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 if (netif_running(bond->dev)) {
1182 bond->send_grat_arp = bond->params.num_grat_arp;
1183 bond_send_gratuitous_arp(bond);
1185 bond->send_unsol_na = bond->params.num_unsol_na;
1186 bond_send_unsolicited_na(bond);
1189 write_unlock_bh(&bond->curr_slave_lock);
1190 read_unlock(&bond->lock);
1192 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1194 read_lock(&bond->lock);
1195 write_lock_bh(&bond->curr_slave_lock);
1199 /* resend IGMP joins since active slave has changed or
1200 * all were sent on curr_active_slave */
1201 if (((USES_PRIMARY(bond->params.mode) && new_active) ||
1202 bond->params.mode == BOND_MODE_ROUNDROBIN) &&
1203 netif_running(bond->dev)) {
1204 bond->igmp_retrans = bond->params.resend_igmp;
1205 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1210 * bond_select_active_slave - select a new active slave, if needed
1211 * @bond: our bonding struct
1213 * This functions should be called when one of the following occurs:
1214 * - The old curr_active_slave has been released or lost its link.
1215 * - The primary_slave has got its link back.
1216 * - A slave has got its link back and there's no old curr_active_slave.
1218 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1220 void bond_select_active_slave(struct bonding *bond)
1222 struct slave *best_slave;
1225 best_slave = bond_find_best_slave(bond);
1226 if (best_slave != bond->curr_active_slave) {
1227 bond_change_active_slave(bond, best_slave);
1228 rv = bond_set_carrier(bond);
1232 if (netif_carrier_ok(bond->dev)) {
1233 pr_info("%s: first active interface up!\n",
1236 pr_info("%s: now running without any active interface !\n",
1242 /*--------------------------- slave list handling ---------------------------*/
1245 * This function attaches the slave to the end of list.
1247 * bond->lock held for writing by caller.
1249 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1251 if (bond->first_slave == NULL) { /* attaching the first slave */
1252 new_slave->next = new_slave;
1253 new_slave->prev = new_slave;
1254 bond->first_slave = new_slave;
1256 new_slave->next = bond->first_slave;
1257 new_slave->prev = bond->first_slave->prev;
1258 new_slave->next->prev = new_slave;
1259 new_slave->prev->next = new_slave;
1266 * This function detaches the slave from the list.
1267 * WARNING: no check is made to verify if the slave effectively
1268 * belongs to <bond>.
1269 * Nothing is freed on return, structures are just unchained.
1270 * If any slave pointer in bond was pointing to <slave>,
1271 * it should be changed by the calling function.
1273 * bond->lock held for writing by caller.
1275 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1278 slave->next->prev = slave->prev;
1281 slave->prev->next = slave->next;
1283 if (bond->first_slave == slave) { /* slave is the first slave */
1284 if (bond->slave_cnt > 1) { /* there are more slave */
1285 bond->first_slave = slave->next;
1287 bond->first_slave = NULL; /* slave was the last one */
1296 #ifdef CONFIG_NET_POLL_CONTROLLER
1298 * You must hold read lock on bond->lock before calling this.
1300 static bool slaves_support_netpoll(struct net_device *bond_dev)
1302 struct bonding *bond = netdev_priv(bond_dev);
1303 struct slave *slave;
1307 bond_for_each_slave(bond, slave, i) {
1308 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1309 !slave->dev->netdev_ops->ndo_poll_controller)
1312 return i != 0 && ret;
1315 static void bond_poll_controller(struct net_device *bond_dev)
1317 struct bonding *bond = netdev_priv(bond_dev);
1318 struct slave *slave;
1321 bond_for_each_slave(bond, slave, i) {
1322 if (slave->dev && IS_UP(slave->dev))
1323 netpoll_poll_dev(slave->dev);
1327 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1329 struct bonding *bond = netdev_priv(bond_dev);
1330 struct slave *slave;
1331 const struct net_device_ops *ops;
1334 read_lock(&bond->lock);
1335 bond_dev->npinfo = NULL;
1336 bond_for_each_slave(bond, slave, i) {
1338 ops = slave->dev->netdev_ops;
1339 if (ops->ndo_netpoll_cleanup)
1340 ops->ndo_netpoll_cleanup(slave->dev);
1342 slave->dev->npinfo = NULL;
1345 read_unlock(&bond->lock);
1350 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1356 /*---------------------------------- IOCTL ----------------------------------*/
1358 static int bond_sethwaddr(struct net_device *bond_dev,
1359 struct net_device *slave_dev)
1361 pr_debug("bond_dev=%p\n", bond_dev);
1362 pr_debug("slave_dev=%p\n", slave_dev);
1363 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1364 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1368 #define BOND_VLAN_FEATURES \
1369 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1370 NETIF_F_HW_VLAN_FILTER)
1373 * Compute the common dev->feature set available to all slaves. Some
1374 * feature bits are managed elsewhere, so preserve those feature bits
1375 * on the master device.
1377 static int bond_compute_features(struct bonding *bond)
1379 struct slave *slave;
1380 struct net_device *bond_dev = bond->dev;
1381 unsigned long features = bond_dev->features;
1382 unsigned long vlan_features = 0;
1383 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1384 bond_dev->hard_header_len);
1387 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1388 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1390 if (!bond->first_slave)
1393 features &= ~NETIF_F_ONE_FOR_ALL;
1395 vlan_features = bond->first_slave->dev->vlan_features;
1396 bond_for_each_slave(bond, slave, i) {
1397 features = netdev_increment_features(features,
1398 slave->dev->features,
1399 NETIF_F_ONE_FOR_ALL);
1400 vlan_features = netdev_increment_features(vlan_features,
1401 slave->dev->vlan_features,
1402 NETIF_F_ONE_FOR_ALL);
1403 if (slave->dev->hard_header_len > max_hard_header_len)
1404 max_hard_header_len = slave->dev->hard_header_len;
1408 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1409 bond_dev->features = netdev_fix_features(features, NULL);
1410 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1411 bond_dev->hard_header_len = max_hard_header_len;
1416 static void bond_setup_by_slave(struct net_device *bond_dev,
1417 struct net_device *slave_dev)
1419 struct bonding *bond = netdev_priv(bond_dev);
1421 bond_dev->header_ops = slave_dev->header_ops;
1423 bond_dev->type = slave_dev->type;
1424 bond_dev->hard_header_len = slave_dev->hard_header_len;
1425 bond_dev->addr_len = slave_dev->addr_len;
1427 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1428 slave_dev->addr_len);
1429 bond->setup_by_slave = 1;
1432 /* enslave device <slave> to bond device <master> */
1433 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1435 struct bonding *bond = netdev_priv(bond_dev);
1436 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1437 struct slave *new_slave = NULL;
1438 struct netdev_hw_addr *ha;
1439 struct sockaddr addr;
1441 int old_features = bond_dev->features;
1444 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1445 slave_ops->ndo_do_ioctl == NULL) {
1446 pr_warning("%s: Warning: no link monitoring support for %s\n",
1447 bond_dev->name, slave_dev->name);
1450 /* bond must be initialized by bond_open() before enslaving */
1451 if (!(bond_dev->flags & IFF_UP)) {
1452 pr_warning("%s: master_dev is not up in bond_enslave\n",
1456 /* already enslaved */
1457 if (slave_dev->flags & IFF_SLAVE) {
1458 pr_debug("Error, Device was already enslaved\n");
1462 /* vlan challenged mutual exclusion */
1463 /* no need to lock since we're protected by rtnl_lock */
1464 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1465 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1467 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1468 bond_dev->name, slave_dev->name, bond_dev->name);
1471 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1472 bond_dev->name, slave_dev->name,
1473 slave_dev->name, bond_dev->name);
1474 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1477 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1478 if (bond->slave_cnt == 0) {
1479 /* First slave, and it is not VLAN challenged,
1480 * so remove the block of adding VLANs over the bond.
1482 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1487 * Old ifenslave binaries are no longer supported. These can
1488 * be identified with moderate accuracy by the state of the slave:
1489 * the current ifenslave will set the interface down prior to
1490 * enslaving it; the old ifenslave will not.
1492 if ((slave_dev->flags & IFF_UP)) {
1493 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1496 goto err_undo_flags;
1499 /* set bonding device ether type by slave - bonding netdevices are
1500 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1501 * there is a need to override some of the type dependent attribs/funcs.
1503 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1504 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1506 if (bond->slave_cnt == 0) {
1507 if (bond_dev->type != slave_dev->type) {
1508 pr_debug("%s: change device type from %d to %d\n",
1510 bond_dev->type, slave_dev->type);
1512 res = netdev_bonding_change(bond_dev,
1513 NETDEV_PRE_TYPE_CHANGE);
1514 res = notifier_to_errno(res);
1516 pr_err("%s: refused to change device type\n",
1519 goto err_undo_flags;
1522 /* Flush unicast and multicast addresses */
1523 dev_uc_flush(bond_dev);
1524 dev_mc_flush(bond_dev);
1526 if (slave_dev->type != ARPHRD_ETHER)
1527 bond_setup_by_slave(bond_dev, slave_dev);
1529 ether_setup(bond_dev);
1531 netdev_bonding_change(bond_dev,
1532 NETDEV_POST_TYPE_CHANGE);
1534 } else if (bond_dev->type != slave_dev->type) {
1535 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1537 slave_dev->type, bond_dev->type);
1539 goto err_undo_flags;
1542 if (slave_ops->ndo_set_mac_address == NULL) {
1543 if (bond->slave_cnt == 0) {
1544 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1546 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1547 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1548 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",
1551 goto err_undo_flags;
1555 /* If this is the first slave, then we need to set the master's hardware
1556 * address to be the same as the slave's. */
1557 if (is_zero_ether_addr(bond->dev->dev_addr))
1558 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1559 slave_dev->addr_len);
1562 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1565 goto err_undo_flags;
1569 * Set the new_slave's queue_id to be zero. Queue ID mapping
1570 * is set via sysfs or module option if desired.
1572 new_slave->queue_id = 0;
1574 /* Save slave's original mtu and then set it to match the bond */
1575 new_slave->original_mtu = slave_dev->mtu;
1576 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1578 pr_debug("Error %d calling dev_set_mtu\n", res);
1583 * Save slave's original ("permanent") mac address for modes
1584 * that need it, and for restoring it upon release, and then
1585 * set it to the master's address
1587 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1589 if (!bond->params.fail_over_mac) {
1591 * Set slave to master's mac address. The application already
1592 * set the master's mac address to that of the first slave
1594 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1595 addr.sa_family = slave_dev->type;
1596 res = dev_set_mac_address(slave_dev, &addr);
1598 pr_debug("Error %d calling set_mac_address\n", res);
1599 goto err_restore_mtu;
1603 res = netdev_set_master(slave_dev, bond_dev);
1605 pr_debug("Error %d calling netdev_set_master\n", res);
1606 goto err_restore_mac;
1608 /* open the slave since the application closed it */
1609 res = dev_open(slave_dev);
1611 pr_debug("Opening slave %s failed\n", slave_dev->name);
1612 goto err_unset_master;
1615 new_slave->dev = slave_dev;
1616 slave_dev->priv_flags |= IFF_BONDING;
1618 if (bond_is_lb(bond)) {
1619 /* bond_alb_init_slave() must be called before all other stages since
1620 * it might fail and we do not want to have to undo everything
1622 res = bond_alb_init_slave(bond, new_slave);
1627 /* If the mode USES_PRIMARY, then the new slave gets the
1628 * master's promisc (and mc) settings only if it becomes the
1629 * curr_active_slave, and that is taken care of later when calling
1630 * bond_change_active()
1632 if (!USES_PRIMARY(bond->params.mode)) {
1633 /* set promiscuity level to new slave */
1634 if (bond_dev->flags & IFF_PROMISC) {
1635 res = dev_set_promiscuity(slave_dev, 1);
1640 /* set allmulti level to new slave */
1641 if (bond_dev->flags & IFF_ALLMULTI) {
1642 res = dev_set_allmulti(slave_dev, 1);
1647 netif_addr_lock_bh(bond_dev);
1648 /* upload master's mc_list to new slave */
1649 netdev_for_each_mc_addr(ha, bond_dev)
1650 dev_mc_add(slave_dev, ha->addr);
1651 netif_addr_unlock_bh(bond_dev);
1654 if (bond->params.mode == BOND_MODE_8023AD) {
1655 /* add lacpdu mc addr to mc list */
1656 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1658 dev_mc_add(slave_dev, lacpdu_multicast);
1661 bond_add_vlans_on_slave(bond, slave_dev);
1663 write_lock_bh(&bond->lock);
1665 bond_attach_slave(bond, new_slave);
1667 new_slave->delay = 0;
1668 new_slave->link_failure_count = 0;
1670 bond_compute_features(bond);
1672 write_unlock_bh(&bond->lock);
1674 read_lock(&bond->lock);
1676 new_slave->last_arp_rx = jiffies;
1678 if (bond->params.miimon && !bond->params.use_carrier) {
1679 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1681 if ((link_reporting == -1) && !bond->params.arp_interval) {
1683 * miimon is set but a bonded network driver
1684 * does not support ETHTOOL/MII and
1685 * arp_interval is not set. Note: if
1686 * use_carrier is enabled, we will never go
1687 * here (because netif_carrier is always
1688 * supported); thus, we don't need to change
1689 * the messages for netif_carrier.
1691 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",
1692 bond_dev->name, slave_dev->name);
1693 } else if (link_reporting == -1) {
1694 /* unable get link status using mii/ethtool */
1695 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",
1696 bond_dev->name, slave_dev->name);
1700 /* check for initial state */
1701 if (!bond->params.miimon ||
1702 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1703 if (bond->params.updelay) {
1704 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1705 new_slave->link = BOND_LINK_BACK;
1706 new_slave->delay = bond->params.updelay;
1708 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1709 new_slave->link = BOND_LINK_UP;
1711 new_slave->jiffies = jiffies;
1713 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1714 new_slave->link = BOND_LINK_DOWN;
1717 if (bond_update_speed_duplex(new_slave) &&
1718 (new_slave->link != BOND_LINK_DOWN)) {
1719 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1720 bond_dev->name, new_slave->dev->name);
1722 if (bond->params.mode == BOND_MODE_8023AD) {
1723 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1728 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1729 /* if there is a primary slave, remember it */
1730 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1731 bond->primary_slave = new_slave;
1732 bond->force_primary = true;
1736 write_lock_bh(&bond->curr_slave_lock);
1738 switch (bond->params.mode) {
1739 case BOND_MODE_ACTIVEBACKUP:
1740 bond_set_slave_inactive_flags(new_slave);
1741 bond_select_active_slave(bond);
1743 case BOND_MODE_8023AD:
1744 /* in 802.3ad mode, the internal mechanism
1745 * will activate the slaves in the selected
1748 bond_set_slave_inactive_flags(new_slave);
1749 /* if this is the first slave */
1750 if (bond->slave_cnt == 1) {
1751 SLAVE_AD_INFO(new_slave).id = 1;
1752 /* Initialize AD with the number of times that the AD timer is called in 1 second
1753 * can be called only after the mac address of the bond is set
1755 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1756 bond->params.lacp_fast);
1758 SLAVE_AD_INFO(new_slave).id =
1759 SLAVE_AD_INFO(new_slave->prev).id + 1;
1762 bond_3ad_bind_slave(new_slave);
1766 new_slave->state = BOND_STATE_ACTIVE;
1767 bond_set_slave_inactive_flags(new_slave);
1768 bond_select_active_slave(bond);
1771 pr_debug("This slave is always active in trunk mode\n");
1773 /* always active in trunk mode */
1774 new_slave->state = BOND_STATE_ACTIVE;
1776 /* In trunking mode there is little meaning to curr_active_slave
1777 * anyway (it holds no special properties of the bond device),
1778 * so we can change it without calling change_active_interface()
1780 if (!bond->curr_active_slave)
1781 bond->curr_active_slave = new_slave;
1784 } /* switch(bond_mode) */
1786 write_unlock_bh(&bond->curr_slave_lock);
1788 bond_set_carrier(bond);
1790 #ifdef CONFIG_NET_POLL_CONTROLLER
1791 if (slaves_support_netpoll(bond_dev)) {
1792 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1793 if (bond_dev->npinfo)
1794 slave_dev->npinfo = bond_dev->npinfo;
1795 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1796 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1797 pr_info("New slave device %s does not support netpoll\n",
1799 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1802 read_unlock(&bond->lock);
1804 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1808 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1809 bond_dev->name, slave_dev->name,
1810 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1811 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1813 /* enslave is successful */
1816 /* Undo stages on error */
1818 dev_close(slave_dev);
1821 netdev_set_master(slave_dev, NULL);
1824 if (!bond->params.fail_over_mac) {
1825 /* XXX TODO - fom follow mode needs to change master's
1826 * MAC if this slave's MAC is in use by the bond, or at
1827 * least print a warning.
1829 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1830 addr.sa_family = slave_dev->type;
1831 dev_set_mac_address(slave_dev, &addr);
1835 dev_set_mtu(slave_dev, new_slave->original_mtu);
1841 bond_dev->features = old_features;
1847 * Try to release the slave device <slave> from the bond device <master>
1848 * It is legal to access curr_active_slave without a lock because all the function
1851 * The rules for slave state should be:
1852 * for Active/Backup:
1853 * Active stays on all backups go down
1854 * for Bonded connections:
1855 * The first up interface should be left on and all others downed.
1857 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1859 struct bonding *bond = netdev_priv(bond_dev);
1860 struct slave *slave, *oldcurrent;
1861 struct sockaddr addr;
1863 /* slave is not a slave or master is not master of this slave */
1864 if (!(slave_dev->flags & IFF_SLAVE) ||
1865 (slave_dev->master != bond_dev)) {
1866 pr_err("%s: Error: cannot release %s.\n",
1867 bond_dev->name, slave_dev->name);
1872 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1873 write_lock_bh(&bond->lock);
1875 slave = bond_get_slave_by_dev(bond, slave_dev);
1877 /* not a slave of this bond */
1878 pr_info("%s: %s not enslaved\n",
1879 bond_dev->name, slave_dev->name);
1880 write_unlock_bh(&bond->lock);
1881 unblock_netpoll_tx();
1885 if (!bond->params.fail_over_mac) {
1886 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1887 bond->slave_cnt > 1)
1888 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",
1889 bond_dev->name, slave_dev->name,
1891 bond_dev->name, slave_dev->name);
1894 /* Inform AD package of unbinding of slave. */
1895 if (bond->params.mode == BOND_MODE_8023AD) {
1896 /* must be called before the slave is
1897 * detached from the list
1899 bond_3ad_unbind_slave(slave);
1902 pr_info("%s: releasing %s interface %s\n",
1904 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1907 oldcurrent = bond->curr_active_slave;
1909 bond->current_arp_slave = NULL;
1911 /* release the slave from its bond */
1912 bond_detach_slave(bond, slave);
1914 bond_compute_features(bond);
1916 if (bond->primary_slave == slave)
1917 bond->primary_slave = NULL;
1919 if (oldcurrent == slave)
1920 bond_change_active_slave(bond, NULL);
1922 if (bond_is_lb(bond)) {
1923 /* Must be called only after the slave has been
1924 * detached from the list and the curr_active_slave
1925 * has been cleared (if our_slave == old_current),
1926 * but before a new active slave is selected.
1928 write_unlock_bh(&bond->lock);
1929 bond_alb_deinit_slave(bond, slave);
1930 write_lock_bh(&bond->lock);
1933 if (oldcurrent == slave) {
1935 * Note that we hold RTNL over this sequence, so there
1936 * is no concern that another slave add/remove event
1939 write_unlock_bh(&bond->lock);
1940 read_lock(&bond->lock);
1941 write_lock_bh(&bond->curr_slave_lock);
1943 bond_select_active_slave(bond);
1945 write_unlock_bh(&bond->curr_slave_lock);
1946 read_unlock(&bond->lock);
1947 write_lock_bh(&bond->lock);
1950 if (bond->slave_cnt == 0) {
1951 bond_set_carrier(bond);
1953 /* if the last slave was removed, zero the mac address
1954 * of the master so it will be set by the application
1955 * to the mac address of the first slave
1957 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1960 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1962 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1963 bond_dev->name, bond_dev->name);
1964 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1967 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1968 !bond_has_challenged_slaves(bond)) {
1969 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1970 bond_dev->name, slave_dev->name, bond_dev->name);
1971 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1974 write_unlock_bh(&bond->lock);
1975 unblock_netpoll_tx();
1977 /* must do this from outside any spinlocks */
1978 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1980 bond_del_vlans_from_slave(bond, slave_dev);
1982 /* If the mode USES_PRIMARY, then we should only remove its
1983 * promisc and mc settings if it was the curr_active_slave, but that was
1984 * already taken care of above when we detached the slave
1986 if (!USES_PRIMARY(bond->params.mode)) {
1987 /* unset promiscuity level from slave */
1988 if (bond_dev->flags & IFF_PROMISC)
1989 dev_set_promiscuity(slave_dev, -1);
1991 /* unset allmulti level from slave */
1992 if (bond_dev->flags & IFF_ALLMULTI)
1993 dev_set_allmulti(slave_dev, -1);
1995 /* flush master's mc_list from slave */
1996 netif_addr_lock_bh(bond_dev);
1997 bond_mc_list_flush(bond_dev, slave_dev);
1998 netif_addr_unlock_bh(bond_dev);
2001 netdev_set_master(slave_dev, NULL);
2003 #ifdef CONFIG_NET_POLL_CONTROLLER
2004 read_lock_bh(&bond->lock);
2006 if (slaves_support_netpoll(bond_dev))
2007 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2008 read_unlock_bh(&bond->lock);
2009 if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2010 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2012 slave_dev->npinfo = NULL;
2015 /* close slave before restoring its mac address */
2016 dev_close(slave_dev);
2018 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2019 /* restore original ("permanent") mac address */
2020 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2021 addr.sa_family = slave_dev->type;
2022 dev_set_mac_address(slave_dev, &addr);
2025 dev_set_mtu(slave_dev, slave->original_mtu);
2027 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2028 IFF_SLAVE_INACTIVE | IFF_BONDING |
2033 return 0; /* deletion OK */
2037 * First release a slave and than destroy the bond if no more slaves are left.
2038 * Must be under rtnl_lock when this function is called.
2040 static int bond_release_and_destroy(struct net_device *bond_dev,
2041 struct net_device *slave_dev)
2043 struct bonding *bond = netdev_priv(bond_dev);
2046 ret = bond_release(bond_dev, slave_dev);
2047 if ((ret == 0) && (bond->slave_cnt == 0)) {
2048 pr_info("%s: destroying bond %s.\n",
2049 bond_dev->name, bond_dev->name);
2050 unregister_netdevice(bond_dev);
2056 * This function releases all slaves.
2058 static int bond_release_all(struct net_device *bond_dev)
2060 struct bonding *bond = netdev_priv(bond_dev);
2061 struct slave *slave;
2062 struct net_device *slave_dev;
2063 struct sockaddr addr;
2065 write_lock_bh(&bond->lock);
2067 netif_carrier_off(bond_dev);
2069 if (bond->slave_cnt == 0)
2072 bond->current_arp_slave = NULL;
2073 bond->primary_slave = NULL;
2074 bond_change_active_slave(bond, NULL);
2076 while ((slave = bond->first_slave) != NULL) {
2077 /* Inform AD package of unbinding of slave
2078 * before slave is detached from the list.
2080 if (bond->params.mode == BOND_MODE_8023AD)
2081 bond_3ad_unbind_slave(slave);
2083 slave_dev = slave->dev;
2084 bond_detach_slave(bond, slave);
2086 /* now that the slave is detached, unlock and perform
2087 * all the undo steps that should not be called from
2090 write_unlock_bh(&bond->lock);
2092 if (bond_is_lb(bond)) {
2093 /* must be called only after the slave
2094 * has been detached from the list
2096 bond_alb_deinit_slave(bond, slave);
2099 bond_compute_features(bond);
2101 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2102 bond_del_vlans_from_slave(bond, slave_dev);
2104 /* If the mode USES_PRIMARY, then we should only remove its
2105 * promisc and mc settings if it was the curr_active_slave, but that was
2106 * already taken care of above when we detached the slave
2108 if (!USES_PRIMARY(bond->params.mode)) {
2109 /* unset promiscuity level from slave */
2110 if (bond_dev->flags & IFF_PROMISC)
2111 dev_set_promiscuity(slave_dev, -1);
2113 /* unset allmulti level from slave */
2114 if (bond_dev->flags & IFF_ALLMULTI)
2115 dev_set_allmulti(slave_dev, -1);
2117 /* flush master's mc_list from slave */
2118 netif_addr_lock_bh(bond_dev);
2119 bond_mc_list_flush(bond_dev, slave_dev);
2120 netif_addr_unlock_bh(bond_dev);
2123 netdev_set_master(slave_dev, NULL);
2125 /* close slave before restoring its mac address */
2126 dev_close(slave_dev);
2128 if (!bond->params.fail_over_mac) {
2129 /* restore original ("permanent") mac address*/
2130 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2131 addr.sa_family = slave_dev->type;
2132 dev_set_mac_address(slave_dev, &addr);
2135 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2136 IFF_SLAVE_INACTIVE);
2140 /* re-acquire the lock before getting the next slave */
2141 write_lock_bh(&bond->lock);
2144 /* zero the mac address of the master so it will be
2145 * set by the application to the mac address of the
2148 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2151 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2153 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2154 bond_dev->name, bond_dev->name);
2155 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2159 pr_info("%s: released all slaves\n", bond_dev->name);
2162 write_unlock_bh(&bond->lock);
2167 * This function changes the active slave to slave <slave_dev>.
2168 * It returns -EINVAL in the following cases.
2169 * - <slave_dev> is not found in the list.
2170 * - There is not active slave now.
2171 * - <slave_dev> is already active.
2172 * - The link state of <slave_dev> is not BOND_LINK_UP.
2173 * - <slave_dev> is not running.
2174 * In these cases, this function does nothing.
2175 * In the other cases, current_slave pointer is changed and 0 is returned.
2177 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2179 struct bonding *bond = netdev_priv(bond_dev);
2180 struct slave *old_active = NULL;
2181 struct slave *new_active = NULL;
2184 if (!USES_PRIMARY(bond->params.mode))
2187 /* Verify that master_dev is indeed the master of slave_dev */
2188 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2191 read_lock(&bond->lock);
2193 read_lock(&bond->curr_slave_lock);
2194 old_active = bond->curr_active_slave;
2195 read_unlock(&bond->curr_slave_lock);
2197 new_active = bond_get_slave_by_dev(bond, slave_dev);
2200 * Changing to the current active: do nothing; return success.
2202 if (new_active && (new_active == old_active)) {
2203 read_unlock(&bond->lock);
2209 (new_active->link == BOND_LINK_UP) &&
2210 IS_UP(new_active->dev)) {
2212 write_lock_bh(&bond->curr_slave_lock);
2213 bond_change_active_slave(bond, new_active);
2214 write_unlock_bh(&bond->curr_slave_lock);
2215 unblock_netpoll_tx();
2219 read_unlock(&bond->lock);
2224 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2226 struct bonding *bond = netdev_priv(bond_dev);
2228 info->bond_mode = bond->params.mode;
2229 info->miimon = bond->params.miimon;
2231 read_lock(&bond->lock);
2232 info->num_slaves = bond->slave_cnt;
2233 read_unlock(&bond->lock);
2238 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2240 struct bonding *bond = netdev_priv(bond_dev);
2241 struct slave *slave;
2242 int i, res = -ENODEV;
2244 read_lock(&bond->lock);
2246 bond_for_each_slave(bond, slave, i) {
2247 if (i == (int)info->slave_id) {
2249 strcpy(info->slave_name, slave->dev->name);
2250 info->link = slave->link;
2251 info->state = slave->state;
2252 info->link_failure_count = slave->link_failure_count;
2257 read_unlock(&bond->lock);
2262 /*-------------------------------- Monitoring -------------------------------*/
2265 static int bond_miimon_inspect(struct bonding *bond)
2267 struct slave *slave;
2268 int i, link_state, commit = 0;
2269 bool ignore_updelay;
2271 ignore_updelay = !bond->curr_active_slave ? true : false;
2273 bond_for_each_slave(bond, slave, i) {
2274 slave->new_link = BOND_LINK_NOCHANGE;
2276 link_state = bond_check_dev_link(bond, slave->dev, 0);
2278 switch (slave->link) {
2283 slave->link = BOND_LINK_FAIL;
2284 slave->delay = bond->params.downdelay;
2286 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2288 (bond->params.mode ==
2289 BOND_MODE_ACTIVEBACKUP) ?
2290 ((slave->state == BOND_STATE_ACTIVE) ?
2291 "active " : "backup ") : "",
2293 bond->params.downdelay * bond->params.miimon);
2296 case BOND_LINK_FAIL:
2299 * recovered before downdelay expired
2301 slave->link = BOND_LINK_UP;
2302 slave->jiffies = jiffies;
2303 pr_info("%s: link status up again after %d ms for interface %s.\n",
2305 (bond->params.downdelay - slave->delay) *
2306 bond->params.miimon,
2311 if (slave->delay <= 0) {
2312 slave->new_link = BOND_LINK_DOWN;
2320 case BOND_LINK_DOWN:
2324 slave->link = BOND_LINK_BACK;
2325 slave->delay = bond->params.updelay;
2328 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2329 bond->dev->name, slave->dev->name,
2330 ignore_updelay ? 0 :
2331 bond->params.updelay *
2332 bond->params.miimon);
2335 case BOND_LINK_BACK:
2337 slave->link = BOND_LINK_DOWN;
2338 pr_info("%s: link status down again after %d ms for interface %s.\n",
2340 (bond->params.updelay - slave->delay) *
2341 bond->params.miimon,
2350 if (slave->delay <= 0) {
2351 slave->new_link = BOND_LINK_UP;
2353 ignore_updelay = false;
2365 static void bond_miimon_commit(struct bonding *bond)
2367 struct slave *slave;
2370 bond_for_each_slave(bond, slave, i) {
2371 switch (slave->new_link) {
2372 case BOND_LINK_NOCHANGE:
2376 slave->link = BOND_LINK_UP;
2377 slave->jiffies = jiffies;
2379 if (bond->params.mode == BOND_MODE_8023AD) {
2380 /* prevent it from being the active one */
2381 slave->state = BOND_STATE_BACKUP;
2382 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2383 /* make it immediately active */
2384 slave->state = BOND_STATE_ACTIVE;
2385 } else if (slave != bond->primary_slave) {
2386 /* prevent it from being the active one */
2387 slave->state = BOND_STATE_BACKUP;
2390 bond_update_speed_duplex(slave);
2392 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2393 bond->dev->name, slave->dev->name,
2394 slave->speed, slave->duplex ? "full" : "half");
2396 /* notify ad that the link status has changed */
2397 if (bond->params.mode == BOND_MODE_8023AD)
2398 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2400 if (bond_is_lb(bond))
2401 bond_alb_handle_link_change(bond, slave,
2404 if (!bond->curr_active_slave ||
2405 (slave == bond->primary_slave))
2410 case BOND_LINK_DOWN:
2411 if (slave->link_failure_count < UINT_MAX)
2412 slave->link_failure_count++;
2414 slave->link = BOND_LINK_DOWN;
2416 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2417 bond->params.mode == BOND_MODE_8023AD)
2418 bond_set_slave_inactive_flags(slave);
2420 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2421 bond->dev->name, slave->dev->name);
2423 if (bond->params.mode == BOND_MODE_8023AD)
2424 bond_3ad_handle_link_change(slave,
2427 if (bond_is_lb(bond))
2428 bond_alb_handle_link_change(bond, slave,
2431 if (slave == bond->curr_active_slave)
2437 pr_err("%s: invalid new link %d on slave %s\n",
2438 bond->dev->name, slave->new_link,
2440 slave->new_link = BOND_LINK_NOCHANGE;
2448 write_lock_bh(&bond->curr_slave_lock);
2449 bond_select_active_slave(bond);
2450 write_unlock_bh(&bond->curr_slave_lock);
2451 unblock_netpoll_tx();
2454 bond_set_carrier(bond);
2460 * Really a wrapper that splits the mii monitor into two phases: an
2461 * inspection, then (if inspection indicates something needs to be done)
2462 * an acquisition of appropriate locks followed by a commit phase to
2463 * implement whatever link state changes are indicated.
2465 void bond_mii_monitor(struct work_struct *work)
2467 struct bonding *bond = container_of(work, struct bonding,
2470 read_lock(&bond->lock);
2471 if (bond->kill_timers)
2474 if (bond->slave_cnt == 0)
2477 if (bond->send_grat_arp) {
2478 read_lock(&bond->curr_slave_lock);
2479 bond_send_gratuitous_arp(bond);
2480 read_unlock(&bond->curr_slave_lock);
2483 if (bond->send_unsol_na) {
2484 read_lock(&bond->curr_slave_lock);
2485 bond_send_unsolicited_na(bond);
2486 read_unlock(&bond->curr_slave_lock);
2489 if (bond_miimon_inspect(bond)) {
2490 read_unlock(&bond->lock);
2492 read_lock(&bond->lock);
2494 bond_miimon_commit(bond);
2496 read_unlock(&bond->lock);
2497 rtnl_unlock(); /* might sleep, hold no other locks */
2498 read_lock(&bond->lock);
2502 if (bond->params.miimon)
2503 queue_delayed_work(bond->wq, &bond->mii_work,
2504 msecs_to_jiffies(bond->params.miimon));
2506 read_unlock(&bond->lock);
2509 static __be32 bond_glean_dev_ip(struct net_device *dev)
2511 struct in_device *idev;
2512 struct in_ifaddr *ifa;
2519 idev = __in_dev_get_rcu(dev);
2523 ifa = idev->ifa_list;
2527 addr = ifa->ifa_local;
2533 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2535 struct vlan_entry *vlan;
2537 if (ip == bond->master_ip)
2540 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2541 if (ip == vlan->vlan_ip)
2549 * We go to the (large) trouble of VLAN tagging ARP frames because
2550 * switches in VLAN mode (especially if ports are configured as
2551 * "native" to a VLAN) might not pass non-tagged frames.
2553 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2555 struct sk_buff *skb;
2557 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2558 slave_dev->name, dest_ip, src_ip, vlan_id);
2560 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2561 NULL, slave_dev->dev_addr, NULL);
2564 pr_err("ARP packet allocation failed\n");
2568 skb = vlan_put_tag(skb, vlan_id);
2570 pr_err("failed to insert VLAN tag\n");
2578 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2581 __be32 *targets = bond->params.arp_targets;
2582 struct vlan_entry *vlan;
2583 struct net_device *vlan_dev;
2587 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2590 pr_debug("basa: target %x\n", targets[i]);
2592 pr_debug("basa: empty vlan: arp_send\n");
2593 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2594 bond->master_ip, 0);
2599 * If VLANs are configured, we do a route lookup to
2600 * determine which VLAN interface would be used, so we
2601 * can tag the ARP with the proper VLAN tag.
2603 memset(&fl, 0, sizeof(fl));
2604 fl.fl4_dst = targets[i];
2605 fl.fl4_tos = RTO_ONLINK;
2607 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2609 if (net_ratelimit()) {
2610 pr_warning("%s: no route to arp_ip_target %pI4\n",
2611 bond->dev->name, &fl.fl4_dst);
2617 * This target is not on a VLAN
2619 if (rt->dst.dev == bond->dev) {
2621 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2622 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2623 bond->master_ip, 0);
2628 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2629 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2630 if (vlan_dev == rt->dst.dev) {
2631 vlan_id = vlan->vlan_id;
2632 pr_debug("basa: vlan match on %s %d\n",
2633 vlan_dev->name, vlan_id);
2640 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2641 vlan->vlan_ip, vlan_id);
2645 if (net_ratelimit()) {
2646 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2647 bond->dev->name, &fl.fl4_dst,
2648 rt->dst.dev ? rt->dst.dev->name : "NULL");
2655 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2656 * for each VLAN above us.
2658 * Caller must hold curr_slave_lock for read or better
2660 static void bond_send_gratuitous_arp(struct bonding *bond)
2662 struct slave *slave = bond->curr_active_slave;
2663 struct vlan_entry *vlan;
2664 struct net_device *vlan_dev;
2666 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2667 bond->dev->name, slave ? slave->dev->name : "NULL");
2669 if (!slave || !bond->send_grat_arp ||
2670 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2673 bond->send_grat_arp--;
2675 if (bond->master_ip) {
2676 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2677 bond->master_ip, 0);
2683 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2684 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2685 if (vlan->vlan_ip) {
2686 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2687 vlan->vlan_ip, vlan->vlan_id);
2692 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2695 __be32 *targets = bond->params.arp_targets;
2697 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2698 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2699 &sip, &tip, i, &targets[i],
2700 bond_has_this_ip(bond, tip));
2701 if (sip == targets[i]) {
2702 if (bond_has_this_ip(bond, tip))
2703 slave->last_arp_rx = jiffies;
2709 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2712 struct slave *slave;
2713 struct bonding *bond;
2714 unsigned char *arp_ptr;
2717 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2719 * When using VLANS and bonding, dev and oriv_dev may be
2720 * incorrect if the physical interface supports VLAN
2721 * acceleration. With this change ARP validation now
2722 * works for hosts only reachable on the VLAN interface.
2724 dev = vlan_dev_real_dev(dev);
2725 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2728 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2731 bond = netdev_priv(dev);
2732 read_lock(&bond->lock);
2734 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2735 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2736 orig_dev ? orig_dev->name : "NULL");
2738 slave = bond_get_slave_by_dev(bond, orig_dev);
2739 if (!slave || !slave_do_arp_validate(bond, slave))
2742 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2746 if (arp->ar_hln != dev->addr_len ||
2747 skb->pkt_type == PACKET_OTHERHOST ||
2748 skb->pkt_type == PACKET_LOOPBACK ||
2749 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2750 arp->ar_pro != htons(ETH_P_IP) ||
2754 arp_ptr = (unsigned char *)(arp + 1);
2755 arp_ptr += dev->addr_len;
2756 memcpy(&sip, arp_ptr, 4);
2757 arp_ptr += 4 + dev->addr_len;
2758 memcpy(&tip, arp_ptr, 4);
2760 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2761 bond->dev->name, slave->dev->name, slave->state,
2762 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2766 * Backup slaves won't see the ARP reply, but do come through
2767 * here for each ARP probe (so we swap the sip/tip to validate
2768 * the probe). In a "redundant switch, common router" type of
2769 * configuration, the ARP probe will (hopefully) travel from
2770 * the active, through one switch, the router, then the other
2771 * switch before reaching the backup.
2773 if (slave->state == BOND_STATE_ACTIVE)
2774 bond_validate_arp(bond, slave, sip, tip);
2776 bond_validate_arp(bond, slave, tip, sip);
2779 read_unlock(&bond->lock);
2782 return NET_RX_SUCCESS;
2786 * this function is called regularly to monitor each slave's link
2787 * ensuring that traffic is being sent and received when arp monitoring
2788 * is used in load-balancing mode. if the adapter has been dormant, then an
2789 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2790 * arp monitoring in active backup mode.
2792 void bond_loadbalance_arp_mon(struct work_struct *work)
2794 struct bonding *bond = container_of(work, struct bonding,
2796 struct slave *slave, *oldcurrent;
2797 int do_failover = 0;
2801 read_lock(&bond->lock);
2803 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2805 if (bond->kill_timers)
2808 if (bond->slave_cnt == 0)
2811 read_lock(&bond->curr_slave_lock);
2812 oldcurrent = bond->curr_active_slave;
2813 read_unlock(&bond->curr_slave_lock);
2815 /* see if any of the previous devices are up now (i.e. they have
2816 * xmt and rcv traffic). the curr_active_slave does not come into
2817 * the picture unless it is null. also, slave->jiffies is not needed
2818 * here because we send an arp on each slave and give a slave as
2819 * long as it needs to get the tx/rx within the delta.
2820 * TODO: what about up/down delay in arp mode? it wasn't here before
2823 bond_for_each_slave(bond, slave, i) {
2824 unsigned long trans_start = dev_trans_start(slave->dev);
2826 if (slave->link != BOND_LINK_UP) {
2827 if (time_in_range(jiffies,
2828 trans_start - delta_in_ticks,
2829 trans_start + delta_in_ticks) &&
2830 time_in_range(jiffies,
2831 slave->dev->last_rx - delta_in_ticks,
2832 slave->dev->last_rx + delta_in_ticks)) {
2834 slave->link = BOND_LINK_UP;
2835 slave->state = BOND_STATE_ACTIVE;
2837 /* primary_slave has no meaning in round-robin
2838 * mode. the window of a slave being up and
2839 * curr_active_slave being null after enslaving
2843 pr_info("%s: link status definitely up for interface %s, ",
2848 pr_info("%s: interface %s is now up\n",
2854 /* slave->link == BOND_LINK_UP */
2856 /* not all switches will respond to an arp request
2857 * when the source ip is 0, so don't take the link down
2858 * if we don't know our ip yet
2860 if (!time_in_range(jiffies,
2861 trans_start - delta_in_ticks,
2862 trans_start + 2 * delta_in_ticks) ||
2863 !time_in_range(jiffies,
2864 slave->dev->last_rx - delta_in_ticks,
2865 slave->dev->last_rx + 2 * delta_in_ticks)) {
2867 slave->link = BOND_LINK_DOWN;
2868 slave->state = BOND_STATE_BACKUP;
2870 if (slave->link_failure_count < UINT_MAX)
2871 slave->link_failure_count++;
2873 pr_info("%s: interface %s is now down.\n",
2877 if (slave == oldcurrent)
2882 /* note: if switch is in round-robin mode, all links
2883 * must tx arp to ensure all links rx an arp - otherwise
2884 * links may oscillate or not come up at all; if switch is
2885 * in something like xor mode, there is nothing we can
2886 * do - all replies will be rx'ed on same link causing slaves
2887 * to be unstable during low/no traffic periods
2889 if (IS_UP(slave->dev))
2890 bond_arp_send_all(bond, slave);
2895 write_lock_bh(&bond->curr_slave_lock);
2897 bond_select_active_slave(bond);
2899 write_unlock_bh(&bond->curr_slave_lock);
2900 unblock_netpoll_tx();
2904 if (bond->params.arp_interval)
2905 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2907 read_unlock(&bond->lock);
2911 * Called to inspect slaves for active-backup mode ARP monitor link state
2912 * changes. Sets new_link in slaves to specify what action should take
2913 * place for the slave. Returns 0 if no changes are found, >0 if changes
2914 * to link states must be committed.
2916 * Called with bond->lock held for read.
2918 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2920 struct slave *slave;
2922 unsigned long trans_start;
2924 bond_for_each_slave(bond, slave, i) {
2925 slave->new_link = BOND_LINK_NOCHANGE;
2927 if (slave->link != BOND_LINK_UP) {
2928 if (time_in_range(jiffies,
2929 slave_last_rx(bond, slave) - delta_in_ticks,
2930 slave_last_rx(bond, slave) + delta_in_ticks)) {
2932 slave->new_link = BOND_LINK_UP;
2940 * Give slaves 2*delta after being enslaved or made
2941 * active. This avoids bouncing, as the last receive
2942 * times need a full ARP monitor cycle to be updated.
2944 if (time_in_range(jiffies,
2945 slave->jiffies - delta_in_ticks,
2946 slave->jiffies + 2 * delta_in_ticks))
2950 * Backup slave is down if:
2951 * - No current_arp_slave AND
2952 * - more than 3*delta since last receive AND
2953 * - the bond has an IP address
2955 * Note: a non-null current_arp_slave indicates
2956 * the curr_active_slave went down and we are
2957 * searching for a new one; under this condition
2958 * we only take the curr_active_slave down - this
2959 * gives each slave a chance to tx/rx traffic
2960 * before being taken out
2962 if (slave->state == BOND_STATE_BACKUP &&
2963 !bond->current_arp_slave &&
2964 !time_in_range(jiffies,
2965 slave_last_rx(bond, slave) - delta_in_ticks,
2966 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2968 slave->new_link = BOND_LINK_DOWN;
2973 * Active slave is down if:
2974 * - more than 2*delta since transmitting OR
2975 * - (more than 2*delta since receive AND
2976 * the bond has an IP address)
2978 trans_start = dev_trans_start(slave->dev);
2979 if ((slave->state == BOND_STATE_ACTIVE) &&
2980 (!time_in_range(jiffies,
2981 trans_start - delta_in_ticks,
2982 trans_start + 2 * delta_in_ticks) ||
2983 !time_in_range(jiffies,
2984 slave_last_rx(bond, slave) - delta_in_ticks,
2985 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2987 slave->new_link = BOND_LINK_DOWN;
2996 * Called to commit link state changes noted by inspection step of
2997 * active-backup mode ARP monitor.
2999 * Called with RTNL and bond->lock for read.
3001 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3003 struct slave *slave;
3005 unsigned long trans_start;
3007 bond_for_each_slave(bond, slave, i) {
3008 switch (slave->new_link) {
3009 case BOND_LINK_NOCHANGE:
3013 trans_start = dev_trans_start(slave->dev);
3014 if ((!bond->curr_active_slave &&
3015 time_in_range(jiffies,
3016 trans_start - delta_in_ticks,
3017 trans_start + delta_in_ticks)) ||
3018 bond->curr_active_slave != slave) {
3019 slave->link = BOND_LINK_UP;
3020 bond->current_arp_slave = NULL;
3022 pr_info("%s: link status definitely up for interface %s.\n",
3023 bond->dev->name, slave->dev->name);
3025 if (!bond->curr_active_slave ||
3026 (slave == bond->primary_slave))
3033 case BOND_LINK_DOWN:
3034 if (slave->link_failure_count < UINT_MAX)
3035 slave->link_failure_count++;
3037 slave->link = BOND_LINK_DOWN;
3038 bond_set_slave_inactive_flags(slave);
3040 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3041 bond->dev->name, slave->dev->name);
3043 if (slave == bond->curr_active_slave) {
3044 bond->current_arp_slave = NULL;
3051 pr_err("%s: impossible: new_link %d on slave %s\n",
3052 bond->dev->name, slave->new_link,
3060 write_lock_bh(&bond->curr_slave_lock);
3061 bond_select_active_slave(bond);
3062 write_unlock_bh(&bond->curr_slave_lock);
3063 unblock_netpoll_tx();
3066 bond_set_carrier(bond);
3070 * Send ARP probes for active-backup mode ARP monitor.
3072 * Called with bond->lock held for read.
3074 static void bond_ab_arp_probe(struct bonding *bond)
3076 struct slave *slave;
3079 read_lock(&bond->curr_slave_lock);
3081 if (bond->current_arp_slave && bond->curr_active_slave)
3082 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3083 bond->current_arp_slave->dev->name,
3084 bond->curr_active_slave->dev->name);
3086 if (bond->curr_active_slave) {
3087 bond_arp_send_all(bond, bond->curr_active_slave);
3088 read_unlock(&bond->curr_slave_lock);
3092 read_unlock(&bond->curr_slave_lock);
3094 /* if we don't have a curr_active_slave, search for the next available
3095 * backup slave from the current_arp_slave and make it the candidate
3096 * for becoming the curr_active_slave
3099 if (!bond->current_arp_slave) {
3100 bond->current_arp_slave = bond->first_slave;
3101 if (!bond->current_arp_slave)
3105 bond_set_slave_inactive_flags(bond->current_arp_slave);
3107 /* search for next candidate */
3108 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3109 if (IS_UP(slave->dev)) {
3110 slave->link = BOND_LINK_BACK;
3111 bond_set_slave_active_flags(slave);
3112 bond_arp_send_all(bond, slave);
3113 slave->jiffies = jiffies;
3114 bond->current_arp_slave = slave;
3118 /* if the link state is up at this point, we
3119 * mark it down - this can happen if we have
3120 * simultaneous link failures and
3121 * reselect_active_interface doesn't make this
3122 * one the current slave so it is still marked
3123 * up when it is actually down
3125 if (slave->link == BOND_LINK_UP) {
3126 slave->link = BOND_LINK_DOWN;
3127 if (slave->link_failure_count < UINT_MAX)
3128 slave->link_failure_count++;
3130 bond_set_slave_inactive_flags(slave);
3132 pr_info("%s: backup interface %s is now down.\n",
3133 bond->dev->name, slave->dev->name);
3138 void bond_activebackup_arp_mon(struct work_struct *work)
3140 struct bonding *bond = container_of(work, struct bonding,
3144 read_lock(&bond->lock);
3146 if (bond->kill_timers)
3149 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3151 if (bond->slave_cnt == 0)
3154 if (bond->send_grat_arp) {
3155 read_lock(&bond->curr_slave_lock);
3156 bond_send_gratuitous_arp(bond);
3157 read_unlock(&bond->curr_slave_lock);
3160 if (bond->send_unsol_na) {
3161 read_lock(&bond->curr_slave_lock);
3162 bond_send_unsolicited_na(bond);
3163 read_unlock(&bond->curr_slave_lock);
3166 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3167 read_unlock(&bond->lock);
3169 read_lock(&bond->lock);
3171 bond_ab_arp_commit(bond, delta_in_ticks);
3173 read_unlock(&bond->lock);
3175 read_lock(&bond->lock);
3178 bond_ab_arp_probe(bond);
3181 if (bond->params.arp_interval)
3182 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3184 read_unlock(&bond->lock);
3187 /*------------------------------ proc/seq_file-------------------------------*/
3189 #ifdef CONFIG_PROC_FS
3191 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3192 __acquires(&dev_base_lock)
3193 __acquires(&bond->lock)
3195 struct bonding *bond = seq->private;
3197 struct slave *slave;
3200 /* make sure the bond won't be taken away */
3201 read_lock(&dev_base_lock);
3202 read_lock(&bond->lock);
3205 return SEQ_START_TOKEN;
3207 bond_for_each_slave(bond, slave, i) {
3215 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3217 struct bonding *bond = seq->private;
3218 struct slave *slave = v;
3221 if (v == SEQ_START_TOKEN)
3222 return bond->first_slave;
3224 slave = slave->next;
3226 return (slave == bond->first_slave) ? NULL : slave;
3229 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3230 __releases(&bond->lock)
3231 __releases(&dev_base_lock)
3233 struct bonding *bond = seq->private;
3235 read_unlock(&bond->lock);
3236 read_unlock(&dev_base_lock);
3239 static void bond_info_show_master(struct seq_file *seq)
3241 struct bonding *bond = seq->private;
3245 read_lock(&bond->curr_slave_lock);
3246 curr = bond->curr_active_slave;
3247 read_unlock(&bond->curr_slave_lock);
3249 seq_printf(seq, "Bonding Mode: %s",
3250 bond_mode_name(bond->params.mode));
3252 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3253 bond->params.fail_over_mac)
3254 seq_printf(seq, " (fail_over_mac %s)",
3255 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3257 seq_printf(seq, "\n");
3259 if (bond->params.mode == BOND_MODE_XOR ||
3260 bond->params.mode == BOND_MODE_8023AD) {
3261 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3262 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3263 bond->params.xmit_policy);
3266 if (USES_PRIMARY(bond->params.mode)) {
3267 seq_printf(seq, "Primary Slave: %s",
3268 (bond->primary_slave) ?
3269 bond->primary_slave->dev->name : "None");
3270 if (bond->primary_slave)
3271 seq_printf(seq, " (primary_reselect %s)",
3272 pri_reselect_tbl[bond->params.primary_reselect].modename);
3274 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3275 (curr) ? curr->dev->name : "None");
3278 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3280 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3281 seq_printf(seq, "Up Delay (ms): %d\n",
3282 bond->params.updelay * bond->params.miimon);
3283 seq_printf(seq, "Down Delay (ms): %d\n",
3284 bond->params.downdelay * bond->params.miimon);
3287 /* ARP information */
3288 if (bond->params.arp_interval > 0) {
3290 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3291 bond->params.arp_interval);
3293 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3295 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3296 if (!bond->params.arp_targets[i])
3299 seq_printf(seq, ",");
3300 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3303 seq_printf(seq, "\n");
3306 if (bond->params.mode == BOND_MODE_8023AD) {
3307 struct ad_info ad_info;
3309 seq_puts(seq, "\n802.3ad info\n");
3310 seq_printf(seq, "LACP rate: %s\n",
3311 (bond->params.lacp_fast) ? "fast" : "slow");
3312 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3313 ad_select_tbl[bond->params.ad_select].modename);
3315 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3316 seq_printf(seq, "bond %s has no active aggregator\n",
3319 seq_printf(seq, "Active Aggregator Info:\n");
3321 seq_printf(seq, "\tAggregator ID: %d\n",
3322 ad_info.aggregator_id);
3323 seq_printf(seq, "\tNumber of ports: %d\n",
3325 seq_printf(seq, "\tActor Key: %d\n",
3327 seq_printf(seq, "\tPartner Key: %d\n",
3328 ad_info.partner_key);
3329 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3330 ad_info.partner_system);
3335 static void bond_info_show_slave(struct seq_file *seq,
3336 const struct slave *slave)
3338 struct bonding *bond = seq->private;
3340 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3341 seq_printf(seq, "MII Status: %s\n",
3342 (slave->link == BOND_LINK_UP) ? "up" : "down");
3343 seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3344 seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3345 seq_printf(seq, "Link Failure Count: %u\n",
3346 slave->link_failure_count);
3348 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3350 if (bond->params.mode == BOND_MODE_8023AD) {
3351 const struct aggregator *agg
3352 = SLAVE_AD_INFO(slave).port.aggregator;
3355 seq_printf(seq, "Aggregator ID: %d\n",
3356 agg->aggregator_identifier);
3358 seq_puts(seq, "Aggregator ID: N/A\n");
3360 seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3363 static int bond_info_seq_show(struct seq_file *seq, void *v)
3365 if (v == SEQ_START_TOKEN) {
3366 seq_printf(seq, "%s\n", version);
3367 bond_info_show_master(seq);
3369 bond_info_show_slave(seq, v);
3374 static const struct seq_operations bond_info_seq_ops = {
3375 .start = bond_info_seq_start,
3376 .next = bond_info_seq_next,
3377 .stop = bond_info_seq_stop,
3378 .show = bond_info_seq_show,
3381 static int bond_info_open(struct inode *inode, struct file *file)
3383 struct seq_file *seq;
3384 struct proc_dir_entry *proc;
3387 res = seq_open(file, &bond_info_seq_ops);
3389 /* recover the pointer buried in proc_dir_entry data */
3390 seq = file->private_data;
3392 seq->private = proc->data;
3398 static const struct file_operations bond_info_fops = {
3399 .owner = THIS_MODULE,
3400 .open = bond_info_open,
3402 .llseek = seq_lseek,
3403 .release = seq_release,
3406 static void bond_create_proc_entry(struct bonding *bond)
3408 struct net_device *bond_dev = bond->dev;
3409 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3412 bond->proc_entry = proc_create_data(bond_dev->name,
3413 S_IRUGO, bn->proc_dir,
3414 &bond_info_fops, bond);
3415 if (bond->proc_entry == NULL)
3416 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3417 DRV_NAME, bond_dev->name);
3419 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3423 static void bond_remove_proc_entry(struct bonding *bond)
3425 struct net_device *bond_dev = bond->dev;
3426 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3428 if (bn->proc_dir && bond->proc_entry) {
3429 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3430 memset(bond->proc_file_name, 0, IFNAMSIZ);
3431 bond->proc_entry = NULL;
3435 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3436 * Caller must hold rtnl_lock.
3438 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3440 if (!bn->proc_dir) {
3441 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3443 pr_warning("Warning: cannot create /proc/net/%s\n",
3448 /* Destroy the bonding directory under /proc/net, if empty.
3449 * Caller must hold rtnl_lock.
3451 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3454 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3455 bn->proc_dir = NULL;
3459 #else /* !CONFIG_PROC_FS */
3461 static void bond_create_proc_entry(struct bonding *bond)
3465 static void bond_remove_proc_entry(struct bonding *bond)
3469 static inline void bond_create_proc_dir(struct bond_net *bn)
3473 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3477 #endif /* CONFIG_PROC_FS */
3480 /*-------------------------- netdev event handling --------------------------*/
3483 * Change device name
3485 static int bond_event_changename(struct bonding *bond)
3487 bond_remove_proc_entry(bond);
3488 bond_create_proc_entry(bond);
3493 static int bond_master_netdev_event(unsigned long event,
3494 struct net_device *bond_dev)
3496 struct bonding *event_bond = netdev_priv(bond_dev);
3499 case NETDEV_CHANGENAME:
3500 return bond_event_changename(event_bond);
3508 static int bond_slave_netdev_event(unsigned long event,
3509 struct net_device *slave_dev)
3511 struct net_device *bond_dev = slave_dev->master;
3512 struct bonding *bond = netdev_priv(bond_dev);
3515 case NETDEV_UNREGISTER:
3517 if (bond->setup_by_slave)
3518 bond_release_and_destroy(bond_dev, slave_dev);
3520 bond_release(bond_dev, slave_dev);
3524 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3525 struct slave *slave;
3527 slave = bond_get_slave_by_dev(bond, slave_dev);
3529 u16 old_speed = slave->speed;
3530 u16 old_duplex = slave->duplex;
3532 bond_update_speed_duplex(slave);
3534 if (bond_is_lb(bond))
3537 if (old_speed != slave->speed)
3538 bond_3ad_adapter_speed_changed(slave);
3539 if (old_duplex != slave->duplex)
3540 bond_3ad_adapter_duplex_changed(slave);
3547 * ... Or is it this?
3550 case NETDEV_CHANGEMTU:
3552 * TODO: Should slaves be allowed to
3553 * independently alter their MTU? For
3554 * an active-backup bond, slaves need
3555 * not be the same type of device, so
3556 * MTUs may vary. For other modes,
3557 * slaves arguably should have the
3558 * same MTUs. To do this, we'd need to
3559 * take over the slave's change_mtu
3560 * function for the duration of their
3564 case NETDEV_CHANGENAME:
3566 * TODO: handle changing the primary's name
3569 case NETDEV_FEAT_CHANGE:
3570 bond_compute_features(bond);
3580 * bond_netdev_event: handle netdev notifier chain events.
3582 * This function receives events for the netdev chain. The caller (an
3583 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3584 * locks for us to safely manipulate the slave devices (RTNL lock,
3587 static int bond_netdev_event(struct notifier_block *this,
3588 unsigned long event, void *ptr)
3590 struct net_device *event_dev = (struct net_device *)ptr;
3592 pr_debug("event_dev: %s, event: %lx\n",
3593 event_dev ? event_dev->name : "None",
3596 if (!(event_dev->priv_flags & IFF_BONDING))
3599 if (event_dev->flags & IFF_MASTER) {
3600 pr_debug("IFF_MASTER\n");
3601 return bond_master_netdev_event(event, event_dev);
3604 if (event_dev->flags & IFF_SLAVE) {
3605 pr_debug("IFF_SLAVE\n");
3606 return bond_slave_netdev_event(event, event_dev);
3613 * bond_inetaddr_event: handle inetaddr notifier chain events.
3615 * We keep track of device IPs primarily to use as source addresses in
3616 * ARP monitor probes (rather than spewing out broadcasts all the time).
3618 * We track one IP for the main device (if it has one), plus one per VLAN.
3620 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3622 struct in_ifaddr *ifa = ptr;
3623 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3624 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3625 struct bonding *bond;
3626 struct vlan_entry *vlan;
3628 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3629 if (bond->dev == event_dev) {
3632 bond->master_ip = ifa->ifa_local;
3635 bond->master_ip = bond_glean_dev_ip(bond->dev);
3642 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3645 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3646 if (vlan_dev == event_dev) {
3649 vlan->vlan_ip = ifa->ifa_local;
3653 bond_glean_dev_ip(vlan_dev);
3664 static struct notifier_block bond_netdev_notifier = {
3665 .notifier_call = bond_netdev_event,
3668 static struct notifier_block bond_inetaddr_notifier = {
3669 .notifier_call = bond_inetaddr_event,
3672 /*-------------------------- Packet type handling ---------------------------*/
3674 /* register to receive lacpdus on a bond */
3675 static void bond_register_lacpdu(struct bonding *bond)
3677 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3679 /* initialize packet type */
3680 pk_type->type = PKT_TYPE_LACPDU;
3681 pk_type->dev = bond->dev;
3682 pk_type->func = bond_3ad_lacpdu_recv;
3684 dev_add_pack(pk_type);
3687 /* unregister to receive lacpdus on a bond */
3688 static void bond_unregister_lacpdu(struct bonding *bond)
3690 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3693 void bond_register_arp(struct bonding *bond)
3695 struct packet_type *pt = &bond->arp_mon_pt;
3700 pt->type = htons(ETH_P_ARP);
3701 pt->dev = bond->dev;
3702 pt->func = bond_arp_rcv;
3706 void bond_unregister_arp(struct bonding *bond)
3708 struct packet_type *pt = &bond->arp_mon_pt;
3710 dev_remove_pack(pt);
3714 /*---------------------------- Hashing Policies -----------------------------*/
3717 * Hash for the output device based upon layer 2 and layer 3 data. If
3718 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3720 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3722 struct ethhdr *data = (struct ethhdr *)skb->data;
3723 struct iphdr *iph = ip_hdr(skb);
3725 if (skb->protocol == htons(ETH_P_IP)) {
3726 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3727 (data->h_dest[5] ^ data->h_source[5])) % count;
3730 return (data->h_dest[5] ^ data->h_source[5]) % count;
3734 * Hash for the output device based upon layer 3 and layer 4 data. If
3735 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3736 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3738 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3740 struct ethhdr *data = (struct ethhdr *)skb->data;
3741 struct iphdr *iph = ip_hdr(skb);
3742 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3745 if (skb->protocol == htons(ETH_P_IP)) {
3746 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3747 (iph->protocol == IPPROTO_TCP ||
3748 iph->protocol == IPPROTO_UDP)) {
3749 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3751 return (layer4_xor ^
3752 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3756 return (data->h_dest[5] ^ data->h_source[5]) % count;
3760 * Hash for the output device based upon layer 2 data
3762 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3764 struct ethhdr *data = (struct ethhdr *)skb->data;
3766 return (data->h_dest[5] ^ data->h_source[5]) % count;
3769 /*-------------------------- Device entry points ----------------------------*/
3771 static int bond_open(struct net_device *bond_dev)
3773 struct bonding *bond = netdev_priv(bond_dev);
3775 bond->kill_timers = 0;
3777 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3779 if (bond_is_lb(bond)) {
3780 /* bond_alb_initialize must be called before the timer
3783 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3784 /* something went wrong - fail the open operation */
3788 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3789 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3792 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3793 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3794 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3797 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3798 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3799 INIT_DELAYED_WORK(&bond->arp_work,
3800 bond_activebackup_arp_mon);
3802 INIT_DELAYED_WORK(&bond->arp_work,
3803 bond_loadbalance_arp_mon);
3805 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3806 if (bond->params.arp_validate)
3807 bond_register_arp(bond);
3810 if (bond->params.mode == BOND_MODE_8023AD) {
3811 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3812 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3813 /* register to receive LACPDUs */
3814 bond_register_lacpdu(bond);
3815 bond_3ad_initiate_agg_selection(bond, 1);
3821 static int bond_close(struct net_device *bond_dev)
3823 struct bonding *bond = netdev_priv(bond_dev);
3825 if (bond->params.mode == BOND_MODE_8023AD) {
3826 /* Unregister the receive of LACPDUs */
3827 bond_unregister_lacpdu(bond);
3830 if (bond->params.arp_validate)
3831 bond_unregister_arp(bond);
3833 write_lock_bh(&bond->lock);
3835 bond->send_grat_arp = 0;
3836 bond->send_unsol_na = 0;
3838 /* signal timers not to re-arm */
3839 bond->kill_timers = 1;
3841 write_unlock_bh(&bond->lock);
3843 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3844 cancel_delayed_work(&bond->mii_work);
3847 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3848 cancel_delayed_work(&bond->arp_work);
3851 switch (bond->params.mode) {
3852 case BOND_MODE_8023AD:
3853 cancel_delayed_work(&bond->ad_work);
3857 cancel_delayed_work(&bond->alb_work);
3863 if (delayed_work_pending(&bond->mcast_work))
3864 cancel_delayed_work(&bond->mcast_work);
3866 if (bond_is_lb(bond)) {
3867 /* Must be called only after all
3868 * slaves have been released
3870 bond_alb_deinitialize(bond);
3876 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3877 struct rtnl_link_stats64 *stats)
3879 struct bonding *bond = netdev_priv(bond_dev);
3880 struct rtnl_link_stats64 temp;
3881 struct slave *slave;
3884 memset(stats, 0, sizeof(*stats));
3886 read_lock_bh(&bond->lock);
3888 bond_for_each_slave(bond, slave, i) {
3889 const struct rtnl_link_stats64 *sstats =
3890 dev_get_stats(slave->dev, &temp);
3892 stats->rx_packets += sstats->rx_packets;
3893 stats->rx_bytes += sstats->rx_bytes;
3894 stats->rx_errors += sstats->rx_errors;
3895 stats->rx_dropped += sstats->rx_dropped;
3897 stats->tx_packets += sstats->tx_packets;
3898 stats->tx_bytes += sstats->tx_bytes;
3899 stats->tx_errors += sstats->tx_errors;
3900 stats->tx_dropped += sstats->tx_dropped;
3902 stats->multicast += sstats->multicast;
3903 stats->collisions += sstats->collisions;
3905 stats->rx_length_errors += sstats->rx_length_errors;
3906 stats->rx_over_errors += sstats->rx_over_errors;
3907 stats->rx_crc_errors += sstats->rx_crc_errors;
3908 stats->rx_frame_errors += sstats->rx_frame_errors;
3909 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3910 stats->rx_missed_errors += sstats->rx_missed_errors;
3912 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3913 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3914 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3915 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3916 stats->tx_window_errors += sstats->tx_window_errors;
3919 read_unlock_bh(&bond->lock);
3924 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3926 struct net_device *slave_dev = NULL;
3927 struct ifbond k_binfo;
3928 struct ifbond __user *u_binfo = NULL;
3929 struct ifslave k_sinfo;
3930 struct ifslave __user *u_sinfo = NULL;
3931 struct mii_ioctl_data *mii = NULL;
3934 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3946 * We do this again just in case we were called by SIOCGMIIREG
3947 * instead of SIOCGMIIPHY.
3954 if (mii->reg_num == 1) {
3955 struct bonding *bond = netdev_priv(bond_dev);
3957 read_lock(&bond->lock);
3958 read_lock(&bond->curr_slave_lock);
3959 if (netif_carrier_ok(bond->dev))
3960 mii->val_out = BMSR_LSTATUS;
3962 read_unlock(&bond->curr_slave_lock);
3963 read_unlock(&bond->lock);
3967 case BOND_INFO_QUERY_OLD:
3968 case SIOCBONDINFOQUERY:
3969 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3971 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3974 res = bond_info_query(bond_dev, &k_binfo);
3976 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3980 case BOND_SLAVE_INFO_QUERY_OLD:
3981 case SIOCBONDSLAVEINFOQUERY:
3982 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3984 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3987 res = bond_slave_info_query(bond_dev, &k_sinfo);
3989 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3998 if (!capable(CAP_NET_ADMIN))
4001 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4003 pr_debug("slave_dev=%p:\n", slave_dev);
4008 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4010 case BOND_ENSLAVE_OLD:
4011 case SIOCBONDENSLAVE:
4012 res = bond_enslave(bond_dev, slave_dev);
4014 case BOND_RELEASE_OLD:
4015 case SIOCBONDRELEASE:
4016 res = bond_release(bond_dev, slave_dev);
4018 case BOND_SETHWADDR_OLD:
4019 case SIOCBONDSETHWADDR:
4020 res = bond_sethwaddr(bond_dev, slave_dev);
4022 case BOND_CHANGE_ACTIVE_OLD:
4023 case SIOCBONDCHANGEACTIVE:
4024 res = bond_ioctl_change_active(bond_dev, slave_dev);
4036 static bool bond_addr_in_mc_list(unsigned char *addr,
4037 struct netdev_hw_addr_list *list,
4040 struct netdev_hw_addr *ha;
4042 netdev_hw_addr_list_for_each(ha, list)
4043 if (!memcmp(ha->addr, addr, addrlen))
4049 static void bond_set_multicast_list(struct net_device *bond_dev)
4051 struct bonding *bond = netdev_priv(bond_dev);
4052 struct netdev_hw_addr *ha;
4056 * Do promisc before checking multicast_mode
4058 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4060 * FIXME: Need to handle the error when one of the multi-slaves
4063 bond_set_promiscuity(bond, 1);
4066 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4067 bond_set_promiscuity(bond, -1);
4070 /* set allmulti flag to slaves */
4071 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4073 * FIXME: Need to handle the error when one of the multi-slaves
4076 bond_set_allmulti(bond, 1);
4079 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4080 bond_set_allmulti(bond, -1);
4083 read_lock(&bond->lock);
4085 bond->flags = bond_dev->flags;
4087 /* looking for addresses to add to slaves' mc list */
4088 netdev_for_each_mc_addr(ha, bond_dev) {
4089 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4090 bond_dev->addr_len);
4092 bond_mc_add(bond, ha->addr);
4095 /* looking for addresses to delete from slaves' list */
4096 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4097 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4098 bond_dev->addr_len);
4100 bond_mc_del(bond, ha->addr);
4103 /* save master's multicast list */
4104 __hw_addr_flush(&bond->mc_list);
4105 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4106 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4108 read_unlock(&bond->lock);
4111 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4113 struct bonding *bond = netdev_priv(dev);
4114 struct slave *slave = bond->first_slave;
4117 const struct net_device_ops *slave_ops
4118 = slave->dev->netdev_ops;
4119 if (slave_ops->ndo_neigh_setup)
4120 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4126 * Change the MTU of all of a master's slaves to match the master
4128 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4130 struct bonding *bond = netdev_priv(bond_dev);
4131 struct slave *slave, *stop_at;
4135 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4136 (bond_dev ? bond_dev->name : "None"), new_mtu);
4138 /* Can't hold bond->lock with bh disabled here since
4139 * some base drivers panic. On the other hand we can't
4140 * hold bond->lock without bh disabled because we'll
4141 * deadlock. The only solution is to rely on the fact
4142 * that we're under rtnl_lock here, and the slaves
4143 * list won't change. This doesn't solve the problem
4144 * of setting the slave's MTU while it is
4145 * transmitting, but the assumption is that the base
4146 * driver can handle that.
4148 * TODO: figure out a way to safely iterate the slaves
4149 * list, but without holding a lock around the actual
4150 * call to the base driver.
4153 bond_for_each_slave(bond, slave, i) {
4154 pr_debug("s %p s->p %p c_m %p\n",
4157 slave->dev->netdev_ops->ndo_change_mtu);
4159 res = dev_set_mtu(slave->dev, new_mtu);
4162 /* If we failed to set the slave's mtu to the new value
4163 * we must abort the operation even in ACTIVE_BACKUP
4164 * mode, because if we allow the backup slaves to have
4165 * different mtu values than the active slave we'll
4166 * need to change their mtu when doing a failover. That
4167 * means changing their mtu from timer context, which
4168 * is probably not a good idea.
4170 pr_debug("err %d %s\n", res, slave->dev->name);
4175 bond_dev->mtu = new_mtu;
4180 /* unwind from head to the slave that failed */
4182 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4185 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4187 pr_debug("unwind err %d dev %s\n",
4188 tmp_res, slave->dev->name);
4198 * Note that many devices must be down to change the HW address, and
4199 * downing the master releases all slaves. We can make bonds full of
4200 * bonding devices to test this, however.
4202 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4204 struct bonding *bond = netdev_priv(bond_dev);
4205 struct sockaddr *sa = addr, tmp_sa;
4206 struct slave *slave, *stop_at;
4210 if (bond->params.mode == BOND_MODE_ALB)
4211 return bond_alb_set_mac_address(bond_dev, addr);
4214 pr_debug("bond=%p, name=%s\n",
4215 bond, bond_dev ? bond_dev->name : "None");
4218 * If fail_over_mac is set to active, do nothing and return
4219 * success. Returning an error causes ifenslave to fail.
4221 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4224 if (!is_valid_ether_addr(sa->sa_data))
4225 return -EADDRNOTAVAIL;
4227 /* Can't hold bond->lock with bh disabled here since
4228 * some base drivers panic. On the other hand we can't
4229 * hold bond->lock without bh disabled because we'll
4230 * deadlock. The only solution is to rely on the fact
4231 * that we're under rtnl_lock here, and the slaves
4232 * list won't change. This doesn't solve the problem
4233 * of setting the slave's hw address while it is
4234 * transmitting, but the assumption is that the base
4235 * driver can handle that.
4237 * TODO: figure out a way to safely iterate the slaves
4238 * list, but without holding a lock around the actual
4239 * call to the base driver.
4242 bond_for_each_slave(bond, slave, i) {
4243 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4244 pr_debug("slave %p %s\n", slave, slave->dev->name);
4246 if (slave_ops->ndo_set_mac_address == NULL) {
4248 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4252 res = dev_set_mac_address(slave->dev, addr);
4254 /* TODO: consider downing the slave
4256 * User should expect communications
4257 * breakage anyway until ARP finish
4260 pr_debug("err %d %s\n", res, slave->dev->name);
4266 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4270 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4271 tmp_sa.sa_family = bond_dev->type;
4273 /* unwind from head to the slave that failed */
4275 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4278 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4280 pr_debug("unwind err %d dev %s\n",
4281 tmp_res, slave->dev->name);
4288 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4290 struct bonding *bond = netdev_priv(bond_dev);
4291 struct slave *slave, *start_at;
4292 int i, slave_no, res = 1;
4293 struct iphdr *iph = ip_hdr(skb);
4295 read_lock(&bond->lock);
4297 if (!BOND_IS_OK(bond))
4300 * Start with the curr_active_slave that joined the bond as the
4301 * default for sending IGMP traffic. For failover purposes one
4302 * needs to maintain some consistency for the interface that will
4303 * send the join/membership reports. The curr_active_slave found
4304 * will send all of this type of traffic.
4306 if ((iph->protocol == IPPROTO_IGMP) &&
4307 (skb->protocol == htons(ETH_P_IP))) {
4309 read_lock(&bond->curr_slave_lock);
4310 slave = bond->curr_active_slave;
4311 read_unlock(&bond->curr_slave_lock);
4317 * Concurrent TX may collide on rr_tx_counter; we accept
4318 * that as being rare enough not to justify using an
4321 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4323 bond_for_each_slave(bond, slave, i) {
4331 bond_for_each_slave_from(bond, slave, i, start_at) {
4332 if (IS_UP(slave->dev) &&
4333 (slave->link == BOND_LINK_UP) &&
4334 (slave->state == BOND_STATE_ACTIVE)) {
4335 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4342 /* no suitable interface, frame not sent */
4345 read_unlock(&bond->lock);
4346 return NETDEV_TX_OK;
4351 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4352 * the bond has a usable interface.
4354 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4356 struct bonding *bond = netdev_priv(bond_dev);
4359 read_lock(&bond->lock);
4360 read_lock(&bond->curr_slave_lock);
4362 if (!BOND_IS_OK(bond))
4365 if (!bond->curr_active_slave)
4368 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4372 /* no suitable interface, frame not sent */
4375 read_unlock(&bond->curr_slave_lock);
4376 read_unlock(&bond->lock);
4377 return NETDEV_TX_OK;
4381 * In bond_xmit_xor() , we determine the output device by using a pre-
4382 * determined xmit_hash_policy(), If the selected device is not enabled,
4383 * find the next active slave.
4385 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4387 struct bonding *bond = netdev_priv(bond_dev);
4388 struct slave *slave, *start_at;
4393 read_lock(&bond->lock);
4395 if (!BOND_IS_OK(bond))
4398 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4400 bond_for_each_slave(bond, slave, i) {
4408 bond_for_each_slave_from(bond, slave, i, start_at) {
4409 if (IS_UP(slave->dev) &&
4410 (slave->link == BOND_LINK_UP) &&
4411 (slave->state == BOND_STATE_ACTIVE)) {
4412 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4419 /* no suitable interface, frame not sent */
4422 read_unlock(&bond->lock);
4423 return NETDEV_TX_OK;
4427 * in broadcast mode, we send everything to all usable interfaces.
4429 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4431 struct bonding *bond = netdev_priv(bond_dev);
4432 struct slave *slave, *start_at;
4433 struct net_device *tx_dev = NULL;
4437 read_lock(&bond->lock);
4439 if (!BOND_IS_OK(bond))
4442 read_lock(&bond->curr_slave_lock);
4443 start_at = bond->curr_active_slave;
4444 read_unlock(&bond->curr_slave_lock);
4449 bond_for_each_slave_from(bond, slave, i, start_at) {
4450 if (IS_UP(slave->dev) &&
4451 (slave->link == BOND_LINK_UP) &&
4452 (slave->state == BOND_STATE_ACTIVE)) {
4454 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4456 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4461 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4463 dev_kfree_skb(skb2);
4467 tx_dev = slave->dev;
4472 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4476 /* no suitable interface, frame not sent */
4479 /* frame sent to all suitable interfaces */
4480 read_unlock(&bond->lock);
4481 return NETDEV_TX_OK;
4484 /*------------------------- Device initialization ---------------------------*/
4486 static void bond_set_xmit_hash_policy(struct bonding *bond)
4488 switch (bond->params.xmit_policy) {
4489 case BOND_XMIT_POLICY_LAYER23:
4490 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4492 case BOND_XMIT_POLICY_LAYER34:
4493 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4495 case BOND_XMIT_POLICY_LAYER2:
4497 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4503 * Lookup the slave that corresponds to a qid
4505 static inline int bond_slave_override(struct bonding *bond,
4506 struct sk_buff *skb)
4509 struct slave *slave = NULL;
4510 struct slave *check_slave;
4512 read_lock(&bond->lock);
4514 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4517 /* Find out if any slaves have the same mapping as this skb. */
4518 bond_for_each_slave(bond, check_slave, i) {
4519 if (check_slave->queue_id == skb->queue_mapping) {
4520 slave = check_slave;
4525 /* If the slave isn't UP, use default transmit policy. */
4526 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4527 (slave->link == BOND_LINK_UP)) {
4528 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4532 read_unlock(&bond->lock);
4536 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4539 * This helper function exists to help dev_pick_tx get the correct
4540 * destination queue. Using a helper function skips the a call to
4541 * skb_tx_hash and will put the skbs in the queue we expect on their
4542 * way down to the bonding driver.
4544 return skb->queue_mapping;
4547 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4549 struct bonding *bond = netdev_priv(dev);
4552 * If we risk deadlock from transmitting this in the
4553 * netpoll path, tell netpoll to queue the frame for later tx
4555 if (is_netpoll_tx_blocked(dev))
4556 return NETDEV_TX_BUSY;
4558 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4559 if (!bond_slave_override(bond, skb))
4560 return NETDEV_TX_OK;
4563 switch (bond->params.mode) {
4564 case BOND_MODE_ROUNDROBIN:
4565 return bond_xmit_roundrobin(skb, dev);
4566 case BOND_MODE_ACTIVEBACKUP:
4567 return bond_xmit_activebackup(skb, dev);
4569 return bond_xmit_xor(skb, dev);
4570 case BOND_MODE_BROADCAST:
4571 return bond_xmit_broadcast(skb, dev);
4572 case BOND_MODE_8023AD:
4573 return bond_3ad_xmit_xor(skb, dev);
4576 return bond_alb_xmit(skb, dev);
4578 /* Should never happen, mode already checked */
4579 pr_err("%s: Error: Unknown bonding mode %d\n",
4580 dev->name, bond->params.mode);
4583 return NETDEV_TX_OK;
4589 * set bond mode specific net device operations
4591 void bond_set_mode_ops(struct bonding *bond, int mode)
4593 struct net_device *bond_dev = bond->dev;
4596 case BOND_MODE_ROUNDROBIN:
4598 case BOND_MODE_ACTIVEBACKUP:
4601 bond_set_xmit_hash_policy(bond);
4603 case BOND_MODE_BROADCAST:
4605 case BOND_MODE_8023AD:
4606 bond_set_master_3ad_flags(bond);
4607 bond_set_xmit_hash_policy(bond);
4610 bond_set_master_alb_flags(bond);
4615 /* Should never happen, mode already checked */
4616 pr_err("%s: Error: Unknown bonding mode %d\n",
4617 bond_dev->name, mode);
4622 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4623 struct ethtool_drvinfo *drvinfo)
4625 strncpy(drvinfo->driver, DRV_NAME, 32);
4626 strncpy(drvinfo->version, DRV_VERSION, 32);
4627 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4630 static const struct ethtool_ops bond_ethtool_ops = {
4631 .get_drvinfo = bond_ethtool_get_drvinfo,
4632 .get_link = ethtool_op_get_link,
4633 .get_tx_csum = ethtool_op_get_tx_csum,
4634 .get_sg = ethtool_op_get_sg,
4635 .get_tso = ethtool_op_get_tso,
4636 .get_ufo = ethtool_op_get_ufo,
4637 .get_flags = ethtool_op_get_flags,
4640 static const struct net_device_ops bond_netdev_ops = {
4641 .ndo_init = bond_init,
4642 .ndo_uninit = bond_uninit,
4643 .ndo_open = bond_open,
4644 .ndo_stop = bond_close,
4645 .ndo_start_xmit = bond_start_xmit,
4646 .ndo_select_queue = bond_select_queue,
4647 .ndo_get_stats64 = bond_get_stats,
4648 .ndo_do_ioctl = bond_do_ioctl,
4649 .ndo_set_multicast_list = bond_set_multicast_list,
4650 .ndo_change_mtu = bond_change_mtu,
4651 .ndo_set_mac_address = bond_set_mac_address,
4652 .ndo_neigh_setup = bond_neigh_setup,
4653 .ndo_vlan_rx_register = bond_vlan_rx_register,
4654 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4655 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4656 #ifdef CONFIG_NET_POLL_CONTROLLER
4657 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4658 .ndo_poll_controller = bond_poll_controller,
4662 static void bond_destructor(struct net_device *bond_dev)
4664 struct bonding *bond = netdev_priv(bond_dev);
4666 destroy_workqueue(bond->wq);
4667 free_netdev(bond_dev);
4670 static void bond_setup(struct net_device *bond_dev)
4672 struct bonding *bond = netdev_priv(bond_dev);
4674 /* initialize rwlocks */
4675 rwlock_init(&bond->lock);
4676 rwlock_init(&bond->curr_slave_lock);
4678 bond->params = bonding_defaults;
4680 /* Initialize pointers */
4681 bond->dev = bond_dev;
4682 INIT_LIST_HEAD(&bond->vlan_list);
4684 /* Initialize the device entry points */
4685 ether_setup(bond_dev);
4686 bond_dev->netdev_ops = &bond_netdev_ops;
4687 bond_dev->ethtool_ops = &bond_ethtool_ops;
4688 bond_set_mode_ops(bond, bond->params.mode);
4690 bond_dev->destructor = bond_destructor;
4692 /* Initialize the device options */
4693 bond_dev->tx_queue_len = 0;
4694 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4695 bond_dev->priv_flags |= IFF_BONDING;
4696 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4698 if (bond->params.arp_interval)
4699 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4701 /* At first, we block adding VLANs. That's the only way to
4702 * prevent problems that occur when adding VLANs over an
4703 * empty bond. The block will be removed once non-challenged
4704 * slaves are enslaved.
4706 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4708 /* don't acquire bond device's netif_tx_lock when
4710 bond_dev->features |= NETIF_F_LLTX;
4712 /* By default, we declare the bond to be fully
4713 * VLAN hardware accelerated capable. Special
4714 * care is taken in the various xmit functions
4715 * when there are slaves that are not hw accel
4718 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4719 NETIF_F_HW_VLAN_RX |
4720 NETIF_F_HW_VLAN_FILTER);
4722 /* By default, we enable GRO on bonding devices.
4723 * Actual support requires lowlevel drivers are GRO ready.
4725 bond_dev->features |= NETIF_F_GRO;
4728 static void bond_work_cancel_all(struct bonding *bond)
4730 write_lock_bh(&bond->lock);
4731 bond->kill_timers = 1;
4732 write_unlock_bh(&bond->lock);
4734 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4735 cancel_delayed_work(&bond->mii_work);
4737 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4738 cancel_delayed_work(&bond->arp_work);
4740 if (bond->params.mode == BOND_MODE_ALB &&
4741 delayed_work_pending(&bond->alb_work))
4742 cancel_delayed_work(&bond->alb_work);
4744 if (bond->params.mode == BOND_MODE_8023AD &&
4745 delayed_work_pending(&bond->ad_work))
4746 cancel_delayed_work(&bond->ad_work);
4748 if (delayed_work_pending(&bond->mcast_work))
4749 cancel_delayed_work(&bond->mcast_work);
4753 * Destroy a bonding device.
4754 * Must be under rtnl_lock when this function is called.
4756 static void bond_uninit(struct net_device *bond_dev)
4758 struct bonding *bond = netdev_priv(bond_dev);
4759 struct vlan_entry *vlan, *tmp;
4761 bond_netpoll_cleanup(bond_dev);
4763 /* Release the bonded slaves */
4764 bond_release_all(bond_dev);
4766 list_del(&bond->bond_list);
4768 bond_work_cancel_all(bond);
4770 bond_remove_proc_entry(bond);
4772 __hw_addr_flush(&bond->mc_list);
4774 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4775 list_del(&vlan->vlan_list);
4780 /*------------------------- Module initialization ---------------------------*/
4783 * Convert string input module parms. Accept either the
4784 * number of the mode or its string name. A bit complicated because
4785 * some mode names are substrings of other names, and calls from sysfs
4786 * may have whitespace in the name (trailing newlines, for example).
4788 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4790 int modeint = -1, i, rv;
4791 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4793 for (p = (char *)buf; *p; p++)
4794 if (!(isdigit(*p) || isspace(*p)))
4798 rv = sscanf(buf, "%20s", modestr);
4800 rv = sscanf(buf, "%d", &modeint);
4805 for (i = 0; tbl[i].modename; i++) {
4806 if (modeint == tbl[i].mode)
4808 if (strcmp(modestr, tbl[i].modename) == 0)
4815 static int bond_check_params(struct bond_params *params)
4817 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4820 * Convert string parameters.
4823 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4824 if (bond_mode == -1) {
4825 pr_err("Error: Invalid bonding mode \"%s\"\n",
4826 mode == NULL ? "NULL" : mode);
4831 if (xmit_hash_policy) {
4832 if ((bond_mode != BOND_MODE_XOR) &&
4833 (bond_mode != BOND_MODE_8023AD)) {
4834 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4835 bond_mode_name(bond_mode));
4837 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4839 if (xmit_hashtype == -1) {
4840 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4841 xmit_hash_policy == NULL ? "NULL" :
4849 if (bond_mode != BOND_MODE_8023AD) {
4850 pr_info("lacp_rate param is irrelevant in mode %s\n",
4851 bond_mode_name(bond_mode));
4853 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4854 if (lacp_fast == -1) {
4855 pr_err("Error: Invalid lacp rate \"%s\"\n",
4856 lacp_rate == NULL ? "NULL" : lacp_rate);
4863 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4864 if (params->ad_select == -1) {
4865 pr_err("Error: Invalid ad_select \"%s\"\n",
4866 ad_select == NULL ? "NULL" : ad_select);
4870 if (bond_mode != BOND_MODE_8023AD) {
4871 pr_warning("ad_select param only affects 802.3ad mode\n");
4874 params->ad_select = BOND_AD_STABLE;
4877 if (max_bonds < 0) {
4878 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4879 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4880 max_bonds = BOND_DEFAULT_MAX_BONDS;
4884 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4885 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4886 miimon = BOND_LINK_MON_INTERV;
4890 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4895 if (downdelay < 0) {
4896 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4897 downdelay, INT_MAX);
4901 if ((use_carrier != 0) && (use_carrier != 1)) {
4902 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4907 if (num_grat_arp < 0 || num_grat_arp > 255) {
4908 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4913 if (num_unsol_na < 0 || num_unsol_na > 255) {
4914 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4919 /* reset values for 802.3ad */
4920 if (bond_mode == BOND_MODE_8023AD) {
4922 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");
4923 pr_warning("Forcing miimon to 100msec\n");
4928 if (tx_queues < 1 || tx_queues > 255) {
4929 pr_warning("Warning: tx_queues (%d) should be between "
4930 "1 and 255, resetting to %d\n",
4931 tx_queues, BOND_DEFAULT_TX_QUEUES);
4932 tx_queues = BOND_DEFAULT_TX_QUEUES;
4935 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4936 pr_warning("Warning: all_slaves_active module parameter (%d), "
4937 "not of valid value (0/1), so it was set to "
4938 "0\n", all_slaves_active);
4939 all_slaves_active = 0;
4942 if (resend_igmp < 0 || resend_igmp > 255) {
4943 pr_warning("Warning: resend_igmp (%d) should be between "
4944 "0 and 255, resetting to %d\n",
4945 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4946 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4949 /* reset values for TLB/ALB */
4950 if ((bond_mode == BOND_MODE_TLB) ||
4951 (bond_mode == BOND_MODE_ALB)) {
4953 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");
4954 pr_warning("Forcing miimon to 100msec\n");
4959 if (bond_mode == BOND_MODE_ALB) {
4960 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",
4965 if (updelay || downdelay) {
4966 /* just warn the user the up/down delay will have
4967 * no effect since miimon is zero...
4969 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",
4970 updelay, downdelay);
4973 /* don't allow arp monitoring */
4975 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4976 miimon, arp_interval);
4980 if ((updelay % miimon) != 0) {
4981 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4983 (updelay / miimon) * miimon);
4988 if ((downdelay % miimon) != 0) {
4989 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4991 (downdelay / miimon) * miimon);
4994 downdelay /= miimon;
4997 if (arp_interval < 0) {
4998 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4999 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5000 arp_interval = BOND_LINK_ARP_INTERV;
5003 for (arp_ip_count = 0;
5004 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5006 /* not complete check, but should be good enough to
5008 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5009 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5010 arp_ip_target[arp_ip_count]);
5013 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5014 arp_target[arp_ip_count] = ip;
5018 if (arp_interval && !arp_ip_count) {
5019 /* don't allow arping if no arp_ip_target given... */
5020 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5026 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5027 pr_err("arp_validate only supported in active-backup mode\n");
5030 if (!arp_interval) {
5031 pr_err("arp_validate requires arp_interval\n");
5035 arp_validate_value = bond_parse_parm(arp_validate,
5037 if (arp_validate_value == -1) {
5038 pr_err("Error: invalid arp_validate \"%s\"\n",
5039 arp_validate == NULL ? "NULL" : arp_validate);
5043 arp_validate_value = 0;
5046 pr_info("MII link monitoring set to %d ms\n", miimon);
5047 } else if (arp_interval) {
5050 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5052 arp_validate_tbl[arp_validate_value].modename,
5055 for (i = 0; i < arp_ip_count; i++)
5056 pr_info(" %s", arp_ip_target[i]);
5060 } else if (max_bonds) {
5061 /* miimon and arp_interval not set, we need one so things
5062 * work as expected, see bonding.txt for details
5064 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");
5067 if (primary && !USES_PRIMARY(bond_mode)) {
5068 /* currently, using a primary only makes sense
5069 * in active backup, TLB or ALB modes
5071 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5072 primary, bond_mode_name(bond_mode));
5076 if (primary && primary_reselect) {
5077 primary_reselect_value = bond_parse_parm(primary_reselect,
5079 if (primary_reselect_value == -1) {
5080 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5082 NULL ? "NULL" : primary_reselect);
5086 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5089 if (fail_over_mac) {
5090 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5092 if (fail_over_mac_value == -1) {
5093 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5094 arp_validate == NULL ? "NULL" : arp_validate);
5098 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5099 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5101 fail_over_mac_value = BOND_FOM_NONE;
5104 /* fill params struct with the proper values */
5105 params->mode = bond_mode;
5106 params->xmit_policy = xmit_hashtype;
5107 params->miimon = miimon;
5108 params->num_grat_arp = num_grat_arp;
5109 params->num_unsol_na = num_unsol_na;
5110 params->arp_interval = arp_interval;
5111 params->arp_validate = arp_validate_value;
5112 params->updelay = updelay;
5113 params->downdelay = downdelay;
5114 params->use_carrier = use_carrier;
5115 params->lacp_fast = lacp_fast;
5116 params->primary[0] = 0;
5117 params->primary_reselect = primary_reselect_value;
5118 params->fail_over_mac = fail_over_mac_value;
5119 params->tx_queues = tx_queues;
5120 params->all_slaves_active = all_slaves_active;
5121 params->resend_igmp = resend_igmp;
5124 strncpy(params->primary, primary, IFNAMSIZ);
5125 params->primary[IFNAMSIZ - 1] = 0;
5128 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5133 static struct lock_class_key bonding_netdev_xmit_lock_key;
5134 static struct lock_class_key bonding_netdev_addr_lock_key;
5136 static void bond_set_lockdep_class_one(struct net_device *dev,
5137 struct netdev_queue *txq,
5140 lockdep_set_class(&txq->_xmit_lock,
5141 &bonding_netdev_xmit_lock_key);
5144 static void bond_set_lockdep_class(struct net_device *dev)
5146 lockdep_set_class(&dev->addr_list_lock,
5147 &bonding_netdev_addr_lock_key);
5148 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5152 * Called from registration process
5154 static int bond_init(struct net_device *bond_dev)
5156 struct bonding *bond = netdev_priv(bond_dev);
5157 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5159 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5161 bond->wq = create_singlethread_workqueue(bond_dev->name);
5165 bond_set_lockdep_class(bond_dev);
5167 netif_carrier_off(bond_dev);
5169 bond_create_proc_entry(bond);
5170 list_add_tail(&bond->bond_list, &bn->dev_list);
5172 bond_prepare_sysfs_group(bond);
5174 __hw_addr_init(&bond->mc_list);
5178 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5180 if (tb[IFLA_ADDRESS]) {
5181 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5183 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5184 return -EADDRNOTAVAIL;
5189 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5191 .priv_size = sizeof(struct bonding),
5192 .setup = bond_setup,
5193 .validate = bond_validate,
5196 /* Create a new bond based on the specified name and bonding parameters.
5197 * If name is NULL, obtain a suitable "bond%d" name for us.
5198 * Caller must NOT hold rtnl_lock; we need to release it here before we
5199 * set up our sysfs entries.
5201 int bond_create(struct net *net, const char *name)
5203 struct net_device *bond_dev;
5208 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5209 bond_setup, tx_queues);
5211 pr_err("%s: eek! can't alloc netdev!\n", name);
5216 dev_net_set(bond_dev, net);
5217 bond_dev->rtnl_link_ops = &bond_link_ops;
5220 res = dev_alloc_name(bond_dev, "bond%d");
5225 * If we're given a name to register
5226 * we need to ensure that its not already
5230 if (__dev_get_by_name(net, name) != NULL)
5234 res = register_netdevice(bond_dev);
5239 bond_destructor(bond_dev);
5243 static int __net_init bond_net_init(struct net *net)
5245 struct bond_net *bn = net_generic(net, bond_net_id);
5248 INIT_LIST_HEAD(&bn->dev_list);
5250 bond_create_proc_dir(bn);
5255 static void __net_exit bond_net_exit(struct net *net)
5257 struct bond_net *bn = net_generic(net, bond_net_id);
5259 bond_destroy_proc_dir(bn);
5262 static struct pernet_operations bond_net_ops = {
5263 .init = bond_net_init,
5264 .exit = bond_net_exit,
5266 .size = sizeof(struct bond_net),
5269 static int __init bonding_init(void)
5274 pr_info("%s", version);
5276 res = bond_check_params(&bonding_defaults);
5280 res = register_pernet_subsys(&bond_net_ops);
5284 res = rtnl_link_register(&bond_link_ops);
5288 for (i = 0; i < max_bonds; i++) {
5289 res = bond_create(&init_net, NULL);
5294 res = bond_create_sysfs();
5299 register_netdevice_notifier(&bond_netdev_notifier);
5300 register_inetaddr_notifier(&bond_inetaddr_notifier);
5301 bond_register_ipv6_notifier();
5305 rtnl_link_unregister(&bond_link_ops);
5307 unregister_pernet_subsys(&bond_net_ops);
5312 static void __exit bonding_exit(void)
5314 unregister_netdevice_notifier(&bond_netdev_notifier);
5315 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5316 bond_unregister_ipv6_notifier();
5318 bond_destroy_sysfs();
5320 rtnl_link_unregister(&bond_link_ops);
5321 unregister_pernet_subsys(&bond_net_ops);
5323 #ifdef CONFIG_NET_POLL_CONTROLLER
5325 * Make sure we don't have an imbalance on our netpoll blocking
5327 WARN_ON(atomic_read(&netpoll_block_tx));
5331 module_init(bonding_init);
5332 module_exit(bonding_exit);
5333 MODULE_LICENSE("GPL");
5334 MODULE_VERSION(DRV_VERSION);
5335 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5336 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5337 MODULE_ALIAS_RTNL_LINK("bond");