2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
57 #include <asm/system.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <linux/preempt.h>
77 #include <net/route.h>
78 #include <net/net_namespace.h>
79 #include <net/netns/generic.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV 0
88 #define BOND_LINK_ARP_INTERV 0
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
93 static int miimon = BOND_LINK_MON_INTERV;
96 static int use_carrier = 1;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval = BOND_LINK_ARP_INTERV;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *fail_over_mac;
108 static int all_slaves_active = 0;
109 static struct bond_params bonding_defaults;
110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 module_param(max_bonds, int, 0);
113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
114 module_param(tx_queues, int, 0);
115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
118 "failover event (alias of num_unsol_na)");
119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
121 "failover event (alias of num_grat_arp)");
122 module_param(miimon, int, 0);
123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124 module_param(updelay, int, 0);
125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126 module_param(downdelay, int, 0);
127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
129 module_param(use_carrier, int, 0);
130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132 module_param(mode, charp, 0);
133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
137 module_param(primary, charp, 0);
138 MODULE_PARM_DESC(primary, "Primary network device to use");
139 module_param(primary_reselect, charp, 0);
140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
145 "2 for only on active slave "
147 module_param(lacp_rate, charp, 0);
148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
149 "0 for slow, 1 for fast");
150 module_param(ad_select, charp, 0);
151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
152 "0 for stable (default), 1 for bandwidth, "
154 module_param(min_links, int, 0);
155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
157 module_param(xmit_hash_policy, charp, 0);
158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
159 "0 for layer 2 (default), 1 for layer 3+4, "
161 module_param(arp_interval, int, 0);
162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
163 module_param_array(arp_ip_target, charp, NULL, 0);
164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
165 module_param(arp_validate, charp, 0);
166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
167 "0 for none (default), 1 for active, "
168 "2 for backup, 3 for all");
169 module_param(fail_over_mac, charp, 0);
170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
171 "the same MAC; 0 for none (default), "
172 "1 for active, 2 for follow");
173 module_param(all_slaves_active, int, 0);
174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
175 "by setting active flag for all slaves; "
176 "0 for never (default), 1 for always.");
177 module_param(resend_igmp, int, 0);
178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
181 /*----------------------------- Global variables ----------------------------*/
183 #ifdef CONFIG_NET_POLL_CONTROLLER
184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
187 int bond_net_id __read_mostly;
189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
190 static int arp_ip_count;
191 static int bond_mode = BOND_MODE_ROUNDROBIN;
192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
193 static int lacp_fast;
195 const struct bond_parm_tbl bond_lacp_tbl[] = {
196 { "slow", AD_LACP_SLOW},
197 { "fast", AD_LACP_FAST},
201 const struct bond_parm_tbl bond_mode_tbl[] = {
202 { "balance-rr", BOND_MODE_ROUNDROBIN},
203 { "active-backup", BOND_MODE_ACTIVEBACKUP},
204 { "balance-xor", BOND_MODE_XOR},
205 { "broadcast", BOND_MODE_BROADCAST},
206 { "802.3ad", BOND_MODE_8023AD},
207 { "balance-tlb", BOND_MODE_TLB},
208 { "balance-alb", BOND_MODE_ALB},
212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
213 { "layer2", BOND_XMIT_POLICY_LAYER2},
214 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
215 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
219 const struct bond_parm_tbl arp_validate_tbl[] = {
220 { "none", BOND_ARP_VALIDATE_NONE},
221 { "active", BOND_ARP_VALIDATE_ACTIVE},
222 { "backup", BOND_ARP_VALIDATE_BACKUP},
223 { "all", BOND_ARP_VALIDATE_ALL},
227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
228 { "none", BOND_FOM_NONE},
229 { "active", BOND_FOM_ACTIVE},
230 { "follow", BOND_FOM_FOLLOW},
234 const struct bond_parm_tbl pri_reselect_tbl[] = {
235 { "always", BOND_PRI_RESELECT_ALWAYS},
236 { "better", BOND_PRI_RESELECT_BETTER},
237 { "failure", BOND_PRI_RESELECT_FAILURE},
241 struct bond_parm_tbl ad_select_tbl[] = {
242 { "stable", BOND_AD_STABLE},
243 { "bandwidth", BOND_AD_BANDWIDTH},
244 { "count", BOND_AD_COUNT},
248 /*-------------------------- Forward declarations ---------------------------*/
250 static int bond_init(struct net_device *bond_dev);
251 static void bond_uninit(struct net_device *bond_dev);
253 /*---------------------------- General routines -----------------------------*/
255 const char *bond_mode_name(int mode)
257 static const char *names[] = {
258 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
259 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
260 [BOND_MODE_XOR] = "load balancing (xor)",
261 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
262 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
263 [BOND_MODE_TLB] = "transmit load balancing",
264 [BOND_MODE_ALB] = "adaptive load balancing",
267 if (mode < 0 || mode > BOND_MODE_ALB)
273 /*---------------------------------- VLAN -----------------------------------*/
276 * bond_add_vlan - add a new vlan id on bond
277 * @bond: bond that got the notification
278 * @vlan_id: the vlan id to add
280 * Returns -ENOMEM if allocation failed.
282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
284 struct vlan_entry *vlan;
286 pr_debug("bond: %s, vlan id %d\n",
287 (bond ? bond->dev->name : "None"), vlan_id);
289 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
293 INIT_LIST_HEAD(&vlan->vlan_list);
294 vlan->vlan_id = vlan_id;
296 write_lock_bh(&bond->lock);
298 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
300 write_unlock_bh(&bond->lock);
302 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
308 * bond_del_vlan - delete a vlan id from bond
309 * @bond: bond that got the notification
310 * @vlan_id: the vlan id to delete
312 * returns -ENODEV if @vlan_id was not found in @bond.
314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
316 struct vlan_entry *vlan;
319 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
322 write_lock_bh(&bond->lock);
324 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
325 if (vlan->vlan_id == vlan_id) {
326 list_del(&vlan->vlan_list);
328 if (bond_is_lb(bond))
329 bond_alb_clear_vlan(bond, vlan_id);
331 pr_debug("removed VLAN ID %d from bond %s\n",
332 vlan_id, bond->dev->name);
341 pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 vlan_id, bond->dev->name);
345 write_unlock_bh(&bond->lock);
346 unblock_netpoll_tx();
351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
358 * Caller must hold bond->lock
360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
362 struct vlan_entry *next, *last;
364 if (list_empty(&bond->vlan_list))
368 next = list_entry(bond->vlan_list.next,
369 struct vlan_entry, vlan_list);
371 last = list_entry(bond->vlan_list.prev,
372 struct vlan_entry, vlan_list);
374 next = list_entry(bond->vlan_list.next,
375 struct vlan_entry, vlan_list);
377 next = list_entry(curr->vlan_list.next,
378 struct vlan_entry, vlan_list);
385 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
388 * bond_dev_queue_xmit - Prepare skb for xmit.
390 * @bond: bond device that got this skb for tx.
391 * @skb: hw accel VLAN tagged skb to transmit
392 * @slave_dev: slave that is supposed to xmit this skbuff
394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
395 struct net_device *slave_dev)
397 skb->dev = slave_dev;
399 skb->queue_mapping = bond_queue_mapping(skb);
401 if (unlikely(netpoll_tx_running(slave_dev)))
402 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
410 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411 * We don't protect the slave list iteration with a lock because:
412 * a. This operation is performed in IOCTL context,
413 * b. The operation is protected by the RTNL semaphore in the 8021q code,
414 * c. Holding a lock with BH disabled while directly calling a base driver
415 * entry point is generally a BAD idea.
417 * The design of synchronization/protection for this operation in the 8021q
418 * module is good for one or more VLAN devices over a single physical device
419 * and cannot be extended for a teaming solution like bonding, so there is a
420 * potential race condition here where a net device from the vlan group might
421 * be referenced (either by a base driver or the 8021q code) while it is being
422 * removed from the system. However, it turns out we're not making matters
423 * worse, and if it works for regular VLAN usage it will work here too.
427 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428 * @bond_dev: bonding net device that got called
429 * @vid: vlan id being added
431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
433 struct bonding *bond = netdev_priv(bond_dev);
434 struct slave *slave, *stop_at;
437 bond_for_each_slave(bond, slave, i) {
438 res = vlan_vid_add(slave->dev, vid);
443 res = bond_add_vlan(bond, vid);
445 pr_err("%s: Error: Failed to add vlan id %d\n",
446 bond_dev->name, vid);
453 /* unwind from head to the slave that failed */
455 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
456 vlan_vid_del(slave->dev, vid);
462 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
463 * @bond_dev: bonding net device that got called
464 * @vid: vlan id being removed
466 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
468 struct bonding *bond = netdev_priv(bond_dev);
472 bond_for_each_slave(bond, slave, i)
473 vlan_vid_del(slave->dev, vid);
475 res = bond_del_vlan(bond, vid);
477 pr_err("%s: Error: Failed to remove vlan id %d\n",
478 bond_dev->name, vid);
485 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
487 struct vlan_entry *vlan;
490 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
491 res = vlan_vid_add(slave_dev, vlan->vlan_id);
493 pr_warning("%s: Failed to add vlan id %d to device %s\n",
494 bond->dev->name, vlan->vlan_id,
499 static void bond_del_vlans_from_slave(struct bonding *bond,
500 struct net_device *slave_dev)
502 struct vlan_entry *vlan;
504 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
507 vlan_vid_del(slave_dev, vlan->vlan_id);
511 /*------------------------------- Link status -------------------------------*/
514 * Set the carrier state for the master according to the state of its
515 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
516 * do special 802.3ad magic.
518 * Returns zero if carrier state does not change, nonzero if it does.
520 static int bond_set_carrier(struct bonding *bond)
525 if (bond->slave_cnt == 0)
528 if (bond->params.mode == BOND_MODE_8023AD)
529 return bond_3ad_set_carrier(bond);
531 bond_for_each_slave(bond, slave, i) {
532 if (slave->link == BOND_LINK_UP) {
533 if (!netif_carrier_ok(bond->dev)) {
534 netif_carrier_on(bond->dev);
542 if (netif_carrier_ok(bond->dev)) {
543 netif_carrier_off(bond->dev);
550 * Get link speed and duplex from the slave's base driver
551 * using ethtool. If for some reason the call fails or the
552 * values are invalid, set speed and duplex to -1,
555 static int bond_update_speed_duplex(struct slave *slave)
557 struct net_device *slave_dev = slave->dev;
558 struct ethtool_cmd ecmd;
562 slave->speed = SPEED_UNKNOWN;
563 slave->duplex = DUPLEX_UNKNOWN;
565 res = __ethtool_get_settings(slave_dev, &ecmd);
569 slave_speed = ethtool_cmd_speed(&ecmd);
570 if (slave_speed == 0 || slave_speed == ((__u32) -1))
573 switch (ecmd.duplex) {
581 slave->speed = slave_speed;
582 slave->duplex = ecmd.duplex;
588 * if <dev> supports MII link status reporting, check its link status.
590 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
591 * depending upon the setting of the use_carrier parameter.
593 * Return either BMSR_LSTATUS, meaning that the link is up (or we
594 * can't tell and just pretend it is), or 0, meaning that the link is
597 * If reporting is non-zero, instead of faking link up, return -1 if
598 * both ETHTOOL and MII ioctls fail (meaning the device does not
599 * support them). If use_carrier is set, return whatever it says.
600 * It'd be nice if there was a good way to tell if a driver supports
601 * netif_carrier, but there really isn't.
603 static int bond_check_dev_link(struct bonding *bond,
604 struct net_device *slave_dev, int reporting)
606 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
607 int (*ioctl)(struct net_device *, struct ifreq *, int);
609 struct mii_ioctl_data *mii;
611 if (!reporting && !netif_running(slave_dev))
614 if (bond->params.use_carrier)
615 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
617 /* Try to get link status using Ethtool first. */
618 if (slave_dev->ethtool_ops) {
619 if (slave_dev->ethtool_ops->get_link) {
622 link = slave_dev->ethtool_ops->get_link(slave_dev);
624 return link ? BMSR_LSTATUS : 0;
628 /* Ethtool can't be used, fallback to MII ioctls. */
629 ioctl = slave_ops->ndo_do_ioctl;
631 /* TODO: set pointer to correct ioctl on a per team member */
632 /* bases to make this more efficient. that is, once */
633 /* we determine the correct ioctl, we will always */
634 /* call it and not the others for that team */
638 * We cannot assume that SIOCGMIIPHY will also read a
639 * register; not all network drivers (e.g., e100)
643 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
644 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
646 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
647 mii->reg_num = MII_BMSR;
648 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
649 return mii->val_out & BMSR_LSTATUS;
654 * If reporting, report that either there's no dev->do_ioctl,
655 * or both SIOCGMIIREG and get_link failed (meaning that we
656 * cannot report link status). If not reporting, pretend
659 return reporting ? -1 : BMSR_LSTATUS;
662 /*----------------------------- Multicast list ------------------------------*/
665 * Push the promiscuity flag down to appropriate slaves
667 static int bond_set_promiscuity(struct bonding *bond, int inc)
670 if (USES_PRIMARY(bond->params.mode)) {
671 /* write lock already acquired */
672 if (bond->curr_active_slave) {
673 err = dev_set_promiscuity(bond->curr_active_slave->dev,
679 bond_for_each_slave(bond, slave, i) {
680 err = dev_set_promiscuity(slave->dev, inc);
689 * Push the allmulti flag down to all slaves
691 static int bond_set_allmulti(struct bonding *bond, int inc)
694 if (USES_PRIMARY(bond->params.mode)) {
695 /* write lock already acquired */
696 if (bond->curr_active_slave) {
697 err = dev_set_allmulti(bond->curr_active_slave->dev,
703 bond_for_each_slave(bond, slave, i) {
704 err = dev_set_allmulti(slave->dev, inc);
713 * Add a Multicast address to slaves
716 static void bond_mc_add(struct bonding *bond, void *addr)
718 if (USES_PRIMARY(bond->params.mode)) {
719 /* write lock already acquired */
720 if (bond->curr_active_slave)
721 dev_mc_add(bond->curr_active_slave->dev, addr);
726 bond_for_each_slave(bond, slave, i)
727 dev_mc_add(slave->dev, addr);
732 * Remove a multicast address from slave
735 static void bond_mc_del(struct bonding *bond, void *addr)
737 if (USES_PRIMARY(bond->params.mode)) {
738 /* write lock already acquired */
739 if (bond->curr_active_slave)
740 dev_mc_del(bond->curr_active_slave->dev, addr);
744 bond_for_each_slave(bond, slave, i) {
745 dev_mc_del(slave->dev, addr);
751 static void __bond_resend_igmp_join_requests(struct net_device *dev)
753 struct in_device *in_dev;
756 in_dev = __in_dev_get_rcu(dev);
758 ip_mc_rejoin_groups(in_dev);
763 * Retrieve the list of registered multicast addresses for the bonding
764 * device and retransmit an IGMP JOIN request to the current active
767 static void bond_resend_igmp_join_requests(struct bonding *bond)
769 struct net_device *vlan_dev;
770 struct vlan_entry *vlan;
772 read_lock(&bond->lock);
774 /* rejoin all groups on bond device */
775 __bond_resend_igmp_join_requests(bond->dev);
777 /* rejoin all groups on vlan devices */
778 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
780 vlan_dev = __vlan_find_dev_deep(bond->dev,
784 __bond_resend_igmp_join_requests(vlan_dev);
787 if (--bond->igmp_retrans > 0)
788 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
790 read_unlock(&bond->lock);
793 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
795 struct bonding *bond = container_of(work, struct bonding,
797 bond_resend_igmp_join_requests(bond);
801 * flush all members of flush->mc_list from device dev->mc_list
803 static void bond_mc_list_flush(struct net_device *bond_dev,
804 struct net_device *slave_dev)
806 struct bonding *bond = netdev_priv(bond_dev);
807 struct netdev_hw_addr *ha;
809 netdev_for_each_mc_addr(ha, bond_dev)
810 dev_mc_del(slave_dev, ha->addr);
812 if (bond->params.mode == BOND_MODE_8023AD) {
813 /* del lacpdu mc addr from mc list */
814 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
816 dev_mc_del(slave_dev, lacpdu_multicast);
820 /*--------------------------- Active slave change ---------------------------*/
823 * Update the mc list and multicast-related flags for the new and
824 * old active slaves (if any) according to the multicast mode, and
825 * promiscuous flags unconditionally.
827 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
828 struct slave *old_active)
830 struct netdev_hw_addr *ha;
832 if (!USES_PRIMARY(bond->params.mode))
833 /* nothing to do - mc list is already up-to-date on
839 if (bond->dev->flags & IFF_PROMISC)
840 dev_set_promiscuity(old_active->dev, -1);
842 if (bond->dev->flags & IFF_ALLMULTI)
843 dev_set_allmulti(old_active->dev, -1);
845 netdev_for_each_mc_addr(ha, bond->dev)
846 dev_mc_del(old_active->dev, ha->addr);
850 /* FIXME: Signal errors upstream. */
851 if (bond->dev->flags & IFF_PROMISC)
852 dev_set_promiscuity(new_active->dev, 1);
854 if (bond->dev->flags & IFF_ALLMULTI)
855 dev_set_allmulti(new_active->dev, 1);
857 netdev_for_each_mc_addr(ha, bond->dev)
858 dev_mc_add(new_active->dev, ha->addr);
863 * bond_do_fail_over_mac
865 * Perform special MAC address swapping for fail_over_mac settings
867 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
869 static void bond_do_fail_over_mac(struct bonding *bond,
870 struct slave *new_active,
871 struct slave *old_active)
872 __releases(&bond->curr_slave_lock)
873 __releases(&bond->lock)
874 __acquires(&bond->lock)
875 __acquires(&bond->curr_slave_lock)
877 u8 tmp_mac[ETH_ALEN];
878 struct sockaddr saddr;
881 switch (bond->params.fail_over_mac) {
882 case BOND_FOM_ACTIVE:
884 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
885 new_active->dev->addr_len);
887 case BOND_FOM_FOLLOW:
889 * if new_active && old_active, swap them
890 * if just old_active, do nothing (going to no active slave)
891 * if just new_active, set new_active to bond's MAC
896 write_unlock_bh(&bond->curr_slave_lock);
897 read_unlock(&bond->lock);
900 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
901 memcpy(saddr.sa_data, old_active->dev->dev_addr,
903 saddr.sa_family = new_active->dev->type;
905 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
906 saddr.sa_family = bond->dev->type;
909 rv = dev_set_mac_address(new_active->dev, &saddr);
911 pr_err("%s: Error %d setting MAC of slave %s\n",
912 bond->dev->name, -rv, new_active->dev->name);
919 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
920 saddr.sa_family = old_active->dev->type;
922 rv = dev_set_mac_address(old_active->dev, &saddr);
924 pr_err("%s: Error %d setting MAC of slave %s\n",
925 bond->dev->name, -rv, new_active->dev->name);
927 read_lock(&bond->lock);
928 write_lock_bh(&bond->curr_slave_lock);
931 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
932 bond->dev->name, bond->params.fail_over_mac);
938 static bool bond_should_change_active(struct bonding *bond)
940 struct slave *prim = bond->primary_slave;
941 struct slave *curr = bond->curr_active_slave;
943 if (!prim || !curr || curr->link != BOND_LINK_UP)
945 if (bond->force_primary) {
946 bond->force_primary = false;
949 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
950 (prim->speed < curr->speed ||
951 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
953 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
959 * find_best_interface - select the best available slave to be the active one
960 * @bond: our bonding struct
962 * Warning: Caller must hold curr_slave_lock for writing.
964 static struct slave *bond_find_best_slave(struct bonding *bond)
966 struct slave *new_active, *old_active;
967 struct slave *bestslave = NULL;
968 int mintime = bond->params.updelay;
971 new_active = bond->curr_active_slave;
973 if (!new_active) { /* there were no active slaves left */
974 if (bond->slave_cnt > 0) /* found one slave */
975 new_active = bond->first_slave;
977 return NULL; /* still no slave, return NULL */
980 if ((bond->primary_slave) &&
981 bond->primary_slave->link == BOND_LINK_UP &&
982 bond_should_change_active(bond)) {
983 new_active = bond->primary_slave;
986 /* remember where to stop iterating over the slaves */
987 old_active = new_active;
989 bond_for_each_slave_from(bond, new_active, i, old_active) {
990 if (new_active->link == BOND_LINK_UP) {
992 } else if (new_active->link == BOND_LINK_BACK &&
993 IS_UP(new_active->dev)) {
994 /* link up, but waiting for stabilization */
995 if (new_active->delay < mintime) {
996 mintime = new_active->delay;
997 bestslave = new_active;
1005 static bool bond_should_notify_peers(struct bonding *bond)
1007 struct slave *slave = bond->curr_active_slave;
1009 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1010 bond->dev->name, slave ? slave->dev->name : "NULL");
1012 if (!slave || !bond->send_peer_notif ||
1013 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1016 bond->send_peer_notif--;
1021 * change_active_interface - change the active slave into the specified one
1022 * @bond: our bonding struct
1023 * @new: the new slave to make the active one
1025 * Set the new slave to the bond's settings and unset them on the old
1026 * curr_active_slave.
1027 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1029 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1030 * because it is apparently the best available slave we have, even though its
1031 * updelay hasn't timed out yet.
1033 * If new_active is not NULL, caller must hold bond->lock for read and
1034 * curr_slave_lock for write_bh.
1036 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1038 struct slave *old_active = bond->curr_active_slave;
1040 if (old_active == new_active)
1044 new_active->jiffies = jiffies;
1046 if (new_active->link == BOND_LINK_BACK) {
1047 if (USES_PRIMARY(bond->params.mode)) {
1048 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1049 bond->dev->name, new_active->dev->name,
1050 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1053 new_active->delay = 0;
1054 new_active->link = BOND_LINK_UP;
1056 if (bond->params.mode == BOND_MODE_8023AD)
1057 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1059 if (bond_is_lb(bond))
1060 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1062 if (USES_PRIMARY(bond->params.mode)) {
1063 pr_info("%s: making interface %s the new active one.\n",
1064 bond->dev->name, new_active->dev->name);
1069 if (USES_PRIMARY(bond->params.mode))
1070 bond_mc_swap(bond, new_active, old_active);
1072 if (bond_is_lb(bond)) {
1073 bond_alb_handle_active_change(bond, new_active);
1075 bond_set_slave_inactive_flags(old_active);
1077 bond_set_slave_active_flags(new_active);
1079 bond->curr_active_slave = new_active;
1082 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1084 bond_set_slave_inactive_flags(old_active);
1087 bool should_notify_peers = false;
1089 bond_set_slave_active_flags(new_active);
1091 if (bond->params.fail_over_mac)
1092 bond_do_fail_over_mac(bond, new_active,
1095 if (netif_running(bond->dev)) {
1096 bond->send_peer_notif =
1097 bond->params.num_peer_notif;
1098 should_notify_peers =
1099 bond_should_notify_peers(bond);
1102 write_unlock_bh(&bond->curr_slave_lock);
1103 read_unlock(&bond->lock);
1105 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1106 if (should_notify_peers)
1107 netdev_bonding_change(bond->dev,
1108 NETDEV_NOTIFY_PEERS);
1110 read_lock(&bond->lock);
1111 write_lock_bh(&bond->curr_slave_lock);
1115 /* resend IGMP joins since active slave has changed or
1116 * all were sent on curr_active_slave.
1117 * resend only if bond is brought up with the affected
1118 * bonding modes and the retransmission is enabled */
1119 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1120 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1121 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1122 bond->igmp_retrans = bond->params.resend_igmp;
1123 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1128 * bond_select_active_slave - select a new active slave, if needed
1129 * @bond: our bonding struct
1131 * This functions should be called when one of the following occurs:
1132 * - The old curr_active_slave has been released or lost its link.
1133 * - The primary_slave has got its link back.
1134 * - A slave has got its link back and there's no old curr_active_slave.
1136 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1138 void bond_select_active_slave(struct bonding *bond)
1140 struct slave *best_slave;
1143 best_slave = bond_find_best_slave(bond);
1144 if (best_slave != bond->curr_active_slave) {
1145 bond_change_active_slave(bond, best_slave);
1146 rv = bond_set_carrier(bond);
1150 if (netif_carrier_ok(bond->dev)) {
1151 pr_info("%s: first active interface up!\n",
1154 pr_info("%s: now running without any active interface !\n",
1160 /*--------------------------- slave list handling ---------------------------*/
1163 * This function attaches the slave to the end of list.
1165 * bond->lock held for writing by caller.
1167 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1169 if (bond->first_slave == NULL) { /* attaching the first slave */
1170 new_slave->next = new_slave;
1171 new_slave->prev = new_slave;
1172 bond->first_slave = new_slave;
1174 new_slave->next = bond->first_slave;
1175 new_slave->prev = bond->first_slave->prev;
1176 new_slave->next->prev = new_slave;
1177 new_slave->prev->next = new_slave;
1184 * This function detaches the slave from the list.
1185 * WARNING: no check is made to verify if the slave effectively
1186 * belongs to <bond>.
1187 * Nothing is freed on return, structures are just unchained.
1188 * If any slave pointer in bond was pointing to <slave>,
1189 * it should be changed by the calling function.
1191 * bond->lock held for writing by caller.
1193 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1196 slave->next->prev = slave->prev;
1199 slave->prev->next = slave->next;
1201 if (bond->first_slave == slave) { /* slave is the first slave */
1202 if (bond->slave_cnt > 1) { /* there are more slave */
1203 bond->first_slave = slave->next;
1205 bond->first_slave = NULL; /* slave was the last one */
1214 #ifdef CONFIG_NET_POLL_CONTROLLER
1215 static inline int slave_enable_netpoll(struct slave *slave)
1220 np = kzalloc(sizeof(*np), GFP_KERNEL);
1225 np->dev = slave->dev;
1226 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1227 err = __netpoll_setup(np);
1236 static inline void slave_disable_netpoll(struct slave *slave)
1238 struct netpoll *np = slave->np;
1244 synchronize_rcu_bh();
1245 __netpoll_cleanup(np);
1248 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1250 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1252 if (!slave_dev->netdev_ops->ndo_poll_controller)
1257 static void bond_poll_controller(struct net_device *bond_dev)
1261 static void __bond_netpoll_cleanup(struct bonding *bond)
1263 struct slave *slave;
1266 bond_for_each_slave(bond, slave, i)
1267 if (IS_UP(slave->dev))
1268 slave_disable_netpoll(slave);
1270 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1272 struct bonding *bond = netdev_priv(bond_dev);
1274 read_lock(&bond->lock);
1275 __bond_netpoll_cleanup(bond);
1276 read_unlock(&bond->lock);
1279 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1281 struct bonding *bond = netdev_priv(dev);
1282 struct slave *slave;
1285 read_lock(&bond->lock);
1286 bond_for_each_slave(bond, slave, i) {
1287 err = slave_enable_netpoll(slave);
1289 __bond_netpoll_cleanup(bond);
1293 read_unlock(&bond->lock);
1297 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1299 return bond->dev->npinfo;
1303 static inline int slave_enable_netpoll(struct slave *slave)
1307 static inline void slave_disable_netpoll(struct slave *slave)
1310 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1315 /*---------------------------------- IOCTL ----------------------------------*/
1317 static int bond_sethwaddr(struct net_device *bond_dev,
1318 struct net_device *slave_dev)
1320 pr_debug("bond_dev=%p\n", bond_dev);
1321 pr_debug("slave_dev=%p\n", slave_dev);
1322 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1323 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1327 static netdev_features_t bond_fix_features(struct net_device *dev,
1328 netdev_features_t features)
1330 struct slave *slave;
1331 struct bonding *bond = netdev_priv(dev);
1332 netdev_features_t mask;
1335 read_lock(&bond->lock);
1337 if (!bond->first_slave) {
1338 /* Disable adding VLANs to empty bond. But why? --mq */
1339 features |= NETIF_F_VLAN_CHALLENGED;
1344 features &= ~NETIF_F_ONE_FOR_ALL;
1345 features |= NETIF_F_ALL_FOR_ALL;
1347 bond_for_each_slave(bond, slave, i) {
1348 features = netdev_increment_features(features,
1349 slave->dev->features,
1354 read_unlock(&bond->lock);
1358 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1359 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1360 NETIF_F_HIGHDMA | NETIF_F_LRO)
1362 static void bond_compute_features(struct bonding *bond)
1364 struct slave *slave;
1365 struct net_device *bond_dev = bond->dev;
1366 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1367 unsigned short max_hard_header_len = ETH_HLEN;
1370 read_lock(&bond->lock);
1372 if (!bond->first_slave)
1375 bond_for_each_slave(bond, slave, i) {
1376 vlan_features = netdev_increment_features(vlan_features,
1377 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1379 if (slave->dev->hard_header_len > max_hard_header_len)
1380 max_hard_header_len = slave->dev->hard_header_len;
1384 bond_dev->vlan_features = vlan_features;
1385 bond_dev->hard_header_len = max_hard_header_len;
1387 read_unlock(&bond->lock);
1389 netdev_change_features(bond_dev);
1392 static void bond_setup_by_slave(struct net_device *bond_dev,
1393 struct net_device *slave_dev)
1395 struct bonding *bond = netdev_priv(bond_dev);
1397 bond_dev->header_ops = slave_dev->header_ops;
1399 bond_dev->type = slave_dev->type;
1400 bond_dev->hard_header_len = slave_dev->hard_header_len;
1401 bond_dev->addr_len = slave_dev->addr_len;
1403 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1404 slave_dev->addr_len);
1405 bond->setup_by_slave = 1;
1408 /* On bonding slaves other than the currently active slave, suppress
1409 * duplicates except for alb non-mcast/bcast.
1411 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1412 struct slave *slave,
1413 struct bonding *bond)
1415 if (bond_is_slave_inactive(slave)) {
1416 if (bond->params.mode == BOND_MODE_ALB &&
1417 skb->pkt_type != PACKET_BROADCAST &&
1418 skb->pkt_type != PACKET_MULTICAST)
1425 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1427 struct sk_buff *skb = *pskb;
1428 struct slave *slave;
1429 struct bonding *bond;
1430 void (*recv_probe)(struct sk_buff *, struct bonding *,
1433 skb = skb_share_check(skb, GFP_ATOMIC);
1435 return RX_HANDLER_CONSUMED;
1439 slave = bond_slave_get_rcu(skb->dev);
1442 if (bond->params.arp_interval)
1443 slave->dev->last_rx = jiffies;
1445 recv_probe = ACCESS_ONCE(bond->recv_probe);
1447 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1450 recv_probe(nskb, bond, slave);
1451 dev_kfree_skb(nskb);
1455 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1456 return RX_HANDLER_EXACT;
1459 skb->dev = bond->dev;
1461 if (bond->params.mode == BOND_MODE_ALB &&
1462 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1463 skb->pkt_type == PACKET_HOST) {
1465 if (unlikely(skb_cow_head(skb,
1466 skb->data - skb_mac_header(skb)))) {
1468 return RX_HANDLER_CONSUMED;
1470 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1473 return RX_HANDLER_ANOTHER;
1476 /* enslave device <slave> to bond device <master> */
1477 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1479 struct bonding *bond = netdev_priv(bond_dev);
1480 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1481 struct slave *new_slave = NULL;
1482 struct netdev_hw_addr *ha;
1483 struct sockaddr addr;
1487 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1488 slave_ops->ndo_do_ioctl == NULL) {
1489 pr_warning("%s: Warning: no link monitoring support for %s\n",
1490 bond_dev->name, slave_dev->name);
1493 /* already enslaved */
1494 if (slave_dev->flags & IFF_SLAVE) {
1495 pr_debug("Error, Device was already enslaved\n");
1499 /* vlan challenged mutual exclusion */
1500 /* no need to lock since we're protected by rtnl_lock */
1501 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1502 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1503 if (bond_vlan_used(bond)) {
1504 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1505 bond_dev->name, slave_dev->name, bond_dev->name);
1508 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1509 bond_dev->name, slave_dev->name,
1510 slave_dev->name, bond_dev->name);
1513 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1517 * Old ifenslave binaries are no longer supported. These can
1518 * be identified with moderate accuracy by the state of the slave:
1519 * the current ifenslave will set the interface down prior to
1520 * enslaving it; the old ifenslave will not.
1522 if ((slave_dev->flags & IFF_UP)) {
1523 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1526 goto err_undo_flags;
1529 /* set bonding device ether type by slave - bonding netdevices are
1530 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1531 * there is a need to override some of the type dependent attribs/funcs.
1533 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1534 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1536 if (bond->slave_cnt == 0) {
1537 if (bond_dev->type != slave_dev->type) {
1538 pr_debug("%s: change device type from %d to %d\n",
1540 bond_dev->type, slave_dev->type);
1542 res = netdev_bonding_change(bond_dev,
1543 NETDEV_PRE_TYPE_CHANGE);
1544 res = notifier_to_errno(res);
1546 pr_err("%s: refused to change device type\n",
1549 goto err_undo_flags;
1552 /* Flush unicast and multicast addresses */
1553 dev_uc_flush(bond_dev);
1554 dev_mc_flush(bond_dev);
1556 if (slave_dev->type != ARPHRD_ETHER)
1557 bond_setup_by_slave(bond_dev, slave_dev);
1559 ether_setup(bond_dev);
1560 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1563 netdev_bonding_change(bond_dev,
1564 NETDEV_POST_TYPE_CHANGE);
1566 } else if (bond_dev->type != slave_dev->type) {
1567 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1569 slave_dev->type, bond_dev->type);
1571 goto err_undo_flags;
1574 if (slave_ops->ndo_set_mac_address == NULL) {
1575 if (bond->slave_cnt == 0) {
1576 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1578 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1579 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1580 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",
1583 goto err_undo_flags;
1587 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1589 /* If this is the first slave, then we need to set the master's hardware
1590 * address to be the same as the slave's. */
1591 if (is_zero_ether_addr(bond->dev->dev_addr))
1592 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1593 slave_dev->addr_len);
1596 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1599 goto err_undo_flags;
1603 * Set the new_slave's queue_id to be zero. Queue ID mapping
1604 * is set via sysfs or module option if desired.
1606 new_slave->queue_id = 0;
1608 /* Save slave's original mtu and then set it to match the bond */
1609 new_slave->original_mtu = slave_dev->mtu;
1610 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1612 pr_debug("Error %d calling dev_set_mtu\n", res);
1617 * Save slave's original ("permanent") mac address for modes
1618 * that need it, and for restoring it upon release, and then
1619 * set it to the master's address
1621 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1623 if (!bond->params.fail_over_mac) {
1625 * Set slave to master's mac address. The application already
1626 * set the master's mac address to that of the first slave
1628 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1629 addr.sa_family = slave_dev->type;
1630 res = dev_set_mac_address(slave_dev, &addr);
1632 pr_debug("Error %d calling set_mac_address\n", res);
1633 goto err_restore_mtu;
1637 res = netdev_set_bond_master(slave_dev, bond_dev);
1639 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1640 goto err_restore_mac;
1643 /* open the slave since the application closed it */
1644 res = dev_open(slave_dev);
1646 pr_debug("Opening slave %s failed\n", slave_dev->name);
1647 goto err_unset_master;
1650 new_slave->bond = bond;
1651 new_slave->dev = slave_dev;
1652 slave_dev->priv_flags |= IFF_BONDING;
1654 if (bond_is_lb(bond)) {
1655 /* bond_alb_init_slave() must be called before all other stages since
1656 * it might fail and we do not want to have to undo everything
1658 res = bond_alb_init_slave(bond, new_slave);
1663 /* If the mode USES_PRIMARY, then the new slave gets the
1664 * master's promisc (and mc) settings only if it becomes the
1665 * curr_active_slave, and that is taken care of later when calling
1666 * bond_change_active()
1668 if (!USES_PRIMARY(bond->params.mode)) {
1669 /* set promiscuity level to new slave */
1670 if (bond_dev->flags & IFF_PROMISC) {
1671 res = dev_set_promiscuity(slave_dev, 1);
1676 /* set allmulti level to new slave */
1677 if (bond_dev->flags & IFF_ALLMULTI) {
1678 res = dev_set_allmulti(slave_dev, 1);
1683 netif_addr_lock_bh(bond_dev);
1684 /* upload master's mc_list to new slave */
1685 netdev_for_each_mc_addr(ha, bond_dev)
1686 dev_mc_add(slave_dev, ha->addr);
1687 netif_addr_unlock_bh(bond_dev);
1690 if (bond->params.mode == BOND_MODE_8023AD) {
1691 /* add lacpdu mc addr to mc list */
1692 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1694 dev_mc_add(slave_dev, lacpdu_multicast);
1697 bond_add_vlans_on_slave(bond, slave_dev);
1699 write_lock_bh(&bond->lock);
1701 bond_attach_slave(bond, new_slave);
1703 new_slave->delay = 0;
1704 new_slave->link_failure_count = 0;
1706 write_unlock_bh(&bond->lock);
1708 bond_compute_features(bond);
1710 read_lock(&bond->lock);
1712 new_slave->last_arp_rx = jiffies;
1714 if (bond->params.miimon && !bond->params.use_carrier) {
1715 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1717 if ((link_reporting == -1) && !bond->params.arp_interval) {
1719 * miimon is set but a bonded network driver
1720 * does not support ETHTOOL/MII and
1721 * arp_interval is not set. Note: if
1722 * use_carrier is enabled, we will never go
1723 * here (because netif_carrier is always
1724 * supported); thus, we don't need to change
1725 * the messages for netif_carrier.
1727 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",
1728 bond_dev->name, slave_dev->name);
1729 } else if (link_reporting == -1) {
1730 /* unable get link status using mii/ethtool */
1731 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",
1732 bond_dev->name, slave_dev->name);
1736 /* check for initial state */
1737 if (!bond->params.miimon ||
1738 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1739 if (bond->params.updelay) {
1740 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1741 new_slave->link = BOND_LINK_BACK;
1742 new_slave->delay = bond->params.updelay;
1744 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1745 new_slave->link = BOND_LINK_UP;
1747 new_slave->jiffies = jiffies;
1749 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1750 new_slave->link = BOND_LINK_DOWN;
1753 bond_update_speed_duplex(new_slave);
1755 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1756 /* if there is a primary slave, remember it */
1757 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1758 bond->primary_slave = new_slave;
1759 bond->force_primary = true;
1763 write_lock_bh(&bond->curr_slave_lock);
1765 switch (bond->params.mode) {
1766 case BOND_MODE_ACTIVEBACKUP:
1767 bond_set_slave_inactive_flags(new_slave);
1768 bond_select_active_slave(bond);
1770 case BOND_MODE_8023AD:
1771 /* in 802.3ad mode, the internal mechanism
1772 * will activate the slaves in the selected
1775 bond_set_slave_inactive_flags(new_slave);
1776 /* if this is the first slave */
1777 if (bond->slave_cnt == 1) {
1778 SLAVE_AD_INFO(new_slave).id = 1;
1779 /* Initialize AD with the number of times that the AD timer is called in 1 second
1780 * can be called only after the mac address of the bond is set
1782 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1784 SLAVE_AD_INFO(new_slave).id =
1785 SLAVE_AD_INFO(new_slave->prev).id + 1;
1788 bond_3ad_bind_slave(new_slave);
1792 bond_set_active_slave(new_slave);
1793 bond_set_slave_inactive_flags(new_slave);
1794 bond_select_active_slave(bond);
1797 pr_debug("This slave is always active in trunk mode\n");
1799 /* always active in trunk mode */
1800 bond_set_active_slave(new_slave);
1802 /* In trunking mode there is little meaning to curr_active_slave
1803 * anyway (it holds no special properties of the bond device),
1804 * so we can change it without calling change_active_interface()
1806 if (!bond->curr_active_slave)
1807 bond->curr_active_slave = new_slave;
1810 } /* switch(bond_mode) */
1812 write_unlock_bh(&bond->curr_slave_lock);
1814 bond_set_carrier(bond);
1816 #ifdef CONFIG_NET_POLL_CONTROLLER
1817 slave_dev->npinfo = bond_netpoll_info(bond);
1818 if (slave_dev->npinfo) {
1819 if (slave_enable_netpoll(new_slave)) {
1820 read_unlock(&bond->lock);
1821 pr_info("Error, %s: master_dev is using netpoll, "
1822 "but new slave device does not support netpoll.\n",
1830 read_unlock(&bond->lock);
1832 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1836 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1839 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1840 goto err_dest_symlinks;
1843 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1844 bond_dev->name, slave_dev->name,
1845 bond_is_active_slave(new_slave) ? "n active" : " backup",
1846 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1848 /* enslave is successful */
1851 /* Undo stages on error */
1853 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1856 dev_close(slave_dev);
1859 netdev_set_bond_master(slave_dev, NULL);
1862 if (!bond->params.fail_over_mac) {
1863 /* XXX TODO - fom follow mode needs to change master's
1864 * MAC if this slave's MAC is in use by the bond, or at
1865 * least print a warning.
1867 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1868 addr.sa_family = slave_dev->type;
1869 dev_set_mac_address(slave_dev, &addr);
1873 dev_set_mtu(slave_dev, new_slave->original_mtu);
1879 bond_compute_features(bond);
1885 * Try to release the slave device <slave> from the bond device <master>
1886 * It is legal to access curr_active_slave without a lock because all the function
1889 * The rules for slave state should be:
1890 * for Active/Backup:
1891 * Active stays on all backups go down
1892 * for Bonded connections:
1893 * The first up interface should be left on and all others downed.
1895 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1897 struct bonding *bond = netdev_priv(bond_dev);
1898 struct slave *slave, *oldcurrent;
1899 struct sockaddr addr;
1900 netdev_features_t old_features = bond_dev->features;
1902 /* slave is not a slave or master is not master of this slave */
1903 if (!(slave_dev->flags & IFF_SLAVE) ||
1904 (slave_dev->master != bond_dev)) {
1905 pr_err("%s: Error: cannot release %s.\n",
1906 bond_dev->name, slave_dev->name);
1911 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1912 write_lock_bh(&bond->lock);
1914 slave = bond_get_slave_by_dev(bond, slave_dev);
1916 /* not a slave of this bond */
1917 pr_info("%s: %s not enslaved\n",
1918 bond_dev->name, slave_dev->name);
1919 write_unlock_bh(&bond->lock);
1920 unblock_netpoll_tx();
1924 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1925 * for this slave anymore.
1927 netdev_rx_handler_unregister(slave_dev);
1928 write_unlock_bh(&bond->lock);
1930 write_lock_bh(&bond->lock);
1932 if (!bond->params.fail_over_mac) {
1933 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1934 bond->slave_cnt > 1)
1935 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",
1936 bond_dev->name, slave_dev->name,
1938 bond_dev->name, slave_dev->name);
1941 /* Inform AD package of unbinding of slave. */
1942 if (bond->params.mode == BOND_MODE_8023AD) {
1943 /* must be called before the slave is
1944 * detached from the list
1946 bond_3ad_unbind_slave(slave);
1949 pr_info("%s: releasing %s interface %s\n",
1951 bond_is_active_slave(slave) ? "active" : "backup",
1954 oldcurrent = bond->curr_active_slave;
1956 bond->current_arp_slave = NULL;
1958 /* release the slave from its bond */
1959 bond_detach_slave(bond, slave);
1961 if (bond->primary_slave == slave)
1962 bond->primary_slave = NULL;
1964 if (oldcurrent == slave)
1965 bond_change_active_slave(bond, NULL);
1967 if (bond_is_lb(bond)) {
1968 /* Must be called only after the slave has been
1969 * detached from the list and the curr_active_slave
1970 * has been cleared (if our_slave == old_current),
1971 * but before a new active slave is selected.
1973 write_unlock_bh(&bond->lock);
1974 bond_alb_deinit_slave(bond, slave);
1975 write_lock_bh(&bond->lock);
1978 if (oldcurrent == slave) {
1980 * Note that we hold RTNL over this sequence, so there
1981 * is no concern that another slave add/remove event
1984 write_unlock_bh(&bond->lock);
1985 read_lock(&bond->lock);
1986 write_lock_bh(&bond->curr_slave_lock);
1988 bond_select_active_slave(bond);
1990 write_unlock_bh(&bond->curr_slave_lock);
1991 read_unlock(&bond->lock);
1992 write_lock_bh(&bond->lock);
1995 if (bond->slave_cnt == 0) {
1996 bond_set_carrier(bond);
1998 /* if the last slave was removed, zero the mac address
1999 * of the master so it will be set by the application
2000 * to the mac address of the first slave
2002 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2004 if (bond_vlan_used(bond)) {
2005 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2006 bond_dev->name, bond_dev->name);
2007 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2012 write_unlock_bh(&bond->lock);
2013 unblock_netpoll_tx();
2015 bond_compute_features(bond);
2016 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2017 (old_features & NETIF_F_VLAN_CHALLENGED))
2018 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2019 bond_dev->name, slave_dev->name, bond_dev->name);
2021 /* must do this from outside any spinlocks */
2022 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2024 bond_del_vlans_from_slave(bond, slave_dev);
2026 /* If the mode USES_PRIMARY, then we should only remove its
2027 * promisc and mc settings if it was the curr_active_slave, but that was
2028 * already taken care of above when we detached the slave
2030 if (!USES_PRIMARY(bond->params.mode)) {
2031 /* unset promiscuity level from slave */
2032 if (bond_dev->flags & IFF_PROMISC)
2033 dev_set_promiscuity(slave_dev, -1);
2035 /* unset allmulti level from slave */
2036 if (bond_dev->flags & IFF_ALLMULTI)
2037 dev_set_allmulti(slave_dev, -1);
2039 /* flush master's mc_list from slave */
2040 netif_addr_lock_bh(bond_dev);
2041 bond_mc_list_flush(bond_dev, slave_dev);
2042 netif_addr_unlock_bh(bond_dev);
2045 netdev_set_bond_master(slave_dev, NULL);
2047 slave_disable_netpoll(slave);
2049 /* close slave before restoring its mac address */
2050 dev_close(slave_dev);
2052 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2053 /* restore original ("permanent") mac address */
2054 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2055 addr.sa_family = slave_dev->type;
2056 dev_set_mac_address(slave_dev, &addr);
2059 dev_set_mtu(slave_dev, slave->original_mtu);
2061 slave_dev->priv_flags &= ~IFF_BONDING;
2065 return 0; /* deletion OK */
2069 * First release a slave and then destroy the bond if no more slaves are left.
2070 * Must be under rtnl_lock when this function is called.
2072 static int bond_release_and_destroy(struct net_device *bond_dev,
2073 struct net_device *slave_dev)
2075 struct bonding *bond = netdev_priv(bond_dev);
2078 ret = bond_release(bond_dev, slave_dev);
2079 if ((ret == 0) && (bond->slave_cnt == 0)) {
2080 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2081 pr_info("%s: destroying bond %s.\n",
2082 bond_dev->name, bond_dev->name);
2083 unregister_netdevice(bond_dev);
2089 * This function releases all slaves.
2091 static int bond_release_all(struct net_device *bond_dev)
2093 struct bonding *bond = netdev_priv(bond_dev);
2094 struct slave *slave;
2095 struct net_device *slave_dev;
2096 struct sockaddr addr;
2098 write_lock_bh(&bond->lock);
2100 netif_carrier_off(bond_dev);
2102 if (bond->slave_cnt == 0)
2105 bond->current_arp_slave = NULL;
2106 bond->primary_slave = NULL;
2107 bond_change_active_slave(bond, NULL);
2109 while ((slave = bond->first_slave) != NULL) {
2110 /* Inform AD package of unbinding of slave
2111 * before slave is detached from the list.
2113 if (bond->params.mode == BOND_MODE_8023AD)
2114 bond_3ad_unbind_slave(slave);
2116 slave_dev = slave->dev;
2117 bond_detach_slave(bond, slave);
2119 /* now that the slave is detached, unlock and perform
2120 * all the undo steps that should not be called from
2123 write_unlock_bh(&bond->lock);
2125 /* unregister rx_handler early so bond_handle_frame wouldn't
2126 * be called for this slave anymore.
2128 netdev_rx_handler_unregister(slave_dev);
2131 if (bond_is_lb(bond)) {
2132 /* must be called only after the slave
2133 * has been detached from the list
2135 bond_alb_deinit_slave(bond, slave);
2138 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2139 bond_del_vlans_from_slave(bond, slave_dev);
2141 /* If the mode USES_PRIMARY, then we should only remove its
2142 * promisc and mc settings if it was the curr_active_slave, but that was
2143 * already taken care of above when we detached the slave
2145 if (!USES_PRIMARY(bond->params.mode)) {
2146 /* unset promiscuity level from slave */
2147 if (bond_dev->flags & IFF_PROMISC)
2148 dev_set_promiscuity(slave_dev, -1);
2150 /* unset allmulti level from slave */
2151 if (bond_dev->flags & IFF_ALLMULTI)
2152 dev_set_allmulti(slave_dev, -1);
2154 /* flush master's mc_list from slave */
2155 netif_addr_lock_bh(bond_dev);
2156 bond_mc_list_flush(bond_dev, slave_dev);
2157 netif_addr_unlock_bh(bond_dev);
2160 netdev_set_bond_master(slave_dev, NULL);
2162 slave_disable_netpoll(slave);
2164 /* close slave before restoring its mac address */
2165 dev_close(slave_dev);
2167 if (!bond->params.fail_over_mac) {
2168 /* restore original ("permanent") mac address*/
2169 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2170 addr.sa_family = slave_dev->type;
2171 dev_set_mac_address(slave_dev, &addr);
2176 /* re-acquire the lock before getting the next slave */
2177 write_lock_bh(&bond->lock);
2180 /* zero the mac address of the master so it will be
2181 * set by the application to the mac address of the
2184 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2186 if (bond_vlan_used(bond)) {
2187 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2188 bond_dev->name, bond_dev->name);
2189 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2193 pr_info("%s: released all slaves\n", bond_dev->name);
2196 write_unlock_bh(&bond->lock);
2198 bond_compute_features(bond);
2204 * This function changes the active slave to slave <slave_dev>.
2205 * It returns -EINVAL in the following cases.
2206 * - <slave_dev> is not found in the list.
2207 * - There is not active slave now.
2208 * - <slave_dev> is already active.
2209 * - The link state of <slave_dev> is not BOND_LINK_UP.
2210 * - <slave_dev> is not running.
2211 * In these cases, this function does nothing.
2212 * In the other cases, current_slave pointer is changed and 0 is returned.
2214 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2216 struct bonding *bond = netdev_priv(bond_dev);
2217 struct slave *old_active = NULL;
2218 struct slave *new_active = NULL;
2221 if (!USES_PRIMARY(bond->params.mode))
2224 /* Verify that master_dev is indeed the master of slave_dev */
2225 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2228 read_lock(&bond->lock);
2230 read_lock(&bond->curr_slave_lock);
2231 old_active = bond->curr_active_slave;
2232 read_unlock(&bond->curr_slave_lock);
2234 new_active = bond_get_slave_by_dev(bond, slave_dev);
2237 * Changing to the current active: do nothing; return success.
2239 if (new_active && (new_active == old_active)) {
2240 read_unlock(&bond->lock);
2246 (new_active->link == BOND_LINK_UP) &&
2247 IS_UP(new_active->dev)) {
2249 write_lock_bh(&bond->curr_slave_lock);
2250 bond_change_active_slave(bond, new_active);
2251 write_unlock_bh(&bond->curr_slave_lock);
2252 unblock_netpoll_tx();
2256 read_unlock(&bond->lock);
2261 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2263 struct bonding *bond = netdev_priv(bond_dev);
2265 info->bond_mode = bond->params.mode;
2266 info->miimon = bond->params.miimon;
2268 read_lock(&bond->lock);
2269 info->num_slaves = bond->slave_cnt;
2270 read_unlock(&bond->lock);
2275 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2277 struct bonding *bond = netdev_priv(bond_dev);
2278 struct slave *slave;
2279 int i, res = -ENODEV;
2281 read_lock(&bond->lock);
2283 bond_for_each_slave(bond, slave, i) {
2284 if (i == (int)info->slave_id) {
2286 strcpy(info->slave_name, slave->dev->name);
2287 info->link = slave->link;
2288 info->state = bond_slave_state(slave);
2289 info->link_failure_count = slave->link_failure_count;
2294 read_unlock(&bond->lock);
2299 /*-------------------------------- Monitoring -------------------------------*/
2302 static int bond_miimon_inspect(struct bonding *bond)
2304 struct slave *slave;
2305 int i, link_state, commit = 0;
2306 bool ignore_updelay;
2308 ignore_updelay = !bond->curr_active_slave ? true : false;
2310 bond_for_each_slave(bond, slave, i) {
2311 slave->new_link = BOND_LINK_NOCHANGE;
2313 link_state = bond_check_dev_link(bond, slave->dev, 0);
2315 switch (slave->link) {
2320 slave->link = BOND_LINK_FAIL;
2321 slave->delay = bond->params.downdelay;
2323 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2325 (bond->params.mode ==
2326 BOND_MODE_ACTIVEBACKUP) ?
2327 (bond_is_active_slave(slave) ?
2328 "active " : "backup ") : "",
2330 bond->params.downdelay * bond->params.miimon);
2333 case BOND_LINK_FAIL:
2336 * recovered before downdelay expired
2338 slave->link = BOND_LINK_UP;
2339 slave->jiffies = jiffies;
2340 pr_info("%s: link status up again after %d ms for interface %s.\n",
2342 (bond->params.downdelay - slave->delay) *
2343 bond->params.miimon,
2348 if (slave->delay <= 0) {
2349 slave->new_link = BOND_LINK_DOWN;
2357 case BOND_LINK_DOWN:
2361 slave->link = BOND_LINK_BACK;
2362 slave->delay = bond->params.updelay;
2365 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2366 bond->dev->name, slave->dev->name,
2367 ignore_updelay ? 0 :
2368 bond->params.updelay *
2369 bond->params.miimon);
2372 case BOND_LINK_BACK:
2374 slave->link = BOND_LINK_DOWN;
2375 pr_info("%s: link status down again after %d ms for interface %s.\n",
2377 (bond->params.updelay - slave->delay) *
2378 bond->params.miimon,
2387 if (slave->delay <= 0) {
2388 slave->new_link = BOND_LINK_UP;
2390 ignore_updelay = false;
2402 static void bond_miimon_commit(struct bonding *bond)
2404 struct slave *slave;
2407 bond_for_each_slave(bond, slave, i) {
2408 switch (slave->new_link) {
2409 case BOND_LINK_NOCHANGE:
2413 slave->link = BOND_LINK_UP;
2414 slave->jiffies = jiffies;
2416 if (bond->params.mode == BOND_MODE_8023AD) {
2417 /* prevent it from being the active one */
2418 bond_set_backup_slave(slave);
2419 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2420 /* make it immediately active */
2421 bond_set_active_slave(slave);
2422 } else if (slave != bond->primary_slave) {
2423 /* prevent it from being the active one */
2424 bond_set_backup_slave(slave);
2427 bond_update_speed_duplex(slave);
2429 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2430 bond->dev->name, slave->dev->name,
2431 slave->speed, slave->duplex ? "full" : "half");
2433 /* notify ad that the link status has changed */
2434 if (bond->params.mode == BOND_MODE_8023AD)
2435 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2437 if (bond_is_lb(bond))
2438 bond_alb_handle_link_change(bond, slave,
2441 if (!bond->curr_active_slave ||
2442 (slave == bond->primary_slave))
2447 case BOND_LINK_DOWN:
2448 if (slave->link_failure_count < UINT_MAX)
2449 slave->link_failure_count++;
2451 slave->link = BOND_LINK_DOWN;
2453 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2454 bond->params.mode == BOND_MODE_8023AD)
2455 bond_set_slave_inactive_flags(slave);
2457 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2458 bond->dev->name, slave->dev->name);
2460 if (bond->params.mode == BOND_MODE_8023AD)
2461 bond_3ad_handle_link_change(slave,
2464 if (bond_is_lb(bond))
2465 bond_alb_handle_link_change(bond, slave,
2468 if (slave == bond->curr_active_slave)
2474 pr_err("%s: invalid new link %d on slave %s\n",
2475 bond->dev->name, slave->new_link,
2477 slave->new_link = BOND_LINK_NOCHANGE;
2485 write_lock_bh(&bond->curr_slave_lock);
2486 bond_select_active_slave(bond);
2487 write_unlock_bh(&bond->curr_slave_lock);
2488 unblock_netpoll_tx();
2491 bond_set_carrier(bond);
2497 * Really a wrapper that splits the mii monitor into two phases: an
2498 * inspection, then (if inspection indicates something needs to be done)
2499 * an acquisition of appropriate locks followed by a commit phase to
2500 * implement whatever link state changes are indicated.
2502 void bond_mii_monitor(struct work_struct *work)
2504 struct bonding *bond = container_of(work, struct bonding,
2506 bool should_notify_peers = false;
2507 unsigned long delay;
2509 read_lock(&bond->lock);
2511 delay = msecs_to_jiffies(bond->params.miimon);
2513 if (bond->slave_cnt == 0)
2516 should_notify_peers = bond_should_notify_peers(bond);
2518 if (bond_miimon_inspect(bond)) {
2519 read_unlock(&bond->lock);
2521 /* Race avoidance with bond_close cancel of workqueue */
2522 if (!rtnl_trylock()) {
2523 read_lock(&bond->lock);
2525 should_notify_peers = false;
2529 read_lock(&bond->lock);
2531 bond_miimon_commit(bond);
2533 read_unlock(&bond->lock);
2534 rtnl_unlock(); /* might sleep, hold no other locks */
2535 read_lock(&bond->lock);
2539 if (bond->params.miimon)
2540 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2542 read_unlock(&bond->lock);
2544 if (should_notify_peers) {
2545 if (!rtnl_trylock()) {
2546 read_lock(&bond->lock);
2547 bond->send_peer_notif++;
2548 read_unlock(&bond->lock);
2551 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2556 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2558 struct vlan_entry *vlan;
2560 if (ip == bond->master_ip)
2563 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2564 if (ip == vlan->vlan_ip)
2572 * We go to the (large) trouble of VLAN tagging ARP frames because
2573 * switches in VLAN mode (especially if ports are configured as
2574 * "native" to a VLAN) might not pass non-tagged frames.
2576 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2578 struct sk_buff *skb;
2580 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2581 slave_dev->name, dest_ip, src_ip, vlan_id);
2583 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2584 NULL, slave_dev->dev_addr, NULL);
2587 pr_err("ARP packet allocation failed\n");
2591 skb = vlan_put_tag(skb, vlan_id);
2593 pr_err("failed to insert VLAN tag\n");
2601 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2604 __be32 *targets = bond->params.arp_targets;
2605 struct vlan_entry *vlan;
2606 struct net_device *vlan_dev;
2609 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2612 pr_debug("basa: target %x\n", targets[i]);
2613 if (!bond_vlan_used(bond)) {
2614 pr_debug("basa: empty vlan: arp_send\n");
2615 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2616 bond->master_ip, 0);
2621 * If VLANs are configured, we do a route lookup to
2622 * determine which VLAN interface would be used, so we
2623 * can tag the ARP with the proper VLAN tag.
2625 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2628 if (net_ratelimit()) {
2629 pr_warning("%s: no route to arp_ip_target %pI4\n",
2630 bond->dev->name, &targets[i]);
2636 * This target is not on a VLAN
2638 if (rt->dst.dev == bond->dev) {
2640 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2641 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2642 bond->master_ip, 0);
2647 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2649 vlan_dev = __vlan_find_dev_deep(bond->dev,
2652 if (vlan_dev == rt->dst.dev) {
2653 vlan_id = vlan->vlan_id;
2654 pr_debug("basa: vlan match on %s %d\n",
2655 vlan_dev->name, vlan_id);
2662 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2663 vlan->vlan_ip, vlan_id);
2667 if (net_ratelimit()) {
2668 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2669 bond->dev->name, &targets[i],
2670 rt->dst.dev ? rt->dst.dev->name : "NULL");
2676 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2679 __be32 *targets = bond->params.arp_targets;
2681 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2682 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2683 &sip, &tip, i, &targets[i],
2684 bond_has_this_ip(bond, tip));
2685 if (sip == targets[i]) {
2686 if (bond_has_this_ip(bond, tip))
2687 slave->last_arp_rx = jiffies;
2693 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2694 struct slave *slave)
2697 unsigned char *arp_ptr;
2700 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2703 read_lock(&bond->lock);
2705 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2706 bond->dev->name, skb->dev->name);
2708 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2712 if (arp->ar_hln != bond->dev->addr_len ||
2713 skb->pkt_type == PACKET_OTHERHOST ||
2714 skb->pkt_type == PACKET_LOOPBACK ||
2715 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2716 arp->ar_pro != htons(ETH_P_IP) ||
2720 arp_ptr = (unsigned char *)(arp + 1);
2721 arp_ptr += bond->dev->addr_len;
2722 memcpy(&sip, arp_ptr, 4);
2723 arp_ptr += 4 + bond->dev->addr_len;
2724 memcpy(&tip, arp_ptr, 4);
2726 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2727 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2728 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2732 * Backup slaves won't see the ARP reply, but do come through
2733 * here for each ARP probe (so we swap the sip/tip to validate
2734 * the probe). In a "redundant switch, common router" type of
2735 * configuration, the ARP probe will (hopefully) travel from
2736 * the active, through one switch, the router, then the other
2737 * switch before reaching the backup.
2739 if (bond_is_active_slave(slave))
2740 bond_validate_arp(bond, slave, sip, tip);
2742 bond_validate_arp(bond, slave, tip, sip);
2745 read_unlock(&bond->lock);
2749 * this function is called regularly to monitor each slave's link
2750 * ensuring that traffic is being sent and received when arp monitoring
2751 * is used in load-balancing mode. if the adapter has been dormant, then an
2752 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2753 * arp monitoring in active backup mode.
2755 void bond_loadbalance_arp_mon(struct work_struct *work)
2757 struct bonding *bond = container_of(work, struct bonding,
2759 struct slave *slave, *oldcurrent;
2760 int do_failover = 0;
2764 read_lock(&bond->lock);
2766 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2768 if (bond->slave_cnt == 0)
2771 read_lock(&bond->curr_slave_lock);
2772 oldcurrent = bond->curr_active_slave;
2773 read_unlock(&bond->curr_slave_lock);
2775 /* see if any of the previous devices are up now (i.e. they have
2776 * xmt and rcv traffic). the curr_active_slave does not come into
2777 * the picture unless it is null. also, slave->jiffies is not needed
2778 * here because we send an arp on each slave and give a slave as
2779 * long as it needs to get the tx/rx within the delta.
2780 * TODO: what about up/down delay in arp mode? it wasn't here before
2783 bond_for_each_slave(bond, slave, i) {
2784 unsigned long trans_start = dev_trans_start(slave->dev);
2786 if (slave->link != BOND_LINK_UP) {
2787 if (time_in_range(jiffies,
2788 trans_start - delta_in_ticks,
2789 trans_start + delta_in_ticks) &&
2790 time_in_range(jiffies,
2791 slave->dev->last_rx - delta_in_ticks,
2792 slave->dev->last_rx + delta_in_ticks)) {
2794 slave->link = BOND_LINK_UP;
2795 bond_set_active_slave(slave);
2797 /* primary_slave has no meaning in round-robin
2798 * mode. the window of a slave being up and
2799 * curr_active_slave being null after enslaving
2803 pr_info("%s: link status definitely up for interface %s, ",
2808 pr_info("%s: interface %s is now up\n",
2814 /* slave->link == BOND_LINK_UP */
2816 /* not all switches will respond to an arp request
2817 * when the source ip is 0, so don't take the link down
2818 * if we don't know our ip yet
2820 if (!time_in_range(jiffies,
2821 trans_start - delta_in_ticks,
2822 trans_start + 2 * delta_in_ticks) ||
2823 !time_in_range(jiffies,
2824 slave->dev->last_rx - delta_in_ticks,
2825 slave->dev->last_rx + 2 * delta_in_ticks)) {
2827 slave->link = BOND_LINK_DOWN;
2828 bond_set_backup_slave(slave);
2830 if (slave->link_failure_count < UINT_MAX)
2831 slave->link_failure_count++;
2833 pr_info("%s: interface %s is now down.\n",
2837 if (slave == oldcurrent)
2842 /* note: if switch is in round-robin mode, all links
2843 * must tx arp to ensure all links rx an arp - otherwise
2844 * links may oscillate or not come up at all; if switch is
2845 * in something like xor mode, there is nothing we can
2846 * do - all replies will be rx'ed on same link causing slaves
2847 * to be unstable during low/no traffic periods
2849 if (IS_UP(slave->dev))
2850 bond_arp_send_all(bond, slave);
2855 write_lock_bh(&bond->curr_slave_lock);
2857 bond_select_active_slave(bond);
2859 write_unlock_bh(&bond->curr_slave_lock);
2860 unblock_netpoll_tx();
2864 if (bond->params.arp_interval)
2865 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2867 read_unlock(&bond->lock);
2871 * Called to inspect slaves for active-backup mode ARP monitor link state
2872 * changes. Sets new_link in slaves to specify what action should take
2873 * place for the slave. Returns 0 if no changes are found, >0 if changes
2874 * to link states must be committed.
2876 * Called with bond->lock held for read.
2878 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2880 struct slave *slave;
2882 unsigned long trans_start;
2884 bond_for_each_slave(bond, slave, i) {
2885 slave->new_link = BOND_LINK_NOCHANGE;
2887 if (slave->link != BOND_LINK_UP) {
2888 if (time_in_range(jiffies,
2889 slave_last_rx(bond, slave) - delta_in_ticks,
2890 slave_last_rx(bond, slave) + delta_in_ticks)) {
2892 slave->new_link = BOND_LINK_UP;
2900 * Give slaves 2*delta after being enslaved or made
2901 * active. This avoids bouncing, as the last receive
2902 * times need a full ARP monitor cycle to be updated.
2904 if (time_in_range(jiffies,
2905 slave->jiffies - delta_in_ticks,
2906 slave->jiffies + 2 * delta_in_ticks))
2910 * Backup slave is down if:
2911 * - No current_arp_slave AND
2912 * - more than 3*delta since last receive AND
2913 * - the bond has an IP address
2915 * Note: a non-null current_arp_slave indicates
2916 * the curr_active_slave went down and we are
2917 * searching for a new one; under this condition
2918 * we only take the curr_active_slave down - this
2919 * gives each slave a chance to tx/rx traffic
2920 * before being taken out
2922 if (!bond_is_active_slave(slave) &&
2923 !bond->current_arp_slave &&
2924 !time_in_range(jiffies,
2925 slave_last_rx(bond, slave) - delta_in_ticks,
2926 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2928 slave->new_link = BOND_LINK_DOWN;
2933 * Active slave is down if:
2934 * - more than 2*delta since transmitting OR
2935 * - (more than 2*delta since receive AND
2936 * the bond has an IP address)
2938 trans_start = dev_trans_start(slave->dev);
2939 if (bond_is_active_slave(slave) &&
2940 (!time_in_range(jiffies,
2941 trans_start - delta_in_ticks,
2942 trans_start + 2 * delta_in_ticks) ||
2943 !time_in_range(jiffies,
2944 slave_last_rx(bond, slave) - delta_in_ticks,
2945 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2947 slave->new_link = BOND_LINK_DOWN;
2956 * Called to commit link state changes noted by inspection step of
2957 * active-backup mode ARP monitor.
2959 * Called with RTNL and bond->lock for read.
2961 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2963 struct slave *slave;
2965 unsigned long trans_start;
2967 bond_for_each_slave(bond, slave, i) {
2968 switch (slave->new_link) {
2969 case BOND_LINK_NOCHANGE:
2973 trans_start = dev_trans_start(slave->dev);
2974 if ((!bond->curr_active_slave &&
2975 time_in_range(jiffies,
2976 trans_start - delta_in_ticks,
2977 trans_start + delta_in_ticks)) ||
2978 bond->curr_active_slave != slave) {
2979 slave->link = BOND_LINK_UP;
2980 bond->current_arp_slave = NULL;
2982 pr_info("%s: link status definitely up for interface %s.\n",
2983 bond->dev->name, slave->dev->name);
2985 if (!bond->curr_active_slave ||
2986 (slave == bond->primary_slave))
2993 case BOND_LINK_DOWN:
2994 if (slave->link_failure_count < UINT_MAX)
2995 slave->link_failure_count++;
2997 slave->link = BOND_LINK_DOWN;
2998 bond_set_slave_inactive_flags(slave);
3000 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3001 bond->dev->name, slave->dev->name);
3003 if (slave == bond->curr_active_slave) {
3004 bond->current_arp_slave = NULL;
3011 pr_err("%s: impossible: new_link %d on slave %s\n",
3012 bond->dev->name, slave->new_link,
3020 write_lock_bh(&bond->curr_slave_lock);
3021 bond_select_active_slave(bond);
3022 write_unlock_bh(&bond->curr_slave_lock);
3023 unblock_netpoll_tx();
3026 bond_set_carrier(bond);
3030 * Send ARP probes for active-backup mode ARP monitor.
3032 * Called with bond->lock held for read.
3034 static void bond_ab_arp_probe(struct bonding *bond)
3036 struct slave *slave;
3039 read_lock(&bond->curr_slave_lock);
3041 if (bond->current_arp_slave && bond->curr_active_slave)
3042 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3043 bond->current_arp_slave->dev->name,
3044 bond->curr_active_slave->dev->name);
3046 if (bond->curr_active_slave) {
3047 bond_arp_send_all(bond, bond->curr_active_slave);
3048 read_unlock(&bond->curr_slave_lock);
3052 read_unlock(&bond->curr_slave_lock);
3054 /* if we don't have a curr_active_slave, search for the next available
3055 * backup slave from the current_arp_slave and make it the candidate
3056 * for becoming the curr_active_slave
3059 if (!bond->current_arp_slave) {
3060 bond->current_arp_slave = bond->first_slave;
3061 if (!bond->current_arp_slave)
3065 bond_set_slave_inactive_flags(bond->current_arp_slave);
3067 /* search for next candidate */
3068 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3069 if (IS_UP(slave->dev)) {
3070 slave->link = BOND_LINK_BACK;
3071 bond_set_slave_active_flags(slave);
3072 bond_arp_send_all(bond, slave);
3073 slave->jiffies = jiffies;
3074 bond->current_arp_slave = slave;
3078 /* if the link state is up at this point, we
3079 * mark it down - this can happen if we have
3080 * simultaneous link failures and
3081 * reselect_active_interface doesn't make this
3082 * one the current slave so it is still marked
3083 * up when it is actually down
3085 if (slave->link == BOND_LINK_UP) {
3086 slave->link = BOND_LINK_DOWN;
3087 if (slave->link_failure_count < UINT_MAX)
3088 slave->link_failure_count++;
3090 bond_set_slave_inactive_flags(slave);
3092 pr_info("%s: backup interface %s is now down.\n",
3093 bond->dev->name, slave->dev->name);
3098 void bond_activebackup_arp_mon(struct work_struct *work)
3100 struct bonding *bond = container_of(work, struct bonding,
3102 bool should_notify_peers = false;
3105 read_lock(&bond->lock);
3107 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3109 if (bond->slave_cnt == 0)
3112 should_notify_peers = bond_should_notify_peers(bond);
3114 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3115 read_unlock(&bond->lock);
3117 /* Race avoidance with bond_close flush of workqueue */
3118 if (!rtnl_trylock()) {
3119 read_lock(&bond->lock);
3121 should_notify_peers = false;
3125 read_lock(&bond->lock);
3127 bond_ab_arp_commit(bond, delta_in_ticks);
3129 read_unlock(&bond->lock);
3131 read_lock(&bond->lock);
3134 bond_ab_arp_probe(bond);
3137 if (bond->params.arp_interval)
3138 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3140 read_unlock(&bond->lock);
3142 if (should_notify_peers) {
3143 if (!rtnl_trylock()) {
3144 read_lock(&bond->lock);
3145 bond->send_peer_notif++;
3146 read_unlock(&bond->lock);
3149 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3154 /*-------------------------- netdev event handling --------------------------*/
3157 * Change device name
3159 static int bond_event_changename(struct bonding *bond)
3161 bond_remove_proc_entry(bond);
3162 bond_create_proc_entry(bond);
3164 bond_debug_reregister(bond);
3169 static int bond_master_netdev_event(unsigned long event,
3170 struct net_device *bond_dev)
3172 struct bonding *event_bond = netdev_priv(bond_dev);
3175 case NETDEV_CHANGENAME:
3176 return bond_event_changename(event_bond);
3184 static int bond_slave_netdev_event(unsigned long event,
3185 struct net_device *slave_dev)
3187 struct net_device *bond_dev = slave_dev->master;
3188 struct bonding *bond = netdev_priv(bond_dev);
3189 struct slave *slave = NULL;
3192 case NETDEV_UNREGISTER:
3194 if (bond->setup_by_slave)
3195 bond_release_and_destroy(bond_dev, slave_dev);
3197 bond_release(bond_dev, slave_dev);
3202 slave = bond_get_slave_by_dev(bond, slave_dev);
3204 u32 old_speed = slave->speed;
3205 u8 old_duplex = slave->duplex;
3207 bond_update_speed_duplex(slave);
3209 if (bond->params.mode == BOND_MODE_8023AD) {
3210 if (old_speed != slave->speed)
3211 bond_3ad_adapter_speed_changed(slave);
3212 if (old_duplex != slave->duplex)
3213 bond_3ad_adapter_duplex_changed(slave);
3220 * ... Or is it this?
3223 case NETDEV_CHANGEMTU:
3225 * TODO: Should slaves be allowed to
3226 * independently alter their MTU? For
3227 * an active-backup bond, slaves need
3228 * not be the same type of device, so
3229 * MTUs may vary. For other modes,
3230 * slaves arguably should have the
3231 * same MTUs. To do this, we'd need to
3232 * take over the slave's change_mtu
3233 * function for the duration of their
3237 case NETDEV_CHANGENAME:
3239 * TODO: handle changing the primary's name
3242 case NETDEV_FEAT_CHANGE:
3243 bond_compute_features(bond);
3253 * bond_netdev_event: handle netdev notifier chain events.
3255 * This function receives events for the netdev chain. The caller (an
3256 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3257 * locks for us to safely manipulate the slave devices (RTNL lock,
3260 static int bond_netdev_event(struct notifier_block *this,
3261 unsigned long event, void *ptr)
3263 struct net_device *event_dev = (struct net_device *)ptr;
3265 pr_debug("event_dev: %s, event: %lx\n",
3266 event_dev ? event_dev->name : "None",
3269 if (!(event_dev->priv_flags & IFF_BONDING))
3272 if (event_dev->flags & IFF_MASTER) {
3273 pr_debug("IFF_MASTER\n");
3274 return bond_master_netdev_event(event, event_dev);
3277 if (event_dev->flags & IFF_SLAVE) {
3278 pr_debug("IFF_SLAVE\n");
3279 return bond_slave_netdev_event(event, event_dev);
3286 * bond_inetaddr_event: handle inetaddr notifier chain events.
3288 * We keep track of device IPs primarily to use as source addresses in
3289 * ARP monitor probes (rather than spewing out broadcasts all the time).
3291 * We track one IP for the main device (if it has one), plus one per VLAN.
3293 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3295 struct in_ifaddr *ifa = ptr;
3296 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3297 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3298 struct bonding *bond;
3299 struct vlan_entry *vlan;
3301 /* we only care about primary address */
3302 if(ifa->ifa_flags & IFA_F_SECONDARY)
3305 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3306 if (bond->dev == event_dev) {
3309 bond->master_ip = ifa->ifa_local;
3312 bond->master_ip = 0;
3319 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3320 vlan_dev = __vlan_find_dev_deep(bond->dev,
3322 if (vlan_dev == event_dev) {
3325 vlan->vlan_ip = ifa->ifa_local;
3339 static struct notifier_block bond_netdev_notifier = {
3340 .notifier_call = bond_netdev_event,
3343 static struct notifier_block bond_inetaddr_notifier = {
3344 .notifier_call = bond_inetaddr_event,
3347 /*---------------------------- Hashing Policies -----------------------------*/
3350 * Hash for the output device based upon layer 2 and layer 3 data. If
3351 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3353 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3355 struct ethhdr *data = (struct ethhdr *)skb->data;
3356 struct iphdr *iph = ip_hdr(skb);
3358 if (skb->protocol == htons(ETH_P_IP)) {
3359 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3360 (data->h_dest[5] ^ data->h_source[5])) % count;
3363 return (data->h_dest[5] ^ data->h_source[5]) % count;
3367 * Hash for the output device based upon layer 3 and layer 4 data. If
3368 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3369 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3371 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3373 struct ethhdr *data = (struct ethhdr *)skb->data;
3374 struct iphdr *iph = ip_hdr(skb);
3375 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3378 if (skb->protocol == htons(ETH_P_IP)) {
3379 if (!ip_is_fragment(iph) &&
3380 (iph->protocol == IPPROTO_TCP ||
3381 iph->protocol == IPPROTO_UDP)) {
3382 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3384 return (layer4_xor ^
3385 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3389 return (data->h_dest[5] ^ data->h_source[5]) % count;
3393 * Hash for the output device based upon layer 2 data
3395 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3397 struct ethhdr *data = (struct ethhdr *)skb->data;
3399 return (data->h_dest[5] ^ data->h_source[5]) % count;
3402 /*-------------------------- Device entry points ----------------------------*/
3404 static int bond_open(struct net_device *bond_dev)
3406 struct bonding *bond = netdev_priv(bond_dev);
3407 struct slave *slave;
3410 /* reset slave->backup and slave->inactive */
3411 read_lock(&bond->lock);
3412 if (bond->slave_cnt > 0) {
3413 read_lock(&bond->curr_slave_lock);
3414 bond_for_each_slave(bond, slave, i) {
3415 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3416 && (slave != bond->curr_active_slave)) {
3417 bond_set_slave_inactive_flags(slave);
3419 bond_set_slave_active_flags(slave);
3422 read_unlock(&bond->curr_slave_lock);
3424 read_unlock(&bond->lock);
3426 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3428 if (bond_is_lb(bond)) {
3429 /* bond_alb_initialize must be called before the timer
3432 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3433 /* something went wrong - fail the open operation */
3437 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3438 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3441 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3442 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3443 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3446 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3447 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3448 INIT_DELAYED_WORK(&bond->arp_work,
3449 bond_activebackup_arp_mon);
3451 INIT_DELAYED_WORK(&bond->arp_work,
3452 bond_loadbalance_arp_mon);
3454 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3455 if (bond->params.arp_validate)
3456 bond->recv_probe = bond_arp_rcv;
3459 if (bond->params.mode == BOND_MODE_8023AD) {
3460 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3461 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3462 /* register to receive LACPDUs */
3463 bond->recv_probe = bond_3ad_lacpdu_recv;
3464 bond_3ad_initiate_agg_selection(bond, 1);
3470 static int bond_close(struct net_device *bond_dev)
3472 struct bonding *bond = netdev_priv(bond_dev);
3474 write_lock_bh(&bond->lock);
3476 bond->send_peer_notif = 0;
3478 write_unlock_bh(&bond->lock);
3480 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3481 cancel_delayed_work_sync(&bond->mii_work);
3484 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3485 cancel_delayed_work_sync(&bond->arp_work);
3488 switch (bond->params.mode) {
3489 case BOND_MODE_8023AD:
3490 cancel_delayed_work_sync(&bond->ad_work);
3494 cancel_delayed_work_sync(&bond->alb_work);
3500 if (delayed_work_pending(&bond->mcast_work))
3501 cancel_delayed_work_sync(&bond->mcast_work);
3503 if (bond_is_lb(bond)) {
3504 /* Must be called only after all
3505 * slaves have been released
3507 bond_alb_deinitialize(bond);
3509 bond->recv_probe = NULL;
3514 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3515 struct rtnl_link_stats64 *stats)
3517 struct bonding *bond = netdev_priv(bond_dev);
3518 struct rtnl_link_stats64 temp;
3519 struct slave *slave;
3522 memset(stats, 0, sizeof(*stats));
3524 read_lock_bh(&bond->lock);
3526 bond_for_each_slave(bond, slave, i) {
3527 const struct rtnl_link_stats64 *sstats =
3528 dev_get_stats(slave->dev, &temp);
3530 stats->rx_packets += sstats->rx_packets;
3531 stats->rx_bytes += sstats->rx_bytes;
3532 stats->rx_errors += sstats->rx_errors;
3533 stats->rx_dropped += sstats->rx_dropped;
3535 stats->tx_packets += sstats->tx_packets;
3536 stats->tx_bytes += sstats->tx_bytes;
3537 stats->tx_errors += sstats->tx_errors;
3538 stats->tx_dropped += sstats->tx_dropped;
3540 stats->multicast += sstats->multicast;
3541 stats->collisions += sstats->collisions;
3543 stats->rx_length_errors += sstats->rx_length_errors;
3544 stats->rx_over_errors += sstats->rx_over_errors;
3545 stats->rx_crc_errors += sstats->rx_crc_errors;
3546 stats->rx_frame_errors += sstats->rx_frame_errors;
3547 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3548 stats->rx_missed_errors += sstats->rx_missed_errors;
3550 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3551 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3552 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3553 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3554 stats->tx_window_errors += sstats->tx_window_errors;
3557 read_unlock_bh(&bond->lock);
3562 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3564 struct net_device *slave_dev = NULL;
3565 struct ifbond k_binfo;
3566 struct ifbond __user *u_binfo = NULL;
3567 struct ifslave k_sinfo;
3568 struct ifslave __user *u_sinfo = NULL;
3569 struct mii_ioctl_data *mii = NULL;
3572 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3584 * We do this again just in case we were called by SIOCGMIIREG
3585 * instead of SIOCGMIIPHY.
3592 if (mii->reg_num == 1) {
3593 struct bonding *bond = netdev_priv(bond_dev);
3595 read_lock(&bond->lock);
3596 read_lock(&bond->curr_slave_lock);
3597 if (netif_carrier_ok(bond->dev))
3598 mii->val_out = BMSR_LSTATUS;
3600 read_unlock(&bond->curr_slave_lock);
3601 read_unlock(&bond->lock);
3605 case BOND_INFO_QUERY_OLD:
3606 case SIOCBONDINFOQUERY:
3607 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3609 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3612 res = bond_info_query(bond_dev, &k_binfo);
3614 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3618 case BOND_SLAVE_INFO_QUERY_OLD:
3619 case SIOCBONDSLAVEINFOQUERY:
3620 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3622 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3625 res = bond_slave_info_query(bond_dev, &k_sinfo);
3627 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3636 if (!capable(CAP_NET_ADMIN))
3639 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3641 pr_debug("slave_dev=%p:\n", slave_dev);
3646 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3648 case BOND_ENSLAVE_OLD:
3649 case SIOCBONDENSLAVE:
3650 res = bond_enslave(bond_dev, slave_dev);
3652 case BOND_RELEASE_OLD:
3653 case SIOCBONDRELEASE:
3654 res = bond_release(bond_dev, slave_dev);
3656 case BOND_SETHWADDR_OLD:
3657 case SIOCBONDSETHWADDR:
3658 res = bond_sethwaddr(bond_dev, slave_dev);
3660 case BOND_CHANGE_ACTIVE_OLD:
3661 case SIOCBONDCHANGEACTIVE:
3662 res = bond_ioctl_change_active(bond_dev, slave_dev);
3674 static bool bond_addr_in_mc_list(unsigned char *addr,
3675 struct netdev_hw_addr_list *list,
3678 struct netdev_hw_addr *ha;
3680 netdev_hw_addr_list_for_each(ha, list)
3681 if (!memcmp(ha->addr, addr, addrlen))
3687 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3689 struct bonding *bond = netdev_priv(bond_dev);
3691 if (change & IFF_PROMISC)
3692 bond_set_promiscuity(bond,
3693 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3695 if (change & IFF_ALLMULTI)
3696 bond_set_allmulti(bond,
3697 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3700 static void bond_set_multicast_list(struct net_device *bond_dev)
3702 struct bonding *bond = netdev_priv(bond_dev);
3703 struct netdev_hw_addr *ha;
3706 read_lock(&bond->lock);
3708 /* looking for addresses to add to slaves' mc list */
3709 netdev_for_each_mc_addr(ha, bond_dev) {
3710 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3711 bond_dev->addr_len);
3713 bond_mc_add(bond, ha->addr);
3716 /* looking for addresses to delete from slaves' list */
3717 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3718 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3719 bond_dev->addr_len);
3721 bond_mc_del(bond, ha->addr);
3724 /* save master's multicast list */
3725 __hw_addr_flush(&bond->mc_list);
3726 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3727 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3729 read_unlock(&bond->lock);
3732 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3734 struct bonding *bond = netdev_priv(dev);
3735 struct slave *slave = bond->first_slave;
3738 const struct net_device_ops *slave_ops
3739 = slave->dev->netdev_ops;
3740 if (slave_ops->ndo_neigh_setup)
3741 return slave_ops->ndo_neigh_setup(slave->dev, parms);
3747 * Change the MTU of all of a master's slaves to match the master
3749 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3751 struct bonding *bond = netdev_priv(bond_dev);
3752 struct slave *slave, *stop_at;
3756 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3757 (bond_dev ? bond_dev->name : "None"), new_mtu);
3759 /* Can't hold bond->lock with bh disabled here since
3760 * some base drivers panic. On the other hand we can't
3761 * hold bond->lock without bh disabled because we'll
3762 * deadlock. The only solution is to rely on the fact
3763 * that we're under rtnl_lock here, and the slaves
3764 * list won't change. This doesn't solve the problem
3765 * of setting the slave's MTU while it is
3766 * transmitting, but the assumption is that the base
3767 * driver can handle that.
3769 * TODO: figure out a way to safely iterate the slaves
3770 * list, but without holding a lock around the actual
3771 * call to the base driver.
3774 bond_for_each_slave(bond, slave, i) {
3775 pr_debug("s %p s->p %p c_m %p\n",
3778 slave->dev->netdev_ops->ndo_change_mtu);
3780 res = dev_set_mtu(slave->dev, new_mtu);
3783 /* If we failed to set the slave's mtu to the new value
3784 * we must abort the operation even in ACTIVE_BACKUP
3785 * mode, because if we allow the backup slaves to have
3786 * different mtu values than the active slave we'll
3787 * need to change their mtu when doing a failover. That
3788 * means changing their mtu from timer context, which
3789 * is probably not a good idea.
3791 pr_debug("err %d %s\n", res, slave->dev->name);
3796 bond_dev->mtu = new_mtu;
3801 /* unwind from head to the slave that failed */
3803 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3806 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3808 pr_debug("unwind err %d dev %s\n",
3809 tmp_res, slave->dev->name);
3819 * Note that many devices must be down to change the HW address, and
3820 * downing the master releases all slaves. We can make bonds full of
3821 * bonding devices to test this, however.
3823 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3825 struct bonding *bond = netdev_priv(bond_dev);
3826 struct sockaddr *sa = addr, tmp_sa;
3827 struct slave *slave, *stop_at;
3831 if (bond->params.mode == BOND_MODE_ALB)
3832 return bond_alb_set_mac_address(bond_dev, addr);
3835 pr_debug("bond=%p, name=%s\n",
3836 bond, bond_dev ? bond_dev->name : "None");
3839 * If fail_over_mac is set to active, do nothing and return
3840 * success. Returning an error causes ifenslave to fail.
3842 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3845 if (!is_valid_ether_addr(sa->sa_data))
3846 return -EADDRNOTAVAIL;
3848 /* Can't hold bond->lock with bh disabled here since
3849 * some base drivers panic. On the other hand we can't
3850 * hold bond->lock without bh disabled because we'll
3851 * deadlock. The only solution is to rely on the fact
3852 * that we're under rtnl_lock here, and the slaves
3853 * list won't change. This doesn't solve the problem
3854 * of setting the slave's hw address while it is
3855 * transmitting, but the assumption is that the base
3856 * driver can handle that.
3858 * TODO: figure out a way to safely iterate the slaves
3859 * list, but without holding a lock around the actual
3860 * call to the base driver.
3863 bond_for_each_slave(bond, slave, i) {
3864 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3865 pr_debug("slave %p %s\n", slave, slave->dev->name);
3867 if (slave_ops->ndo_set_mac_address == NULL) {
3869 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3873 res = dev_set_mac_address(slave->dev, addr);
3875 /* TODO: consider downing the slave
3877 * User should expect communications
3878 * breakage anyway until ARP finish
3881 pr_debug("err %d %s\n", res, slave->dev->name);
3887 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3891 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3892 tmp_sa.sa_family = bond_dev->type;
3894 /* unwind from head to the slave that failed */
3896 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3899 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3901 pr_debug("unwind err %d dev %s\n",
3902 tmp_res, slave->dev->name);
3909 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3911 struct bonding *bond = netdev_priv(bond_dev);
3912 struct slave *slave, *start_at;
3913 int i, slave_no, res = 1;
3914 struct iphdr *iph = ip_hdr(skb);
3917 * Start with the curr_active_slave that joined the bond as the
3918 * default for sending IGMP traffic. For failover purposes one
3919 * needs to maintain some consistency for the interface that will
3920 * send the join/membership reports. The curr_active_slave found
3921 * will send all of this type of traffic.
3923 if ((iph->protocol == IPPROTO_IGMP) &&
3924 (skb->protocol == htons(ETH_P_IP))) {
3926 read_lock(&bond->curr_slave_lock);
3927 slave = bond->curr_active_slave;
3928 read_unlock(&bond->curr_slave_lock);
3934 * Concurrent TX may collide on rr_tx_counter; we accept
3935 * that as being rare enough not to justify using an
3938 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3940 bond_for_each_slave(bond, slave, i) {
3948 bond_for_each_slave_from(bond, slave, i, start_at) {
3949 if (IS_UP(slave->dev) &&
3950 (slave->link == BOND_LINK_UP) &&
3951 bond_is_active_slave(slave)) {
3952 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3959 /* no suitable interface, frame not sent */
3963 return NETDEV_TX_OK;
3968 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3969 * the bond has a usable interface.
3971 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3973 struct bonding *bond = netdev_priv(bond_dev);
3976 read_lock(&bond->curr_slave_lock);
3978 if (bond->curr_active_slave)
3979 res = bond_dev_queue_xmit(bond, skb,
3980 bond->curr_active_slave->dev);
3983 /* no suitable interface, frame not sent */
3986 read_unlock(&bond->curr_slave_lock);
3988 return NETDEV_TX_OK;
3992 * In bond_xmit_xor() , we determine the output device by using a pre-
3993 * determined xmit_hash_policy(), If the selected device is not enabled,
3994 * find the next active slave.
3996 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3998 struct bonding *bond = netdev_priv(bond_dev);
3999 struct slave *slave, *start_at;
4004 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4006 bond_for_each_slave(bond, slave, i) {
4014 bond_for_each_slave_from(bond, slave, i, start_at) {
4015 if (IS_UP(slave->dev) &&
4016 (slave->link == BOND_LINK_UP) &&
4017 bond_is_active_slave(slave)) {
4018 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4024 /* no suitable interface, frame not sent */
4028 return NETDEV_TX_OK;
4032 * in broadcast mode, we send everything to all usable interfaces.
4034 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4036 struct bonding *bond = netdev_priv(bond_dev);
4037 struct slave *slave, *start_at;
4038 struct net_device *tx_dev = NULL;
4042 read_lock(&bond->curr_slave_lock);
4043 start_at = bond->curr_active_slave;
4044 read_unlock(&bond->curr_slave_lock);
4049 bond_for_each_slave_from(bond, slave, i, start_at) {
4050 if (IS_UP(slave->dev) &&
4051 (slave->link == BOND_LINK_UP) &&
4052 bond_is_active_slave(slave)) {
4054 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4056 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4061 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4063 dev_kfree_skb(skb2);
4067 tx_dev = slave->dev;
4072 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4076 /* no suitable interface, frame not sent */
4079 /* frame sent to all suitable interfaces */
4080 return NETDEV_TX_OK;
4083 /*------------------------- Device initialization ---------------------------*/
4085 static void bond_set_xmit_hash_policy(struct bonding *bond)
4087 switch (bond->params.xmit_policy) {
4088 case BOND_XMIT_POLICY_LAYER23:
4089 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4091 case BOND_XMIT_POLICY_LAYER34:
4092 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4094 case BOND_XMIT_POLICY_LAYER2:
4096 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4102 * Lookup the slave that corresponds to a qid
4104 static inline int bond_slave_override(struct bonding *bond,
4105 struct sk_buff *skb)
4108 struct slave *slave = NULL;
4109 struct slave *check_slave;
4111 if (!skb->queue_mapping)
4114 /* Find out if any slaves have the same mapping as this skb. */
4115 bond_for_each_slave(bond, check_slave, i) {
4116 if (check_slave->queue_id == skb->queue_mapping) {
4117 slave = check_slave;
4122 /* If the slave isn't UP, use default transmit policy. */
4123 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4124 (slave->link == BOND_LINK_UP)) {
4125 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4132 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4135 * This helper function exists to help dev_pick_tx get the correct
4136 * destination queue. Using a helper function skips a call to
4137 * skb_tx_hash and will put the skbs in the queue we expect on their
4138 * way down to the bonding driver.
4140 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4143 * Save the original txq to restore before passing to the driver
4145 bond_queue_mapping(skb) = skb->queue_mapping;
4147 if (unlikely(txq >= dev->real_num_tx_queues)) {
4149 txq -= dev->real_num_tx_queues;
4150 } while (txq >= dev->real_num_tx_queues);
4155 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4157 struct bonding *bond = netdev_priv(dev);
4159 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4160 if (!bond_slave_override(bond, skb))
4161 return NETDEV_TX_OK;
4164 switch (bond->params.mode) {
4165 case BOND_MODE_ROUNDROBIN:
4166 return bond_xmit_roundrobin(skb, dev);
4167 case BOND_MODE_ACTIVEBACKUP:
4168 return bond_xmit_activebackup(skb, dev);
4170 return bond_xmit_xor(skb, dev);
4171 case BOND_MODE_BROADCAST:
4172 return bond_xmit_broadcast(skb, dev);
4173 case BOND_MODE_8023AD:
4174 return bond_3ad_xmit_xor(skb, dev);
4177 return bond_alb_xmit(skb, dev);
4179 /* Should never happen, mode already checked */
4180 pr_err("%s: Error: Unknown bonding mode %d\n",
4181 dev->name, bond->params.mode);
4184 return NETDEV_TX_OK;
4188 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4190 struct bonding *bond = netdev_priv(dev);
4191 netdev_tx_t ret = NETDEV_TX_OK;
4194 * If we risk deadlock from transmitting this in the
4195 * netpoll path, tell netpoll to queue the frame for later tx
4197 if (is_netpoll_tx_blocked(dev))
4198 return NETDEV_TX_BUSY;
4200 read_lock(&bond->lock);
4202 if (bond->slave_cnt)
4203 ret = __bond_start_xmit(skb, dev);
4207 read_unlock(&bond->lock);
4213 * set bond mode specific net device operations
4215 void bond_set_mode_ops(struct bonding *bond, int mode)
4217 struct net_device *bond_dev = bond->dev;
4220 case BOND_MODE_ROUNDROBIN:
4222 case BOND_MODE_ACTIVEBACKUP:
4225 bond_set_xmit_hash_policy(bond);
4227 case BOND_MODE_BROADCAST:
4229 case BOND_MODE_8023AD:
4230 bond_set_xmit_hash_policy(bond);
4237 /* Should never happen, mode already checked */
4238 pr_err("%s: Error: Unknown bonding mode %d\n",
4239 bond_dev->name, mode);
4244 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4245 struct ethtool_drvinfo *drvinfo)
4247 strncpy(drvinfo->driver, DRV_NAME, 32);
4248 strncpy(drvinfo->version, DRV_VERSION, 32);
4249 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4252 static const struct ethtool_ops bond_ethtool_ops = {
4253 .get_drvinfo = bond_ethtool_get_drvinfo,
4254 .get_link = ethtool_op_get_link,
4257 static const struct net_device_ops bond_netdev_ops = {
4258 .ndo_init = bond_init,
4259 .ndo_uninit = bond_uninit,
4260 .ndo_open = bond_open,
4261 .ndo_stop = bond_close,
4262 .ndo_start_xmit = bond_start_xmit,
4263 .ndo_select_queue = bond_select_queue,
4264 .ndo_get_stats64 = bond_get_stats,
4265 .ndo_do_ioctl = bond_do_ioctl,
4266 .ndo_change_rx_flags = bond_change_rx_flags,
4267 .ndo_set_rx_mode = bond_set_multicast_list,
4268 .ndo_change_mtu = bond_change_mtu,
4269 .ndo_set_mac_address = bond_set_mac_address,
4270 .ndo_neigh_setup = bond_neigh_setup,
4271 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4272 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4273 #ifdef CONFIG_NET_POLL_CONTROLLER
4274 .ndo_netpoll_setup = bond_netpoll_setup,
4275 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4276 .ndo_poll_controller = bond_poll_controller,
4278 .ndo_add_slave = bond_enslave,
4279 .ndo_del_slave = bond_release,
4280 .ndo_fix_features = bond_fix_features,
4283 static void bond_destructor(struct net_device *bond_dev)
4285 struct bonding *bond = netdev_priv(bond_dev);
4287 destroy_workqueue(bond->wq);
4288 free_netdev(bond_dev);
4291 static void bond_setup(struct net_device *bond_dev)
4293 struct bonding *bond = netdev_priv(bond_dev);
4295 /* initialize rwlocks */
4296 rwlock_init(&bond->lock);
4297 rwlock_init(&bond->curr_slave_lock);
4299 bond->params = bonding_defaults;
4301 /* Initialize pointers */
4302 bond->dev = bond_dev;
4303 INIT_LIST_HEAD(&bond->vlan_list);
4305 /* Initialize the device entry points */
4306 ether_setup(bond_dev);
4307 bond_dev->netdev_ops = &bond_netdev_ops;
4308 bond_dev->ethtool_ops = &bond_ethtool_ops;
4309 bond_set_mode_ops(bond, bond->params.mode);
4311 bond_dev->destructor = bond_destructor;
4313 /* Initialize the device options */
4314 bond_dev->tx_queue_len = 0;
4315 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4316 bond_dev->priv_flags |= IFF_BONDING;
4317 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4319 /* At first, we block adding VLANs. That's the only way to
4320 * prevent problems that occur when adding VLANs over an
4321 * empty bond. The block will be removed once non-challenged
4322 * slaves are enslaved.
4324 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4326 /* don't acquire bond device's netif_tx_lock when
4328 bond_dev->features |= NETIF_F_LLTX;
4330 /* By default, we declare the bond to be fully
4331 * VLAN hardware accelerated capable. Special
4332 * care is taken in the various xmit functions
4333 * when there are slaves that are not hw accel
4337 bond_dev->hw_features = BOND_VLAN_FEATURES |
4338 NETIF_F_HW_VLAN_TX |
4339 NETIF_F_HW_VLAN_RX |
4340 NETIF_F_HW_VLAN_FILTER;
4342 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4343 bond_dev->features |= bond_dev->hw_features;
4346 static void bond_work_cancel_all(struct bonding *bond)
4348 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4349 cancel_delayed_work_sync(&bond->mii_work);
4351 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4352 cancel_delayed_work_sync(&bond->arp_work);
4354 if (bond->params.mode == BOND_MODE_ALB &&
4355 delayed_work_pending(&bond->alb_work))
4356 cancel_delayed_work_sync(&bond->alb_work);
4358 if (bond->params.mode == BOND_MODE_8023AD &&
4359 delayed_work_pending(&bond->ad_work))
4360 cancel_delayed_work_sync(&bond->ad_work);
4362 if (delayed_work_pending(&bond->mcast_work))
4363 cancel_delayed_work_sync(&bond->mcast_work);
4367 * Destroy a bonding device.
4368 * Must be under rtnl_lock when this function is called.
4370 static void bond_uninit(struct net_device *bond_dev)
4372 struct bonding *bond = netdev_priv(bond_dev);
4373 struct vlan_entry *vlan, *tmp;
4375 bond_netpoll_cleanup(bond_dev);
4377 /* Release the bonded slaves */
4378 bond_release_all(bond_dev);
4380 list_del(&bond->bond_list);
4382 bond_work_cancel_all(bond);
4384 bond_remove_proc_entry(bond);
4386 bond_debug_unregister(bond);
4388 __hw_addr_flush(&bond->mc_list);
4390 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4391 list_del(&vlan->vlan_list);
4396 /*------------------------- Module initialization ---------------------------*/
4399 * Convert string input module parms. Accept either the
4400 * number of the mode or its string name. A bit complicated because
4401 * some mode names are substrings of other names, and calls from sysfs
4402 * may have whitespace in the name (trailing newlines, for example).
4404 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4406 int modeint = -1, i, rv;
4407 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4409 for (p = (char *)buf; *p; p++)
4410 if (!(isdigit(*p) || isspace(*p)))
4414 rv = sscanf(buf, "%20s", modestr);
4416 rv = sscanf(buf, "%d", &modeint);
4421 for (i = 0; tbl[i].modename; i++) {
4422 if (modeint == tbl[i].mode)
4424 if (strcmp(modestr, tbl[i].modename) == 0)
4431 static int bond_check_params(struct bond_params *params)
4433 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4436 * Convert string parameters.
4439 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4440 if (bond_mode == -1) {
4441 pr_err("Error: Invalid bonding mode \"%s\"\n",
4442 mode == NULL ? "NULL" : mode);
4447 if (xmit_hash_policy) {
4448 if ((bond_mode != BOND_MODE_XOR) &&
4449 (bond_mode != BOND_MODE_8023AD)) {
4450 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4451 bond_mode_name(bond_mode));
4453 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4455 if (xmit_hashtype == -1) {
4456 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4457 xmit_hash_policy == NULL ? "NULL" :
4465 if (bond_mode != BOND_MODE_8023AD) {
4466 pr_info("lacp_rate param is irrelevant in mode %s\n",
4467 bond_mode_name(bond_mode));
4469 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4470 if (lacp_fast == -1) {
4471 pr_err("Error: Invalid lacp rate \"%s\"\n",
4472 lacp_rate == NULL ? "NULL" : lacp_rate);
4479 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4480 if (params->ad_select == -1) {
4481 pr_err("Error: Invalid ad_select \"%s\"\n",
4482 ad_select == NULL ? "NULL" : ad_select);
4486 if (bond_mode != BOND_MODE_8023AD) {
4487 pr_warning("ad_select param only affects 802.3ad mode\n");
4490 params->ad_select = BOND_AD_STABLE;
4493 if (max_bonds < 0) {
4494 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4495 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4496 max_bonds = BOND_DEFAULT_MAX_BONDS;
4500 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4501 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4502 miimon = BOND_LINK_MON_INTERV;
4506 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4511 if (downdelay < 0) {
4512 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4513 downdelay, INT_MAX);
4517 if ((use_carrier != 0) && (use_carrier != 1)) {
4518 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4523 if (num_peer_notif < 0 || num_peer_notif > 255) {
4524 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4529 /* reset values for 802.3ad */
4530 if (bond_mode == BOND_MODE_8023AD) {
4532 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");
4533 pr_warning("Forcing miimon to 100msec\n");
4538 if (tx_queues < 1 || tx_queues > 255) {
4539 pr_warning("Warning: tx_queues (%d) should be between "
4540 "1 and 255, resetting to %d\n",
4541 tx_queues, BOND_DEFAULT_TX_QUEUES);
4542 tx_queues = BOND_DEFAULT_TX_QUEUES;
4545 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4546 pr_warning("Warning: all_slaves_active module parameter (%d), "
4547 "not of valid value (0/1), so it was set to "
4548 "0\n", all_slaves_active);
4549 all_slaves_active = 0;
4552 if (resend_igmp < 0 || resend_igmp > 255) {
4553 pr_warning("Warning: resend_igmp (%d) should be between "
4554 "0 and 255, resetting to %d\n",
4555 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4556 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4559 /* reset values for TLB/ALB */
4560 if ((bond_mode == BOND_MODE_TLB) ||
4561 (bond_mode == BOND_MODE_ALB)) {
4563 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");
4564 pr_warning("Forcing miimon to 100msec\n");
4569 if (bond_mode == BOND_MODE_ALB) {
4570 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",
4575 if (updelay || downdelay) {
4576 /* just warn the user the up/down delay will have
4577 * no effect since miimon is zero...
4579 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",
4580 updelay, downdelay);
4583 /* don't allow arp monitoring */
4585 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4586 miimon, arp_interval);
4590 if ((updelay % miimon) != 0) {
4591 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4593 (updelay / miimon) * miimon);
4598 if ((downdelay % miimon) != 0) {
4599 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4601 (downdelay / miimon) * miimon);
4604 downdelay /= miimon;
4607 if (arp_interval < 0) {
4608 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4609 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4610 arp_interval = BOND_LINK_ARP_INTERV;
4613 for (arp_ip_count = 0;
4614 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4616 /* not complete check, but should be good enough to
4618 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4619 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4620 arp_ip_target[arp_ip_count]);
4623 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4624 arp_target[arp_ip_count] = ip;
4628 if (arp_interval && !arp_ip_count) {
4629 /* don't allow arping if no arp_ip_target given... */
4630 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4636 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4637 pr_err("arp_validate only supported in active-backup mode\n");
4640 if (!arp_interval) {
4641 pr_err("arp_validate requires arp_interval\n");
4645 arp_validate_value = bond_parse_parm(arp_validate,
4647 if (arp_validate_value == -1) {
4648 pr_err("Error: invalid arp_validate \"%s\"\n",
4649 arp_validate == NULL ? "NULL" : arp_validate);
4653 arp_validate_value = 0;
4656 pr_info("MII link monitoring set to %d ms\n", miimon);
4657 } else if (arp_interval) {
4660 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4662 arp_validate_tbl[arp_validate_value].modename,
4665 for (i = 0; i < arp_ip_count; i++)
4666 pr_info(" %s", arp_ip_target[i]);
4670 } else if (max_bonds) {
4671 /* miimon and arp_interval not set, we need one so things
4672 * work as expected, see bonding.txt for details
4674 pr_debug("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");
4677 if (primary && !USES_PRIMARY(bond_mode)) {
4678 /* currently, using a primary only makes sense
4679 * in active backup, TLB or ALB modes
4681 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4682 primary, bond_mode_name(bond_mode));
4686 if (primary && primary_reselect) {
4687 primary_reselect_value = bond_parse_parm(primary_reselect,
4689 if (primary_reselect_value == -1) {
4690 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4692 NULL ? "NULL" : primary_reselect);
4696 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4699 if (fail_over_mac) {
4700 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4702 if (fail_over_mac_value == -1) {
4703 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4704 arp_validate == NULL ? "NULL" : arp_validate);
4708 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4709 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4711 fail_over_mac_value = BOND_FOM_NONE;
4714 /* fill params struct with the proper values */
4715 params->mode = bond_mode;
4716 params->xmit_policy = xmit_hashtype;
4717 params->miimon = miimon;
4718 params->num_peer_notif = num_peer_notif;
4719 params->arp_interval = arp_interval;
4720 params->arp_validate = arp_validate_value;
4721 params->updelay = updelay;
4722 params->downdelay = downdelay;
4723 params->use_carrier = use_carrier;
4724 params->lacp_fast = lacp_fast;
4725 params->primary[0] = 0;
4726 params->primary_reselect = primary_reselect_value;
4727 params->fail_over_mac = fail_over_mac_value;
4728 params->tx_queues = tx_queues;
4729 params->all_slaves_active = all_slaves_active;
4730 params->resend_igmp = resend_igmp;
4731 params->min_links = min_links;
4734 strncpy(params->primary, primary, IFNAMSIZ);
4735 params->primary[IFNAMSIZ - 1] = 0;
4738 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4743 static struct lock_class_key bonding_netdev_xmit_lock_key;
4744 static struct lock_class_key bonding_netdev_addr_lock_key;
4746 static void bond_set_lockdep_class_one(struct net_device *dev,
4747 struct netdev_queue *txq,
4750 lockdep_set_class(&txq->_xmit_lock,
4751 &bonding_netdev_xmit_lock_key);
4754 static void bond_set_lockdep_class(struct net_device *dev)
4756 lockdep_set_class(&dev->addr_list_lock,
4757 &bonding_netdev_addr_lock_key);
4758 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4762 * Called from registration process
4764 static int bond_init(struct net_device *bond_dev)
4766 struct bonding *bond = netdev_priv(bond_dev);
4767 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4768 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4770 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4773 * Initialize locks that may be required during
4774 * en/deslave operations. All of the bond_open work
4775 * (of which this is part) should really be moved to
4776 * a phase prior to dev_open
4778 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4779 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4781 bond->wq = create_singlethread_workqueue(bond_dev->name);
4785 bond_set_lockdep_class(bond_dev);
4787 bond_create_proc_entry(bond);
4788 list_add_tail(&bond->bond_list, &bn->dev_list);
4790 bond_prepare_sysfs_group(bond);
4792 bond_debug_register(bond);
4794 __hw_addr_init(&bond->mc_list);
4798 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4800 if (tb[IFLA_ADDRESS]) {
4801 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4803 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4804 return -EADDRNOTAVAIL;
4809 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[],
4810 unsigned int *num_queues,
4811 unsigned int *real_num_queues)
4813 *num_queues = tx_queues;
4817 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4819 .priv_size = sizeof(struct bonding),
4820 .setup = bond_setup,
4821 .validate = bond_validate,
4822 .get_tx_queues = bond_get_tx_queues,
4825 /* Create a new bond based on the specified name and bonding parameters.
4826 * If name is NULL, obtain a suitable "bond%d" name for us.
4827 * Caller must NOT hold rtnl_lock; we need to release it here before we
4828 * set up our sysfs entries.
4830 int bond_create(struct net *net, const char *name)
4832 struct net_device *bond_dev;
4837 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4838 name ? name : "bond%d",
4839 bond_setup, tx_queues);
4841 pr_err("%s: eek! can't alloc netdev!\n", name);
4846 dev_net_set(bond_dev, net);
4847 bond_dev->rtnl_link_ops = &bond_link_ops;
4849 res = register_netdevice(bond_dev);
4851 netif_carrier_off(bond_dev);
4855 bond_destructor(bond_dev);
4859 static int __net_init bond_net_init(struct net *net)
4861 struct bond_net *bn = net_generic(net, bond_net_id);
4864 INIT_LIST_HEAD(&bn->dev_list);
4866 bond_create_proc_dir(bn);
4867 bond_create_sysfs(bn);
4872 static void __net_exit bond_net_exit(struct net *net)
4874 struct bond_net *bn = net_generic(net, bond_net_id);
4876 bond_destroy_sysfs(bn);
4877 bond_destroy_proc_dir(bn);
4880 static struct pernet_operations bond_net_ops = {
4881 .init = bond_net_init,
4882 .exit = bond_net_exit,
4884 .size = sizeof(struct bond_net),
4887 static int __init bonding_init(void)
4892 pr_info("%s", bond_version);
4894 res = bond_check_params(&bonding_defaults);
4898 res = register_pernet_subsys(&bond_net_ops);
4902 res = rtnl_link_register(&bond_link_ops);
4906 bond_create_debugfs();
4908 for (i = 0; i < max_bonds; i++) {
4909 res = bond_create(&init_net, NULL);
4914 register_netdevice_notifier(&bond_netdev_notifier);
4915 register_inetaddr_notifier(&bond_inetaddr_notifier);
4919 rtnl_link_unregister(&bond_link_ops);
4921 unregister_pernet_subsys(&bond_net_ops);
4926 static void __exit bonding_exit(void)
4928 unregister_netdevice_notifier(&bond_netdev_notifier);
4929 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4931 bond_destroy_debugfs();
4933 rtnl_link_unregister(&bond_link_ops);
4934 unregister_pernet_subsys(&bond_net_ops);
4936 #ifdef CONFIG_NET_POLL_CONTROLLER
4938 * Make sure we don't have an imbalance on our netpoll blocking
4940 WARN_ON(atomic_read(&netpoll_block_tx));
4944 module_init(bonding_init);
4945 module_exit(bonding_exit);
4946 MODULE_LICENSE("GPL");
4947 MODULE_VERSION(DRV_VERSION);
4948 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4949 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4950 MODULE_ALIAS_RTNL_LINK("bond");