blob: a2e25de98bdeaf21a89dcda794923c6017fadf68 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
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.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
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.
31 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 */
33
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/kernel.h>
35#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/types.h>
37#include <linux/fcntl.h>
38#include <linux/interrupt.h>
39#include <linux/ptrace.h>
40#include <linux/ioport.h>
41#include <linux/in.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040042#include <net/ip.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/ip.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040044#include <linux/tcp.h>
45#include <linux/udp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#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>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000054#include <linux/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <asm/dma.h>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000056#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <linux/errno.h>
58#include <linux/netdevice.h>
59#include <linux/inetdevice.h>
Jay Vosburgha816c7c2007-02-28 17:03:37 -080060#include <linux/igmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#include <linux/etherdevice.h>
62#include <linux/skbuff.h>
63#include <net/sock.h>
64#include <linux/rtnetlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#include <linux/smp.h>
66#include <linux/if_ether.h>
67#include <net/arp.h>
68#include <linux/mii.h>
69#include <linux/ethtool.h>
70#include <linux/if_vlan.h>
71#include <linux/if_bonding.h>
David Sterbab63bb732007-12-06 23:40:33 -080072#include <linux/jiffies.h>
Neil Hormane843fa52010-10-13 16:01:50 +000073#include <linux/preempt.h>
Jay Vosburghc3ade5c2005-06-26 17:52:20 -040074#include <net/route.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020075#include <net/net_namespace.h>
Eric W. Biedermanec87fd32009-10-29 14:18:26 +000076#include <net/netns/generic.h>
Eric Dumazet5ee31c682012-06-12 06:03:51 +000077#include <net/pkt_sched.h>
nikolay@redhat.com278b2082013-08-01 16:54:51 +020078#include <linux/rculist.h>
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +020079#include <net/flow_keys.h>
Roopa Prabhuc158cba2015-01-29 22:40:16 -080080#include <net/switchdev.h>
David S. Miller1ef80192014-11-10 13:27:49 -050081#include <net/bonding.h>
82#include <net/bond_3ad.h>
83#include <net/bond_alb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Matan Barak73b5a6f22015-04-26 15:55:57 +030085#include "bonding_priv.h"
86
Linus Torvalds1da177e2005-04-16 15:20:36 -070087/*---------------------------- Module parameters ----------------------------*/
88
89/* monitor all links that often (in milliseconds). <=0 disables monitoring */
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
91static int max_bonds = BOND_DEFAULT_MAX_BONDS;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +000092static int tx_queues = BOND_DEFAULT_TX_QUEUES;
Ben Hutchingsad246c92011-04-26 15:25:52 +000093static int num_peer_notif = 1;
Nikolay Aleksandrovb98d9c62014-01-22 14:53:31 +010094static int miimon;
Stephen Hemminger3d632c32009-06-12 19:02:48 +000095static int updelay;
96static int downdelay;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097static int use_carrier = 1;
Stephen Hemminger3d632c32009-06-12 19:02:48 +000098static char *mode;
99static char *primary;
Jiri Pirkoa5499522009-09-25 03:28:09 +0000100static char *primary_reselect;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000101static char *lacp_rate;
stephen hemminger655f8912011-06-22 09:54:39 +0000102static int min_links;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000103static char *ad_select;
104static char *xmit_hash_policy;
Nikolay Aleksandrov7bdb04e2014-01-22 14:53:23 +0100105static int arp_interval;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000106static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107static char *arp_validate;
Veaceslav Falico8599b522013-06-24 11:49:34 +0200108static char *arp_all_targets;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000109static char *fail_over_mac;
dingtianhongb07ea072013-07-23 15:25:47 +0800110static int all_slaves_active;
Stephen Hemmingerd2991f72009-06-12 19:02:44 +0000111static struct bond_params bonding_defaults;
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000112static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +0100113static int packets_per_slave = 1;
dingtianhong3a7129e2013-12-21 14:40:12 +0800114static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116module_param(max_bonds, int, 0);
117MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
Andy Gospodarekbb1d9122010-06-02 08:40:18 +0000118module_param(tx_queues, int, 0);
119MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
Ben Hutchingsad246c92011-04-26 15:25:52 +0000120module_param_named(num_grat_arp, num_peer_notif, int, 0644);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000121MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
Ben Hutchingsad246c92011-04-26 15:25:52 +0000123module_param_named(num_unsol_na, num_peer_notif, int, 0644);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000124MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126module_param(miimon, int, 0);
127MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128module_param(updelay, int, 0);
129MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130module_param(downdelay, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800131MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 "in milliseconds");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133module_param(use_carrier, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800134MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136module_param(mode, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000137MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
Mitch Williams2ac47662005-11-09 10:35:03 -0800138 "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");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141module_param(primary, charp, 0);
142MODULE_PARM_DESC(primary, "Primary network device to use");
Jiri Pirkoa5499522009-09-25 03:28:09 +0000143module_param(primary_reselect, charp, 0);
144MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "once it comes up; "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
148 "better, "
149 "2 for only on active slave "
150 "failure");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151module_param(lacp_rate, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000152MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
Jay Vosburghfd989c82008-11-04 17:51:16 -0800154module_param(ad_select, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000155MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
157 "2 for count");
stephen hemminger655f8912011-06-22 09:54:39 +0000158module_param(min_links, int, 0);
159MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400161module_param(xmit_hash_policy, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000162MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +0200164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166module_param(arp_interval, int, 0);
167MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168module_param_array(arp_ip_target, charp, NULL, 0);
169MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700170module_param(arp_validate, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000171MODULE_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");
Veaceslav Falico8599b522013-06-24 11:49:34 +0200174module_param(arp_all_targets, charp, 0);
175MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700176module_param(fail_over_mac, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000177MODULE_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");
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000180module_param(all_slaves_active, int, 0);
Masanari Iida37b70212014-09-09 18:07:55 +0900181MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
Andy Gospodarek90e62472011-05-25 04:41:59 +0000182 "by setting active flag for all slaves; "
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000183 "0 for never (default), 1 for always.");
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000184module_param(resend_igmp, int, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000185MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 "link failure");
Nikolay Aleksandrov73958322013-11-05 13:51:41 +0100187module_param(packets_per_slave, int, 0);
188MODULE_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.");
dingtianhong3a7129e2013-12-21 14:40:12 +0800191module_param(lp_interval, uint, 0);
192MODULE_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.");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196/*----------------------------- Global variables ----------------------------*/
197
Neil Hormane843fa52010-10-13 16:01:50 +0000198#ifdef CONFIG_NET_POLL_CONTROLLER
Neil Hormanfb4fa762010-12-06 09:05:50 +0000199atomic_t netpoll_block_tx = ATOMIC_INIT(0);
Neil Hormane843fa52010-10-13 16:01:50 +0000200#endif
201
Eric Dumazetf99189b2009-11-17 10:42:49 +0000202int bond_net_id __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000204static __be32 arp_target[BOND_MAX_ARP_TARGETS];
205static int arp_ip_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206static int bond_mode = BOND_MODE_ROUNDROBIN;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000207static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
208static int lacp_fast;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210/*-------------------------- Forward declarations ---------------------------*/
211
Stephen Hemminger181470f2009-06-12 19:02:52 +0000212static int bond_init(struct net_device *bond_dev);
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +0000213static void bond_uninit(struct net_device *bond_dev);
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -0400214static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
215 struct rtnl_link_stats64 *stats);
Mahesh Bandewaree637712014-10-04 17:45:01 -0700216static void bond_slave_arr_handler(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218/*---------------------------- General routines -----------------------------*/
219
Amerigo Wangbd33acc2011-03-06 21:58:46 +0000220const char *bond_mode_name(int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800222 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)",
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000227 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800228 [BOND_MODE_TLB] = "transmit load balancing",
229 [BOND_MODE_ALB] = "adaptive load balancing",
230 };
231
Wang Sheng-Huif5280942013-07-24 14:53:26 +0800232 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 return "unknown";
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800234
235 return names[mode];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236}
237
238/*---------------------------------- VLAN -----------------------------------*/
239
240/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 * bond_dev_queue_xmit - Prepare skb for xmit.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000242 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 * @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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 */
dingtianhongd316ded2014-01-02 09:13:09 +0800247void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000248 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249{
Ben Hutchings83874512010-12-13 08:19:28 +0000250 skb->dev = slave_dev;
Neil Horman374eeb52011-06-03 10:35:52 +0000251
Eric Dumazet5ee31c682012-06-12 06:03:51 +0000252 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
Jiri Pirkodf4ab5b2012-07-20 02:28:49 +0000253 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
254 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
Neil Horman374eeb52011-06-03 10:35:52 +0000255
Amerigo Wange15c3c222012-08-10 01:24:45 +0000256 if (unlikely(netpoll_tx_running(bond->dev)))
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000257 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
Amerigo Wang080e4132011-02-17 23:43:33 +0000258 else
WANG Congf6dc31a2010-05-06 00:48:51 -0700259 dev_queue_xmit(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260}
261
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200262/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
Jiri Pirkocc0e4072011-07-20 04:54:46 +0000263 * We don't protect the slave list iteration with a lock because:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 * 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.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000268 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 * 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.
276*/
277
278/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 * 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
282 */
Patrick McHardy80d5c362013-04-19 02:04:28 +0000283static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
284 __be16 proto, u16 vid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285{
Wang Chen454d7c92008-11-12 23:37:49 -0800286 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico81f23b12013-09-25 09:20:13 +0200287 struct slave *slave, *rollback_slave;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200288 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200289 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200291 bond_for_each_slave(bond, slave, iter) {
Patrick McHardy80d5c362013-04-19 02:04:28 +0000292 res = vlan_vid_add(slave->dev, proto, vid);
Jiri Pirko87002b02011-12-08 04:11:17 +0000293 if (res)
294 goto unwind;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 }
296
Jiri Pirko8e586132011-12-08 19:52:37 -0500297 return 0;
Jiri Pirko87002b02011-12-08 04:11:17 +0000298
299unwind:
Veaceslav Falico81f23b12013-09-25 09:20:13 +0200300 /* unwind to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200301 bond_for_each_slave(bond, rollback_slave, iter) {
Veaceslav Falico81f23b12013-09-25 09:20:13 +0200302 if (rollback_slave == slave)
303 break;
304
305 vlan_vid_del(rollback_slave->dev, proto, vid);
306 }
Jiri Pirko87002b02011-12-08 04:11:17 +0000307
308 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
311/**
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
315 */
Patrick McHardy80d5c362013-04-19 02:04:28 +0000316static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
317 __be16 proto, u16 vid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
Wang Chen454d7c92008-11-12 23:37:49 -0800319 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200320 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200323 bond_for_each_slave(bond, slave, iter)
Patrick McHardy80d5c362013-04-19 02:04:28 +0000324 vlan_vid_del(slave->dev, proto, vid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
Veaceslav Falicoe868b0c2013-08-28 23:25:15 +0200326 if (bond_is_lb(bond))
327 bond_alb_clear_vlan(bond, vid);
Jiri Pirko8e586132011-12-08 19:52:37 -0500328
329 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
331
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332/*------------------------------- Link status -------------------------------*/
333
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200334/* Set the carrier state for the master according to the state of its
Jay Vosburghff59c452006-03-27 13:27:43 -0800335 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
336 * do special 802.3ad magic.
337 *
338 * Returns zero if carrier state does not change, nonzero if it does.
339 */
Jonathan Toppins2477bc92015-01-26 01:16:57 -0500340int bond_set_carrier(struct bonding *bond)
Jay Vosburghff59c452006-03-27 13:27:43 -0800341{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200342 struct list_head *iter;
Jay Vosburghff59c452006-03-27 13:27:43 -0800343 struct slave *slave;
Jay Vosburghff59c452006-03-27 13:27:43 -0800344
Veaceslav Falico0965a1f2013-09-25 09:20:21 +0200345 if (!bond_has_slaves(bond))
Jay Vosburghff59c452006-03-27 13:27:43 -0800346 goto down;
347
Veaceslav Falico01844092014-05-15 21:39:55 +0200348 if (BOND_MODE(bond) == BOND_MODE_8023AD)
Jay Vosburghff59c452006-03-27 13:27:43 -0800349 return bond_3ad_set_carrier(bond);
350
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200351 bond_for_each_slave(bond, slave, iter) {
Jay Vosburghff59c452006-03-27 13:27:43 -0800352 if (slave->link == BOND_LINK_UP) {
353 if (!netif_carrier_ok(bond->dev)) {
354 netif_carrier_on(bond->dev);
355 return 1;
356 }
357 return 0;
358 }
359 }
360
361down:
362 if (netif_carrier_ok(bond->dev)) {
363 netif_carrier_off(bond->dev);
364 return 1;
365 }
366 return 0;
367}
368
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200369/* Get link speed and duplex from the slave's base driver
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 * using ethtool. If for some reason the call fails or the
Weiping Pan98f41f62011-10-31 17:20:48 +0000371 * values are invalid, set speed and duplex to -1,
Rick Jones13b95fb2012-04-26 11:20:30 +0000372 * and return.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 */
Rick Jones13b95fb2012-04-26 11:20:30 +0000374static void bond_update_speed_duplex(struct slave *slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
376 struct net_device *slave_dev = slave->dev;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000377 struct ethtool_cmd ecmd;
David Decotigny5d305302011-04-13 15:22:31 +0000378 u32 slave_speed;
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700379 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Dan Carpenter589665f2011-11-04 08:21:38 +0000381 slave->speed = SPEED_UNKNOWN;
382 slave->duplex = DUPLEX_UNKNOWN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000384 res = __ethtool_get_settings(slave_dev, &ecmd);
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700385 if (res < 0)
Rick Jones13b95fb2012-04-26 11:20:30 +0000386 return;
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700387
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000388 slave_speed = ethtool_cmd_speed(&ecmd);
Jiri Pirko6f92c662011-06-01 10:36:33 +0000389 if (slave_speed == 0 || slave_speed == ((__u32) -1))
Rick Jones13b95fb2012-04-26 11:20:30 +0000390 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000392 switch (ecmd.duplex) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 case DUPLEX_FULL:
394 case DUPLEX_HALF:
395 break;
396 default:
Rick Jones13b95fb2012-04-26 11:20:30 +0000397 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 }
399
David Decotigny5d305302011-04-13 15:22:31 +0000400 slave->speed = slave_speed;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000401 slave->duplex = ecmd.duplex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Rick Jones13b95fb2012-04-26 11:20:30 +0000403 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404}
405
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -0800406const char *bond_slave_link_status(s8 link)
407{
408 switch (link) {
409 case BOND_LINK_UP:
410 return "up";
411 case BOND_LINK_FAIL:
412 return "going down";
413 case BOND_LINK_DOWN:
414 return "down";
415 case BOND_LINK_BACK:
416 return "going back";
417 default:
418 return "unknown";
419 }
420}
421
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200422/* if <dev> supports MII link status reporting, check its link status.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 *
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000425 * depending upon the setting of the use_carrier parameter.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 *
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
429 * down.
430 *
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.
436 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000437static int bond_check_dev_link(struct bonding *bond,
438 struct net_device *slave_dev, int reporting)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Jiri Bohacd9d52832009-10-28 22:23:54 -0700441 int (*ioctl)(struct net_device *, struct ifreq *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 struct ifreq ifr;
443 struct mii_ioctl_data *mii;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Petri Gynther6c988852009-08-28 12:05:15 +0000445 if (!reporting && !netif_running(slave_dev))
446 return 0;
447
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800448 if (bond->params.use_carrier)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Jiri Pirko29112f42009-04-24 01:58:23 +0000451 /* Try to get link status using Ethtool first. */
Ben Hutchingsc772dde2012-12-07 06:15:32 +0000452 if (slave_dev->ethtool_ops->get_link)
453 return slave_dev->ethtool_ops->get_link(slave_dev) ?
454 BMSR_LSTATUS : 0;
Jiri Pirko29112f42009-04-24 01:58:23 +0000455
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000456 /* Ethtool can't be used, fallback to MII ioctls. */
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800457 ioctl = slave_ops->ndo_do_ioctl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 if (ioctl) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200459 /* 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
463 * member.
464 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200466 /* We cannot assume that SIOCGMIIPHY will also read a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 * register; not all network drivers (e.g., e100)
468 * support that.
469 */
470
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
473 mii = if_mii(&ifr);
474 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
475 mii->reg_num = MII_BMSR;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000476 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
477 return mii->val_out & BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
479 }
480
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200481 /* If reporting, report that either there's no dev->do_ioctl,
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700482 * or both SIOCGMIIREG and get_link failed (meaning that we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 * cannot report link status). If not reporting, pretend
484 * we're ok.
485 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000486 return reporting ? -1 : BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487}
488
489/*----------------------------- Multicast list ------------------------------*/
490
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200491/* Push the promiscuity flag down to appropriate slaves */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700492static int bond_set_promiscuity(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200494 struct list_head *iter;
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700495 int err = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200496
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200497 if (bond_uses_primary(bond)) {
Veaceslav Falico14056e72014-07-16 18:32:01 +0200498 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
Eric Dumazet4740d632014-07-15 06:56:55 -0700499
Eric Dumazet4740d632014-07-15 06:56:55 -0700500 if (curr_active)
501 err = dev_set_promiscuity(curr_active->dev, inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 } else {
503 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200504
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200505 bond_for_each_slave(bond, slave, iter) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700506 err = dev_set_promiscuity(slave->dev, inc);
507 if (err)
508 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 }
510 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700511 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512}
513
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200514/* Push the allmulti flag down to all slaves */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700515static int bond_set_allmulti(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200517 struct list_head *iter;
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700518 int err = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200519
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200520 if (bond_uses_primary(bond)) {
Veaceslav Falico14056e72014-07-16 18:32:01 +0200521 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
Eric Dumazet4740d632014-07-15 06:56:55 -0700522
Eric Dumazet4740d632014-07-15 06:56:55 -0700523 if (curr_active)
524 err = dev_set_allmulti(curr_active->dev, inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 } else {
526 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200527
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200528 bond_for_each_slave(bond, slave, iter) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700529 err = dev_set_allmulti(slave->dev, inc);
530 if (err)
531 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 }
533 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700534 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535}
536
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200537/* Retrieve the list of registered multicast addresses for the bonding
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800538 * device and retransmit an IGMP JOIN request to the current active
539 * slave.
540 */
stephen hemminger379b7382010-10-15 11:02:56 +0000541static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000542{
543 struct bonding *bond = container_of(work, struct bonding,
Flavio Leitner94265cf2011-05-25 08:38:58 +0000544 mcast_work.work);
Veaceslav Falicoad999ee2013-03-26 04:10:02 +0000545
dingtianhongf2369102013-12-13 10:20:26 +0800546 if (!rtnl_trylock()) {
547 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
548 return;
549 }
550 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
551
552 if (bond->igmp_retrans > 1) {
553 bond->igmp_retrans--;
554 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
555 }
556 rtnl_unlock();
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800557}
558
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200559/* Flush bond's hardware addresses from slave */
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000560static void bond_hw_addr_flush(struct net_device *bond_dev,
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000561 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
Wang Chen454d7c92008-11-12 23:37:49 -0800563 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000565 dev_uc_unsync(slave_dev, bond_dev);
566 dev_mc_unsync(slave_dev, bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Veaceslav Falico01844092014-05-15 21:39:55 +0200568 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 /* del lacpdu mc addr from mc list */
570 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
571
Jiri Pirko22bedad32010-04-01 21:22:57 +0000572 dev_mc_del(slave_dev, lacpdu_multicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 }
574}
575
576/*--------------------------- Active slave change ---------------------------*/
577
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000578/* Update the hardware address list and promisc/allmulti for the new and
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200579 * 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
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000581 * this function to swap these settings during a failover.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 */
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000583static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
584 struct slave *old_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (old_active) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000587 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 dev_set_promiscuity(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000590 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 dev_set_allmulti(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000593 bond_hw_addr_flush(bond->dev, old_active->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595
596 if (new_active) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700597 /* FIXME: Signal errors upstream. */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000598 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 dev_set_promiscuity(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000601 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 dev_set_allmulti(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
nikolay@redhat.comd632ce92013-04-18 07:33:38 +0000604 netif_addr_lock_bh(bond->dev);
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000605 dev_uc_sync(new_active->dev, bond->dev);
606 dev_mc_sync(new_active->dev, bond->dev);
nikolay@redhat.comd632ce92013-04-18 07:33:38 +0000607 netif_addr_unlock_bh(bond->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 }
609}
610
nikolay@redhat.comae0d6752013-06-26 17:13:39 +0200611/**
612 * bond_set_dev_addr - clone slave's address to bond
613 * @bond_dev: bond net device
614 * @slave_dev: slave net device
615 *
616 * Should be called with RTNL held.
617 */
618static void bond_set_dev_addr(struct net_device *bond_dev,
619 struct net_device *slave_dev)
620{
Veaceslav Falico76444f52014-07-15 19:35:58 +0200621 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
622 bond_dev, slave_dev, slave_dev->addr_len);
nikolay@redhat.comae0d6752013-06-26 17:13:39 +0200623 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);
626}
627
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200628/* bond_do_fail_over_mac
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700629 *
630 * Perform special MAC address swapping for fail_over_mac settings
631 *
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200632 * Called with RTNL
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700633 */
634static void bond_do_fail_over_mac(struct bonding *bond,
635 struct slave *new_active,
636 struct slave *old_active)
637{
638 u8 tmp_mac[ETH_ALEN];
639 struct sockaddr saddr;
640 int rv;
641
642 switch (bond->params.fail_over_mac) {
643 case BOND_FOM_ACTIVE:
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200644 if (new_active)
nikolay@redhat.comae0d6752013-06-26 17:13:39 +0200645 bond_set_dev_addr(bond->dev, new_active->dev);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700646 break;
647 case BOND_FOM_FOLLOW:
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200648 /* if new_active && old_active, swap them
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700649 * if just old_active, do nothing (going to no active slave)
650 * if just new_active, set new_active to bond's MAC
651 */
652 if (!new_active)
653 return;
654
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700655 if (old_active) {
Joe Perchesada0f862014-02-15 16:02:17 -0800656 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
Joe Perches2a7c1832014-02-18 09:42:45 -0800657 ether_addr_copy(saddr.sa_data,
658 old_active->dev->dev_addr);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700659 saddr.sa_family = new_active->dev->type;
660 } else {
Joe Perchesada0f862014-02-15 16:02:17 -0800661 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700662 saddr.sa_family = bond->dev->type;
663 }
664
665 rv = dev_set_mac_address(new_active->dev, &saddr);
666 if (rv) {
Veaceslav Falico76444f52014-07-15 19:35:58 +0200667 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
668 -rv, new_active->dev->name);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700669 goto out;
670 }
671
672 if (!old_active)
673 goto out;
674
Joe Perchesada0f862014-02-15 16:02:17 -0800675 ether_addr_copy(saddr.sa_data, tmp_mac);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700676 saddr.sa_family = old_active->dev->type;
677
678 rv = dev_set_mac_address(old_active->dev, &saddr);
679 if (rv)
Veaceslav Falico76444f52014-07-15 19:35:58 +0200680 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
681 -rv, new_active->dev->name);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700682out:
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700683 break;
684 default:
Veaceslav Falico76444f52014-07-15 19:35:58 +0200685 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
686 bond->params.fail_over_mac);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700687 break;
688 }
689
690}
691
Jiri Pirkoa5499522009-09-25 03:28:09 +0000692static bool bond_should_change_active(struct bonding *bond)
693{
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +0200694 struct slave *prim = rtnl_dereference(bond->primary_slave);
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200695 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
Jiri Pirkoa5499522009-09-25 03:28:09 +0000696
697 if (!prim || !curr || curr->link != BOND_LINK_UP)
698 return true;
699 if (bond->force_primary) {
700 bond->force_primary = false;
701 return true;
702 }
703 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
704 (prim->speed < curr->speed ||
705 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
706 return false;
707 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
708 return false;
709 return true;
710}
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712/**
713 * find_best_interface - select the best available slave to be the active one
714 * @bond: our bonding struct
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 */
716static struct slave *bond_find_best_slave(struct bonding *bond)
717{
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +0200718 struct slave *slave, *bestslave = NULL, *primary;
Veaceslav Falico77140d22013-09-25 09:20:18 +0200719 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 int mintime = bond->params.updelay;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +0200722 primary = rtnl_dereference(bond->primary_slave);
723 if (primary && primary->link == BOND_LINK_UP &&
Veaceslav Falico77140d22013-09-25 09:20:18 +0200724 bond_should_change_active(bond))
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +0200725 return primary;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
Veaceslav Falico77140d22013-09-25 09:20:18 +0200727 bond_for_each_slave(bond, slave, iter) {
728 if (slave->link == BOND_LINK_UP)
729 return slave;
Veaceslav Falicob6adc612014-05-15 21:39:57 +0200730 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
Veaceslav Falico77140d22013-09-25 09:20:18 +0200731 slave->delay < mintime) {
732 mintime = slave->delay;
733 bestslave = slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 }
735 }
736
737 return bestslave;
738}
739
Ben Hutchingsad246c92011-04-26 15:25:52 +0000740static bool bond_should_notify_peers(struct bonding *bond)
741{
dingtianhong4cb4f972013-12-13 10:19:39 +0800742 struct slave *slave;
743
744 rcu_read_lock();
745 slave = rcu_dereference(bond->curr_active_slave);
746 rcu_read_unlock();
Ben Hutchingsad246c92011-04-26 15:25:52 +0000747
Veaceslav Falico76444f52014-07-15 19:35:58 +0200748 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
749 slave ? slave->dev->name : "NULL");
Ben Hutchingsad246c92011-04-26 15:25:52 +0000750
751 if (!slave || !bond->send_peer_notif ||
752 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
753 return false;
754
Ben Hutchingsad246c92011-04-26 15:25:52 +0000755 return true;
756}
757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758/**
759 * change_active_interface - change the active slave into the specified one
760 * @bond: our bonding struct
761 * @new: the new slave to make the active one
762 *
763 * Set the new slave to the bond's settings and unset them on the old
764 * curr_active_slave.
765 * Setting include flags, mc-list, promiscuity, allmulti, etc.
766 *
767 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
768 * because it is apparently the best available slave we have, even though its
769 * updelay hasn't timed out yet.
770 *
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200771 * Caller must hold RTNL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 */
Mitch Williamsa77b5322005-11-09 10:35:51 -0800773void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774{
Eric Dumazet4740d632014-07-15 06:56:55 -0700775 struct slave *old_active;
776
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200777 ASSERT_RTNL();
778
779 old_active = rtnl_dereference(bond->curr_active_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000781 if (old_active == new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
784 if (new_active) {
Veaceslav Falico8e603462014-02-18 07:48:46 +0100785 new_active->last_link_up = jiffies;
Jay Vosburghb2220ca2008-05-17 21:10:13 -0700786
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 if (new_active->link == BOND_LINK_BACK) {
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200788 if (bond_uses_primary(bond)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +0200789 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
790 new_active->dev->name,
791 (bond->params.updelay - new_active->delay) * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 }
793
794 new_active->delay = 0;
Moni Shoua69a23382015-02-03 16:48:30 +0200795 bond_set_slave_link_state(new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Veaceslav Falico01844092014-05-15 21:39:55 +0200797 if (BOND_MODE(bond) == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Holger Eitzenberger58402052008-12-09 23:07:13 -0800800 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 } else {
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200803 if (bond_uses_primary(bond)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +0200804 netdev_info(bond->dev, "making interface %s the new active one\n",
805 new_active->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 }
807 }
808 }
809
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200810 if (bond_uses_primary(bond))
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000811 bond_hw_addr_swap(bond, new_active, old_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
Holger Eitzenberger58402052008-12-09 23:07:13 -0800813 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 bond_alb_handle_active_change(bond, new_active);
Jay Vosburgh8f903c72006-02-21 16:36:44 -0800815 if (old_active)
dingtianhong5e5b0662014-02-26 11:05:22 +0800816 bond_set_slave_inactive_flags(old_active,
817 BOND_SLAVE_NOTIFY_NOW);
Jay Vosburgh8f903c72006-02-21 16:36:44 -0800818 if (new_active)
dingtianhong5e5b0662014-02-26 11:05:22 +0800819 bond_set_slave_active_flags(new_active,
820 BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 } else {
nikolay@redhat.com278b2082013-08-01 16:54:51 +0200822 rcu_assign_pointer(bond->curr_active_slave, new_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400824
Veaceslav Falico01844092014-05-15 21:39:55 +0200825 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000826 if (old_active)
dingtianhong5e5b0662014-02-26 11:05:22 +0800827 bond_set_slave_inactive_flags(old_active,
828 BOND_SLAVE_NOTIFY_NOW);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400829
830 if (new_active) {
Ben Hutchingsad246c92011-04-26 15:25:52 +0000831 bool should_notify_peers = false;
832
dingtianhong5e5b0662014-02-26 11:05:22 +0800833 bond_set_slave_active_flags(new_active,
834 BOND_SLAVE_NOTIFY_NOW);
Moni Shoua2ab82852007-10-09 19:43:39 -0700835
Or Gerlitz709f8a42008-06-13 18:12:01 -0700836 if (bond->params.fail_over_mac)
837 bond_do_fail_over_mac(bond, new_active,
838 old_active);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700839
Ben Hutchingsad246c92011-04-26 15:25:52 +0000840 if (netif_running(bond->dev)) {
841 bond->send_peer_notif =
842 bond->params.num_peer_notif;
843 should_notify_peers =
844 bond_should_notify_peers(bond);
845 }
846
Amerigo Wangb7bc2a52012-08-09 22:14:57 +0000847 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
Ben Hutchingsad246c92011-04-26 15:25:52 +0000848 if (should_notify_peers)
Amerigo Wangb7bc2a52012-08-09 22:14:57 +0000849 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
850 bond->dev);
Moni Shoua7893b242008-05-17 21:10:12 -0700851 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400852 }
Andy Gospodareka2fd9402010-03-25 14:49:05 +0000853
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000854 /* resend IGMP joins since active slave has changed or
Flavio Leitner94265cf2011-05-25 08:38:58 +0000855 * all were sent on curr_active_slave.
856 * resend only if bond is brought up with the affected
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200857 * bonding modes and the retransmission is enabled
858 */
Flavio Leitner94265cf2011-05-25 08:38:58 +0000859 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200860 ((bond_uses_primary(bond) && new_active) ||
Veaceslav Falico01844092014-05-15 21:39:55 +0200861 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000862 bond->igmp_retrans = bond->params.resend_igmp;
Nikolay Aleksandrov4beac022013-08-01 11:51:42 +0200863 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
Andy Gospodareka2fd9402010-03-25 14:49:05 +0000864 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
866
867/**
868 * bond_select_active_slave - select a new active slave, if needed
869 * @bond: our bonding struct
870 *
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000871 * This functions should be called when one of the following occurs:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 * - The old curr_active_slave has been released or lost its link.
873 * - The primary_slave has got its link back.
874 * - A slave has got its link back and there's no old curr_active_slave.
875 *
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200876 * Caller must hold RTNL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 */
Mitch Williamsa77b5322005-11-09 10:35:51 -0800878void bond_select_active_slave(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879{
880 struct slave *best_slave;
Jay Vosburghff59c452006-03-27 13:27:43 -0800881 int rv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
Nikolay Aleksandrove0974582014-09-15 17:19:35 +0200883 ASSERT_RTNL();
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 best_slave = bond_find_best_slave(bond);
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200886 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 bond_change_active_slave(bond, best_slave);
Jay Vosburghff59c452006-03-27 13:27:43 -0800888 rv = bond_set_carrier(bond);
889 if (!rv)
890 return;
891
892 if (netif_carrier_ok(bond->dev)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +0200893 netdev_info(bond->dev, "first active interface up!\n");
Jay Vosburghff59c452006-03-27 13:27:43 -0800894 } else {
Veaceslav Falico76444f52014-07-15 19:35:58 +0200895 netdev_info(bond->dev, "now running without any active interface!\n");
Jay Vosburghff59c452006-03-27 13:27:43 -0800896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 }
898}
899
WANG Congf6dc31a2010-05-06 00:48:51 -0700900#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000901static inline int slave_enable_netpoll(struct slave *slave)
WANG Congf6dc31a2010-05-06 00:48:51 -0700902{
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000903 struct netpoll *np;
904 int err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -0700905
Eric W. Biedermana8779ec2014-03-27 15:36:38 -0700906 np = kzalloc(sizeof(*np), GFP_KERNEL);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000907 err = -ENOMEM;
908 if (!np)
909 goto out;
910
Eric W. Biedermana8779ec2014-03-27 15:36:38 -0700911 err = __netpoll_setup(np, slave->dev);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000912 if (err) {
913 kfree(np);
914 goto out;
WANG Congf6dc31a2010-05-06 00:48:51 -0700915 }
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000916 slave->np = np;
917out:
918 return err;
919}
920static inline void slave_disable_netpoll(struct slave *slave)
921{
922 struct netpoll *np = slave->np;
923
924 if (!np)
925 return;
926
927 slave->np = NULL;
Neil Horman2cde6ac2013-02-11 10:25:30 +0000928 __netpoll_free_async(np);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000929}
WANG Congf6dc31a2010-05-06 00:48:51 -0700930
931static void bond_poll_controller(struct net_device *bond_dev)
932{
Mahesh Bandewar616f4542015-03-04 21:57:52 -0800933 struct bonding *bond = netdev_priv(bond_dev);
934 struct slave *slave = NULL;
935 struct list_head *iter;
936 struct ad_info ad_info;
937 struct netpoll_info *ni;
938 const struct net_device_ops *ops;
939
940 if (BOND_MODE(bond) == BOND_MODE_8023AD)
941 if (bond_3ad_get_active_agg_info(bond, &ad_info))
942 return;
943
944 rcu_read_lock_bh();
945 bond_for_each_slave_rcu(bond, slave, iter) {
946 ops = slave->dev->netdev_ops;
947 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
948 continue;
949
950 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
951 struct aggregator *agg =
952 SLAVE_AD_INFO(slave)->port.aggregator;
953
954 if (agg &&
955 agg->aggregator_identifier != ad_info.aggregator_id)
956 continue;
957 }
958
959 ni = rcu_dereference_bh(slave->dev->npinfo);
960 if (down_trylock(&ni->dev_lock))
961 continue;
962 ops->ndo_poll_controller(slave->dev);
963 up(&ni->dev_lock);
964 }
965 rcu_read_unlock_bh();
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000966}
967
dingtianhongc4cdef92013-07-23 15:25:27 +0800968static void bond_netpoll_cleanup(struct net_device *bond_dev)
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000969{
dingtianhongc4cdef92013-07-23 15:25:27 +0800970 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200971 struct list_head *iter;
Neil Hormanc2355e12010-10-13 16:01:49 +0000972 struct slave *slave;
Neil Hormanc2355e12010-10-13 16:01:49 +0000973
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200974 bond_for_each_slave(bond, slave, iter)
Veaceslav Falicob6adc612014-05-15 21:39:57 +0200975 if (bond_slave_is_up(slave))
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000976 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -0700977}
WANG Congf6dc31a2010-05-06 00:48:51 -0700978
Eric W. Biedermana8779ec2014-03-27 15:36:38 -0700979static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000980{
981 struct bonding *bond = netdev_priv(dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200982 struct list_head *iter;
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000983 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200984 int err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -0700985
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200986 bond_for_each_slave(bond, slave, iter) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000987 err = slave_enable_netpoll(slave);
988 if (err) {
dingtianhongc4cdef92013-07-23 15:25:27 +0800989 bond_netpoll_cleanup(dev);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000990 break;
991 }
992 }
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000993 return err;
994}
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000995#else
996static inline int slave_enable_netpoll(struct slave *slave)
997{
998 return 0;
999}
1000static inline void slave_disable_netpoll(struct slave *slave)
1001{
1002}
WANG Congf6dc31a2010-05-06 00:48:51 -07001003static void bond_netpoll_cleanup(struct net_device *bond_dev)
1004{
1005}
WANG Congf6dc31a2010-05-06 00:48:51 -07001006#endif
1007
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008/*---------------------------------- IOCTL ----------------------------------*/
1009
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001010static netdev_features_t bond_fix_features(struct net_device *dev,
nikolay@redhat.com9b7b1652013-09-02 13:51:41 +02001011 netdev_features_t features)
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001012{
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001013 struct bonding *bond = netdev_priv(dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001014 struct list_head *iter;
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001015 netdev_features_t mask;
nikolay@redhat.com9b7b1652013-09-02 13:51:41 +02001016 struct slave *slave;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001017
Scott Feldman7889cbe2015-05-10 09:48:07 -07001018 mask = features;
Roopa Prabhuc158cba2015-01-29 22:40:16 -08001019
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001020 features &= ~NETIF_F_ONE_FOR_ALL;
1021 features |= NETIF_F_ALL_FOR_ALL;
1022
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001023 bond_for_each_slave(bond, slave, iter) {
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001024 features = netdev_increment_features(features,
1025 slave->dev->features,
1026 mask);
1027 }
Eric Dumazetb0ce3502013-05-16 07:34:53 +00001028 features = netdev_add_tso_features(features, mask);
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001029
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001030 return features;
1031}
1032
Michał Mirosław62f2a3a2011-07-13 14:10:29 +00001033#define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1034 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1035 NETIF_F_HIGHDMA | NETIF_F_LRO)
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001036
Or Gerlitz5a7baa72014-06-17 16:11:09 +03001037#define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
Eric Dumazet24f87d42015-01-25 23:27:14 -08001038 NETIF_F_TSO)
Or Gerlitz5a7baa72014-06-17 16:11:09 +03001039
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001040static void bond_compute_features(struct bonding *bond)
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001041{
Eric Dumazet02875872014-10-05 18:38:35 -07001042 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1043 IFF_XMIT_DST_RELEASE_PERM;
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001044 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
Or Gerlitz5a7baa72014-06-17 16:11:09 +03001045 netdev_features_t enc_features = BOND_ENC_FEATURES;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001046 struct net_device *bond_dev = bond->dev;
1047 struct list_head *iter;
1048 struct slave *slave;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001049 unsigned short max_hard_header_len = ETH_HLEN;
Sarveshwar Bandi0e376bd2012-11-21 04:35:03 +00001050 unsigned int gso_max_size = GSO_MAX_SIZE;
1051 u16 gso_max_segs = GSO_MAX_SEGS;
Herbert Xub63365a2008-10-23 01:11:29 -07001052
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001053 if (!bond_has_slaves(bond))
Herbert Xub63365a2008-10-23 01:11:29 -07001054 goto done;
Michal Kubečeka9b3ace2014-05-20 08:29:35 +02001055 vlan_features &= NETIF_F_ALL_FOR_ALL;
Herbert Xub63365a2008-10-23 01:11:29 -07001056
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001057 bond_for_each_slave(bond, slave, iter) {
Jay Vosburgh278339a2009-08-28 12:05:12 +00001058 vlan_features = netdev_increment_features(vlan_features,
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001059 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1060
Or Gerlitz5a7baa72014-06-17 16:11:09 +03001061 enc_features = netdev_increment_features(enc_features,
1062 slave->dev->hw_enc_features,
1063 BOND_ENC_FEATURES);
Eric Dumazetb6fe83e2012-07-17 12:19:48 +00001064 dst_release_flag &= slave->dev->priv_flags;
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001065 if (slave->dev->hard_header_len > max_hard_header_len)
1066 max_hard_header_len = slave->dev->hard_header_len;
Sarveshwar Bandi0e376bd2012-11-21 04:35:03 +00001067
1068 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1069 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001070 }
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001071
Herbert Xub63365a2008-10-23 01:11:29 -07001072done:
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001073 bond_dev->vlan_features = vlan_features;
Eric Dumazet24f87d42015-01-25 23:27:14 -08001074 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001075 bond_dev->hard_header_len = max_hard_header_len;
Sarveshwar Bandi0e376bd2012-11-21 04:35:03 +00001076 bond_dev->gso_max_segs = gso_max_segs;
1077 netif_set_gso_max_size(bond_dev, gso_max_size);
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001078
Eric Dumazet02875872014-10-05 18:38:35 -07001079 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1080 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1081 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1082 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
Eric Dumazetb6fe83e2012-07-17 12:19:48 +00001083
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001084 netdev_change_features(bond_dev);
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001085}
1086
Moni Shoua872254d2007-10-09 19:43:38 -07001087static void bond_setup_by_slave(struct net_device *bond_dev,
1088 struct net_device *slave_dev)
1089{
Stephen Hemminger00829822008-11-20 20:14:53 -08001090 bond_dev->header_ops = slave_dev->header_ops;
Moni Shoua872254d2007-10-09 19:43:38 -07001091
1092 bond_dev->type = slave_dev->type;
1093 bond_dev->hard_header_len = slave_dev->hard_header_len;
1094 bond_dev->addr_len = slave_dev->addr_len;
1095
1096 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1097 slave_dev->addr_len);
1098}
1099
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001100/* On bonding slaves other than the currently active slave, suppress
Jiri Pirko3aba8912011-04-19 03:48:16 +00001101 * duplicates except for alb non-mcast/bcast.
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001102 */
1103static bool bond_should_deliver_exact_match(struct sk_buff *skb,
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001104 struct slave *slave,
1105 struct bonding *bond)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001106{
Jiri Pirko2d7011c2011-03-16 08:46:43 +00001107 if (bond_is_slave_inactive(slave)) {
Veaceslav Falico01844092014-05-15 21:39:55 +02001108 if (BOND_MODE(bond) == BOND_MODE_ALB &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001109 skb->pkt_type != PACKET_BROADCAST &&
1110 skb->pkt_type != PACKET_MULTICAST)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001111 return false;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001112 return true;
1113 }
1114 return false;
1115}
1116
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001117static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001118{
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001119 struct sk_buff *skb = *pskb;
Jiri Pirkof1c17752011-03-12 03:14:35 +00001120 struct slave *slave;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001121 struct bonding *bond;
Eric Dumazetde063b72012-06-11 19:23:07 +00001122 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1123 struct slave *);
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00001124 int ret = RX_HANDLER_ANOTHER;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001125
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001126 skb = skb_share_check(skb, GFP_ATOMIC);
1127 if (unlikely(!skb))
1128 return RX_HANDLER_CONSUMED;
1129
1130 *pskb = skb;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001131
Jiri Pirko35d48902011-03-22 02:38:12 +00001132 slave = bond_slave_get_rcu(skb->dev);
1133 bond = slave->bond;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001134
Mitsuo Hayasaka4d974802011-10-12 16:04:29 +00001135 recv_probe = ACCESS_ONCE(bond->recv_probe);
1136 if (recv_probe) {
Eric Dumazetde063b72012-06-11 19:23:07 +00001137 ret = recv_probe(skb, bond, slave);
1138 if (ret == RX_HANDLER_CONSUMED) {
1139 consume_skb(skb);
1140 return ret;
Jiri Pirko3aba8912011-04-19 03:48:16 +00001141 }
1142 }
1143
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001144 if (bond_should_deliver_exact_match(skb, slave, bond)) {
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001145 return RX_HANDLER_EXACT;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001146 }
1147
Jiri Pirko35d48902011-03-22 02:38:12 +00001148 skb->dev = bond->dev;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001149
Veaceslav Falico01844092014-05-15 21:39:55 +02001150 if (BOND_MODE(bond) == BOND_MODE_ALB &&
Jiri Pirko35d48902011-03-22 02:38:12 +00001151 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001152 skb->pkt_type == PACKET_HOST) {
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001153
Changli Gao541ac7c2011-03-02 21:07:14 +00001154 if (unlikely(skb_cow_head(skb,
1155 skb->data - skb_mac_header(skb)))) {
1156 kfree_skb(skb);
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001157 return RX_HANDLER_CONSUMED;
Changli Gao541ac7c2011-03-02 21:07:14 +00001158 }
Joe Perches2a7c1832014-02-18 09:42:45 -08001159 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001160 }
1161
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00001162 return ret;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001163}
1164
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001165static int bond_master_upper_dev_link(struct net_device *bond_dev,
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001166 struct net_device *slave_dev,
1167 struct slave *slave)
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001168{
1169 int err;
1170
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001171 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001172 if (err)
1173 return err;
1174 slave_dev->flags |= IFF_SLAVE;
Alexei Starovoitov7f294052013-10-23 16:02:42 -07001175 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001176 return 0;
1177}
1178
1179static void bond_upper_dev_unlink(struct net_device *bond_dev,
1180 struct net_device *slave_dev)
1181{
1182 netdev_upper_dev_unlink(slave_dev, bond_dev);
1183 slave_dev->flags &= ~IFF_SLAVE;
Alexei Starovoitov7f294052013-10-23 16:02:42 -07001184 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001185}
1186
dingtianhong3fdddd82014-05-12 15:08:43 +08001187static struct slave *bond_alloc_slave(struct bonding *bond)
1188{
1189 struct slave *slave = NULL;
1190
1191 slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1192 if (!slave)
1193 return NULL;
1194
Veaceslav Falico01844092014-05-15 21:39:55 +02001195 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
dingtianhong3fdddd82014-05-12 15:08:43 +08001196 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1197 GFP_KERNEL);
1198 if (!SLAVE_AD_INFO(slave)) {
1199 kfree(slave);
1200 return NULL;
1201 }
1202 }
1203 return slave;
1204}
1205
1206static void bond_free_slave(struct slave *slave)
1207{
1208 struct bonding *bond = bond_get_bond_by_slave(slave);
1209
Veaceslav Falico01844092014-05-15 21:39:55 +02001210 if (BOND_MODE(bond) == BOND_MODE_8023AD)
dingtianhong3fdddd82014-05-12 15:08:43 +08001211 kfree(SLAVE_AD_INFO(slave));
1212
1213 kfree(slave);
1214}
1215
Moni Shoua69a23382015-02-03 16:48:30 +02001216static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1217{
1218 info->bond_mode = BOND_MODE(bond);
1219 info->miimon = bond->params.miimon;
1220 info->num_slaves = bond->slave_cnt;
1221}
1222
1223static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1224{
1225 strcpy(info->slave_name, slave->dev->name);
1226 info->link = slave->link;
1227 info->state = bond_slave_state(slave);
1228 info->link_failure_count = slave->link_failure_count;
1229}
1230
Moni Shoua92e584f2015-02-08 11:49:32 +02001231static void bond_netdev_notify(struct net_device *dev,
1232 struct netdev_bonding_info *info)
Moni Shoua69e61132015-02-03 16:48:31 +02001233{
Moni Shoua69e61132015-02-03 16:48:31 +02001234 rtnl_lock();
Moni Shoua92e584f2015-02-08 11:49:32 +02001235 netdev_bonding_info_change(dev, info);
Moni Shoua69e61132015-02-03 16:48:31 +02001236 rtnl_unlock();
1237}
1238
1239static void bond_netdev_notify_work(struct work_struct *_work)
1240{
1241 struct netdev_notify_work *w =
1242 container_of(_work, struct netdev_notify_work, work.work);
1243
Moni Shoua92e584f2015-02-08 11:49:32 +02001244 bond_netdev_notify(w->dev, &w->bonding_info);
Moni Shoua69e61132015-02-03 16:48:31 +02001245 dev_put(w->dev);
Moni Shoua92e584f2015-02-08 11:49:32 +02001246 kfree(w);
Moni Shoua69e61132015-02-03 16:48:31 +02001247}
1248
1249void bond_queue_slave_event(struct slave *slave)
1250{
Moni Shoua92e584f2015-02-08 11:49:32 +02001251 struct bonding *bond = slave->bond;
Moni Shoua69e61132015-02-03 16:48:31 +02001252 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1253
1254 if (!nnw)
1255 return;
1256
Moni Shoua92e584f2015-02-08 11:49:32 +02001257 dev_hold(slave->dev);
Moni Shoua69e61132015-02-03 16:48:31 +02001258 nnw->dev = slave->dev;
Moni Shoua92e584f2015-02-08 11:49:32 +02001259 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1260 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1261 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
Moni Shoua69e61132015-02-03 16:48:31 +02001262
Moni Shoua92e584f2015-02-08 11:49:32 +02001263 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
Moni Shoua69e61132015-02-03 16:48:31 +02001264}
1265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266/* enslave device <slave> to bond device <master> */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001267int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268{
Wang Chen454d7c92008-11-12 23:37:49 -08001269 struct bonding *bond = netdev_priv(bond_dev);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001270 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Veaceslav Falicoc8c23902013-09-25 09:20:25 +02001271 struct slave *new_slave = NULL, *prev_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 struct sockaddr addr;
1273 int link_reporting;
Veaceslav Falico8599b522013-06-24 11:49:34 +02001274 int res = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Ben Hutchingsc772dde2012-12-07 06:15:32 +00001276 if (!bond->params.use_carrier &&
1277 slave_dev->ethtool_ops->get_link == NULL &&
1278 slave_ops->ndo_do_ioctl == NULL) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001279 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1280 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 }
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 /* already enslaved */
1284 if (slave_dev->flags & IFF_SLAVE) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001285 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 return -EBUSY;
1287 }
1288
Jiri Bohac09a89c22014-02-26 18:20:13 +01001289 if (bond_dev == slave_dev) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001290 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
Jiri Bohac09a89c22014-02-26 18:20:13 +01001291 return -EPERM;
1292 }
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 /* vlan challenged mutual exclusion */
1295 /* no need to lock since we're protected by rtnl_lock */
1296 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001297 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1298 slave_dev->name);
Jiri Pirko55462cf2012-10-14 04:30:56 +00001299 if (vlan_uses_dev(bond_dev)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001300 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1301 slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 return -EPERM;
1303 } else {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001304 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1305 slave_dev->name, slave_dev->name,
1306 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 }
1308 } else {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001309 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1310 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 }
1312
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001313 /* Old ifenslave binaries are no longer supported. These can
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001314 * be identified with moderate accuracy by the state of the slave:
Jay Vosburgh217df672005-09-26 16:11:50 -07001315 * the current ifenslave will set the interface down prior to
1316 * enslaving it; the old ifenslave will not.
1317 */
1318 if ((slave_dev->flags & IFF_UP)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001319 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1320 slave_dev->name);
Jay Vosburgh217df672005-09-26 16:11:50 -07001321 res = -EPERM;
1322 goto err_undo_flags;
1323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
Moni Shoua872254d2007-10-09 19:43:38 -07001325 /* set bonding device ether type by slave - bonding netdevices are
1326 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1327 * there is a need to override some of the type dependent attribs/funcs.
1328 *
1329 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1330 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1331 */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001332 if (!bond_has_slaves(bond)) {
Moni Shouae36b9d12009-07-15 04:56:31 +00001333 if (bond_dev->type != slave_dev->type) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001334 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1335 bond_dev->type, slave_dev->type);
Moni Shoua75c78502009-09-15 02:37:40 -07001336
Amerigo Wangb7bc2a52012-08-09 22:14:57 +00001337 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1338 bond_dev);
Jiri Pirko3ca5b402010-03-10 10:29:35 +00001339 res = notifier_to_errno(res);
1340 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001341 netdev_err(bond_dev, "refused to change device type\n");
Jiri Pirko3ca5b402010-03-10 10:29:35 +00001342 res = -EBUSY;
1343 goto err_undo_flags;
1344 }
Moni Shoua75c78502009-09-15 02:37:40 -07001345
Jiri Pirko32a806c2010-03-19 04:00:23 +00001346 /* Flush unicast and multicast addresses */
Jiri Pirkoa748ee22010-04-01 21:22:09 +00001347 dev_uc_flush(bond_dev);
Jiri Pirko22bedad32010-04-01 21:22:57 +00001348 dev_mc_flush(bond_dev);
Jiri Pirko32a806c2010-03-19 04:00:23 +00001349
Moni Shouae36b9d12009-07-15 04:56:31 +00001350 if (slave_dev->type != ARPHRD_ETHER)
1351 bond_setup_by_slave(bond_dev, slave_dev);
Neil Horman550fd082011-07-26 06:05:38 +00001352 else {
Moni Shouae36b9d12009-07-15 04:56:31 +00001353 ether_setup(bond_dev);
Neil Horman550fd082011-07-26 06:05:38 +00001354 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1355 }
Moni Shoua75c78502009-09-15 02:37:40 -07001356
Amerigo Wangb7bc2a52012-08-09 22:14:57 +00001357 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1358 bond_dev);
Moni Shouae36b9d12009-07-15 04:56:31 +00001359 }
Moni Shoua872254d2007-10-09 19:43:38 -07001360 } else if (bond_dev->type != slave_dev->type) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001361 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1362 slave_dev->name, slave_dev->type, bond_dev->type);
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001363 res = -EINVAL;
1364 goto err_undo_flags;
Moni Shoua872254d2007-10-09 19:43:38 -07001365 }
1366
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001367 if (slave_ops->ndo_set_mac_address == NULL) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001368 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
Veaceslav Falicof5442442014-07-15 13:26:01 +02001369 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1370 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1371 if (!bond_has_slaves(bond)) {
dingtianhong00503b62014-01-25 13:00:29 +08001372 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
Veaceslav Falico76444f52014-07-15 19:35:58 +02001373 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
Veaceslav Falicof5442442014-07-15 13:26:01 +02001374 } else {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001375 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");
Veaceslav Falicof5442442014-07-15 13:26:01 +02001376 res = -EOPNOTSUPP;
1377 goto err_undo_flags;
dingtianhong00503b62014-01-25 13:00:29 +08001378 }
Moni Shoua2ab82852007-10-09 19:43:39 -07001379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 }
1381
Amerigo Wang8d8fc292011-05-19 21:39:10 +00001382 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1383
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001384 /* If this is the first slave, then we need to set the master's hardware
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001385 * address to be the same as the slave's.
1386 */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001387 if (!bond_has_slaves(bond) &&
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001388 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
Jiri Pirko409cc1f2013-01-30 11:08:11 +01001389 bond_set_dev_addr(bond->dev, slave_dev);
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001390
dingtianhong3fdddd82014-05-12 15:08:43 +08001391 new_slave = bond_alloc_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 if (!new_slave) {
1393 res = -ENOMEM;
1394 goto err_undo_flags;
1395 }
dingtianhong3fdddd82014-05-12 15:08:43 +08001396
Veaceslav Falicodc73c412014-05-21 17:42:00 +02001397 new_slave->bond = bond;
1398 new_slave->dev = slave_dev;
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001399 /* Set the new_slave's queue_id to be zero. Queue ID mapping
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00001400 * is set via sysfs or module option if desired.
1401 */
1402 new_slave->queue_id = 0;
1403
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001404 /* Save slave's original mtu and then set it to match the bond */
1405 new_slave->original_mtu = slave_dev->mtu;
1406 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1407 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001408 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001409 goto err_free;
1410 }
1411
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001412 /* Save slave's original ("permanent") mac address for modes
Jay Vosburgh217df672005-09-26 16:11:50 -07001413 * that need it, and for restoring it upon release, and then
1414 * set it to the master's address
1415 */
Joe Perchesada0f862014-02-15 16:02:17 -08001416 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
dingtianhong00503b62014-01-25 13:00:29 +08001418 if (!bond->params.fail_over_mac ||
Veaceslav Falico01844092014-05-15 21:39:55 +02001419 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001420 /* Set slave to master's mac address. The application already
Moni Shoua2ab82852007-10-09 19:43:39 -07001421 * set the master's mac address to that of the first slave
1422 */
1423 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1424 addr.sa_family = slave_dev->type;
1425 res = dev_set_mac_address(slave_dev, &addr);
1426 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001427 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001428 goto err_restore_mtu;
Moni Shoua2ab82852007-10-09 19:43:39 -07001429 }
Jay Vosburgh217df672005-09-26 16:11:50 -07001430 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
Jay Vosburgh217df672005-09-26 16:11:50 -07001432 /* open the slave since the application closed it */
1433 res = dev_open(slave_dev);
1434 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001435 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001436 goto err_restore_mac;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 }
1438
Jay Vosburgh0b680e72006-09-22 21:54:10 -07001439 slave_dev->priv_flags |= IFF_BONDING;
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04001440 /* initialize slave stats */
1441 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Holger Eitzenberger58402052008-12-09 23:07:13 -08001443 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 /* bond_alb_init_slave() must be called before all other stages since
1445 * it might fail and we do not want to have to undo everything
1446 */
1447 res = bond_alb_init_slave(bond, new_slave);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001448 if (res)
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001449 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 }
1451
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02001452 /* If the mode uses primary, then the following is handled by
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001453 * bond_change_active_slave().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 */
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02001455 if (!bond_uses_primary(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 /* set promiscuity level to new slave */
1457 if (bond_dev->flags & IFF_PROMISC) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -07001458 res = dev_set_promiscuity(slave_dev, 1);
1459 if (res)
1460 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 }
1462
1463 /* set allmulti level to new slave */
1464 if (bond_dev->flags & IFF_ALLMULTI) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -07001465 res = dev_set_allmulti(slave_dev, 1);
1466 if (res)
1467 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 }
1469
David S. Millerb9e40852008-07-15 00:15:08 -07001470 netif_addr_lock_bh(bond_dev);
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001471
1472 dev_mc_sync_multiple(slave_dev, bond_dev);
1473 dev_uc_sync_multiple(slave_dev, bond_dev);
1474
David S. Millerb9e40852008-07-15 00:15:08 -07001475 netif_addr_unlock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 }
1477
Veaceslav Falico01844092014-05-15 21:39:55 +02001478 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 /* add lacpdu mc addr to mc list */
1480 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1481
Jiri Pirko22bedad32010-04-01 21:22:57 +00001482 dev_mc_add(slave_dev, lacpdu_multicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 }
1484
Wei Yongjunb8e2fde2013-08-23 10:45:07 +08001485 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1486 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001487 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1488 slave_dev->name);
nikolay@redhat.com1ff412a2013-08-06 12:40:15 +02001489 goto err_close;
1490 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491
Veaceslav Falicoc8c23902013-09-25 09:20:25 +02001492 prev_slave = bond_last_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
1494 new_slave->delay = 0;
1495 new_slave->link_failure_count = 0;
1496
Veaceslav Falico876254a2013-03-12 06:31:32 +00001497 bond_update_speed_duplex(new_slave);
1498
Veaceslav Falico49f17de2014-02-18 07:48:47 +01001499 new_slave->last_rx = jiffies -
Michal Kubečekf31c7932012-04-17 02:02:06 +00001500 (msecs_to_jiffies(bond->params.arp_interval) + 1);
Veaceslav Falico8599b522013-06-24 11:49:34 +02001501 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
Veaceslav Falico49f17de2014-02-18 07:48:47 +01001502 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07001503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 if (bond->params.miimon && !bond->params.use_carrier) {
1505 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1506
1507 if ((link_reporting == -1) && !bond->params.arp_interval) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001508 /* miimon is set but a bonded network driver
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 * does not support ETHTOOL/MII and
1510 * arp_interval is not set. Note: if
1511 * use_carrier is enabled, we will never go
1512 * here (because netif_carrier is always
1513 * supported); thus, we don't need to change
1514 * the messages for netif_carrier.
1515 */
Veaceslav Falico76444f52014-07-15 19:35:58 +02001516 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",
1517 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 } else if (link_reporting == -1) {
1519 /* unable get link status using mii/ethtool */
Veaceslav Falico76444f52014-07-15 19:35:58 +02001520 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",
1521 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 }
1523 }
1524
1525 /* check for initial state */
Michal Kubečekf31c7932012-04-17 02:02:06 +00001526 if (bond->params.miimon) {
1527 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1528 if (bond->params.updelay) {
Moni Shoua69a23382015-02-03 16:48:30 +02001529 bond_set_slave_link_state(new_slave,
1530 BOND_LINK_BACK);
Michal Kubečekf31c7932012-04-17 02:02:06 +00001531 new_slave->delay = bond->params.updelay;
1532 } else {
Moni Shoua69a23382015-02-03 16:48:30 +02001533 bond_set_slave_link_state(new_slave,
1534 BOND_LINK_UP);
Michal Kubečekf31c7932012-04-17 02:02:06 +00001535 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 } else {
Moni Shoua69a23382015-02-03 16:48:30 +02001537 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 }
Michal Kubečekf31c7932012-04-17 02:02:06 +00001539 } else if (bond->params.arp_interval) {
Moni Shoua69a23382015-02-03 16:48:30 +02001540 bond_set_slave_link_state(new_slave,
1541 (netif_carrier_ok(slave_dev) ?
1542 BOND_LINK_UP : BOND_LINK_DOWN));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 } else {
Moni Shoua69a23382015-02-03 16:48:30 +02001544 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 }
1546
Michal Kubečekf31c7932012-04-17 02:02:06 +00001547 if (new_slave->link != BOND_LINK_DOWN)
Veaceslav Falico8e603462014-02-18 07:48:46 +01001548 new_slave->last_link_up = jiffies;
Veaceslav Falico76444f52014-07-15 19:35:58 +02001549 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1550 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1551 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
Michal Kubečekf31c7932012-04-17 02:02:06 +00001552
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02001553 if (bond_uses_primary(bond) && bond->params.primary[0]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 /* if there is a primary slave, remember it */
Jiri Pirkoa5499522009-09-25 03:28:09 +00001555 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02001556 rcu_assign_pointer(bond->primary_slave, new_slave);
Jiri Pirkoa5499522009-09-25 03:28:09 +00001557 bond->force_primary = true;
1558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 }
1560
Veaceslav Falico01844092014-05-15 21:39:55 +02001561 switch (BOND_MODE(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 case BOND_MODE_ACTIVEBACKUP:
dingtianhong5e5b0662014-02-26 11:05:22 +08001563 bond_set_slave_inactive_flags(new_slave,
1564 BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 break;
1566 case BOND_MODE_8023AD:
1567 /* in 802.3ad mode, the internal mechanism
1568 * will activate the slaves in the selected
1569 * aggregator
1570 */
dingtianhong5e5b0662014-02-26 11:05:22 +08001571 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 /* if this is the first slave */
Veaceslav Falico23c147e2013-09-27 15:10:57 +02001573 if (!prev_slave) {
dingtianhong3fdddd82014-05-12 15:08:43 +08001574 SLAVE_AD_INFO(new_slave)->id = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 /* Initialize AD with the number of times that the AD timer is called in 1 second
1576 * can be called only after the mac address of the bond is set
1577 */
Peter Pan(潘卫平)56d00c672011-06-08 21:19:02 +00001578 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 } else {
dingtianhong3fdddd82014-05-12 15:08:43 +08001580 SLAVE_AD_INFO(new_slave)->id =
1581 SLAVE_AD_INFO(prev_slave)->id + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 }
1583
1584 bond_3ad_bind_slave(new_slave);
1585 break;
1586 case BOND_MODE_TLB:
1587 case BOND_MODE_ALB:
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001588 bond_set_active_slave(new_slave);
dingtianhong5e5b0662014-02-26 11:05:22 +08001589 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 break;
1591 default:
Veaceslav Falico76444f52014-07-15 19:35:58 +02001592 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
1594 /* always active in trunk mode */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001595 bond_set_active_slave(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597 /* In trunking mode there is little meaning to curr_active_slave
1598 * anyway (it holds no special properties of the bond device),
1599 * so we can change it without calling change_active_interface()
1600 */
Eric Dumazet4740d632014-07-15 06:56:55 -07001601 if (!rcu_access_pointer(bond->curr_active_slave) &&
1602 new_slave->link == BOND_LINK_UP)
nikolay@redhat.com278b2082013-08-01 16:54:51 +02001603 rcu_assign_pointer(bond->curr_active_slave, new_slave);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001604
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 break;
1606 } /* switch(bond_mode) */
1607
WANG Congf6dc31a2010-05-06 00:48:51 -07001608#ifdef CONFIG_NET_POLL_CONTROLLER
stephen hemminger10eccb42013-07-24 11:53:57 -07001609 slave_dev->npinfo = bond->dev->npinfo;
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001610 if (slave_dev->npinfo) {
1611 if (slave_enable_netpoll(new_slave)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001612 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001613 res = -EBUSY;
stephen hemmingerf7d98212011-12-31 13:26:46 +00001614 goto err_detach;
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001615 }
WANG Congf6dc31a2010-05-06 00:48:51 -07001616 }
1617#endif
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001618
Michal Kubečekfbe168b2014-11-13 07:54:50 +01001619 if (!(bond_dev->features & NETIF_F_LRO))
1620 dev_disable_lro(slave_dev);
1621
Jiri Pirko35d48902011-03-22 02:38:12 +00001622 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1623 new_slave);
1624 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001625 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
Veaceslav Falico5831d662013-09-25 09:20:32 +02001626 goto err_detach;
Jiri Pirko35d48902011-03-22 02:38:12 +00001627 }
1628
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001629 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1630 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001631 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001632 goto err_unregister;
1633 }
1634
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001635 res = bond_sysfs_slave_add(new_slave);
1636 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001637 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001638 goto err_upper_unlink;
1639 }
1640
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001641 bond->slave_cnt++;
1642 bond_compute_features(bond);
1643 bond_set_carrier(bond);
1644
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02001645 if (bond_uses_primary(bond)) {
dingtianhongf80889a2014-02-12 12:06:40 +08001646 block_netpoll_tx();
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001647 bond_select_active_slave(bond);
dingtianhongf80889a2014-02-12 12:06:40 +08001648 unblock_netpoll_tx();
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001649 }
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001650
Mahesh Bandewaree637712014-10-04 17:45:01 -07001651 if (bond_mode_uses_xmit_hash(bond))
1652 bond_update_slave_arr(bond, NULL);
1653
Veaceslav Falico76444f52014-07-15 19:35:58 +02001654 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1655 slave_dev->name,
1656 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1657 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659 /* enslave is successful */
Moni Shoua69e61132015-02-03 16:48:31 +02001660 bond_queue_slave_event(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 return 0;
1662
1663/* Undo stages on error */
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001664err_upper_unlink:
1665 bond_upper_dev_unlink(bond_dev, slave_dev);
1666
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001667err_unregister:
1668 netdev_rx_handler_unregister(slave_dev);
1669
stephen hemmingerf7d98212011-12-31 13:26:46 +00001670err_detach:
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02001671 if (!bond_uses_primary(bond))
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001672 bond_hw_addr_flush(bond_dev, slave_dev);
1673
nikolay@redhat.com1ff412a2013-08-06 12:40:15 +02001674 vlan_vids_del_by_dev(slave_dev, bond_dev);
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02001675 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1676 RCU_INIT_POINTER(bond->primary_slave, NULL);
Eric Dumazet4740d632014-07-15 06:56:55 -07001677 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
dingtianhongf80889a2014-02-12 12:06:40 +08001678 block_netpoll_tx();
dingtianhongc8517032013-12-13 10:20:07 +08001679 bond_change_active_slave(bond, NULL);
nikolay@redhat.com3c5913b532013-04-18 07:33:36 +00001680 bond_select_active_slave(bond);
dingtianhongf80889a2014-02-12 12:06:40 +08001681 unblock_netpoll_tx();
nikolay@redhat.com3c5913b532013-04-18 07:33:36 +00001682 }
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02001683 /* either primary_slave or curr_active_slave might've changed */
1684 synchronize_rcu();
nikolay@redhat.comfc7a72a2013-04-18 07:33:37 +00001685 slave_disable_netpoll(new_slave);
stephen hemmingerf7d98212011-12-31 13:26:46 +00001686
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687err_close:
nikolay@redhat.comb6a5a7b2013-04-11 09:18:56 +00001688 slave_dev->priv_flags &= ~IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 dev_close(slave_dev);
1690
1691err_restore_mac:
dingtianhong00503b62014-01-25 13:00:29 +08001692 if (!bond->params.fail_over_mac ||
Veaceslav Falico01844092014-05-15 21:39:55 +02001693 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001694 /* XXX TODO - fom follow mode needs to change master's
1695 * MAC if this slave's MAC is in use by the bond, or at
1696 * least print a warning.
1697 */
Joe Perchesada0f862014-02-15 16:02:17 -08001698 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
Moni Shoua2ab82852007-10-09 19:43:39 -07001699 addr.sa_family = slave_dev->type;
1700 dev_set_mac_address(slave_dev, &addr);
1701 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001703err_restore_mtu:
1704 dev_set_mtu(slave_dev, new_slave->original_mtu);
1705
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706err_free:
dingtianhong3fdddd82014-05-12 15:08:43 +08001707 bond_free_slave(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
1709err_undo_flags:
Nikolay Aleksandrovb8fad452013-06-12 00:07:01 +02001710 /* Enslave of first slave has failed and we need to fix master's mac */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001711 if (!bond_has_slaves(bond) &&
dingtianhong844223a2014-01-02 09:13:16 +08001712 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
Nikolay Aleksandrovb8fad452013-06-12 00:07:01 +02001713 eth_hw_addr_random(bond_dev);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 return res;
1716}
1717
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001718/* Try to release the slave device <slave> from the bond device <master>
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 * It is legal to access curr_active_slave without a lock because all the function
Nikolay Aleksandrov8c0bc552014-09-11 22:49:28 +02001720 * is RTNL-locked. If "all" is true it means that the function is being called
nikolay@redhat.com08963412013-02-18 14:09:42 +00001721 * while destroying a bond interface and all slaves are being released.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 *
1723 * The rules for slave state should be:
1724 * for Active/Backup:
1725 * Active stays on all backups go down
1726 * for Bonded connections:
1727 * The first up interface should be left on and all others downed.
1728 */
nikolay@redhat.com08963412013-02-18 14:09:42 +00001729static int __bond_release_one(struct net_device *bond_dev,
1730 struct net_device *slave_dev,
1731 bool all)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732{
Wang Chen454d7c92008-11-12 23:37:49 -08001733 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 struct slave *slave, *oldcurrent;
1735 struct sockaddr addr;
Neil Horman5a0068d2013-09-27 12:22:15 -04001736 int old_flags = bond_dev->flags;
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001737 netdev_features_t old_features = bond_dev->features;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738
1739 /* slave is not a slave or master is not master of this slave */
1740 if (!(slave_dev->flags & IFF_SLAVE) ||
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001741 !netdev_has_upper_dev(slave_dev, bond_dev)) {
Wengang Wanga22a9e42014-12-23 09:24:36 +08001742 netdev_dbg(bond_dev, "cannot release %s\n",
Veaceslav Falico76444f52014-07-15 19:35:58 +02001743 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 return -EINVAL;
1745 }
1746
Neil Hormane843fa52010-10-13 16:01:50 +00001747 block_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748
1749 slave = bond_get_slave_by_dev(bond, slave_dev);
1750 if (!slave) {
1751 /* not a slave of this bond */
Veaceslav Falico76444f52014-07-15 19:35:58 +02001752 netdev_info(bond_dev, "%s not enslaved\n",
1753 slave_dev->name);
Neil Hormane843fa52010-10-13 16:01:50 +00001754 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 return -EINVAL;
1756 }
1757
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001758 bond_sysfs_slave_del(slave);
1759
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04001760 /* recompute stats just before removing the slave */
1761 bond_get_stats(bond->dev, &bond->bond_stats);
1762
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001763 bond_upper_dev_unlink(bond_dev, slave_dev);
Jiri Pirko35d48902011-03-22 02:38:12 +00001764 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1765 * for this slave anymore.
1766 */
1767 netdev_rx_handler_unregister(slave_dev);
Jiri Pirko35d48902011-03-22 02:38:12 +00001768
Nikolay Aleksandrove4702592014-09-11 22:49:27 +02001769 if (BOND_MODE(bond) == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 bond_3ad_unbind_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
Mahesh Bandewaree637712014-10-04 17:45:01 -07001772 if (bond_mode_uses_xmit_hash(bond))
1773 bond_update_slave_arr(bond, slave);
1774
Veaceslav Falico76444f52014-07-15 19:35:58 +02001775 netdev_info(bond_dev, "Releasing %s interface %s\n",
1776 bond_is_active_slave(slave) ? "active" : "backup",
1777 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
Eric Dumazet4740d632014-07-15 06:56:55 -07001779 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
Eric Dumazet85741712014-07-15 06:56:56 -07001781 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
dingtianhong00503b62014-01-25 13:00:29 +08001783 if (!all && (!bond->params.fail_over_mac ||
Veaceslav Falico01844092014-05-15 21:39:55 +02001784 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
dingtianhong844223a2014-01-02 09:13:16 +08001785 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001786 bond_has_slaves(bond))
Veaceslav Falico76444f52014-07-15 19:35:58 +02001787 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",
1788 slave_dev->name, slave->perm_hwaddr,
1789 bond_dev->name, slave_dev->name);
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001790 }
1791
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02001792 if (rtnl_dereference(bond->primary_slave) == slave)
1793 RCU_INIT_POINTER(bond->primary_slave, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +02001795 if (oldcurrent == slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 bond_change_active_slave(bond, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
Holger Eitzenberger58402052008-12-09 23:07:13 -08001798 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 /* Must be called only after the slave has been
1800 * detached from the list and the curr_active_slave
1801 * has been cleared (if our_slave == old_current),
1802 * but before a new active slave is selected.
1803 */
1804 bond_alb_deinit_slave(bond, slave);
1805 }
1806
nikolay@redhat.com08963412013-02-18 14:09:42 +00001807 if (all) {
Paul E. McKenney36708b82013-12-09 15:19:53 -08001808 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
nikolay@redhat.com08963412013-02-18 14:09:42 +00001809 } else if (oldcurrent == slave) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001810 /* Note that we hold RTNL over this sequence, so there
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001811 * is no concern that another slave add/remove event
1812 * will interfere.
1813 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 bond_select_active_slave(bond);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001815 }
1816
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001817 if (!bond_has_slaves(bond)) {
Jay Vosburghff59c452006-03-27 13:27:43 -08001818 bond_set_carrier(bond);
Jiri Pirko409cc1f2013-01-30 11:08:11 +01001819 eth_hw_addr_random(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 }
1821
Neil Hormane843fa52010-10-13 16:01:50 +00001822 unblock_netpoll_tx();
nikolay@redhat.com278b2082013-08-01 16:54:51 +02001823 synchronize_rcu();
Nikolay Aleksandrovee6154e2014-02-26 14:20:30 +01001824 bond->slave_cnt--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001826 if (!bond_has_slaves(bond)) {
Shlomo Pongratz2af73d42012-04-03 22:56:19 +00001827 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
Veaceslav Falico80028ea2013-03-06 07:10:32 +00001828 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1829 }
Shlomo Pongratz2af73d42012-04-03 22:56:19 +00001830
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001831 bond_compute_features(bond);
1832 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1833 (old_features & NETIF_F_VLAN_CHALLENGED))
Veaceslav Falico76444f52014-07-15 19:35:58 +02001834 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1835 slave_dev->name, bond_dev->name);
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001836
nikolay@redhat.com1ff412a2013-08-06 12:40:15 +02001837 vlan_vids_del_by_dev(slave_dev, bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001839 /* If the mode uses primary, then this case was handled above by
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001840 * bond_change_active_slave(..., NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 */
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02001842 if (!bond_uses_primary(bond)) {
Neil Horman5a0068d2013-09-27 12:22:15 -04001843 /* unset promiscuity level from slave
1844 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1845 * of the IFF_PROMISC flag in the bond_dev, but we need the
1846 * value of that flag before that change, as that was the value
1847 * when this slave was attached, so we cache at the start of the
1848 * function and use it here. Same goes for ALLMULTI below
1849 */
1850 if (old_flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 dev_set_promiscuity(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
1853 /* unset allmulti level from slave */
Neil Horman5a0068d2013-09-27 12:22:15 -04001854 if (old_flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 dev_set_allmulti(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001857 bond_hw_addr_flush(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 }
1859
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001860 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -07001861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 /* close slave before restoring its mac address */
1863 dev_close(slave_dev);
1864
dingtianhong00503b62014-01-25 13:00:29 +08001865 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
Veaceslav Falico01844092014-05-15 21:39:55 +02001866 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
Moni Shoua2ab82852007-10-09 19:43:39 -07001867 /* restore original ("permanent") mac address */
Joe Perchesada0f862014-02-15 16:02:17 -08001868 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
Moni Shoua2ab82852007-10-09 19:43:39 -07001869 addr.sa_family = slave_dev->type;
1870 dev_set_mac_address(slave_dev, &addr);
1871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001873 dev_set_mtu(slave_dev, slave->original_mtu);
1874
Jiri Pirko2d7011c2011-03-16 08:46:43 +00001875 slave_dev->priv_flags &= ~IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
dingtianhong3fdddd82014-05-12 15:08:43 +08001877 bond_free_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001879 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880}
1881
nikolay@redhat.com08963412013-02-18 14:09:42 +00001882/* A wrapper used because of ndo_del_link */
1883int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1884{
1885 return __bond_release_one(bond_dev, slave_dev, false);
1886}
1887
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001888/* First release a slave and then destroy the bond if no more slaves are left.
1889 * Must be under rtnl_lock when this function is called.
1890 */
stephen hemminger26d8ee72010-10-15 05:09:34 +00001891static int bond_release_and_destroy(struct net_device *bond_dev,
1892 struct net_device *slave_dev)
Moni Shouad90a1622007-10-09 19:43:43 -07001893{
Wang Chen454d7c92008-11-12 23:37:49 -08001894 struct bonding *bond = netdev_priv(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07001895 int ret;
1896
1897 ret = bond_release(bond_dev, slave_dev);
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001898 if (ret == 0 && !bond_has_slaves(bond)) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001899 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
Veaceslav Falico76444f52014-07-15 19:35:58 +02001900 netdev_info(bond_dev, "Destroying bond %s\n",
1901 bond_dev->name);
Stephen Hemminger9e716262009-06-12 19:02:47 +00001902 unregister_netdevice(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07001903 }
1904 return ret;
1905}
1906
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1908{
Wang Chen454d7c92008-11-12 23:37:49 -08001909 struct bonding *bond = netdev_priv(bond_dev);
Moni Shoua69a23382015-02-03 16:48:30 +02001910 bond_fill_ifbond(bond, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 return 0;
1912}
1913
1914static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1915{
Wang Chen454d7c92008-11-12 23:37:49 -08001916 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001917 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001918 int i = 0, res = -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001921 bond_for_each_slave(bond, slave, iter) {
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001922 if (i++ == (int)info->slave_id) {
Eric Dumazet689c96c2009-04-23 03:39:04 +00001923 res = 0;
Moni Shoua69a23382015-02-03 16:48:30 +02001924 bond_fill_ifslave(slave, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 break;
1926 }
1927 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
Eric Dumazet689c96c2009-04-23 03:39:04 +00001929 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930}
1931
1932/*-------------------------------- Monitoring -------------------------------*/
1933
Eric Dumazet4740d632014-07-15 06:56:55 -07001934/* called with rcu_read_lock() */
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001935static int bond_miimon_inspect(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936{
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001937 int link_state, commit = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001938 struct list_head *iter;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001939 struct slave *slave;
Jiri Pirko41f89102009-04-24 03:57:29 +00001940 bool ignore_updelay;
1941
Eric Dumazet4740d632014-07-15 06:56:55 -07001942 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
dingtianhong4cb4f972013-12-13 10:19:39 +08001944 bond_for_each_slave_rcu(bond, slave, iter) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001945 slave->new_link = BOND_LINK_NOCHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001947 link_state = bond_check_dev_link(bond, slave->dev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948
1949 switch (slave->link) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001950 case BOND_LINK_UP:
1951 if (link_state)
1952 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
Moni Shoua69a23382015-02-03 16:48:30 +02001954 bond_set_slave_link_state(slave, BOND_LINK_FAIL);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001955 slave->delay = bond->params.downdelay;
1956 if (slave->delay) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001957 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1958 (BOND_MODE(bond) ==
1959 BOND_MODE_ACTIVEBACKUP) ?
1960 (bond_is_active_slave(slave) ?
1961 "active " : "backup ") : "",
1962 slave->dev->name,
1963 bond->params.downdelay * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001965 /*FALLTHRU*/
1966 case BOND_LINK_FAIL:
1967 if (link_state) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001968 /* recovered before downdelay expired */
Moni Shoua69a23382015-02-03 16:48:30 +02001969 bond_set_slave_link_state(slave, BOND_LINK_UP);
Veaceslav Falico8e603462014-02-18 07:48:46 +01001970 slave->last_link_up = jiffies;
Veaceslav Falico76444f52014-07-15 19:35:58 +02001971 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1972 (bond->params.downdelay - slave->delay) *
1973 bond->params.miimon,
1974 slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001975 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001977
1978 if (slave->delay <= 0) {
1979 slave->new_link = BOND_LINK_DOWN;
1980 commit++;
1981 continue;
1982 }
1983
1984 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001987 case BOND_LINK_DOWN:
1988 if (!link_state)
1989 continue;
1990
Moni Shoua69a23382015-02-03 16:48:30 +02001991 bond_set_slave_link_state(slave, BOND_LINK_BACK);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001992 slave->delay = bond->params.updelay;
1993
1994 if (slave->delay) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001995 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1996 slave->dev->name,
1997 ignore_updelay ? 0 :
1998 bond->params.updelay *
1999 bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002001 /*FALLTHRU*/
2002 case BOND_LINK_BACK:
2003 if (!link_state) {
Moni Shoua69a23382015-02-03 16:48:30 +02002004 bond_set_slave_link_state(slave,
2005 BOND_LINK_DOWN);
Veaceslav Falico76444f52014-07-15 19:35:58 +02002006 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2007 (bond->params.updelay - slave->delay) *
2008 bond->params.miimon,
2009 slave->dev->name);
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002010
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002011 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002013
Jiri Pirko41f89102009-04-24 03:57:29 +00002014 if (ignore_updelay)
2015 slave->delay = 0;
2016
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002017 if (slave->delay <= 0) {
2018 slave->new_link = BOND_LINK_UP;
2019 commit++;
Jiri Pirko41f89102009-04-24 03:57:29 +00002020 ignore_updelay = false;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002021 continue;
2022 }
2023
2024 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 break;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002026 }
2027 }
2028
2029 return commit;
2030}
2031
2032static void bond_miimon_commit(struct bonding *bond)
2033{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002034 struct list_head *iter;
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002035 struct slave *slave, *primary;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002036
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002037 bond_for_each_slave(bond, slave, iter) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002038 switch (slave->new_link) {
2039 case BOND_LINK_NOCHANGE:
2040 continue;
2041
2042 case BOND_LINK_UP:
Moni Shoua69a23382015-02-03 16:48:30 +02002043 bond_set_slave_link_state(slave, BOND_LINK_UP);
Veaceslav Falico8e603462014-02-18 07:48:46 +01002044 slave->last_link_up = jiffies;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002045
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002046 primary = rtnl_dereference(bond->primary_slave);
Veaceslav Falico01844092014-05-15 21:39:55 +02002047 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002048 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002049 bond_set_backup_slave(slave);
Veaceslav Falico01844092014-05-15 21:39:55 +02002050 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002051 /* make it immediately active */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002052 bond_set_active_slave(slave);
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002053 } else if (slave != primary) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002054 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002055 bond_set_backup_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002056 }
2057
Veaceslav Falico76444f52014-07-15 19:35:58 +02002058 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2059 slave->dev->name,
2060 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2061 slave->duplex ? "full" : "half");
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002062
2063 /* notify ad that the link status has changed */
Veaceslav Falico01844092014-05-15 21:39:55 +02002064 if (BOND_MODE(bond) == BOND_MODE_8023AD)
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002065 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2066
Holger Eitzenberger58402052008-12-09 23:07:13 -08002067 if (bond_is_lb(bond))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002068 bond_alb_handle_link_change(bond, slave,
2069 BOND_LINK_UP);
2070
Mahesh Bandewaree637712014-10-04 17:45:01 -07002071 if (BOND_MODE(bond) == BOND_MODE_XOR)
2072 bond_update_slave_arr(bond, NULL);
2073
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002074 if (!bond->curr_active_slave || slave == primary)
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002075 goto do_failover;
2076
2077 continue;
2078
2079 case BOND_LINK_DOWN:
Jay Vosburghfba4acd2008-10-30 17:41:14 -07002080 if (slave->link_failure_count < UINT_MAX)
2081 slave->link_failure_count++;
2082
Moni Shoua69a23382015-02-03 16:48:30 +02002083 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002084
Veaceslav Falico01844092014-05-15 21:39:55 +02002085 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2086 BOND_MODE(bond) == BOND_MODE_8023AD)
dingtianhong5e5b0662014-02-26 11:05:22 +08002087 bond_set_slave_inactive_flags(slave,
2088 BOND_SLAVE_NOTIFY_NOW);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002089
Veaceslav Falico76444f52014-07-15 19:35:58 +02002090 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2091 slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002092
Veaceslav Falico01844092014-05-15 21:39:55 +02002093 if (BOND_MODE(bond) == BOND_MODE_8023AD)
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002094 bond_3ad_handle_link_change(slave,
2095 BOND_LINK_DOWN);
2096
Jiri Pirkoae63e802009-05-27 05:42:36 +00002097 if (bond_is_lb(bond))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002098 bond_alb_handle_link_change(bond, slave,
2099 BOND_LINK_DOWN);
2100
Mahesh Bandewaree637712014-10-04 17:45:01 -07002101 if (BOND_MODE(bond) == BOND_MODE_XOR)
2102 bond_update_slave_arr(bond, NULL);
2103
Eric Dumazet4740d632014-07-15 06:56:55 -07002104 if (slave == rcu_access_pointer(bond->curr_active_slave))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002105 goto do_failover;
2106
2107 continue;
2108
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 default:
Veaceslav Falico76444f52014-07-15 19:35:58 +02002110 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2111 slave->new_link, slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002112 slave->new_link = BOND_LINK_NOCHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002114 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 }
2116
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002117do_failover:
Neil Hormane843fa52010-10-13 16:01:50 +00002118 block_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 bond_select_active_slave(bond);
Neil Hormane843fa52010-10-13 16:01:50 +00002120 unblock_netpoll_tx();
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002121 }
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002122
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002123 bond_set_carrier(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124}
2125
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002126/* bond_mii_monitor
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002127 *
2128 * Really a wrapper that splits the mii monitor into two phases: an
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002129 * inspection, then (if inspection indicates something needs to be done)
2130 * an acquisition of appropriate locks followed by a commit phase to
2131 * implement whatever link state changes are indicated.
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002132 */
stephen hemminger6da67d22013-12-30 10:43:41 -08002133static void bond_mii_monitor(struct work_struct *work)
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002134{
2135 struct bonding *bond = container_of(work, struct bonding,
2136 mii_work.work);
Ben Hutchingsad246c92011-04-26 15:25:52 +00002137 bool should_notify_peers = false;
David S. Miller1f2cd842013-10-28 00:11:22 -04002138 unsigned long delay;
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002139
David S. Miller1f2cd842013-10-28 00:11:22 -04002140 delay = msecs_to_jiffies(bond->params.miimon);
2141
2142 if (!bond_has_slaves(bond))
dingtianhong6b6c5262013-10-24 11:09:03 +08002143 goto re_arm;
dingtianhong6b6c5262013-10-24 11:09:03 +08002144
dingtianhong4cb4f972013-12-13 10:19:39 +08002145 rcu_read_lock();
2146
Ben Hutchingsad246c92011-04-26 15:25:52 +00002147 should_notify_peers = bond_should_notify_peers(bond);
2148
David S. Miller1f2cd842013-10-28 00:11:22 -04002149 if (bond_miimon_inspect(bond)) {
dingtianhong4cb4f972013-12-13 10:19:39 +08002150 rcu_read_unlock();
David S. Miller1f2cd842013-10-28 00:11:22 -04002151
2152 /* Race avoidance with bond_close cancel of workqueue */
2153 if (!rtnl_trylock()) {
David S. Miller1f2cd842013-10-28 00:11:22 -04002154 delay = 1;
2155 should_notify_peers = false;
2156 goto re_arm;
2157 }
2158
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002159 bond_miimon_commit(bond);
2160
David S. Miller1f2cd842013-10-28 00:11:22 -04002161 rtnl_unlock(); /* might sleep, hold no other locks */
dingtianhong4cb4f972013-12-13 10:19:39 +08002162 } else
2163 rcu_read_unlock();
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002164
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002165re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00002166 if (bond->params.miimon)
David S. Miller1f2cd842013-10-28 00:11:22 -04002167 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2168
David S. Miller1f2cd842013-10-28 00:11:22 -04002169 if (should_notify_peers) {
2170 if (!rtnl_trylock())
2171 return;
2172 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2173 rtnl_unlock();
2174 }
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002175}
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002176
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002177static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002178{
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002179 struct net_device *upper;
2180 struct list_head *iter;
2181 bool ret = false;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002182
Andy Gospodarekeaddcd72012-03-22 16:14:29 +00002183 if (ip == bond_confirm_addr(bond->dev, 0, ip))
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002184 return true;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002185
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002186 rcu_read_lock();
Veaceslav Falico2f268f12013-09-25 09:20:07 +02002187 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002188 if (ip == bond_confirm_addr(upper, 0, ip)) {
2189 ret = true;
2190 break;
2191 }
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002192 }
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002193 rcu_read_unlock();
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002194
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002195 return ret;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002196}
2197
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002198/* We go to the (large) trouble of VLAN tagging ARP frames because
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002199 * switches in VLAN mode (especially if ports are configured as
2200 * "native" to a VLAN) might not pass non-tagged frames.
2201 */
dingtianhongfbd929f2014-03-25 17:44:43 +08002202static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2203 __be32 dest_ip, __be32 src_ip,
Vlad Yasevich44a40852014-05-16 17:20:38 -04002204 struct bond_vlan_tag *tags)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002205{
2206 struct sk_buff *skb;
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002207 struct bond_vlan_tag *outer_tag = tags;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002208
Veaceslav Falico76444f52014-07-15 19:35:58 +02002209 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2210 arp_op, slave_dev->name, &dest_ip, &src_ip);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002211
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002212 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2213 NULL, slave_dev->dev_addr, NULL);
2214
2215 if (!skb) {
dingtianhong4873ac32014-03-25 17:44:44 +08002216 net_err_ratelimited("ARP packet allocation failed\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002217 return;
2218 }
dingtianhongfbd929f2014-03-25 17:44:43 +08002219
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002220 if (!tags || tags->vlan_proto == VLAN_N_VID)
2221 goto xmit;
2222
2223 tags++;
2224
Vlad Yasevich44a40852014-05-16 17:20:38 -04002225 /* Go through all the tags backwards and add them to the packet */
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002226 while (tags->vlan_proto != VLAN_N_VID) {
2227 if (!tags->vlan_id) {
2228 tags++;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002229 continue;
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002230 }
Vlad Yasevich44a40852014-05-16 17:20:38 -04002231
Veaceslav Falico76444f52014-07-15 19:35:58 +02002232 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002233 ntohs(outer_tag->vlan_proto), tags->vlan_id);
Jiri Pirko62749e22014-11-19 14:04:58 +01002234 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2235 tags->vlan_id);
Vlad Yasevich44a40852014-05-16 17:20:38 -04002236 if (!skb) {
2237 net_err_ratelimited("failed to insert inner VLAN tag\n");
2238 return;
2239 }
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002240
2241 tags++;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002242 }
2243 /* Set the outer tag */
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002244 if (outer_tag->vlan_id) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002245 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002246 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
Jiri Pirkob4bef1b2014-11-19 14:04:57 +01002247 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2248 outer_tag->vlan_id);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002249 }
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002250
2251xmit:
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002252 arp_xmit(skb);
2253}
2254
Vlad Yasevich44a40852014-05-16 17:20:38 -04002255/* Validate the device path between the @start_dev and the @end_dev.
2256 * The path is valid if the @end_dev is reachable through device
2257 * stacking.
2258 * When the path is validated, collect any vlan information in the
2259 * path.
2260 */
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002261struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2262 struct net_device *end_dev,
2263 int level)
Vlad Yasevich44a40852014-05-16 17:20:38 -04002264{
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002265 struct bond_vlan_tag *tags;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002266 struct net_device *upper;
2267 struct list_head *iter;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002268
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002269 if (start_dev == end_dev) {
2270 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2271 if (!tags)
2272 return ERR_PTR(-ENOMEM);
2273 tags[level].vlan_proto = VLAN_N_VID;
2274 return tags;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002275 }
2276
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002277 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2278 tags = bond_verify_device_path(upper, end_dev, level + 1);
2279 if (IS_ERR_OR_NULL(tags)) {
2280 if (IS_ERR(tags))
2281 return tags;
2282 continue;
2283 }
2284 if (is_vlan_dev(upper)) {
2285 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2286 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2287 }
2288
2289 return tags;
2290 }
2291
2292 return NULL;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002293}
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002294
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2296{
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002297 struct rtable *rt;
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002298 struct bond_vlan_tag *tags;
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002299 __be32 *targets = bond->params.arp_targets, addr;
dingtianhongfbd929f2014-03-25 17:44:43 +08002300 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002302 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002303 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002304 tags = NULL;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002305
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002306 /* Find out through which dev should the packet go */
David S. Miller78fbfd82011-03-12 00:00:52 -05002307 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2308 RTO_ONLINK, 0);
David S. Millerb23dd4f2011-03-02 14:31:35 -08002309 if (IS_ERR(rt)) {
Veaceslav Falico28572762014-02-28 12:39:19 +01002310 /* there's no route to target - try to send arp
2311 * probe to generate any traffic (arp_validate=0)
2312 */
dingtianhong4873ac32014-03-25 17:44:44 +08002313 if (bond->params.arp_validate)
2314 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2315 bond->dev->name,
2316 &targets[i]);
Vlad Yasevich44a40852014-05-16 17:20:38 -04002317 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2318 0, tags);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002319 continue;
2320 }
2321
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002322 /* bond device itself */
2323 if (rt->dst.dev == bond->dev)
2324 goto found;
2325
2326 rcu_read_lock();
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002327 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002328 rcu_read_unlock();
2329
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002330 if (!IS_ERR_OR_NULL(tags))
Vlad Yasevich44a40852014-05-16 17:20:38 -04002331 goto found;
2332
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002333 /* Not our device - skip */
Veaceslav Falico76444f52014-07-15 19:35:58 +02002334 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2335 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
Veaceslav Falico3e325822013-08-28 23:25:16 +02002336
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002337 ip_rt_put(rt);
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002338 continue;
2339
2340found:
2341 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2342 ip_rt_put(rt);
2343 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
Vlad Yasevich44a40852014-05-16 17:20:38 -04002344 addr, tags);
Dan Carpentera67eed52014-07-25 15:21:21 +03002345 kfree(tags);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002346 }
2347}
2348
Al Virod3bb52b2007-08-22 20:06:58 -04002349static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002350{
Veaceslav Falico8599b522013-06-24 11:49:34 +02002351 int i;
2352
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002353 if (!sip || !bond_has_this_ip(bond, tip)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002354 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2355 &sip, &tip);
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002356 return;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002357 }
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002358
Veaceslav Falico8599b522013-06-24 11:49:34 +02002359 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2360 if (i == -1) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002361 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2362 &sip);
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002363 return;
2364 }
Veaceslav Falico49f17de2014-02-18 07:48:47 +01002365 slave->last_rx = jiffies;
Veaceslav Falico8599b522013-06-24 11:49:34 +02002366 slave->target_last_arp_rx[i] = jiffies;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002367}
2368
nikolay@redhat.com5bb9e0b2013-09-07 00:00:26 +02002369int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2370 struct slave *slave)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002371{
Eric Dumazetde063b72012-06-11 19:23:07 +00002372 struct arphdr *arp = (struct arphdr *)skb->data;
Veaceslav Falico010d3c32014-02-20 12:07:57 +01002373 struct slave *curr_active_slave;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002374 unsigned char *arp_ptr;
Al Virod3bb52b2007-08-22 20:06:58 -04002375 __be32 sip, tip;
Veaceslav Falicof2cb6912014-02-18 07:48:42 +01002376 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002377
Veaceslav Falicof2cb6912014-02-18 07:48:42 +01002378 if (!slave_do_arp_validate(bond, slave)) {
dingtianhongbedabf92014-05-07 22:10:20 +08002379 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2380 !slave_do_arp_validate_only(bond))
Veaceslav Falico49f17de2014-02-18 07:48:47 +01002381 slave->last_rx = jiffies;
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00002382 return RX_HANDLER_ANOTHER;
Veaceslav Falicof2cb6912014-02-18 07:48:42 +01002383 } else if (!is_arp) {
2384 return RX_HANDLER_ANOTHER;
2385 }
Veaceslav Falico2c146102013-06-24 11:49:31 +02002386
Eric Dumazetde063b72012-06-11 19:23:07 +00002387 alen = arp_hdr_len(bond->dev);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002388
Veaceslav Falico76444f52014-07-15 19:35:58 +02002389 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2390 skb->dev->name);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002391
Eric Dumazetde063b72012-06-11 19:23:07 +00002392 if (alen > skb_headlen(skb)) {
2393 arp = kmalloc(alen, GFP_ATOMIC);
2394 if (!arp)
2395 goto out_unlock;
2396 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2397 goto out_unlock;
2398 }
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002399
Jiri Pirko3aba8912011-04-19 03:48:16 +00002400 if (arp->ar_hln != bond->dev->addr_len ||
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002401 skb->pkt_type == PACKET_OTHERHOST ||
2402 skb->pkt_type == PACKET_LOOPBACK ||
2403 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2404 arp->ar_pro != htons(ETH_P_IP) ||
2405 arp->ar_pln != 4)
2406 goto out_unlock;
2407
2408 arp_ptr = (unsigned char *)(arp + 1);
Jiri Pirko3aba8912011-04-19 03:48:16 +00002409 arp_ptr += bond->dev->addr_len;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002410 memcpy(&sip, arp_ptr, 4);
Jiri Pirko3aba8912011-04-19 03:48:16 +00002411 arp_ptr += 4 + bond->dev->addr_len;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002412 memcpy(&tip, arp_ptr, 4);
2413
Veaceslav Falico76444f52014-07-15 19:35:58 +02002414 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2415 slave->dev->name, bond_slave_state(slave),
2416 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2417 &sip, &tip);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002418
Veaceslav Falico010d3c32014-02-20 12:07:57 +01002419 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2420
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002421 /* Backup slaves won't see the ARP reply, but do come through
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002422 * here for each ARP probe (so we swap the sip/tip to validate
2423 * the probe). In a "redundant switch, common router" type of
2424 * configuration, the ARP probe will (hopefully) travel from
2425 * the active, through one switch, the router, then the other
2426 * switch before reaching the backup.
Veaceslav Falicoaeea64a2013-06-24 11:49:32 +02002427 *
2428 * We 'trust' the arp requests if there is an active slave and
2429 * it received valid arp reply(s) after it became active. This
2430 * is done to avoid endless looping when we can't reach the
2431 * arp_ip_target and fool ourselves with our own arp requests.
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002432 */
Veaceslav Falico010d3c32014-02-20 12:07:57 +01002433
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002434 if (bond_is_active_slave(slave))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002435 bond_validate_arp(bond, slave, sip, tip);
Veaceslav Falico010d3c32014-02-20 12:07:57 +01002436 else if (curr_active_slave &&
2437 time_after(slave_last_rx(bond, curr_active_slave),
2438 curr_active_slave->last_link_up))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002439 bond_validate_arp(bond, slave, tip, sip);
2440
2441out_unlock:
Eric Dumazetde063b72012-06-11 19:23:07 +00002442 if (arp != (struct arphdr *)skb->data)
2443 kfree(arp);
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00002444 return RX_HANDLER_ANOTHER;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002445}
2446
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002447/* function to verify if we're in the arp_interval timeslice, returns true if
2448 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2449 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2450 */
2451static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2452 int mod)
2453{
2454 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2455
2456 return time_in_range(jiffies,
2457 last_act - delta_in_ticks,
2458 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2459}
2460
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002461/* This function is called regularly to monitor each slave's link
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 * ensuring that traffic is being sent and received when arp monitoring
2463 * is used in load-balancing mode. if the adapter has been dormant, then an
2464 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2465 * arp monitoring in active backup mode.
2466 */
stephen hemminger6da67d22013-12-30 10:43:41 -08002467static void bond_loadbalance_arp_mon(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468{
Jay Vosburgh1b76b312007-10-17 17:37:45 -07002469 struct bonding *bond = container_of(work, struct bonding,
2470 arp_work.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 struct slave *slave, *oldcurrent;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002472 struct list_head *iter;
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002473 int do_failover = 0, slave_state_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
David S. Miller1f2cd842013-10-28 00:11:22 -04002475 if (!bond_has_slaves(bond))
dingtianhong7f1bb572013-10-24 11:09:17 +08002476 goto re_arm;
dingtianhong7f1bb572013-10-24 11:09:17 +08002477
dingtianhong2e52f4f2013-12-13 10:19:50 +08002478 rcu_read_lock();
2479
Eric Dumazet4740d632014-07-15 06:56:55 -07002480 oldcurrent = rcu_dereference(bond->curr_active_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 /* see if any of the previous devices are up now (i.e. they have
2482 * xmt and rcv traffic). the curr_active_slave does not come into
Veaceslav Falico8e603462014-02-18 07:48:46 +01002483 * the picture unless it is null. also, slave->last_link_up is not
2484 * needed here because we send an arp on each slave and give a slave
2485 * as long as it needs to get the tx/rx within the delta.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 * TODO: what about up/down delay in arp mode? it wasn't here before
2487 * so it can wait
2488 */
dingtianhong2e52f4f2013-12-13 10:19:50 +08002489 bond_for_each_slave_rcu(bond, slave, iter) {
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002490 unsigned long trans_start = dev_trans_start(slave->dev);
2491
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 if (slave->link != BOND_LINK_UP) {
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002493 if (bond_time_in_interval(bond, trans_start, 1) &&
Veaceslav Falico49f17de2014-02-18 07:48:47 +01002494 bond_time_in_interval(bond, slave->last_rx, 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
2496 slave->link = BOND_LINK_UP;
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002497 slave_state_changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498
2499 /* primary_slave has no meaning in round-robin
2500 * mode. the window of a slave being up and
2501 * curr_active_slave being null after enslaving
2502 * is closed.
2503 */
2504 if (!oldcurrent) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002505 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2506 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 do_failover = 1;
2508 } else {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002509 netdev_info(bond->dev, "interface %s is now up\n",
2510 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 }
2512 }
2513 } else {
2514 /* slave->link == BOND_LINK_UP */
2515
2516 /* not all switches will respond to an arp request
2517 * when the source ip is 0, so don't take the link down
2518 * if we don't know our ip yet
2519 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002520 if (!bond_time_in_interval(bond, trans_start, 2) ||
Veaceslav Falico49f17de2014-02-18 07:48:47 +01002521 !bond_time_in_interval(bond, slave->last_rx, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
2523 slave->link = BOND_LINK_DOWN;
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002524 slave_state_changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002526 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
Veaceslav Falico76444f52014-07-15 19:35:58 +02002529 netdev_info(bond->dev, "interface %s is now down\n",
2530 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002532 if (slave == oldcurrent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 do_failover = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 }
2535 }
2536
2537 /* note: if switch is in round-robin mode, all links
2538 * must tx arp to ensure all links rx an arp - otherwise
2539 * links may oscillate or not come up at all; if switch is
2540 * in something like xor mode, there is nothing we can
2541 * do - all replies will be rx'ed on same link causing slaves
2542 * to be unstable during low/no traffic periods
2543 */
Veaceslav Falicob6adc612014-05-15 21:39:57 +02002544 if (bond_slave_is_up(slave))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 bond_arp_send_all(bond, slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 }
2547
dingtianhong2e52f4f2013-12-13 10:19:50 +08002548 rcu_read_unlock();
2549
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002550 if (do_failover || slave_state_changed) {
dingtianhong2e52f4f2013-12-13 10:19:50 +08002551 if (!rtnl_trylock())
2552 goto re_arm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002554 if (slave_state_changed) {
2555 bond_slave_state_change(bond);
Mahesh Bandewaree637712014-10-04 17:45:01 -07002556 if (BOND_MODE(bond) == BOND_MODE_XOR)
2557 bond_update_slave_arr(bond, NULL);
Nikolay Aleksandrovb8e45002014-11-18 15:14:44 +01002558 }
2559 if (do_failover) {
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002560 block_netpoll_tx();
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002561 bond_select_active_slave(bond);
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002562 unblock_netpoll_tx();
2563 }
dingtianhong2e52f4f2013-12-13 10:19:50 +08002564 rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 }
2566
2567re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00002568 if (bond->params.arp_interval)
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002569 queue_delayed_work(bond->wq, &bond->arp_work,
2570 msecs_to_jiffies(bond->params.arp_interval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571}
2572
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002573/* Called to inspect slaves for active-backup mode ARP monitor link state
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002574 * changes. Sets new_link in slaves to specify what action should take
2575 * place for the slave. Returns 0 if no changes are found, >0 if changes
2576 * to link states must be committed.
2577 *
Nikolay Aleksandrov8c0bc552014-09-11 22:49:28 +02002578 * Called with rcu_read_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002580static int bond_ab_arp_inspect(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581{
Veaceslav Falicodef44602013-08-03 03:50:35 +02002582 unsigned long trans_start, last_rx;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002583 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002584 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002585 int commit = 0;
Jiri Bohacda210f52012-08-30 12:02:47 +00002586
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002587 bond_for_each_slave_rcu(bond, slave, iter) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002588 slave->new_link = BOND_LINK_NOCHANGE;
Veaceslav Falicodef44602013-08-03 03:50:35 +02002589 last_rx = slave_last_rx(bond, slave);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002590
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 if (slave->link != BOND_LINK_UP) {
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002592 if (bond_time_in_interval(bond, last_rx, 1)) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002593 slave->new_link = BOND_LINK_UP;
2594 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 }
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002596 continue;
2597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002599 /* Give slaves 2*delta after being enslaved or made
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002600 * active. This avoids bouncing, as the last receive
2601 * times need a full ARP monitor cycle to be updated.
2602 */
Veaceslav Falico8e603462014-02-18 07:48:46 +01002603 if (bond_time_in_interval(bond, slave->last_link_up, 2))
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002604 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002606 /* Backup slave is down if:
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002607 * - No current_arp_slave AND
2608 * - more than 3*delta since last receive AND
2609 * - the bond has an IP address
2610 *
2611 * Note: a non-null current_arp_slave indicates
2612 * the curr_active_slave went down and we are
2613 * searching for a new one; under this condition
2614 * we only take the curr_active_slave down - this
2615 * gives each slave a chance to tx/rx traffic
2616 * before being taken out
2617 */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002618 if (!bond_is_active_slave(slave) &&
Eric Dumazet85741712014-07-15 06:56:56 -07002619 !rcu_access_pointer(bond->current_arp_slave) &&
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002620 !bond_time_in_interval(bond, last_rx, 3)) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002621 slave->new_link = BOND_LINK_DOWN;
2622 commit++;
2623 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002625 /* Active slave is down if:
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002626 * - more than 2*delta since transmitting OR
2627 * - (more than 2*delta since receive AND
2628 * the bond has an IP address)
2629 */
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002630 trans_start = dev_trans_start(slave->dev);
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002631 if (bond_is_active_slave(slave) &&
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002632 (!bond_time_in_interval(bond, trans_start, 2) ||
2633 !bond_time_in_interval(bond, last_rx, 2))) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002634 slave->new_link = BOND_LINK_DOWN;
2635 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 }
2637 }
2638
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002639 return commit;
2640}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002642/* Called to commit link state changes noted by inspection step of
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002643 * active-backup mode ARP monitor.
2644 *
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002645 * Called with RTNL hold.
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002646 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002647static void bond_ab_arp_commit(struct bonding *bond)
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002648{
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002649 unsigned long trans_start;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002650 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002651 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002653 bond_for_each_slave(bond, slave, iter) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002654 switch (slave->new_link) {
2655 case BOND_LINK_NOCHANGE:
2656 continue;
2657
2658 case BOND_LINK_UP:
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002659 trans_start = dev_trans_start(slave->dev);
Eric Dumazet4740d632014-07-15 06:56:55 -07002660 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2661 (!rtnl_dereference(bond->curr_active_slave) &&
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002662 bond_time_in_interval(bond, trans_start, 1))) {
Eric Dumazet85741712014-07-15 06:56:56 -07002663 struct slave *current_arp_slave;
2664
2665 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
Moni Shoua69a23382015-02-03 16:48:30 +02002666 bond_set_slave_link_state(slave, BOND_LINK_UP);
Eric Dumazet85741712014-07-15 06:56:56 -07002667 if (current_arp_slave) {
Veaceslav Falico5a430972012-04-05 03:47:43 +00002668 bond_set_slave_inactive_flags(
Eric Dumazet85741712014-07-15 06:56:56 -07002669 current_arp_slave,
dingtianhong5e5b0662014-02-26 11:05:22 +08002670 BOND_SLAVE_NOTIFY_NOW);
Eric Dumazet85741712014-07-15 06:56:56 -07002671 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
Veaceslav Falico5a430972012-04-05 03:47:43 +00002672 }
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002673
Veaceslav Falico76444f52014-07-15 19:35:58 +02002674 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2675 slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002676
Eric Dumazet4740d632014-07-15 06:56:55 -07002677 if (!rtnl_dereference(bond->curr_active_slave) ||
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002678 slave == rtnl_dereference(bond->primary_slave))
Jiri Pirkob9f60252009-08-31 11:09:38 +00002679 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002680
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002681 }
2682
Jiri Pirkob9f60252009-08-31 11:09:38 +00002683 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002684
2685 case BOND_LINK_DOWN:
2686 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688
Moni Shoua69a23382015-02-03 16:48:30 +02002689 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
dingtianhong5e5b0662014-02-26 11:05:22 +08002690 bond_set_slave_inactive_flags(slave,
2691 BOND_SLAVE_NOTIFY_NOW);
Jiri Pirkob9f60252009-08-31 11:09:38 +00002692
Veaceslav Falico76444f52014-07-15 19:35:58 +02002693 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2694 slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002695
Eric Dumazet4740d632014-07-15 06:56:55 -07002696 if (slave == rtnl_dereference(bond->curr_active_slave)) {
Eric Dumazet85741712014-07-15 06:56:56 -07002697 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
Jiri Pirkob9f60252009-08-31 11:09:38 +00002698 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002699 }
Jiri Pirkob9f60252009-08-31 11:09:38 +00002700
2701 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002702
2703 default:
Veaceslav Falico76444f52014-07-15 19:35:58 +02002704 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2705 slave->new_link, slave->dev->name);
Jiri Pirkob9f60252009-08-31 11:09:38 +00002706 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708
Jiri Pirkob9f60252009-08-31 11:09:38 +00002709do_failover:
Neil Hormane843fa52010-10-13 16:01:50 +00002710 block_netpoll_tx();
Jiri Pirkob9f60252009-08-31 11:09:38 +00002711 bond_select_active_slave(bond);
Neil Hormane843fa52010-10-13 16:01:50 +00002712 unblock_netpoll_tx();
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002713 }
2714
2715 bond_set_carrier(bond);
2716}
2717
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002718/* Send ARP probes for active-backup mode ARP monitor.
dingtianhongb0929912014-02-26 11:05:23 +08002719 *
Nikolay Aleksandrov8c0bc552014-09-11 22:49:28 +02002720 * Called with rcu_read_lock held.
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002721 */
Veaceslav Falicof2ebd472014-01-27 14:37:32 +01002722static bool bond_ab_arp_probe(struct bonding *bond)
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002723{
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002724 struct slave *slave, *before = NULL, *new_slave = NULL,
dingtianhongb0929912014-02-26 11:05:23 +08002725 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2726 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
Veaceslav Falico4087df872013-09-25 09:20:19 +02002727 struct list_head *iter;
2728 bool found = false;
dingtianhongb0929912014-02-26 11:05:23 +08002729 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
Veaceslav Falicof2ebd472014-01-27 14:37:32 +01002730
Veaceslav Falico98b90f22014-01-27 14:37:31 +01002731 if (curr_arp_slave && curr_active_slave)
Veaceslav Falico76444f52014-07-15 19:35:58 +02002732 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2733 curr_arp_slave->dev->name,
2734 curr_active_slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002735
Veaceslav Falico98b90f22014-01-27 14:37:31 +01002736 if (curr_active_slave) {
2737 bond_arp_send_all(bond, curr_active_slave);
dingtianhongb0929912014-02-26 11:05:23 +08002738 return should_notify_rtnl;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002739 }
2740
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 /* if we don't have a curr_active_slave, search for the next available
2742 * backup slave from the current_arp_slave and make it the candidate
2743 * for becoming the curr_active_slave
2744 */
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002745
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002746 if (!curr_arp_slave) {
dingtianhongb0929912014-02-26 11:05:23 +08002747 curr_arp_slave = bond_first_slave_rcu(bond);
2748 if (!curr_arp_slave)
2749 return should_notify_rtnl;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002750 }
2751
dingtianhongb0929912014-02-26 11:05:23 +08002752 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002753
dingtianhongb0929912014-02-26 11:05:23 +08002754 bond_for_each_slave_rcu(bond, slave, iter) {
Veaceslav Falicob6adc612014-05-15 21:39:57 +02002755 if (!found && !before && bond_slave_is_up(slave))
Veaceslav Falico4087df872013-09-25 09:20:19 +02002756 before = slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757
Veaceslav Falicob6adc612014-05-15 21:39:57 +02002758 if (found && !new_slave && bond_slave_is_up(slave))
Veaceslav Falico4087df872013-09-25 09:20:19 +02002759 new_slave = slave;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002760 /* if the link state is up at this point, we
2761 * mark it down - this can happen if we have
2762 * simultaneous link failures and
2763 * reselect_active_interface doesn't make this
2764 * one the current slave so it is still marked
2765 * up when it is actually down
2766 */
Veaceslav Falicob6adc612014-05-15 21:39:57 +02002767 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
Moni Shoua69a23382015-02-03 16:48:30 +02002768 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002769 if (slave->link_failure_count < UINT_MAX)
2770 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771
dingtianhong5e5b0662014-02-26 11:05:22 +08002772 bond_set_slave_inactive_flags(slave,
dingtianhongb0929912014-02-26 11:05:23 +08002773 BOND_SLAVE_NOTIFY_LATER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
Veaceslav Falico76444f52014-07-15 19:35:58 +02002775 netdev_info(bond->dev, "backup interface %s is now down\n",
2776 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 }
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002778 if (slave == curr_arp_slave)
Veaceslav Falico4087df872013-09-25 09:20:19 +02002779 found = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 }
Veaceslav Falico4087df872013-09-25 09:20:19 +02002781
2782 if (!new_slave && before)
2783 new_slave = before;
2784
dingtianhongb0929912014-02-26 11:05:23 +08002785 if (!new_slave)
2786 goto check_state;
Veaceslav Falico4087df872013-09-25 09:20:19 +02002787
Moni Shoua69a23382015-02-03 16:48:30 +02002788 bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
dingtianhongb0929912014-02-26 11:05:23 +08002789 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
Veaceslav Falico4087df872013-09-25 09:20:19 +02002790 bond_arp_send_all(bond, new_slave);
Veaceslav Falico8e603462014-02-18 07:48:46 +01002791 new_slave->last_link_up = jiffies;
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002792 rcu_assign_pointer(bond->current_arp_slave, new_slave);
Veaceslav Falicof2ebd472014-01-27 14:37:32 +01002793
dingtianhongb0929912014-02-26 11:05:23 +08002794check_state:
2795 bond_for_each_slave_rcu(bond, slave, iter) {
2796 if (slave->should_notify) {
2797 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2798 break;
2799 }
2800 }
2801 return should_notify_rtnl;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002802}
2803
stephen hemminger6da67d22013-12-30 10:43:41 -08002804static void bond_activebackup_arp_mon(struct work_struct *work)
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002805{
2806 struct bonding *bond = container_of(work, struct bonding,
2807 arp_work.work);
dingtianhongb0929912014-02-26 11:05:23 +08002808 bool should_notify_peers = false;
2809 bool should_notify_rtnl = false;
David S. Miller1f2cd842013-10-28 00:11:22 -04002810 int delta_in_ticks;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002811
David S. Miller1f2cd842013-10-28 00:11:22 -04002812 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2813
2814 if (!bond_has_slaves(bond))
dingtianhong80b9d232013-10-24 11:09:25 +08002815 goto re_arm;
dingtianhong80b9d232013-10-24 11:09:25 +08002816
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002817 rcu_read_lock();
Ben Hutchingsad246c92011-04-26 15:25:52 +00002818
dingtianhongb0929912014-02-26 11:05:23 +08002819 should_notify_peers = bond_should_notify_peers(bond);
2820
2821 if (bond_ab_arp_inspect(bond)) {
2822 rcu_read_unlock();
2823
David S. Miller1f2cd842013-10-28 00:11:22 -04002824 /* Race avoidance with bond_close flush of workqueue */
2825 if (!rtnl_trylock()) {
David S. Miller1f2cd842013-10-28 00:11:22 -04002826 delta_in_ticks = 1;
2827 should_notify_peers = false;
2828 goto re_arm;
2829 }
2830
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002831 bond_ab_arp_commit(bond);
dingtianhongb0929912014-02-26 11:05:23 +08002832
David S. Miller1f2cd842013-10-28 00:11:22 -04002833 rtnl_unlock();
dingtianhongb0929912014-02-26 11:05:23 +08002834 rcu_read_lock();
David S. Miller1f2cd842013-10-28 00:11:22 -04002835 }
2836
dingtianhongb0929912014-02-26 11:05:23 +08002837 should_notify_rtnl = bond_ab_arp_probe(bond);
2838 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839
2840re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00002841 if (bond->params.arp_interval)
David S. Miller1f2cd842013-10-28 00:11:22 -04002842 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2843
dingtianhongb0929912014-02-26 11:05:23 +08002844 if (should_notify_peers || should_notify_rtnl) {
David S. Miller1f2cd842013-10-28 00:11:22 -04002845 if (!rtnl_trylock())
2846 return;
dingtianhongb0929912014-02-26 11:05:23 +08002847
2848 if (should_notify_peers)
2849 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2850 bond->dev);
2851 if (should_notify_rtnl)
2852 bond_slave_state_notify(bond);
2853
David S. Miller1f2cd842013-10-28 00:11:22 -04002854 rtnl_unlock();
2855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856}
2857
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858/*-------------------------- netdev event handling --------------------------*/
2859
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002860/* Change device name */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861static int bond_event_changename(struct bonding *bond)
2862{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 bond_remove_proc_entry(bond);
2864 bond_create_proc_entry(bond);
Stephen Hemminger7e083842009-06-12 19:02:46 +00002865
Taku Izumif073c7c2010-12-09 15:17:13 +00002866 bond_debug_reregister(bond);
2867
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 return NOTIFY_DONE;
2869}
2870
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002871static int bond_master_netdev_event(unsigned long event,
2872 struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873{
Wang Chen454d7c92008-11-12 23:37:49 -08002874 struct bonding *event_bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875
2876 switch (event) {
2877 case NETDEV_CHANGENAME:
2878 return bond_event_changename(event_bond);
Eric W. Biedermana64d49c2012-07-09 10:51:45 +00002879 case NETDEV_UNREGISTER:
2880 bond_remove_proc_entry(event_bond);
2881 break;
2882 case NETDEV_REGISTER:
2883 bond_create_proc_entry(event_bond);
2884 break;
nikolay@redhat.com6c8d23f2013-09-02 13:51:38 +02002885 case NETDEV_NOTIFY_PEERS:
2886 if (event_bond->send_peer_notif)
2887 event_bond->send_peer_notif--;
2888 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 default:
2890 break;
2891 }
2892
2893 return NOTIFY_DONE;
2894}
2895
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002896static int bond_slave_netdev_event(unsigned long event,
2897 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898{
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002899 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
nikolay@redhat.com61013912013-04-11 09:18:55 +00002900 struct bonding *bond;
2901 struct net_device *bond_dev;
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002902 u32 old_speed;
2903 u8 old_duplex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904
nikolay@redhat.com61013912013-04-11 09:18:55 +00002905 /* A netdev event can be generated while enslaving a device
2906 * before netdev_rx_handler_register is called in which case
2907 * slave will be NULL
2908 */
2909 if (!slave)
2910 return NOTIFY_DONE;
2911 bond_dev = slave->bond->dev;
2912 bond = slave->bond;
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002913 primary = rtnl_dereference(bond->primary_slave);
nikolay@redhat.com61013912013-04-11 09:18:55 +00002914
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 switch (event) {
2916 case NETDEV_UNREGISTER:
nikolay@redhat.com8d2ada72013-06-26 17:13:37 +02002917 if (bond_dev->type != ARPHRD_ETHER)
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002918 bond_release_and_destroy(bond_dev, slave_dev);
2919 else
2920 bond_release(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 break;
Weiping Pan98f41f62011-10-31 17:20:48 +00002922 case NETDEV_UP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 case NETDEV_CHANGE:
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002924 old_speed = slave->speed;
2925 old_duplex = slave->duplex;
Jay Vosburgh17d04502009-03-18 18:38:25 -07002926
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002927 bond_update_speed_duplex(slave);
Jay Vosburgh17d04502009-03-18 18:38:25 -07002928
Veaceslav Falico01844092014-05-15 21:39:55 +02002929 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002930 if (old_speed != slave->speed)
2931 bond_3ad_adapter_speed_changed(slave);
2932 if (old_duplex != slave->duplex)
2933 bond_3ad_adapter_duplex_changed(slave);
Jay Vosburgh17d04502009-03-18 18:38:25 -07002934 }
Mahesh Bandewar950ddcb2015-02-19 10:13:25 -08002935 /* Fallthrough */
2936 case NETDEV_DOWN:
Mahesh Bandewaree637712014-10-04 17:45:01 -07002937 /* Refresh slave-array if applicable!
2938 * If the setup does not use miimon or arpmon (mode-specific!),
2939 * then these events will not cause the slave-array to be
2940 * refreshed. This will cause xmit to use a slave that is not
2941 * usable. Avoid such situation by refeshing the array at these
2942 * events. If these (miimon/arpmon) parameters are configured
2943 * then array gets refreshed twice and that should be fine!
2944 */
2945 if (bond_mode_uses_xmit_hash(bond))
2946 bond_update_slave_arr(bond, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 case NETDEV_CHANGEMTU:
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002949 /* TODO: Should slaves be allowed to
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 * independently alter their MTU? For
2951 * an active-backup bond, slaves need
2952 * not be the same type of device, so
2953 * MTUs may vary. For other modes,
2954 * slaves arguably should have the
2955 * same MTUs. To do this, we'd need to
2956 * take over the slave's change_mtu
2957 * function for the duration of their
2958 * servitude.
2959 */
2960 break;
2961 case NETDEV_CHANGENAME:
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01002962 /* we don't care if we don't have primary set */
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02002963 if (!bond_uses_primary(bond) ||
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01002964 !bond->params.primary[0])
2965 break;
2966
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002967 if (slave == primary) {
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01002968 /* slave's name changed - he's no longer primary */
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002969 RCU_INIT_POINTER(bond->primary_slave, NULL);
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01002970 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2971 /* we have a new primary slave */
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002972 rcu_assign_pointer(bond->primary_slave, slave);
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01002973 } else { /* we didn't change primary - exit */
2974 break;
2975 }
2976
Veaceslav Falico76444f52014-07-15 19:35:58 +02002977 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002978 primary ? slave_dev->name : "none");
dingtianhongf80889a2014-02-12 12:06:40 +08002979
2980 block_netpoll_tx();
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01002981 bond_select_active_slave(bond);
dingtianhongf80889a2014-02-12 12:06:40 +08002982 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 break;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04002984 case NETDEV_FEAT_CHANGE:
2985 bond_compute_features(bond);
2986 break;
Jiri Pirko4aa5dee2013-07-20 12:13:53 +02002987 case NETDEV_RESEND_IGMP:
2988 /* Propagate to master device */
2989 call_netdevice_notifiers(event, slave->bond->dev);
2990 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 default:
2992 break;
2993 }
2994
2995 return NOTIFY_DONE;
2996}
2997
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002998/* bond_netdev_event: handle netdev notifier chain events.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 *
3000 * This function receives events for the netdev chain. The caller (an
Alan Sterne041c682006-03-27 01:16:30 -08003001 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 * locks for us to safely manipulate the slave devices (RTNL lock,
3003 * dev_probe_lock).
3004 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003005static int bond_netdev_event(struct notifier_block *this,
3006 unsigned long event, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007{
Jiri Pirko351638e2013-05-28 01:30:21 +00003008 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009
Veaceslav Falico76444f52014-07-15 19:35:58 +02003010 netdev_dbg(event_dev, "event: %lx\n", event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011
Jay Vosburgh0b680e72006-09-22 21:54:10 -07003012 if (!(event_dev->priv_flags & IFF_BONDING))
3013 return NOTIFY_DONE;
3014
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 if (event_dev->flags & IFF_MASTER) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003016 netdev_dbg(event_dev, "IFF_MASTER\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 return bond_master_netdev_event(event, event_dev);
3018 }
3019
3020 if (event_dev->flags & IFF_SLAVE) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003021 netdev_dbg(event_dev, "IFF_SLAVE\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 return bond_slave_netdev_event(event, event_dev);
3023 }
3024
3025 return NOTIFY_DONE;
3026}
3027
3028static struct notifier_block bond_netdev_notifier = {
3029 .notifier_call = bond_netdev_event,
3030};
3031
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003032/*---------------------------- Hashing Policies -----------------------------*/
3033
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003034/* L2 hash helper */
3035static inline u32 bond_eth_hash(struct sk_buff *skb)
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003036{
Jianhua Xiece04d632014-07-17 14:16:25 +08003037 struct ethhdr *ep, hdr_tmp;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003038
Jianhua Xiece04d632014-07-17 14:16:25 +08003039 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3040 if (ep)
3041 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003042 return 0;
3043}
3044
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003045/* Extract the appropriate headers based on bond's xmit policy */
3046static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3047 struct flow_keys *fk)
John Eaglesham6b923cb2012-08-21 20:43:35 +00003048{
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003049 const struct ipv6hdr *iph6;
Eric Dumazet43945422013-04-15 17:03:24 +00003050 const struct iphdr *iph;
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003051 int noff, proto = -1;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003052
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003053 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3054 return skb_flow_dissect(skb, fk);
3055
3056 fk->ports = 0;
3057 noff = skb_network_offset(skb);
3058 if (skb->protocol == htons(ETH_P_IP)) {
dingtianhong054bb882014-03-25 17:00:09 +08003059 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003060 return false;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003061 iph = ip_hdr(skb);
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003062 fk->src = iph->saddr;
3063 fk->dst = iph->daddr;
3064 noff += iph->ihl << 2;
3065 if (!ip_is_fragment(iph))
3066 proto = iph->protocol;
3067 } else if (skb->protocol == htons(ETH_P_IPV6)) {
dingtianhong054bb882014-03-25 17:00:09 +08003068 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003069 return false;
3070 iph6 = ipv6_hdr(skb);
3071 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3072 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3073 noff += sizeof(*iph6);
3074 proto = iph6->nexthdr;
3075 } else {
3076 return false;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003077 }
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003078 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3079 fk->ports = skb_flow_get_ports(skb, noff, proto);
John Eaglesham6b923cb2012-08-21 20:43:35 +00003080
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003081 return true;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003082}
3083
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003084/**
3085 * bond_xmit_hash - generate a hash value based on the xmit policy
3086 * @bond: bonding device
3087 * @skb: buffer to use for headers
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003088 *
3089 * This function will extract the necessary headers from the skb buffer and use
3090 * them to generate a hash based on the xmit_policy set in the bonding device
John Eaglesham6b923cb2012-08-21 20:43:35 +00003091 */
Mahesh Bandewaree62e862014-04-22 16:30:15 -07003092u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
John Eaglesham6b923cb2012-08-21 20:43:35 +00003093{
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003094 struct flow_keys flow;
3095 u32 hash;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003096
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003097 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3098 !bond_flow_dissect(bond, skb, &flow))
Mahesh Bandewaree62e862014-04-22 16:30:15 -07003099 return bond_eth_hash(skb);
Eric Dumazet43945422013-04-15 17:03:24 +00003100
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003101 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3102 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3103 hash = bond_eth_hash(skb);
3104 else
3105 hash = (__force u32)flow.ports;
3106 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3107 hash ^= (hash >> 16);
3108 hash ^= (hash >> 8);
John Eaglesham6b923cb2012-08-21 20:43:35 +00003109
Mahesh Bandewaree62e862014-04-22 16:30:15 -07003110 return hash;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003111}
3112
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113/*-------------------------- Device entry points ----------------------------*/
3114
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003115static void bond_work_init_all(struct bonding *bond)
3116{
3117 INIT_DELAYED_WORK(&bond->mcast_work,
3118 bond_resend_igmp_join_requests_delayed);
3119 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3120 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
Veaceslav Falico01844092014-05-15 21:39:55 +02003121 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003122 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3123 else
3124 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3125 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
Mahesh Bandewaree637712014-10-04 17:45:01 -07003126 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003127}
3128
3129static void bond_work_cancel_all(struct bonding *bond)
3130{
3131 cancel_delayed_work_sync(&bond->mii_work);
3132 cancel_delayed_work_sync(&bond->arp_work);
3133 cancel_delayed_work_sync(&bond->alb_work);
3134 cancel_delayed_work_sync(&bond->ad_work);
3135 cancel_delayed_work_sync(&bond->mcast_work);
Mahesh Bandewaree637712014-10-04 17:45:01 -07003136 cancel_delayed_work_sync(&bond->slave_arr_work);
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003137}
3138
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139static int bond_open(struct net_device *bond_dev)
3140{
Wang Chen454d7c92008-11-12 23:37:49 -08003141 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003142 struct list_head *iter;
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003143 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003145 /* reset slave->backup and slave->inactive */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02003146 if (bond_has_slaves(bond)) {
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003147 bond_for_each_slave(bond, slave, iter) {
Eric Dumazet4740d632014-07-15 06:56:55 -07003148 if (bond_uses_primary(bond) &&
3149 slave != rcu_access_pointer(bond->curr_active_slave)) {
dingtianhong5e5b0662014-02-26 11:05:22 +08003150 bond_set_slave_inactive_flags(slave,
3151 BOND_SLAVE_NOTIFY_NOW);
Wilson Kok8bbe71a2015-01-26 01:16:58 -05003152 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
dingtianhong5e5b0662014-02-26 11:05:22 +08003153 bond_set_slave_active_flags(slave,
3154 BOND_SLAVE_NOTIFY_NOW);
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003155 }
3156 }
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003157 }
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003158
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003159 bond_work_init_all(bond);
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00003160
Holger Eitzenberger58402052008-12-09 23:07:13 -08003161 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 /* bond_alb_initialize must be called before the timer
3163 * is started.
3164 */
Veaceslav Falico01844092014-05-15 21:39:55 +02003165 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
stephen hemmingerb4739462010-01-25 23:34:15 +00003166 return -ENOMEM;
Mahesh Bandeware9f0fb82014-04-22 16:30:22 -07003167 if (bond->params.tlb_dynamic_lb)
3168 queue_delayed_work(bond->wq, &bond->alb_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169 }
3170
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003171 if (bond->params.miimon) /* link check interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003172 queue_delayed_work(bond->wq, &bond->mii_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173
3174 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003175 queue_delayed_work(bond->wq, &bond->arp_work, 0);
Veaceslav Falico3fe68df2014-02-18 07:48:39 +01003176 bond->recv_probe = bond_arp_rcv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 }
3178
Veaceslav Falico01844092014-05-15 21:39:55 +02003179 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003180 queue_delayed_work(bond->wq, &bond->ad_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 /* register to receive LACPDUs */
Jiri Pirko3aba8912011-04-19 03:48:16 +00003182 bond->recv_probe = bond_3ad_lacpdu_recv;
Jay Vosburghfd989c82008-11-04 17:51:16 -08003183 bond_3ad_initiate_agg_selection(bond, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 }
3185
Mahesh Bandewaree637712014-10-04 17:45:01 -07003186 if (bond_mode_uses_xmit_hash(bond))
3187 bond_update_slave_arr(bond, NULL);
3188
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189 return 0;
3190}
3191
3192static int bond_close(struct net_device *bond_dev)
3193{
Wang Chen454d7c92008-11-12 23:37:49 -08003194 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003196 bond_work_cancel_all(bond);
nikolay@redhat.com6c8d23f2013-09-02 13:51:38 +02003197 bond->send_peer_notif = 0;
nikolay@redhat.comee8487c2013-09-02 13:51:39 +02003198 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199 bond_alb_deinitialize(bond);
Jiri Pirko3aba8912011-04-19 03:48:16 +00003200 bond->recv_probe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201
3202 return 0;
3203}
3204
Eric Dumazet28172732010-07-07 14:58:56 -07003205static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3206 struct rtnl_link_stats64 *stats)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207{
Wang Chen454d7c92008-11-12 23:37:49 -08003208 struct bonding *bond = netdev_priv(bond_dev);
Eric Dumazet28172732010-07-07 14:58:56 -07003209 struct rtnl_link_stats64 temp;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003210 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003213 memcpy(stats, &bond->bond_stats, sizeof(*stats));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003215 bond_for_each_slave(bond, slave, iter) {
Ben Hutchingsbe1f3c22010-06-08 07:19:54 +00003216 const struct rtnl_link_stats64 *sstats =
Eric Dumazet28172732010-07-07 14:58:56 -07003217 dev_get_stats(slave->dev, &temp);
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003218 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
Stephen Hemmingereeda3fd2008-11-19 21:40:23 -08003219
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003220 stats->rx_packets += sstats->rx_packets - pstats->rx_packets;
3221 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3222 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3223 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003225 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3226 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3227 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3228 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003230 stats->multicast += sstats->multicast - pstats->multicast;
3231 stats->collisions += sstats->collisions - pstats->collisions;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003233 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3234 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3235 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3236 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3237 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3238 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003240 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3241 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3242 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3243 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3244 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3245
3246 /* save off the slave stats for the next run */
3247 memcpy(pstats, sstats, sizeof(*sstats));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 }
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003249 memcpy(&bond->bond_stats, stats, sizeof(*stats));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250
3251 return stats;
3252}
3253
3254static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3255{
Jiri Pirko080a06e2013-10-18 17:43:36 +02003256 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 struct net_device *slave_dev = NULL;
3258 struct ifbond k_binfo;
3259 struct ifbond __user *u_binfo = NULL;
3260 struct ifslave k_sinfo;
3261 struct ifslave __user *u_sinfo = NULL;
3262 struct mii_ioctl_data *mii = NULL;
Nikolay Aleksandrovd1fbd3e2014-01-22 14:53:35 +01003263 struct bond_opt_value newval;
Gao feng387ff9112013-01-31 16:31:00 +00003264 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 int res = 0;
3266
Veaceslav Falico76444f52014-07-15 19:35:58 +02003267 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268
3269 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270 case SIOCGMIIPHY:
3271 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003272 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003274
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 mii->phy_id = 0;
3276 /* Fall Through */
3277 case SIOCGMIIREG:
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003278 /* We do this again just in case we were called by SIOCGMIIREG
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279 * instead of SIOCGMIIPHY.
3280 */
3281 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003282 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003284
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 if (mii->reg_num == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286 mii->val_out = 0;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003287 if (netif_carrier_ok(bond->dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288 mii->val_out = BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289 }
3290
3291 return 0;
3292 case BOND_INFO_QUERY_OLD:
3293 case SIOCBONDINFOQUERY:
3294 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3295
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003296 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298
3299 res = bond_info_query(bond_dev, &k_binfo);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003300 if (res == 0 &&
3301 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3302 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303
3304 return res;
3305 case BOND_SLAVE_INFO_QUERY_OLD:
3306 case SIOCBONDSLAVEINFOQUERY:
3307 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3308
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003309 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311
3312 res = bond_slave_info_query(bond_dev, &k_sinfo);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003313 if (res == 0 &&
3314 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3315 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316
3317 return res;
3318 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 break;
3320 }
3321
Gao feng387ff9112013-01-31 16:31:00 +00003322 net = dev_net(bond_dev);
3323
3324 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326
Ying Xue0917b932014-01-15 10:23:37 +08003327 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328
Veaceslav Falico76444f52014-07-15 19:35:58 +02003329 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003331 if (!slave_dev)
Ying Xue0917b932014-01-15 10:23:37 +08003332 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333
Veaceslav Falico76444f52014-07-15 19:35:58 +02003334 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
Ying Xue0917b932014-01-15 10:23:37 +08003335 switch (cmd) {
3336 case BOND_ENSLAVE_OLD:
3337 case SIOCBONDENSLAVE:
3338 res = bond_enslave(bond_dev, slave_dev);
3339 break;
3340 case BOND_RELEASE_OLD:
3341 case SIOCBONDRELEASE:
3342 res = bond_release(bond_dev, slave_dev);
3343 break;
3344 case BOND_SETHWADDR_OLD:
3345 case SIOCBONDSETHWADDR:
3346 bond_set_dev_addr(bond_dev, slave_dev);
3347 res = 0;
3348 break;
3349 case BOND_CHANGE_ACTIVE_OLD:
3350 case SIOCBONDCHANGEACTIVE:
Nikolay Aleksandrovd1fbd3e2014-01-22 14:53:35 +01003351 bond_opt_initstr(&newval, slave_dev->name);
3352 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
Ying Xue0917b932014-01-15 10:23:37 +08003353 break;
3354 default:
3355 res = -EOPNOTSUPP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356 }
3357
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 return res;
3359}
3360
Jiri Pirkod03462b2011-08-16 03:15:04 +00003361static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3362{
3363 struct bonding *bond = netdev_priv(bond_dev);
3364
3365 if (change & IFF_PROMISC)
3366 bond_set_promiscuity(bond,
3367 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3368
3369 if (change & IFF_ALLMULTI)
3370 bond_set_allmulti(bond,
3371 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3372}
3373
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003374static void bond_set_rx_mode(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375{
Wang Chen454d7c92008-11-12 23:37:49 -08003376 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003377 struct list_head *iter;
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003378 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379
Veaceslav Falicob3241872013-09-28 21:18:56 +02003380 rcu_read_lock();
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02003381 if (bond_uses_primary(bond)) {
Veaceslav Falicob3241872013-09-28 21:18:56 +02003382 slave = rcu_dereference(bond->curr_active_slave);
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003383 if (slave) {
3384 dev_uc_sync(slave->dev, bond_dev);
3385 dev_mc_sync(slave->dev, bond_dev);
3386 }
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003387 } else {
Veaceslav Falicob3241872013-09-28 21:18:56 +02003388 bond_for_each_slave_rcu(bond, slave, iter) {
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003389 dev_uc_sync_multiple(slave->dev, bond_dev);
3390 dev_mc_sync_multiple(slave->dev, bond_dev);
3391 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392 }
Veaceslav Falicob3241872013-09-28 21:18:56 +02003393 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394}
3395
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003396static int bond_neigh_init(struct neighbour *n)
Stephen Hemminger00829822008-11-20 20:14:53 -08003397{
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003398 struct bonding *bond = netdev_priv(n->dev);
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003399 const struct net_device_ops *slave_ops;
3400 struct neigh_parms parms;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003401 struct slave *slave;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003402 int ret;
Stephen Hemminger00829822008-11-20 20:14:53 -08003403
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003404 slave = bond_first_slave(bond);
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003405 if (!slave)
3406 return 0;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003407 slave_ops = slave->dev->netdev_ops;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003408 if (!slave_ops->ndo_neigh_setup)
3409 return 0;
3410
3411 parms.neigh_setup = NULL;
3412 parms.neigh_cleanup = NULL;
3413 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3414 if (ret)
3415 return ret;
3416
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003417 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003418 * after the last slave has been detached. Assumes that all slaves
3419 * utilize the same neigh_cleanup (true at this writing as only user
3420 * is ipoib).
3421 */
3422 n->parms->neigh_cleanup = parms.neigh_cleanup;
3423
3424 if (!parms.neigh_setup)
3425 return 0;
3426
3427 return parms.neigh_setup(n);
3428}
3429
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003430/* The bonding ndo_neigh_setup is called at init time beofre any
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003431 * slave exists. So we must declare proxy setup function which will
3432 * be used at run time to resolve the actual slave neigh param setup.
Veaceslav Falico9918d5b2013-08-02 19:07:39 +02003433 *
3434 * It's also called by master devices (such as vlans) to setup their
3435 * underlying devices. In that case - do nothing, we're already set up from
3436 * our init.
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003437 */
3438static int bond_neigh_setup(struct net_device *dev,
3439 struct neigh_parms *parms)
3440{
Veaceslav Falico9918d5b2013-08-02 19:07:39 +02003441 /* modify only our neigh_parms */
3442 if (parms->dev == dev)
3443 parms->neigh_setup = bond_neigh_init;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003444
Stephen Hemminger00829822008-11-20 20:14:53 -08003445 return 0;
3446}
3447
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003448/* Change the MTU of all of a master's slaves to match the master */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3450{
Wang Chen454d7c92008-11-12 23:37:49 -08003451 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003452 struct slave *slave, *rollback_slave;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003453 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454 int res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003455
Veaceslav Falico76444f52014-07-15 19:35:58 +02003456 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003458 bond_for_each_slave(bond, slave, iter) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003459 netdev_dbg(bond_dev, "s %p c_m %p\n",
3460 slave, slave->dev->netdev_ops->ndo_change_mtu);
Mitch Williamse944ef72005-11-09 10:36:50 -08003461
Linus Torvalds1da177e2005-04-16 15:20:36 -07003462 res = dev_set_mtu(slave->dev, new_mtu);
3463
3464 if (res) {
3465 /* If we failed to set the slave's mtu to the new value
3466 * we must abort the operation even in ACTIVE_BACKUP
3467 * mode, because if we allow the backup slaves to have
3468 * different mtu values than the active slave we'll
3469 * need to change their mtu when doing a failover. That
3470 * means changing their mtu from timer context, which
3471 * is probably not a good idea.
3472 */
Veaceslav Falico76444f52014-07-15 19:35:58 +02003473 netdev_dbg(bond_dev, "err %d %s\n", res,
3474 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 goto unwind;
3476 }
3477 }
3478
3479 bond_dev->mtu = new_mtu;
3480
3481 return 0;
3482
3483unwind:
3484 /* unwind from head to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003485 bond_for_each_slave(bond, rollback_slave, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003486 int tmp_res;
3487
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003488 if (rollback_slave == slave)
3489 break;
3490
3491 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492 if (tmp_res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003493 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3494 tmp_res, rollback_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495 }
3496 }
3497
3498 return res;
3499}
3500
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003501/* Change HW address
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 *
3503 * Note that many devices must be down to change the HW address, and
3504 * downing the master releases all slaves. We can make bonds full of
3505 * bonding devices to test this, however.
3506 */
3507static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3508{
Wang Chen454d7c92008-11-12 23:37:49 -08003509 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003510 struct slave *slave, *rollback_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511 struct sockaddr *sa = addr, tmp_sa;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003512 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513 int res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514
Veaceslav Falico01844092014-05-15 21:39:55 +02003515 if (BOND_MODE(bond) == BOND_MODE_ALB)
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003516 return bond_alb_set_mac_address(bond_dev, addr);
3517
3518
Veaceslav Falico76444f52014-07-15 19:35:58 +02003519 netdev_dbg(bond_dev, "bond=%p\n", bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520
Jay Vosburgh1b5acd22013-05-31 11:57:31 +00003521 /* If fail_over_mac is enabled, do nothing and return success.
3522 * Returning an error causes ifenslave to fail.
Jay Vosburghdd957c52007-10-09 19:57:24 -07003523 */
dingtianhongcc689aa2014-01-25 13:00:57 +08003524 if (bond->params.fail_over_mac &&
Veaceslav Falico01844092014-05-15 21:39:55 +02003525 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
Jay Vosburghdd957c52007-10-09 19:57:24 -07003526 return 0;
Moni Shoua2ab82852007-10-09 19:43:39 -07003527
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003528 if (!is_valid_ether_addr(sa->sa_data))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529 return -EADDRNOTAVAIL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003531 bond_for_each_slave(bond, slave, iter) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003532 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533 res = dev_set_mac_address(slave->dev, addr);
3534 if (res) {
3535 /* TODO: consider downing the slave
3536 * and retry ?
3537 * User should expect communications
3538 * breakage anyway until ARP finish
3539 * updating, so...
3540 */
Veaceslav Falico76444f52014-07-15 19:35:58 +02003541 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542 goto unwind;
3543 }
3544 }
3545
3546 /* success */
3547 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3548 return 0;
3549
3550unwind:
3551 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3552 tmp_sa.sa_family = bond_dev->type;
3553
3554 /* unwind from head to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003555 bond_for_each_slave(bond, rollback_slave, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556 int tmp_res;
3557
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003558 if (rollback_slave == slave)
3559 break;
3560
3561 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 if (tmp_res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003563 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3564 tmp_res, rollback_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565 }
3566 }
3567
3568 return res;
3569}
3570
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003571/**
3572 * bond_xmit_slave_id - transmit skb through slave with slave_id
3573 * @bond: bonding device that is transmitting
3574 * @skb: buffer to transmit
3575 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3576 *
3577 * This function tries to transmit through slave with slave_id but in case
3578 * it fails, it tries to find the first available slave for transmission.
3579 * The skb is consumed in all cases, thus the function is void.
3580 */
stephen hemmingerda131dd2013-12-29 11:41:25 -08003581static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003582{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003583 struct list_head *iter;
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003584 struct slave *slave;
3585 int i = slave_id;
3586
3587 /* Here we start from the slave with slave_id */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003588 bond_for_each_slave_rcu(bond, slave, iter) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003589 if (--i < 0) {
Veaceslav Falico891ab542014-05-15 21:39:58 +02003590 if (bond_slave_can_tx(slave)) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003591 bond_dev_queue_xmit(bond, skb, slave->dev);
3592 return;
3593 }
3594 }
3595 }
3596
3597 /* Here we start from the first slave up to slave_id */
3598 i = slave_id;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003599 bond_for_each_slave_rcu(bond, slave, iter) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003600 if (--i < 0)
3601 break;
Veaceslav Falico891ab542014-05-15 21:39:58 +02003602 if (bond_slave_can_tx(slave)) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003603 bond_dev_queue_xmit(bond, skb, slave->dev);
3604 return;
3605 }
3606 }
3607 /* no slave that can tx has been found */
Eric Dumazet31aa8602014-10-31 11:47:54 -07003608 bond_tx_drop(bond->dev, skb);
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003609}
3610
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003611/**
3612 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3613 * @bond: bonding device to use
3614 *
3615 * Based on the value of the bonding device's packets_per_slave parameter
3616 * this function generates a slave id, which is usually used as the next
3617 * slave to transmit through.
3618 */
3619static u32 bond_rr_gen_slave_id(struct bonding *bond)
3620{
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003621 u32 slave_id;
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01003622 struct reciprocal_value reciprocal_packets_per_slave;
3623 int packets_per_slave = bond->params.packets_per_slave;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003624
3625 switch (packets_per_slave) {
3626 case 0:
3627 slave_id = prandom_u32();
3628 break;
3629 case 1:
3630 slave_id = bond->rr_tx_counter;
3631 break;
3632 default:
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01003633 reciprocal_packets_per_slave =
3634 bond->params.reciprocal_packets_per_slave;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003635 slave_id = reciprocal_divide(bond->rr_tx_counter,
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01003636 reciprocal_packets_per_slave);
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003637 break;
3638 }
3639 bond->rr_tx_counter++;
3640
3641 return slave_id;
3642}
3643
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3645{
Wang Chen454d7c92008-11-12 23:37:49 -08003646 struct bonding *bond = netdev_priv(bond_dev);
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003647 struct iphdr *iph = ip_hdr(skb);
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003648 struct slave *slave;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003649 u32 slave_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003651 /* Start with the curr_active_slave that joined the bond as the
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003652 * default for sending IGMP traffic. For failover purposes one
3653 * needs to maintain some consistency for the interface that will
3654 * send the join/membership reports. The curr_active_slave found
3655 * will send all of this type of traffic.
Jay Vosburghcf5f9042007-10-17 17:37:47 -07003656 */
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003657 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003658 slave = rcu_dereference(bond->curr_active_slave);
dingtianhonga64d0442014-09-19 21:04:57 +08003659 if (slave)
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003660 bond_dev_queue_xmit(bond, skb, slave->dev);
3661 else
3662 bond_xmit_slave_id(bond, skb, 0);
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003663 } else {
Nikolay Aleksandrov9a72c2d2014-09-12 17:38:18 +02003664 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3665
3666 if (likely(slave_cnt)) {
3667 slave_id = bond_rr_gen_slave_id(bond);
3668 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3669 } else {
Eric Dumazet31aa8602014-10-31 11:47:54 -07003670 bond_tx_drop(bond_dev, skb);
Nikolay Aleksandrov9a72c2d2014-09-12 17:38:18 +02003671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 }
Michał Mirosław0693e882011-05-07 01:48:02 +00003673
Patrick McHardyec634fe2009-07-05 19:23:38 -07003674 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675}
3676
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003677/* In active-backup mode, we know that bond->curr_active_slave is always valid if
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678 * the bond has a usable interface.
3679 */
3680static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3681{
Wang Chen454d7c92008-11-12 23:37:49 -08003682 struct bonding *bond = netdev_priv(bond_dev);
nikolay@redhat.com71bc3b22013-08-01 16:54:48 +02003683 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003685 slave = rcu_dereference(bond->curr_active_slave);
nikolay@redhat.com71bc3b22013-08-01 16:54:48 +02003686 if (slave)
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003687 bond_dev_queue_xmit(bond, skb, slave->dev);
3688 else
Eric Dumazet31aa8602014-10-31 11:47:54 -07003689 bond_tx_drop(bond_dev, skb);
Michał Mirosław0693e882011-05-07 01:48:02 +00003690
Patrick McHardyec634fe2009-07-05 19:23:38 -07003691 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692}
3693
Mahesh Bandewaree637712014-10-04 17:45:01 -07003694/* Use this to update slave_array when (a) it's not appropriate to update
3695 * slave_array right away (note that update_slave_array() may sleep)
3696 * and / or (b) RTNL is not held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697 */
Mahesh Bandewaree637712014-10-04 17:45:01 -07003698void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699{
Mahesh Bandewaree637712014-10-04 17:45:01 -07003700 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3701}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702
Mahesh Bandewaree637712014-10-04 17:45:01 -07003703/* Slave array work handler. Holds only RTNL */
3704static void bond_slave_arr_handler(struct work_struct *work)
3705{
3706 struct bonding *bond = container_of(work, struct bonding,
3707 slave_arr_work.work);
3708 int ret;
3709
3710 if (!rtnl_trylock())
3711 goto err;
3712
3713 ret = bond_update_slave_arr(bond, NULL);
3714 rtnl_unlock();
3715 if (ret) {
3716 pr_warn_ratelimited("Failed to update slave array from WT\n");
3717 goto err;
3718 }
3719 return;
3720
3721err:
3722 bond_slave_arr_work_rearm(bond, 1);
3723}
3724
3725/* Build the usable slaves array in control path for modes that use xmit-hash
3726 * to determine the slave interface -
3727 * (a) BOND_MODE_8023AD
3728 * (b) BOND_MODE_XOR
3729 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3730 *
3731 * The caller is expected to hold RTNL only and NO other lock!
3732 */
3733int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3734{
3735 struct slave *slave;
3736 struct list_head *iter;
3737 struct bond_up_slave *new_arr, *old_arr;
3738 int slaves_in_agg;
3739 int agg_id = 0;
3740 int ret = 0;
3741
3742#ifdef CONFIG_LOCKDEP
3743 WARN_ON(lockdep_is_held(&bond->mode_lock));
3744#endif
3745
3746 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3747 GFP_KERNEL);
3748 if (!new_arr) {
3749 ret = -ENOMEM;
3750 pr_err("Failed to build slave-array.\n");
3751 goto out;
3752 }
3753 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3754 struct ad_info ad_info;
3755
3756 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3757 pr_debug("bond_3ad_get_active_agg_info failed\n");
3758 kfree_rcu(new_arr, rcu);
3759 /* No active aggragator means it's not safe to use
3760 * the previous array.
3761 */
3762 old_arr = rtnl_dereference(bond->slave_arr);
3763 if (old_arr) {
3764 RCU_INIT_POINTER(bond->slave_arr, NULL);
3765 kfree_rcu(old_arr, rcu);
3766 }
3767 goto out;
3768 }
3769 slaves_in_agg = ad_info.ports;
3770 agg_id = ad_info.aggregator_id;
3771 }
3772 bond_for_each_slave(bond, slave, iter) {
3773 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3774 struct aggregator *agg;
3775
3776 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3777 if (!agg || agg->aggregator_identifier != agg_id)
3778 continue;
3779 }
3780 if (!bond_slave_can_tx(slave))
3781 continue;
3782 if (skipslave == slave)
3783 continue;
3784 new_arr->arr[new_arr->count++] = slave;
3785 }
3786
3787 old_arr = rtnl_dereference(bond->slave_arr);
3788 rcu_assign_pointer(bond->slave_arr, new_arr);
3789 if (old_arr)
3790 kfree_rcu(old_arr, rcu);
3791out:
3792 if (ret != 0 && skipslave) {
3793 int idx;
3794
3795 /* Rare situation where caller has asked to skip a specific
3796 * slave but allocation failed (most likely!). BTW this is
3797 * only possible when the call is initiated from
3798 * __bond_release_one(). In this situation; overwrite the
3799 * skipslave entry in the array with the last entry from the
3800 * array to avoid a situation where the xmit path may choose
3801 * this to-be-skipped slave to send a packet out.
3802 */
3803 old_arr = rtnl_dereference(bond->slave_arr);
3804 for (idx = 0; idx < old_arr->count; idx++) {
3805 if (skipslave == old_arr->arr[idx]) {
3806 old_arr->arr[idx] =
3807 old_arr->arr[old_arr->count-1];
3808 old_arr->count--;
3809 break;
3810 }
3811 }
3812 }
3813 return ret;
3814}
3815
3816/* Use this Xmit function for 3AD as well as XOR modes. The current
3817 * usable slave array is formed in the control path. The xmit function
3818 * just calculates hash and sends the packet out.
3819 */
Jonathan Toppins30369102015-01-26 01:17:01 -05003820static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
Mahesh Bandewaree637712014-10-04 17:45:01 -07003821{
3822 struct bonding *bond = netdev_priv(dev);
3823 struct slave *slave;
3824 struct bond_up_slave *slaves;
3825 unsigned int count;
3826
3827 slaves = rcu_dereference(bond->slave_arr);
3828 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3829 if (likely(count)) {
3830 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3831 bond_dev_queue_xmit(bond, skb, slave->dev);
3832 } else {
Eric Dumazet31aa8602014-10-31 11:47:54 -07003833 bond_tx_drop(dev, skb);
Mahesh Bandewaree637712014-10-04 17:45:01 -07003834 }
Michał Mirosław0693e882011-05-07 01:48:02 +00003835
Patrick McHardyec634fe2009-07-05 19:23:38 -07003836 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837}
3838
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003839/* in broadcast mode, we send everything to all usable interfaces. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3841{
Wang Chen454d7c92008-11-12 23:37:49 -08003842 struct bonding *bond = netdev_priv(bond_dev);
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003843 struct slave *slave = NULL;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003844 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003846 bond_for_each_slave_rcu(bond, slave, iter) {
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003847 if (bond_is_last_slave(bond, slave))
3848 break;
Veaceslav Falicob6adc612014-05-15 21:39:57 +02003849 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003850 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003852 if (!skb2) {
dingtianhong9152e262014-03-25 17:00:10 +08003853 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3854 bond_dev->name, __func__);
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003855 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856 }
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003857 bond_dev_queue_xmit(bond, skb2, slave->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858 }
3859 }
Veaceslav Falicob6adc612014-05-15 21:39:57 +02003860 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003861 bond_dev_queue_xmit(bond, skb, slave->dev);
3862 else
Eric Dumazet31aa8602014-10-31 11:47:54 -07003863 bond_tx_drop(bond_dev, skb);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003864
Patrick McHardyec634fe2009-07-05 19:23:38 -07003865 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866}
3867
3868/*------------------------- Device initialization ---------------------------*/
3869
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003870/* Lookup the slave that corresponds to a qid */
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003871static inline int bond_slave_override(struct bonding *bond,
3872 struct sk_buff *skb)
3873{
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003874 struct slave *slave = NULL;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003875 struct list_head *iter;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003876
Michał Mirosław0693e882011-05-07 01:48:02 +00003877 if (!skb->queue_mapping)
3878 return 1;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003879
3880 /* Find out if any slaves have the same mapping as this skb. */
dingtianhong3900f292014-01-02 09:13:06 +08003881 bond_for_each_slave_rcu(bond, slave, iter) {
3882 if (slave->queue_id == skb->queue_mapping) {
Anton Nayshtutf5e2dc52015-03-29 14:20:25 +03003883 if (bond_slave_is_up(slave) &&
3884 slave->link == BOND_LINK_UP) {
dingtianhong3900f292014-01-02 09:13:06 +08003885 bond_dev_queue_xmit(bond, skb, slave->dev);
3886 return 0;
3887 }
3888 /* If the slave isn't UP, use default transmit policy. */
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003889 break;
3890 }
3891 }
3892
dingtianhong3900f292014-01-02 09:13:06 +08003893 return 1;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003894}
3895
Neil Horman374eeb52011-06-03 10:35:52 +00003896
Jason Wangf663dd92014-01-10 16:18:26 +08003897static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
Daniel Borkmann99932d42014-02-16 15:55:20 +01003898 void *accel_priv, select_queue_fallback_t fallback)
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003899{
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003900 /* This helper function exists to help dev_pick_tx get the correct
Phil Oesterfd0e4352011-03-14 06:22:04 +00003901 * destination queue. Using a helper function skips a call to
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003902 * skb_tx_hash and will put the skbs in the queue we expect on their
3903 * way down to the bonding driver.
3904 */
Phil Oesterfd0e4352011-03-14 06:22:04 +00003905 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3906
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003907 /* Save the original txq to restore before passing to the driver */
Jiri Pirkodf4ab5b2012-07-20 02:28:49 +00003908 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
Neil Horman374eeb52011-06-03 10:35:52 +00003909
Phil Oesterfd0e4352011-03-14 06:22:04 +00003910 if (unlikely(txq >= dev->real_num_tx_queues)) {
David Decotignyd30ee672011-04-13 15:22:29 +00003911 do {
Phil Oesterfd0e4352011-03-14 06:22:04 +00003912 txq -= dev->real_num_tx_queues;
David Decotignyd30ee672011-04-13 15:22:29 +00003913 } while (txq >= dev->real_num_tx_queues);
Phil Oesterfd0e4352011-03-14 06:22:04 +00003914 }
3915 return txq;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003916}
3917
Michał Mirosław0693e882011-05-07 01:48:02 +00003918static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
Stephen Hemminger00829822008-11-20 20:14:53 -08003919{
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003920 struct bonding *bond = netdev_priv(dev);
3921
Veaceslav Falicod1e2e5c2014-05-15 21:39:52 +02003922 if (bond_should_override_tx_queue(bond) &&
3923 !bond_slave_override(bond, skb))
3924 return NETDEV_TX_OK;
Stephen Hemminger00829822008-11-20 20:14:53 -08003925
Veaceslav Falico01844092014-05-15 21:39:55 +02003926 switch (BOND_MODE(bond)) {
Stephen Hemminger00829822008-11-20 20:14:53 -08003927 case BOND_MODE_ROUNDROBIN:
3928 return bond_xmit_roundrobin(skb, dev);
3929 case BOND_MODE_ACTIVEBACKUP:
3930 return bond_xmit_activebackup(skb, dev);
Mahesh Bandewaree637712014-10-04 17:45:01 -07003931 case BOND_MODE_8023AD:
Stephen Hemminger00829822008-11-20 20:14:53 -08003932 case BOND_MODE_XOR:
Mahesh Bandewaree637712014-10-04 17:45:01 -07003933 return bond_3ad_xor_xmit(skb, dev);
Stephen Hemminger00829822008-11-20 20:14:53 -08003934 case BOND_MODE_BROADCAST:
3935 return bond_xmit_broadcast(skb, dev);
Stephen Hemminger00829822008-11-20 20:14:53 -08003936 case BOND_MODE_ALB:
Stephen Hemminger00829822008-11-20 20:14:53 -08003937 return bond_alb_xmit(skb, dev);
Mahesh Bandewarf05b42e2014-04-22 16:30:20 -07003938 case BOND_MODE_TLB:
3939 return bond_tlb_xmit(skb, dev);
Stephen Hemminger00829822008-11-20 20:14:53 -08003940 default:
3941 /* Should never happen, mode already checked */
Veaceslav Falico76444f52014-07-15 19:35:58 +02003942 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
Stephen Hemminger00829822008-11-20 20:14:53 -08003943 WARN_ON_ONCE(1);
Eric Dumazet31aa8602014-10-31 11:47:54 -07003944 bond_tx_drop(dev, skb);
Stephen Hemminger00829822008-11-20 20:14:53 -08003945 return NETDEV_TX_OK;
3946 }
3947}
3948
Michał Mirosław0693e882011-05-07 01:48:02 +00003949static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3950{
3951 struct bonding *bond = netdev_priv(dev);
3952 netdev_tx_t ret = NETDEV_TX_OK;
3953
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003954 /* If we risk deadlock from transmitting this in the
Michał Mirosław0693e882011-05-07 01:48:02 +00003955 * netpoll path, tell netpoll to queue the frame for later tx
3956 */
dingtianhong054bb882014-03-25 17:00:09 +08003957 if (unlikely(is_netpoll_tx_blocked(dev)))
Michał Mirosław0693e882011-05-07 01:48:02 +00003958 return NETDEV_TX_BUSY;
3959
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003960 rcu_read_lock();
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02003961 if (bond_has_slaves(bond))
Michał Mirosław0693e882011-05-07 01:48:02 +00003962 ret = __bond_start_xmit(skb, dev);
3963 else
Eric Dumazet31aa8602014-10-31 11:47:54 -07003964 bond_tx_drop(dev, skb);
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003965 rcu_read_unlock();
Michał Mirosław0693e882011-05-07 01:48:02 +00003966
3967 return ret;
3968}
Stephen Hemminger00829822008-11-20 20:14:53 -08003969
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003970static int bond_ethtool_get_settings(struct net_device *bond_dev,
3971 struct ethtool_cmd *ecmd)
3972{
3973 struct bonding *bond = netdev_priv(bond_dev);
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003974 unsigned long speed = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003975 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003976 struct slave *slave;
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003977
3978 ecmd->duplex = DUPLEX_UNKNOWN;
3979 ecmd->port = PORT_OTHER;
3980
Veaceslav Falico8557cd72014-05-15 21:39:59 +02003981 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003982 * do not need to check mode. Though link speed might not represent
3983 * the true receive or transmit bandwidth (not all modes are symmetric)
3984 * this is an accurate maximum.
3985 */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003986 bond_for_each_slave(bond, slave, iter) {
Veaceslav Falico8557cd72014-05-15 21:39:59 +02003987 if (bond_slave_can_tx(slave)) {
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003988 if (slave->speed != SPEED_UNKNOWN)
3989 speed += slave->speed;
3990 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3991 slave->duplex != DUPLEX_UNKNOWN)
3992 ecmd->duplex = slave->duplex;
3993 }
3994 }
3995 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003996
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003997 return 0;
3998}
3999
Jay Vosburgh217df672005-09-26 16:11:50 -07004000static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
Jiri Pirko7826d432013-01-06 00:44:26 +00004001 struct ethtool_drvinfo *drvinfo)
Jay Vosburgh217df672005-09-26 16:11:50 -07004002{
Jiri Pirko7826d432013-01-06 00:44:26 +00004003 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4004 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4005 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4006 BOND_ABI_VERSION);
Jay Vosburgh217df672005-09-26 16:11:50 -07004007}
4008
Jeff Garzik7282d492006-09-13 14:30:00 -04004009static const struct ethtool_ops bond_ethtool_ops = {
Jay Vosburgh217df672005-09-26 16:11:50 -07004010 .get_drvinfo = bond_ethtool_get_drvinfo,
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004011 .get_settings = bond_ethtool_get_settings,
Stephen Hemmingerfa53eba2008-09-13 21:17:09 -04004012 .get_link = ethtool_op_get_link,
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004013};
4014
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004015static const struct net_device_ops bond_netdev_ops = {
Stephen Hemminger181470f2009-06-12 19:02:52 +00004016 .ndo_init = bond_init,
Stephen Hemminger9e716262009-06-12 19:02:47 +00004017 .ndo_uninit = bond_uninit,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004018 .ndo_open = bond_open,
4019 .ndo_stop = bond_close,
Stephen Hemminger00829822008-11-20 20:14:53 -08004020 .ndo_start_xmit = bond_start_xmit,
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004021 .ndo_select_queue = bond_select_queue,
Ben Hutchingsbe1f3c22010-06-08 07:19:54 +00004022 .ndo_get_stats64 = bond_get_stats,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004023 .ndo_do_ioctl = bond_do_ioctl,
Jiri Pirkod03462b2011-08-16 03:15:04 +00004024 .ndo_change_rx_flags = bond_change_rx_flags,
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00004025 .ndo_set_rx_mode = bond_set_rx_mode,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004026 .ndo_change_mtu = bond_change_mtu,
Jiri Pirkocc0e4072011-07-20 04:54:46 +00004027 .ndo_set_mac_address = bond_set_mac_address,
Stephen Hemminger00829822008-11-20 20:14:53 -08004028 .ndo_neigh_setup = bond_neigh_setup,
Jiri Pirkocc0e4072011-07-20 04:54:46 +00004029 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004030 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
WANG Congf6dc31a2010-05-06 00:48:51 -07004031#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +00004032 .ndo_netpoll_setup = bond_netpoll_setup,
WANG Congf6dc31a2010-05-06 00:48:51 -07004033 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4034 .ndo_poll_controller = bond_poll_controller,
4035#endif
Jiri Pirko9232ecc2011-02-13 09:33:01 +00004036 .ndo_add_slave = bond_enslave,
4037 .ndo_del_slave = bond_release,
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004038 .ndo_fix_features = bond_fix_features,
Scott Feldmanfc8f40d2015-05-10 09:47:57 -07004039 .ndo_bridge_setlink = switchdev_port_bridge_setlink,
Scott Feldman85fdb952015-05-10 09:48:05 -07004040 .ndo_bridge_getlink = switchdev_port_bridge_getlink,
Scott Feldman54ba5a02015-05-10 09:48:01 -07004041 .ndo_bridge_dellink = switchdev_port_bridge_dellink,
Toshiaki Makita4847f042015-03-27 14:31:14 +09004042 .ndo_features_check = passthru_features_check,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004043};
4044
Doug Goldsteinb3f92b62013-02-18 14:59:23 +00004045static const struct device_type bond_type = {
4046 .name = "bond",
4047};
4048
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004049static void bond_destructor(struct net_device *bond_dev)
4050{
4051 struct bonding *bond = netdev_priv(bond_dev);
4052 if (bond->wq)
4053 destroy_workqueue(bond->wq);
4054 free_netdev(bond_dev);
4055}
4056
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004057void bond_setup(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058{
Wang Chen454d7c92008-11-12 23:37:49 -08004059 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060
Nikolay Aleksandrovb7435622014-09-11 22:49:25 +02004061 spin_lock_init(&bond->mode_lock);
Stephen Hemmingerd2991f72009-06-12 19:02:44 +00004062 bond->params = bonding_defaults;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063
4064 /* Initialize pointers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065 bond->dev = bond_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066
4067 /* Initialize the device entry points */
Stephen Hemminger181470f2009-06-12 19:02:52 +00004068 ether_setup(bond_dev);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004069 bond_dev->netdev_ops = &bond_netdev_ops;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004070 bond_dev->ethtool_ops = &bond_ethtool_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004072 bond_dev->destructor = bond_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073
Doug Goldsteinb3f92b62013-02-18 14:59:23 +00004074 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4075
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076 /* Initialize the device options */
4077 bond_dev->tx_queue_len = 0;
4078 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
Vlad Yasevichc565b482014-06-04 16:23:36 -04004079 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
Neil Horman550fd082011-07-26 06:05:38 +00004080 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004081
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004082 /* don't acquire bond device's netif_tx_lock when transmitting */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004083 bond_dev->features |= NETIF_F_LLTX;
4084
4085 /* By default, we declare the bond to be fully
4086 * VLAN hardware accelerated capable. Special
4087 * care is taken in the various xmit functions
4088 * when there are slaves that are not hw accel
4089 * capable
4090 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091
Weilong Chenf9399812014-01-22 17:16:30 +08004092 /* Don't allow bond devices to change network namespaces. */
4093 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4094
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004095 bond_dev->hw_features = BOND_VLAN_FEATURES |
Patrick McHardyf6469682013-04-19 02:04:27 +00004096 NETIF_F_HW_VLAN_CTAG_TX |
4097 NETIF_F_HW_VLAN_CTAG_RX |
4098 NETIF_F_HW_VLAN_CTAG_FILTER;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004099
Michał Mirosław34324dc2011-11-15 15:29:55 +00004100 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
Eric Dumazet24f87d42015-01-25 23:27:14 -08004101 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004102 bond_dev->features |= bond_dev->hw_features;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103}
4104
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004105/* Destroy a bonding device.
4106 * Must be under rtnl_lock when this function is called.
4107 */
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004108static void bond_uninit(struct net_device *bond_dev)
Jay Vosburgha434e432008-10-30 17:41:15 -07004109{
Wang Chen454d7c92008-11-12 23:37:49 -08004110 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico544a0282013-09-25 09:20:15 +02004111 struct list_head *iter;
4112 struct slave *slave;
Mahesh Bandewaree637712014-10-04 17:45:01 -07004113 struct bond_up_slave *arr;
Jay Vosburgha434e432008-10-30 17:41:15 -07004114
WANG Congf6dc31a2010-05-06 00:48:51 -07004115 bond_netpoll_cleanup(bond_dev);
4116
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004117 /* Release the bonded slaves */
Veaceslav Falico544a0282013-09-25 09:20:15 +02004118 bond_for_each_slave(bond, slave, iter)
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02004119 __bond_release_one(bond_dev, slave->dev, true);
Veaceslav Falico76444f52014-07-15 19:35:58 +02004120 netdev_info(bond_dev, "Released all slaves\n");
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004121
Mahesh Bandewaree637712014-10-04 17:45:01 -07004122 arr = rtnl_dereference(bond->slave_arr);
4123 if (arr) {
4124 RCU_INIT_POINTER(bond->slave_arr, NULL);
4125 kfree_rcu(arr, rcu);
4126 }
4127
Jay Vosburgha434e432008-10-30 17:41:15 -07004128 list_del(&bond->bond_list);
4129
Taku Izumif073c7c2010-12-09 15:17:13 +00004130 bond_debug_unregister(bond);
Jay Vosburgha434e432008-10-30 17:41:15 -07004131}
4132
Linus Torvalds1da177e2005-04-16 15:20:36 -07004133/*------------------------- Module initialization ---------------------------*/
4134
Linus Torvalds1da177e2005-04-16 15:20:36 -07004135static int bond_check_params(struct bond_params *params)
4136{
nikolay@redhat.com5a5c5fd2013-05-18 01:18:30 +00004137 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
stephen hemmingerf3253332014-03-04 16:36:44 -08004138 struct bond_opt_value newval;
4139 const struct bond_opt_value *valptr;
Veaceslav Falico8599b522013-06-24 11:49:34 +02004140 int arp_all_targets_value;
Mahesh Bandewar6791e462015-05-09 00:01:55 -07004141 u16 ad_actor_sys_prio = 0;
Mahesh Bandeward22a5fc2015-05-09 00:01:57 -07004142 u16 ad_user_port_key = 0;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004143
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004144 /* Convert string parameters. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004145 if (mode) {
Nikolay Aleksandrov2b3798d2014-01-22 14:53:17 +01004146 bond_opt_initstr(&newval, mode);
4147 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4148 if (!valptr) {
4149 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004150 return -EINVAL;
4151 }
Nikolay Aleksandrov2b3798d2014-01-22 14:53:17 +01004152 bond_mode = valptr->value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 }
4154
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004155 if (xmit_hash_policy) {
4156 if ((bond_mode != BOND_MODE_XOR) &&
Mahesh Bandewarf05b42e2014-04-22 16:30:20 -07004157 (bond_mode != BOND_MODE_8023AD) &&
4158 (bond_mode != BOND_MODE_TLB)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004159 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
Joe Perches90194262014-02-15 16:01:45 -08004160 bond_mode_name(bond_mode));
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004161 } else {
Nikolay Aleksandrova4b32ce2014-01-22 14:53:19 +01004162 bond_opt_initstr(&newval, xmit_hash_policy);
4163 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4164 &newval);
4165 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004166 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004167 xmit_hash_policy);
4168 return -EINVAL;
4169 }
Nikolay Aleksandrova4b32ce2014-01-22 14:53:19 +01004170 xmit_hashtype = valptr->value;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004171 }
4172 }
4173
Linus Torvalds1da177e2005-04-16 15:20:36 -07004174 if (lacp_rate) {
4175 if (bond_mode != BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004176 pr_info("lacp_rate param is irrelevant in mode %s\n",
4177 bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178 } else {
Nikolay Aleksandrovd3131de2014-01-22 14:53:27 +01004179 bond_opt_initstr(&newval, lacp_rate);
4180 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4181 &newval);
4182 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004183 pr_err("Error: Invalid lacp rate \"%s\"\n",
Nikolay Aleksandrovd3131de2014-01-22 14:53:27 +01004184 lacp_rate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185 return -EINVAL;
4186 }
Nikolay Aleksandrovd3131de2014-01-22 14:53:27 +01004187 lacp_fast = valptr->value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 }
4189 }
4190
Jay Vosburghfd989c82008-11-04 17:51:16 -08004191 if (ad_select) {
Nikolay Aleksandrov548d28b2014-07-13 09:47:47 +02004192 bond_opt_initstr(&newval, ad_select);
Nikolay Aleksandrov9e5f5ee2014-01-22 14:53:29 +01004193 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4194 &newval);
4195 if (!valptr) {
4196 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
Jay Vosburghfd989c82008-11-04 17:51:16 -08004197 return -EINVAL;
4198 }
Nikolay Aleksandrov9e5f5ee2014-01-22 14:53:29 +01004199 params->ad_select = valptr->value;
4200 if (bond_mode != BOND_MODE_8023AD)
Joe Perches91565eb2014-02-15 15:57:04 -08004201 pr_warn("ad_select param only affects 802.3ad mode\n");
Jay Vosburghfd989c82008-11-04 17:51:16 -08004202 } else {
4203 params->ad_select = BOND_AD_STABLE;
4204 }
4205
Nicolas de Pesloüanf5841302009-08-28 13:18:34 +00004206 if (max_bonds < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004207 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4208 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004209 max_bonds = BOND_DEFAULT_MAX_BONDS;
4210 }
4211
4212 if (miimon < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004213 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4214 miimon, INT_MAX);
Nikolay Aleksandrovb98d9c62014-01-22 14:53:31 +01004215 miimon = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004216 }
4217
4218 if (updelay < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004219 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4220 updelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004221 updelay = 0;
4222 }
4223
4224 if (downdelay < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004225 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4226 downdelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227 downdelay = 0;
4228 }
4229
4230 if ((use_carrier != 0) && (use_carrier != 1)) {
Joe Perches91565eb2014-02-15 15:57:04 -08004231 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4232 use_carrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004233 use_carrier = 1;
4234 }
4235
Ben Hutchingsad246c92011-04-26 15:25:52 +00004236 if (num_peer_notif < 0 || num_peer_notif > 255) {
Joe Perches91565eb2014-02-15 15:57:04 -08004237 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4238 num_peer_notif);
Ben Hutchingsad246c92011-04-26 15:25:52 +00004239 num_peer_notif = 1;
4240 }
4241
dingtianhong834db4b2013-12-21 14:40:17 +08004242 /* reset values for 802.3ad/TLB/ALB */
Veaceslav Falico267bed72014-05-15 21:39:53 +02004243 if (!bond_mode_uses_arp(bond_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244 if (!miimon) {
Joe Perches91565eb2014-02-15 15:57:04 -08004245 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");
4246 pr_warn("Forcing miimon to 100msec\n");
dingtianhongfe9d04a2013-11-22 22:28:43 +08004247 miimon = BOND_DEFAULT_MIIMON;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248 }
4249 }
4250
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004251 if (tx_queues < 1 || tx_queues > 255) {
Joe Perches91565eb2014-02-15 15:57:04 -08004252 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4253 tx_queues, BOND_DEFAULT_TX_QUEUES);
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004254 tx_queues = BOND_DEFAULT_TX_QUEUES;
4255 }
4256
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004257 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
Joe Perches91565eb2014-02-15 15:57:04 -08004258 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4259 all_slaves_active);
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004260 all_slaves_active = 0;
4261 }
4262
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004263 if (resend_igmp < 0 || resend_igmp > 255) {
Joe Perches91565eb2014-02-15 15:57:04 -08004264 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4265 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004266 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4267 }
4268
Nikolay Aleksandrovaa59d852014-01-22 14:53:18 +01004269 bond_opt_initval(&newval, packets_per_slave);
4270 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01004271 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4272 packets_per_slave, USHRT_MAX);
4273 packets_per_slave = 1;
4274 }
4275
Linus Torvalds1da177e2005-04-16 15:20:36 -07004276 if (bond_mode == BOND_MODE_ALB) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004277 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",
4278 updelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004279 }
4280
4281 if (!miimon) {
4282 if (updelay || downdelay) {
4283 /* just warn the user the up/down delay will have
4284 * no effect since miimon is zero...
4285 */
Joe Perches91565eb2014-02-15 15:57:04 -08004286 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",
4287 updelay, downdelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288 }
4289 } else {
4290 /* don't allow arp monitoring */
4291 if (arp_interval) {
Joe Perches91565eb2014-02-15 15:57:04 -08004292 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4293 miimon, arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004294 arp_interval = 0;
4295 }
4296
4297 if ((updelay % miimon) != 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004298 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4299 updelay, miimon, (updelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004300 }
4301
4302 updelay /= miimon;
4303
4304 if ((downdelay % miimon) != 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004305 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4306 downdelay, miimon,
4307 (downdelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308 }
4309
4310 downdelay /= miimon;
4311 }
4312
4313 if (arp_interval < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004314 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4315 arp_interval, INT_MAX);
Nikolay Aleksandrov7bdb04e2014-01-22 14:53:23 +01004316 arp_interval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004317 }
4318
nikolay@redhat.com5a5c5fd2013-05-18 01:18:30 +00004319 for (arp_ip_count = 0, i = 0;
4320 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
dingtianhong89015c12013-12-04 18:59:31 +08004321 __be32 ip;
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004322
4323 /* not a complete check, but good enough to catch mistakes */
dingtianhong89015c12013-12-04 18:59:31 +08004324 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
Veaceslav Falico2807a9f2014-05-15 21:39:56 +02004325 !bond_is_ip_target_ok(ip)) {
Joe Perches91565eb2014-02-15 15:57:04 -08004326 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4327 arp_ip_target[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004328 arp_interval = 0;
4329 } else {
Veaceslav Falico0afee4e2013-06-24 11:49:30 +02004330 if (bond_get_targets_ip(arp_target, ip) == -1)
4331 arp_target[arp_ip_count++] = ip;
4332 else
Joe Perches91565eb2014-02-15 15:57:04 -08004333 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4334 &ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335 }
4336 }
4337
4338 if (arp_interval && !arp_ip_count) {
4339 /* don't allow arping if no arp_ip_target given... */
Joe Perches91565eb2014-02-15 15:57:04 -08004340 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4341 arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004342 arp_interval = 0;
4343 }
4344
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004345 if (arp_validate) {
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004346 if (!arp_interval) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004347 pr_err("arp_validate requires arp_interval\n");
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004348 return -EINVAL;
4349 }
4350
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004351 bond_opt_initstr(&newval, arp_validate);
4352 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4353 &newval);
4354 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004355 pr_err("Error: invalid arp_validate \"%s\"\n",
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004356 arp_validate);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004357 return -EINVAL;
4358 }
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004359 arp_validate_value = valptr->value;
4360 } else {
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004361 arp_validate_value = 0;
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004362 }
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004363
Veaceslav Falico8599b522013-06-24 11:49:34 +02004364 arp_all_targets_value = 0;
4365 if (arp_all_targets) {
Nikolay Aleksandrovedf36b22014-01-22 14:53:21 +01004366 bond_opt_initstr(&newval, arp_all_targets);
4367 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4368 &newval);
4369 if (!valptr) {
Veaceslav Falico8599b522013-06-24 11:49:34 +02004370 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4371 arp_all_targets);
4372 arp_all_targets_value = 0;
Nikolay Aleksandrovedf36b22014-01-22 14:53:21 +01004373 } else {
4374 arp_all_targets_value = valptr->value;
Veaceslav Falico8599b522013-06-24 11:49:34 +02004375 }
4376 }
4377
Linus Torvalds1da177e2005-04-16 15:20:36 -07004378 if (miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004379 pr_info("MII link monitoring set to %d ms\n", miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004380 } else if (arp_interval) {
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004381 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4382 arp_validate_value);
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004383 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004384 arp_interval, valptr->string, arp_ip_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004385
4386 for (i = 0; i < arp_ip_count; i++)
Joe Perches90194262014-02-15 16:01:45 -08004387 pr_cont(" %s", arp_ip_target[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004388
Joe Perches90194262014-02-15 16:01:45 -08004389 pr_cont("\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004390
Jay Vosburghb8a97872008-06-13 18:12:04 -07004391 } else if (max_bonds) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392 /* miimon and arp_interval not set, we need one so things
4393 * work as expected, see bonding.txt for details
4394 */
Joe Perches90194262014-02-15 16:01:45 -08004395 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");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004396 }
4397
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02004398 if (primary && !bond_mode_uses_primary(bond_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004399 /* currently, using a primary only makes sense
4400 * in active backup, TLB or ALB modes
4401 */
Joe Perches91565eb2014-02-15 15:57:04 -08004402 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4403 primary, bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004404 primary = NULL;
4405 }
4406
Jiri Pirkoa5499522009-09-25 03:28:09 +00004407 if (primary && primary_reselect) {
Nikolay Aleksandrov388d3a62014-01-22 14:53:33 +01004408 bond_opt_initstr(&newval, primary_reselect);
4409 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4410 &newval);
4411 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004412 pr_err("Error: Invalid primary_reselect \"%s\"\n",
Nikolay Aleksandrov388d3a62014-01-22 14:53:33 +01004413 primary_reselect);
Jiri Pirkoa5499522009-09-25 03:28:09 +00004414 return -EINVAL;
4415 }
Nikolay Aleksandrov388d3a62014-01-22 14:53:33 +01004416 primary_reselect_value = valptr->value;
Jiri Pirkoa5499522009-09-25 03:28:09 +00004417 } else {
4418 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4419 }
4420
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004421 if (fail_over_mac) {
Nikolay Aleksandrov1df6b6a2014-01-22 14:53:22 +01004422 bond_opt_initstr(&newval, fail_over_mac);
4423 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4424 &newval);
4425 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004426 pr_err("Error: invalid fail_over_mac \"%s\"\n",
Nikolay Aleksandrov1df6b6a2014-01-22 14:53:22 +01004427 fail_over_mac);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004428 return -EINVAL;
4429 }
Nikolay Aleksandrov1df6b6a2014-01-22 14:53:22 +01004430 fail_over_mac_value = valptr->value;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004431 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
Joe Perches91565eb2014-02-15 15:57:04 -08004432 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004433 } else {
4434 fail_over_mac_value = BOND_FOM_NONE;
4435 }
Jay Vosburghdd957c52007-10-09 19:57:24 -07004436
Mahesh Bandewar6791e462015-05-09 00:01:55 -07004437 bond_opt_initstr(&newval, "default");
4438 valptr = bond_opt_parse(
4439 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4440 &newval);
4441 if (!valptr) {
4442 pr_err("Error: No ad_actor_sys_prio default value");
4443 return -EINVAL;
4444 }
4445 ad_actor_sys_prio = valptr->value;
4446
Mahesh Bandeward22a5fc2015-05-09 00:01:57 -07004447 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4448 &newval);
4449 if (!valptr) {
4450 pr_err("Error: No ad_user_port_key default value");
4451 return -EINVAL;
4452 }
4453 ad_user_port_key = valptr->value;
4454
dingtianhong3a7129e2013-12-21 14:40:12 +08004455 if (lp_interval == 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004456 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4457 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
dingtianhong3a7129e2013-12-21 14:40:12 +08004458 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4459 }
4460
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461 /* fill params struct with the proper values */
4462 params->mode = bond_mode;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004463 params->xmit_policy = xmit_hashtype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004464 params->miimon = miimon;
Ben Hutchingsad246c92011-04-26 15:25:52 +00004465 params->num_peer_notif = num_peer_notif;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004466 params->arp_interval = arp_interval;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004467 params->arp_validate = arp_validate_value;
Veaceslav Falico8599b522013-06-24 11:49:34 +02004468 params->arp_all_targets = arp_all_targets_value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004469 params->updelay = updelay;
4470 params->downdelay = downdelay;
4471 params->use_carrier = use_carrier;
4472 params->lacp_fast = lacp_fast;
4473 params->primary[0] = 0;
Jiri Pirkoa5499522009-09-25 03:28:09 +00004474 params->primary_reselect = primary_reselect_value;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004475 params->fail_over_mac = fail_over_mac_value;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004476 params->tx_queues = tx_queues;
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004477 params->all_slaves_active = all_slaves_active;
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004478 params->resend_igmp = resend_igmp;
stephen hemminger655f8912011-06-22 09:54:39 +00004479 params->min_links = min_links;
dingtianhong3a7129e2013-12-21 14:40:12 +08004480 params->lp_interval = lp_interval;
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01004481 params->packets_per_slave = packets_per_slave;
Mahesh Bandeware9f0fb82014-04-22 16:30:22 -07004482 params->tlb_dynamic_lb = 1; /* Default value */
Mahesh Bandewar6791e462015-05-09 00:01:55 -07004483 params->ad_actor_sys_prio = ad_actor_sys_prio;
Mahesh Bandewar74514952015-05-09 00:01:56 -07004484 eth_zero_addr(params->ad_actor_system);
Mahesh Bandeward22a5fc2015-05-09 00:01:57 -07004485 params->ad_user_port_key = ad_user_port_key;
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01004486 if (packets_per_slave > 0) {
4487 params->reciprocal_packets_per_slave =
4488 reciprocal_value(packets_per_slave);
4489 } else {
4490 /* reciprocal_packets_per_slave is unused if
4491 * packets_per_slave is 0 or 1, just initialize it
4492 */
4493 params->reciprocal_packets_per_slave =
4494 (struct reciprocal_value) { 0 };
4495 }
4496
Linus Torvalds1da177e2005-04-16 15:20:36 -07004497 if (primary) {
4498 strncpy(params->primary, primary, IFNAMSIZ);
4499 params->primary[IFNAMSIZ - 1] = 0;
4500 }
4501
4502 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4503
4504 return 0;
4505}
4506
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004507static struct lock_class_key bonding_netdev_xmit_lock_key;
David S. Millercf508b12008-07-22 14:16:42 -07004508static struct lock_class_key bonding_netdev_addr_lock_key;
Eric Dumazet49ee4922012-10-03 23:05:26 +00004509static struct lock_class_key bonding_tx_busylock_key;
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004510
David S. Millere8a04642008-07-17 00:34:19 -07004511static void bond_set_lockdep_class_one(struct net_device *dev,
4512 struct netdev_queue *txq,
4513 void *_unused)
David S. Millerc773e842008-07-08 23:13:53 -07004514{
4515 lockdep_set_class(&txq->_xmit_lock,
4516 &bonding_netdev_xmit_lock_key);
4517}
4518
4519static void bond_set_lockdep_class(struct net_device *dev)
4520{
David S. Millercf508b12008-07-22 14:16:42 -07004521 lockdep_set_class(&dev->addr_list_lock,
4522 &bonding_netdev_addr_lock_key);
David S. Millere8a04642008-07-17 00:34:19 -07004523 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
Eric Dumazet49ee4922012-10-03 23:05:26 +00004524 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
David S. Millerc773e842008-07-08 23:13:53 -07004525}
4526
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004527/* Called from registration process */
Stephen Hemminger181470f2009-06-12 19:02:52 +00004528static int bond_init(struct net_device *bond_dev)
4529{
4530 struct bonding *bond = netdev_priv(bond_dev);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004531 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004532
Veaceslav Falico76444f52014-07-15 19:35:58 +02004533 netdev_dbg(bond_dev, "Begin bond_init\n");
Stephen Hemminger181470f2009-06-12 19:02:52 +00004534
4535 bond->wq = create_singlethread_workqueue(bond_dev->name);
4536 if (!bond->wq)
4537 return -ENOMEM;
4538
4539 bond_set_lockdep_class(bond_dev);
4540
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004541 list_add_tail(&bond->bond_list, &bn->dev_list);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004542
Eric W. Biederman6151b3d2009-10-29 14:18:22 +00004543 bond_prepare_sysfs_group(bond);
Jiri Pirko22bedad32010-04-01 21:22:57 +00004544
Taku Izumif073c7c2010-12-09 15:17:13 +00004545 bond_debug_register(bond);
4546
Jiri Pirko409cc1f2013-01-30 11:08:11 +01004547 /* Ensure valid dev_addr */
4548 if (is_zero_ether_addr(bond_dev->dev_addr) &&
nikolay@redhat.com97a1e632013-06-26 17:13:38 +02004549 bond_dev->addr_assign_type == NET_ADDR_PERM)
Jiri Pirko409cc1f2013-01-30 11:08:11 +01004550 eth_hw_addr_random(bond_dev);
Jiri Pirko409cc1f2013-01-30 11:08:11 +01004551
Stephen Hemminger181470f2009-06-12 19:02:52 +00004552 return 0;
4553}
4554
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004555unsigned int bond_get_num_tx_queues(void)
Jiri Pirkod5da4512011-08-10 06:09:44 +00004556{
stephen hemmingerefacb302012-04-10 18:34:43 +00004557 return tx_queues;
Jiri Pirkod5da4512011-08-10 06:09:44 +00004558}
4559
Mitch Williamsdfe60392005-11-09 10:36:04 -08004560/* Create a new bond based on the specified name and bonding parameters.
Jay Vosburghe4b91c42007-01-19 18:15:31 -08004561 * If name is NULL, obtain a suitable "bond%d" name for us.
Mitch Williamsdfe60392005-11-09 10:36:04 -08004562 * Caller must NOT hold rtnl_lock; we need to release it here before we
4563 * set up our sysfs entries.
4564 */
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004565int bond_create(struct net *net, const char *name)
Mitch Williamsdfe60392005-11-09 10:36:04 -08004566{
4567 struct net_device *bond_dev;
Paie913fb22015-04-29 14:24:23 -04004568 struct bonding *bond;
4569 struct alb_bond_info *bond_info;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004570 int res;
4571
4572 rtnl_lock();
Jay Vosburgh027ea042008-01-17 16:25:02 -08004573
Jiri Pirko1c5cae82011-04-30 01:21:32 +00004574 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
Tom Gundersenc835a672014-07-14 16:37:24 +02004575 name ? name : "bond%d", NET_NAME_UNKNOWN,
Jiri Pirko1c5cae82011-04-30 01:21:32 +00004576 bond_setup, tx_queues);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004577 if (!bond_dev) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004578 pr_err("%s: eek! can't alloc netdev!\n", name);
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004579 rtnl_unlock();
4580 return -ENOMEM;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004581 }
4582
Paie913fb22015-04-29 14:24:23 -04004583 /*
4584 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4585 * It is set to 0 by default which is wrong.
4586 */
4587 bond = netdev_priv(bond_dev);
4588 bond_info = &(BOND_ALB_INFO(bond));
4589 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4590
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004591 dev_net_set(bond_dev, net);
Eric W. Biederman88ead972009-10-29 14:18:25 +00004592 bond_dev->rtnl_link_ops = &bond_link_ops;
4593
Mitch Williamsdfe60392005-11-09 10:36:04 -08004594 res = register_netdevice(bond_dev);
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004595
Phil Oestere826eaf2011-03-14 06:22:05 +00004596 netif_carrier_off(bond_dev);
4597
Mitch Williamsdfe60392005-11-09 10:36:04 -08004598 rtnl_unlock();
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004599 if (res < 0)
4600 bond_destructor(bond_dev);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004601 return res;
4602}
4603
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004604static int __net_init bond_net_init(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004605{
Eric W. Biederman15449742009-11-29 15:46:04 +00004606 struct bond_net *bn = net_generic(net, bond_net_id);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004607
4608 bn->net = net;
4609 INIT_LIST_HEAD(&bn->dev_list);
4610
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004611 bond_create_proc_dir(bn);
Eric W. Biederman4c224002011-10-12 21:56:25 +00004612 bond_create_sysfs(bn);
Veaceslav Falico87a7b842013-06-24 11:49:29 +02004613
Eric W. Biederman15449742009-11-29 15:46:04 +00004614 return 0;
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004615}
4616
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004617static void __net_exit bond_net_exit(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004618{
Eric W. Biederman15449742009-11-29 15:46:04 +00004619 struct bond_net *bn = net_generic(net, bond_net_id);
nikolay@redhat.com69b02162013-04-06 00:54:38 +00004620 struct bonding *bond, *tmp_bond;
4621 LIST_HEAD(list);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004622
Eric W. Biederman4c224002011-10-12 21:56:25 +00004623 bond_destroy_sysfs(bn);
nikolay@redhat.com69b02162013-04-06 00:54:38 +00004624
4625 /* Kill off any bonds created after unregistering bond rtnl ops */
4626 rtnl_lock();
4627 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4628 unregister_netdevice_queue(bond->dev, &list);
4629 unregister_netdevice_many(&list);
4630 rtnl_unlock();
Veaceslav Falico23fa5c22014-07-17 12:04:08 +02004631
4632 bond_destroy_proc_dir(bn);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004633}
4634
4635static struct pernet_operations bond_net_ops = {
4636 .init = bond_net_init,
4637 .exit = bond_net_exit,
Eric W. Biederman15449742009-11-29 15:46:04 +00004638 .id = &bond_net_id,
4639 .size = sizeof(struct bond_net),
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004640};
4641
Linus Torvalds1da177e2005-04-16 15:20:36 -07004642static int __init bonding_init(void)
4643{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004644 int i;
4645 int res;
4646
Amerigo Wangbd33acc2011-03-06 21:58:46 +00004647 pr_info("%s", bond_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004648
Mitch Williamsdfe60392005-11-09 10:36:04 -08004649 res = bond_check_params(&bonding_defaults);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004650 if (res)
Mitch Williamsdfe60392005-11-09 10:36:04 -08004651 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004652
Eric W. Biederman15449742009-11-29 15:46:04 +00004653 res = register_pernet_subsys(&bond_net_ops);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004654 if (res)
4655 goto out;
Jay Vosburgh027ea042008-01-17 16:25:02 -08004656
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004657 res = bond_netlink_init();
Eric W. Biederman88ead972009-10-29 14:18:25 +00004658 if (res)
Eric W. Biederman66391042009-10-29 23:58:54 +00004659 goto err_link;
Eric W. Biederman88ead972009-10-29 14:18:25 +00004660
Taku Izumif073c7c2010-12-09 15:17:13 +00004661 bond_create_debugfs();
4662
Linus Torvalds1da177e2005-04-16 15:20:36 -07004663 for (i = 0; i < max_bonds; i++) {
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004664 res = bond_create(&init_net, NULL);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004665 if (res)
4666 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004667 }
4668
Linus Torvalds1da177e2005-04-16 15:20:36 -07004669 register_netdevice_notifier(&bond_netdev_notifier);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004670out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004671 return res;
Eric W. Biederman88ead972009-10-29 14:18:25 +00004672err:
Thomas Richterdb298682014-04-09 12:52:59 +02004673 bond_destroy_debugfs();
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004674 bond_netlink_fini();
Eric W. Biederman66391042009-10-29 23:58:54 +00004675err_link:
Eric W. Biederman15449742009-11-29 15:46:04 +00004676 unregister_pernet_subsys(&bond_net_ops);
Eric W. Biederman88ead972009-10-29 14:18:25 +00004677 goto out;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004678
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679}
4680
4681static void __exit bonding_exit(void)
4682{
4683 unregister_netdevice_notifier(&bond_netdev_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004684
Taku Izumif073c7c2010-12-09 15:17:13 +00004685 bond_destroy_debugfs();
Pavel Emelyanovae68c392008-05-02 17:49:39 -07004686
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004687 bond_netlink_fini();
nikolay@redhat.comffcdedb2013-04-06 00:54:37 +00004688 unregister_pernet_subsys(&bond_net_ops);
Neil Hormane843fa52010-10-13 16:01:50 +00004689
4690#ifdef CONFIG_NET_POLL_CONTROLLER
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004691 /* Make sure we don't have an imbalance on our netpoll blocking */
Neil Hormanfb4fa762010-12-06 09:05:50 +00004692 WARN_ON(atomic_read(&netpoll_block_tx));
Neil Hormane843fa52010-10-13 16:01:50 +00004693#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004694}
4695
4696module_init(bonding_init);
4697module_exit(bonding_exit);
4698MODULE_LICENSE("GPL");
4699MODULE_VERSION(DRV_VERSION);
4700MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4701MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");