_netdev_notifier);
5165 register_inetaddr_notifier(&bond_inetaddr_notifier);
5166 bond_register_ipv6_notifier();
5178 static void __exit bonding_exit(void)
5180 unregister_netdevice_notifier(&bond_netdev_notifier);
5181 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5182 bond_unregister_ipv6_notifier();
5184 bond_destroy_sysfs();
5191 module_init(bonding_init);
5192 module_exit(bonding_exit);
5193 MODULE_LICENSE("GPL");
5194 MODULE_VERSION(DRV_VERSION);
5195 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5196 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");