2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
57 #include <asm/system.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
80 #include <net/netns/generic.h>
85 /*---------------------------- Module parameters ----------------------------*/
87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 #define BOND_LINK_MON_INTERV 0
89 #define BOND_LINK_ARP_INTERV 0
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int num_grat_arp = 1;
93 static int num_unsol_na = 1;
94 static int miimon = BOND_LINK_MON_INTERV;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval = BOND_LINK_ARP_INTERV;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *fail_over_mac;
108 static struct bond_params bonding_defaults;
110 module_param(max_bonds, int, 0);
111 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
112 module_param(num_grat_arp, int, 0644);
113 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
114 module_param(num_unsol_na, int, 0644);
115 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
116 module_param(miimon, int, 0);
117 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
118 module_param(updelay, int, 0);
119 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
120 module_param(downdelay, int, 0);
121 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
123 module_param(use_carrier, int, 0);
124 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
125 "0 for off, 1 for on (default)");
126 module_param(mode, charp, 0);
127 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
128 "1 for active-backup, 2 for balance-xor, "
129 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
130 "6 for balance-alb");
131 module_param(primary, charp, 0);
132 MODULE_PARM_DESC(primary, "Primary network device to use");
133 module_param(primary_reselect, charp, 0);
134 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
136 "0 for always (default), "
137 "1 for only if speed of primary is "
139 "2 for only on active slave "
141 module_param(lacp_rate, charp, 0);
142 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
144 module_param(ad_select, charp, 0);
145 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
146 module_param(xmit_hash_policy, charp, 0);
147 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
148 ", 1 for layer 3+4");
149 module_param(arp_interval, int, 0);
150 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
151 module_param_array(arp_ip_target, charp, NULL, 0);
152 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
153 module_param(arp_validate, charp, 0);
154 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
155 module_param(fail_over_mac, charp, 0);
156 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
158 /*----------------------------- Global variables ----------------------------*/
160 static const char * const version =
161 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
163 int bond_net_id __read_mostly;
165 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
166 static int arp_ip_count;
167 static int bond_mode = BOND_MODE_ROUNDROBIN;
168 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
169 static int lacp_fast;
172 const struct bond_parm_tbl bond_lacp_tbl[] = {
173 { "slow", AD_LACP_SLOW},
174 { "fast", AD_LACP_FAST},
178 const struct bond_parm_tbl bond_mode_tbl[] = {
179 { "balance-rr", BOND_MODE_ROUNDROBIN},
180 { "active-backup", BOND_MODE_ACTIVEBACKUP},
181 { "balance-xor", BOND_MODE_XOR},
182 { "broadcast", BOND_MODE_BROADCAST},
183 { "802.3ad", BOND_MODE_8023AD},
184 { "balance-tlb", BOND_MODE_TLB},
185 { "balance-alb", BOND_MODE_ALB},
189 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
190 { "layer2", BOND_XMIT_POLICY_LAYER2},
191 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
192 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
196 const struct bond_parm_tbl arp_validate_tbl[] = {
197 { "none", BOND_ARP_VALIDATE_NONE},
198 { "active", BOND_ARP_VALIDATE_ACTIVE},
199 { "backup", BOND_ARP_VALIDATE_BACKUP},
200 { "all", BOND_ARP_VALIDATE_ALL},
204 const struct bond_parm_tbl fail_over_mac_tbl[] = {
205 { "none", BOND_FOM_NONE},
206 { "active", BOND_FOM_ACTIVE},
207 { "follow", BOND_FOM_FOLLOW},
211 const struct bond_parm_tbl pri_reselect_tbl[] = {
212 { "always", BOND_PRI_RESELECT_ALWAYS},
213 { "better", BOND_PRI_RESELECT_BETTER},
214 { "failure", BOND_PRI_RESELECT_FAILURE},
218 struct bond_parm_tbl ad_select_tbl[] = {
219 { "stable", BOND_AD_STABLE},
220 { "bandwidth", BOND_AD_BANDWIDTH},
221 { "count", BOND_AD_COUNT},
225 /*-------------------------- Forward declarations ---------------------------*/
227 static void bond_send_gratuitous_arp(struct bonding *bond);
228 static int bond_init(struct net_device *bond_dev);
229 static void bond_uninit(struct net_device *bond_dev);
231 /*---------------------------- General routines -----------------------------*/
233 static const char *bond_mode_name(int mode)
235 static const char *names[] = {
236 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
237 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
238 [BOND_MODE_XOR] = "load balancing (xor)",
239 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
240 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
241 [BOND_MODE_TLB] = "transmit load balancing",
242 [BOND_MODE_ALB] = "adaptive load balancing",
245 if (mode < 0 || mode > BOND_MODE_ALB)
251 /*---------------------------------- VLAN -----------------------------------*/
254 * bond_add_vlan - add a new vlan id on bond
255 * @bond: bond that got the notification
256 * @vlan_id: the vlan id to add
258 * Returns -ENOMEM if allocation failed.
260 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
262 struct vlan_entry *vlan;
264 pr_debug("bond: %s, vlan id %d\n",
265 (bond ? bond->dev->name : "None"), vlan_id);
267 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
271 INIT_LIST_HEAD(&vlan->vlan_list);
272 vlan->vlan_id = vlan_id;
274 write_lock_bh(&bond->lock);
276 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
278 write_unlock_bh(&bond->lock);
280 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
286 * bond_del_vlan - delete a vlan id from bond
287 * @bond: bond that got the notification
288 * @vlan_id: the vlan id to delete
290 * returns -ENODEV if @vlan_id was not found in @bond.
292 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
294 struct vlan_entry *vlan;
297 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
299 write_lock_bh(&bond->lock);
301 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
302 if (vlan->vlan_id == vlan_id) {
303 list_del(&vlan->vlan_list);
305 if (bond_is_lb(bond))
306 bond_alb_clear_vlan(bond, vlan_id);
308 pr_debug("removed VLAN ID %d from bond %s\n",
309 vlan_id, bond->dev->name);
313 if (list_empty(&bond->vlan_list) &&
314 (bond->slave_cnt == 0)) {
315 /* Last VLAN removed and no slaves, so
316 * restore block on adding VLANs. This will
317 * be removed once new slaves that are not
318 * VLAN challenged will be added.
320 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
328 pr_debug("couldn't find VLAN ID %d in bond %s\n",
329 vlan_id, bond->dev->name);
332 write_unlock_bh(&bond->lock);
337 * bond_has_challenged_slaves
338 * @bond: the bond we're working on
340 * Searches the slave list. Returns 1 if a vlan challenged slave
341 * was found, 0 otherwise.
343 * Assumes bond->lock is held.
345 static int bond_has_challenged_slaves(struct bonding *bond)
350 bond_for_each_slave(bond, slave, i) {
351 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
352 pr_debug("found VLAN challenged slave - %s\n",
358 pr_debug("no VLAN challenged slaves found\n");
363 * bond_next_vlan - safely skip to the next item in the vlans list.
364 * @bond: the bond we're working on
365 * @curr: item we're advancing from
367 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
368 * or @curr->next otherwise (even if it is @curr itself again).
370 * Caller must hold bond->lock
372 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
374 struct vlan_entry *next, *last;
376 if (list_empty(&bond->vlan_list))
380 next = list_entry(bond->vlan_list.next,
381 struct vlan_entry, vlan_list);
383 last = list_entry(bond->vlan_list.prev,
384 struct vlan_entry, vlan_list);
386 next = list_entry(bond->vlan_list.next,
387 struct vlan_entry, vlan_list);
389 next = list_entry(curr->vlan_list.next,
390 struct vlan_entry, vlan_list);
398 * bond_dev_queue_xmit - Prepare skb for xmit.
400 * @bond: bond device that got this skb for tx.
401 * @skb: hw accel VLAN tagged skb to transmit
402 * @slave_dev: slave that is supposed to xmit this skbuff
404 * When the bond gets an skb to transmit that is
405 * already hardware accelerated VLAN tagged, and it
406 * needs to relay this skb to a slave that is not
407 * hw accel capable, the skb needs to be "unaccelerated",
408 * i.e. strip the hwaccel tag and re-insert it as part
411 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
412 struct net_device *slave_dev)
414 unsigned short uninitialized_var(vlan_id);
416 if (!list_empty(&bond->vlan_list) &&
417 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
418 vlan_get_tag(skb, &vlan_id) == 0) {
419 skb->dev = slave_dev;
420 skb = vlan_put_tag(skb, vlan_id);
422 /* vlan_put_tag() frees the skb in case of error,
423 * so return success here so the calling functions
424 * won't attempt to free is again.
429 skb->dev = slave_dev;
439 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
440 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
442 * a. This operation is performed in IOCTL context,
443 * b. The operation is protected by the RTNL semaphore in the 8021q code,
444 * c. Holding a lock with BH disabled while directly calling a base driver
445 * entry point is generally a BAD idea.
447 * The design of synchronization/protection for this operation in the 8021q
448 * module is good for one or more VLAN devices over a single physical device
449 * and cannot be extended for a teaming solution like bonding, so there is a
450 * potential race condition here where a net device from the vlan group might
451 * be referenced (either by a base driver or the 8021q code) while it is being
452 * removed from the system. However, it turns out we're not making matters
453 * worse, and if it works for regular VLAN usage it will work here too.
457 * bond_vlan_rx_register - Propagates registration to slaves
458 * @bond_dev: bonding net device that got called
459 * @grp: vlan group being registered
461 static void bond_vlan_rx_register(struct net_device *bond_dev,
462 struct vlan_group *grp)
464 struct bonding *bond = netdev_priv(bond_dev);
470 bond_for_each_slave(bond, slave, i) {
471 struct net_device *slave_dev = slave->dev;
472 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
474 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
475 slave_ops->ndo_vlan_rx_register) {
476 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
482 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
483 * @bond_dev: bonding net device that got called
484 * @vid: vlan id being added
486 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
488 struct bonding *bond = netdev_priv(bond_dev);
492 bond_for_each_slave(bond, slave, i) {
493 struct net_device *slave_dev = slave->dev;
494 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
496 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
497 slave_ops->ndo_vlan_rx_add_vid) {
498 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
502 res = bond_add_vlan(bond, vid);
504 pr_err("%s: Error: Failed to add vlan id %d\n",
505 bond_dev->name, vid);
510 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
511 * @bond_dev: bonding net device that got called
512 * @vid: vlan id being removed
514 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
516 struct bonding *bond = netdev_priv(bond_dev);
518 struct net_device *vlan_dev;
521 bond_for_each_slave(bond, slave, i) {
522 struct net_device *slave_dev = slave->dev;
523 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
525 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
526 slave_ops->ndo_vlan_rx_kill_vid) {
527 /* Save and then restore vlan_dev in the grp array,
528 * since the slave's driver might clear it.
530 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
531 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
532 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
536 res = bond_del_vlan(bond, vid);
538 pr_err("%s: Error: Failed to remove vlan id %d\n",
539 bond_dev->name, vid);
543 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
545 struct vlan_entry *vlan;
546 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
548 write_lock_bh(&bond->lock);
550 if (list_empty(&bond->vlan_list))
553 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
554 slave_ops->ndo_vlan_rx_register)
555 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
557 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
558 !(slave_ops->ndo_vlan_rx_add_vid))
561 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
562 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
565 write_unlock_bh(&bond->lock);
568 static void bond_del_vlans_from_slave(struct bonding *bond,
569 struct net_device *slave_dev)
571 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
572 struct vlan_entry *vlan;
573 struct net_device *vlan_dev;
575 write_lock_bh(&bond->lock);
577 if (list_empty(&bond->vlan_list))
580 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
581 !(slave_ops->ndo_vlan_rx_kill_vid))
584 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
585 /* Save and then restore vlan_dev in the grp array,
586 * since the slave's driver might clear it.
588 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
589 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
590 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
594 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
595 slave_ops->ndo_vlan_rx_register)
596 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
599 write_unlock_bh(&bond->lock);
602 /*------------------------------- Link status -------------------------------*/
605 * Set the carrier state for the master according to the state of its
606 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
607 * do special 802.3ad magic.
609 * Returns zero if carrier state does not change, nonzero if it does.
611 static int bond_set_carrier(struct bonding *bond)
616 if (bond->slave_cnt == 0)
619 if (bond->params.mode == BOND_MODE_8023AD)
620 return bond_3ad_set_carrier(bond);
622 bond_for_each_slave(bond, slave, i) {
623 if (slave->link == BOND_LINK_UP) {
624 if (!netif_carrier_ok(bond->dev)) {
625 netif_carrier_on(bond->dev);
633 if (netif_carrier_ok(bond->dev)) {
634 netif_carrier_off(bond->dev);
641 * Get link speed and duplex from the slave's base driver
642 * using ethtool. If for some reason the call fails or the
643 * values are invalid, fake speed and duplex to 100/Full
646 static int bond_update_speed_duplex(struct slave *slave)
648 struct net_device *slave_dev = slave->dev;
649 struct ethtool_cmd etool;
652 /* Fake speed and duplex */
653 slave->speed = SPEED_100;
654 slave->duplex = DUPLEX_FULL;
656 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
659 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
663 switch (etool.speed) {
673 switch (etool.duplex) {
681 slave->speed = etool.speed;
682 slave->duplex = etool.duplex;
688 * if <dev> supports MII link status reporting, check its link status.
690 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
691 * depending upon the setting of the use_carrier parameter.
693 * Return either BMSR_LSTATUS, meaning that the link is up (or we
694 * can't tell and just pretend it is), or 0, meaning that the link is
697 * If reporting is non-zero, instead of faking link up, return -1 if
698 * both ETHTOOL and MII ioctls fail (meaning the device does not
699 * support them). If use_carrier is set, return whatever it says.
700 * It'd be nice if there was a good way to tell if a driver supports
701 * netif_carrier, but there really isn't.
703 static int bond_check_dev_link(struct bonding *bond,
704 struct net_device *slave_dev, int reporting)
706 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
707 int (*ioctl)(struct net_device *, struct ifreq *, int);
709 struct mii_ioctl_data *mii;
711 if (!reporting && !netif_running(slave_dev))
714 if (bond->params.use_carrier)
715 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
717 /* Try to get link status using Ethtool first. */
718 if (slave_dev->ethtool_ops) {
719 if (slave_dev->ethtool_ops->get_link) {
722 link = slave_dev->ethtool_ops->get_link(slave_dev);
724 return link ? BMSR_LSTATUS : 0;
728 /* Ethtool can't be used, fallback to MII ioctls. */
729 ioctl = slave_ops->ndo_do_ioctl;
731 /* TODO: set pointer to correct ioctl on a per team member */
732 /* bases to make this more efficient. that is, once */
733 /* we determine the correct ioctl, we will always */
734 /* call it and not the others for that team */
738 * We cannot assume that SIOCGMIIPHY will also read a
739 * register; not all network drivers (e.g., e100)
743 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
744 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
746 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
747 mii->reg_num = MII_BMSR;
748 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
749 return mii->val_out & BMSR_LSTATUS;
754 * If reporting, report that either there's no dev->do_ioctl,
755 * or both SIOCGMIIREG and get_link failed (meaning that we
756 * cannot report link status). If not reporting, pretend
759 return reporting ? -1 : BMSR_LSTATUS;
762 /*----------------------------- Multicast list ------------------------------*/
765 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
767 static inline int bond_is_dmi_same(const struct dev_mc_list *dmi1,
768 const struct dev_mc_list *dmi2)
770 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
771 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
775 * returns dmi entry if found, NULL otherwise
777 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi,
778 struct dev_mc_list *mc_list)
780 struct dev_mc_list *idmi;
782 for (idmi = mc_list; idmi; idmi = idmi->next) {
783 if (bond_is_dmi_same(dmi, idmi))
791 * Push the promiscuity flag down to appropriate slaves
793 static int bond_set_promiscuity(struct bonding *bond, int inc)
796 if (USES_PRIMARY(bond->params.mode)) {
797 /* write lock already acquired */
798 if (bond->curr_active_slave) {
799 err = dev_set_promiscuity(bond->curr_active_slave->dev,
805 bond_for_each_slave(bond, slave, i) {
806 err = dev_set_promiscuity(slave->dev, inc);
815 * Push the allmulti flag down to all slaves
817 static int bond_set_allmulti(struct bonding *bond, int inc)
820 if (USES_PRIMARY(bond->params.mode)) {
821 /* write lock already acquired */
822 if (bond->curr_active_slave) {
823 err = dev_set_allmulti(bond->curr_active_slave->dev,
829 bond_for_each_slave(bond, slave, i) {
830 err = dev_set_allmulti(slave->dev, inc);
839 * Add a Multicast address to slaves
842 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
844 if (USES_PRIMARY(bond->params.mode)) {
845 /* write lock already acquired */
846 if (bond->curr_active_slave)
847 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
852 bond_for_each_slave(bond, slave, i)
853 dev_mc_add(slave->dev, addr, alen, 0);
858 * Remove a multicast address from slave
861 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
863 if (USES_PRIMARY(bond->params.mode)) {
864 /* write lock already acquired */
865 if (bond->curr_active_slave)
866 dev_mc_delete(bond->curr_active_slave->dev, addr,
871 bond_for_each_slave(bond, slave, i) {
872 dev_mc_delete(slave->dev, addr, alen, 0);
879 * Retrieve the list of registered multicast addresses for the bonding
880 * device and retransmit an IGMP JOIN request to the current active
883 static void bond_resend_igmp_join_requests(struct bonding *bond)
885 struct in_device *in_dev;
886 struct ip_mc_list *im;
889 in_dev = __in_dev_get_rcu(bond->dev);
891 for (im = in_dev->mc_list; im; im = im->next)
892 ip_mc_rejoin_group(im);
899 * Totally destroys the mc_list in bond
901 static void bond_mc_list_destroy(struct bonding *bond)
903 struct dev_mc_list *dmi;
907 bond->mc_list = dmi->next;
912 bond->mc_list = NULL;
916 * Copy all the Multicast addresses from src to the bonding device dst
918 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
921 struct dev_mc_list *dmi, *new_dmi;
923 for (dmi = mc_list; dmi; dmi = dmi->next) {
924 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
927 /* FIXME: Potential memory leak !!! */
931 new_dmi->next = bond->mc_list;
932 bond->mc_list = new_dmi;
933 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
934 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
935 new_dmi->dmi_users = dmi->dmi_users;
936 new_dmi->dmi_gusers = dmi->dmi_gusers;
943 * flush all members of flush->mc_list from device dev->mc_list
945 static void bond_mc_list_flush(struct net_device *bond_dev,
946 struct net_device *slave_dev)
948 struct bonding *bond = netdev_priv(bond_dev);
949 struct dev_mc_list *dmi;
951 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
952 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
954 if (bond->params.mode == BOND_MODE_8023AD) {
955 /* del lacpdu mc addr from mc list */
956 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
958 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
962 /*--------------------------- Active slave change ---------------------------*/
965 * Update the mc list and multicast-related flags for the new and
966 * old active slaves (if any) according to the multicast mode, and
967 * promiscuous flags unconditionally.
969 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
970 struct slave *old_active)
972 struct dev_mc_list *dmi;
974 if (!USES_PRIMARY(bond->params.mode))
975 /* nothing to do - mc list is already up-to-date on
981 if (bond->dev->flags & IFF_PROMISC)
982 dev_set_promiscuity(old_active->dev, -1);
984 if (bond->dev->flags & IFF_ALLMULTI)
985 dev_set_allmulti(old_active->dev, -1);
987 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
988 dev_mc_delete(old_active->dev, dmi->dmi_addr,
989 dmi->dmi_addrlen, 0);
993 /* FIXME: Signal errors upstream. */
994 if (bond->dev->flags & IFF_PROMISC)
995 dev_set_promiscuity(new_active->dev, 1);
997 if (bond->dev->flags & IFF_ALLMULTI)
998 dev_set_allmulti(new_active->dev, 1);
1000 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
1001 dev_mc_add(new_active->dev, dmi->dmi_addr,
1002 dmi->dmi_addrlen, 0);
1003 bond_resend_igmp_join_requests(bond);
1008 * bond_do_fail_over_mac
1010 * Perform special MAC address swapping for fail_over_mac settings
1012 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1014 static void bond_do_fail_over_mac(struct bonding *bond,
1015 struct slave *new_active,
1016 struct slave *old_active)
1017 __releases(&bond->curr_slave_lock)
1018 __releases(&bond->lock)
1019 __acquires(&bond->lock)
1020 __acquires(&bond->curr_slave_lock)
1022 u8 tmp_mac[ETH_ALEN];
1023 struct sockaddr saddr;
1026 switch (bond->params.fail_over_mac) {
1027 case BOND_FOM_ACTIVE:
1029 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1030 new_active->dev->addr_len);
1032 case BOND_FOM_FOLLOW:
1034 * if new_active && old_active, swap them
1035 * if just old_active, do nothing (going to no active slave)
1036 * if just new_active, set new_active to bond's MAC
1041 write_unlock_bh(&bond->curr_slave_lock);
1042 read_unlock(&bond->lock);
1045 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1046 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1048 saddr.sa_family = new_active->dev->type;
1050 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1051 saddr.sa_family = bond->dev->type;
1054 rv = dev_set_mac_address(new_active->dev, &saddr);
1056 pr_err("%s: Error %d setting MAC of slave %s\n",
1057 bond->dev->name, -rv, new_active->dev->name);
1064 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1065 saddr.sa_family = old_active->dev->type;
1067 rv = dev_set_mac_address(old_active->dev, &saddr);
1069 pr_err("%s: Error %d setting MAC of slave %s\n",
1070 bond->dev->name, -rv, new_active->dev->name);
1072 read_lock(&bond->lock);
1073 write_lock_bh(&bond->curr_slave_lock);
1076 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1077 bond->dev->name, bond->params.fail_over_mac);
1083 static bool bond_should_change_active(struct bonding *bond)
1085 struct slave *prim = bond->primary_slave;
1086 struct slave *curr = bond->curr_active_slave;
1088 if (!prim || !curr || curr->link != BOND_LINK_UP)
1090 if (bond->force_primary) {
1091 bond->force_primary = false;
1094 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1095 (prim->speed < curr->speed ||
1096 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1098 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1104 * find_best_interface - select the best available slave to be the active one
1105 * @bond: our bonding struct
1107 * Warning: Caller must hold curr_slave_lock for writing.
1109 static struct slave *bond_find_best_slave(struct bonding *bond)
1111 struct slave *new_active, *old_active;
1112 struct slave *bestslave = NULL;
1113 int mintime = bond->params.updelay;
1116 new_active = bond->curr_active_slave;
1118 if (!new_active) { /* there were no active slaves left */
1119 if (bond->slave_cnt > 0) /* found one slave */
1120 new_active = bond->first_slave;
1122 return NULL; /* still no slave, return NULL */
1125 if ((bond->primary_slave) &&
1126 bond->primary_slave->link == BOND_LINK_UP &&
1127 bond_should_change_active(bond)) {
1128 new_active = bond->primary_slave;
1131 /* remember where to stop iterating over the slaves */
1132 old_active = new_active;
1134 bond_for_each_slave_from(bond, new_active, i, old_active) {
1135 if (new_active->link == BOND_LINK_UP) {
1137 } else if (new_active->link == BOND_LINK_BACK &&
1138 IS_UP(new_active->dev)) {
1139 /* link up, but waiting for stabilization */
1140 if (new_active->delay < mintime) {
1141 mintime = new_active->delay;
1142 bestslave = new_active;
1151 * change_active_interface - change the active slave into the specified one
1152 * @bond: our bonding struct
1153 * @new: the new slave to make the active one
1155 * Set the new slave to the bond's settings and unset them on the old
1156 * curr_active_slave.
1157 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1159 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1160 * because it is apparently the best available slave we have, even though its
1161 * updelay hasn't timed out yet.
1163 * If new_active is not NULL, caller must hold bond->lock for read and
1164 * curr_slave_lock for write_bh.
1166 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1168 struct slave *old_active = bond->curr_active_slave;
1170 if (old_active == new_active)
1174 new_active->jiffies = jiffies;
1176 if (new_active->link == BOND_LINK_BACK) {
1177 if (USES_PRIMARY(bond->params.mode)) {
1178 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1179 bond->dev->name, new_active->dev->name,
1180 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1183 new_active->delay = 0;
1184 new_active->link = BOND_LINK_UP;
1186 if (bond->params.mode == BOND_MODE_8023AD)
1187 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1189 if (bond_is_lb(bond))
1190 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1192 if (USES_PRIMARY(bond->params.mode)) {
1193 pr_info("%s: making interface %s the new active one.\n",
1194 bond->dev->name, new_active->dev->name);
1199 if (USES_PRIMARY(bond->params.mode))
1200 bond_mc_swap(bond, new_active, old_active);
1202 if (bond_is_lb(bond)) {
1203 bond_alb_handle_active_change(bond, new_active);
1205 bond_set_slave_inactive_flags(old_active);
1207 bond_set_slave_active_flags(new_active);
1209 bond->curr_active_slave = new_active;
1212 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1214 bond_set_slave_inactive_flags(old_active);
1217 bond_set_slave_active_flags(new_active);
1219 if (bond->params.fail_over_mac)
1220 bond_do_fail_over_mac(bond, new_active,
1223 bond->send_grat_arp = bond->params.num_grat_arp;
1224 bond_send_gratuitous_arp(bond);
1226 bond->send_unsol_na = bond->params.num_unsol_na;
1227 bond_send_unsolicited_na(bond);
1229 write_unlock_bh(&bond->curr_slave_lock);
1230 read_unlock(&bond->lock);
1232 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1234 read_lock(&bond->lock);
1235 write_lock_bh(&bond->curr_slave_lock);
1241 * bond_select_active_slave - select a new active slave, if needed
1242 * @bond: our bonding struct
1244 * This functions should be called when one of the following occurs:
1245 * - The old curr_active_slave has been released or lost its link.
1246 * - The primary_slave has got its link back.
1247 * - A slave has got its link back and there's no old curr_active_slave.
1249 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1251 void bond_select_active_slave(struct bonding *bond)
1253 struct slave *best_slave;
1256 best_slave = bond_find_best_slave(bond);
1257 if (best_slave != bond->curr_active_slave) {
1258 bond_change_active_slave(bond, best_slave);
1259 rv = bond_set_carrier(bond);
1263 if (netif_carrier_ok(bond->dev)) {
1264 pr_info("%s: first active interface up!\n",
1267 pr_info("%s: now running without any active interface !\n",
1273 /*--------------------------- slave list handling ---------------------------*/
1276 * This function attaches the slave to the end of list.
1278 * bond->lock held for writing by caller.
1280 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1282 if (bond->first_slave == NULL) { /* attaching the first slave */
1283 new_slave->next = new_slave;
1284 new_slave->prev = new_slave;
1285 bond->first_slave = new_slave;
1287 new_slave->next = bond->first_slave;
1288 new_slave->prev = bond->first_slave->prev;
1289 new_slave->next->prev = new_slave;
1290 new_slave->prev->next = new_slave;
1297 * This function detaches the slave from the list.
1298 * WARNING: no check is made to verify if the slave effectively
1299 * belongs to <bond>.
1300 * Nothing is freed on return, structures are just unchained.
1301 * If any slave pointer in bond was pointing to <slave>,
1302 * it should be changed by the calling function.
1304 * bond->lock held for writing by caller.
1306 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1309 slave->next->prev = slave->prev;
1312 slave->prev->next = slave->next;
1314 if (bond->first_slave == slave) { /* slave is the first slave */
1315 if (bond->slave_cnt > 1) { /* there are more slave */
1316 bond->first_slave = slave->next;
1318 bond->first_slave = NULL; /* slave was the last one */
1327 /*---------------------------------- IOCTL ----------------------------------*/
1329 static int bond_sethwaddr(struct net_device *bond_dev,
1330 struct net_device *slave_dev)
1332 pr_debug("bond_dev=%p\n", bond_dev);
1333 pr_debug("slave_dev=%p\n", slave_dev);
1334 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1335 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1339 #define BOND_VLAN_FEATURES \
1340 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1341 NETIF_F_HW_VLAN_FILTER)
1344 * Compute the common dev->feature set available to all slaves. Some
1345 * feature bits are managed elsewhere, so preserve those feature bits
1346 * on the master device.
1348 static int bond_compute_features(struct bonding *bond)
1350 struct slave *slave;
1351 struct net_device *bond_dev = bond->dev;
1352 unsigned long features = bond_dev->features;
1353 unsigned long vlan_features = 0;
1354 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1355 bond_dev->hard_header_len);
1358 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1359 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1361 if (!bond->first_slave)
1364 features &= ~NETIF_F_ONE_FOR_ALL;
1366 vlan_features = bond->first_slave->dev->vlan_features;
1367 bond_for_each_slave(bond, slave, i) {
1368 features = netdev_increment_features(features,
1369 slave->dev->features,
1370 NETIF_F_ONE_FOR_ALL);
1371 vlan_features = netdev_increment_features(vlan_features,
1372 slave->dev->vlan_features,
1373 NETIF_F_ONE_FOR_ALL);
1374 if (slave->dev->hard_header_len > max_hard_header_len)
1375 max_hard_header_len = slave->dev->hard_header_len;
1379 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1380 bond_dev->features = netdev_fix_features(features, NULL);
1381 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1382 bond_dev->hard_header_len = max_hard_header_len;
1387 static void bond_setup_by_slave(struct net_device *bond_dev,
1388 struct net_device *slave_dev)
1390 struct bonding *bond = netdev_priv(bond_dev);
1392 bond_dev->header_ops = slave_dev->header_ops;
1394 bond_dev->type = slave_dev->type;
1395 bond_dev->hard_header_len = slave_dev->hard_header_len;
1396 bond_dev->addr_len = slave_dev->addr_len;
1398 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1399 slave_dev->addr_len);
1400 bond->setup_by_slave = 1;
1403 /* enslave device <slave> to bond device <master> */
1404 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1406 struct bonding *bond = netdev_priv(bond_dev);
1407 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1408 struct slave *new_slave = NULL;
1409 struct dev_mc_list *dmi;
1410 struct sockaddr addr;
1412 int old_features = bond_dev->features;
1415 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1416 slave_ops->ndo_do_ioctl == NULL) {
1417 pr_warning("%s: Warning: no link monitoring support for %s\n",
1418 bond_dev->name, slave_dev->name);
1421 /* bond must be initialized by bond_open() before enslaving */
1422 if (!(bond_dev->flags & IFF_UP)) {
1423 pr_warning("%s: master_dev is not up in bond_enslave\n",
1427 /* already enslaved */
1428 if (slave_dev->flags & IFF_SLAVE) {
1429 pr_debug("Error, Device was already enslaved\n");
1433 /* vlan challenged mutual exclusion */
1434 /* no need to lock since we're protected by rtnl_lock */
1435 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1436 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1437 if (!list_empty(&bond->vlan_list)) {
1438 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1439 bond_dev->name, slave_dev->name, bond_dev->name);
1442 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1443 bond_dev->name, slave_dev->name,
1444 slave_dev->name, bond_dev->name);
1445 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1448 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1449 if (bond->slave_cnt == 0) {
1450 /* First slave, and it is not VLAN challenged,
1451 * so remove the block of adding VLANs over the bond.
1453 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1458 * Old ifenslave binaries are no longer supported. These can
1459 * be identified with moderate accuracy by the state of the slave:
1460 * the current ifenslave will set the interface down prior to
1461 * enslaving it; the old ifenslave will not.
1463 if ((slave_dev->flags & IFF_UP)) {
1464 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1467 goto err_undo_flags;
1470 /* set bonding device ether type by slave - bonding netdevices are
1471 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1472 * there is a need to override some of the type dependent attribs/funcs.
1474 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1475 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1477 if (bond->slave_cnt == 0) {
1478 if (bond_dev->type != slave_dev->type) {
1479 pr_debug("%s: change device type from %d to %d\n",
1481 bond_dev->type, slave_dev->type);
1483 netdev_bonding_change(bond_dev, NETDEV_BONDING_OLDTYPE);
1485 if (slave_dev->type != ARPHRD_ETHER)
1486 bond_setup_by_slave(bond_dev, slave_dev);
1488 ether_setup(bond_dev);
1490 netdev_bonding_change(bond_dev, NETDEV_BONDING_NEWTYPE);
1492 } else if (bond_dev->type != slave_dev->type) {
1493 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1495 slave_dev->type, bond_dev->type);
1497 goto err_undo_flags;
1500 if (slave_ops->ndo_set_mac_address == NULL) {
1501 if (bond->slave_cnt == 0) {
1502 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1504 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1505 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1506 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1509 goto err_undo_flags;
1513 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1516 goto err_undo_flags;
1519 /* save slave's original flags before calling
1520 * netdev_set_master and dev_open
1522 new_slave->original_flags = slave_dev->flags;
1525 * Save slave's original ("permanent") mac address for modes
1526 * that need it, and for restoring it upon release, and then
1527 * set it to the master's address
1529 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1531 if (!bond->params.fail_over_mac) {
1533 * Set slave to master's mac address. The application already
1534 * set the master's mac address to that of the first slave
1536 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1537 addr.sa_family = slave_dev->type;
1538 res = dev_set_mac_address(slave_dev, &addr);
1540 pr_debug("Error %d calling set_mac_address\n", res);
1545 res = netdev_set_master(slave_dev, bond_dev);
1547 pr_debug("Error %d calling netdev_set_master\n", res);
1548 goto err_restore_mac;
1550 /* open the slave since the application closed it */
1551 res = dev_open(slave_dev);
1553 pr_debug("Opening slave %s failed\n", slave_dev->name);
1554 goto err_unset_master;
1557 new_slave->dev = slave_dev;
1558 slave_dev->priv_flags |= IFF_BONDING;
1560 if (bond_is_lb(bond)) {
1561 /* bond_alb_init_slave() must be called before all other stages since
1562 * it might fail and we do not want to have to undo everything
1564 res = bond_alb_init_slave(bond, new_slave);
1569 /* If the mode USES_PRIMARY, then the new slave gets the
1570 * master's promisc (and mc) settings only if it becomes the
1571 * curr_active_slave, and that is taken care of later when calling
1572 * bond_change_active()
1574 if (!USES_PRIMARY(bond->params.mode)) {
1575 /* set promiscuity level to new slave */
1576 if (bond_dev->flags & IFF_PROMISC) {
1577 res = dev_set_promiscuity(slave_dev, 1);
1582 /* set allmulti level to new slave */
1583 if (bond_dev->flags & IFF_ALLMULTI) {
1584 res = dev_set_allmulti(slave_dev, 1);
1589 netif_addr_lock_bh(bond_dev);
1590 /* upload master's mc_list to new slave */
1591 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1592 dev_mc_add(slave_dev, dmi->dmi_addr,
1593 dmi->dmi_addrlen, 0);
1594 netif_addr_unlock_bh(bond_dev);
1597 if (bond->params.mode == BOND_MODE_8023AD) {
1598 /* add lacpdu mc addr to mc list */
1599 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1601 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1604 bond_add_vlans_on_slave(bond, slave_dev);
1606 write_lock_bh(&bond->lock);
1608 bond_attach_slave(bond, new_slave);
1610 new_slave->delay = 0;
1611 new_slave->link_failure_count = 0;
1613 bond_compute_features(bond);
1615 write_unlock_bh(&bond->lock);
1617 read_lock(&bond->lock);
1619 new_slave->last_arp_rx = jiffies;
1621 if (bond->params.miimon && !bond->params.use_carrier) {
1622 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1624 if ((link_reporting == -1) && !bond->params.arp_interval) {
1626 * miimon is set but a bonded network driver
1627 * does not support ETHTOOL/MII and
1628 * arp_interval is not set. Note: if
1629 * use_carrier is enabled, we will never go
1630 * here (because netif_carrier is always
1631 * supported); thus, we don't need to change
1632 * the messages for netif_carrier.
1634 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1635 bond_dev->name, slave_dev->name);
1636 } else if (link_reporting == -1) {
1637 /* unable get link status using mii/ethtool */
1638 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1639 bond_dev->name, slave_dev->name);
1643 /* check for initial state */
1644 if (!bond->params.miimon ||
1645 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1646 if (bond->params.updelay) {
1647 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1648 new_slave->link = BOND_LINK_BACK;
1649 new_slave->delay = bond->params.updelay;
1651 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1652 new_slave->link = BOND_LINK_UP;
1654 new_slave->jiffies = jiffies;
1656 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1657 new_slave->link = BOND_LINK_DOWN;
1660 if (bond_update_speed_duplex(new_slave) &&
1661 (new_slave->link != BOND_LINK_DOWN)) {
1662 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1663 bond_dev->name, new_slave->dev->name);
1665 if (bond->params.mode == BOND_MODE_8023AD) {
1666 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1671 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1672 /* if there is a primary slave, remember it */
1673 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1674 bond->primary_slave = new_slave;
1675 bond->force_primary = true;
1679 write_lock_bh(&bond->curr_slave_lock);
1681 switch (bond->params.mode) {
1682 case BOND_MODE_ACTIVEBACKUP:
1683 bond_set_slave_inactive_flags(new_slave);
1684 bond_select_active_slave(bond);
1686 case BOND_MODE_8023AD:
1687 /* in 802.3ad mode, the internal mechanism
1688 * will activate the slaves in the selected
1691 bond_set_slave_inactive_flags(new_slave);
1692 /* if this is the first slave */
1693 if (bond->slave_cnt == 1) {
1694 SLAVE_AD_INFO(new_slave).id = 1;
1695 /* Initialize AD with the number of times that the AD timer is called in 1 second
1696 * can be called only after the mac address of the bond is set
1698 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1699 bond->params.lacp_fast);
1701 SLAVE_AD_INFO(new_slave).id =
1702 SLAVE_AD_INFO(new_slave->prev).id + 1;
1705 bond_3ad_bind_slave(new_slave);
1709 new_slave->state = BOND_STATE_ACTIVE;
1710 bond_set_slave_inactive_flags(new_slave);
1711 bond_select_active_slave(bond);
1714 pr_debug("This slave is always active in trunk mode\n");
1716 /* always active in trunk mode */
1717 new_slave->state = BOND_STATE_ACTIVE;
1719 /* In trunking mode there is little meaning to curr_active_slave
1720 * anyway (it holds no special properties of the bond device),
1721 * so we can change it without calling change_active_interface()
1723 if (!bond->curr_active_slave)
1724 bond->curr_active_slave = new_slave;
1727 } /* switch(bond_mode) */
1729 write_unlock_bh(&bond->curr_slave_lock);
1731 bond_set_carrier(bond);
1733 read_unlock(&bond->lock);
1735 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1739 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1740 bond_dev->name, slave_dev->name,
1741 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1742 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1744 /* enslave is successful */
1747 /* Undo stages on error */
1749 dev_close(slave_dev);
1752 netdev_set_master(slave_dev, NULL);
1755 if (!bond->params.fail_over_mac) {
1756 /* XXX TODO - fom follow mode needs to change master's
1757 * MAC if this slave's MAC is in use by the bond, or at
1758 * least print a warning.
1760 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1761 addr.sa_family = slave_dev->type;
1762 dev_set_mac_address(slave_dev, &addr);
1769 bond_dev->features = old_features;
1775 * Try to release the slave device <slave> from the bond device <master>
1776 * It is legal to access curr_active_slave without a lock because all the function
1779 * The rules for slave state should be:
1780 * for Active/Backup:
1781 * Active stays on all backups go down
1782 * for Bonded connections:
1783 * The first up interface should be left on and all others downed.
1785 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1787 struct bonding *bond = netdev_priv(bond_dev);
1788 struct slave *slave, *oldcurrent;
1789 struct sockaddr addr;
1791 /* slave is not a slave or master is not master of this slave */
1792 if (!(slave_dev->flags & IFF_SLAVE) ||
1793 (slave_dev->master != bond_dev)) {
1794 pr_err("%s: Error: cannot release %s.\n",
1795 bond_dev->name, slave_dev->name);
1799 write_lock_bh(&bond->lock);
1801 slave = bond_get_slave_by_dev(bond, slave_dev);
1803 /* not a slave of this bond */
1804 pr_info("%s: %s not enslaved\n",
1805 bond_dev->name, slave_dev->name);
1806 write_unlock_bh(&bond->lock);
1810 if (!bond->params.fail_over_mac) {
1811 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1812 bond->slave_cnt > 1)
1813 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1814 bond_dev->name, slave_dev->name,
1816 bond_dev->name, slave_dev->name);
1819 /* Inform AD package of unbinding of slave. */
1820 if (bond->params.mode == BOND_MODE_8023AD) {
1821 /* must be called before the slave is
1822 * detached from the list
1824 bond_3ad_unbind_slave(slave);
1827 pr_info("%s: releasing %s interface %s\n",
1829 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1832 oldcurrent = bond->curr_active_slave;
1834 bond->current_arp_slave = NULL;
1836 /* release the slave from its bond */
1837 bond_detach_slave(bond, slave);
1839 bond_compute_features(bond);
1841 if (bond->primary_slave == slave)
1842 bond->primary_slave = NULL;
1844 if (oldcurrent == slave)
1845 bond_change_active_slave(bond, NULL);
1847 if (bond_is_lb(bond)) {
1848 /* Must be called only after the slave has been
1849 * detached from the list and the curr_active_slave
1850 * has been cleared (if our_slave == old_current),
1851 * but before a new active slave is selected.
1853 write_unlock_bh(&bond->lock);
1854 bond_alb_deinit_slave(bond, slave);
1855 write_lock_bh(&bond->lock);
1858 if (oldcurrent == slave) {
1860 * Note that we hold RTNL over this sequence, so there
1861 * is no concern that another slave add/remove event
1864 write_unlock_bh(&bond->lock);
1865 read_lock(&bond->lock);
1866 write_lock_bh(&bond->curr_slave_lock);
1868 bond_select_active_slave(bond);
1870 write_unlock_bh(&bond->curr_slave_lock);
1871 read_unlock(&bond->lock);
1872 write_lock_bh(&bond->lock);
1875 if (bond->slave_cnt == 0) {
1876 bond_set_carrier(bond);
1878 /* if the last slave was removed, zero the mac address
1879 * of the master so it will be set by the application
1880 * to the mac address of the first slave
1882 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1884 if (list_empty(&bond->vlan_list)) {
1885 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1887 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1888 bond_dev->name, bond_dev->name);
1889 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1892 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1893 !bond_has_challenged_slaves(bond)) {
1894 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1895 bond_dev->name, slave_dev->name, bond_dev->name);
1896 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1899 write_unlock_bh(&bond->lock);
1901 /* must do this from outside any spinlocks */
1902 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1904 bond_del_vlans_from_slave(bond, slave_dev);
1906 /* If the mode USES_PRIMARY, then we should only remove its
1907 * promisc and mc settings if it was the curr_active_slave, but that was
1908 * already taken care of above when we detached the slave
1910 if (!USES_PRIMARY(bond->params.mode)) {
1911 /* unset promiscuity level from slave */
1912 if (bond_dev->flags & IFF_PROMISC)
1913 dev_set_promiscuity(slave_dev, -1);
1915 /* unset allmulti level from slave */
1916 if (bond_dev->flags & IFF_ALLMULTI)
1917 dev_set_allmulti(slave_dev, -1);
1919 /* flush master's mc_list from slave */
1920 netif_addr_lock_bh(bond_dev);
1921 bond_mc_list_flush(bond_dev, slave_dev);
1922 netif_addr_unlock_bh(bond_dev);
1925 netdev_set_master(slave_dev, NULL);
1927 /* close slave before restoring its mac address */
1928 dev_close(slave_dev);
1930 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1931 /* restore original ("permanent") mac address */
1932 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1933 addr.sa_family = slave_dev->type;
1934 dev_set_mac_address(slave_dev, &addr);
1937 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1938 IFF_SLAVE_INACTIVE | IFF_BONDING |
1943 return 0; /* deletion OK */
1947 * First release a slave and than destroy the bond if no more slaves are left.
1948 * Must be under rtnl_lock when this function is called.
1950 int bond_release_and_destroy(struct net_device *bond_dev,
1951 struct net_device *slave_dev)
1953 struct bonding *bond = netdev_priv(bond_dev);
1956 ret = bond_release(bond_dev, slave_dev);
1957 if ((ret == 0) && (bond->slave_cnt == 0)) {
1958 pr_info("%s: destroying bond %s.\n",
1959 bond_dev->name, bond_dev->name);
1960 unregister_netdevice(bond_dev);
1966 * This function releases all slaves.
1968 static int bond_release_all(struct net_device *bond_dev)
1970 struct bonding *bond = netdev_priv(bond_dev);
1971 struct slave *slave;
1972 struct net_device *slave_dev;
1973 struct sockaddr addr;
1975 write_lock_bh(&bond->lock);
1977 netif_carrier_off(bond_dev);
1979 if (bond->slave_cnt == 0)
1982 bond->current_arp_slave = NULL;
1983 bond->primary_slave = NULL;
1984 bond_change_active_slave(bond, NULL);
1986 while ((slave = bond->first_slave) != NULL) {
1987 /* Inform AD package of unbinding of slave
1988 * before slave is detached from the list.
1990 if (bond->params.mode == BOND_MODE_8023AD)
1991 bond_3ad_unbind_slave(slave);
1993 slave_dev = slave->dev;
1994 bond_detach_slave(bond, slave);
1996 /* now that the slave is detached, unlock and perform
1997 * all the undo steps that should not be called from
2000 write_unlock_bh(&bond->lock);
2002 if (bond_is_lb(bond)) {
2003 /* must be called only after the slave
2004 * has been detached from the list
2006 bond_alb_deinit_slave(bond, slave);
2009 bond_compute_features(bond);
2011 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2012 bond_del_vlans_from_slave(bond, slave_dev);
2014 /* If the mode USES_PRIMARY, then we should only remove its
2015 * promisc and mc settings if it was the curr_active_slave, but that was
2016 * already taken care of above when we detached the slave
2018 if (!USES_PRIMARY(bond->params.mode)) {
2019 /* unset promiscuity level from slave */
2020 if (bond_dev->flags & IFF_PROMISC)
2021 dev_set_promiscuity(slave_dev, -1);
2023 /* unset allmulti level from slave */
2024 if (bond_dev->flags & IFF_ALLMULTI)
2025 dev_set_allmulti(slave_dev, -1);
2027 /* flush master's mc_list from slave */
2028 netif_addr_lock_bh(bond_dev);
2029 bond_mc_list_flush(bond_dev, slave_dev);
2030 netif_addr_unlock_bh(bond_dev);
2033 netdev_set_master(slave_dev, NULL);
2035 /* close slave before restoring its mac address */
2036 dev_close(slave_dev);
2038 if (!bond->params.fail_over_mac) {
2039 /* restore original ("permanent") mac address*/
2040 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2041 addr.sa_family = slave_dev->type;
2042 dev_set_mac_address(slave_dev, &addr);
2045 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2046 IFF_SLAVE_INACTIVE);
2050 /* re-acquire the lock before getting the next slave */
2051 write_lock_bh(&bond->lock);
2054 /* zero the mac address of the master so it will be
2055 * set by the application to the mac address of the
2058 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2060 if (list_empty(&bond->vlan_list))
2061 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2063 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2064 bond_dev->name, bond_dev->name);
2065 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2069 pr_info("%s: released all slaves\n", bond_dev->name);
2072 write_unlock_bh(&bond->lock);
2078 * This function changes the active slave to slave <slave_dev>.
2079 * It returns -EINVAL in the following cases.
2080 * - <slave_dev> is not found in the list.
2081 * - There is not active slave now.
2082 * - <slave_dev> is already active.
2083 * - The link state of <slave_dev> is not BOND_LINK_UP.
2084 * - <slave_dev> is not running.
2085 * In these cases, this function does nothing.
2086 * In the other cases, current_slave pointer is changed and 0 is returned.
2088 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2090 struct bonding *bond = netdev_priv(bond_dev);
2091 struct slave *old_active = NULL;
2092 struct slave *new_active = NULL;
2095 if (!USES_PRIMARY(bond->params.mode))
2098 /* Verify that master_dev is indeed the master of slave_dev */
2099 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2102 read_lock(&bond->lock);
2104 read_lock(&bond->curr_slave_lock);
2105 old_active = bond->curr_active_slave;
2106 read_unlock(&bond->curr_slave_lock);
2108 new_active = bond_get_slave_by_dev(bond, slave_dev);
2111 * Changing to the current active: do nothing; return success.
2113 if (new_active && (new_active == old_active)) {
2114 read_unlock(&bond->lock);
2120 (new_active->link == BOND_LINK_UP) &&
2121 IS_UP(new_active->dev)) {
2122 write_lock_bh(&bond->curr_slave_lock);
2123 bond_change_active_slave(bond, new_active);
2124 write_unlock_bh(&bond->curr_slave_lock);
2128 read_unlock(&bond->lock);
2133 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2135 struct bonding *bond = netdev_priv(bond_dev);
2137 info->bond_mode = bond->params.mode;
2138 info->miimon = bond->params.miimon;
2140 read_lock(&bond->lock);
2141 info->num_slaves = bond->slave_cnt;
2142 read_unlock(&bond->lock);
2147 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2149 struct bonding *bond = netdev_priv(bond_dev);
2150 struct slave *slave;
2151 int i, res = -ENODEV;
2153 read_lock(&bond->lock);
2155 bond_for_each_slave(bond, slave, i) {
2156 if (i == (int)info->slave_id) {
2158 strcpy(info->slave_name, slave->dev->name);
2159 info->link = slave->link;
2160 info->state = slave->state;
2161 info->link_failure_count = slave->link_failure_count;
2166 read_unlock(&bond->lock);
2171 /*-------------------------------- Monitoring -------------------------------*/
2174 static int bond_miimon_inspect(struct bonding *bond)
2176 struct slave *slave;
2177 int i, link_state, commit = 0;
2178 bool ignore_updelay;
2180 ignore_updelay = !bond->curr_active_slave ? true : false;
2182 bond_for_each_slave(bond, slave, i) {
2183 slave->new_link = BOND_LINK_NOCHANGE;
2185 link_state = bond_check_dev_link(bond, slave->dev, 0);
2187 switch (slave->link) {
2192 slave->link = BOND_LINK_FAIL;
2193 slave->delay = bond->params.downdelay;
2195 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2197 (bond->params.mode ==
2198 BOND_MODE_ACTIVEBACKUP) ?
2199 ((slave->state == BOND_STATE_ACTIVE) ?
2200 "active " : "backup ") : "",
2202 bond->params.downdelay * bond->params.miimon);
2205 case BOND_LINK_FAIL:
2208 * recovered before downdelay expired
2210 slave->link = BOND_LINK_UP;
2211 slave->jiffies = jiffies;
2212 pr_info("%s: link status up again after %d ms for interface %s.\n",
2214 (bond->params.downdelay - slave->delay) *
2215 bond->params.miimon,
2220 if (slave->delay <= 0) {
2221 slave->new_link = BOND_LINK_DOWN;
2229 case BOND_LINK_DOWN:
2233 slave->link = BOND_LINK_BACK;
2234 slave->delay = bond->params.updelay;
2237 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2238 bond->dev->name, slave->dev->name,
2239 ignore_updelay ? 0 :
2240 bond->params.updelay *
2241 bond->params.miimon);
2244 case BOND_LINK_BACK:
2246 slave->link = BOND_LINK_DOWN;
2247 pr_info("%s: link status down again after %d ms for interface %s.\n",
2249 (bond->params.updelay - slave->delay) *
2250 bond->params.miimon,
2259 if (slave->delay <= 0) {
2260 slave->new_link = BOND_LINK_UP;
2262 ignore_updelay = false;
2274 static void bond_miimon_commit(struct bonding *bond)
2276 struct slave *slave;
2279 bond_for_each_slave(bond, slave, i) {
2280 switch (slave->new_link) {
2281 case BOND_LINK_NOCHANGE:
2285 slave->link = BOND_LINK_UP;
2286 slave->jiffies = jiffies;
2288 if (bond->params.mode == BOND_MODE_8023AD) {
2289 /* prevent it from being the active one */
2290 slave->state = BOND_STATE_BACKUP;
2291 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2292 /* make it immediately active */
2293 slave->state = BOND_STATE_ACTIVE;
2294 } else if (slave != bond->primary_slave) {
2295 /* prevent it from being the active one */
2296 slave->state = BOND_STATE_BACKUP;
2299 pr_info("%s: link status definitely up for interface %s.\n",
2300 bond->dev->name, slave->dev->name);
2302 /* notify ad that the link status has changed */
2303 if (bond->params.mode == BOND_MODE_8023AD)
2304 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2306 if (bond_is_lb(bond))
2307 bond_alb_handle_link_change(bond, slave,
2310 if (!bond->curr_active_slave ||
2311 (slave == bond->primary_slave))
2316 case BOND_LINK_DOWN:
2317 if (slave->link_failure_count < UINT_MAX)
2318 slave->link_failure_count++;
2320 slave->link = BOND_LINK_DOWN;
2322 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2323 bond->params.mode == BOND_MODE_8023AD)
2324 bond_set_slave_inactive_flags(slave);
2326 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2327 bond->dev->name, slave->dev->name);
2329 if (bond->params.mode == BOND_MODE_8023AD)
2330 bond_3ad_handle_link_change(slave,
2333 if (bond_is_lb(bond))
2334 bond_alb_handle_link_change(bond, slave,
2337 if (slave == bond->curr_active_slave)
2343 pr_err("%s: invalid new link %d on slave %s\n",
2344 bond->dev->name, slave->new_link,
2346 slave->new_link = BOND_LINK_NOCHANGE;
2353 write_lock_bh(&bond->curr_slave_lock);
2354 bond_select_active_slave(bond);
2355 write_unlock_bh(&bond->curr_slave_lock);
2358 bond_set_carrier(bond);
2364 * Really a wrapper that splits the mii monitor into two phases: an
2365 * inspection, then (if inspection indicates something needs to be done)
2366 * an acquisition of appropriate locks followed by a commit phase to
2367 * implement whatever link state changes are indicated.
2369 void bond_mii_monitor(struct work_struct *work)
2371 struct bonding *bond = container_of(work, struct bonding,
2374 read_lock(&bond->lock);
2375 if (bond->kill_timers)
2378 if (bond->slave_cnt == 0)
2381 if (bond->send_grat_arp) {
2382 read_lock(&bond->curr_slave_lock);
2383 bond_send_gratuitous_arp(bond);
2384 read_unlock(&bond->curr_slave_lock);
2387 if (bond->send_unsol_na) {
2388 read_lock(&bond->curr_slave_lock);
2389 bond_send_unsolicited_na(bond);
2390 read_unlock(&bond->curr_slave_lock);
2393 if (bond_miimon_inspect(bond)) {
2394 read_unlock(&bond->lock);
2396 read_lock(&bond->lock);
2398 bond_miimon_commit(bond);
2400 read_unlock(&bond->lock);
2401 rtnl_unlock(); /* might sleep, hold no other locks */
2402 read_lock(&bond->lock);
2406 if (bond->params.miimon)
2407 queue_delayed_work(bond->wq, &bond->mii_work,
2408 msecs_to_jiffies(bond->params.miimon));
2410 read_unlock(&bond->lock);
2413 static __be32 bond_glean_dev_ip(struct net_device *dev)
2415 struct in_device *idev;
2416 struct in_ifaddr *ifa;
2423 idev = __in_dev_get_rcu(dev);
2427 ifa = idev->ifa_list;
2431 addr = ifa->ifa_local;
2437 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2439 struct vlan_entry *vlan;
2441 if (ip == bond->master_ip)
2444 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2445 if (ip == vlan->vlan_ip)
2453 * We go to the (large) trouble of VLAN tagging ARP frames because
2454 * switches in VLAN mode (especially if ports are configured as
2455 * "native" to a VLAN) might not pass non-tagged frames.
2457 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2459 struct sk_buff *skb;
2461 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2462 slave_dev->name, dest_ip, src_ip, vlan_id);
2464 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2465 NULL, slave_dev->dev_addr, NULL);
2468 pr_err("ARP packet allocation failed\n");
2472 skb = vlan_put_tag(skb, vlan_id);
2474 pr_err("failed to insert VLAN tag\n");
2482 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2485 __be32 *targets = bond->params.arp_targets;
2486 struct vlan_entry *vlan;
2487 struct net_device *vlan_dev;
2491 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2494 pr_debug("basa: target %x\n", targets[i]);
2495 if (list_empty(&bond->vlan_list)) {
2496 pr_debug("basa: empty vlan: arp_send\n");
2497 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2498 bond->master_ip, 0);
2503 * If VLANs are configured, we do a route lookup to
2504 * determine which VLAN interface would be used, so we
2505 * can tag the ARP with the proper VLAN tag.
2507 memset(&fl, 0, sizeof(fl));
2508 fl.fl4_dst = targets[i];
2509 fl.fl4_tos = RTO_ONLINK;
2511 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2513 if (net_ratelimit()) {
2514 pr_warning("%s: no route to arp_ip_target %pI4\n",
2515 bond->dev->name, &fl.fl4_dst);
2521 * This target is not on a VLAN
2523 if (rt->u.dst.dev == bond->dev) {
2525 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2526 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2527 bond->master_ip, 0);
2532 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2533 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2534 if (vlan_dev == rt->u.dst.dev) {
2535 vlan_id = vlan->vlan_id;
2536 pr_debug("basa: vlan match on %s %d\n",
2537 vlan_dev->name, vlan_id);
2544 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2545 vlan->vlan_ip, vlan_id);
2549 if (net_ratelimit()) {
2550 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2551 bond->dev->name, &fl.fl4_dst,
2552 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2559 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2560 * for each VLAN above us.
2562 * Caller must hold curr_slave_lock for read or better
2564 static void bond_send_gratuitous_arp(struct bonding *bond)
2566 struct slave *slave = bond->curr_active_slave;
2567 struct vlan_entry *vlan;
2568 struct net_device *vlan_dev;
2570 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2571 bond->dev->name, slave ? slave->dev->name : "NULL");
2573 if (!slave || !bond->send_grat_arp ||
2574 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2577 bond->send_grat_arp--;
2579 if (bond->master_ip) {
2580 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2581 bond->master_ip, 0);
2584 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2585 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2586 if (vlan->vlan_ip) {
2587 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2588 vlan->vlan_ip, vlan->vlan_id);
2593 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2596 __be32 *targets = bond->params.arp_targets;
2598 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2599 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2600 &sip, &tip, i, &targets[i],
2601 bond_has_this_ip(bond, tip));
2602 if (sip == targets[i]) {
2603 if (bond_has_this_ip(bond, tip))
2604 slave->last_arp_rx = jiffies;
2610 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2613 struct slave *slave;
2614 struct bonding *bond;
2615 unsigned char *arp_ptr;
2618 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2620 * When using VLANS and bonding, dev and oriv_dev may be
2621 * incorrect if the physical interface supports VLAN
2622 * acceleration. With this change ARP validation now
2623 * works for hosts only reachable on the VLAN interface.
2625 dev = vlan_dev_real_dev(dev);
2626 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2629 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2632 bond = netdev_priv(dev);
2633 read_lock(&bond->lock);
2635 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2636 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2637 orig_dev ? orig_dev->name : "NULL");
2639 slave = bond_get_slave_by_dev(bond, orig_dev);
2640 if (!slave || !slave_do_arp_validate(bond, slave))
2643 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2647 if (arp->ar_hln != dev->addr_len ||
2648 skb->pkt_type == PACKET_OTHERHOST ||
2649 skb->pkt_type == PACKET_LOOPBACK ||
2650 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2651 arp->ar_pro != htons(ETH_P_IP) ||
2655 arp_ptr = (unsigned char *)(arp + 1);
2656 arp_ptr += dev->addr_len;
2657 memcpy(&sip, arp_ptr, 4);
2658 arp_ptr += 4 + dev->addr_len;
2659 memcpy(&tip, arp_ptr, 4);
2661 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2662 bond->dev->name, slave->dev->name, slave->state,
2663 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2667 * Backup slaves won't see the ARP reply, but do come through
2668 * here for each ARP probe (so we swap the sip/tip to validate
2669 * the probe). In a "redundant switch, common router" type of
2670 * configuration, the ARP probe will (hopefully) travel from
2671 * the active, through one switch, the router, then the other
2672 * switch before reaching the backup.
2674 if (slave->state == BOND_STATE_ACTIVE)
2675 bond_validate_arp(bond, slave, sip, tip);
2677 bond_validate_arp(bond, slave, tip, sip);
2680 read_unlock(&bond->lock);
2683 return NET_RX_SUCCESS;
2687 * this function is called regularly to monitor each slave's link
2688 * ensuring that traffic is being sent and received when arp monitoring
2689 * is used in load-balancing mode. if the adapter has been dormant, then an
2690 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2691 * arp monitoring in active backup mode.
2693 void bond_loadbalance_arp_mon(struct work_struct *work)
2695 struct bonding *bond = container_of(work, struct bonding,
2697 struct slave *slave, *oldcurrent;
2698 int do_failover = 0;
2702 read_lock(&bond->lock);
2704 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2706 if (bond->kill_timers)
2709 if (bond->slave_cnt == 0)
2712 read_lock(&bond->curr_slave_lock);
2713 oldcurrent = bond->curr_active_slave;
2714 read_unlock(&bond->curr_slave_lock);
2716 /* see if any of the previous devices are up now (i.e. they have
2717 * xmt and rcv traffic). the curr_active_slave does not come into
2718 * the picture unless it is null. also, slave->jiffies is not needed
2719 * here because we send an arp on each slave and give a slave as
2720 * long as it needs to get the tx/rx within the delta.
2721 * TODO: what about up/down delay in arp mode? it wasn't here before
2724 bond_for_each_slave(bond, slave, i) {
2725 if (slave->link != BOND_LINK_UP) {
2726 if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2727 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2729 slave->link = BOND_LINK_UP;
2730 slave->state = BOND_STATE_ACTIVE;
2732 /* primary_slave has no meaning in round-robin
2733 * mode. the window of a slave being up and
2734 * curr_active_slave being null after enslaving
2738 pr_info("%s: link status definitely up for interface %s, ",
2743 pr_info("%s: interface %s is now up\n",
2749 /* slave->link == BOND_LINK_UP */
2751 /* not all switches will respond to an arp request
2752 * when the source ip is 0, so don't take the link down
2753 * if we don't know our ip yet
2755 if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2756 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2758 slave->link = BOND_LINK_DOWN;
2759 slave->state = BOND_STATE_BACKUP;
2761 if (slave->link_failure_count < UINT_MAX)
2762 slave->link_failure_count++;
2764 pr_info("%s: interface %s is now down.\n",
2768 if (slave == oldcurrent)
2773 /* note: if switch is in round-robin mode, all links
2774 * must tx arp to ensure all links rx an arp - otherwise
2775 * links may oscillate or not come up at all; if switch is
2776 * in something like xor mode, there is nothing we can
2777 * do - all replies will be rx'ed on same link causing slaves
2778 * to be unstable during low/no traffic periods
2780 if (IS_UP(slave->dev))
2781 bond_arp_send_all(bond, slave);
2785 write_lock_bh(&bond->curr_slave_lock);
2787 bond_select_active_slave(bond);
2789 write_unlock_bh(&bond->curr_slave_lock);
2793 if (bond->params.arp_interval)
2794 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2796 read_unlock(&bond->lock);
2800 * Called to inspect slaves for active-backup mode ARP monitor link state
2801 * changes. Sets new_link in slaves to specify what action should take
2802 * place for the slave. Returns 0 if no changes are found, >0 if changes
2803 * to link states must be committed.
2805 * Called with bond->lock held for read.
2807 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2809 struct slave *slave;
2812 bond_for_each_slave(bond, slave, i) {
2813 slave->new_link = BOND_LINK_NOCHANGE;
2815 if (slave->link != BOND_LINK_UP) {
2816 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2818 slave->new_link = BOND_LINK_UP;
2826 * Give slaves 2*delta after being enslaved or made
2827 * active. This avoids bouncing, as the last receive
2828 * times need a full ARP monitor cycle to be updated.
2830 if (!time_after_eq(jiffies, slave->jiffies +
2831 2 * delta_in_ticks))
2835 * Backup slave is down if:
2836 * - No current_arp_slave AND
2837 * - more than 3*delta since last receive AND
2838 * - the bond has an IP address
2840 * Note: a non-null current_arp_slave indicates
2841 * the curr_active_slave went down and we are
2842 * searching for a new one; under this condition
2843 * we only take the curr_active_slave down - this
2844 * gives each slave a chance to tx/rx traffic
2845 * before being taken out
2847 if (slave->state == BOND_STATE_BACKUP &&
2848 !bond->current_arp_slave &&
2849 time_after(jiffies, slave_last_rx(bond, slave) +
2850 3 * delta_in_ticks)) {
2851 slave->new_link = BOND_LINK_DOWN;
2856 * Active slave is down if:
2857 * - more than 2*delta since transmitting OR
2858 * - (more than 2*delta since receive AND
2859 * the bond has an IP address)
2861 if ((slave->state == BOND_STATE_ACTIVE) &&
2862 (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2863 2 * delta_in_ticks) ||
2864 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2865 + 2 * delta_in_ticks)))) {
2866 slave->new_link = BOND_LINK_DOWN;
2875 * Called to commit link state changes noted by inspection step of
2876 * active-backup mode ARP monitor.
2878 * Called with RTNL and bond->lock for read.
2880 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2882 struct slave *slave;
2885 bond_for_each_slave(bond, slave, i) {
2886 switch (slave->new_link) {
2887 case BOND_LINK_NOCHANGE:
2891 if ((!bond->curr_active_slave &&
2892 time_before_eq(jiffies,
2893 dev_trans_start(slave->dev) +
2895 bond->curr_active_slave != slave) {
2896 slave->link = BOND_LINK_UP;
2897 bond->current_arp_slave = NULL;
2899 pr_info("%s: link status definitely up for interface %s.\n",
2900 bond->dev->name, slave->dev->name);
2902 if (!bond->curr_active_slave ||
2903 (slave == bond->primary_slave))
2910 case BOND_LINK_DOWN:
2911 if (slave->link_failure_count < UINT_MAX)
2912 slave->link_failure_count++;
2914 slave->link = BOND_LINK_DOWN;
2915 bond_set_slave_inactive_flags(slave);
2917 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2918 bond->dev->name, slave->dev->name);
2920 if (slave == bond->curr_active_slave) {
2921 bond->current_arp_slave = NULL;
2928 pr_err("%s: impossible: new_link %d on slave %s\n",
2929 bond->dev->name, slave->new_link,
2936 write_lock_bh(&bond->curr_slave_lock);
2937 bond_select_active_slave(bond);
2938 write_unlock_bh(&bond->curr_slave_lock);
2941 bond_set_carrier(bond);
2945 * Send ARP probes for active-backup mode ARP monitor.
2947 * Called with bond->lock held for read.
2949 static void bond_ab_arp_probe(struct bonding *bond)
2951 struct slave *slave;
2954 read_lock(&bond->curr_slave_lock);
2956 if (bond->current_arp_slave && bond->curr_active_slave)
2957 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2958 bond->current_arp_slave->dev->name,
2959 bond->curr_active_slave->dev->name);
2961 if (bond->curr_active_slave) {
2962 bond_arp_send_all(bond, bond->curr_active_slave);
2963 read_unlock(&bond->curr_slave_lock);
2967 read_unlock(&bond->curr_slave_lock);
2969 /* if we don't have a curr_active_slave, search for the next available
2970 * backup slave from the current_arp_slave and make it the candidate
2971 * for becoming the curr_active_slave
2974 if (!bond->current_arp_slave) {
2975 bond->current_arp_slave = bond->first_slave;
2976 if (!bond->current_arp_slave)
2980 bond_set_slave_inactive_flags(bond->current_arp_slave);
2982 /* search for next candidate */
2983 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2984 if (IS_UP(slave->dev)) {
2985 slave->link = BOND_LINK_BACK;
2986 bond_set_slave_active_flags(slave);
2987 bond_arp_send_all(bond, slave);
2988 slave->jiffies = jiffies;
2989 bond->current_arp_slave = slave;
2993 /* if the link state is up at this point, we
2994 * mark it down - this can happen if we have
2995 * simultaneous link failures and
2996 * reselect_active_interface doesn't make this
2997 * one the current slave so it is still marked
2998 * up when it is actually down
3000 if (slave->link == BOND_LINK_UP) {
3001 slave->link = BOND_LINK_DOWN;
3002 if (slave->link_failure_count < UINT_MAX)
3003 slave->link_failure_count++;
3005 bond_set_slave_inactive_flags(slave);
3007 pr_info("%s: backup interface %s is now down.\n",
3008 bond->dev->name, slave->dev->name);
3013 void bond_activebackup_arp_mon(struct work_struct *work)
3015 struct bonding *bond = container_of(work, struct bonding,
3019 read_lock(&bond->lock);
3021 if (bond->kill_timers)
3024 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3026 if (bond->slave_cnt == 0)
3029 if (bond->send_grat_arp) {
3030 read_lock(&bond->curr_slave_lock);
3031 bond_send_gratuitous_arp(bond);
3032 read_unlock(&bond->curr_slave_lock);
3035 if (bond->send_unsol_na) {
3036 read_lock(&bond->curr_slave_lock);
3037 bond_send_unsolicited_na(bond);
3038 read_unlock(&bond->curr_slave_lock);
3041 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3042 read_unlock(&bond->lock);
3044 read_lock(&bond->lock);
3046 bond_ab_arp_commit(bond, delta_in_ticks);
3048 read_unlock(&bond->lock);
3050 read_lock(&bond->lock);
3053 bond_ab_arp_probe(bond);
3056 if (bond->params.arp_interval)
3057 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3059 read_unlock(&bond->lock);
3062 /*------------------------------ proc/seq_file-------------------------------*/
3064 #ifdef CONFIG_PROC_FS
3066 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3067 __acquires(&dev_base_lock)
3068 __acquires(&bond->lock)
3070 struct bonding *bond = seq->private;
3072 struct slave *slave;
3075 /* make sure the bond won't be taken away */
3076 read_lock(&dev_base_lock);
3077 read_lock(&bond->lock);
3080 return SEQ_START_TOKEN;
3082 bond_for_each_slave(bond, slave, i) {
3090 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3092 struct bonding *bond = seq->private;
3093 struct slave *slave = v;
3096 if (v == SEQ_START_TOKEN)
3097 return bond->first_slave;
3099 slave = slave->next;
3101 return (slave == bond->first_slave) ? NULL : slave;
3104 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3105 __releases(&bond->lock)
3106 __releases(&dev_base_lock)
3108 struct bonding *bond = seq->private;
3110 read_unlock(&bond->lock);
3111 read_unlock(&dev_base_lock);
3114 static void bond_info_show_master(struct seq_file *seq)
3116 struct bonding *bond = seq->private;
3120 read_lock(&bond->curr_slave_lock);
3121 curr = bond->curr_active_slave;
3122 read_unlock(&bond->curr_slave_lock);
3124 seq_printf(seq, "Bonding Mode: %s",
3125 bond_mode_name(bond->params.mode));
3127 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3128 bond->params.fail_over_mac)
3129 seq_printf(seq, " (fail_over_mac %s)",
3130 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3132 seq_printf(seq, "\n");
3134 if (bond->params.mode == BOND_MODE_XOR ||
3135 bond->params.mode == BOND_MODE_8023AD) {
3136 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3137 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3138 bond->params.xmit_policy);
3141 if (USES_PRIMARY(bond->params.mode)) {
3142 seq_printf(seq, "Primary Slave: %s",
3143 (bond->primary_slave) ?
3144 bond->primary_slave->dev->name : "None");
3145 if (bond->primary_slave)
3146 seq_printf(seq, " (primary_reselect %s)",
3147 pri_reselect_tbl[bond->params.primary_reselect].modename);
3149 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3150 (curr) ? curr->dev->name : "None");
3153 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3155 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3156 seq_printf(seq, "Up Delay (ms): %d\n",
3157 bond->params.updelay * bond->params.miimon);
3158 seq_printf(seq, "Down Delay (ms): %d\n",
3159 bond->params.downdelay * bond->params.miimon);
3162 /* ARP information */
3163 if (bond->params.arp_interval > 0) {
3165 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3166 bond->params.arp_interval);
3168 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3170 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3171 if (!bond->params.arp_targets[i])
3174 seq_printf(seq, ",");
3175 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3178 seq_printf(seq, "\n");
3181 if (bond->params.mode == BOND_MODE_8023AD) {
3182 struct ad_info ad_info;
3184 seq_puts(seq, "\n802.3ad info\n");
3185 seq_printf(seq, "LACP rate: %s\n",
3186 (bond->params.lacp_fast) ? "fast" : "slow");
3187 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3188 ad_select_tbl[bond->params.ad_select].modename);
3190 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3191 seq_printf(seq, "bond %s has no active aggregator\n",
3194 seq_printf(seq, "Active Aggregator Info:\n");
3196 seq_printf(seq, "\tAggregator ID: %d\n",
3197 ad_info.aggregator_id);
3198 seq_printf(seq, "\tNumber of ports: %d\n",
3200 seq_printf(seq, "\tActor Key: %d\n",
3202 seq_printf(seq, "\tPartner Key: %d\n",
3203 ad_info.partner_key);
3204 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3205 ad_info.partner_system);
3210 static void bond_info_show_slave(struct seq_file *seq,
3211 const struct slave *slave)
3213 struct bonding *bond = seq->private;
3215 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3216 seq_printf(seq, "MII Status: %s\n",
3217 (slave->link == BOND_LINK_UP) ? "up" : "down");
3218 seq_printf(seq, "Link Failure Count: %u\n",
3219 slave->link_failure_count);
3221 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3223 if (bond->params.mode == BOND_MODE_8023AD) {
3224 const struct aggregator *agg
3225 = SLAVE_AD_INFO(slave).port.aggregator;
3228 seq_printf(seq, "Aggregator ID: %d\n",
3229 agg->aggregator_identifier);
3231 seq_puts(seq, "Aggregator ID: N/A\n");
3235 static int bond_info_seq_show(struct seq_file *seq, void *v)
3237 if (v == SEQ_START_TOKEN) {
3238 seq_printf(seq, "%s\n", version);
3239 bond_info_show_master(seq);
3241 bond_info_show_slave(seq, v);
3246 static const struct seq_operations bond_info_seq_ops = {
3247 .start = bond_info_seq_start,
3248 .next = bond_info_seq_next,
3249 .stop = bond_info_seq_stop,
3250 .show = bond_info_seq_show,
3253 static int bond_info_open(struct inode *inode, struct file *file)
3255 struct seq_file *seq;
3256 struct proc_dir_entry *proc;
3259 res = seq_open(file, &bond_info_seq_ops);
3261 /* recover the pointer buried in proc_dir_entry data */
3262 seq = file->private_data;
3264 seq->private = proc->data;
3270 static const struct file_operations bond_info_fops = {
3271 .owner = THIS_MODULE,
3272 .open = bond_info_open,
3274 .llseek = seq_lseek,
3275 .release = seq_release,
3278 static void bond_create_proc_entry(struct bonding *bond)
3280 struct net_device *bond_dev = bond->dev;
3281 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3284 bond->proc_entry = proc_create_data(bond_dev->name,
3285 S_IRUGO, bn->proc_dir,
3286 &bond_info_fops, bond);
3287 if (bond->proc_entry == NULL)
3288 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3289 DRV_NAME, bond_dev->name);
3291 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3295 static void bond_remove_proc_entry(struct bonding *bond)
3297 struct net_device *bond_dev = bond->dev;
3298 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3300 if (bn->proc_dir && bond->proc_entry) {
3301 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3302 memset(bond->proc_file_name, 0, IFNAMSIZ);
3303 bond->proc_entry = NULL;
3307 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3308 * Caller must hold rtnl_lock.
3310 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3312 if (!bn->proc_dir) {
3313 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3315 pr_warning("Warning: cannot create /proc/net/%s\n",
3320 /* Destroy the bonding directory under /proc/net, if empty.
3321 * Caller must hold rtnl_lock.
3323 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3326 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3327 bn->proc_dir = NULL;
3331 #else /* !CONFIG_PROC_FS */
3333 static void bond_create_proc_entry(struct bonding *bond)
3337 static void bond_remove_proc_entry(struct bonding *bond)
3341 static inline void bond_create_proc_dir(struct bond_net *bn)
3345 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3349 #endif /* CONFIG_PROC_FS */
3352 /*-------------------------- netdev event handling --------------------------*/
3355 * Change device name
3357 static int bond_event_changename(struct bonding *bond)
3359 bond_remove_proc_entry(bond);
3360 bond_create_proc_entry(bond);
3365 static int bond_master_netdev_event(unsigned long event,
3366 struct net_device *bond_dev)
3368 struct bonding *event_bond = netdev_priv(bond_dev);
3371 case NETDEV_CHANGENAME:
3372 return bond_event_changename(event_bond);
3380 static int bond_slave_netdev_event(unsigned long event,
3381 struct net_device *slave_dev)
3383 struct net_device *bond_dev = slave_dev->master;
3384 struct bonding *bond = netdev_priv(bond_dev);
3387 case NETDEV_UNREGISTER:
3389 if (bond->setup_by_slave)
3390 bond_release_and_destroy(bond_dev, slave_dev);
3392 bond_release(bond_dev, slave_dev);
3396 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3397 struct slave *slave;
3399 slave = bond_get_slave_by_dev(bond, slave_dev);
3401 u16 old_speed = slave->speed;
3402 u16 old_duplex = slave->duplex;
3404 bond_update_speed_duplex(slave);
3406 if (bond_is_lb(bond))
3409 if (old_speed != slave->speed)
3410 bond_3ad_adapter_speed_changed(slave);
3411 if (old_duplex != slave->duplex)
3412 bond_3ad_adapter_duplex_changed(slave);
3419 * ... Or is it this?
3422 case NETDEV_CHANGEMTU:
3424 * TODO: Should slaves be allowed to
3425 * independently alter their MTU? For
3426 * an active-backup bond, slaves need
3427 * not be the same type of device, so
3428 * MTUs may vary. For other modes,
3429 * slaves arguably should have the
3430 * same MTUs. To do this, we'd need to
3431 * take over the slave's change_mtu
3432 * function for the duration of their
3436 case NETDEV_CHANGENAME:
3438 * TODO: handle changing the primary's name
3441 case NETDEV_FEAT_CHANGE:
3442 bond_compute_features(bond);
3452 * bond_netdev_event: handle netdev notifier chain events.
3454 * This function receives events for the netdev chain. The caller (an
3455 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3456 * locks for us to safely manipulate the slave devices (RTNL lock,
3459 static int bond_netdev_event(struct notifier_block *this,
3460 unsigned long event, void *ptr)
3462 struct net_device *event_dev = (struct net_device *)ptr;
3464 pr_debug("event_dev: %s, event: %lx\n",
3465 event_dev ? event_dev->name : "None",
3468 if (!(event_dev->priv_flags & IFF_BONDING))
3471 if (event_dev->flags & IFF_MASTER) {
3472 pr_debug("IFF_MASTER\n");
3473 return bond_master_netdev_event(event, event_dev);
3476 if (event_dev->flags & IFF_SLAVE) {
3477 pr_debug("IFF_SLAVE\n");
3478 return bond_slave_netdev_event(event, event_dev);
3485 * bond_inetaddr_event: handle inetaddr notifier chain events.
3487 * We keep track of device IPs primarily to use as source addresses in
3488 * ARP monitor probes (rather than spewing out broadcasts all the time).
3490 * We track one IP for the main device (if it has one), plus one per VLAN.
3492 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3494 struct in_ifaddr *ifa = ptr;
3495 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3496 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3497 struct bonding *bond;
3498 struct vlan_entry *vlan;
3500 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3501 if (bond->dev == event_dev) {
3504 bond->master_ip = ifa->ifa_local;
3507 bond->master_ip = bond_glean_dev_ip(bond->dev);
3514 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3515 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3516 if (vlan_dev == event_dev) {
3519 vlan->vlan_ip = ifa->ifa_local;
3523 bond_glean_dev_ip(vlan_dev);
3534 static struct notifier_block bond_netdev_notifier = {
3535 .notifier_call = bond_netdev_event,
3538 static struct notifier_block bond_inetaddr_notifier = {
3539 .notifier_call = bond_inetaddr_event,
3542 /*-------------------------- Packet type handling ---------------------------*/
3544 /* register to receive lacpdus on a bond */
3545 static void bond_register_lacpdu(struct bonding *bond)
3547 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3549 /* initialize packet type */
3550 pk_type->type = PKT_TYPE_LACPDU;
3551 pk_type->dev = bond->dev;
3552 pk_type->func = bond_3ad_lacpdu_recv;
3554 dev_add_pack(pk_type);
3557 /* unregister to receive lacpdus on a bond */
3558 static void bond_unregister_lacpdu(struct bonding *bond)
3560 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3563 void bond_register_arp(struct bonding *bond)
3565 struct packet_type *pt = &bond->arp_mon_pt;
3570 pt->type = htons(ETH_P_ARP);
3571 pt->dev = bond->dev;
3572 pt->func = bond_arp_rcv;
3576 void bond_unregister_arp(struct bonding *bond)
3578 struct packet_type *pt = &bond->arp_mon_pt;
3580 dev_remove_pack(pt);
3584 /*---------------------------- Hashing Policies -----------------------------*/
3587 * Hash for the output device based upon layer 2 and layer 3 data. If
3588 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3590 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3592 struct ethhdr *data = (struct ethhdr *)skb->data;
3593 struct iphdr *iph = ip_hdr(skb);
3595 if (skb->protocol == htons(ETH_P_IP)) {
3596 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3597 (data->h_dest[5] ^ data->h_source[5])) % count;
3600 return (data->h_dest[5] ^ data->h_source[5]) % count;
3604 * Hash for the output device based upon layer 3 and layer 4 data. If
3605 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3606 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3608 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3610 struct ethhdr *data = (struct ethhdr *)skb->data;
3611 struct iphdr *iph = ip_hdr(skb);
3612 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3615 if (skb->protocol == htons(ETH_P_IP)) {
3616 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3617 (iph->protocol == IPPROTO_TCP ||
3618 iph->protocol == IPPROTO_UDP)) {
3619 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3621 return (layer4_xor ^
3622 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3626 return (data->h_dest[5] ^ data->h_source[5]) % count;
3630 * Hash for the output device based upon layer 2 data
3632 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3634 struct ethhdr *data = (struct ethhdr *)skb->data;
3636 return (data->h_dest[5] ^ data->h_source[5]) % count;
3639 /*-------------------------- Device entry points ----------------------------*/
3641 static int bond_open(struct net_device *bond_dev)
3643 struct bonding *bond = netdev_priv(bond_dev);
3645 bond->kill_timers = 0;
3647 if (bond_is_lb(bond)) {
3648 /* bond_alb_initialize must be called before the timer
3651 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3652 /* something went wrong - fail the open operation */
3656 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3657 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3660 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3661 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3662 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3665 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3666 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3667 INIT_DELAYED_WORK(&bond->arp_work,
3668 bond_activebackup_arp_mon);
3670 INIT_DELAYED_WORK(&bond->arp_work,
3671 bond_loadbalance_arp_mon);
3673 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3674 if (bond->params.arp_validate)
3675 bond_register_arp(bond);
3678 if (bond->params.mode == BOND_MODE_8023AD) {
3679 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3680 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3681 /* register to receive LACPDUs */
3682 bond_register_lacpdu(bond);
3683 bond_3ad_initiate_agg_selection(bond, 1);
3689 static int bond_close(struct net_device *bond_dev)
3691 struct bonding *bond = netdev_priv(bond_dev);
3693 if (bond->params.mode == BOND_MODE_8023AD) {
3694 /* Unregister the receive of LACPDUs */
3695 bond_unregister_lacpdu(bond);
3698 if (bond->params.arp_validate)
3699 bond_unregister_arp(bond);
3701 write_lock_bh(&bond->lock);
3703 bond->send_grat_arp = 0;
3704 bond->send_unsol_na = 0;
3706 /* signal timers not to re-arm */
3707 bond->kill_timers = 1;
3709 write_unlock_bh(&bond->lock);
3711 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3712 cancel_delayed_work(&bond->mii_work);
3715 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3716 cancel_delayed_work(&bond->arp_work);
3719 switch (bond->params.mode) {
3720 case BOND_MODE_8023AD:
3721 cancel_delayed_work(&bond->ad_work);
3725 cancel_delayed_work(&bond->alb_work);
3732 if (bond_is_lb(bond)) {
3733 /* Must be called only after all
3734 * slaves have been released
3736 bond_alb_deinitialize(bond);
3742 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3744 struct bonding *bond = netdev_priv(bond_dev);
3745 struct net_device_stats *stats = &bond_dev->stats;
3746 struct net_device_stats local_stats;
3747 struct slave *slave;
3750 memset(&local_stats, 0, sizeof(struct net_device_stats));
3752 read_lock_bh(&bond->lock);
3754 bond_for_each_slave(bond, slave, i) {
3755 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3757 local_stats.rx_packets += sstats->rx_packets;
3758 local_stats.rx_bytes += sstats->rx_bytes;
3759 local_stats.rx_errors += sstats->rx_errors;
3760 local_stats.rx_dropped += sstats->rx_dropped;
3762 local_stats.tx_packets += sstats->tx_packets;
3763 local_stats.tx_bytes += sstats->tx_bytes;
3764 local_stats.tx_errors += sstats->tx_errors;
3765 local_stats.tx_dropped += sstats->tx_dropped;
3767 local_stats.multicast += sstats->multicast;
3768 local_stats.collisions += sstats->collisions;
3770 local_stats.rx_length_errors += sstats->rx_length_errors;
3771 local_stats.rx_over_errors += sstats->rx_over_errors;
3772 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3773 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3774 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3775 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3777 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3778 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3779 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3780 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3781 local_stats.tx_window_errors += sstats->tx_window_errors;
3784 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3786 read_unlock_bh(&bond->lock);
3791 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3793 struct net_device *slave_dev = NULL;
3794 struct ifbond k_binfo;
3795 struct ifbond __user *u_binfo = NULL;
3796 struct ifslave k_sinfo;
3797 struct ifslave __user *u_sinfo = NULL;
3798 struct mii_ioctl_data *mii = NULL;
3801 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3813 * We do this again just in case we were called by SIOCGMIIREG
3814 * instead of SIOCGMIIPHY.
3821 if (mii->reg_num == 1) {
3822 struct bonding *bond = netdev_priv(bond_dev);
3824 read_lock(&bond->lock);
3825 read_lock(&bond->curr_slave_lock);
3826 if (netif_carrier_ok(bond->dev))
3827 mii->val_out = BMSR_LSTATUS;
3829 read_unlock(&bond->curr_slave_lock);
3830 read_unlock(&bond->lock);
3834 case BOND_INFO_QUERY_OLD:
3835 case SIOCBONDINFOQUERY:
3836 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3838 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3841 res = bond_info_query(bond_dev, &k_binfo);
3843 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3847 case BOND_SLAVE_INFO_QUERY_OLD:
3848 case SIOCBONDSLAVEINFOQUERY:
3849 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3851 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3854 res = bond_slave_info_query(bond_dev, &k_sinfo);
3856 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3865 if (!capable(CAP_NET_ADMIN))
3868 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3870 pr_debug("slave_dev=%p:\n", slave_dev);
3875 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3877 case BOND_ENSLAVE_OLD:
3878 case SIOCBONDENSLAVE:
3879 res = bond_enslave(bond_dev, slave_dev);
3881 case BOND_RELEASE_OLD:
3882 case SIOCBONDRELEASE:
3883 res = bond_release(bond_dev, slave_dev);
3885 case BOND_SETHWADDR_OLD:
3886 case SIOCBONDSETHWADDR:
3887 res = bond_sethwaddr(bond_dev, slave_dev);
3889 case BOND_CHANGE_ACTIVE_OLD:
3890 case SIOCBONDCHANGEACTIVE:
3891 res = bond_ioctl_change_active(bond_dev, slave_dev);
3903 static void bond_set_multicast_list(struct net_device *bond_dev)
3905 struct bonding *bond = netdev_priv(bond_dev);
3906 struct dev_mc_list *dmi;
3909 * Do promisc before checking multicast_mode
3911 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3913 * FIXME: Need to handle the error when one of the multi-slaves
3916 bond_set_promiscuity(bond, 1);
3919 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3920 bond_set_promiscuity(bond, -1);
3923 /* set allmulti flag to slaves */
3924 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3926 * FIXME: Need to handle the error when one of the multi-slaves
3929 bond_set_allmulti(bond, 1);
3932 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3933 bond_set_allmulti(bond, -1);
3936 read_lock(&bond->lock);
3938 bond->flags = bond_dev->flags;
3940 /* looking for addresses to add to slaves' mc list */
3941 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3942 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
3943 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3946 /* looking for addresses to delete from slaves' list */
3947 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3948 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
3949 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3952 /* save master's multicast list */
3953 bond_mc_list_destroy(bond);
3954 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3956 read_unlock(&bond->lock);
3959 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3961 struct bonding *bond = netdev_priv(dev);
3962 struct slave *slave = bond->first_slave;
3965 const struct net_device_ops *slave_ops
3966 = slave->dev->netdev_ops;
3967 if (slave_ops->ndo_neigh_setup)
3968 return slave_ops->ndo_neigh_setup(slave->dev, parms);
3974 * Change the MTU of all of a master's slaves to match the master
3976 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3978 struct bonding *bond = netdev_priv(bond_dev);
3979 struct slave *slave, *stop_at;
3983 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3984 (bond_dev ? bond_dev->name : "None"), new_mtu);
3986 /* Can't hold bond->lock with bh disabled here since
3987 * some base drivers panic. On the other hand we can't
3988 * hold bond->lock without bh disabled because we'll
3989 * deadlock. The only solution is to rely on the fact
3990 * that we're under rtnl_lock here, and the slaves
3991 * list won't change. This doesn't solve the problem
3992 * of setting the slave's MTU while it is
3993 * transmitting, but the assumption is that the base
3994 * driver can handle that.
3996 * TODO: figure out a way to safely iterate the slaves
3997 * list, but without holding a lock around the actual
3998 * call to the base driver.
4001 bond_for_each_slave(bond, slave, i) {
4002 pr_debug("s %p s->p %p c_m %p\n",
4005 slave->dev->netdev_ops->ndo_change_mtu);
4007 res = dev_set_mtu(slave->dev, new_mtu);
4010 /* If we failed to set the slave's mtu to the new value
4011 * we must abort the operation even in ACTIVE_BACKUP
4012 * mode, because if we allow the backup slaves to have
4013 * different mtu values than the active slave we'll
4014 * need to change their mtu when doing a failover. That
4015 * means changing their mtu from timer context, which
4016 * is probably not a good idea.
4018 pr_debug("err %d %s\n", res, slave->dev->name);
4023 bond_dev->mtu = new_mtu;
4028 /* unwind from head to the slave that failed */
4030 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4033 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4035 pr_debug("unwind err %d dev %s\n",
4036 tmp_res, slave->dev->name);
4046 * Note that many devices must be down to change the HW address, and
4047 * downing the master releases all slaves. We can make bonds full of
4048 * bonding devices to test this, however.
4050 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4052 struct bonding *bond = netdev_priv(bond_dev);
4053 struct sockaddr *sa = addr, tmp_sa;
4054 struct slave *slave, *stop_at;
4058 if (bond->params.mode == BOND_MODE_ALB)
4059 return bond_alb_set_mac_address(bond_dev, addr);
4062 pr_debug("bond=%p, name=%s\n",
4063 bond, bond_dev ? bond_dev->name : "None");
4066 * If fail_over_mac is set to active, do nothing and return
4067 * success. Returning an error causes ifenslave to fail.
4069 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4072 if (!is_valid_ether_addr(sa->sa_data))
4073 return -EADDRNOTAVAIL;
4075 /* Can't hold bond->lock with bh disabled here since
4076 * some base drivers panic. On the other hand we can't
4077 * hold bond->lock without bh disabled because we'll
4078 * deadlock. The only solution is to rely on the fact
4079 * that we're under rtnl_lock here, and the slaves
4080 * list won't change. This doesn't solve the problem
4081 * of setting the slave's hw address while it is
4082 * transmitting, but the assumption is that the base
4083 * driver can handle that.
4085 * TODO: figure out a way to safely iterate the slaves
4086 * list, but without holding a lock around the actual
4087 * call to the base driver.
4090 bond_for_each_slave(bond, slave, i) {
4091 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4092 pr_debug("slave %p %s\n", slave, slave->dev->name);
4094 if (slave_ops->ndo_set_mac_address == NULL) {
4096 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4100 res = dev_set_mac_address(slave->dev, addr);
4102 /* TODO: consider downing the slave
4104 * User should expect communications
4105 * breakage anyway until ARP finish
4108 pr_debug("err %d %s\n", res, slave->dev->name);
4114 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4118 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4119 tmp_sa.sa_family = bond_dev->type;
4121 /* unwind from head to the slave that failed */
4123 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4126 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4128 pr_debug("unwind err %d dev %s\n",
4129 tmp_res, slave->dev->name);
4136 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4138 struct bonding *bond = netdev_priv(bond_dev);
4139 struct slave *slave, *start_at;
4140 int i, slave_no, res = 1;
4142 read_lock(&bond->lock);
4144 if (!BOND_IS_OK(bond))
4148 * Concurrent TX may collide on rr_tx_counter; we accept that
4149 * as being rare enough not to justify using an atomic op here
4151 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4153 bond_for_each_slave(bond, slave, i) {
4160 bond_for_each_slave_from(bond, slave, i, start_at) {
4161 if (IS_UP(slave->dev) &&
4162 (slave->link == BOND_LINK_UP) &&
4163 (slave->state == BOND_STATE_ACTIVE)) {
4164 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4171 /* no suitable interface, frame not sent */
4174 read_unlock(&bond->lock);
4175 return NETDEV_TX_OK;
4180 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4181 * the bond has a usable interface.
4183 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4185 struct bonding *bond = netdev_priv(bond_dev);
4188 read_lock(&bond->lock);
4189 read_lock(&bond->curr_slave_lock);
4191 if (!BOND_IS_OK(bond))
4194 if (!bond->curr_active_slave)
4197 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4201 /* no suitable interface, frame not sent */
4204 read_unlock(&bond->curr_slave_lock);
4205 read_unlock(&bond->lock);
4206 return NETDEV_TX_OK;
4210 * In bond_xmit_xor() , we determine the output device by using a pre-
4211 * determined xmit_hash_policy(), If the selected device is not enabled,
4212 * find the next active slave.
4214 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4216 struct bonding *bond = netdev_priv(bond_dev);
4217 struct slave *slave, *start_at;
4222 read_lock(&bond->lock);
4224 if (!BOND_IS_OK(bond))
4227 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4229 bond_for_each_slave(bond, slave, i) {
4237 bond_for_each_slave_from(bond, slave, i, start_at) {
4238 if (IS_UP(slave->dev) &&
4239 (slave->link == BOND_LINK_UP) &&
4240 (slave->state == BOND_STATE_ACTIVE)) {
4241 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4248 /* no suitable interface, frame not sent */
4251 read_unlock(&bond->lock);
4252 return NETDEV_TX_OK;
4256 * in broadcast mode, we send everything to all usable interfaces.
4258 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4260 struct bonding *bond = netdev_priv(bond_dev);
4261 struct slave *slave, *start_at;
4262 struct net_device *tx_dev = NULL;
4266 read_lock(&bond->lock);
4268 if (!BOND_IS_OK(bond))
4271 read_lock(&bond->curr_slave_lock);
4272 start_at = bond->curr_active_slave;
4273 read_unlock(&bond->curr_slave_lock);
4278 bond_for_each_slave_from(bond, slave, i, start_at) {
4279 if (IS_UP(slave->dev) &&
4280 (slave->link == BOND_LINK_UP) &&
4281 (slave->state == BOND_STATE_ACTIVE)) {
4283 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4285 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4290 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4292 dev_kfree_skb(skb2);
4296 tx_dev = slave->dev;
4301 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4305 /* no suitable interface, frame not sent */
4308 /* frame sent to all suitable interfaces */
4309 read_unlock(&bond->lock);
4310 return NETDEV_TX_OK;
4313 /*------------------------- Device initialization ---------------------------*/
4315 static void bond_set_xmit_hash_policy(struct bonding *bond)
4317 switch (bond->params.xmit_policy) {
4318 case BOND_XMIT_POLICY_LAYER23:
4319 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4321 case BOND_XMIT_POLICY_LAYER34:
4322 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4324 case BOND_XMIT_POLICY_LAYER2:
4326 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4331 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4333 const struct bonding *bond = netdev_priv(dev);
4335 switch (bond->params.mode) {
4336 case BOND_MODE_ROUNDROBIN:
4337 return bond_xmit_roundrobin(skb, dev);
4338 case BOND_MODE_ACTIVEBACKUP:
4339 return bond_xmit_activebackup(skb, dev);
4341 return bond_xmit_xor(skb, dev);
4342 case BOND_MODE_BROADCAST:
4343 return bond_xmit_broadcast(skb, dev);
4344 case BOND_MODE_8023AD:
4345 return bond_3ad_xmit_xor(skb, dev);
4348 return bond_alb_xmit(skb, dev);
4350 /* Should never happen, mode already checked */
4351 pr_err("%s: Error: Unknown bonding mode %d\n",
4352 dev->name, bond->params.mode);
4355 return NETDEV_TX_OK;
4361 * set bond mode specific net device operations
4363 void bond_set_mode_ops(struct bonding *bond, int mode)
4365 struct net_device *bond_dev = bond->dev;
4368 case BOND_MODE_ROUNDROBIN:
4370 case BOND_MODE_ACTIVEBACKUP:
4373 bond_set_xmit_hash_policy(bond);
4375 case BOND_MODE_BROADCAST:
4377 case BOND_MODE_8023AD:
4378 bond_set_master_3ad_flags(bond);
4379 bond_set_xmit_hash_policy(bond);
4382 bond_set_master_alb_flags(bond);
4387 /* Should never happen, mode already checked */
4388 pr_err("%s: Error: Unknown bonding mode %d\n",
4389 bond_dev->name, mode);
4394 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4395 struct ethtool_drvinfo *drvinfo)
4397 strncpy(drvinfo->driver, DRV_NAME, 32);
4398 strncpy(drvinfo->version, DRV_VERSION, 32);
4399 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4402 static const struct ethtool_ops bond_ethtool_ops = {
4403 .get_drvinfo = bond_ethtool_get_drvinfo,
4404 .get_link = ethtool_op_get_link,
4405 .get_tx_csum = ethtool_op_get_tx_csum,
4406 .get_sg = ethtool_op_get_sg,
4407 .get_tso = ethtool_op_get_tso,
4408 .get_ufo = ethtool_op_get_ufo,
4409 .get_flags = ethtool_op_get_flags,
4412 static const struct net_device_ops bond_netdev_ops = {
4413 .ndo_init = bond_init,
4414 .ndo_uninit = bond_uninit,
4415 .ndo_open = bond_open,
4416 .ndo_stop = bond_close,
4417 .ndo_start_xmit = bond_start_xmit,
4418 .ndo_get_stats = bond_get_stats,
4419 .ndo_do_ioctl = bond_do_ioctl,
4420 .ndo_set_multicast_list = bond_set_multicast_list,
4421 .ndo_change_mtu = bond_change_mtu,
4422 .ndo_set_mac_address = bond_set_mac_address,
4423 .ndo_neigh_setup = bond_neigh_setup,
4424 .ndo_vlan_rx_register = bond_vlan_rx_register,
4425 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4426 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4429 static void bond_setup(struct net_device *bond_dev)
4431 struct bonding *bond = netdev_priv(bond_dev);
4433 /* initialize rwlocks */
4434 rwlock_init(&bond->lock);
4435 rwlock_init(&bond->curr_slave_lock);
4437 bond->params = bonding_defaults;
4439 /* Initialize pointers */
4440 bond->dev = bond_dev;
4441 INIT_LIST_HEAD(&bond->vlan_list);
4443 /* Initialize the device entry points */
4444 ether_setup(bond_dev);
4445 bond_dev->netdev_ops = &bond_netdev_ops;
4446 bond_dev->ethtool_ops = &bond_ethtool_ops;
4447 bond_set_mode_ops(bond, bond->params.mode);
4449 bond_dev->destructor = free_netdev;
4451 /* Initialize the device options */
4452 bond_dev->tx_queue_len = 0;
4453 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4454 bond_dev->priv_flags |= IFF_BONDING;
4455 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4457 if (bond->params.arp_interval)
4458 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4460 /* At first, we block adding VLANs. That's the only way to
4461 * prevent problems that occur when adding VLANs over an
4462 * empty bond. The block will be removed once non-challenged
4463 * slaves are enslaved.
4465 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4467 /* don't acquire bond device's netif_tx_lock when
4469 bond_dev->features |= NETIF_F_LLTX;
4471 /* By default, we declare the bond to be fully
4472 * VLAN hardware accelerated capable. Special
4473 * care is taken in the various xmit functions
4474 * when there are slaves that are not hw accel
4477 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4478 NETIF_F_HW_VLAN_RX |
4479 NETIF_F_HW_VLAN_FILTER);
4483 static void bond_work_cancel_all(struct bonding *bond)
4485 write_lock_bh(&bond->lock);
4486 bond->kill_timers = 1;
4487 write_unlock_bh(&bond->lock);
4489 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4490 cancel_delayed_work(&bond->mii_work);
4492 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4493 cancel_delayed_work(&bond->arp_work);
4495 if (bond->params.mode == BOND_MODE_ALB &&
4496 delayed_work_pending(&bond->alb_work))
4497 cancel_delayed_work(&bond->alb_work);
4499 if (bond->params.mode == BOND_MODE_8023AD &&
4500 delayed_work_pending(&bond->ad_work))
4501 cancel_delayed_work(&bond->ad_work);
4505 * Destroy a bonding device.
4506 * Must be under rtnl_lock when this function is called.
4508 static void bond_uninit(struct net_device *bond_dev)
4510 struct bonding *bond = netdev_priv(bond_dev);
4512 /* Release the bonded slaves */
4513 bond_release_all(bond_dev);
4515 list_del(&bond->bond_list);
4517 bond_work_cancel_all(bond);
4519 bond_remove_proc_entry(bond);
4522 destroy_workqueue(bond->wq);
4524 netif_addr_lock_bh(bond_dev);
4525 bond_mc_list_destroy(bond);
4526 netif_addr_unlock_bh(bond_dev);
4529 /*------------------------- Module initialization ---------------------------*/
4532 * Convert string input module parms. Accept either the
4533 * number of the mode or its string name. A bit complicated because
4534 * some mode names are substrings of other names, and calls from sysfs
4535 * may have whitespace in the name (trailing newlines, for example).
4537 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4539 int modeint = -1, i, rv;
4540 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4542 for (p = (char *)buf; *p; p++)
4543 if (!(isdigit(*p) || isspace(*p)))
4547 rv = sscanf(buf, "%20s", modestr);
4549 rv = sscanf(buf, "%d", &modeint);
4554 for (i = 0; tbl[i].modename; i++) {
4555 if (modeint == tbl[i].mode)
4557 if (strcmp(modestr, tbl[i].modename) == 0)
4564 static int bond_check_params(struct bond_params *params)
4566 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4569 * Convert string parameters.
4572 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4573 if (bond_mode == -1) {
4574 pr_err("Error: Invalid bonding mode \"%s\"\n",
4575 mode == NULL ? "NULL" : mode);
4580 if (xmit_hash_policy) {
4581 if ((bond_mode != BOND_MODE_XOR) &&
4582 (bond_mode != BOND_MODE_8023AD)) {
4583 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4584 bond_mode_name(bond_mode));
4586 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4588 if (xmit_hashtype == -1) {
4589 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4590 xmit_hash_policy == NULL ? "NULL" :
4598 if (bond_mode != BOND_MODE_8023AD) {
4599 pr_info("lacp_rate param is irrelevant in mode %s\n",
4600 bond_mode_name(bond_mode));
4602 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4603 if (lacp_fast == -1) {
4604 pr_err("Error: Invalid lacp rate \"%s\"\n",
4605 lacp_rate == NULL ? "NULL" : lacp_rate);
4612 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4613 if (params->ad_select == -1) {
4614 pr_err("Error: Invalid ad_select \"%s\"\n",
4615 ad_select == NULL ? "NULL" : ad_select);
4619 if (bond_mode != BOND_MODE_8023AD) {
4620 pr_warning("ad_select param only affects 802.3ad mode\n");
4623 params->ad_select = BOND_AD_STABLE;
4626 if (max_bonds < 0) {
4627 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4628 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4629 max_bonds = BOND_DEFAULT_MAX_BONDS;
4633 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4634 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4635 miimon = BOND_LINK_MON_INTERV;
4639 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4644 if (downdelay < 0) {
4645 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4646 downdelay, INT_MAX);
4650 if ((use_carrier != 0) && (use_carrier != 1)) {
4651 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4656 if (num_grat_arp < 0 || num_grat_arp > 255) {
4657 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1 \n",
4662 if (num_unsol_na < 0 || num_unsol_na > 255) {
4663 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1 \n",
4668 /* reset values for 802.3ad */
4669 if (bond_mode == BOND_MODE_8023AD) {
4671 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4672 pr_warning("Forcing miimon to 100msec\n");
4677 /* reset values for TLB/ALB */
4678 if ((bond_mode == BOND_MODE_TLB) ||
4679 (bond_mode == BOND_MODE_ALB)) {
4681 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4682 pr_warning("Forcing miimon to 100msec\n");
4687 if (bond_mode == BOND_MODE_ALB) {
4688 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4693 if (updelay || downdelay) {
4694 /* just warn the user the up/down delay will have
4695 * no effect since miimon is zero...
4697 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4698 updelay, downdelay);
4701 /* don't allow arp monitoring */
4703 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4704 miimon, arp_interval);
4708 if ((updelay % miimon) != 0) {
4709 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4711 (updelay / miimon) * miimon);
4716 if ((downdelay % miimon) != 0) {
4717 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4719 (downdelay / miimon) * miimon);
4722 downdelay /= miimon;
4725 if (arp_interval < 0) {
4726 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4727 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4728 arp_interval = BOND_LINK_ARP_INTERV;
4731 for (arp_ip_count = 0;
4732 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4734 /* not complete check, but should be good enough to
4736 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4737 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4738 arp_ip_target[arp_ip_count]);
4741 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4742 arp_target[arp_ip_count] = ip;
4746 if (arp_interval && !arp_ip_count) {
4747 /* don't allow arping if no arp_ip_target given... */
4748 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4754 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4755 pr_err("arp_validate only supported in active-backup mode\n");
4758 if (!arp_interval) {
4759 pr_err("arp_validate requires arp_interval\n");
4763 arp_validate_value = bond_parse_parm(arp_validate,
4765 if (arp_validate_value == -1) {
4766 pr_err("Error: invalid arp_validate \"%s\"\n",
4767 arp_validate == NULL ? "NULL" : arp_validate);
4771 arp_validate_value = 0;
4774 pr_info("MII link monitoring set to %d ms\n", miimon);
4775 } else if (arp_interval) {
4778 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4780 arp_validate_tbl[arp_validate_value].modename,
4783 for (i = 0; i < arp_ip_count; i++)
4784 pr_info(" %s", arp_ip_target[i]);
4788 } else if (max_bonds) {
4789 /* miimon and arp_interval not set, we need one so things
4790 * work as expected, see bonding.txt for details
4792 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4795 if (primary && !USES_PRIMARY(bond_mode)) {
4796 /* currently, using a primary only makes sense
4797 * in active backup, TLB or ALB modes
4799 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4800 primary, bond_mode_name(bond_mode));
4804 if (primary && primary_reselect) {
4805 primary_reselect_value = bond_parse_parm(primary_reselect,
4807 if (primary_reselect_value == -1) {
4808 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4810 NULL ? "NULL" : primary_reselect);
4814 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4817 if (fail_over_mac) {
4818 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4820 if (fail_over_mac_value == -1) {
4821 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4822 arp_validate == NULL ? "NULL" : arp_validate);
4826 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4827 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4829 fail_over_mac_value = BOND_FOM_NONE;
4832 /* fill params struct with the proper values */
4833 params->mode = bond_mode;
4834 params->xmit_policy = xmit_hashtype;
4835 params->miimon = miimon;
4836 params->num_grat_arp = num_grat_arp;
4837 params->num_unsol_na = num_unsol_na;
4838 params->arp_interval = arp_interval;
4839 params->arp_validate = arp_validate_value;
4840 params->updelay = updelay;
4841 params->downdelay = downdelay;
4842 params->use_carrier = use_carrier;
4843 params->lacp_fast = lacp_fast;
4844 params->primary[0] = 0;
4845 params->primary_reselect = primary_reselect_value;
4846 params->fail_over_mac = fail_over_mac_value;
4849 strncpy(params->primary, primary, IFNAMSIZ);
4850 params->primary[IFNAMSIZ - 1] = 0;
4853 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4858 static struct lock_class_key bonding_netdev_xmit_lock_key;
4859 static struct lock_class_key bonding_netdev_addr_lock_key;
4861 static void bond_set_lockdep_class_one(struct net_device *dev,
4862 struct netdev_queue *txq,
4865 lockdep_set_class(&txq->_xmit_lock,
4866 &bonding_netdev_xmit_lock_key);
4869 static void bond_set_lockdep_class(struct net_device *dev)
4871 lockdep_set_class(&dev->addr_list_lock,
4872 &bonding_netdev_addr_lock_key);
4873 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4877 * Called from registration process
4879 static int bond_init(struct net_device *bond_dev)
4881 struct bonding *bond = netdev_priv(bond_dev);
4882 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4884 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4886 bond->wq = create_singlethread_workqueue(bond_dev->name);
4890 bond_set_lockdep_class(bond_dev);
4892 netif_carrier_off(bond_dev);
4894 bond_create_proc_entry(bond);
4895 list_add_tail(&bond->bond_list, &bn->dev_list);
4897 bond_prepare_sysfs_group(bond);
4901 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4903 if (tb[IFLA_ADDRESS]) {
4904 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4906 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4907 return -EADDRNOTAVAIL;
4912 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4914 .priv_size = sizeof(struct bonding),
4915 .setup = bond_setup,
4916 .validate = bond_validate,
4919 /* Create a new bond based on the specified name and bonding parameters.
4920 * If name is NULL, obtain a suitable "bond%d" name for us.
4921 * Caller must NOT hold rtnl_lock; we need to release it here before we
4922 * set up our sysfs entries.
4924 int bond_create(struct net *net, const char *name)
4926 struct net_device *bond_dev;
4931 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4934 pr_err("%s: eek! can't alloc netdev!\n", name);
4939 dev_net_set(bond_dev, net);
4940 bond_dev->rtnl_link_ops = &bond_link_ops;
4943 res = dev_alloc_name(bond_dev, "bond%d");
4948 res = register_netdevice(bond_dev);
4956 free_netdev(bond_dev);
4960 static int __net_init bond_net_init(struct net *net)
4962 struct bond_net *bn = net_generic(net, bond_net_id);
4965 INIT_LIST_HEAD(&bn->dev_list);
4967 bond_create_proc_dir(bn);
4972 static void __net_exit bond_net_exit(struct net *net)
4974 struct bond_net *bn = net_generic(net, bond_net_id);
4976 bond_destroy_proc_dir(bn);
4979 static struct pernet_operations bond_net_ops = {
4980 .init = bond_net_init,
4981 .exit = bond_net_exit,
4983 .size = sizeof(struct bond_net),
4986 static int __init bonding_init(void)
4991 pr_info("%s", version);
4993 res = bond_check_params(&bonding_defaults);
4997 res = register_pernet_subsys(&bond_net_ops);
5001 res = rtnl_link_register(&bond_link_ops);
5005 for (i = 0; i < max_bonds; i++) {
5006 res = bond_create(&init_net, NULL);
5011 res = bond_create_sysfs();
5015 register_netdevice_notifier(&bond_netdev_notifier);
5016 register_inetaddr_notifier(&bond_inetaddr_notifier);
5017 bond_register_ipv6_notifier();
5021 rtnl_link_unregister(&bond_link_ops);
5023 unregister_pernet_subsys(&bond_net_ops);
5028 static void __exit bonding_exit(void)
5030 unregister_netdevice_notifier(&bond_netdev_notifier);
5031 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5032 bond_unregister_ipv6_notifier();
5034 bond_destroy_sysfs();
5036 rtnl_link_unregister(&bond_link_ops);
5037 unregister_pernet_subsys(&bond_net_ops);
5040 module_init(bonding_init);
5041 module_exit(bonding_exit);
5042 MODULE_LICENSE("GPL");
5043 MODULE_VERSION(DRV_VERSION);
5044 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5045 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5046 MODULE_ALIAS_RTNL_LINK("bond");