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/netpoll.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <linux/jiffies.h>
79 #include <linux/preempt.h>
80 #include <net/route.h>
81 #include <net/net_namespace.h>
82 #include <net/netns/generic.h>
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 #define BOND_LINK_MON_INTERV 0
91 #define BOND_LINK_ARP_INTERV 0
93 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
94 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
95 static int num_grat_arp = 1;
96 static int num_unsol_na = 1;
97 static int miimon = BOND_LINK_MON_INTERV;
100 static int use_carrier = 1;
102 static char *primary;
103 static char *primary_reselect;
104 static char *lacp_rate;
105 static char *ad_select;
106 static char *xmit_hash_policy;
107 static int arp_interval = BOND_LINK_ARP_INTERV;
108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
109 static char *arp_validate;
110 static char *fail_over_mac;
111 static int all_slaves_active = 0;
112 static struct bond_params bonding_defaults;
113 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param(num_grat_arp, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
121 module_param(num_unsol_na, int, 0644);
122 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
123 module_param(miimon, int, 0);
124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
125 module_param(updelay, int, 0);
126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
127 module_param(downdelay, int, 0);
128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
130 module_param(use_carrier, int, 0);
131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132 "0 for off, 1 for on (default)");
133 module_param(mode, charp, 0);
134 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
135 "1 for active-backup, 2 for balance-xor, "
136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137 "6 for balance-alb");
138 module_param(primary, charp, 0);
139 MODULE_PARM_DESC(primary, "Primary network device to use");
140 module_param(primary_reselect, charp, 0);
141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
143 "0 for always (default), "
144 "1 for only if speed of primary is "
146 "2 for only on active slave "
148 module_param(lacp_rate, charp, 0);
149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
151 module_param(ad_select, charp, 0);
152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
153 module_param(xmit_hash_policy, charp, 0);
154 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
155 ", 1 for layer 3+4");
156 module_param(arp_interval, int, 0);
157 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
158 module_param_array(arp_ip_target, charp, NULL, 0);
159 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
160 module_param(arp_validate, charp, 0);
161 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
162 module_param(fail_over_mac, charp, 0);
163 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
164 module_param(all_slaves_active, int, 0);
165 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
166 "by setting active flag for all slaves. "
167 "0 for never (default), 1 for always.");
168 module_param(resend_igmp, int, 0);
169 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
171 /*----------------------------- Global variables ----------------------------*/
173 #ifdef CONFIG_NET_POLL_CONTROLLER
174 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
177 static const char * const version =
178 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
180 int bond_net_id __read_mostly;
182 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
183 static int arp_ip_count;
184 static int bond_mode = BOND_MODE_ROUNDROBIN;
185 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
186 static int lacp_fast;
188 const struct bond_parm_tbl bond_lacp_tbl[] = {
189 { "slow", AD_LACP_SLOW},
190 { "fast", AD_LACP_FAST},
194 const struct bond_parm_tbl bond_mode_tbl[] = {
195 { "balance-rr", BOND_MODE_ROUNDROBIN},
196 { "active-backup", BOND_MODE_ACTIVEBACKUP},
197 { "balance-xor", BOND_MODE_XOR},
198 { "broadcast", BOND_MODE_BROADCAST},
199 { "802.3ad", BOND_MODE_8023AD},
200 { "balance-tlb", BOND_MODE_TLB},
201 { "balance-alb", BOND_MODE_ALB},
205 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
206 { "layer2", BOND_XMIT_POLICY_LAYER2},
207 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
208 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
212 const struct bond_parm_tbl arp_validate_tbl[] = {
213 { "none", BOND_ARP_VALIDATE_NONE},
214 { "active", BOND_ARP_VALIDATE_ACTIVE},
215 { "backup", BOND_ARP_VALIDATE_BACKUP},
216 { "all", BOND_ARP_VALIDATE_ALL},
220 const struct bond_parm_tbl fail_over_mac_tbl[] = {
221 { "none", BOND_FOM_NONE},
222 { "active", BOND_FOM_ACTIVE},
223 { "follow", BOND_FOM_FOLLOW},
227 const struct bond_parm_tbl pri_reselect_tbl[] = {
228 { "always", BOND_PRI_RESELECT_ALWAYS},
229 { "better", BOND_PRI_RESELECT_BETTER},
230 { "failure", BOND_PRI_RESELECT_FAILURE},
234 struct bond_parm_tbl ad_select_tbl[] = {
235 { "stable", BOND_AD_STABLE},
236 { "bandwidth", BOND_AD_BANDWIDTH},
237 { "count", BOND_AD_COUNT},
241 /*-------------------------- Forward declarations ---------------------------*/
243 static void bond_send_gratuitous_arp(struct bonding *bond);
244 static int bond_init(struct net_device *bond_dev);
245 static void bond_uninit(struct net_device *bond_dev);
247 /*---------------------------- General routines -----------------------------*/
249 static const char *bond_mode_name(int mode)
251 static const char *names[] = {
252 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
253 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
254 [BOND_MODE_XOR] = "load balancing (xor)",
255 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
256 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
257 [BOND_MODE_TLB] = "transmit load balancing",
258 [BOND_MODE_ALB] = "adaptive load balancing",
261 if (mode < 0 || mode > BOND_MODE_ALB)
267 /*---------------------------------- VLAN -----------------------------------*/
270 * bond_add_vlan - add a new vlan id on bond
271 * @bond: bond that got the notification
272 * @vlan_id: the vlan id to add
274 * Returns -ENOMEM if allocation failed.
276 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
278 struct vlan_entry *vlan;
280 pr_debug("bond: %s, vlan id %d\n",
281 (bond ? bond->dev->name : "None"), vlan_id);
283 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
287 INIT_LIST_HEAD(&vlan->vlan_list);
288 vlan->vlan_id = vlan_id;
290 write_lock_bh(&bond->lock);
292 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
294 write_unlock_bh(&bond->lock);
296 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
302 * bond_del_vlan - delete a vlan id from bond
303 * @bond: bond that got the notification
304 * @vlan_id: the vlan id to delete
306 * returns -ENODEV if @vlan_id was not found in @bond.
308 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
310 struct vlan_entry *vlan;
313 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
316 write_lock_bh(&bond->lock);
318 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
319 if (vlan->vlan_id == vlan_id) {
320 list_del(&vlan->vlan_list);
322 if (bond_is_lb(bond))
323 bond_alb_clear_vlan(bond, vlan_id);
325 pr_debug("removed VLAN ID %d from bond %s\n",
326 vlan_id, bond->dev->name);
330 if (list_empty(&bond->vlan_list) &&
331 (bond->slave_cnt == 0)) {
332 /* Last VLAN removed and no slaves, so
333 * restore block on adding VLANs. This will
334 * be removed once new slaves that are not
335 * VLAN challenged will be added.
337 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
345 pr_debug("couldn't find VLAN ID %d in bond %s\n",
346 vlan_id, bond->dev->name);
349 write_unlock_bh(&bond->lock);
350 unblock_netpoll_tx();
355 * bond_has_challenged_slaves
356 * @bond: the bond we're working on
358 * Searches the slave list. Returns 1 if a vlan challenged slave
359 * was found, 0 otherwise.
361 * Assumes bond->lock is held.
363 static int bond_has_challenged_slaves(struct bonding *bond)
368 bond_for_each_slave(bond, slave, i) {
369 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
370 pr_debug("found VLAN challenged slave - %s\n",
376 pr_debug("no VLAN challenged slaves found\n");
381 * bond_next_vlan - safely skip to the next item in the vlans list.
382 * @bond: the bond we're working on
383 * @curr: item we're advancing from
385 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
386 * or @curr->next otherwise (even if it is @curr itself again).
388 * Caller must hold bond->lock
390 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
392 struct vlan_entry *next, *last;
394 if (list_empty(&bond->vlan_list))
398 next = list_entry(bond->vlan_list.next,
399 struct vlan_entry, vlan_list);
401 last = list_entry(bond->vlan_list.prev,
402 struct vlan_entry, vlan_list);
404 next = list_entry(bond->vlan_list.next,
405 struct vlan_entry, vlan_list);
407 next = list_entry(curr->vlan_list.next,
408 struct vlan_entry, vlan_list);
416 * bond_dev_queue_xmit - Prepare skb for xmit.
418 * @bond: bond device that got this skb for tx.
419 * @skb: hw accel VLAN tagged skb to transmit
420 * @slave_dev: slave that is supposed to xmit this skbuff
422 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
423 struct net_device *slave_dev)
425 skb->dev = slave_dev;
427 #ifdef CONFIG_NET_POLL_CONTROLLER
428 if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
429 struct netpoll *np = bond->dev->npinfo->netpoll;
430 slave_dev->npinfo = bond->dev->npinfo;
431 slave_dev->priv_flags |= IFF_IN_NETPOLL;
432 netpoll_send_skb_on_dev(np, skb, slave_dev);
433 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
442 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
443 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
445 * a. This operation is performed in IOCTL context,
446 * b. The operation is protected by the RTNL semaphore in the 8021q code,
447 * c. Holding a lock with BH disabled while directly calling a base driver
448 * entry point is generally a BAD idea.
450 * The design of synchronization/protection for this operation in the 8021q
451 * module is good for one or more VLAN devices over a single physical device
452 * and cannot be extended for a teaming solution like bonding, so there is a
453 * potential race condition here where a net device from the vlan group might
454 * be referenced (either by a base driver or the 8021q code) while it is being
455 * removed from the system. However, it turns out we're not making matters
456 * worse, and if it works for regular VLAN usage it will work here too.
460 * bond_vlan_rx_register - Propagates registration to slaves
461 * @bond_dev: bonding net device that got called
462 * @grp: vlan group being registered
464 static void bond_vlan_rx_register(struct net_device *bond_dev,
465 struct vlan_group *grp)
467 struct bonding *bond = netdev_priv(bond_dev);
471 write_lock_bh(&bond->lock);
473 write_unlock_bh(&bond->lock);
475 bond_for_each_slave(bond, slave, i) {
476 struct net_device *slave_dev = slave->dev;
477 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
479 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
480 slave_ops->ndo_vlan_rx_register) {
481 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
487 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
488 * @bond_dev: bonding net device that got called
489 * @vid: vlan id being added
491 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
493 struct bonding *bond = netdev_priv(bond_dev);
497 bond_for_each_slave(bond, slave, i) {
498 struct net_device *slave_dev = slave->dev;
499 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
501 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
502 slave_ops->ndo_vlan_rx_add_vid) {
503 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
507 res = bond_add_vlan(bond, vid);
509 pr_err("%s: Error: Failed to add vlan id %d\n",
510 bond_dev->name, vid);
515 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
516 * @bond_dev: bonding net device that got called
517 * @vid: vlan id being removed
519 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
521 struct bonding *bond = netdev_priv(bond_dev);
523 struct net_device *vlan_dev;
526 bond_for_each_slave(bond, slave, i) {
527 struct net_device *slave_dev = slave->dev;
528 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
530 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
531 slave_ops->ndo_vlan_rx_kill_vid) {
532 /* Save and then restore vlan_dev in the grp array,
533 * since the slave's driver might clear it.
535 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
536 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
537 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
541 res = bond_del_vlan(bond, vid);
543 pr_err("%s: Error: Failed to remove vlan id %d\n",
544 bond_dev->name, vid);
548 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
550 struct vlan_entry *vlan;
551 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
556 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
557 slave_ops->ndo_vlan_rx_register)
558 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
560 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
561 !(slave_ops->ndo_vlan_rx_add_vid))
564 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
565 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
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;
578 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
579 !(slave_ops->ndo_vlan_rx_kill_vid))
582 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 /*------------------------------- Link status -------------------------------*/
602 * Set the carrier state for the master according to the state of its
603 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
604 * do special 802.3ad magic.
606 * Returns zero if carrier state does not change, nonzero if it does.
608 static int bond_set_carrier(struct bonding *bond)
613 if (bond->slave_cnt == 0)
616 if (bond->params.mode == BOND_MODE_8023AD)
617 return bond_3ad_set_carrier(bond);
619 bond_for_each_slave(bond, slave, i) {
620 if (slave->link == BOND_LINK_UP) {
621 if (!netif_carrier_ok(bond->dev)) {
622 netif_carrier_on(bond->dev);
630 if (netif_carrier_ok(bond->dev)) {
631 netif_carrier_off(bond->dev);
638 * Get link speed and duplex from the slave's base driver
639 * using ethtool. If for some reason the call fails or the
640 * values are invalid, fake speed and duplex to 100/Full
643 static int bond_update_speed_duplex(struct slave *slave)
645 struct net_device *slave_dev = slave->dev;
646 struct ethtool_cmd etool;
649 /* Fake speed and duplex */
650 slave->speed = SPEED_100;
651 slave->duplex = DUPLEX_FULL;
653 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
656 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
660 switch (etool.speed) {
670 switch (etool.duplex) {
678 slave->speed = etool.speed;
679 slave->duplex = etool.duplex;
685 * if <dev> supports MII link status reporting, check its link status.
687 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
688 * depending upon the setting of the use_carrier parameter.
690 * Return either BMSR_LSTATUS, meaning that the link is up (or we
691 * can't tell and just pretend it is), or 0, meaning that the link is
694 * If reporting is non-zero, instead of faking link up, return -1 if
695 * both ETHTOOL and MII ioctls fail (meaning the device does not
696 * support them). If use_carrier is set, return whatever it says.
697 * It'd be nice if there was a good way to tell if a driver supports
698 * netif_carrier, but there really isn't.
700 static int bond_check_dev_link(struct bonding *bond,
701 struct net_device *slave_dev, int reporting)
703 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
704 int (*ioctl)(struct net_device *, struct ifreq *, int);
706 struct mii_ioctl_data *mii;
708 if (!reporting && !netif_running(slave_dev))
711 if (bond->params.use_carrier)
712 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
714 /* Try to get link status using Ethtool first. */
715 if (slave_dev->ethtool_ops) {
716 if (slave_dev->ethtool_ops->get_link) {
719 link = slave_dev->ethtool_ops->get_link(slave_dev);
721 return link ? BMSR_LSTATUS : 0;
725 /* Ethtool can't be used, fallback to MII ioctls. */
726 ioctl = slave_ops->ndo_do_ioctl;
728 /* TODO: set pointer to correct ioctl on a per team member */
729 /* bases to make this more efficient. that is, once */
730 /* we determine the correct ioctl, we will always */
731 /* call it and not the others for that team */
735 * We cannot assume that SIOCGMIIPHY will also read a
736 * register; not all network drivers (e.g., e100)
740 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
741 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
743 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
744 mii->reg_num = MII_BMSR;
745 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
746 return mii->val_out & BMSR_LSTATUS;
751 * If reporting, report that either there's no dev->do_ioctl,
752 * or both SIOCGMIIREG and get_link failed (meaning that we
753 * cannot report link status). If not reporting, pretend
756 return reporting ? -1 : BMSR_LSTATUS;
759 /*----------------------------- Multicast list ------------------------------*/
762 * Push the promiscuity flag down to appropriate slaves
764 static int bond_set_promiscuity(struct bonding *bond, int inc)
767 if (USES_PRIMARY(bond->params.mode)) {
768 /* write lock already acquired */
769 if (bond->curr_active_slave) {
770 err = dev_set_promiscuity(bond->curr_active_slave->dev,
776 bond_for_each_slave(bond, slave, i) {
777 err = dev_set_promiscuity(slave->dev, inc);
786 * Push the allmulti flag down to all slaves
788 static int bond_set_allmulti(struct bonding *bond, int inc)
791 if (USES_PRIMARY(bond->params.mode)) {
792 /* write lock already acquired */
793 if (bond->curr_active_slave) {
794 err = dev_set_allmulti(bond->curr_active_slave->dev,
800 bond_for_each_slave(bond, slave, i) {
801 err = dev_set_allmulti(slave->dev, inc);
810 * Add a Multicast address to slaves
813 static void bond_mc_add(struct bonding *bond, void *addr)
815 if (USES_PRIMARY(bond->params.mode)) {
816 /* write lock already acquired */
817 if (bond->curr_active_slave)
818 dev_mc_add(bond->curr_active_slave->dev, addr);
823 bond_for_each_slave(bond, slave, i)
824 dev_mc_add(slave->dev, addr);
829 * Remove a multicast address from slave
832 static void bond_mc_del(struct bonding *bond, void *addr)
834 if (USES_PRIMARY(bond->params.mode)) {
835 /* write lock already acquired */
836 if (bond->curr_active_slave)
837 dev_mc_del(bond->curr_active_slave->dev, addr);
841 bond_for_each_slave(bond, slave, i) {
842 dev_mc_del(slave->dev, addr);
848 static void __bond_resend_igmp_join_requests(struct net_device *dev)
850 struct in_device *in_dev;
851 struct ip_mc_list *im;
854 in_dev = __in_dev_get_rcu(dev);
856 read_lock(&in_dev->mc_list_lock);
857 for (im = in_dev->mc_list; im; im = im->next)
858 ip_mc_rejoin_group(im);
859 read_unlock(&in_dev->mc_list_lock);
866 * Retrieve the list of registered multicast addresses for the bonding
867 * device and retransmit an IGMP JOIN request to the current active
870 static void bond_resend_igmp_join_requests(struct bonding *bond)
872 struct net_device *vlan_dev;
873 struct vlan_entry *vlan;
875 read_lock(&bond->lock);
877 /* rejoin all groups on bond device */
878 __bond_resend_igmp_join_requests(bond->dev);
880 /* rejoin all groups on vlan devices */
882 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
883 vlan_dev = vlan_group_get_device(bond->vlgrp,
886 __bond_resend_igmp_join_requests(vlan_dev);
890 if (--bond->igmp_retrans > 0)
891 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
893 read_unlock(&bond->lock);
896 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
898 struct bonding *bond = container_of(work, struct bonding,
900 bond_resend_igmp_join_requests(bond);
904 * flush all members of flush->mc_list from device dev->mc_list
906 static void bond_mc_list_flush(struct net_device *bond_dev,
907 struct net_device *slave_dev)
909 struct bonding *bond = netdev_priv(bond_dev);
910 struct netdev_hw_addr *ha;
912 netdev_for_each_mc_addr(ha, bond_dev)
913 dev_mc_del(slave_dev, ha->addr);
915 if (bond->params.mode == BOND_MODE_8023AD) {
916 /* del lacpdu mc addr from mc list */
917 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
919 dev_mc_del(slave_dev, lacpdu_multicast);
923 /*--------------------------- Active slave change ---------------------------*/
926 * Update the mc list and multicast-related flags for the new and
927 * old active slaves (if any) according to the multicast mode, and
928 * promiscuous flags unconditionally.
930 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
931 struct slave *old_active)
933 struct netdev_hw_addr *ha;
935 if (!USES_PRIMARY(bond->params.mode))
936 /* nothing to do - mc list is already up-to-date on
942 if (bond->dev->flags & IFF_PROMISC)
943 dev_set_promiscuity(old_active->dev, -1);
945 if (bond->dev->flags & IFF_ALLMULTI)
946 dev_set_allmulti(old_active->dev, -1);
948 netdev_for_each_mc_addr(ha, bond->dev)
949 dev_mc_del(old_active->dev, ha->addr);
953 /* FIXME: Signal errors upstream. */
954 if (bond->dev->flags & IFF_PROMISC)
955 dev_set_promiscuity(new_active->dev, 1);
957 if (bond->dev->flags & IFF_ALLMULTI)
958 dev_set_allmulti(new_active->dev, 1);
960 netdev_for_each_mc_addr(ha, bond->dev)
961 dev_mc_add(new_active->dev, ha->addr);
966 * bond_do_fail_over_mac
968 * Perform special MAC address swapping for fail_over_mac settings
970 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
972 static void bond_do_fail_over_mac(struct bonding *bond,
973 struct slave *new_active,
974 struct slave *old_active)
975 __releases(&bond->curr_slave_lock)
976 __releases(&bond->lock)
977 __acquires(&bond->lock)
978 __acquires(&bond->curr_slave_lock)
980 u8 tmp_mac[ETH_ALEN];
981 struct sockaddr saddr;
984 switch (bond->params.fail_over_mac) {
985 case BOND_FOM_ACTIVE:
987 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
988 new_active->dev->addr_len);
990 case BOND_FOM_FOLLOW:
992 * if new_active && old_active, swap them
993 * if just old_active, do nothing (going to no active slave)
994 * if just new_active, set new_active to bond's MAC
999 write_unlock_bh(&bond->curr_slave_lock);
1000 read_unlock(&bond->lock);
1003 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1004 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1006 saddr.sa_family = new_active->dev->type;
1008 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1009 saddr.sa_family = bond->dev->type;
1012 rv = dev_set_mac_address(new_active->dev, &saddr);
1014 pr_err("%s: Error %d setting MAC of slave %s\n",
1015 bond->dev->name, -rv, new_active->dev->name);
1022 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1023 saddr.sa_family = old_active->dev->type;
1025 rv = dev_set_mac_address(old_active->dev, &saddr);
1027 pr_err("%s: Error %d setting MAC of slave %s\n",
1028 bond->dev->name, -rv, new_active->dev->name);
1030 read_lock(&bond->lock);
1031 write_lock_bh(&bond->curr_slave_lock);
1034 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1035 bond->dev->name, bond->params.fail_over_mac);
1041 static bool bond_should_change_active(struct bonding *bond)
1043 struct slave *prim = bond->primary_slave;
1044 struct slave *curr = bond->curr_active_slave;
1046 if (!prim || !curr || curr->link != BOND_LINK_UP)
1048 if (bond->force_primary) {
1049 bond->force_primary = false;
1052 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1053 (prim->speed < curr->speed ||
1054 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1056 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1062 * find_best_interface - select the best available slave to be the active one
1063 * @bond: our bonding struct
1065 * Warning: Caller must hold curr_slave_lock for writing.
1067 static struct slave *bond_find_best_slave(struct bonding *bond)
1069 struct slave *new_active, *old_active;
1070 struct slave *bestslave = NULL;
1071 int mintime = bond->params.updelay;
1074 new_active = bond->curr_active_slave;
1076 if (!new_active) { /* there were no active slaves left */
1077 if (bond->slave_cnt > 0) /* found one slave */
1078 new_active = bond->first_slave;
1080 return NULL; /* still no slave, return NULL */
1083 if ((bond->primary_slave) &&
1084 bond->primary_slave->link == BOND_LINK_UP &&
1085 bond_should_change_active(bond)) {
1086 new_active = bond->primary_slave;
1089 /* remember where to stop iterating over the slaves */
1090 old_active = new_active;
1092 bond_for_each_slave_from(bond, new_active, i, old_active) {
1093 if (new_active->link == BOND_LINK_UP) {
1095 } else if (new_active->link == BOND_LINK_BACK &&
1096 IS_UP(new_active->dev)) {
1097 /* link up, but waiting for stabilization */
1098 if (new_active->delay < mintime) {
1099 mintime = new_active->delay;
1100 bestslave = new_active;
1109 * change_active_interface - change the active slave into the specified one
1110 * @bond: our bonding struct
1111 * @new: the new slave to make the active one
1113 * Set the new slave to the bond's settings and unset them on the old
1114 * curr_active_slave.
1115 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1117 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1118 * because it is apparently the best available slave we have, even though its
1119 * updelay hasn't timed out yet.
1121 * If new_active is not NULL, caller must hold bond->lock for read and
1122 * curr_slave_lock for write_bh.
1124 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1126 struct slave *old_active = bond->curr_active_slave;
1128 if (old_active == new_active)
1132 new_active->jiffies = jiffies;
1134 if (new_active->link == BOND_LINK_BACK) {
1135 if (USES_PRIMARY(bond->params.mode)) {
1136 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1137 bond->dev->name, new_active->dev->name,
1138 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1141 new_active->delay = 0;
1142 new_active->link = BOND_LINK_UP;
1144 if (bond->params.mode == BOND_MODE_8023AD)
1145 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1147 if (bond_is_lb(bond))
1148 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1150 if (USES_PRIMARY(bond->params.mode)) {
1151 pr_info("%s: making interface %s the new active one.\n",
1152 bond->dev->name, new_active->dev->name);
1157 if (USES_PRIMARY(bond->params.mode))
1158 bond_mc_swap(bond, new_active, old_active);
1160 if (bond_is_lb(bond)) {
1161 bond_alb_handle_active_change(bond, new_active);
1163 bond_set_slave_inactive_flags(old_active);
1165 bond_set_slave_active_flags(new_active);
1167 bond->curr_active_slave = new_active;
1170 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1172 bond_set_slave_inactive_flags(old_active);
1175 bond_set_slave_active_flags(new_active);
1177 if (bond->params.fail_over_mac)
1178 bond_do_fail_over_mac(bond, new_active,
1181 bond->send_grat_arp = bond->params.num_grat_arp;
1182 bond_send_gratuitous_arp(bond);
1184 bond->send_unsol_na = bond->params.num_unsol_na;
1185 bond_send_unsolicited_na(bond);
1187 write_unlock_bh(&bond->curr_slave_lock);
1188 read_unlock(&bond->lock);
1190 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1192 read_lock(&bond->lock);
1193 write_lock_bh(&bond->curr_slave_lock);
1197 /* resend IGMP joins since active slave has changed or
1198 * all were sent on curr_active_slave */
1199 if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1200 bond->params.mode == BOND_MODE_ROUNDROBIN) {
1201 bond->igmp_retrans = bond->params.resend_igmp;
1202 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1207 * bond_select_active_slave - select a new active slave, if needed
1208 * @bond: our bonding struct
1210 * This functions should be called when one of the following occurs:
1211 * - The old curr_active_slave has been released or lost its link.
1212 * - The primary_slave has got its link back.
1213 * - A slave has got its link back and there's no old curr_active_slave.
1215 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1217 void bond_select_active_slave(struct bonding *bond)
1219 struct slave *best_slave;
1222 best_slave = bond_find_best_slave(bond);
1223 if (best_slave != bond->curr_active_slave) {
1224 bond_change_active_slave(bond, best_slave);
1225 rv = bond_set_carrier(bond);
1229 if (netif_carrier_ok(bond->dev)) {
1230 pr_info("%s: first active interface up!\n",
1233 pr_info("%s: now running without any active interface !\n",
1239 /*--------------------------- slave list handling ---------------------------*/
1242 * This function attaches the slave to the end of list.
1244 * bond->lock held for writing by caller.
1246 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1248 if (bond->first_slave == NULL) { /* attaching the first slave */
1249 new_slave->next = new_slave;
1250 new_slave->prev = new_slave;
1251 bond->first_slave = new_slave;
1253 new_slave->next = bond->first_slave;
1254 new_slave->prev = bond->first_slave->prev;
1255 new_slave->next->prev = new_slave;
1256 new_slave->prev->next = new_slave;
1263 * This function detaches the slave from the list.
1264 * WARNING: no check is made to verify if the slave effectively
1265 * belongs to <bond>.
1266 * Nothing is freed on return, structures are just unchained.
1267 * If any slave pointer in bond was pointing to <slave>,
1268 * it should be changed by the calling function.
1270 * bond->lock held for writing by caller.
1272 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1275 slave->next->prev = slave->prev;
1278 slave->prev->next = slave->next;
1280 if (bond->first_slave == slave) { /* slave is the first slave */
1281 if (bond->slave_cnt > 1) { /* there are more slave */
1282 bond->first_slave = slave->next;
1284 bond->first_slave = NULL; /* slave was the last one */
1293 #ifdef CONFIG_NET_POLL_CONTROLLER
1295 * You must hold read lock on bond->lock before calling this.
1297 static bool slaves_support_netpoll(struct net_device *bond_dev)
1299 struct bonding *bond = netdev_priv(bond_dev);
1300 struct slave *slave;
1304 bond_for_each_slave(bond, slave, i) {
1305 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1306 !slave->dev->netdev_ops->ndo_poll_controller)
1309 return i != 0 && ret;
1312 static void bond_poll_controller(struct net_device *bond_dev)
1314 struct bonding *bond = netdev_priv(bond_dev);
1315 struct slave *slave;
1318 bond_for_each_slave(bond, slave, i) {
1319 if (slave->dev && IS_UP(slave->dev))
1320 netpoll_poll_dev(slave->dev);
1324 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1326 struct bonding *bond = netdev_priv(bond_dev);
1327 struct slave *slave;
1328 const struct net_device_ops *ops;
1331 read_lock(&bond->lock);
1332 bond_dev->npinfo = NULL;
1333 bond_for_each_slave(bond, slave, i) {
1335 ops = slave->dev->netdev_ops;
1336 if (ops->ndo_netpoll_cleanup)
1337 ops->ndo_netpoll_cleanup(slave->dev);
1339 slave->dev->npinfo = NULL;
1342 read_unlock(&bond->lock);
1347 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1353 /*---------------------------------- IOCTL ----------------------------------*/
1355 static int bond_sethwaddr(struct net_device *bond_dev,
1356 struct net_device *slave_dev)
1358 pr_debug("bond_dev=%p\n", bond_dev);
1359 pr_debug("slave_dev=%p\n", slave_dev);
1360 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1361 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1365 #define BOND_VLAN_FEATURES \
1366 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1367 NETIF_F_HW_VLAN_FILTER)
1370 * Compute the common dev->feature set available to all slaves. Some
1371 * feature bits are managed elsewhere, so preserve those feature bits
1372 * on the master device.
1374 static int bond_compute_features(struct bonding *bond)
1376 struct slave *slave;
1377 struct net_device *bond_dev = bond->dev;
1378 unsigned long features = bond_dev->features;
1379 unsigned long vlan_features = 0;
1380 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1381 bond_dev->hard_header_len);
1384 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1385 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1387 if (!bond->first_slave)
1390 features &= ~NETIF_F_ONE_FOR_ALL;
1392 vlan_features = bond->first_slave->dev->vlan_features;
1393 bond_for_each_slave(bond, slave, i) {
1394 features = netdev_increment_features(features,
1395 slave->dev->features,
1396 NETIF_F_ONE_FOR_ALL);
1397 vlan_features = netdev_increment_features(vlan_features,
1398 slave->dev->vlan_features,
1399 NETIF_F_ONE_FOR_ALL);
1400 if (slave->dev->hard_header_len > max_hard_header_len)
1401 max_hard_header_len = slave->dev->hard_header_len;
1405 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1406 bond_dev->features = netdev_fix_features(features, NULL);
1407 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1408 bond_dev->hard_header_len = max_hard_header_len;
1413 static void bond_setup_by_slave(struct net_device *bond_dev,
1414 struct net_device *slave_dev)
1416 struct bonding *bond = netdev_priv(bond_dev);
1418 bond_dev->header_ops = slave_dev->header_ops;
1420 bond_dev->type = slave_dev->type;
1421 bond_dev->hard_header_len = slave_dev->hard_header_len;
1422 bond_dev->addr_len = slave_dev->addr_len;
1424 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1425 slave_dev->addr_len);
1426 bond->setup_by_slave = 1;
1429 /* enslave device <slave> to bond device <master> */
1430 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1432 struct bonding *bond = netdev_priv(bond_dev);
1433 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1434 struct slave *new_slave = NULL;
1435 struct netdev_hw_addr *ha;
1436 struct sockaddr addr;
1438 int old_features = bond_dev->features;
1441 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1442 slave_ops->ndo_do_ioctl == NULL) {
1443 pr_warning("%s: Warning: no link monitoring support for %s\n",
1444 bond_dev->name, slave_dev->name);
1447 /* bond must be initialized by bond_open() before enslaving */
1448 if (!(bond_dev->flags & IFF_UP)) {
1449 pr_warning("%s: master_dev is not up in bond_enslave\n",
1453 /* already enslaved */
1454 if (slave_dev->flags & IFF_SLAVE) {
1455 pr_debug("Error, Device was already enslaved\n");
1459 /* vlan challenged mutual exclusion */
1460 /* no need to lock since we're protected by rtnl_lock */
1461 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1462 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1464 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1465 bond_dev->name, slave_dev->name, bond_dev->name);
1468 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1469 bond_dev->name, slave_dev->name,
1470 slave_dev->name, bond_dev->name);
1471 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1474 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1475 if (bond->slave_cnt == 0) {
1476 /* First slave, and it is not VLAN challenged,
1477 * so remove the block of adding VLANs over the bond.
1479 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1484 * Old ifenslave binaries are no longer supported. These can
1485 * be identified with moderate accuracy by the state of the slave:
1486 * the current ifenslave will set the interface down prior to
1487 * enslaving it; the old ifenslave will not.
1489 if ((slave_dev->flags & IFF_UP)) {
1490 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1493 goto err_undo_flags;
1496 /* set bonding device ether type by slave - bonding netdevices are
1497 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1498 * there is a need to override some of the type dependent attribs/funcs.
1500 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1501 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1503 if (bond->slave_cnt == 0) {
1504 if (bond_dev->type != slave_dev->type) {
1505 pr_debug("%s: change device type from %d to %d\n",
1507 bond_dev->type, slave_dev->type);
1509 res = netdev_bonding_change(bond_dev,
1510 NETDEV_PRE_TYPE_CHANGE);
1511 res = notifier_to_errno(res);
1513 pr_err("%s: refused to change device type\n",
1516 goto err_undo_flags;
1519 /* Flush unicast and multicast addresses */
1520 dev_uc_flush(bond_dev);
1521 dev_mc_flush(bond_dev);
1523 if (slave_dev->type != ARPHRD_ETHER)
1524 bond_setup_by_slave(bond_dev, slave_dev);
1526 ether_setup(bond_dev);
1528 netdev_bonding_change(bond_dev,
1529 NETDEV_POST_TYPE_CHANGE);
1531 } else if (bond_dev->type != slave_dev->type) {
1532 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1534 slave_dev->type, bond_dev->type);
1536 goto err_undo_flags;
1539 if (slave_ops->ndo_set_mac_address == NULL) {
1540 if (bond->slave_cnt == 0) {
1541 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1543 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1544 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1545 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",
1548 goto err_undo_flags;
1552 /* If this is the first slave, then we need to set the master's hardware
1553 * address to be the same as the slave's. */
1554 if (is_zero_ether_addr(bond->dev->dev_addr))
1555 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1556 slave_dev->addr_len);
1559 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1562 goto err_undo_flags;
1566 * Set the new_slave's queue_id to be zero. Queue ID mapping
1567 * is set via sysfs or module option if desired.
1569 new_slave->queue_id = 0;
1571 /* Save slave's original mtu and then set it to match the bond */
1572 new_slave->original_mtu = slave_dev->mtu;
1573 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1575 pr_debug("Error %d calling dev_set_mtu\n", res);
1580 * Save slave's original ("permanent") mac address for modes
1581 * that need it, and for restoring it upon release, and then
1582 * set it to the master's address
1584 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1586 if (!bond->params.fail_over_mac) {
1588 * Set slave to master's mac address. The application already
1589 * set the master's mac address to that of the first slave
1591 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1592 addr.sa_family = slave_dev->type;
1593 res = dev_set_mac_address(slave_dev, &addr);
1595 pr_debug("Error %d calling set_mac_address\n", res);
1596 goto err_restore_mtu;
1600 res = netdev_set_master(slave_dev, bond_dev);
1602 pr_debug("Error %d calling netdev_set_master\n", res);
1603 goto err_restore_mac;
1605 /* open the slave since the application closed it */
1606 res = dev_open(slave_dev);
1608 pr_debug("Opening slave %s failed\n", slave_dev->name);
1609 goto err_unset_master;
1612 new_slave->dev = slave_dev;
1613 slave_dev->priv_flags |= IFF_BONDING;
1615 if (bond_is_lb(bond)) {
1616 /* bond_alb_init_slave() must be called before all other stages since
1617 * it might fail and we do not want to have to undo everything
1619 res = bond_alb_init_slave(bond, new_slave);
1624 /* If the mode USES_PRIMARY, then the new slave gets the
1625 * master's promisc (and mc) settings only if it becomes the
1626 * curr_active_slave, and that is taken care of later when calling
1627 * bond_change_active()
1629 if (!USES_PRIMARY(bond->params.mode)) {
1630 /* set promiscuity level to new slave */
1631 if (bond_dev->flags & IFF_PROMISC) {
1632 res = dev_set_promiscuity(slave_dev, 1);
1637 /* set allmulti level to new slave */
1638 if (bond_dev->flags & IFF_ALLMULTI) {
1639 res = dev_set_allmulti(slave_dev, 1);
1644 netif_addr_lock_bh(bond_dev);
1645 /* upload master's mc_list to new slave */
1646 netdev_for_each_mc_addr(ha, bond_dev)
1647 dev_mc_add(slave_dev, ha->addr);
1648 netif_addr_unlock_bh(bond_dev);
1651 if (bond->params.mode == BOND_MODE_8023AD) {
1652 /* add lacpdu mc addr to mc list */
1653 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1655 dev_mc_add(slave_dev, lacpdu_multicast);
1658 bond_add_vlans_on_slave(bond, slave_dev);
1660 write_lock_bh(&bond->lock);
1662 bond_attach_slave(bond, new_slave);
1664 new_slave->delay = 0;
1665 new_slave->link_failure_count = 0;
1667 bond_compute_features(bond);
1669 write_unlock_bh(&bond->lock);
1671 read_lock(&bond->lock);
1673 new_slave->last_arp_rx = jiffies;
1675 if (bond->params.miimon && !bond->params.use_carrier) {
1676 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1678 if ((link_reporting == -1) && !bond->params.arp_interval) {
1680 * miimon is set but a bonded network driver
1681 * does not support ETHTOOL/MII and
1682 * arp_interval is not set. Note: if
1683 * use_carrier is enabled, we will never go
1684 * here (because netif_carrier is always
1685 * supported); thus, we don't need to change
1686 * the messages for netif_carrier.
1688 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",
1689 bond_dev->name, slave_dev->name);
1690 } else if (link_reporting == -1) {
1691 /* unable get link status using mii/ethtool */
1692 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",
1693 bond_dev->name, slave_dev->name);
1697 /* check for initial state */
1698 if (!bond->params.miimon ||
1699 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1700 if (bond->params.updelay) {
1701 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1702 new_slave->link = BOND_LINK_BACK;
1703 new_slave->delay = bond->params.updelay;
1705 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1706 new_slave->link = BOND_LINK_UP;
1708 new_slave->jiffies = jiffies;
1710 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1711 new_slave->link = BOND_LINK_DOWN;
1714 if (bond_update_speed_duplex(new_slave) &&
1715 (new_slave->link != BOND_LINK_DOWN)) {
1716 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1717 bond_dev->name, new_slave->dev->name);
1719 if (bond->params.mode == BOND_MODE_8023AD) {
1720 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1725 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1726 /* if there is a primary slave, remember it */
1727 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1728 bond->primary_slave = new_slave;
1729 bond->force_primary = true;
1733 write_lock_bh(&bond->curr_slave_lock);
1735 switch (bond->params.mode) {
1736 case BOND_MODE_ACTIVEBACKUP:
1737 bond_set_slave_inactive_flags(new_slave);
1738 bond_select_active_slave(bond);
1740 case BOND_MODE_8023AD:
1741 /* in 802.3ad mode, the internal mechanism
1742 * will activate the slaves in the selected
1745 bond_set_slave_inactive_flags(new_slave);
1746 /* if this is the first slave */
1747 if (bond->slave_cnt == 1) {
1748 SLAVE_AD_INFO(new_slave).id = 1;
1749 /* Initialize AD with the number of times that the AD timer is called in 1 second
1750 * can be called only after the mac address of the bond is set
1752 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1753 bond->params.lacp_fast);
1755 SLAVE_AD_INFO(new_slave).id =
1756 SLAVE_AD_INFO(new_slave->prev).id + 1;
1759 bond_3ad_bind_slave(new_slave);
1763 new_slave->state = BOND_STATE_ACTIVE;
1764 bond_set_slave_inactive_flags(new_slave);
1765 bond_select_active_slave(bond);
1768 pr_debug("This slave is always active in trunk mode\n");
1770 /* always active in trunk mode */
1771 new_slave->state = BOND_STATE_ACTIVE;
1773 /* In trunking mode there is little meaning to curr_active_slave
1774 * anyway (it holds no special properties of the bond device),
1775 * so we can change it without calling change_active_interface()
1777 if (!bond->curr_active_slave)
1778 bond->curr_active_slave = new_slave;
1781 } /* switch(bond_mode) */
1783 write_unlock_bh(&bond->curr_slave_lock);
1785 bond_set_carrier(bond);
1787 #ifdef CONFIG_NET_POLL_CONTROLLER
1788 if (slaves_support_netpoll(bond_dev)) {
1789 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1790 if (bond_dev->npinfo)
1791 slave_dev->npinfo = bond_dev->npinfo;
1792 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1793 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1794 pr_info("New slave device %s does not support netpoll\n",
1796 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1799 read_unlock(&bond->lock);
1801 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1805 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1806 bond_dev->name, slave_dev->name,
1807 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1808 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1810 /* enslave is successful */
1813 /* Undo stages on error */
1815 dev_close(slave_dev);
1818 netdev_set_master(slave_dev, NULL);
1821 if (!bond->params.fail_over_mac) {
1822 /* XXX TODO - fom follow mode needs to change master's
1823 * MAC if this slave's MAC is in use by the bond, or at
1824 * least print a warning.
1826 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1827 addr.sa_family = slave_dev->type;
1828 dev_set_mac_address(slave_dev, &addr);
1832 dev_set_mtu(slave_dev, new_slave->original_mtu);
1838 bond_dev->features = old_features;
1844 * Try to release the slave device <slave> from the bond device <master>
1845 * It is legal to access curr_active_slave without a lock because all the function
1848 * The rules for slave state should be:
1849 * for Active/Backup:
1850 * Active stays on all backups go down
1851 * for Bonded connections:
1852 * The first up interface should be left on and all others downed.
1854 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1856 struct bonding *bond = netdev_priv(bond_dev);
1857 struct slave *slave, *oldcurrent;
1858 struct sockaddr addr;
1860 /* slave is not a slave or master is not master of this slave */
1861 if (!(slave_dev->flags & IFF_SLAVE) ||
1862 (slave_dev->master != bond_dev)) {
1863 pr_err("%s: Error: cannot release %s.\n",
1864 bond_dev->name, slave_dev->name);
1869 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1870 write_lock_bh(&bond->lock);
1872 slave = bond_get_slave_by_dev(bond, slave_dev);
1874 /* not a slave of this bond */
1875 pr_info("%s: %s not enslaved\n",
1876 bond_dev->name, slave_dev->name);
1877 write_unlock_bh(&bond->lock);
1878 unblock_netpoll_tx();
1882 if (!bond->params.fail_over_mac) {
1883 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1884 bond->slave_cnt > 1)
1885 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",
1886 bond_dev->name, slave_dev->name,
1888 bond_dev->name, slave_dev->name);
1891 /* Inform AD package of unbinding of slave. */
1892 if (bond->params.mode == BOND_MODE_8023AD) {
1893 /* must be called before the slave is
1894 * detached from the list
1896 bond_3ad_unbind_slave(slave);
1899 pr_info("%s: releasing %s interface %s\n",
1901 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1904 oldcurrent = bond->curr_active_slave;
1906 bond->current_arp_slave = NULL;
1908 /* release the slave from its bond */
1909 bond_detach_slave(bond, slave);
1911 bond_compute_features(bond);
1913 if (bond->primary_slave == slave)
1914 bond->primary_slave = NULL;
1916 if (oldcurrent == slave)
1917 bond_change_active_slave(bond, NULL);
1919 if (bond_is_lb(bond)) {
1920 /* Must be called only after the slave has been
1921 * detached from the list and the curr_active_slave
1922 * has been cleared (if our_slave == old_current),
1923 * but before a new active slave is selected.
1925 write_unlock_bh(&bond->lock);
1926 bond_alb_deinit_slave(bond, slave);
1927 write_lock_bh(&bond->lock);
1930 if (oldcurrent == slave) {
1932 * Note that we hold RTNL over this sequence, so there
1933 * is no concern that another slave add/remove event
1936 write_unlock_bh(&bond->lock);
1937 read_lock(&bond->lock);
1938 write_lock_bh(&bond->curr_slave_lock);
1940 bond_select_active_slave(bond);
1942 write_unlock_bh(&bond->curr_slave_lock);
1943 read_unlock(&bond->lock);
1944 write_lock_bh(&bond->lock);
1947 if (bond->slave_cnt == 0) {
1948 bond_set_carrier(bond);
1950 /* if the last slave was removed, zero the mac address
1951 * of the master so it will be set by the application
1952 * to the mac address of the first slave
1954 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1957 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1959 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1960 bond_dev->name, bond_dev->name);
1961 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1964 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1965 !bond_has_challenged_slaves(bond)) {
1966 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1967 bond_dev->name, slave_dev->name, bond_dev->name);
1968 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1971 write_unlock_bh(&bond->lock);
1972 unblock_netpoll_tx();
1974 /* must do this from outside any spinlocks */
1975 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1977 bond_del_vlans_from_slave(bond, slave_dev);
1979 /* If the mode USES_PRIMARY, then we should only remove its
1980 * promisc and mc settings if it was the curr_active_slave, but that was
1981 * already taken care of above when we detached the slave
1983 if (!USES_PRIMARY(bond->params.mode)) {
1984 /* unset promiscuity level from slave */
1985 if (bond_dev->flags & IFF_PROMISC)
1986 dev_set_promiscuity(slave_dev, -1);
1988 /* unset allmulti level from slave */
1989 if (bond_dev->flags & IFF_ALLMULTI)
1990 dev_set_allmulti(slave_dev, -1);
1992 /* flush master's mc_list from slave */
1993 netif_addr_lock_bh(bond_dev);
1994 bond_mc_list_flush(bond_dev, slave_dev);
1995 netif_addr_unlock_bh(bond_dev);
1998 netdev_set_master(slave_dev, NULL);
2000 #ifdef CONFIG_NET_POLL_CONTROLLER
2001 read_lock_bh(&bond->lock);
2003 if (slaves_support_netpoll(bond_dev))
2004 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2005 read_unlock_bh(&bond->lock);
2006 if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2007 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2009 slave_dev->npinfo = NULL;
2012 /* close slave before restoring its mac address */
2013 dev_close(slave_dev);
2015 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2016 /* restore original ("permanent") mac address */
2017 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2018 addr.sa_family = slave_dev->type;
2019 dev_set_mac_address(slave_dev, &addr);
2022 dev_set_mtu(slave_dev, slave->original_mtu);
2024 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2025 IFF_SLAVE_INACTIVE | IFF_BONDING |
2030 return 0; /* deletion OK */
2034 * First release a slave and than destroy the bond if no more slaves are left.
2035 * Must be under rtnl_lock when this function is called.
2037 static int bond_release_and_destroy(struct net_device *bond_dev,
2038 struct net_device *slave_dev)
2040 struct bonding *bond = netdev_priv(bond_dev);
2043 ret = bond_release(bond_dev, slave_dev);
2044 if ((ret == 0) && (bond->slave_cnt == 0)) {
2045 pr_info("%s: destroying bond %s.\n",
2046 bond_dev->name, bond_dev->name);
2047 unregister_netdevice(bond_dev);
2053 * This function releases all slaves.
2055 static int bond_release_all(struct net_device *bond_dev)
2057 struct bonding *bond = netdev_priv(bond_dev);
2058 struct slave *slave;
2059 struct net_device *slave_dev;
2060 struct sockaddr addr;
2062 write_lock_bh(&bond->lock);
2064 netif_carrier_off(bond_dev);
2066 if (bond->slave_cnt == 0)
2069 bond->current_arp_slave = NULL;
2070 bond->primary_slave = NULL;
2071 bond_change_active_slave(bond, NULL);
2073 while ((slave = bond->first_slave) != NULL) {
2074 /* Inform AD package of unbinding of slave
2075 * before slave is detached from the list.
2077 if (bond->params.mode == BOND_MODE_8023AD)
2078 bond_3ad_unbind_slave(slave);
2080 slave_dev = slave->dev;
2081 bond_detach_slave(bond, slave);
2083 /* now that the slave is detached, unlock and perform
2084 * all the undo steps that should not be called from
2087 write_unlock_bh(&bond->lock);
2089 if (bond_is_lb(bond)) {
2090 /* must be called only after the slave
2091 * has been detached from the list
2093 bond_alb_deinit_slave(bond, slave);
2096 bond_compute_features(bond);
2098 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2099 bond_del_vlans_from_slave(bond, slave_dev);
2101 /* If the mode USES_PRIMARY, then we should only remove its
2102 * promisc and mc settings if it was the curr_active_slave, but that was
2103 * already taken care of above when we detached the slave
2105 if (!USES_PRIMARY(bond->params.mode)) {
2106 /* unset promiscuity level from slave */
2107 if (bond_dev->flags & IFF_PROMISC)
2108 dev_set_promiscuity(slave_dev, -1);
2110 /* unset allmulti level from slave */
2111 if (bond_dev->flags & IFF_ALLMULTI)
2112 dev_set_allmulti(slave_dev, -1);
2114 /* flush master's mc_list from slave */
2115 netif_addr_lock_bh(bond_dev);
2116 bond_mc_list_flush(bond_dev, slave_dev);
2117 netif_addr_unlock_bh(bond_dev);
2120 netdev_set_master(slave_dev, NULL);
2122 /* close slave before restoring its mac address */
2123 dev_close(slave_dev);
2125 if (!bond->params.fail_over_mac) {
2126 /* restore original ("permanent") mac address*/
2127 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2128 addr.sa_family = slave_dev->type;
2129 dev_set_mac_address(slave_dev, &addr);
2132 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2133 IFF_SLAVE_INACTIVE);
2137 /* re-acquire the lock before getting the next slave */
2138 write_lock_bh(&bond->lock);
2141 /* zero the mac address of the master so it will be
2142 * set by the application to the mac address of the
2145 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2148 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2150 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2151 bond_dev->name, bond_dev->name);
2152 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2156 pr_info("%s: released all slaves\n", bond_dev->name);
2159 write_unlock_bh(&bond->lock);
2164 * This function changes the active slave to slave <slave_dev>.
2165 * It returns -EINVAL in the following cases.
2166 * - <slave_dev> is not found in the list.
2167 * - There is not active slave now.
2168 * - <slave_dev> is already active.
2169 * - The link state of <slave_dev> is not BOND_LINK_UP.
2170 * - <slave_dev> is not running.
2171 * In these cases, this function does nothing.
2172 * In the other cases, current_slave pointer is changed and 0 is returned.
2174 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2176 struct bonding *bond = netdev_priv(bond_dev);
2177 struct slave *old_active = NULL;
2178 struct slave *new_active = NULL;
2181 if (!USES_PRIMARY(bond->params.mode))
2184 /* Verify that master_dev is indeed the master of slave_dev */
2185 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2188 read_lock(&bond->lock);
2190 read_lock(&bond->curr_slave_lock);
2191 old_active = bond->curr_active_slave;
2192 read_unlock(&bond->curr_slave_lock);
2194 new_active = bond_get_slave_by_dev(bond, slave_dev);
2197 * Changing to the current active: do nothing; return success.
2199 if (new_active && (new_active == old_active)) {
2200 read_unlock(&bond->lock);
2206 (new_active->link == BOND_LINK_UP) &&
2207 IS_UP(new_active->dev)) {
2209 write_lock_bh(&bond->curr_slave_lock);
2210 bond_change_active_slave(bond, new_active);
2211 write_unlock_bh(&bond->curr_slave_lock);
2212 unblock_netpoll_tx();
2216 read_unlock(&bond->lock);
2221 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2223 struct bonding *bond = netdev_priv(bond_dev);
2225 info->bond_mode = bond->params.mode;
2226 info->miimon = bond->params.miimon;
2228 read_lock(&bond->lock);
2229 info->num_slaves = bond->slave_cnt;
2230 read_unlock(&bond->lock);
2235 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2237 struct bonding *bond = netdev_priv(bond_dev);
2238 struct slave *slave;
2239 int i, res = -ENODEV;
2241 read_lock(&bond->lock);
2243 bond_for_each_slave(bond, slave, i) {
2244 if (i == (int)info->slave_id) {
2246 strcpy(info->slave_name, slave->dev->name);
2247 info->link = slave->link;
2248 info->state = slave->state;
2249 info->link_failure_count = slave->link_failure_count;
2254 read_unlock(&bond->lock);
2259 /*-------------------------------- Monitoring -------------------------------*/
2262 static int bond_miimon_inspect(struct bonding *bond)
2264 struct slave *slave;
2265 int i, link_state, commit = 0;
2266 bool ignore_updelay;
2268 ignore_updelay = !bond->curr_active_slave ? true : false;
2270 bond_for_each_slave(bond, slave, i) {
2271 slave->new_link = BOND_LINK_NOCHANGE;
2273 link_state = bond_check_dev_link(bond, slave->dev, 0);
2275 switch (slave->link) {
2280 slave->link = BOND_LINK_FAIL;
2281 slave->delay = bond->params.downdelay;
2283 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2285 (bond->params.mode ==
2286 BOND_MODE_ACTIVEBACKUP) ?
2287 ((slave->state == BOND_STATE_ACTIVE) ?
2288 "active " : "backup ") : "",
2290 bond->params.downdelay * bond->params.miimon);
2293 case BOND_LINK_FAIL:
2296 * recovered before downdelay expired
2298 slave->link = BOND_LINK_UP;
2299 slave->jiffies = jiffies;
2300 pr_info("%s: link status up again after %d ms for interface %s.\n",
2302 (bond->params.downdelay - slave->delay) *
2303 bond->params.miimon,
2308 if (slave->delay <= 0) {
2309 slave->new_link = BOND_LINK_DOWN;
2317 case BOND_LINK_DOWN:
2321 slave->link = BOND_LINK_BACK;
2322 slave->delay = bond->params.updelay;
2325 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2326 bond->dev->name, slave->dev->name,
2327 ignore_updelay ? 0 :
2328 bond->params.updelay *
2329 bond->params.miimon);
2332 case BOND_LINK_BACK:
2334 slave->link = BOND_LINK_DOWN;
2335 pr_info("%s: link status down again after %d ms for interface %s.\n",
2337 (bond->params.updelay - slave->delay) *
2338 bond->params.miimon,
2347 if (slave->delay <= 0) {
2348 slave->new_link = BOND_LINK_UP;
2350 ignore_updelay = false;
2362 static void bond_miimon_commit(struct bonding *bond)
2364 struct slave *slave;
2367 bond_for_each_slave(bond, slave, i) {
2368 switch (slave->new_link) {
2369 case BOND_LINK_NOCHANGE:
2373 slave->link = BOND_LINK_UP;
2374 slave->jiffies = jiffies;
2376 if (bond->params.mode == BOND_MODE_8023AD) {
2377 /* prevent it from being the active one */
2378 slave->state = BOND_STATE_BACKUP;
2379 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2380 /* make it immediately active */
2381 slave->state = BOND_STATE_ACTIVE;
2382 } else if (slave != bond->primary_slave) {
2383 /* prevent it from being the active one */
2384 slave->state = BOND_STATE_BACKUP;
2387 bond_update_speed_duplex(slave);
2389 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2390 bond->dev->name, slave->dev->name,
2391 slave->speed, slave->duplex ? "full" : "half");
2393 /* notify ad that the link status has changed */
2394 if (bond->params.mode == BOND_MODE_8023AD)
2395 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2397 if (bond_is_lb(bond))
2398 bond_alb_handle_link_change(bond, slave,
2401 if (!bond->curr_active_slave ||
2402 (slave == bond->primary_slave))
2407 case BOND_LINK_DOWN:
2408 if (slave->link_failure_count < UINT_MAX)
2409 slave->link_failure_count++;
2411 slave->link = BOND_LINK_DOWN;
2413 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2414 bond->params.mode == BOND_MODE_8023AD)
2415 bond_set_slave_inactive_flags(slave);
2417 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2418 bond->dev->name, slave->dev->name);
2420 if (bond->params.mode == BOND_MODE_8023AD)
2421 bond_3ad_handle_link_change(slave,
2424 if (bond_is_lb(bond))
2425 bond_alb_handle_link_change(bond, slave,
2428 if (slave == bond->curr_active_slave)
2434 pr_err("%s: invalid new link %d on slave %s\n",
2435 bond->dev->name, slave->new_link,
2437 slave->new_link = BOND_LINK_NOCHANGE;
2445 write_lock_bh(&bond->curr_slave_lock);
2446 bond_select_active_slave(bond);
2447 write_unlock_bh(&bond->curr_slave_lock);
2448 unblock_netpoll_tx();
2451 bond_set_carrier(bond);
2457 * Really a wrapper that splits the mii monitor into two phases: an
2458 * inspection, then (if inspection indicates something needs to be done)
2459 * an acquisition of appropriate locks followed by a commit phase to
2460 * implement whatever link state changes are indicated.
2462 void bond_mii_monitor(struct work_struct *work)
2464 struct bonding *bond = container_of(work, struct bonding,
2467 read_lock(&bond->lock);
2468 if (bond->kill_timers)
2471 if (bond->slave_cnt == 0)
2474 if (bond->send_grat_arp) {
2475 read_lock(&bond->curr_slave_lock);
2476 bond_send_gratuitous_arp(bond);
2477 read_unlock(&bond->curr_slave_lock);
2480 if (bond->send_unsol_na) {
2481 read_lock(&bond->curr_slave_lock);
2482 bond_send_unsolicited_na(bond);
2483 read_unlock(&bond->curr_slave_lock);
2486 if (bond_miimon_inspect(bond)) {
2487 read_unlock(&bond->lock);
2489 read_lock(&bond->lock);
2491 bond_miimon_commit(bond);
2493 read_unlock(&bond->lock);
2494 rtnl_unlock(); /* might sleep, hold no other locks */
2495 read_lock(&bond->lock);
2499 if (bond->params.miimon)
2500 queue_delayed_work(bond->wq, &bond->mii_work,
2501 msecs_to_jiffies(bond->params.miimon));
2503 read_unlock(&bond->lock);
2506 static __be32 bond_glean_dev_ip(struct net_device *dev)
2508 struct in_device *idev;
2509 struct in_ifaddr *ifa;
2516 idev = __in_dev_get_rcu(dev);
2520 ifa = idev->ifa_list;
2524 addr = ifa->ifa_local;
2530 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2532 struct vlan_entry *vlan;
2534 if (ip == bond->master_ip)
2537 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2538 if (ip == vlan->vlan_ip)
2546 * We go to the (large) trouble of VLAN tagging ARP frames because
2547 * switches in VLAN mode (especially if ports are configured as
2548 * "native" to a VLAN) might not pass non-tagged frames.
2550 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2552 struct sk_buff *skb;
2554 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2555 slave_dev->name, dest_ip, src_ip, vlan_id);
2557 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2558 NULL, slave_dev->dev_addr, NULL);
2561 pr_err("ARP packet allocation failed\n");
2565 skb = vlan_put_tag(skb, vlan_id);
2567 pr_err("failed to insert VLAN tag\n");
2575 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2578 __be32 *targets = bond->params.arp_targets;
2579 struct vlan_entry *vlan;
2580 struct net_device *vlan_dev;
2584 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2587 pr_debug("basa: target %x\n", targets[i]);
2589 pr_debug("basa: empty vlan: arp_send\n");
2590 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2591 bond->master_ip, 0);
2596 * If VLANs are configured, we do a route lookup to
2597 * determine which VLAN interface would be used, so we
2598 * can tag the ARP with the proper VLAN tag.
2600 memset(&fl, 0, sizeof(fl));
2601 fl.fl4_dst = targets[i];
2602 fl.fl4_tos = RTO_ONLINK;
2604 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2606 if (net_ratelimit()) {
2607 pr_warning("%s: no route to arp_ip_target %pI4\n",
2608 bond->dev->name, &fl.fl4_dst);
2614 * This target is not on a VLAN
2616 if (rt->dst.dev == bond->dev) {
2618 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2619 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2620 bond->master_ip, 0);
2625 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2626 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2627 if (vlan_dev == rt->dst.dev) {
2628 vlan_id = vlan->vlan_id;
2629 pr_debug("basa: vlan match on %s %d\n",
2630 vlan_dev->name, vlan_id);
2637 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2638 vlan->vlan_ip, vlan_id);
2642 if (net_ratelimit()) {
2643 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2644 bond->dev->name, &fl.fl4_dst,
2645 rt->dst.dev ? rt->dst.dev->name : "NULL");
2652 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2653 * for each VLAN above us.
2655 * Caller must hold curr_slave_lock for read or better
2657 static void bond_send_gratuitous_arp(struct bonding *bond)
2659 struct slave *slave = bond->curr_active_slave;
2660 struct vlan_entry *vlan;
2661 struct net_device *vlan_dev;
2663 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2664 bond->dev->name, slave ? slave->dev->name : "NULL");
2666 if (!slave || !bond->send_grat_arp ||
2667 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2670 bond->send_grat_arp--;
2672 if (bond->master_ip) {
2673 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2674 bond->master_ip, 0);
2680 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2681 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2682 if (vlan->vlan_ip) {
2683 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2684 vlan->vlan_ip, vlan->vlan_id);
2689 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2692 __be32 *targets = bond->params.arp_targets;
2694 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2695 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2696 &sip, &tip, i, &targets[i],
2697 bond_has_this_ip(bond, tip));
2698 if (sip == targets[i]) {
2699 if (bond_has_this_ip(bond, tip))
2700 slave->last_arp_rx = jiffies;
2706 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2709 struct slave *slave;
2710 struct bonding *bond;
2711 unsigned char *arp_ptr;
2714 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2716 * When using VLANS and bonding, dev and oriv_dev may be
2717 * incorrect if the physical interface supports VLAN
2718 * acceleration. With this change ARP validation now
2719 * works for hosts only reachable on the VLAN interface.
2721 dev = vlan_dev_real_dev(dev);
2722 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2725 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2728 bond = netdev_priv(dev);
2729 read_lock(&bond->lock);
2731 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2732 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2733 orig_dev ? orig_dev->name : "NULL");
2735 slave = bond_get_slave_by_dev(bond, orig_dev);
2736 if (!slave || !slave_do_arp_validate(bond, slave))
2739 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2743 if (arp->ar_hln != dev->addr_len ||
2744 skb->pkt_type == PACKET_OTHERHOST ||
2745 skb->pkt_type == PACKET_LOOPBACK ||
2746 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2747 arp->ar_pro != htons(ETH_P_IP) ||
2751 arp_ptr = (unsigned char *)(arp + 1);
2752 arp_ptr += dev->addr_len;
2753 memcpy(&sip, arp_ptr, 4);
2754 arp_ptr += 4 + dev->addr_len;
2755 memcpy(&tip, arp_ptr, 4);
2757 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2758 bond->dev->name, slave->dev->name, slave->state,
2759 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2763 * Backup slaves won't see the ARP reply, but do come through
2764 * here for each ARP probe (so we swap the sip/tip to validate
2765 * the probe). In a "redundant switch, common router" type of
2766 * configuration, the ARP probe will (hopefully) travel from
2767 * the active, through one switch, the router, then the other
2768 * switch before reaching the backup.
2770 if (slave->state == BOND_STATE_ACTIVE)
2771 bond_validate_arp(bond, slave, sip, tip);
2773 bond_validate_arp(bond, slave, tip, sip);
2776 read_unlock(&bond->lock);
2779 return NET_RX_SUCCESS;
2783 * this function is called regularly to monitor each slave's link
2784 * ensuring that traffic is being sent and received when arp monitoring
2785 * is used in load-balancing mode. if the adapter has been dormant, then an
2786 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2787 * arp monitoring in active backup mode.
2789 void bond_loadbalance_arp_mon(struct work_struct *work)
2791 struct bonding *bond = container_of(work, struct bonding,
2793 struct slave *slave, *oldcurrent;
2794 int do_failover = 0;
2798 read_lock(&bond->lock);
2800 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2802 if (bond->kill_timers)
2805 if (bond->slave_cnt == 0)
2808 read_lock(&bond->curr_slave_lock);
2809 oldcurrent = bond->curr_active_slave;
2810 read_unlock(&bond->curr_slave_lock);
2812 /* see if any of the previous devices are up now (i.e. they have
2813 * xmt and rcv traffic). the curr_active_slave does not come into
2814 * the picture unless it is null. also, slave->jiffies is not needed
2815 * here because we send an arp on each slave and give a slave as
2816 * long as it needs to get the tx/rx within the delta.
2817 * TODO: what about up/down delay in arp mode? it wasn't here before
2820 bond_for_each_slave(bond, slave, i) {
2821 unsigned long trans_start = dev_trans_start(slave->dev);
2823 if (slave->link != BOND_LINK_UP) {
2824 if (time_in_range(jiffies,
2825 trans_start - delta_in_ticks,
2826 trans_start + delta_in_ticks) &&
2827 time_in_range(jiffies,
2828 slave->dev->last_rx - delta_in_ticks,
2829 slave->dev->last_rx + delta_in_ticks)) {
2831 slave->link = BOND_LINK_UP;
2832 slave->state = BOND_STATE_ACTIVE;
2834 /* primary_slave has no meaning in round-robin
2835 * mode. the window of a slave being up and
2836 * curr_active_slave being null after enslaving
2840 pr_info("%s: link status definitely up for interface %s, ",
2845 pr_info("%s: interface %s is now up\n",
2851 /* slave->link == BOND_LINK_UP */
2853 /* not all switches will respond to an arp request
2854 * when the source ip is 0, so don't take the link down
2855 * if we don't know our ip yet
2857 if (!time_in_range(jiffies,
2858 trans_start - delta_in_ticks,
2859 trans_start + 2 * delta_in_ticks) ||
2860 !time_in_range(jiffies,
2861 slave->dev->last_rx - delta_in_ticks,
2862 slave->dev->last_rx + 2 * delta_in_ticks)) {
2864 slave->link = BOND_LINK_DOWN;
2865 slave->state = BOND_STATE_BACKUP;
2867 if (slave->link_failure_count < UINT_MAX)
2868 slave->link_failure_count++;
2870 pr_info("%s: interface %s is now down.\n",
2874 if (slave == oldcurrent)
2879 /* note: if switch is in round-robin mode, all links
2880 * must tx arp to ensure all links rx an arp - otherwise
2881 * links may oscillate or not come up at all; if switch is
2882 * in something like xor mode, there is nothing we can
2883 * do - all replies will be rx'ed on same link causing slaves
2884 * to be unstable during low/no traffic periods
2886 if (IS_UP(slave->dev))
2887 bond_arp_send_all(bond, slave);
2892 write_lock_bh(&bond->curr_slave_lock);
2894 bond_select_active_slave(bond);
2896 write_unlock_bh(&bond->curr_slave_lock);
2897 unblock_netpoll_tx();
2901 if (bond->params.arp_interval)
2902 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2904 read_unlock(&bond->lock);
2908 * Called to inspect slaves for active-backup mode ARP monitor link state
2909 * changes. Sets new_link in slaves to specify what action should take
2910 * place for the slave. Returns 0 if no changes are found, >0 if changes
2911 * to link states must be committed.
2913 * Called with bond->lock held for read.
2915 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2917 struct slave *slave;
2919 unsigned long trans_start;
2921 bond_for_each_slave(bond, slave, i) {
2922 slave->new_link = BOND_LINK_NOCHANGE;
2924 if (slave->link != BOND_LINK_UP) {
2925 if (time_in_range(jiffies,
2926 slave_last_rx(bond, slave) - delta_in_ticks,
2927 slave_last_rx(bond, slave) + delta_in_ticks)) {
2929 slave->new_link = BOND_LINK_UP;
2937 * Give slaves 2*delta after being enslaved or made
2938 * active. This avoids bouncing, as the last receive
2939 * times need a full ARP monitor cycle to be updated.
2941 if (time_in_range(jiffies,
2942 slave->jiffies - delta_in_ticks,
2943 slave->jiffies + 2 * delta_in_ticks))
2947 * Backup slave is down if:
2948 * - No current_arp_slave AND
2949 * - more than 3*delta since last receive AND
2950 * - the bond has an IP address
2952 * Note: a non-null current_arp_slave indicates
2953 * the curr_active_slave went down and we are
2954 * searching for a new one; under this condition
2955 * we only take the curr_active_slave down - this
2956 * gives each slave a chance to tx/rx traffic
2957 * before being taken out
2959 if (slave->state == BOND_STATE_BACKUP &&
2960 !bond->current_arp_slave &&
2961 !time_in_range(jiffies,
2962 slave_last_rx(bond, slave) - delta_in_ticks,
2963 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2965 slave->new_link = BOND_LINK_DOWN;
2970 * Active slave is down if:
2971 * - more than 2*delta since transmitting OR
2972 * - (more than 2*delta since receive AND
2973 * the bond has an IP address)
2975 trans_start = dev_trans_start(slave->dev);
2976 if ((slave->state == BOND_STATE_ACTIVE) &&
2977 (!time_in_range(jiffies,
2978 trans_start - delta_in_ticks,
2979 trans_start + 2 * delta_in_ticks) ||
2980 !time_in_range(jiffies,
2981 slave_last_rx(bond, slave) - delta_in_ticks,
2982 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2984 slave->new_link = BOND_LINK_DOWN;
2993 * Called to commit link state changes noted by inspection step of
2994 * active-backup mode ARP monitor.
2996 * Called with RTNL and bond->lock for read.
2998 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3000 struct slave *slave;
3002 unsigned long trans_start;
3004 bond_for_each_slave(bond, slave, i) {
3005 switch (slave->new_link) {
3006 case BOND_LINK_NOCHANGE:
3010 trans_start = dev_trans_start(slave->dev);
3011 if ((!bond->curr_active_slave &&
3012 time_in_range(jiffies,
3013 trans_start - delta_in_ticks,
3014 trans_start + delta_in_ticks)) ||
3015 bond->curr_active_slave != slave) {
3016 slave->link = BOND_LINK_UP;
3017 bond->current_arp_slave = NULL;
3019 pr_info("%s: link status definitely up for interface %s.\n",
3020 bond->dev->name, slave->dev->name);
3022 if (!bond->curr_active_slave ||
3023 (slave == bond->primary_slave))
3030 case BOND_LINK_DOWN:
3031 if (slave->link_failure_count < UINT_MAX)
3032 slave->link_failure_count++;
3034 slave->link = BOND_LINK_DOWN;
3035 bond_set_slave_inactive_flags(slave);
3037 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3038 bond->dev->name, slave->dev->name);
3040 if (slave == bond->curr_active_slave) {
3041 bond->current_arp_slave = NULL;
3048 pr_err("%s: impossible: new_link %d on slave %s\n",
3049 bond->dev->name, slave->new_link,
3057 write_lock_bh(&bond->curr_slave_lock);
3058 bond_select_active_slave(bond);
3059 write_unlock_bh(&bond->curr_slave_lock);
3060 unblock_netpoll_tx();
3063 bond_set_carrier(bond);
3067 * Send ARP probes for active-backup mode ARP monitor.
3069 * Called with bond->lock held for read.
3071 static void bond_ab_arp_probe(struct bonding *bond)
3073 struct slave *slave;
3076 read_lock(&bond->curr_slave_lock);
3078 if (bond->current_arp_slave && bond->curr_active_slave)
3079 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3080 bond->current_arp_slave->dev->name,
3081 bond->curr_active_slave->dev->name);
3083 if (bond->curr_active_slave) {
3084 bond_arp_send_all(bond, bond->curr_active_slave);
3085 read_unlock(&bond->curr_slave_lock);
3089 read_unlock(&bond->curr_slave_lock);
3091 /* if we don't have a curr_active_slave, search for the next available
3092 * backup slave from the current_arp_slave and make it the candidate
3093 * for becoming the curr_active_slave
3096 if (!bond->current_arp_slave) {
3097 bond->current_arp_slave = bond->first_slave;
3098 if (!bond->current_arp_slave)
3102 bond_set_slave_inactive_flags(bond->current_arp_slave);
3104 /* search for next candidate */
3105 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3106 if (IS_UP(slave->dev)) {
3107 slave->link = BOND_LINK_BACK;
3108 bond_set_slave_active_flags(slave);
3109 bond_arp_send_all(bond, slave);
3110 slave->jiffies = jiffies;
3111 bond->current_arp_slave = slave;
3115 /* if the link state is up at this point, we
3116 * mark it down - this can happen if we have
3117 * simultaneous link failures and
3118 * reselect_active_interface doesn't make this
3119 * one the current slave so it is still marked
3120 * up when it is actually down
3122 if (slave->link == BOND_LINK_UP) {
3123 slave->link = BOND_LINK_DOWN;
3124 if (slave->link_failure_count < UINT_MAX)
3125 slave->link_failure_count++;
3127 bond_set_slave_inactive_flags(slave);
3129 pr_info("%s: backup interface %s is now down.\n",
3130 bond->dev->name, slave->dev->name);
3135 void bond_activebackup_arp_mon(struct work_struct *work)
3137 struct bonding *bond = container_of(work, struct bonding,
3141 read_lock(&bond->lock);
3143 if (bond->kill_timers)
3146 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3148 if (bond->slave_cnt == 0)
3151 if (bond->send_grat_arp) {
3152 read_lock(&bond->curr_slave_lock);
3153 bond_send_gratuitous_arp(bond);
3154 read_unlock(&bond->curr_slave_lock);
3157 if (bond->send_unsol_na) {
3158 read_lock(&bond->curr_slave_lock);
3159 bond_send_unsolicited_na(bond);
3160 read_unlock(&bond->curr_slave_lock);
3163 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3164 read_unlock(&bond->lock);
3166 read_lock(&bond->lock);
3168 bond_ab_arp_commit(bond, delta_in_ticks);
3170 read_unlock(&bond->lock);
3172 read_lock(&bond->lock);
3175 bond_ab_arp_probe(bond);
3178 if (bond->params.arp_interval)
3179 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3181 read_unlock(&bond->lock);
3184 /*------------------------------ proc/seq_file-------------------------------*/
3186 #ifdef CONFIG_PROC_FS
3188 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3189 __acquires(&dev_base_lock)
3190 __acquires(&bond->lock)
3192 struct bonding *bond = seq->private;
3194 struct slave *slave;
3197 /* make sure the bond won't be taken away */
3198 read_lock(&dev_base_lock);
3199 read_lock(&bond->lock);
3202 return SEQ_START_TOKEN;
3204 bond_for_each_slave(bond, slave, i) {
3212 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3214 struct bonding *bond = seq->private;
3215 struct slave *slave = v;
3218 if (v == SEQ_START_TOKEN)
3219 return bond->first_slave;
3221 slave = slave->next;
3223 return (slave == bond->first_slave) ? NULL : slave;
3226 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3227 __releases(&bond->lock)
3228 __releases(&dev_base_lock)
3230 struct bonding *bond = seq->private;
3232 read_unlock(&bond->lock);
3233 read_unlock(&dev_base_lock);
3236 static void bond_info_show_master(struct seq_file *seq)
3238 struct bonding *bond = seq->private;
3242 read_lock(&bond->curr_slave_lock);
3243 curr = bond->curr_active_slave;
3244 read_unlock(&bond->curr_slave_lock);
3246 seq_printf(seq, "Bonding Mode: %s",
3247 bond_mode_name(bond->params.mode));
3249 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3250 bond->params.fail_over_mac)
3251 seq_printf(seq, " (fail_over_mac %s)",
3252 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3254 seq_printf(seq, "\n");
3256 if (bond->params.mode == BOND_MODE_XOR ||
3257 bond->params.mode == BOND_MODE_8023AD) {
3258 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3259 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3260 bond->params.xmit_policy);
3263 if (USES_PRIMARY(bond->params.mode)) {
3264 seq_printf(seq, "Primary Slave: %s",
3265 (bond->primary_slave) ?
3266 bond->primary_slave->dev->name : "None");
3267 if (bond->primary_slave)
3268 seq_printf(seq, " (primary_reselect %s)",
3269 pri_reselect_tbl[bond->params.primary_reselect].modename);
3271 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3272 (curr) ? curr->dev->name : "None");
3275 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3277 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3278 seq_printf(seq, "Up Delay (ms): %d\n",
3279 bond->params.updelay * bond->params.miimon);
3280 seq_printf(seq, "Down Delay (ms): %d\n",
3281 bond->params.downdelay * bond->params.miimon);
3284 /* ARP information */
3285 if (bond->params.arp_interval > 0) {
3287 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3288 bond->params.arp_interval);
3290 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3292 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3293 if (!bond->params.arp_targets[i])
3296 seq_printf(seq, ",");
3297 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3300 seq_printf(seq, "\n");
3303 if (bond->params.mode == BOND_MODE_8023AD) {
3304 struct ad_info ad_info;
3306 seq_puts(seq, "\n802.3ad info\n");
3307 seq_printf(seq, "LACP rate: %s\n",
3308 (bond->params.lacp_fast) ? "fast" : "slow");
3309 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3310 ad_select_tbl[bond->params.ad_select].modename);
3312 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3313 seq_printf(seq, "bond %s has no active aggregator\n",
3316 seq_printf(seq, "Active Aggregator Info:\n");
3318 seq_printf(seq, "\tAggregator ID: %d\n",
3319 ad_info.aggregator_id);
3320 seq_printf(seq, "\tNumber of ports: %d\n",
3322 seq_printf(seq, "\tActor Key: %d\n",
3324 seq_printf(seq, "\tPartner Key: %d\n",
3325 ad_info.partner_key);
3326 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3327 ad_info.partner_system);
3332 static void bond_info_show_slave(struct seq_file *seq,
3333 const struct slave *slave)
3335 struct bonding *bond = seq->private;
3337 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3338 seq_printf(seq, "MII Status: %s\n",
3339 (slave->link == BOND_LINK_UP) ? "up" : "down");
3340 seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3341 seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3342 seq_printf(seq, "Link Failure Count: %u\n",
3343 slave->link_failure_count);
3345 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3347 if (bond->params.mode == BOND_MODE_8023AD) {
3348 const struct aggregator *agg
3349 = SLAVE_AD_INFO(slave).port.aggregator;
3352 seq_printf(seq, "Aggregator ID: %d\n",
3353 agg->aggregator_identifier);
3355 seq_puts(seq, "Aggregator ID: N/A\n");
3357 seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3360 static int bond_info_seq_show(struct seq_file *seq, void *v)
3362 if (v == SEQ_START_TOKEN) {
3363 seq_printf(seq, "%s\n", version);
3364 bond_info_show_master(seq);
3366 bond_info_show_slave(seq, v);
3371 static const struct seq_operations bond_info_seq_ops = {
3372 .start = bond_info_seq_start,
3373 .next = bond_info_seq_next,
3374 .stop = bond_info_seq_stop,
3375 .show = bond_info_seq_show,
3378 static int bond_info_open(struct inode *inode, struct file *file)
3380 struct seq_file *seq;
3381 struct proc_dir_entry *proc;
3384 res = seq_open(file, &bond_info_seq_ops);
3386 /* recover the pointer buried in proc_dir_entry data */
3387 seq = file->private_data;
3389 seq->private = proc->data;
3395 static const struct file_operations bond_info_fops = {
3396 .owner = THIS_MODULE,
3397 .open = bond_info_open,
3399 .llseek = seq_lseek,
3400 .release = seq_release,
3403 static void bond_create_proc_entry(struct bonding *bond)
3405 struct net_device *bond_dev = bond->dev;
3406 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3409 bond->proc_entry = proc_create_data(bond_dev->name,
3410 S_IRUGO, bn->proc_dir,
3411 &bond_info_fops, bond);
3412 if (bond->proc_entry == NULL)
3413 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3414 DRV_NAME, bond_dev->name);
3416 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3420 static void bond_remove_proc_entry(struct bonding *bond)
3422 struct net_device *bond_dev = bond->dev;
3423 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3425 if (bn->proc_dir && bond->proc_entry) {
3426 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3427 memset(bond->proc_file_name, 0, IFNAMSIZ);
3428 bond->proc_entry = NULL;
3432 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3433 * Caller must hold rtnl_lock.
3435 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3437 if (!bn->proc_dir) {
3438 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3440 pr_warning("Warning: cannot create /proc/net/%s\n",
3445 /* Destroy the bonding directory under /proc/net, if empty.
3446 * Caller must hold rtnl_lock.
3448 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3451 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3452 bn->proc_dir = NULL;
3456 #else /* !CONFIG_PROC_FS */
3458 static void bond_create_proc_entry(struct bonding *bond)
3462 static void bond_remove_proc_entry(struct bonding *bond)
3466 static inline void bond_create_proc_dir(struct bond_net *bn)
3470 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3474 #endif /* CONFIG_PROC_FS */
3477 /*-------------------------- netdev event handling --------------------------*/
3480 * Change device name
3482 static int bond_event_changename(struct bonding *bond)
3484 bond_remove_proc_entry(bond);
3485 bond_create_proc_entry(bond);
3490 static int bond_master_netdev_event(unsigned long event,
3491 struct net_device *bond_dev)
3493 struct bonding *event_bond = netdev_priv(bond_dev);
3496 case NETDEV_CHANGENAME:
3497 return bond_event_changename(event_bond);
3505 static int bond_slave_netdev_event(unsigned long event,
3506 struct net_device *slave_dev)
3508 struct net_device *bond_dev = slave_dev->master;
3509 struct bonding *bond = netdev_priv(bond_dev);
3512 case NETDEV_UNREGISTER:
3514 if (bond->setup_by_slave)
3515 bond_release_and_destroy(bond_dev, slave_dev);
3517 bond_release(bond_dev, slave_dev);
3521 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3522 struct slave *slave;
3524 slave = bond_get_slave_by_dev(bond, slave_dev);
3526 u16 old_speed = slave->speed;
3527 u16 old_duplex = slave->duplex;
3529 bond_update_speed_duplex(slave);
3531 if (bond_is_lb(bond))
3534 if (old_speed != slave->speed)
3535 bond_3ad_adapter_speed_changed(slave);
3536 if (old_duplex != slave->duplex)
3537 bond_3ad_adapter_duplex_changed(slave);
3544 * ... Or is it this?
3547 case NETDEV_CHANGEMTU:
3549 * TODO: Should slaves be allowed to
3550 * independently alter their MTU? For
3551 * an active-backup bond, slaves need
3552 * not be the same type of device, so
3553 * MTUs may vary. For other modes,
3554 * slaves arguably should have the
3555 * same MTUs. To do this, we'd need to
3556 * take over the slave's change_mtu
3557 * function for the duration of their
3561 case NETDEV_CHANGENAME:
3563 * TODO: handle changing the primary's name
3566 case NETDEV_FEAT_CHANGE:
3567 bond_compute_features(bond);
3577 * bond_netdev_event: handle netdev notifier chain events.
3579 * This function receives events for the netdev chain. The caller (an
3580 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3581 * locks for us to safely manipulate the slave devices (RTNL lock,
3584 static int bond_netdev_event(struct notifier_block *this,
3585 unsigned long event, void *ptr)
3587 struct net_device *event_dev = (struct net_device *)ptr;
3589 pr_debug("event_dev: %s, event: %lx\n",
3590 event_dev ? event_dev->name : "None",
3593 if (!(event_dev->priv_flags & IFF_BONDING))
3596 if (event_dev->flags & IFF_MASTER) {
3597 pr_debug("IFF_MASTER\n");
3598 return bond_master_netdev_event(event, event_dev);
3601 if (event_dev->flags & IFF_SLAVE) {
3602 pr_debug("IFF_SLAVE\n");
3603 return bond_slave_netdev_event(event, event_dev);
3610 * bond_inetaddr_event: handle inetaddr notifier chain events.
3612 * We keep track of device IPs primarily to use as source addresses in
3613 * ARP monitor probes (rather than spewing out broadcasts all the time).
3615 * We track one IP for the main device (if it has one), plus one per VLAN.
3617 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3619 struct in_ifaddr *ifa = ptr;
3620 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3621 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3622 struct bonding *bond;
3623 struct vlan_entry *vlan;
3625 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3626 if (bond->dev == event_dev) {
3629 bond->master_ip = ifa->ifa_local;
3632 bond->master_ip = bond_glean_dev_ip(bond->dev);
3639 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3642 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3643 if (vlan_dev == event_dev) {
3646 vlan->vlan_ip = ifa->ifa_local;
3650 bond_glean_dev_ip(vlan_dev);
3661 static struct notifier_block bond_netdev_notifier = {
3662 .notifier_call = bond_netdev_event,
3665 static struct notifier_block bond_inetaddr_notifier = {
3666 .notifier_call = bond_inetaddr_event,
3669 /*-------------------------- Packet type handling ---------------------------*/
3671 /* register to receive lacpdus on a bond */
3672 static void bond_register_lacpdu(struct bonding *bond)
3674 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3676 /* initialize packet type */
3677 pk_type->type = PKT_TYPE_LACPDU;
3678 pk_type->dev = bond->dev;
3679 pk_type->func = bond_3ad_lacpdu_recv;
3681 dev_add_pack(pk_type);
3684 /* unregister to receive lacpdus on a bond */
3685 static void bond_unregister_lacpdu(struct bonding *bond)
3687 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3690 void bond_register_arp(struct bonding *bond)
3692 struct packet_type *pt = &bond->arp_mon_pt;
3697 pt->type = htons(ETH_P_ARP);
3698 pt->dev = bond->dev;
3699 pt->func = bond_arp_rcv;
3703 void bond_unregister_arp(struct bonding *bond)
3705 struct packet_type *pt = &bond->arp_mon_pt;
3707 dev_remove_pack(pt);
3711 /*---------------------------- Hashing Policies -----------------------------*/
3714 * Hash for the output device based upon layer 2 and layer 3 data. If
3715 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3717 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3719 struct ethhdr *data = (struct ethhdr *)skb->data;
3720 struct iphdr *iph = ip_hdr(skb);
3722 if (skb->protocol == htons(ETH_P_IP)) {
3723 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3724 (data->h_dest[5] ^ data->h_source[5])) % count;
3727 return (data->h_dest[5] ^ data->h_source[5]) % count;
3731 * Hash for the output device based upon layer 3 and layer 4 data. If
3732 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3733 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3735 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3737 struct ethhdr *data = (struct ethhdr *)skb->data;
3738 struct iphdr *iph = ip_hdr(skb);
3739 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3742 if (skb->protocol == htons(ETH_P_IP)) {
3743 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3744 (iph->protocol == IPPROTO_TCP ||
3745 iph->protocol == IPPROTO_UDP)) {
3746 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3748 return (layer4_xor ^
3749 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3753 return (data->h_dest[5] ^ data->h_source[5]) % count;
3757 * Hash for the output device based upon layer 2 data
3759 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3761 struct ethhdr *data = (struct ethhdr *)skb->data;
3763 return (data->h_dest[5] ^ data->h_source[5]) % count;
3766 /*-------------------------- Device entry points ----------------------------*/
3768 static int bond_open(struct net_device *bond_dev)
3770 struct bonding *bond = netdev_priv(bond_dev);
3772 bond->kill_timers = 0;
3774 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3776 if (bond_is_lb(bond)) {
3777 /* bond_alb_initialize must be called before the timer
3780 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3781 /* something went wrong - fail the open operation */
3785 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3786 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3789 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3790 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3791 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3794 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3795 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3796 INIT_DELAYED_WORK(&bond->arp_work,
3797 bond_activebackup_arp_mon);
3799 INIT_DELAYED_WORK(&bond->arp_work,
3800 bond_loadbalance_arp_mon);
3802 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3803 if (bond->params.arp_validate)
3804 bond_register_arp(bond);
3807 if (bond->params.mode == BOND_MODE_8023AD) {
3808 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3809 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3810 /* register to receive LACPDUs */
3811 bond_register_lacpdu(bond);
3812 bond_3ad_initiate_agg_selection(bond, 1);
3818 static int bond_close(struct net_device *bond_dev)
3820 struct bonding *bond = netdev_priv(bond_dev);
3822 if (bond->params.mode == BOND_MODE_8023AD) {
3823 /* Unregister the receive of LACPDUs */
3824 bond_unregister_lacpdu(bond);
3827 if (bond->params.arp_validate)
3828 bond_unregister_arp(bond);
3830 write_lock_bh(&bond->lock);
3832 bond->send_grat_arp = 0;
3833 bond->send_unsol_na = 0;
3835 /* signal timers not to re-arm */
3836 bond->kill_timers = 1;
3838 write_unlock_bh(&bond->lock);
3840 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3841 cancel_delayed_work(&bond->mii_work);
3844 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3845 cancel_delayed_work(&bond->arp_work);
3848 switch (bond->params.mode) {
3849 case BOND_MODE_8023AD:
3850 cancel_delayed_work(&bond->ad_work);
3854 cancel_delayed_work(&bond->alb_work);
3860 if (delayed_work_pending(&bond->mcast_work))
3861 cancel_delayed_work(&bond->mcast_work);
3863 if (bond_is_lb(bond)) {
3864 /* Must be called only after all
3865 * slaves have been released
3867 bond_alb_deinitialize(bond);
3873 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3874 struct rtnl_link_stats64 *stats)
3876 struct bonding *bond = netdev_priv(bond_dev);
3877 struct rtnl_link_stats64 temp;
3878 struct slave *slave;
3881 memset(stats, 0, sizeof(*stats));
3883 read_lock_bh(&bond->lock);
3885 bond_for_each_slave(bond, slave, i) {
3886 const struct rtnl_link_stats64 *sstats =
3887 dev_get_stats(slave->dev, &temp);
3889 stats->rx_packets += sstats->rx_packets;
3890 stats->rx_bytes += sstats->rx_bytes;
3891 stats->rx_errors += sstats->rx_errors;
3892 stats->rx_dropped += sstats->rx_dropped;
3894 stats->tx_packets += sstats->tx_packets;
3895 stats->tx_bytes += sstats->tx_bytes;
3896 stats->tx_errors += sstats->tx_errors;
3897 stats->tx_dropped += sstats->tx_dropped;
3899 stats->multicast += sstats->multicast;
3900 stats->collisions += sstats->collisions;
3902 stats->rx_length_errors += sstats->rx_length_errors;
3903 stats->rx_over_errors += sstats->rx_over_errors;
3904 stats->rx_crc_errors += sstats->rx_crc_errors;
3905 stats->rx_frame_errors += sstats->rx_frame_errors;
3906 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3907 stats->rx_missed_errors += sstats->rx_missed_errors;
3909 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3910 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3911 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3912 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3913 stats->tx_window_errors += sstats->tx_window_errors;
3916 read_unlock_bh(&bond->lock);
3921 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3923 struct net_device *slave_dev = NULL;
3924 struct ifbond k_binfo;
3925 struct ifbond __user *u_binfo = NULL;
3926 struct ifslave k_sinfo;
3927 struct ifslave __user *u_sinfo = NULL;
3928 struct mii_ioctl_data *mii = NULL;
3931 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3943 * We do this again just in case we were called by SIOCGMIIREG
3944 * instead of SIOCGMIIPHY.
3951 if (mii->reg_num == 1) {
3952 struct bonding *bond = netdev_priv(bond_dev);
3954 read_lock(&bond->lock);
3955 read_lock(&bond->curr_slave_lock);
3956 if (netif_carrier_ok(bond->dev))
3957 mii->val_out = BMSR_LSTATUS;
3959 read_unlock(&bond->curr_slave_lock);
3960 read_unlock(&bond->lock);
3964 case BOND_INFO_QUERY_OLD:
3965 case SIOCBONDINFOQUERY:
3966 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3968 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3971 res = bond_info_query(bond_dev, &k_binfo);
3973 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3977 case BOND_SLAVE_INFO_QUERY_OLD:
3978 case SIOCBONDSLAVEINFOQUERY:
3979 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3981 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3984 res = bond_slave_info_query(bond_dev, &k_sinfo);
3986 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3995 if (!capable(CAP_NET_ADMIN))
3998 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4000 pr_debug("slave_dev=%p:\n", slave_dev);
4005 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4007 case BOND_ENSLAVE_OLD:
4008 case SIOCBONDENSLAVE:
4009 res = bond_enslave(bond_dev, slave_dev);
4011 case BOND_RELEASE_OLD:
4012 case SIOCBONDRELEASE:
4013 res = bond_release(bond_dev, slave_dev);
4015 case BOND_SETHWADDR_OLD:
4016 case SIOCBONDSETHWADDR:
4017 res = bond_sethwaddr(bond_dev, slave_dev);
4019 case BOND_CHANGE_ACTIVE_OLD:
4020 case SIOCBONDCHANGEACTIVE:
4021 res = bond_ioctl_change_active(bond_dev, slave_dev);
4033 static bool bond_addr_in_mc_list(unsigned char *addr,
4034 struct netdev_hw_addr_list *list,
4037 struct netdev_hw_addr *ha;
4039 netdev_hw_addr_list_for_each(ha, list)
4040 if (!memcmp(ha->addr, addr, addrlen))
4046 static void bond_set_multicast_list(struct net_device *bond_dev)
4048 struct bonding *bond = netdev_priv(bond_dev);
4049 struct netdev_hw_addr *ha;
4053 * Do promisc before checking multicast_mode
4055 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4057 * FIXME: Need to handle the error when one of the multi-slaves
4060 bond_set_promiscuity(bond, 1);
4063 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4064 bond_set_promiscuity(bond, -1);
4067 /* set allmulti flag to slaves */
4068 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4070 * FIXME: Need to handle the error when one of the multi-slaves
4073 bond_set_allmulti(bond, 1);
4076 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4077 bond_set_allmulti(bond, -1);
4080 read_lock(&bond->lock);
4082 bond->flags = bond_dev->flags;
4084 /* looking for addresses to add to slaves' mc list */
4085 netdev_for_each_mc_addr(ha, bond_dev) {
4086 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4087 bond_dev->addr_len);
4089 bond_mc_add(bond, ha->addr);
4092 /* looking for addresses to delete from slaves' list */
4093 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4094 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4095 bond_dev->addr_len);
4097 bond_mc_del(bond, ha->addr);
4100 /* save master's multicast list */
4101 __hw_addr_flush(&bond->mc_list);
4102 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4103 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4105 read_unlock(&bond->lock);
4108 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4110 struct bonding *bond = netdev_priv(dev);
4111 struct slave *slave = bond->first_slave;
4114 const struct net_device_ops *slave_ops
4115 = slave->dev->netdev_ops;
4116 if (slave_ops->ndo_neigh_setup)
4117 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4123 * Change the MTU of all of a master's slaves to match the master
4125 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4127 struct bonding *bond = netdev_priv(bond_dev);
4128 struct slave *slave, *stop_at;
4132 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4133 (bond_dev ? bond_dev->name : "None"), new_mtu);
4135 /* Can't hold bond->lock with bh disabled here since
4136 * some base drivers panic. On the other hand we can't
4137 * hold bond->lock without bh disabled because we'll
4138 * deadlock. The only solution is to rely on the fact
4139 * that we're under rtnl_lock here, and the slaves
4140 * list won't change. This doesn't solve the problem
4141 * of setting the slave's MTU while it is
4142 * transmitting, but the assumption is that the base
4143 * driver can handle that.
4145 * TODO: figure out a way to safely iterate the slaves
4146 * list, but without holding a lock around the actual
4147 * call to the base driver.
4150 bond_for_each_slave(bond, slave, i) {
4151 pr_debug("s %p s->p %p c_m %p\n",
4154 slave->dev->netdev_ops->ndo_change_mtu);
4156 res = dev_set_mtu(slave->dev, new_mtu);
4159 /* If we failed to set the slave's mtu to the new value
4160 * we must abort the operation even in ACTIVE_BACKUP
4161 * mode, because if we allow the backup slaves to have
4162 * different mtu values than the active slave we'll
4163 * need to change their mtu when doing a failover. That
4164 * means changing their mtu from timer context, which
4165 * is probably not a good idea.
4167 pr_debug("err %d %s\n", res, slave->dev->name);
4172 bond_dev->mtu = new_mtu;
4177 /* unwind from head to the slave that failed */
4179 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4182 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4184 pr_debug("unwind err %d dev %s\n",
4185 tmp_res, slave->dev->name);
4195 * Note that many devices must be down to change the HW address, and
4196 * downing the master releases all slaves. We can make bonds full of
4197 * bonding devices to test this, however.
4199 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4201 struct bonding *bond = netdev_priv(bond_dev);
4202 struct sockaddr *sa = addr, tmp_sa;
4203 struct slave *slave, *stop_at;
4207 if (bond->params.mode == BOND_MODE_ALB)
4208 return bond_alb_set_mac_address(bond_dev, addr);
4211 pr_debug("bond=%p, name=%s\n",
4212 bond, bond_dev ? bond_dev->name : "None");
4215 * If fail_over_mac is set to active, do nothing and return
4216 * success. Returning an error causes ifenslave to fail.
4218 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4221 if (!is_valid_ether_addr(sa->sa_data))
4222 return -EADDRNOTAVAIL;
4224 /* Can't hold bond->lock with bh disabled here since
4225 * some base drivers panic. On the other hand we can't
4226 * hold bond->lock without bh disabled because we'll
4227 * deadlock. The only solution is to rely on the fact
4228 * that we're under rtnl_lock here, and the slaves
4229 * list won't change. This doesn't solve the problem
4230 * of setting the slave's hw address while it is
4231 * transmitting, but the assumption is that the base
4232 * driver can handle that.
4234 * TODO: figure out a way to safely iterate the slaves
4235 * list, but without holding a lock around the actual
4236 * call to the base driver.
4239 bond_for_each_slave(bond, slave, i) {
4240 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4241 pr_debug("slave %p %s\n", slave, slave->dev->name);
4243 if (slave_ops->ndo_set_mac_address == NULL) {
4245 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4249 res = dev_set_mac_address(slave->dev, addr);
4251 /* TODO: consider downing the slave
4253 * User should expect communications
4254 * breakage anyway until ARP finish
4257 pr_debug("err %d %s\n", res, slave->dev->name);
4263 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4267 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4268 tmp_sa.sa_family = bond_dev->type;
4270 /* unwind from head to the slave that failed */
4272 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4275 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4277 pr_debug("unwind err %d dev %s\n",
4278 tmp_res, slave->dev->name);
4285 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4287 struct bonding *bond = netdev_priv(bond_dev);
4288 struct slave *slave, *start_at;
4289 int i, slave_no, res = 1;
4290 struct iphdr *iph = ip_hdr(skb);
4292 read_lock(&bond->lock);
4294 if (!BOND_IS_OK(bond))
4297 * Start with the curr_active_slave that joined the bond as the
4298 * default for sending IGMP traffic. For failover purposes one
4299 * needs to maintain some consistency for the interface that will
4300 * send the join/membership reports. The curr_active_slave found
4301 * will send all of this type of traffic.
4303 if ((iph->protocol == IPPROTO_IGMP) &&
4304 (skb->protocol == htons(ETH_P_IP))) {
4306 read_lock(&bond->curr_slave_lock);
4307 slave = bond->curr_active_slave;
4308 read_unlock(&bond->curr_slave_lock);
4314 * Concurrent TX may collide on rr_tx_counter; we accept
4315 * that as being rare enough not to justify using an
4318 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4320 bond_for_each_slave(bond, slave, i) {
4328 bond_for_each_slave_from(bond, slave, i, start_at) {
4329 if (IS_UP(slave->dev) &&
4330 (slave->link == BOND_LINK_UP) &&
4331 (slave->state == BOND_STATE_ACTIVE)) {
4332 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4339 /* no suitable interface, frame not sent */
4342 read_unlock(&bond->lock);
4343 return NETDEV_TX_OK;
4348 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4349 * the bond has a usable interface.
4351 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4353 struct bonding *bond = netdev_priv(bond_dev);
4356 read_lock(&bond->lock);
4357 read_lock(&bond->curr_slave_lock);
4359 if (!BOND_IS_OK(bond))
4362 if (!bond->curr_active_slave)
4365 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4369 /* no suitable interface, frame not sent */
4372 read_unlock(&bond->curr_slave_lock);
4373 read_unlock(&bond->lock);
4374 return NETDEV_TX_OK;
4378 * In bond_xmit_xor() , we determine the output device by using a pre-
4379 * determined xmit_hash_policy(), If the selected device is not enabled,
4380 * find the next active slave.
4382 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4384 struct bonding *bond = netdev_priv(bond_dev);
4385 struct slave *slave, *start_at;
4390 read_lock(&bond->lock);
4392 if (!BOND_IS_OK(bond))
4395 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4397 bond_for_each_slave(bond, slave, i) {
4405 bond_for_each_slave_from(bond, slave, i, start_at) {
4406 if (IS_UP(slave->dev) &&
4407 (slave->link == BOND_LINK_UP) &&
4408 (slave->state == BOND_STATE_ACTIVE)) {
4409 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4416 /* no suitable interface, frame not sent */
4419 read_unlock(&bond->lock);
4420 return NETDEV_TX_OK;
4424 * in broadcast mode, we send everything to all usable interfaces.
4426 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4428 struct bonding *bond = netdev_priv(bond_dev);
4429 struct slave *slave, *start_at;
4430 struct net_device *tx_dev = NULL;
4434 read_lock(&bond->lock);
4436 if (!BOND_IS_OK(bond))
4439 read_lock(&bond->curr_slave_lock);
4440 start_at = bond->curr_active_slave;
4441 read_unlock(&bond->curr_slave_lock);
4446 bond_for_each_slave_from(bond, slave, i, start_at) {
4447 if (IS_UP(slave->dev) &&
4448 (slave->link == BOND_LINK_UP) &&
4449 (slave->state == BOND_STATE_ACTIVE)) {
4451 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4453 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4458 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4460 dev_kfree_skb(skb2);
4464 tx_dev = slave->dev;
4469 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4473 /* no suitable interface, frame not sent */
4476 /* frame sent to all suitable interfaces */
4477 read_unlock(&bond->lock);
4478 return NETDEV_TX_OK;
4481 /*------------------------- Device initialization ---------------------------*/
4483 static void bond_set_xmit_hash_policy(struct bonding *bond)
4485 switch (bond->params.xmit_policy) {
4486 case BOND_XMIT_POLICY_LAYER23:
4487 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4489 case BOND_XMIT_POLICY_LAYER34:
4490 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4492 case BOND_XMIT_POLICY_LAYER2:
4494 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4500 * Lookup the slave that corresponds to a qid
4502 static inline int bond_slave_override(struct bonding *bond,
4503 struct sk_buff *skb)
4506 struct slave *slave = NULL;
4507 struct slave *check_slave;
4509 read_lock(&bond->lock);
4511 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4514 /* Find out if any slaves have the same mapping as this skb. */
4515 bond_for_each_slave(bond, check_slave, i) {
4516 if (check_slave->queue_id == skb->queue_mapping) {
4517 slave = check_slave;
4522 /* If the slave isn't UP, use default transmit policy. */
4523 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4524 (slave->link == BOND_LINK_UP)) {
4525 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4529 read_unlock(&bond->lock);
4533 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4536 * This helper function exists to help dev_pick_tx get the correct
4537 * destination queue. Using a helper function skips the a call to
4538 * skb_tx_hash and will put the skbs in the queue we expect on their
4539 * way down to the bonding driver.
4541 return skb->queue_mapping;
4544 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4546 struct bonding *bond = netdev_priv(dev);
4549 * If we risk deadlock from transmitting this in the
4550 * netpoll path, tell netpoll to queue the frame for later tx
4552 if (is_netpoll_tx_blocked(dev))
4553 return NETDEV_TX_BUSY;
4555 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4556 if (!bond_slave_override(bond, skb))
4557 return NETDEV_TX_OK;
4560 switch (bond->params.mode) {
4561 case BOND_MODE_ROUNDROBIN:
4562 return bond_xmit_roundrobin(skb, dev);
4563 case BOND_MODE_ACTIVEBACKUP:
4564 return bond_xmit_activebackup(skb, dev);
4566 return bond_xmit_xor(skb, dev);
4567 case BOND_MODE_BROADCAST:
4568 return bond_xmit_broadcast(skb, dev);
4569 case BOND_MODE_8023AD:
4570 return bond_3ad_xmit_xor(skb, dev);
4573 return bond_alb_xmit(skb, dev);
4575 /* Should never happen, mode already checked */
4576 pr_err("%s: Error: Unknown bonding mode %d\n",
4577 dev->name, bond->params.mode);
4580 return NETDEV_TX_OK;
4586 * set bond mode specific net device operations
4588 void bond_set_mode_ops(struct bonding *bond, int mode)
4590 struct net_device *bond_dev = bond->dev;
4593 case BOND_MODE_ROUNDROBIN:
4595 case BOND_MODE_ACTIVEBACKUP:
4598 bond_set_xmit_hash_policy(bond);
4600 case BOND_MODE_BROADCAST:
4602 case BOND_MODE_8023AD:
4603 bond_set_master_3ad_flags(bond);
4604 bond_set_xmit_hash_policy(bond);
4607 bond_set_master_alb_flags(bond);
4612 /* Should never happen, mode already checked */
4613 pr_err("%s: Error: Unknown bonding mode %d\n",
4614 bond_dev->name, mode);
4619 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4620 struct ethtool_drvinfo *drvinfo)
4622 strncpy(drvinfo->driver, DRV_NAME, 32);
4623 strncpy(drvinfo->version, DRV_VERSION, 32);
4624 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4627 static const struct ethtool_ops bond_ethtool_ops = {
4628 .get_drvinfo = bond_ethtool_get_drvinfo,
4629 .get_link = ethtool_op_get_link,
4630 .get_tx_csum = ethtool_op_get_tx_csum,
4631 .get_sg = ethtool_op_get_sg,
4632 .get_tso = ethtool_op_get_tso,
4633 .get_ufo = ethtool_op_get_ufo,
4634 .get_flags = ethtool_op_get_flags,
4637 static const struct net_device_ops bond_netdev_ops = {
4638 .ndo_init = bond_init,
4639 .ndo_uninit = bond_uninit,
4640 .ndo_open = bond_open,
4641 .ndo_stop = bond_close,
4642 .ndo_start_xmit = bond_start_xmit,
4643 .ndo_select_queue = bond_select_queue,
4644 .ndo_get_stats64 = bond_get_stats,
4645 .ndo_do_ioctl = bond_do_ioctl,
4646 .ndo_set_multicast_list = bond_set_multicast_list,
4647 .ndo_change_mtu = bond_change_mtu,
4648 .ndo_set_mac_address = bond_set_mac_address,
4649 .ndo_neigh_setup = bond_neigh_setup,
4650 .ndo_vlan_rx_register = bond_vlan_rx_register,
4651 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4652 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4653 #ifdef CONFIG_NET_POLL_CONTROLLER
4654 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4655 .ndo_poll_controller = bond_poll_controller,
4659 static void bond_destructor(struct net_device *bond_dev)
4661 struct bonding *bond = netdev_priv(bond_dev);
4663 destroy_workqueue(bond->wq);
4664 free_netdev(bond_dev);
4667 static void bond_setup(struct net_device *bond_dev)
4669 struct bonding *bond = netdev_priv(bond_dev);
4671 /* initialize rwlocks */
4672 rwlock_init(&bond->lock);
4673 rwlock_init(&bond->curr_slave_lock);
4675 bond->params = bonding_defaults;
4677 /* Initialize pointers */
4678 bond->dev = bond_dev;
4679 INIT_LIST_HEAD(&bond->vlan_list);
4681 /* Initialize the device entry points */
4682 ether_setup(bond_dev);
4683 bond_dev->netdev_ops = &bond_netdev_ops;
4684 bond_dev->ethtool_ops = &bond_ethtool_ops;
4685 bond_set_mode_ops(bond, bond->params.mode);
4687 bond_dev->destructor = bond_destructor;
4689 /* Initialize the device options */
4690 bond_dev->tx_queue_len = 0;
4691 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4692 bond_dev->priv_flags |= IFF_BONDING;
4693 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4695 if (bond->params.arp_interval)
4696 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4698 /* At first, we block adding VLANs. That's the only way to
4699 * prevent problems that occur when adding VLANs over an
4700 * empty bond. The block will be removed once non-challenged
4701 * slaves are enslaved.
4703 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4705 /* don't acquire bond device's netif_tx_lock when
4707 bond_dev->features |= NETIF_F_LLTX;
4709 /* By default, we declare the bond to be fully
4710 * VLAN hardware accelerated capable. Special
4711 * care is taken in the various xmit functions
4712 * when there are slaves that are not hw accel
4715 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4716 NETIF_F_HW_VLAN_RX |
4717 NETIF_F_HW_VLAN_FILTER);
4719 /* By default, we enable GRO on bonding devices.
4720 * Actual support requires lowlevel drivers are GRO ready.
4722 bond_dev->features |= NETIF_F_GRO;
4725 static void bond_work_cancel_all(struct bonding *bond)
4727 write_lock_bh(&bond->lock);
4728 bond->kill_timers = 1;
4729 write_unlock_bh(&bond->lock);
4731 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4732 cancel_delayed_work(&bond->mii_work);
4734 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4735 cancel_delayed_work(&bond->arp_work);
4737 if (bond->params.mode == BOND_MODE_ALB &&
4738 delayed_work_pending(&bond->alb_work))
4739 cancel_delayed_work(&bond->alb_work);
4741 if (bond->params.mode == BOND_MODE_8023AD &&
4742 delayed_work_pending(&bond->ad_work))
4743 cancel_delayed_work(&bond->ad_work);
4745 if (delayed_work_pending(&bond->mcast_work))
4746 cancel_delayed_work(&bond->mcast_work);
4750 * Destroy a bonding device.
4751 * Must be under rtnl_lock when this function is called.
4753 static void bond_uninit(struct net_device *bond_dev)
4755 struct bonding *bond = netdev_priv(bond_dev);
4756 struct vlan_entry *vlan, *tmp;
4758 bond_netpoll_cleanup(bond_dev);
4760 /* Release the bonded slaves */
4761 bond_release_all(bond_dev);
4763 list_del(&bond->bond_list);
4765 bond_work_cancel_all(bond);
4767 bond_remove_proc_entry(bond);
4769 __hw_addr_flush(&bond->mc_list);
4771 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4772 list_del(&vlan->vlan_list);
4777 /*------------------------- Module initialization ---------------------------*/
4780 * Convert string input module parms. Accept either the
4781 * number of the mode or its string name. A bit complicated because
4782 * some mode names are substrings of other names, and calls from sysfs
4783 * may have whitespace in the name (trailing newlines, for example).
4785 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4787 int modeint = -1, i, rv;
4788 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4790 for (p = (char *)buf; *p; p++)
4791 if (!(isdigit(*p) || isspace(*p)))
4795 rv = sscanf(buf, "%20s", modestr);
4797 rv = sscanf(buf, "%d", &modeint);
4802 for (i = 0; tbl[i].modename; i++) {
4803 if (modeint == tbl[i].mode)
4805 if (strcmp(modestr, tbl[i].modename) == 0)
4812 static int bond_check_params(struct bond_params *params)
4814 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4817 * Convert string parameters.
4820 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4821 if (bond_mode == -1) {
4822 pr_err("Error: Invalid bonding mode \"%s\"\n",
4823 mode == NULL ? "NULL" : mode);
4828 if (xmit_hash_policy) {
4829 if ((bond_mode != BOND_MODE_XOR) &&
4830 (bond_mode != BOND_MODE_8023AD)) {
4831 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4832 bond_mode_name(bond_mode));
4834 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4836 if (xmit_hashtype == -1) {
4837 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4838 xmit_hash_policy == NULL ? "NULL" :
4846 if (bond_mode != BOND_MODE_8023AD) {
4847 pr_info("lacp_rate param is irrelevant in mode %s\n",
4848 bond_mode_name(bond_mode));
4850 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4851 if (lacp_fast == -1) {
4852 pr_err("Error: Invalid lacp rate \"%s\"\n",
4853 lacp_rate == NULL ? "NULL" : lacp_rate);
4860 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4861 if (params->ad_select == -1) {
4862 pr_err("Error: Invalid ad_select \"%s\"\n",
4863 ad_select == NULL ? "NULL" : ad_select);
4867 if (bond_mode != BOND_MODE_8023AD) {
4868 pr_warning("ad_select param only affects 802.3ad mode\n");
4871 params->ad_select = BOND_AD_STABLE;
4874 if (max_bonds < 0) {
4875 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4876 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4877 max_bonds = BOND_DEFAULT_MAX_BONDS;
4881 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4882 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4883 miimon = BOND_LINK_MON_INTERV;
4887 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4892 if (downdelay < 0) {
4893 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4894 downdelay, INT_MAX);
4898 if ((use_carrier != 0) && (use_carrier != 1)) {
4899 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4904 if (num_grat_arp < 0 || num_grat_arp > 255) {
4905 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4910 if (num_unsol_na < 0 || num_unsol_na > 255) {
4911 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4916 /* reset values for 802.3ad */
4917 if (bond_mode == BOND_MODE_8023AD) {
4919 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");
4920 pr_warning("Forcing miimon to 100msec\n");
4925 if (tx_queues < 1 || tx_queues > 255) {
4926 pr_warning("Warning: tx_queues (%d) should be between "
4927 "1 and 255, resetting to %d\n",
4928 tx_queues, BOND_DEFAULT_TX_QUEUES);
4929 tx_queues = BOND_DEFAULT_TX_QUEUES;
4932 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4933 pr_warning("Warning: all_slaves_active module parameter (%d), "
4934 "not of valid value (0/1), so it was set to "
4935 "0\n", all_slaves_active);
4936 all_slaves_active = 0;
4939 if (resend_igmp < 0 || resend_igmp > 255) {
4940 pr_warning("Warning: resend_igmp (%d) should be between "
4941 "0 and 255, resetting to %d\n",
4942 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4943 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4946 /* reset values for TLB/ALB */
4947 if ((bond_mode == BOND_MODE_TLB) ||
4948 (bond_mode == BOND_MODE_ALB)) {
4950 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");
4951 pr_warning("Forcing miimon to 100msec\n");
4956 if (bond_mode == BOND_MODE_ALB) {
4957 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",
4962 if (updelay || downdelay) {
4963 /* just warn the user the up/down delay will have
4964 * no effect since miimon is zero...
4966 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",
4967 updelay, downdelay);
4970 /* don't allow arp monitoring */
4972 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4973 miimon, arp_interval);
4977 if ((updelay % miimon) != 0) {
4978 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4980 (updelay / miimon) * miimon);
4985 if ((downdelay % miimon) != 0) {
4986 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4988 (downdelay / miimon) * miimon);
4991 downdelay /= miimon;
4994 if (arp_interval < 0) {
4995 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4996 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4997 arp_interval = BOND_LINK_ARP_INTERV;
5000 for (arp_ip_count = 0;
5001 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5003 /* not complete check, but should be good enough to
5005 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5006 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5007 arp_ip_target[arp_ip_count]);
5010 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5011 arp_target[arp_ip_count] = ip;
5015 if (arp_interval && !arp_ip_count) {
5016 /* don't allow arping if no arp_ip_target given... */
5017 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5023 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5024 pr_err("arp_validate only supported in active-backup mode\n");
5027 if (!arp_interval) {
5028 pr_err("arp_validate requires arp_interval\n");
5032 arp_validate_value = bond_parse_parm(arp_validate,
5034 if (arp_validate_value == -1) {
5035 pr_err("Error: invalid arp_validate \"%s\"\n",
5036 arp_validate == NULL ? "NULL" : arp_validate);
5040 arp_validate_value = 0;
5043 pr_info("MII link monitoring set to %d ms\n", miimon);
5044 } else if (arp_interval) {
5047 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5049 arp_validate_tbl[arp_validate_value].modename,
5052 for (i = 0; i < arp_ip_count; i++)
5053 pr_info(" %s", arp_ip_target[i]);
5057 } else if (max_bonds) {
5058 /* miimon and arp_interval not set, we need one so things
5059 * work as expected, see bonding.txt for details
5061 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");
5064 if (primary && !USES_PRIMARY(bond_mode)) {
5065 /* currently, using a primary only makes sense
5066 * in active backup, TLB or ALB modes
5068 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5069 primary, bond_mode_name(bond_mode));
5073 if (primary && primary_reselect) {
5074 primary_reselect_value = bond_parse_parm(primary_reselect,
5076 if (primary_reselect_value == -1) {
5077 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5079 NULL ? "NULL" : primary_reselect);
5083 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5086 if (fail_over_mac) {
5087 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5089 if (fail_over_mac_value == -1) {
5090 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5091 arp_validate == NULL ? "NULL" : arp_validate);
5095 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5096 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5098 fail_over_mac_value = BOND_FOM_NONE;
5101 /* fill params struct with the proper values */
5102 params->mode = bond_mode;
5103 params->xmit_policy = xmit_hashtype;
5104 params->miimon = miimon;
5105 params->num_grat_arp = num_grat_arp;
5106 params->num_unsol_na = num_unsol_na;
5107 params->arp_interval = arp_interval;
5108 params->arp_validate = arp_validate_value;
5109 params->updelay = updelay;
5110 params->downdelay = downdelay;
5111 params->use_carrier = use_carrier;
5112 params->lacp_fast = lacp_fast;
5113 params->primary[0] = 0;
5114 params->primary_reselect = primary_reselect_value;
5115 params->fail_over_mac = fail_over_mac_value;
5116 params->tx_queues = tx_queues;
5117 params->all_slaves_active = all_slaves_active;
5118 params->resend_igmp = resend_igmp;
5121 strncpy(params->primary, primary, IFNAMSIZ);
5122 params->primary[IFNAMSIZ - 1] = 0;
5125 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5130 static struct lock_class_key bonding_netdev_xmit_lock_key;
5131 static struct lock_class_key bonding_netdev_addr_lock_key;
5133 static void bond_set_lockdep_class_one(struct net_device *dev,
5134 struct netdev_queue *txq,
5137 lockdep_set_class(&txq->_xmit_lock,
5138 &bonding_netdev_xmit_lock_key);
5141 static void bond_set_lockdep_class(struct net_device *dev)
5143 lockdep_set_class(&dev->addr_list_lock,
5144 &bonding_netdev_addr_lock_key);
5145 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5149 * Called from registration process
5151 static int bond_init(struct net_device *bond_dev)
5153 struct bonding *bond = netdev_priv(bond_dev);
5154 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5156 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5158 bond->wq = create_singlethread_workqueue(bond_dev->name);
5162 bond_set_lockdep_class(bond_dev);
5164 netif_carrier_off(bond_dev);
5166 bond_create_proc_entry(bond);
5167 list_add_tail(&bond->bond_list, &bn->dev_list);
5169 bond_prepare_sysfs_group(bond);
5171 __hw_addr_init(&bond->mc_list);
5175 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5177 if (tb[IFLA_ADDRESS]) {
5178 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5180 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5181 return -EADDRNOTAVAIL;
5186 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5188 .priv_size = sizeof(struct bonding),
5189 .setup = bond_setup,
5190 .validate = bond_validate,
5193 /* Create a new bond based on the specified name and bonding parameters.
5194 * If name is NULL, obtain a suitable "bond%d" name for us.
5195 * Caller must NOT hold rtnl_lock; we need to release it here before we
5196 * set up our sysfs entries.
5198 int bond_create(struct net *net, const char *name)
5200 struct net_device *bond_dev;
5205 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5206 bond_setup, tx_queues);
5208 pr_err("%s: eek! can't alloc netdev!\n", name);
5213 dev_net_set(bond_dev, net);
5214 bond_dev->rtnl_link_ops = &bond_link_ops;
5217 res = dev_alloc_name(bond_dev, "bond%d");
5222 * If we're given a name to register
5223 * we need to ensure that its not already
5227 if (__dev_get_by_name(net, name) != NULL)
5231 res = register_netdevice(bond_dev);
5236 bond_destructor(bond_dev);
5240 static int __net_init bond_net_init(struct net *net)
5242 struct bond_net *bn = net_generic(net, bond_net_id);
5245 INIT_LIST_HEAD(&bn->dev_list);
5247 bond_create_proc_dir(bn);
5252 static void __net_exit bond_net_exit(struct net *net)
5254 struct bond_net *bn = net_generic(net, bond_net_id);
5256 bond_destroy_proc_dir(bn);
5259 static struct pernet_operations bond_net_ops = {
5260 .init = bond_net_init,
5261 .exit = bond_net_exit,
5263 .size = sizeof(struct bond_net),
5266 static int __init bonding_init(void)
5271 pr_info("%s", version);
5273 res = bond_check_params(&bonding_defaults);
5277 res = register_pernet_subsys(&bond_net_ops);
5281 res = rtnl_link_register(&bond_link_ops);
5285 for (i = 0; i < max_bonds; i++) {
5286 res = bond_create(&init_net, NULL);
5291 res = bond_create_sysfs();
5296 register_netdevice_notifier(&bond_netdev_notifier);
5297 register_inetaddr_notifier(&bond_inetaddr_notifier);
5298 bond_register_ipv6_notifier();
5302 rtnl_link_unregister(&bond_link_ops);
5304 unregister_pernet_subsys(&bond_net_ops);
5309 static void __exit bonding_exit(void)
5311 unregister_netdevice_notifier(&bond_netdev_notifier);
5312 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5313 bond_unregister_ipv6_notifier();
5315 bond_destroy_sysfs();
5317 rtnl_link_unregister(&bond_link_ops);
5318 unregister_pernet_subsys(&bond_net_ops);
5320 #ifdef CONFIG_NET_POLL_CONTROLLER
5322 * Make sure we don't have an imbalance on our netpoll blocking
5324 WARN_ON(atomic_read(&netpoll_block_tx));
5328 module_init(bonding_init);
5329 module_exit(bonding_exit);
5330 MODULE_LICENSE("GPL");
5331 MODULE_VERSION(DRV_VERSION);
5332 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5333 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5334 MODULE_ALIAS_RTNL_LINK("bond");