2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 #include "bonding_priv.h"
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 static int packets_per_slave = 1;
114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
116 module_param(max_bonds, int, 0);
117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118 module_param(tx_queues, int, 0);
119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126 module_param(miimon, int, 0);
127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128 module_param(updelay, int, 0);
129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130 module_param(downdelay, int, 0);
131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
133 module_param(use_carrier, int, 0);
134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136 module_param(mode, charp, 0);
137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141 module_param(primary, charp, 0);
142 MODULE_PARM_DESC(primary, "Primary network device to use");
143 module_param(primary_reselect, charp, 0);
144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
149 "2 for only on active slave "
151 module_param(lacp_rate, charp, 0);
152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154 module_param(ad_select, charp, 0);
155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
158 module_param(min_links, int, 0);
159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166 module_param(arp_interval, int, 0);
167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168 module_param_array(arp_ip_target, charp, NULL, 0);
169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170 module_param(arp_validate, charp, 0);
171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174 module_param(arp_all_targets, charp, 0);
175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac, charp, 0);
177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180 module_param(all_slaves_active, int, 0);
181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184 module_param(resend_igmp, int, 0);
185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
187 module_param(packets_per_slave, int, 0);
188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191 module_param(lp_interval, uint, 0);
192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
196 /*----------------------------- Global variables ----------------------------*/
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
202 int bond_net_id __read_mostly;
204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
205 static int arp_ip_count;
206 static int bond_mode = BOND_MODE_ROUNDROBIN;
207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
208 static int lacp_fast;
210 /*-------------------------- Forward declarations ---------------------------*/
212 static int bond_init(struct net_device *bond_dev);
213 static void bond_uninit(struct net_device *bond_dev);
214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
215 struct rtnl_link_stats64 *stats);
216 static void bond_slave_arr_handler(struct work_struct *work);
218 /*---------------------------- General routines -----------------------------*/
220 const char *bond_mode_name(int mode)
222 static const char *names[] = {
223 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
224 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
225 [BOND_MODE_XOR] = "load balancing (xor)",
226 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
227 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
228 [BOND_MODE_TLB] = "transmit load balancing",
229 [BOND_MODE_ALB] = "adaptive load balancing",
232 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
238 /*---------------------------------- VLAN -----------------------------------*/
241 * bond_dev_queue_xmit - Prepare skb for xmit.
243 * @bond: bond device that got this skb for tx.
244 * @skb: hw accel VLAN tagged skb to transmit
245 * @slave_dev: slave that is supposed to xmit this skbuff
247 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
248 struct net_device *slave_dev)
250 skb->dev = slave_dev;
252 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
253 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
254 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
256 if (unlikely(netpoll_tx_running(bond->dev)))
257 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
262 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
263 * We don't protect the slave list iteration with a lock because:
264 * a. This operation is performed in IOCTL context,
265 * b. The operation is protected by the RTNL semaphore in the 8021q code,
266 * c. Holding a lock with BH disabled while directly calling a base driver
267 * entry point is generally a BAD idea.
269 * The design of synchronization/protection for this operation in the 8021q
270 * module is good for one or more VLAN devices over a single physical device
271 * and cannot be extended for a teaming solution like bonding, so there is a
272 * potential race condition here where a net device from the vlan group might
273 * be referenced (either by a base driver or the 8021q code) while it is being
274 * removed from the system. However, it turns out we're not making matters
275 * worse, and if it works for regular VLAN usage it will work here too.
279 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
280 * @bond_dev: bonding net device that got called
281 * @vid: vlan id being added
283 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
284 __be16 proto, u16 vid)
286 struct bonding *bond = netdev_priv(bond_dev);
287 struct slave *slave, *rollback_slave;
288 struct list_head *iter;
291 bond_for_each_slave(bond, slave, iter) {
292 res = vlan_vid_add(slave->dev, proto, vid);
300 /* unwind to the slave that failed */
301 bond_for_each_slave(bond, rollback_slave, iter) {
302 if (rollback_slave == slave)
305 vlan_vid_del(rollback_slave->dev, proto, vid);
312 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
313 * @bond_dev: bonding net device that got called
314 * @vid: vlan id being removed
316 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
317 __be16 proto, u16 vid)
319 struct bonding *bond = netdev_priv(bond_dev);
320 struct list_head *iter;
323 bond_for_each_slave(bond, slave, iter)
324 vlan_vid_del(slave->dev, proto, vid);
326 if (bond_is_lb(bond))
327 bond_alb_clear_vlan(bond, vid);
332 /*------------------------------- Link status -------------------------------*/
334 /* Set the carrier state for the master according to the state of its
335 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
336 * do special 802.3ad magic.
338 * Returns zero if carrier state does not change, nonzero if it does.
340 int bond_set_carrier(struct bonding *bond)
342 struct list_head *iter;
345 if (!bond_has_slaves(bond))
348 if (BOND_MODE(bond) == BOND_MODE_8023AD)
349 return bond_3ad_set_carrier(bond);
351 bond_for_each_slave(bond, slave, iter) {
352 if (slave->link == BOND_LINK_UP) {
353 if (!netif_carrier_ok(bond->dev)) {
354 netif_carrier_on(bond->dev);
362 if (netif_carrier_ok(bond->dev)) {
363 netif_carrier_off(bond->dev);
369 /* Get link speed and duplex from the slave's base driver
370 * using ethtool. If for some reason the call fails or the
371 * values are invalid, set speed and duplex to -1,
374 static void bond_update_speed_duplex(struct slave *slave)
376 struct net_device *slave_dev = slave->dev;
377 struct ethtool_cmd ecmd;
381 slave->speed = SPEED_UNKNOWN;
382 slave->duplex = DUPLEX_UNKNOWN;
384 res = __ethtool_get_settings(slave_dev, &ecmd);
388 slave_speed = ethtool_cmd_speed(&ecmd);
389 if (slave_speed == 0 || slave_speed == ((__u32) -1))
392 switch (ecmd.duplex) {
400 slave->speed = slave_speed;
401 slave->duplex = ecmd.duplex;
406 const char *bond_slave_link_status(s8 link)
422 /* if <dev> supports MII link status reporting, check its link status.
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
425 * depending upon the setting of the use_carrier parameter.
427 * Return either BMSR_LSTATUS, meaning that the link is up (or we
428 * can't tell and just pretend it is), or 0, meaning that the link is
431 * If reporting is non-zero, instead of faking link up, return -1 if
432 * both ETHTOOL and MII ioctls fail (meaning the device does not
433 * support them). If use_carrier is set, return whatever it says.
434 * It'd be nice if there was a good way to tell if a driver supports
435 * netif_carrier, but there really isn't.
437 static int bond_check_dev_link(struct bonding *bond,
438 struct net_device *slave_dev, int reporting)
440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441 int (*ioctl)(struct net_device *, struct ifreq *, int);
443 struct mii_ioctl_data *mii;
445 if (!reporting && !netif_running(slave_dev))
448 if (bond->params.use_carrier)
449 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
451 /* Try to get link status using Ethtool first. */
452 if (slave_dev->ethtool_ops->get_link)
453 return slave_dev->ethtool_ops->get_link(slave_dev) ?
456 /* Ethtool can't be used, fallback to MII ioctls. */
457 ioctl = slave_ops->ndo_do_ioctl;
459 /* TODO: set pointer to correct ioctl on a per team member
460 * bases to make this more efficient. that is, once
461 * we determine the correct ioctl, we will always
462 * call it and not the others for that team
466 /* We cannot assume that SIOCGMIIPHY will also read a
467 * register; not all network drivers (e.g., e100)
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
474 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
475 mii->reg_num = MII_BMSR;
476 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
477 return mii->val_out & BMSR_LSTATUS;
481 /* If reporting, report that either there's no dev->do_ioctl,
482 * or both SIOCGMIIREG and get_link failed (meaning that we
483 * cannot report link status). If not reporting, pretend
486 return reporting ? -1 : BMSR_LSTATUS;
489 /*----------------------------- Multicast list ------------------------------*/
491 /* Push the promiscuity flag down to appropriate slaves */
492 static int bond_set_promiscuity(struct bonding *bond, int inc)
494 struct list_head *iter;
497 if (bond_uses_primary(bond)) {
498 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
501 err = dev_set_promiscuity(curr_active->dev, inc);
505 bond_for_each_slave(bond, slave, iter) {
506 err = dev_set_promiscuity(slave->dev, inc);
514 /* Push the allmulti flag down to all slaves */
515 static int bond_set_allmulti(struct bonding *bond, int inc)
517 struct list_head *iter;
520 if (bond_uses_primary(bond)) {
521 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
524 err = dev_set_allmulti(curr_active->dev, inc);
528 bond_for_each_slave(bond, slave, iter) {
529 err = dev_set_allmulti(slave->dev, inc);
537 /* Retrieve the list of registered multicast addresses for the bonding
538 * device and retransmit an IGMP JOIN request to the current active
541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
543 struct bonding *bond = container_of(work, struct bonding,
546 if (!rtnl_trylock()) {
547 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
550 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
552 if (bond->igmp_retrans > 1) {
553 bond->igmp_retrans--;
554 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
559 /* Flush bond's hardware addresses from slave */
560 static void bond_hw_addr_flush(struct net_device *bond_dev,
561 struct net_device *slave_dev)
563 struct bonding *bond = netdev_priv(bond_dev);
565 dev_uc_unsync(slave_dev, bond_dev);
566 dev_mc_unsync(slave_dev, bond_dev);
568 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
569 /* del lacpdu mc addr from mc list */
570 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
572 dev_mc_del(slave_dev, lacpdu_multicast);
576 /*--------------------------- Active slave change ---------------------------*/
578 /* Update the hardware address list and promisc/allmulti for the new and
579 * old active slaves (if any). Modes that are not using primary keep all
580 * slaves up date at all times; only the modes that use primary need to call
581 * this function to swap these settings during a failover.
583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
584 struct slave *old_active)
587 if (bond->dev->flags & IFF_PROMISC)
588 dev_set_promiscuity(old_active->dev, -1);
590 if (bond->dev->flags & IFF_ALLMULTI)
591 dev_set_allmulti(old_active->dev, -1);
593 bond_hw_addr_flush(bond->dev, old_active->dev);
597 /* FIXME: Signal errors upstream. */
598 if (bond->dev->flags & IFF_PROMISC)
599 dev_set_promiscuity(new_active->dev, 1);
601 if (bond->dev->flags & IFF_ALLMULTI)
602 dev_set_allmulti(new_active->dev, 1);
604 netif_addr_lock_bh(bond->dev);
605 dev_uc_sync(new_active->dev, bond->dev);
606 dev_mc_sync(new_active->dev, bond->dev);
607 netif_addr_unlock_bh(bond->dev);
612 * bond_set_dev_addr - clone slave's address to bond
613 * @bond_dev: bond net device
614 * @slave_dev: slave net device
616 * Should be called with RTNL held.
618 static void bond_set_dev_addr(struct net_device *bond_dev,
619 struct net_device *slave_dev)
621 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
622 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
623 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
624 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
625 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
628 static struct slave *bond_get_old_active(struct bonding *bond,
629 struct slave *new_active)
632 struct list_head *iter;
634 bond_for_each_slave(bond, slave, iter) {
635 if (slave == new_active)
638 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
645 /* bond_do_fail_over_mac
647 * Perform special MAC address swapping for fail_over_mac settings
651 static void bond_do_fail_over_mac(struct bonding *bond,
652 struct slave *new_active,
653 struct slave *old_active)
655 u8 tmp_mac[ETH_ALEN];
656 struct sockaddr saddr;
659 switch (bond->params.fail_over_mac) {
660 case BOND_FOM_ACTIVE:
662 bond_set_dev_addr(bond->dev, new_active->dev);
664 case BOND_FOM_FOLLOW:
665 /* if new_active && old_active, swap them
666 * if just old_active, do nothing (going to no active slave)
667 * if just new_active, set new_active to bond's MAC
673 old_active = bond_get_old_active(bond, new_active);
676 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
677 ether_addr_copy(saddr.sa_data,
678 old_active->dev->dev_addr);
679 saddr.sa_family = new_active->dev->type;
681 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
682 saddr.sa_family = bond->dev->type;
685 rv = dev_set_mac_address(new_active->dev, &saddr);
687 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
688 -rv, new_active->dev->name);
695 ether_addr_copy(saddr.sa_data, tmp_mac);
696 saddr.sa_family = old_active->dev->type;
698 rv = dev_set_mac_address(old_active->dev, &saddr);
700 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
701 -rv, new_active->dev->name);
705 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
706 bond->params.fail_over_mac);
712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
714 struct slave *prim = rtnl_dereference(bond->primary_slave);
715 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
717 if (!prim || prim->link != BOND_LINK_UP) {
718 if (!curr || curr->link != BOND_LINK_UP)
723 if (bond->force_primary) {
724 bond->force_primary = false;
728 if (!curr || curr->link != BOND_LINK_UP)
731 /* At this point, prim and curr are both up */
732 switch (bond->params.primary_reselect) {
733 case BOND_PRI_RESELECT_ALWAYS:
735 case BOND_PRI_RESELECT_BETTER:
736 if (prim->speed < curr->speed)
738 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
741 case BOND_PRI_RESELECT_FAILURE:
744 netdev_err(bond->dev, "impossible primary_reselect %d\n",
745 bond->params.primary_reselect);
751 * bond_find_best_slave - select the best available slave to be the active one
752 * @bond: our bonding struct
754 static struct slave *bond_find_best_slave(struct bonding *bond)
756 struct slave *slave, *bestslave = NULL;
757 struct list_head *iter;
758 int mintime = bond->params.updelay;
760 slave = bond_choose_primary_or_current(bond);
764 bond_for_each_slave(bond, slave, iter) {
765 if (slave->link == BOND_LINK_UP)
767 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
768 slave->delay < mintime) {
769 mintime = slave->delay;
777 static bool bond_should_notify_peers(struct bonding *bond)
782 slave = rcu_dereference(bond->curr_active_slave);
785 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
786 slave ? slave->dev->name : "NULL");
788 if (!slave || !bond->send_peer_notif ||
789 !netif_carrier_ok(bond->dev) ||
790 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
797 * change_active_interface - change the active slave into the specified one
798 * @bond: our bonding struct
799 * @new: the new slave to make the active one
801 * Set the new slave to the bond's settings and unset them on the old
803 * Setting include flags, mc-list, promiscuity, allmulti, etc.
805 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
806 * because it is apparently the best available slave we have, even though its
807 * updelay hasn't timed out yet.
809 * Caller must hold RTNL.
811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
813 struct slave *old_active;
817 old_active = rtnl_dereference(bond->curr_active_slave);
819 if (old_active == new_active)
823 new_active->last_link_up = jiffies;
825 if (new_active->link == BOND_LINK_BACK) {
826 if (bond_uses_primary(bond)) {
827 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
828 new_active->dev->name,
829 (bond->params.updelay - new_active->delay) * bond->params.miimon);
832 new_active->delay = 0;
833 bond_set_slave_link_state(new_active, BOND_LINK_UP,
834 BOND_SLAVE_NOTIFY_NOW);
836 if (BOND_MODE(bond) == BOND_MODE_8023AD)
837 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
839 if (bond_is_lb(bond))
840 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
842 if (bond_uses_primary(bond)) {
843 netdev_info(bond->dev, "making interface %s the new active one\n",
844 new_active->dev->name);
849 if (bond_uses_primary(bond))
850 bond_hw_addr_swap(bond, new_active, old_active);
852 if (bond_is_lb(bond)) {
853 bond_alb_handle_active_change(bond, new_active);
855 bond_set_slave_inactive_flags(old_active,
856 BOND_SLAVE_NOTIFY_NOW);
858 bond_set_slave_active_flags(new_active,
859 BOND_SLAVE_NOTIFY_NOW);
861 rcu_assign_pointer(bond->curr_active_slave, new_active);
864 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
866 bond_set_slave_inactive_flags(old_active,
867 BOND_SLAVE_NOTIFY_NOW);
870 bool should_notify_peers = false;
872 bond_set_slave_active_flags(new_active,
873 BOND_SLAVE_NOTIFY_NOW);
875 if (bond->params.fail_over_mac)
876 bond_do_fail_over_mac(bond, new_active,
879 if (netif_running(bond->dev)) {
880 bond->send_peer_notif =
881 bond->params.num_peer_notif;
882 should_notify_peers =
883 bond_should_notify_peers(bond);
886 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
887 if (should_notify_peers)
888 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
893 /* resend IGMP joins since active slave has changed or
894 * all were sent on curr_active_slave.
895 * resend only if bond is brought up with the affected
896 * bonding modes and the retransmission is enabled
898 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
899 ((bond_uses_primary(bond) && new_active) ||
900 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
901 bond->igmp_retrans = bond->params.resend_igmp;
902 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
907 * bond_select_active_slave - select a new active slave, if needed
908 * @bond: our bonding struct
910 * This functions should be called when one of the following occurs:
911 * - The old curr_active_slave has been released or lost its link.
912 * - The primary_slave has got its link back.
913 * - A slave has got its link back and there's no old curr_active_slave.
915 * Caller must hold RTNL.
917 void bond_select_active_slave(struct bonding *bond)
919 struct slave *best_slave;
924 best_slave = bond_find_best_slave(bond);
925 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
926 bond_change_active_slave(bond, best_slave);
927 rv = bond_set_carrier(bond);
931 if (netif_carrier_ok(bond->dev))
932 netdev_info(bond->dev, "first active interface up!\n");
934 netdev_info(bond->dev, "now running without any active interface!\n");
938 #ifdef CONFIG_NET_POLL_CONTROLLER
939 static inline int slave_enable_netpoll(struct slave *slave)
944 np = kzalloc(sizeof(*np), GFP_KERNEL);
949 err = __netpoll_setup(np, slave->dev);
958 static inline void slave_disable_netpoll(struct slave *slave)
960 struct netpoll *np = slave->np;
966 __netpoll_free_async(np);
969 static void bond_poll_controller(struct net_device *bond_dev)
971 struct bonding *bond = netdev_priv(bond_dev);
972 struct slave *slave = NULL;
973 struct list_head *iter;
974 struct ad_info ad_info;
975 struct netpoll_info *ni;
976 const struct net_device_ops *ops;
978 if (BOND_MODE(bond) == BOND_MODE_8023AD)
979 if (bond_3ad_get_active_agg_info(bond, &ad_info))
982 bond_for_each_slave_rcu(bond, slave, iter) {
983 ops = slave->dev->netdev_ops;
984 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
987 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
988 struct aggregator *agg =
989 SLAVE_AD_INFO(slave)->port.aggregator;
992 agg->aggregator_identifier != ad_info.aggregator_id)
996 ni = rcu_dereference_bh(slave->dev->npinfo);
997 if (down_trylock(&ni->dev_lock))
999 ops->ndo_poll_controller(slave->dev);
1004 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1006 struct bonding *bond = netdev_priv(bond_dev);
1007 struct list_head *iter;
1008 struct slave *slave;
1010 bond_for_each_slave(bond, slave, iter)
1011 if (bond_slave_is_up(slave))
1012 slave_disable_netpoll(slave);
1015 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1017 struct bonding *bond = netdev_priv(dev);
1018 struct list_head *iter;
1019 struct slave *slave;
1022 bond_for_each_slave(bond, slave, iter) {
1023 err = slave_enable_netpoll(slave);
1025 bond_netpoll_cleanup(dev);
1032 static inline int slave_enable_netpoll(struct slave *slave)
1036 static inline void slave_disable_netpoll(struct slave *slave)
1039 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1044 /*---------------------------------- IOCTL ----------------------------------*/
1046 static netdev_features_t bond_fix_features(struct net_device *dev,
1047 netdev_features_t features)
1049 struct bonding *bond = netdev_priv(dev);
1050 struct list_head *iter;
1051 netdev_features_t mask;
1052 struct slave *slave;
1056 features &= ~NETIF_F_ONE_FOR_ALL;
1057 features |= NETIF_F_ALL_FOR_ALL;
1059 bond_for_each_slave(bond, slave, iter) {
1060 features = netdev_increment_features(features,
1061 slave->dev->features,
1064 features = netdev_add_tso_features(features, mask);
1069 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1070 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1071 NETIF_F_HIGHDMA | NETIF_F_LRO)
1073 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1074 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1076 static void bond_compute_features(struct bonding *bond)
1078 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1079 IFF_XMIT_DST_RELEASE_PERM;
1080 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1081 netdev_features_t enc_features = BOND_ENC_FEATURES;
1082 struct net_device *bond_dev = bond->dev;
1083 struct list_head *iter;
1084 struct slave *slave;
1085 unsigned short max_hard_header_len = ETH_HLEN;
1086 unsigned int gso_max_size = GSO_MAX_SIZE;
1087 u16 gso_max_segs = GSO_MAX_SEGS;
1089 if (!bond_has_slaves(bond))
1091 vlan_features &= NETIF_F_ALL_FOR_ALL;
1093 bond_for_each_slave(bond, slave, iter) {
1094 vlan_features = netdev_increment_features(vlan_features,
1095 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1097 enc_features = netdev_increment_features(enc_features,
1098 slave->dev->hw_enc_features,
1100 dst_release_flag &= slave->dev->priv_flags;
1101 if (slave->dev->hard_header_len > max_hard_header_len)
1102 max_hard_header_len = slave->dev->hard_header_len;
1104 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1105 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1109 bond_dev->vlan_features = vlan_features;
1110 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1111 bond_dev->hard_header_len = max_hard_header_len;
1112 bond_dev->gso_max_segs = gso_max_segs;
1113 netif_set_gso_max_size(bond_dev, gso_max_size);
1115 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1116 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1117 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1118 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1120 netdev_change_features(bond_dev);
1123 static void bond_setup_by_slave(struct net_device *bond_dev,
1124 struct net_device *slave_dev)
1126 bond_dev->header_ops = slave_dev->header_ops;
1128 bond_dev->type = slave_dev->type;
1129 bond_dev->hard_header_len = slave_dev->hard_header_len;
1130 bond_dev->addr_len = slave_dev->addr_len;
1132 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1133 slave_dev->addr_len);
1136 /* On bonding slaves other than the currently active slave, suppress
1137 * duplicates except for alb non-mcast/bcast.
1139 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1140 struct slave *slave,
1141 struct bonding *bond)
1143 if (bond_is_slave_inactive(slave)) {
1144 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1145 skb->pkt_type != PACKET_BROADCAST &&
1146 skb->pkt_type != PACKET_MULTICAST)
1153 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1155 struct sk_buff *skb = *pskb;
1156 struct slave *slave;
1157 struct bonding *bond;
1158 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1160 int ret = RX_HANDLER_ANOTHER;
1162 skb = skb_share_check(skb, GFP_ATOMIC);
1164 return RX_HANDLER_CONSUMED;
1168 slave = bond_slave_get_rcu(skb->dev);
1171 recv_probe = ACCESS_ONCE(bond->recv_probe);
1173 ret = recv_probe(skb, bond, slave);
1174 if (ret == RX_HANDLER_CONSUMED) {
1180 if (bond_should_deliver_exact_match(skb, slave, bond))
1181 return RX_HANDLER_EXACT;
1183 skb->dev = bond->dev;
1185 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1186 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1187 skb->pkt_type == PACKET_HOST) {
1189 if (unlikely(skb_cow_head(skb,
1190 skb->data - skb_mac_header(skb)))) {
1192 return RX_HANDLER_CONSUMED;
1194 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1200 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1202 switch (BOND_MODE(bond)) {
1203 case BOND_MODE_ROUNDROBIN:
1204 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1205 case BOND_MODE_ACTIVEBACKUP:
1206 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1207 case BOND_MODE_BROADCAST:
1208 return NETDEV_LAG_TX_TYPE_BROADCAST;
1210 case BOND_MODE_8023AD:
1211 return NETDEV_LAG_TX_TYPE_HASH;
1213 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1217 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1219 struct netdev_lag_upper_info lag_upper_info;
1222 lag_upper_info.tx_type = bond_lag_tx_type(bond);
1223 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1227 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1231 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1233 netdev_upper_dev_unlink(slave->dev, bond->dev);
1234 slave->dev->flags &= ~IFF_SLAVE;
1235 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1238 static struct slave *bond_alloc_slave(struct bonding *bond)
1240 struct slave *slave = NULL;
1242 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1246 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1247 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1249 if (!SLAVE_AD_INFO(slave)) {
1257 static void bond_free_slave(struct slave *slave)
1259 struct bonding *bond = bond_get_bond_by_slave(slave);
1261 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1262 kfree(SLAVE_AD_INFO(slave));
1267 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1269 info->bond_mode = BOND_MODE(bond);
1270 info->miimon = bond->params.miimon;
1271 info->num_slaves = bond->slave_cnt;
1274 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1276 strcpy(info->slave_name, slave->dev->name);
1277 info->link = slave->link;
1278 info->state = bond_slave_state(slave);
1279 info->link_failure_count = slave->link_failure_count;
1282 static void bond_netdev_notify(struct net_device *dev,
1283 struct netdev_bonding_info *info)
1286 netdev_bonding_info_change(dev, info);
1290 static void bond_netdev_notify_work(struct work_struct *_work)
1292 struct netdev_notify_work *w =
1293 container_of(_work, struct netdev_notify_work, work.work);
1295 bond_netdev_notify(w->dev, &w->bonding_info);
1300 void bond_queue_slave_event(struct slave *slave)
1302 struct bonding *bond = slave->bond;
1303 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1308 dev_hold(slave->dev);
1309 nnw->dev = slave->dev;
1310 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1311 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1312 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1314 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1317 void bond_lower_state_changed(struct slave *slave)
1319 struct netdev_lag_lower_state_info info;
1321 info.link_up = slave->link == BOND_LINK_UP ||
1322 slave->link == BOND_LINK_FAIL;
1323 info.tx_enabled = bond_is_active_slave(slave);
1324 netdev_lower_state_changed(slave->dev, &info);
1327 /* enslave device <slave> to bond device <master> */
1328 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1330 struct bonding *bond = netdev_priv(bond_dev);
1331 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1332 struct slave *new_slave = NULL, *prev_slave;
1333 struct sockaddr addr;
1337 if (!bond->params.use_carrier &&
1338 slave_dev->ethtool_ops->get_link == NULL &&
1339 slave_ops->ndo_do_ioctl == NULL) {
1340 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1344 /* already enslaved */
1345 if (slave_dev->flags & IFF_SLAVE) {
1346 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1350 if (bond_dev == slave_dev) {
1351 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1355 /* vlan challenged mutual exclusion */
1356 /* no need to lock since we're protected by rtnl_lock */
1357 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1358 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1360 if (vlan_uses_dev(bond_dev)) {
1361 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1362 slave_dev->name, bond_dev->name);
1365 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1366 slave_dev->name, slave_dev->name,
1370 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1374 /* Old ifenslave binaries are no longer supported. These can
1375 * be identified with moderate accuracy by the state of the slave:
1376 * the current ifenslave will set the interface down prior to
1377 * enslaving it; the old ifenslave will not.
1379 if (slave_dev->flags & IFF_UP) {
1380 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1385 /* set bonding device ether type by slave - bonding netdevices are
1386 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1387 * there is a need to override some of the type dependent attribs/funcs.
1389 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1390 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1392 if (!bond_has_slaves(bond)) {
1393 if (bond_dev->type != slave_dev->type) {
1394 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1395 bond_dev->type, slave_dev->type);
1397 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1399 res = notifier_to_errno(res);
1401 netdev_err(bond_dev, "refused to change device type\n");
1405 /* Flush unicast and multicast addresses */
1406 dev_uc_flush(bond_dev);
1407 dev_mc_flush(bond_dev);
1409 if (slave_dev->type != ARPHRD_ETHER)
1410 bond_setup_by_slave(bond_dev, slave_dev);
1412 ether_setup(bond_dev);
1413 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1416 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1419 } else if (bond_dev->type != slave_dev->type) {
1420 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1421 slave_dev->name, slave_dev->type, bond_dev->type);
1425 if (slave_ops->ndo_set_mac_address == NULL) {
1426 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1427 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1428 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1429 if (!bond_has_slaves(bond)) {
1430 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1431 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1433 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1435 goto err_undo_flags;
1440 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1442 /* If this is the first slave, then we need to set the master's hardware
1443 * address to be the same as the slave's.
1445 if (!bond_has_slaves(bond) &&
1446 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1447 bond_set_dev_addr(bond->dev, slave_dev);
1449 new_slave = bond_alloc_slave(bond);
1452 goto err_undo_flags;
1455 new_slave->bond = bond;
1456 new_slave->dev = slave_dev;
1457 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1458 * is set via sysfs or module option if desired.
1460 new_slave->queue_id = 0;
1462 /* Save slave's original mtu and then set it to match the bond */
1463 new_slave->original_mtu = slave_dev->mtu;
1464 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1466 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1470 /* Save slave's original ("permanent") mac address for modes
1471 * that need it, and for restoring it upon release, and then
1472 * set it to the master's address
1474 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1476 if (!bond->params.fail_over_mac ||
1477 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1478 /* Set slave to master's mac address. The application already
1479 * set the master's mac address to that of the first slave
1481 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1482 addr.sa_family = slave_dev->type;
1483 res = dev_set_mac_address(slave_dev, &addr);
1485 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1486 goto err_restore_mtu;
1490 /* set slave flag before open to prevent IPv6 addrconf */
1491 slave_dev->flags |= IFF_SLAVE;
1493 /* open the slave since the application closed it */
1494 res = dev_open(slave_dev);
1496 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1497 goto err_restore_mac;
1500 slave_dev->priv_flags |= IFF_BONDING;
1501 /* initialize slave stats */
1502 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1504 if (bond_is_lb(bond)) {
1505 /* bond_alb_init_slave() must be called before all other stages since
1506 * it might fail and we do not want to have to undo everything
1508 res = bond_alb_init_slave(bond, new_slave);
1513 /* If the mode uses primary, then the following is handled by
1514 * bond_change_active_slave().
1516 if (!bond_uses_primary(bond)) {
1517 /* set promiscuity level to new slave */
1518 if (bond_dev->flags & IFF_PROMISC) {
1519 res = dev_set_promiscuity(slave_dev, 1);
1524 /* set allmulti level to new slave */
1525 if (bond_dev->flags & IFF_ALLMULTI) {
1526 res = dev_set_allmulti(slave_dev, 1);
1531 netif_addr_lock_bh(bond_dev);
1533 dev_mc_sync_multiple(slave_dev, bond_dev);
1534 dev_uc_sync_multiple(slave_dev, bond_dev);
1536 netif_addr_unlock_bh(bond_dev);
1539 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1540 /* add lacpdu mc addr to mc list */
1541 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1543 dev_mc_add(slave_dev, lacpdu_multicast);
1546 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1548 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1553 prev_slave = bond_last_slave(bond);
1555 new_slave->delay = 0;
1556 new_slave->link_failure_count = 0;
1558 bond_update_speed_duplex(new_slave);
1560 new_slave->last_rx = jiffies -
1561 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1562 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1563 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1565 if (bond->params.miimon && !bond->params.use_carrier) {
1566 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1568 if ((link_reporting == -1) && !bond->params.arp_interval) {
1569 /* miimon is set but a bonded network driver
1570 * does not support ETHTOOL/MII and
1571 * arp_interval is not set. Note: if
1572 * use_carrier is enabled, we will never go
1573 * here (because netif_carrier is always
1574 * supported); thus, we don't need to change
1575 * the messages for netif_carrier.
1577 netdev_warn(bond_dev, "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",
1579 } else if (link_reporting == -1) {
1580 /* unable get link status using mii/ethtool */
1581 netdev_warn(bond_dev, "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",
1586 /* check for initial state */
1587 if (bond->params.miimon) {
1588 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1589 if (bond->params.updelay) {
1590 bond_set_slave_link_state(new_slave,
1592 BOND_SLAVE_NOTIFY_NOW);
1593 new_slave->delay = bond->params.updelay;
1595 bond_set_slave_link_state(new_slave,
1597 BOND_SLAVE_NOTIFY_NOW);
1600 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1601 BOND_SLAVE_NOTIFY_NOW);
1603 } else if (bond->params.arp_interval) {
1604 bond_set_slave_link_state(new_slave,
1605 (netif_carrier_ok(slave_dev) ?
1606 BOND_LINK_UP : BOND_LINK_DOWN),
1607 BOND_SLAVE_NOTIFY_NOW);
1609 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1610 BOND_SLAVE_NOTIFY_NOW);
1613 if (new_slave->link != BOND_LINK_DOWN)
1614 new_slave->last_link_up = jiffies;
1615 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1616 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1617 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1619 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1620 /* if there is a primary slave, remember it */
1621 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1622 rcu_assign_pointer(bond->primary_slave, new_slave);
1623 bond->force_primary = true;
1627 switch (BOND_MODE(bond)) {
1628 case BOND_MODE_ACTIVEBACKUP:
1629 bond_set_slave_inactive_flags(new_slave,
1630 BOND_SLAVE_NOTIFY_NOW);
1632 case BOND_MODE_8023AD:
1633 /* in 802.3ad mode, the internal mechanism
1634 * will activate the slaves in the selected
1637 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1638 /* if this is the first slave */
1640 SLAVE_AD_INFO(new_slave)->id = 1;
1641 /* Initialize AD with the number of times that the AD timer is called in 1 second
1642 * can be called only after the mac address of the bond is set
1644 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1646 SLAVE_AD_INFO(new_slave)->id =
1647 SLAVE_AD_INFO(prev_slave)->id + 1;
1650 bond_3ad_bind_slave(new_slave);
1654 bond_set_active_slave(new_slave);
1655 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1658 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1660 /* always active in trunk mode */
1661 bond_set_active_slave(new_slave);
1663 /* In trunking mode there is little meaning to curr_active_slave
1664 * anyway (it holds no special properties of the bond device),
1665 * so we can change it without calling change_active_interface()
1667 if (!rcu_access_pointer(bond->curr_active_slave) &&
1668 new_slave->link == BOND_LINK_UP)
1669 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1672 } /* switch(bond_mode) */
1674 #ifdef CONFIG_NET_POLL_CONTROLLER
1675 slave_dev->npinfo = bond->dev->npinfo;
1676 if (slave_dev->npinfo) {
1677 if (slave_enable_netpoll(new_slave)) {
1678 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1685 if (!(bond_dev->features & NETIF_F_LRO))
1686 dev_disable_lro(slave_dev);
1688 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1691 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1695 res = bond_master_upper_dev_link(bond, new_slave);
1697 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1698 goto err_unregister;
1701 res = bond_sysfs_slave_add(new_slave);
1703 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1704 goto err_upper_unlink;
1708 bond_compute_features(bond);
1709 bond_set_carrier(bond);
1711 if (bond_uses_primary(bond)) {
1713 bond_select_active_slave(bond);
1714 unblock_netpoll_tx();
1717 if (bond_mode_uses_xmit_hash(bond))
1718 bond_update_slave_arr(bond, NULL);
1720 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1722 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1723 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1725 /* enslave is successful */
1726 bond_queue_slave_event(new_slave);
1729 /* Undo stages on error */
1731 bond_upper_dev_unlink(bond, new_slave);
1734 netdev_rx_handler_unregister(slave_dev);
1737 if (!bond_uses_primary(bond))
1738 bond_hw_addr_flush(bond_dev, slave_dev);
1740 vlan_vids_del_by_dev(slave_dev, bond_dev);
1741 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1742 RCU_INIT_POINTER(bond->primary_slave, NULL);
1743 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1745 bond_change_active_slave(bond, NULL);
1746 bond_select_active_slave(bond);
1747 unblock_netpoll_tx();
1749 /* either primary_slave or curr_active_slave might've changed */
1751 slave_disable_netpoll(new_slave);
1754 slave_dev->priv_flags &= ~IFF_BONDING;
1755 dev_close(slave_dev);
1758 slave_dev->flags &= ~IFF_SLAVE;
1759 if (!bond->params.fail_over_mac ||
1760 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1761 /* XXX TODO - fom follow mode needs to change master's
1762 * MAC if this slave's MAC is in use by the bond, or at
1763 * least print a warning.
1765 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1766 addr.sa_family = slave_dev->type;
1767 dev_set_mac_address(slave_dev, &addr);
1771 dev_set_mtu(slave_dev, new_slave->original_mtu);
1774 bond_free_slave(new_slave);
1777 /* Enslave of first slave has failed and we need to fix master's mac */
1778 if (!bond_has_slaves(bond)) {
1779 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1780 slave_dev->dev_addr))
1781 eth_hw_addr_random(bond_dev);
1782 if (bond_dev->type != ARPHRD_ETHER) {
1783 dev_close(bond_dev);
1784 ether_setup(bond_dev);
1785 bond_dev->flags |= IFF_MASTER;
1786 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1793 /* Try to release the slave device <slave> from the bond device <master>
1794 * It is legal to access curr_active_slave without a lock because all the function
1795 * is RTNL-locked. If "all" is true it means that the function is being called
1796 * while destroying a bond interface and all slaves are being released.
1798 * The rules for slave state should be:
1799 * for Active/Backup:
1800 * Active stays on all backups go down
1801 * for Bonded connections:
1802 * The first up interface should be left on and all others downed.
1804 static int __bond_release_one(struct net_device *bond_dev,
1805 struct net_device *slave_dev,
1808 struct bonding *bond = netdev_priv(bond_dev);
1809 struct slave *slave, *oldcurrent;
1810 struct sockaddr addr;
1811 int old_flags = bond_dev->flags;
1812 netdev_features_t old_features = bond_dev->features;
1814 /* slave is not a slave or master is not master of this slave */
1815 if (!(slave_dev->flags & IFF_SLAVE) ||
1816 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1817 netdev_dbg(bond_dev, "cannot release %s\n",
1824 slave = bond_get_slave_by_dev(bond, slave_dev);
1826 /* not a slave of this bond */
1827 netdev_info(bond_dev, "%s not enslaved\n",
1829 unblock_netpoll_tx();
1833 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1835 bond_sysfs_slave_del(slave);
1837 /* recompute stats just before removing the slave */
1838 bond_get_stats(bond->dev, &bond->bond_stats);
1840 bond_upper_dev_unlink(bond, slave);
1841 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1842 * for this slave anymore.
1844 netdev_rx_handler_unregister(slave_dev);
1846 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1847 bond_3ad_unbind_slave(slave);
1849 if (bond_mode_uses_xmit_hash(bond))
1850 bond_update_slave_arr(bond, slave);
1852 netdev_info(bond_dev, "Releasing %s interface %s\n",
1853 bond_is_active_slave(slave) ? "active" : "backup",
1856 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1858 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1860 if (!all && (!bond->params.fail_over_mac ||
1861 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1862 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1863 bond_has_slaves(bond))
1864 netdev_warn(bond_dev, "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",
1865 slave_dev->name, slave->perm_hwaddr,
1866 bond_dev->name, slave_dev->name);
1869 if (rtnl_dereference(bond->primary_slave) == slave)
1870 RCU_INIT_POINTER(bond->primary_slave, NULL);
1872 if (oldcurrent == slave)
1873 bond_change_active_slave(bond, NULL);
1875 if (bond_is_lb(bond)) {
1876 /* Must be called only after the slave has been
1877 * detached from the list and the curr_active_slave
1878 * has been cleared (if our_slave == old_current),
1879 * but before a new active slave is selected.
1881 bond_alb_deinit_slave(bond, slave);
1885 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1886 } else if (oldcurrent == slave) {
1887 /* Note that we hold RTNL over this sequence, so there
1888 * is no concern that another slave add/remove event
1891 bond_select_active_slave(bond);
1894 if (!bond_has_slaves(bond)) {
1895 bond_set_carrier(bond);
1896 eth_hw_addr_random(bond_dev);
1899 unblock_netpoll_tx();
1903 if (!bond_has_slaves(bond)) {
1904 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1905 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1908 bond_compute_features(bond);
1909 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1910 (old_features & NETIF_F_VLAN_CHALLENGED))
1911 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1912 slave_dev->name, bond_dev->name);
1914 vlan_vids_del_by_dev(slave_dev, bond_dev);
1916 /* If the mode uses primary, then this case was handled above by
1917 * bond_change_active_slave(..., NULL)
1919 if (!bond_uses_primary(bond)) {
1920 /* unset promiscuity level from slave
1921 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1922 * of the IFF_PROMISC flag in the bond_dev, but we need the
1923 * value of that flag before that change, as that was the value
1924 * when this slave was attached, so we cache at the start of the
1925 * function and use it here. Same goes for ALLMULTI below
1927 if (old_flags & IFF_PROMISC)
1928 dev_set_promiscuity(slave_dev, -1);
1930 /* unset allmulti level from slave */
1931 if (old_flags & IFF_ALLMULTI)
1932 dev_set_allmulti(slave_dev, -1);
1934 bond_hw_addr_flush(bond_dev, slave_dev);
1937 slave_disable_netpoll(slave);
1939 /* close slave before restoring its mac address */
1940 dev_close(slave_dev);
1942 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1943 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1944 /* restore original ("permanent") mac address */
1945 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1946 addr.sa_family = slave_dev->type;
1947 dev_set_mac_address(slave_dev, &addr);
1950 dev_set_mtu(slave_dev, slave->original_mtu);
1952 slave_dev->priv_flags &= ~IFF_BONDING;
1954 bond_free_slave(slave);
1959 /* A wrapper used because of ndo_del_link */
1960 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1962 return __bond_release_one(bond_dev, slave_dev, false);
1965 /* First release a slave and then destroy the bond if no more slaves are left.
1966 * Must be under rtnl_lock when this function is called.
1968 static int bond_release_and_destroy(struct net_device *bond_dev,
1969 struct net_device *slave_dev)
1971 struct bonding *bond = netdev_priv(bond_dev);
1974 ret = bond_release(bond_dev, slave_dev);
1975 if (ret == 0 && !bond_has_slaves(bond)) {
1976 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1977 netdev_info(bond_dev, "Destroying bond %s\n",
1979 bond_remove_proc_entry(bond);
1980 unregister_netdevice(bond_dev);
1985 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1987 struct bonding *bond = netdev_priv(bond_dev);
1988 bond_fill_ifbond(bond, info);
1992 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1994 struct bonding *bond = netdev_priv(bond_dev);
1995 struct list_head *iter;
1996 int i = 0, res = -ENODEV;
1997 struct slave *slave;
1999 bond_for_each_slave(bond, slave, iter) {
2000 if (i++ == (int)info->slave_id) {
2002 bond_fill_ifslave(slave, info);
2010 /*-------------------------------- Monitoring -------------------------------*/
2012 /* called with rcu_read_lock() */
2013 static int bond_miimon_inspect(struct bonding *bond)
2015 int link_state, commit = 0;
2016 struct list_head *iter;
2017 struct slave *slave;
2018 bool ignore_updelay;
2020 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2022 bond_for_each_slave_rcu(bond, slave, iter) {
2023 slave->new_link = BOND_LINK_NOCHANGE;
2025 link_state = bond_check_dev_link(bond, slave->dev, 0);
2027 switch (slave->link) {
2032 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2033 BOND_SLAVE_NOTIFY_LATER);
2034 slave->delay = bond->params.downdelay;
2036 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2038 BOND_MODE_ACTIVEBACKUP) ?
2039 (bond_is_active_slave(slave) ?
2040 "active " : "backup ") : "",
2042 bond->params.downdelay * bond->params.miimon);
2045 case BOND_LINK_FAIL:
2047 /* recovered before downdelay expired */
2048 bond_set_slave_link_state(slave, BOND_LINK_UP,
2049 BOND_SLAVE_NOTIFY_LATER);
2050 slave->last_link_up = jiffies;
2051 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2052 (bond->params.downdelay - slave->delay) *
2053 bond->params.miimon,
2058 if (slave->delay <= 0) {
2059 slave->new_link = BOND_LINK_DOWN;
2067 case BOND_LINK_DOWN:
2071 bond_set_slave_link_state(slave, BOND_LINK_BACK,
2072 BOND_SLAVE_NOTIFY_LATER);
2073 slave->delay = bond->params.updelay;
2076 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2078 ignore_updelay ? 0 :
2079 bond->params.updelay *
2080 bond->params.miimon);
2083 case BOND_LINK_BACK:
2085 bond_set_slave_link_state(slave,
2087 BOND_SLAVE_NOTIFY_LATER);
2088 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2089 (bond->params.updelay - slave->delay) *
2090 bond->params.miimon,
2099 if (slave->delay <= 0) {
2100 slave->new_link = BOND_LINK_UP;
2102 ignore_updelay = false;
2114 static void bond_miimon_commit(struct bonding *bond)
2116 struct list_head *iter;
2117 struct slave *slave, *primary;
2119 bond_for_each_slave(bond, slave, iter) {
2120 switch (slave->new_link) {
2121 case BOND_LINK_NOCHANGE:
2125 bond_set_slave_link_state(slave, BOND_LINK_UP,
2126 BOND_SLAVE_NOTIFY_NOW);
2127 slave->last_link_up = jiffies;
2129 primary = rtnl_dereference(bond->primary_slave);
2130 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2131 /* prevent it from being the active one */
2132 bond_set_backup_slave(slave);
2133 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2134 /* make it immediately active */
2135 bond_set_active_slave(slave);
2136 } else if (slave != primary) {
2137 /* prevent it from being the active one */
2138 bond_set_backup_slave(slave);
2141 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2143 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2144 slave->duplex ? "full" : "half");
2146 /* notify ad that the link status has changed */
2147 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2148 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2150 if (bond_is_lb(bond))
2151 bond_alb_handle_link_change(bond, slave,
2154 if (BOND_MODE(bond) == BOND_MODE_XOR)
2155 bond_update_slave_arr(bond, NULL);
2157 if (!bond->curr_active_slave || slave == primary)
2162 case BOND_LINK_DOWN:
2163 if (slave->link_failure_count < UINT_MAX)
2164 slave->link_failure_count++;
2166 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2167 BOND_SLAVE_NOTIFY_NOW);
2169 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2170 BOND_MODE(bond) == BOND_MODE_8023AD)
2171 bond_set_slave_inactive_flags(slave,
2172 BOND_SLAVE_NOTIFY_NOW);
2174 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2177 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2178 bond_3ad_handle_link_change(slave,
2181 if (bond_is_lb(bond))
2182 bond_alb_handle_link_change(bond, slave,
2185 if (BOND_MODE(bond) == BOND_MODE_XOR)
2186 bond_update_slave_arr(bond, NULL);
2188 if (slave == rcu_access_pointer(bond->curr_active_slave))
2194 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2195 slave->new_link, slave->dev->name);
2196 slave->new_link = BOND_LINK_NOCHANGE;
2203 bond_select_active_slave(bond);
2204 unblock_netpoll_tx();
2207 bond_set_carrier(bond);
2212 * Really a wrapper that splits the mii monitor into two phases: an
2213 * inspection, then (if inspection indicates something needs to be done)
2214 * an acquisition of appropriate locks followed by a commit phase to
2215 * implement whatever link state changes are indicated.
2217 static void bond_mii_monitor(struct work_struct *work)
2219 struct bonding *bond = container_of(work, struct bonding,
2221 bool should_notify_peers = false;
2222 unsigned long delay;
2224 delay = msecs_to_jiffies(bond->params.miimon);
2226 if (!bond_has_slaves(bond))
2231 should_notify_peers = bond_should_notify_peers(bond);
2233 if (bond_miimon_inspect(bond)) {
2236 /* Race avoidance with bond_close cancel of workqueue */
2237 if (!rtnl_trylock()) {
2239 should_notify_peers = false;
2243 bond_miimon_commit(bond);
2245 rtnl_unlock(); /* might sleep, hold no other locks */
2250 if (bond->params.miimon)
2251 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2253 if (should_notify_peers) {
2254 if (!rtnl_trylock())
2256 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2261 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2263 struct net_device *upper;
2264 struct list_head *iter;
2267 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2271 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2272 if (ip == bond_confirm_addr(upper, 0, ip)) {
2282 /* We go to the (large) trouble of VLAN tagging ARP frames because
2283 * switches in VLAN mode (especially if ports are configured as
2284 * "native" to a VLAN) might not pass non-tagged frames.
2286 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2287 __be32 dest_ip, __be32 src_ip,
2288 struct bond_vlan_tag *tags)
2290 struct sk_buff *skb;
2291 struct bond_vlan_tag *outer_tag = tags;
2293 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2294 arp_op, slave_dev->name, &dest_ip, &src_ip);
2296 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2297 NULL, slave_dev->dev_addr, NULL);
2300 net_err_ratelimited("ARP packet allocation failed\n");
2304 if (!tags || tags->vlan_proto == VLAN_N_VID)
2309 /* Go through all the tags backwards and add them to the packet */
2310 while (tags->vlan_proto != VLAN_N_VID) {
2311 if (!tags->vlan_id) {
2316 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2317 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2318 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2321 net_err_ratelimited("failed to insert inner VLAN tag\n");
2327 /* Set the outer tag */
2328 if (outer_tag->vlan_id) {
2329 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2330 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2331 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2332 outer_tag->vlan_id);
2339 /* Validate the device path between the @start_dev and the @end_dev.
2340 * The path is valid if the @end_dev is reachable through device
2342 * When the path is validated, collect any vlan information in the
2345 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2346 struct net_device *end_dev,
2349 struct bond_vlan_tag *tags;
2350 struct net_device *upper;
2351 struct list_head *iter;
2353 if (start_dev == end_dev) {
2354 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2356 return ERR_PTR(-ENOMEM);
2357 tags[level].vlan_proto = VLAN_N_VID;
2361 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2362 tags = bond_verify_device_path(upper, end_dev, level + 1);
2363 if (IS_ERR_OR_NULL(tags)) {
2368 if (is_vlan_dev(upper)) {
2369 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2370 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2379 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2382 struct bond_vlan_tag *tags;
2383 __be32 *targets = bond->params.arp_targets, addr;
2386 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2387 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2390 /* Find out through which dev should the packet go */
2391 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2394 /* there's no route to target - try to send arp
2395 * probe to generate any traffic (arp_validate=0)
2397 if (bond->params.arp_validate)
2398 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2401 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2406 /* bond device itself */
2407 if (rt->dst.dev == bond->dev)
2411 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2414 if (!IS_ERR_OR_NULL(tags))
2417 /* Not our device - skip */
2418 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2419 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2425 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2427 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2433 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2437 if (!sip || !bond_has_this_ip(bond, tip)) {
2438 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2443 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2445 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2449 slave->last_rx = jiffies;
2450 slave->target_last_arp_rx[i] = jiffies;
2453 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2454 struct slave *slave)
2456 struct arphdr *arp = (struct arphdr *)skb->data;
2457 struct slave *curr_active_slave;
2458 unsigned char *arp_ptr;
2460 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2462 if (!slave_do_arp_validate(bond, slave)) {
2463 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2464 !slave_do_arp_validate_only(bond))
2465 slave->last_rx = jiffies;
2466 return RX_HANDLER_ANOTHER;
2467 } else if (!is_arp) {
2468 return RX_HANDLER_ANOTHER;
2471 alen = arp_hdr_len(bond->dev);
2473 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2476 if (alen > skb_headlen(skb)) {
2477 arp = kmalloc(alen, GFP_ATOMIC);
2480 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2484 if (arp->ar_hln != bond->dev->addr_len ||
2485 skb->pkt_type == PACKET_OTHERHOST ||
2486 skb->pkt_type == PACKET_LOOPBACK ||
2487 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2488 arp->ar_pro != htons(ETH_P_IP) ||
2492 arp_ptr = (unsigned char *)(arp + 1);
2493 arp_ptr += bond->dev->addr_len;
2494 memcpy(&sip, arp_ptr, 4);
2495 arp_ptr += 4 + bond->dev->addr_len;
2496 memcpy(&tip, arp_ptr, 4);
2498 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2499 slave->dev->name, bond_slave_state(slave),
2500 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2503 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2505 /* Backup slaves won't see the ARP reply, but do come through
2506 * here for each ARP probe (so we swap the sip/tip to validate
2507 * the probe). In a "redundant switch, common router" type of
2508 * configuration, the ARP probe will (hopefully) travel from
2509 * the active, through one switch, the router, then the other
2510 * switch before reaching the backup.
2512 * We 'trust' the arp requests if there is an active slave and
2513 * it received valid arp reply(s) after it became active. This
2514 * is done to avoid endless looping when we can't reach the
2515 * arp_ip_target and fool ourselves with our own arp requests.
2518 if (bond_is_active_slave(slave))
2519 bond_validate_arp(bond, slave, sip, tip);
2520 else if (curr_active_slave &&
2521 time_after(slave_last_rx(bond, curr_active_slave),
2522 curr_active_slave->last_link_up))
2523 bond_validate_arp(bond, slave, tip, sip);
2526 if (arp != (struct arphdr *)skb->data)
2528 return RX_HANDLER_ANOTHER;
2531 /* function to verify if we're in the arp_interval timeslice, returns true if
2532 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2533 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2535 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2538 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2540 return time_in_range(jiffies,
2541 last_act - delta_in_ticks,
2542 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2545 /* This function is called regularly to monitor each slave's link
2546 * ensuring that traffic is being sent and received when arp monitoring
2547 * is used in load-balancing mode. if the adapter has been dormant, then an
2548 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2549 * arp monitoring in active backup mode.
2551 static void bond_loadbalance_arp_mon(struct work_struct *work)
2553 struct bonding *bond = container_of(work, struct bonding,
2555 struct slave *slave, *oldcurrent;
2556 struct list_head *iter;
2557 int do_failover = 0, slave_state_changed = 0;
2559 if (!bond_has_slaves(bond))
2564 oldcurrent = rcu_dereference(bond->curr_active_slave);
2565 /* see if any of the previous devices are up now (i.e. they have
2566 * xmt and rcv traffic). the curr_active_slave does not come into
2567 * the picture unless it is null. also, slave->last_link_up is not
2568 * needed here because we send an arp on each slave and give a slave
2569 * as long as it needs to get the tx/rx within the delta.
2570 * TODO: what about up/down delay in arp mode? it wasn't here before
2573 bond_for_each_slave_rcu(bond, slave, iter) {
2574 unsigned long trans_start = dev_trans_start(slave->dev);
2576 if (slave->link != BOND_LINK_UP) {
2577 if (bond_time_in_interval(bond, trans_start, 1) &&
2578 bond_time_in_interval(bond, slave->last_rx, 1)) {
2580 slave->link = BOND_LINK_UP;
2581 slave_state_changed = 1;
2583 /* primary_slave has no meaning in round-robin
2584 * mode. the window of a slave being up and
2585 * curr_active_slave being null after enslaving
2589 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2593 netdev_info(bond->dev, "interface %s is now up\n",
2598 /* slave->link == BOND_LINK_UP */
2600 /* not all switches will respond to an arp request
2601 * when the source ip is 0, so don't take the link down
2602 * if we don't know our ip yet
2604 if (!bond_time_in_interval(bond, trans_start, 2) ||
2605 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2607 slave->link = BOND_LINK_DOWN;
2608 slave_state_changed = 1;
2610 if (slave->link_failure_count < UINT_MAX)
2611 slave->link_failure_count++;
2613 netdev_info(bond->dev, "interface %s is now down\n",
2616 if (slave == oldcurrent)
2621 /* note: if switch is in round-robin mode, all links
2622 * must tx arp to ensure all links rx an arp - otherwise
2623 * links may oscillate or not come up at all; if switch is
2624 * in something like xor mode, there is nothing we can
2625 * do - all replies will be rx'ed on same link causing slaves
2626 * to be unstable during low/no traffic periods
2628 if (bond_slave_is_up(slave))
2629 bond_arp_send_all(bond, slave);
2634 if (do_failover || slave_state_changed) {
2635 if (!rtnl_trylock())
2638 if (slave_state_changed) {
2639 bond_slave_state_change(bond);
2640 if (BOND_MODE(bond) == BOND_MODE_XOR)
2641 bond_update_slave_arr(bond, NULL);
2645 bond_select_active_slave(bond);
2646 unblock_netpoll_tx();
2652 if (bond->params.arp_interval)
2653 queue_delayed_work(bond->wq, &bond->arp_work,
2654 msecs_to_jiffies(bond->params.arp_interval));
2657 /* Called to inspect slaves for active-backup mode ARP monitor link state
2658 * changes. Sets new_link in slaves to specify what action should take
2659 * place for the slave. Returns 0 if no changes are found, >0 if changes
2660 * to link states must be committed.
2662 * Called with rcu_read_lock held.
2664 static int bond_ab_arp_inspect(struct bonding *bond)
2666 unsigned long trans_start, last_rx;
2667 struct list_head *iter;
2668 struct slave *slave;
2671 bond_for_each_slave_rcu(bond, slave, iter) {
2672 slave->new_link = BOND_LINK_NOCHANGE;
2673 last_rx = slave_last_rx(bond, slave);
2675 if (slave->link != BOND_LINK_UP) {
2676 if (bond_time_in_interval(bond, last_rx, 1)) {
2677 slave->new_link = BOND_LINK_UP;
2683 /* Give slaves 2*delta after being enslaved or made
2684 * active. This avoids bouncing, as the last receive
2685 * times need a full ARP monitor cycle to be updated.
2687 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2690 /* Backup slave is down if:
2691 * - No current_arp_slave AND
2692 * - more than 3*delta since last receive AND
2693 * - the bond has an IP address
2695 * Note: a non-null current_arp_slave indicates
2696 * the curr_active_slave went down and we are
2697 * searching for a new one; under this condition
2698 * we only take the curr_active_slave down - this
2699 * gives each slave a chance to tx/rx traffic
2700 * before being taken out
2702 if (!bond_is_active_slave(slave) &&
2703 !rcu_access_pointer(bond->current_arp_slave) &&
2704 !bond_time_in_interval(bond, last_rx, 3)) {
2705 slave->new_link = BOND_LINK_DOWN;
2709 /* Active slave is down if:
2710 * - more than 2*delta since transmitting OR
2711 * - (more than 2*delta since receive AND
2712 * the bond has an IP address)
2714 trans_start = dev_trans_start(slave->dev);
2715 if (bond_is_active_slave(slave) &&
2716 (!bond_time_in_interval(bond, trans_start, 2) ||
2717 !bond_time_in_interval(bond, last_rx, 2))) {
2718 slave->new_link = BOND_LINK_DOWN;
2726 /* Called to commit link state changes noted by inspection step of
2727 * active-backup mode ARP monitor.
2729 * Called with RTNL hold.
2731 static void bond_ab_arp_commit(struct bonding *bond)
2733 unsigned long trans_start;
2734 struct list_head *iter;
2735 struct slave *slave;
2737 bond_for_each_slave(bond, slave, iter) {
2738 switch (slave->new_link) {
2739 case BOND_LINK_NOCHANGE:
2743 trans_start = dev_trans_start(slave->dev);
2744 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2745 (!rtnl_dereference(bond->curr_active_slave) &&
2746 bond_time_in_interval(bond, trans_start, 1))) {
2747 struct slave *current_arp_slave;
2749 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2750 bond_set_slave_link_state(slave, BOND_LINK_UP,
2751 BOND_SLAVE_NOTIFY_NOW);
2752 if (current_arp_slave) {
2753 bond_set_slave_inactive_flags(
2755 BOND_SLAVE_NOTIFY_NOW);
2756 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2759 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2762 if (!rtnl_dereference(bond->curr_active_slave) ||
2763 slave == rtnl_dereference(bond->primary_slave))
2770 case BOND_LINK_DOWN:
2771 if (slave->link_failure_count < UINT_MAX)
2772 slave->link_failure_count++;
2774 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2775 BOND_SLAVE_NOTIFY_NOW);
2776 bond_set_slave_inactive_flags(slave,
2777 BOND_SLAVE_NOTIFY_NOW);
2779 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2782 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2783 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2790 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2791 slave->new_link, slave->dev->name);
2797 bond_select_active_slave(bond);
2798 unblock_netpoll_tx();
2801 bond_set_carrier(bond);
2804 /* Send ARP probes for active-backup mode ARP monitor.
2806 * Called with rcu_read_lock held.
2808 static bool bond_ab_arp_probe(struct bonding *bond)
2810 struct slave *slave, *before = NULL, *new_slave = NULL,
2811 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2812 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2813 struct list_head *iter;
2815 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2817 if (curr_arp_slave && curr_active_slave)
2818 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2819 curr_arp_slave->dev->name,
2820 curr_active_slave->dev->name);
2822 if (curr_active_slave) {
2823 bond_arp_send_all(bond, curr_active_slave);
2824 return should_notify_rtnl;
2827 /* if we don't have a curr_active_slave, search for the next available
2828 * backup slave from the current_arp_slave and make it the candidate
2829 * for becoming the curr_active_slave
2832 if (!curr_arp_slave) {
2833 curr_arp_slave = bond_first_slave_rcu(bond);
2834 if (!curr_arp_slave)
2835 return should_notify_rtnl;
2838 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2840 bond_for_each_slave_rcu(bond, slave, iter) {
2841 if (!found && !before && bond_slave_is_up(slave))
2844 if (found && !new_slave && bond_slave_is_up(slave))
2846 /* if the link state is up at this point, we
2847 * mark it down - this can happen if we have
2848 * simultaneous link failures and
2849 * reselect_active_interface doesn't make this
2850 * one the current slave so it is still marked
2851 * up when it is actually down
2853 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2854 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2855 BOND_SLAVE_NOTIFY_LATER);
2856 if (slave->link_failure_count < UINT_MAX)
2857 slave->link_failure_count++;
2859 bond_set_slave_inactive_flags(slave,
2860 BOND_SLAVE_NOTIFY_LATER);
2862 netdev_info(bond->dev, "backup interface %s is now down\n",
2865 if (slave == curr_arp_slave)
2869 if (!new_slave && before)
2875 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2876 BOND_SLAVE_NOTIFY_LATER);
2877 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2878 bond_arp_send_all(bond, new_slave);
2879 new_slave->last_link_up = jiffies;
2880 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2883 bond_for_each_slave_rcu(bond, slave, iter) {
2884 if (slave->should_notify || slave->should_notify_link) {
2885 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2889 return should_notify_rtnl;
2892 static void bond_activebackup_arp_mon(struct work_struct *work)
2894 struct bonding *bond = container_of(work, struct bonding,
2896 bool should_notify_peers = false;
2897 bool should_notify_rtnl = false;
2900 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2902 if (!bond_has_slaves(bond))
2907 should_notify_peers = bond_should_notify_peers(bond);
2909 if (bond_ab_arp_inspect(bond)) {
2912 /* Race avoidance with bond_close flush of workqueue */
2913 if (!rtnl_trylock()) {
2915 should_notify_peers = false;
2919 bond_ab_arp_commit(bond);
2925 should_notify_rtnl = bond_ab_arp_probe(bond);
2929 if (bond->params.arp_interval)
2930 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2932 if (should_notify_peers || should_notify_rtnl) {
2933 if (!rtnl_trylock())
2936 if (should_notify_peers)
2937 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2939 if (should_notify_rtnl) {
2940 bond_slave_state_notify(bond);
2941 bond_slave_link_notify(bond);
2948 /*-------------------------- netdev event handling --------------------------*/
2950 /* Change device name */
2951 static int bond_event_changename(struct bonding *bond)
2953 bond_remove_proc_entry(bond);
2954 bond_create_proc_entry(bond);
2956 bond_debug_reregister(bond);
2961 static int bond_master_netdev_event(unsigned long event,
2962 struct net_device *bond_dev)
2964 struct bonding *event_bond = netdev_priv(bond_dev);
2967 case NETDEV_CHANGENAME:
2968 return bond_event_changename(event_bond);
2969 case NETDEV_UNREGISTER:
2970 bond_remove_proc_entry(event_bond);
2972 case NETDEV_REGISTER:
2973 bond_create_proc_entry(event_bond);
2975 case NETDEV_NOTIFY_PEERS:
2976 if (event_bond->send_peer_notif)
2977 event_bond->send_peer_notif--;
2986 static int bond_slave_netdev_event(unsigned long event,
2987 struct net_device *slave_dev)
2989 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2990 struct bonding *bond;
2991 struct net_device *bond_dev;
2993 /* A netdev event can be generated while enslaving a device
2994 * before netdev_rx_handler_register is called in which case
2995 * slave will be NULL
2999 bond_dev = slave->bond->dev;
3001 primary = rtnl_dereference(bond->primary_slave);
3004 case NETDEV_UNREGISTER:
3005 if (bond_dev->type != ARPHRD_ETHER)
3006 bond_release_and_destroy(bond_dev, slave_dev);
3008 bond_release(bond_dev, slave_dev);
3012 bond_update_speed_duplex(slave);
3013 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3014 bond_3ad_adapter_speed_duplex_changed(slave);
3017 /* Refresh slave-array if applicable!
3018 * If the setup does not use miimon or arpmon (mode-specific!),
3019 * then these events will not cause the slave-array to be
3020 * refreshed. This will cause xmit to use a slave that is not
3021 * usable. Avoid such situation by refeshing the array at these
3022 * events. If these (miimon/arpmon) parameters are configured
3023 * then array gets refreshed twice and that should be fine!
3025 if (bond_mode_uses_xmit_hash(bond))
3026 bond_update_slave_arr(bond, NULL);
3028 case NETDEV_CHANGEMTU:
3029 /* TODO: Should slaves be allowed to
3030 * independently alter their MTU? For
3031 * an active-backup bond, slaves need
3032 * not be the same type of device, so
3033 * MTUs may vary. For other modes,
3034 * slaves arguably should have the
3035 * same MTUs. To do this, we'd need to
3036 * take over the slave's change_mtu
3037 * function for the duration of their
3041 case NETDEV_CHANGENAME:
3042 /* we don't care if we don't have primary set */
3043 if (!bond_uses_primary(bond) ||
3044 !bond->params.primary[0])
3047 if (slave == primary) {
3048 /* slave's name changed - he's no longer primary */
3049 RCU_INIT_POINTER(bond->primary_slave, NULL);
3050 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3051 /* we have a new primary slave */
3052 rcu_assign_pointer(bond->primary_slave, slave);
3053 } else { /* we didn't change primary - exit */
3057 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3058 primary ? slave_dev->name : "none");
3061 bond_select_active_slave(bond);
3062 unblock_netpoll_tx();
3064 case NETDEV_FEAT_CHANGE:
3065 bond_compute_features(bond);
3067 case NETDEV_RESEND_IGMP:
3068 /* Propagate to master device */
3069 call_netdevice_notifiers(event, slave->bond->dev);
3078 /* bond_netdev_event: handle netdev notifier chain events.
3080 * This function receives events for the netdev chain. The caller (an
3081 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3082 * locks for us to safely manipulate the slave devices (RTNL lock,
3085 static int bond_netdev_event(struct notifier_block *this,
3086 unsigned long event, void *ptr)
3088 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3090 netdev_dbg(event_dev, "event: %lx\n", event);
3092 if (!(event_dev->priv_flags & IFF_BONDING))
3095 if (event_dev->flags & IFF_MASTER) {
3096 netdev_dbg(event_dev, "IFF_MASTER\n");
3097 return bond_master_netdev_event(event, event_dev);
3100 if (event_dev->flags & IFF_SLAVE) {
3101 netdev_dbg(event_dev, "IFF_SLAVE\n");
3102 return bond_slave_netdev_event(event, event_dev);
3108 static struct notifier_block bond_netdev_notifier = {
3109 .notifier_call = bond_netdev_event,
3112 /*---------------------------- Hashing Policies -----------------------------*/
3114 /* L2 hash helper */
3115 static inline u32 bond_eth_hash(struct sk_buff *skb)
3117 struct ethhdr *ep, hdr_tmp;
3119 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3121 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3125 /* Extract the appropriate headers based on bond's xmit policy */
3126 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3127 struct flow_keys *fk)
3129 const struct ipv6hdr *iph6;
3130 const struct iphdr *iph;
3131 int noff, proto = -1;
3133 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3134 return skb_flow_dissect_flow_keys(skb, fk, 0);
3136 fk->ports.ports = 0;
3137 noff = skb_network_offset(skb);
3138 if (skb->protocol == htons(ETH_P_IP)) {
3139 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3142 iph_to_flow_copy_v4addrs(fk, iph);
3143 noff += iph->ihl << 2;
3144 if (!ip_is_fragment(iph))
3145 proto = iph->protocol;
3146 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3147 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3149 iph6 = ipv6_hdr(skb);
3150 iph_to_flow_copy_v6addrs(fk, iph6);
3151 noff += sizeof(*iph6);
3152 proto = iph6->nexthdr;
3156 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3157 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3163 * bond_xmit_hash - generate a hash value based on the xmit policy
3164 * @bond: bonding device
3165 * @skb: buffer to use for headers
3167 * This function will extract the necessary headers from the skb buffer and use
3168 * them to generate a hash based on the xmit_policy set in the bonding device
3170 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3172 struct flow_keys flow;
3175 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3179 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3180 !bond_flow_dissect(bond, skb, &flow))
3181 return bond_eth_hash(skb);
3183 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3184 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3185 hash = bond_eth_hash(skb);
3187 hash = (__force u32)flow.ports.ports;
3188 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3189 (__force u32)flow_get_u32_src(&flow);
3190 hash ^= (hash >> 16);
3191 hash ^= (hash >> 8);
3196 /*-------------------------- Device entry points ----------------------------*/
3198 static void bond_work_init_all(struct bonding *bond)
3200 INIT_DELAYED_WORK(&bond->mcast_work,
3201 bond_resend_igmp_join_requests_delayed);
3202 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3203 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3204 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3205 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3207 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3208 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3209 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3212 static void bond_work_cancel_all(struct bonding *bond)
3214 cancel_delayed_work_sync(&bond->mii_work);
3215 cancel_delayed_work_sync(&bond->arp_work);
3216 cancel_delayed_work_sync(&bond->alb_work);
3217 cancel_delayed_work_sync(&bond->ad_work);
3218 cancel_delayed_work_sync(&bond->mcast_work);
3219 cancel_delayed_work_sync(&bond->slave_arr_work);
3222 static int bond_open(struct net_device *bond_dev)
3224 struct bonding *bond = netdev_priv(bond_dev);
3225 struct list_head *iter;
3226 struct slave *slave;
3228 /* reset slave->backup and slave->inactive */
3229 if (bond_has_slaves(bond)) {
3230 bond_for_each_slave(bond, slave, iter) {
3231 if (bond_uses_primary(bond) &&
3232 slave != rcu_access_pointer(bond->curr_active_slave)) {
3233 bond_set_slave_inactive_flags(slave,
3234 BOND_SLAVE_NOTIFY_NOW);
3235 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3236 bond_set_slave_active_flags(slave,
3237 BOND_SLAVE_NOTIFY_NOW);
3242 bond_work_init_all(bond);
3244 if (bond_is_lb(bond)) {
3245 /* bond_alb_initialize must be called before the timer
3248 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3250 if (bond->params.tlb_dynamic_lb)
3251 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3254 if (bond->params.miimon) /* link check interval, in milliseconds. */
3255 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3257 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3258 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3259 bond->recv_probe = bond_arp_rcv;
3262 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3263 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3264 /* register to receive LACPDUs */
3265 bond->recv_probe = bond_3ad_lacpdu_recv;
3266 bond_3ad_initiate_agg_selection(bond, 1);
3269 if (bond_mode_uses_xmit_hash(bond))
3270 bond_update_slave_arr(bond, NULL);
3275 static int bond_close(struct net_device *bond_dev)
3277 struct bonding *bond = netdev_priv(bond_dev);
3279 bond_work_cancel_all(bond);
3280 bond->send_peer_notif = 0;
3281 if (bond_is_lb(bond))
3282 bond_alb_deinitialize(bond);
3283 bond->recv_probe = NULL;
3288 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3289 struct rtnl_link_stats64 *stats)
3291 struct bonding *bond = netdev_priv(bond_dev);
3292 struct rtnl_link_stats64 temp;
3293 struct list_head *iter;
3294 struct slave *slave;
3296 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3298 bond_for_each_slave(bond, slave, iter) {
3299 const struct rtnl_link_stats64 *sstats =
3300 dev_get_stats(slave->dev, &temp);
3301 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3303 stats->rx_packets += sstats->rx_packets - pstats->rx_packets;
3304 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3305 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3306 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3307 stats->rx_nohandler += sstats->rx_nohandler - pstats->rx_nohandler;
3309 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3310 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3311 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3312 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3314 stats->multicast += sstats->multicast - pstats->multicast;
3315 stats->collisions += sstats->collisions - pstats->collisions;
3317 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3318 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3319 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3320 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3321 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3322 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3324 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3325 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3326 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3327 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3328 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3330 /* save off the slave stats for the next run */
3331 memcpy(pstats, sstats, sizeof(*sstats));
3333 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3338 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3340 struct bonding *bond = netdev_priv(bond_dev);
3341 struct net_device *slave_dev = NULL;
3342 struct ifbond k_binfo;
3343 struct ifbond __user *u_binfo = NULL;
3344 struct ifslave k_sinfo;
3345 struct ifslave __user *u_sinfo = NULL;
3346 struct mii_ioctl_data *mii = NULL;
3347 struct bond_opt_value newval;
3351 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3362 /* We do this again just in case we were called by SIOCGMIIREG
3363 * instead of SIOCGMIIPHY.
3369 if (mii->reg_num == 1) {
3371 if (netif_carrier_ok(bond->dev))
3372 mii->val_out = BMSR_LSTATUS;
3376 case BOND_INFO_QUERY_OLD:
3377 case SIOCBONDINFOQUERY:
3378 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3380 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3383 res = bond_info_query(bond_dev, &k_binfo);
3385 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3389 case BOND_SLAVE_INFO_QUERY_OLD:
3390 case SIOCBONDSLAVEINFOQUERY:
3391 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3393 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3396 res = bond_slave_info_query(bond_dev, &k_sinfo);
3398 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3406 net = dev_net(bond_dev);
3408 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3411 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3413 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3418 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3420 case BOND_ENSLAVE_OLD:
3421 case SIOCBONDENSLAVE:
3422 res = bond_enslave(bond_dev, slave_dev);
3424 case BOND_RELEASE_OLD:
3425 case SIOCBONDRELEASE:
3426 res = bond_release(bond_dev, slave_dev);
3428 case BOND_SETHWADDR_OLD:
3429 case SIOCBONDSETHWADDR:
3430 bond_set_dev_addr(bond_dev, slave_dev);
3433 case BOND_CHANGE_ACTIVE_OLD:
3434 case SIOCBONDCHANGEACTIVE:
3435 bond_opt_initstr(&newval, slave_dev->name);
3436 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3445 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3447 struct bonding *bond = netdev_priv(bond_dev);
3449 if (change & IFF_PROMISC)
3450 bond_set_promiscuity(bond,
3451 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3453 if (change & IFF_ALLMULTI)
3454 bond_set_allmulti(bond,
3455 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3458 static void bond_set_rx_mode(struct net_device *bond_dev)
3460 struct bonding *bond = netdev_priv(bond_dev);
3461 struct list_head *iter;
3462 struct slave *slave;
3465 if (bond_uses_primary(bond)) {
3466 slave = rcu_dereference(bond->curr_active_slave);
3468 dev_uc_sync(slave->dev, bond_dev);
3469 dev_mc_sync(slave->dev, bond_dev);
3472 bond_for_each_slave_rcu(bond, slave, iter) {
3473 dev_uc_sync_multiple(slave->dev, bond_dev);
3474 dev_mc_sync_multiple(slave->dev, bond_dev);
3480 static int bond_neigh_init(struct neighbour *n)
3482 struct bonding *bond = netdev_priv(n->dev);
3483 const struct net_device_ops *slave_ops;
3484 struct neigh_parms parms;
3485 struct slave *slave;
3488 slave = bond_first_slave(bond);
3491 slave_ops = slave->dev->netdev_ops;
3492 if (!slave_ops->ndo_neigh_setup)
3495 parms.neigh_setup = NULL;
3496 parms.neigh_cleanup = NULL;
3497 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3501 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3502 * after the last slave has been detached. Assumes that all slaves
3503 * utilize the same neigh_cleanup (true at this writing as only user
3506 n->parms->neigh_cleanup = parms.neigh_cleanup;
3508 if (!parms.neigh_setup)
3511 return parms.neigh_setup(n);
3514 /* The bonding ndo_neigh_setup is called at init time beofre any
3515 * slave exists. So we must declare proxy setup function which will
3516 * be used at run time to resolve the actual slave neigh param setup.
3518 * It's also called by master devices (such as vlans) to setup their
3519 * underlying devices. In that case - do nothing, we're already set up from
3522 static int bond_neigh_setup(struct net_device *dev,
3523 struct neigh_parms *parms)
3525 /* modify only our neigh_parms */
3526 if (parms->dev == dev)
3527 parms->neigh_setup = bond_neigh_init;
3532 /* Change the MTU of all of a master's slaves to match the master */
3533 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3535 struct bonding *bond = netdev_priv(bond_dev);
3536 struct slave *slave, *rollback_slave;
3537 struct list_head *iter;
3540 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3542 bond_for_each_slave(bond, slave, iter) {
3543 netdev_dbg(bond_dev, "s %p c_m %p\n",
3544 slave, slave->dev->netdev_ops->ndo_change_mtu);
3546 res = dev_set_mtu(slave->dev, new_mtu);
3549 /* If we failed to set the slave's mtu to the new value
3550 * we must abort the operation even in ACTIVE_BACKUP
3551 * mode, because if we allow the backup slaves to have
3552 * different mtu values than the active slave we'll
3553 * need to change their mtu when doing a failover. That
3554 * means changing their mtu from timer context, which
3555 * is probably not a good idea.
3557 netdev_dbg(bond_dev, "err %d %s\n", res,
3563 bond_dev->mtu = new_mtu;
3568 /* unwind from head to the slave that failed */
3569 bond_for_each_slave(bond, rollback_slave, iter) {
3572 if (rollback_slave == slave)
3575 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3577 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3578 tmp_res, rollback_slave->dev->name);
3585 /* Change HW address
3587 * Note that many devices must be down to change the HW address, and
3588 * downing the master releases all slaves. We can make bonds full of
3589 * bonding devices to test this, however.
3591 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3593 struct bonding *bond = netdev_priv(bond_dev);
3594 struct slave *slave, *rollback_slave;
3595 struct sockaddr *sa = addr, tmp_sa;
3596 struct list_head *iter;
3599 if (BOND_MODE(bond) == BOND_MODE_ALB)
3600 return bond_alb_set_mac_address(bond_dev, addr);
3603 netdev_dbg(bond_dev, "bond=%p\n", bond);
3605 /* If fail_over_mac is enabled, do nothing and return success.
3606 * Returning an error causes ifenslave to fail.
3608 if (bond->params.fail_over_mac &&
3609 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3612 if (!is_valid_ether_addr(sa->sa_data))
3613 return -EADDRNOTAVAIL;
3615 bond_for_each_slave(bond, slave, iter) {
3616 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3617 res = dev_set_mac_address(slave->dev, addr);
3619 /* TODO: consider downing the slave
3621 * User should expect communications
3622 * breakage anyway until ARP finish
3625 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3631 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3635 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3636 tmp_sa.sa_family = bond_dev->type;
3638 /* unwind from head to the slave that failed */
3639 bond_for_each_slave(bond, rollback_slave, iter) {
3642 if (rollback_slave == slave)
3645 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3647 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3648 tmp_res, rollback_slave->dev->name);
3656 * bond_xmit_slave_id - transmit skb through slave with slave_id
3657 * @bond: bonding device that is transmitting
3658 * @skb: buffer to transmit
3659 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3661 * This function tries to transmit through slave with slave_id but in case
3662 * it fails, it tries to find the first available slave for transmission.
3663 * The skb is consumed in all cases, thus the function is void.
3665 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3667 struct list_head *iter;
3668 struct slave *slave;
3671 /* Here we start from the slave with slave_id */
3672 bond_for_each_slave_rcu(bond, slave, iter) {
3674 if (bond_slave_can_tx(slave)) {
3675 bond_dev_queue_xmit(bond, skb, slave->dev);
3681 /* Here we start from the first slave up to slave_id */
3683 bond_for_each_slave_rcu(bond, slave, iter) {
3686 if (bond_slave_can_tx(slave)) {
3687 bond_dev_queue_xmit(bond, skb, slave->dev);
3691 /* no slave that can tx has been found */
3692 bond_tx_drop(bond->dev, skb);
3696 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3697 * @bond: bonding device to use
3699 * Based on the value of the bonding device's packets_per_slave parameter
3700 * this function generates a slave id, which is usually used as the next
3701 * slave to transmit through.
3703 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3706 struct reciprocal_value reciprocal_packets_per_slave;
3707 int packets_per_slave = bond->params.packets_per_slave;
3709 switch (packets_per_slave) {
3711 slave_id = prandom_u32();
3714 slave_id = bond->rr_tx_counter;
3717 reciprocal_packets_per_slave =
3718 bond->params.reciprocal_packets_per_slave;
3719 slave_id = reciprocal_divide(bond->rr_tx_counter,
3720 reciprocal_packets_per_slave);
3723 bond->rr_tx_counter++;
3728 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3730 struct bonding *bond = netdev_priv(bond_dev);
3731 struct iphdr *iph = ip_hdr(skb);
3732 struct slave *slave;
3735 /* Start with the curr_active_slave that joined the bond as the
3736 * default for sending IGMP traffic. For failover purposes one
3737 * needs to maintain some consistency for the interface that will
3738 * send the join/membership reports. The curr_active_slave found
3739 * will send all of this type of traffic.
3741 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3742 slave = rcu_dereference(bond->curr_active_slave);
3744 bond_dev_queue_xmit(bond, skb, slave->dev);
3746 bond_xmit_slave_id(bond, skb, 0);
3748 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3750 if (likely(slave_cnt)) {
3751 slave_id = bond_rr_gen_slave_id(bond);
3752 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3754 bond_tx_drop(bond_dev, skb);
3758 return NETDEV_TX_OK;
3761 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3762 * the bond has a usable interface.
3764 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3766 struct bonding *bond = netdev_priv(bond_dev);
3767 struct slave *slave;
3769 slave = rcu_dereference(bond->curr_active_slave);
3771 bond_dev_queue_xmit(bond, skb, slave->dev);
3773 bond_tx_drop(bond_dev, skb);
3775 return NETDEV_TX_OK;
3778 /* Use this to update slave_array when (a) it's not appropriate to update
3779 * slave_array right away (note that update_slave_array() may sleep)
3780 * and / or (b) RTNL is not held.
3782 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3784 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3787 /* Slave array work handler. Holds only RTNL */
3788 static void bond_slave_arr_handler(struct work_struct *work)
3790 struct bonding *bond = container_of(work, struct bonding,
3791 slave_arr_work.work);
3794 if (!rtnl_trylock())
3797 ret = bond_update_slave_arr(bond, NULL);
3800 pr_warn_ratelimited("Failed to update slave array from WT\n");
3806 bond_slave_arr_work_rearm(bond, 1);
3809 /* Build the usable slaves array in control path for modes that use xmit-hash
3810 * to determine the slave interface -
3811 * (a) BOND_MODE_8023AD
3813 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3815 * The caller is expected to hold RTNL only and NO other lock!
3817 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3819 struct slave *slave;
3820 struct list_head *iter;
3821 struct bond_up_slave *new_arr, *old_arr;
3825 #ifdef CONFIG_LOCKDEP
3826 WARN_ON(lockdep_is_held(&bond->mode_lock));
3829 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3833 pr_err("Failed to build slave-array.\n");
3836 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3837 struct ad_info ad_info;
3839 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3840 pr_debug("bond_3ad_get_active_agg_info failed\n");
3841 kfree_rcu(new_arr, rcu);
3842 /* No active aggragator means it's not safe to use
3843 * the previous array.
3845 old_arr = rtnl_dereference(bond->slave_arr);
3847 RCU_INIT_POINTER(bond->slave_arr, NULL);
3848 kfree_rcu(old_arr, rcu);
3852 agg_id = ad_info.aggregator_id;
3854 bond_for_each_slave(bond, slave, iter) {
3855 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3856 struct aggregator *agg;
3858 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3859 if (!agg || agg->aggregator_identifier != agg_id)
3862 if (!bond_slave_can_tx(slave))
3864 if (skipslave == slave)
3866 new_arr->arr[new_arr->count++] = slave;
3869 old_arr = rtnl_dereference(bond->slave_arr);
3870 rcu_assign_pointer(bond->slave_arr, new_arr);
3872 kfree_rcu(old_arr, rcu);
3874 if (ret != 0 && skipslave) {
3877 /* Rare situation where caller has asked to skip a specific
3878 * slave but allocation failed (most likely!). BTW this is
3879 * only possible when the call is initiated from
3880 * __bond_release_one(). In this situation; overwrite the
3881 * skipslave entry in the array with the last entry from the
3882 * array to avoid a situation where the xmit path may choose
3883 * this to-be-skipped slave to send a packet out.
3885 old_arr = rtnl_dereference(bond->slave_arr);
3886 for (idx = 0; idx < old_arr->count; idx++) {
3887 if (skipslave == old_arr->arr[idx]) {
3889 old_arr->arr[old_arr->count-1];
3898 /* Use this Xmit function for 3AD as well as XOR modes. The current
3899 * usable slave array is formed in the control path. The xmit function
3900 * just calculates hash and sends the packet out.
3902 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3904 struct bonding *bond = netdev_priv(dev);
3905 struct slave *slave;
3906 struct bond_up_slave *slaves;
3909 slaves = rcu_dereference(bond->slave_arr);
3910 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3911 if (likely(count)) {
3912 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3913 bond_dev_queue_xmit(bond, skb, slave->dev);
3915 bond_tx_drop(dev, skb);
3918 return NETDEV_TX_OK;
3921 /* in broadcast mode, we send everything to all usable interfaces. */
3922 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3924 struct bonding *bond = netdev_priv(bond_dev);
3925 struct slave *slave = NULL;
3926 struct list_head *iter;
3928 bond_for_each_slave_rcu(bond, slave, iter) {
3929 if (bond_is_last_slave(bond, slave))
3931 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3932 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3935 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3936 bond_dev->name, __func__);
3939 bond_dev_queue_xmit(bond, skb2, slave->dev);
3942 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3943 bond_dev_queue_xmit(bond, skb, slave->dev);
3945 bond_tx_drop(bond_dev, skb);
3947 return NETDEV_TX_OK;
3950 /*------------------------- Device initialization ---------------------------*/
3952 /* Lookup the slave that corresponds to a qid */
3953 static inline int bond_slave_override(struct bonding *bond,
3954 struct sk_buff *skb)
3956 struct slave *slave = NULL;
3957 struct list_head *iter;
3959 if (!skb->queue_mapping)
3962 /* Find out if any slaves have the same mapping as this skb. */
3963 bond_for_each_slave_rcu(bond, slave, iter) {
3964 if (slave->queue_id == skb->queue_mapping) {
3965 if (bond_slave_is_up(slave) &&
3966 slave->link == BOND_LINK_UP) {
3967 bond_dev_queue_xmit(bond, skb, slave->dev);
3970 /* If the slave isn't UP, use default transmit policy. */
3979 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3980 void *accel_priv, select_queue_fallback_t fallback)
3982 /* This helper function exists to help dev_pick_tx get the correct
3983 * destination queue. Using a helper function skips a call to
3984 * skb_tx_hash and will put the skbs in the queue we expect on their
3985 * way down to the bonding driver.
3987 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3989 /* Save the original txq to restore before passing to the driver */
3990 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3992 if (unlikely(txq >= dev->real_num_tx_queues)) {
3994 txq -= dev->real_num_tx_queues;
3995 } while (txq >= dev->real_num_tx_queues);
4000 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4002 struct bonding *bond = netdev_priv(dev);
4004 if (bond_should_override_tx_queue(bond) &&
4005 !bond_slave_override(bond, skb))
4006 return NETDEV_TX_OK;
4008 switch (BOND_MODE(bond)) {
4009 case BOND_MODE_ROUNDROBIN:
4010 return bond_xmit_roundrobin(skb, dev);
4011 case BOND_MODE_ACTIVEBACKUP:
4012 return bond_xmit_activebackup(skb, dev);
4013 case BOND_MODE_8023AD:
4015 return bond_3ad_xor_xmit(skb, dev);
4016 case BOND_MODE_BROADCAST:
4017 return bond_xmit_broadcast(skb, dev);
4019 return bond_alb_xmit(skb, dev);
4021 return bond_tlb_xmit(skb, dev);
4023 /* Should never happen, mode already checked */
4024 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4026 bond_tx_drop(dev, skb);
4027 return NETDEV_TX_OK;
4031 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4033 struct bonding *bond = netdev_priv(dev);
4034 netdev_tx_t ret = NETDEV_TX_OK;
4036 /* If we risk deadlock from transmitting this in the
4037 * netpoll path, tell netpoll to queue the frame for later tx
4039 if (unlikely(is_netpoll_tx_blocked(dev)))
4040 return NETDEV_TX_BUSY;
4043 if (bond_has_slaves(bond))
4044 ret = __bond_start_xmit(skb, dev);
4046 bond_tx_drop(dev, skb);
4052 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4053 struct ethtool_cmd *ecmd)
4055 struct bonding *bond = netdev_priv(bond_dev);
4056 unsigned long speed = 0;
4057 struct list_head *iter;
4058 struct slave *slave;
4060 ecmd->duplex = DUPLEX_UNKNOWN;
4061 ecmd->port = PORT_OTHER;
4063 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4064 * do not need to check mode. Though link speed might not represent
4065 * the true receive or transmit bandwidth (not all modes are symmetric)
4066 * this is an accurate maximum.
4068 bond_for_each_slave(bond, slave, iter) {
4069 if (bond_slave_can_tx(slave)) {
4070 if (slave->speed != SPEED_UNKNOWN)
4071 speed += slave->speed;
4072 if (ecmd->duplex == DUPLEX_UNKNOWN &&
4073 slave->duplex != DUPLEX_UNKNOWN)
4074 ecmd->duplex = slave->duplex;
4077 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4082 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4083 struct ethtool_drvinfo *drvinfo)
4085 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4086 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4087 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4091 static const struct ethtool_ops bond_ethtool_ops = {
4092 .get_drvinfo = bond_ethtool_get_drvinfo,
4093 .get_settings = bond_ethtool_get_settings,
4094 .get_link = ethtool_op_get_link,
4097 static const struct net_device_ops bond_netdev_ops = {
4098 .ndo_init = bond_init,
4099 .ndo_uninit = bond_uninit,
4100 .ndo_open = bond_open,
4101 .ndo_stop = bond_close,
4102 .ndo_start_xmit = bond_start_xmit,
4103 .ndo_select_queue = bond_select_queue,
4104 .ndo_get_stats64 = bond_get_stats,
4105 .ndo_do_ioctl = bond_do_ioctl,
4106 .ndo_change_rx_flags = bond_change_rx_flags,
4107 .ndo_set_rx_mode = bond_set_rx_mode,
4108 .ndo_change_mtu = bond_change_mtu,
4109 .ndo_set_mac_address = bond_set_mac_address,
4110 .ndo_neigh_setup = bond_neigh_setup,
4111 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4112 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4113 #ifdef CONFIG_NET_POLL_CONTROLLER
4114 .ndo_netpoll_setup = bond_netpoll_setup,
4115 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4116 .ndo_poll_controller = bond_poll_controller,
4118 .ndo_add_slave = bond_enslave,
4119 .ndo_del_slave = bond_release,
4120 .ndo_fix_features = bond_fix_features,
4121 .ndo_bridge_setlink = switchdev_port_bridge_setlink,
4122 .ndo_bridge_getlink = switchdev_port_bridge_getlink,
4123 .ndo_bridge_dellink = switchdev_port_bridge_dellink,
4124 .ndo_fdb_add = switchdev_port_fdb_add,
4125 .ndo_fdb_del = switchdev_port_fdb_del,
4126 .ndo_fdb_dump = switchdev_port_fdb_dump,
4127 .ndo_features_check = passthru_features_check,
4130 static const struct device_type bond_type = {
4134 static void bond_destructor(struct net_device *bond_dev)
4136 struct bonding *bond = netdev_priv(bond_dev);
4138 destroy_workqueue(bond->wq);
4139 free_netdev(bond_dev);
4142 void bond_setup(struct net_device *bond_dev)
4144 struct bonding *bond = netdev_priv(bond_dev);
4146 spin_lock_init(&bond->mode_lock);
4147 bond->params = bonding_defaults;
4149 /* Initialize pointers */
4150 bond->dev = bond_dev;
4152 /* Initialize the device entry points */
4153 ether_setup(bond_dev);
4154 bond_dev->netdev_ops = &bond_netdev_ops;
4155 bond_dev->ethtool_ops = &bond_ethtool_ops;
4157 bond_dev->destructor = bond_destructor;
4159 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4161 /* Initialize the device options */
4162 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4163 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4164 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4166 /* don't acquire bond device's netif_tx_lock when transmitting */
4167 bond_dev->features |= NETIF_F_LLTX;
4169 /* By default, we declare the bond to be fully
4170 * VLAN hardware accelerated capable. Special
4171 * care is taken in the various xmit functions
4172 * when there are slaves that are not hw accel
4176 /* Don't allow bond devices to change network namespaces. */
4177 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4179 bond_dev->hw_features = BOND_VLAN_FEATURES |
4180 NETIF_F_HW_VLAN_CTAG_TX |
4181 NETIF_F_HW_VLAN_CTAG_RX |
4182 NETIF_F_HW_VLAN_CTAG_FILTER;
4184 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4185 bond_dev->features |= bond_dev->hw_features;
4188 /* Destroy a bonding device.
4189 * Must be under rtnl_lock when this function is called.
4191 static void bond_uninit(struct net_device *bond_dev)
4193 struct bonding *bond = netdev_priv(bond_dev);
4194 struct list_head *iter;
4195 struct slave *slave;
4196 struct bond_up_slave *arr;
4198 bond_netpoll_cleanup(bond_dev);
4200 /* Release the bonded slaves */
4201 bond_for_each_slave(bond, slave, iter)
4202 __bond_release_one(bond_dev, slave->dev, true);
4203 netdev_info(bond_dev, "Released all slaves\n");
4205 arr = rtnl_dereference(bond->slave_arr);
4207 RCU_INIT_POINTER(bond->slave_arr, NULL);
4208 kfree_rcu(arr, rcu);
4211 list_del(&bond->bond_list);
4213 bond_debug_unregister(bond);
4216 /*------------------------- Module initialization ---------------------------*/
4218 static int bond_check_params(struct bond_params *params)
4220 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4221 struct bond_opt_value newval;
4222 const struct bond_opt_value *valptr;
4223 int arp_all_targets_value;
4224 u16 ad_actor_sys_prio = 0;
4225 u16 ad_user_port_key = 0;
4227 /* Convert string parameters. */
4229 bond_opt_initstr(&newval, mode);
4230 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4232 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4235 bond_mode = valptr->value;
4238 if (xmit_hash_policy) {
4239 if ((bond_mode != BOND_MODE_XOR) &&
4240 (bond_mode != BOND_MODE_8023AD) &&
4241 (bond_mode != BOND_MODE_TLB)) {
4242 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4243 bond_mode_name(bond_mode));
4245 bond_opt_initstr(&newval, xmit_hash_policy);
4246 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4249 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4253 xmit_hashtype = valptr->value;
4258 if (bond_mode != BOND_MODE_8023AD) {
4259 pr_info("lacp_rate param is irrelevant in mode %s\n",
4260 bond_mode_name(bond_mode));
4262 bond_opt_initstr(&newval, lacp_rate);
4263 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4266 pr_err("Error: Invalid lacp rate \"%s\"\n",
4270 lacp_fast = valptr->value;
4275 bond_opt_initstr(&newval, ad_select);
4276 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4279 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4282 params->ad_select = valptr->value;
4283 if (bond_mode != BOND_MODE_8023AD)
4284 pr_warn("ad_select param only affects 802.3ad mode\n");
4286 params->ad_select = BOND_AD_STABLE;
4289 if (max_bonds < 0) {
4290 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4291 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4292 max_bonds = BOND_DEFAULT_MAX_BONDS;
4296 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4302 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4307 if (downdelay < 0) {
4308 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4309 downdelay, INT_MAX);
4313 if ((use_carrier != 0) && (use_carrier != 1)) {
4314 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4319 if (num_peer_notif < 0 || num_peer_notif > 255) {
4320 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4325 /* reset values for 802.3ad/TLB/ALB */
4326 if (!bond_mode_uses_arp(bond_mode)) {
4328 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4329 pr_warn("Forcing miimon to 100msec\n");
4330 miimon = BOND_DEFAULT_MIIMON;
4334 if (tx_queues < 1 || tx_queues > 255) {
4335 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4336 tx_queues, BOND_DEFAULT_TX_QUEUES);
4337 tx_queues = BOND_DEFAULT_TX_QUEUES;
4340 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4341 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4343 all_slaves_active = 0;
4346 if (resend_igmp < 0 || resend_igmp > 255) {
4347 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4348 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4349 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4352 bond_opt_initval(&newval, packets_per_slave);
4353 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4354 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4355 packets_per_slave, USHRT_MAX);
4356 packets_per_slave = 1;
4359 if (bond_mode == BOND_MODE_ALB) {
4360 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",
4365 if (updelay || downdelay) {
4366 /* just warn the user the up/down delay will have
4367 * no effect since miimon is zero...
4369 pr_warn("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",
4370 updelay, downdelay);
4373 /* don't allow arp monitoring */
4375 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4376 miimon, arp_interval);
4380 if ((updelay % miimon) != 0) {
4381 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4382 updelay, miimon, (updelay / miimon) * miimon);
4387 if ((downdelay % miimon) != 0) {
4388 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4390 (downdelay / miimon) * miimon);
4393 downdelay /= miimon;
4396 if (arp_interval < 0) {
4397 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4398 arp_interval, INT_MAX);
4402 for (arp_ip_count = 0, i = 0;
4403 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4406 /* not a complete check, but good enough to catch mistakes */
4407 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4408 !bond_is_ip_target_ok(ip)) {
4409 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4413 if (bond_get_targets_ip(arp_target, ip) == -1)
4414 arp_target[arp_ip_count++] = ip;
4416 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4421 if (arp_interval && !arp_ip_count) {
4422 /* don't allow arping if no arp_ip_target given... */
4423 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4429 if (!arp_interval) {
4430 pr_err("arp_validate requires arp_interval\n");
4434 bond_opt_initstr(&newval, arp_validate);
4435 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4438 pr_err("Error: invalid arp_validate \"%s\"\n",
4442 arp_validate_value = valptr->value;
4444 arp_validate_value = 0;
4447 arp_all_targets_value = 0;
4448 if (arp_all_targets) {
4449 bond_opt_initstr(&newval, arp_all_targets);
4450 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4453 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4455 arp_all_targets_value = 0;
4457 arp_all_targets_value = valptr->value;
4462 pr_info("MII link monitoring set to %d ms\n", miimon);
4463 } else if (arp_interval) {
4464 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4465 arp_validate_value);
4466 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4467 arp_interval, valptr->string, arp_ip_count);
4469 for (i = 0; i < arp_ip_count; i++)
4470 pr_cont(" %s", arp_ip_target[i]);
4474 } else if (max_bonds) {
4475 /* miimon and arp_interval not set, we need one so things
4476 * work as expected, see bonding.txt for details
4478 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");
4481 if (primary && !bond_mode_uses_primary(bond_mode)) {
4482 /* currently, using a primary only makes sense
4483 * in active backup, TLB or ALB modes
4485 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4486 primary, bond_mode_name(bond_mode));
4490 if (primary && primary_reselect) {
4491 bond_opt_initstr(&newval, primary_reselect);
4492 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4495 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4499 primary_reselect_value = valptr->value;
4501 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4504 if (fail_over_mac) {
4505 bond_opt_initstr(&newval, fail_over_mac);
4506 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4509 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4513 fail_over_mac_value = valptr->value;
4514 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4515 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4517 fail_over_mac_value = BOND_FOM_NONE;
4520 bond_opt_initstr(&newval, "default");
4521 valptr = bond_opt_parse(
4522 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4525 pr_err("Error: No ad_actor_sys_prio default value");
4528 ad_actor_sys_prio = valptr->value;
4530 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4533 pr_err("Error: No ad_user_port_key default value");
4536 ad_user_port_key = valptr->value;
4538 if (lp_interval == 0) {
4539 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4540 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4541 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4544 /* fill params struct with the proper values */
4545 params->mode = bond_mode;
4546 params->xmit_policy = xmit_hashtype;
4547 params->miimon = miimon;
4548 params->num_peer_notif = num_peer_notif;
4549 params->arp_interval = arp_interval;
4550 params->arp_validate = arp_validate_value;
4551 params->arp_all_targets = arp_all_targets_value;
4552 params->updelay = updelay;
4553 params->downdelay = downdelay;
4554 params->use_carrier = use_carrier;
4555 params->lacp_fast = lacp_fast;
4556 params->primary[0] = 0;
4557 params->primary_reselect = primary_reselect_value;
4558 params->fail_over_mac = fail_over_mac_value;
4559 params->tx_queues = tx_queues;
4560 params->all_slaves_active = all_slaves_active;
4561 params->resend_igmp = resend_igmp;
4562 params->min_links = min_links;
4563 params->lp_interval = lp_interval;
4564 params->packets_per_slave = packets_per_slave;
4565 params->tlb_dynamic_lb = 1; /* Default value */
4566 params->ad_actor_sys_prio = ad_actor_sys_prio;
4567 eth_zero_addr(params->ad_actor_system);
4568 params->ad_user_port_key = ad_user_port_key;
4569 if (packets_per_slave > 0) {
4570 params->reciprocal_packets_per_slave =
4571 reciprocal_value(packets_per_slave);
4573 /* reciprocal_packets_per_slave is unused if
4574 * packets_per_slave is 0 or 1, just initialize it
4576 params->reciprocal_packets_per_slave =
4577 (struct reciprocal_value) { 0 };
4581 strncpy(params->primary, primary, IFNAMSIZ);
4582 params->primary[IFNAMSIZ - 1] = 0;
4585 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4590 static struct lock_class_key bonding_netdev_xmit_lock_key;
4591 static struct lock_class_key bonding_netdev_addr_lock_key;
4592 static struct lock_class_key bonding_tx_busylock_key;
4594 static void bond_set_lockdep_class_one(struct net_device *dev,
4595 struct netdev_queue *txq,
4598 lockdep_set_class(&txq->_xmit_lock,
4599 &bonding_netdev_xmit_lock_key);
4602 static void bond_set_lockdep_class(struct net_device *dev)
4604 lockdep_set_class(&dev->addr_list_lock,
4605 &bonding_netdev_addr_lock_key);
4606 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4607 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4610 /* Called from registration process */
4611 static int bond_init(struct net_device *bond_dev)
4613 struct bonding *bond = netdev_priv(bond_dev);
4614 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4616 netdev_dbg(bond_dev, "Begin bond_init\n");
4618 bond->wq = create_singlethread_workqueue(bond_dev->name);
4622 bond_set_lockdep_class(bond_dev);
4624 list_add_tail(&bond->bond_list, &bn->dev_list);
4626 bond_prepare_sysfs_group(bond);
4628 bond_debug_register(bond);
4630 /* Ensure valid dev_addr */
4631 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4632 bond_dev->addr_assign_type == NET_ADDR_PERM)
4633 eth_hw_addr_random(bond_dev);
4638 unsigned int bond_get_num_tx_queues(void)
4643 /* Create a new bond based on the specified name and bonding parameters.
4644 * If name is NULL, obtain a suitable "bond%d" name for us.
4645 * Caller must NOT hold rtnl_lock; we need to release it here before we
4646 * set up our sysfs entries.
4648 int bond_create(struct net *net, const char *name)
4650 struct net_device *bond_dev;
4651 struct bonding *bond;
4652 struct alb_bond_info *bond_info;
4657 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4658 name ? name : "bond%d", NET_NAME_UNKNOWN,
4659 bond_setup, tx_queues);
4661 pr_err("%s: eek! can't alloc netdev!\n", name);
4667 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4668 * It is set to 0 by default which is wrong.
4670 bond = netdev_priv(bond_dev);
4671 bond_info = &(BOND_ALB_INFO(bond));
4672 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4674 dev_net_set(bond_dev, net);
4675 bond_dev->rtnl_link_ops = &bond_link_ops;
4677 res = register_netdevice(bond_dev);
4679 netif_carrier_off(bond_dev);
4683 bond_destructor(bond_dev);
4687 static int __net_init bond_net_init(struct net *net)
4689 struct bond_net *bn = net_generic(net, bond_net_id);
4692 INIT_LIST_HEAD(&bn->dev_list);
4694 bond_create_proc_dir(bn);
4695 bond_create_sysfs(bn);
4700 static void __net_exit bond_net_exit(struct net *net)
4702 struct bond_net *bn = net_generic(net, bond_net_id);
4703 struct bonding *bond, *tmp_bond;
4706 bond_destroy_sysfs(bn);
4708 /* Kill off any bonds created after unregistering bond rtnl ops */
4710 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4711 unregister_netdevice_queue(bond->dev, &list);
4712 unregister_netdevice_many(&list);
4715 bond_destroy_proc_dir(bn);
4718 static struct pernet_operations bond_net_ops = {
4719 .init = bond_net_init,
4720 .exit = bond_net_exit,
4722 .size = sizeof(struct bond_net),
4725 static int __init bonding_init(void)
4730 pr_info("%s", bond_version);
4732 res = bond_check_params(&bonding_defaults);
4736 res = register_pernet_subsys(&bond_net_ops);
4740 res = bond_netlink_init();
4744 bond_create_debugfs();
4746 for (i = 0; i < max_bonds; i++) {
4747 res = bond_create(&init_net, NULL);
4752 register_netdevice_notifier(&bond_netdev_notifier);
4756 bond_destroy_debugfs();
4757 bond_netlink_fini();
4759 unregister_pernet_subsys(&bond_net_ops);
4764 static void __exit bonding_exit(void)
4766 unregister_netdevice_notifier(&bond_netdev_notifier);
4768 bond_destroy_debugfs();
4770 bond_netlink_fini();
4771 unregister_pernet_subsys(&bond_net_ops);
4773 #ifdef CONFIG_NET_POLL_CONTROLLER
4774 /* Make sure we don't have an imbalance on our netpoll blocking */
4775 WARN_ON(atomic_read(&netpoll_block_tx));
4779 module_init(bonding_init);
4780 module_exit(bonding_exit);
4781 MODULE_LICENSE("GPL");
4782 MODULE_VERSION(DRV_VERSION);
4783 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4784 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");