blob: f3bfcafc96a47f9e064e2736c0abb276a99b4bbb [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
Joe Perchesa4aee5c2009-12-13 20:06:07 -080034#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/kernel.h>
37#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040044#include <net/ip.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/ip.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040046#include <linux/tcp.h>
47#include <linux/udp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000056#include <linux/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <asm/dma.h>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000058#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#include <linux/errno.h>
60#include <linux/netdevice.h>
61#include <linux/inetdevice.h>
Jay Vosburgha816c7c2007-02-28 17:03:37 -080062#include <linux/igmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#include <linux/etherdevice.h>
64#include <linux/skbuff.h>
65#include <net/sock.h>
66#include <linux/rtnetlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <linux/smp.h>
68#include <linux/if_ether.h>
69#include <net/arp.h>
70#include <linux/mii.h>
71#include <linux/ethtool.h>
72#include <linux/if_vlan.h>
73#include <linux/if_bonding.h>
David Sterbab63bb732007-12-06 23:40:33 -080074#include <linux/jiffies.h>
Neil Hormane843fa52010-10-13 16:01:50 +000075#include <linux/preempt.h>
Jay Vosburghc3ade5c2005-06-26 17:52:20 -040076#include <net/route.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020077#include <net/net_namespace.h>
Eric W. Biedermanec87fd32009-10-29 14:18:26 +000078#include <net/netns/generic.h>
Eric Dumazet5ee31c682012-06-12 06:03:51 +000079#include <net/pkt_sched.h>
nikolay@redhat.com278b2082013-08-01 16:54:51 +020080#include <linux/rculist.h>
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +020081#include <net/flow_keys.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070082#include "bonding.h"
83#include "bond_3ad.h"
84#include "bond_alb.h"
85
86/*---------------------------- Module parameters ----------------------------*/
87
88/* monitor all links that often (in milliseconds). <=0 disables monitoring */
89#define BOND_LINK_MON_INTERV 0
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094static int miimon = BOND_LINK_MON_INTERV;
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);
181MODULE_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
Jiri Pirkoa5499522009-09-25 03:28:09 +0000210const struct bond_parm_tbl pri_reselect_tbl[] = {
211{ "always", BOND_PRI_RESELECT_ALWAYS},
212{ "better", BOND_PRI_RESELECT_BETTER},
213{ "failure", BOND_PRI_RESELECT_FAILURE},
214{ NULL, -1},
215};
216
Jay Vosburghfd989c82008-11-04 17:51:16 -0800217struct bond_parm_tbl ad_select_tbl[] = {
218{ "stable", BOND_AD_STABLE},
219{ "bandwidth", BOND_AD_BANDWIDTH},
220{ "count", BOND_AD_COUNT},
221{ NULL, -1},
222};
223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224/*-------------------------- Forward declarations ---------------------------*/
225
Stephen Hemminger181470f2009-06-12 19:02:52 +0000226static int bond_init(struct net_device *bond_dev);
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +0000227static void bond_uninit(struct net_device *bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
229/*---------------------------- General routines -----------------------------*/
230
Amerigo Wangbd33acc2011-03-06 21:58:46 +0000231const char *bond_mode_name(int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800233 static const char *names[] = {
234 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
235 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
236 [BOND_MODE_XOR] = "load balancing (xor)",
237 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000238 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800239 [BOND_MODE_TLB] = "transmit load balancing",
240 [BOND_MODE_ALB] = "adaptive load balancing",
241 };
242
Wang Sheng-Huif5280942013-07-24 14:53:26 +0800243 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 return "unknown";
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800245
246 return names[mode];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
249/*---------------------------------- VLAN -----------------------------------*/
250
251/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 * bond_dev_queue_xmit - Prepare skb for xmit.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000253 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 * @bond: bond device that got this skb for tx.
255 * @skb: hw accel VLAN tagged skb to transmit
256 * @slave_dev: slave that is supposed to xmit this skbuff
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 */
dingtianhongd316ded2014-01-02 09:13:09 +0800258void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000259 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
Ben Hutchings83874512010-12-13 08:19:28 +0000261 skb->dev = slave_dev;
Neil Horman374eeb52011-06-03 10:35:52 +0000262
Eric Dumazet5ee31c682012-06-12 06:03:51 +0000263 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
Jiri Pirkodf4ab5b2012-07-20 02:28:49 +0000264 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
265 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
Neil Horman374eeb52011-06-03 10:35:52 +0000266
Amerigo Wange15c3c222012-08-10 01:24:45 +0000267 if (unlikely(netpoll_tx_running(bond->dev)))
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000268 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
Amerigo Wang080e4132011-02-17 23:43:33 +0000269 else
WANG Congf6dc31a2010-05-06 00:48:51 -0700270 dev_queue_xmit(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
272
273/*
Jiri Pirkocc0e4072011-07-20 04:54:46 +0000274 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
275 * We don't protect the slave list iteration with a lock because:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 * a. This operation is performed in IOCTL context,
277 * b. The operation is protected by the RTNL semaphore in the 8021q code,
278 * c. Holding a lock with BH disabled while directly calling a base driver
279 * entry point is generally a BAD idea.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000280 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 * The design of synchronization/protection for this operation in the 8021q
282 * module is good for one or more VLAN devices over a single physical device
283 * and cannot be extended for a teaming solution like bonding, so there is a
284 * potential race condition here where a net device from the vlan group might
285 * be referenced (either by a base driver or the 8021q code) while it is being
286 * removed from the system. However, it turns out we're not making matters
287 * worse, and if it works for regular VLAN usage it will work here too.
288*/
289
290/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
292 * @bond_dev: bonding net device that got called
293 * @vid: vlan id being added
294 */
Patrick McHardy80d5c362013-04-19 02:04:28 +0000295static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
296 __be16 proto, u16 vid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297{
Wang Chen454d7c92008-11-12 23:37:49 -0800298 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico81f23b12013-09-25 09:20:13 +0200299 struct slave *slave, *rollback_slave;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200300 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200301 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200303 bond_for_each_slave(bond, slave, iter) {
Patrick McHardy80d5c362013-04-19 02:04:28 +0000304 res = vlan_vid_add(slave->dev, proto, vid);
Jiri Pirko87002b02011-12-08 04:11:17 +0000305 if (res)
306 goto unwind;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 }
308
Jiri Pirko8e586132011-12-08 19:52:37 -0500309 return 0;
Jiri Pirko87002b02011-12-08 04:11:17 +0000310
311unwind:
Veaceslav Falico81f23b12013-09-25 09:20:13 +0200312 /* unwind to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200313 bond_for_each_slave(bond, rollback_slave, iter) {
Veaceslav Falico81f23b12013-09-25 09:20:13 +0200314 if (rollback_slave == slave)
315 break;
316
317 vlan_vid_del(rollback_slave->dev, proto, vid);
318 }
Jiri Pirko87002b02011-12-08 04:11:17 +0000319
320 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
323/**
324 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
325 * @bond_dev: bonding net device that got called
326 * @vid: vlan id being removed
327 */
Patrick McHardy80d5c362013-04-19 02:04:28 +0000328static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
329 __be16 proto, u16 vid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
Wang Chen454d7c92008-11-12 23:37:49 -0800331 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200332 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200335 bond_for_each_slave(bond, slave, iter)
Patrick McHardy80d5c362013-04-19 02:04:28 +0000336 vlan_vid_del(slave->dev, proto, vid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Veaceslav Falicoe868b0c2013-08-28 23:25:15 +0200338 if (bond_is_lb(bond))
339 bond_alb_clear_vlan(bond, vid);
Jiri Pirko8e586132011-12-08 19:52:37 -0500340
341 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344/*------------------------------- Link status -------------------------------*/
345
346/*
Jay Vosburghff59c452006-03-27 13:27:43 -0800347 * Set the carrier state for the master according to the state of its
348 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
349 * do special 802.3ad magic.
350 *
351 * Returns zero if carrier state does not change, nonzero if it does.
352 */
353static int bond_set_carrier(struct bonding *bond)
354{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200355 struct list_head *iter;
Jay Vosburghff59c452006-03-27 13:27:43 -0800356 struct slave *slave;
Jay Vosburghff59c452006-03-27 13:27:43 -0800357
Veaceslav Falico0965a1f2013-09-25 09:20:21 +0200358 if (!bond_has_slaves(bond))
Jay Vosburghff59c452006-03-27 13:27:43 -0800359 goto down;
360
361 if (bond->params.mode == BOND_MODE_8023AD)
362 return bond_3ad_set_carrier(bond);
363
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200364 bond_for_each_slave(bond, slave, iter) {
Jay Vosburghff59c452006-03-27 13:27:43 -0800365 if (slave->link == BOND_LINK_UP) {
366 if (!netif_carrier_ok(bond->dev)) {
367 netif_carrier_on(bond->dev);
368 return 1;
369 }
370 return 0;
371 }
372 }
373
374down:
375 if (netif_carrier_ok(bond->dev)) {
376 netif_carrier_off(bond->dev);
377 return 1;
378 }
379 return 0;
380}
381
382/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 * Get link speed and duplex from the slave's base driver
384 * using ethtool. If for some reason the call fails or the
Weiping Pan98f41f62011-10-31 17:20:48 +0000385 * values are invalid, set speed and duplex to -1,
Rick Jones13b95fb2012-04-26 11:20:30 +0000386 * and return.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 */
Rick Jones13b95fb2012-04-26 11:20:30 +0000388static void bond_update_speed_duplex(struct slave *slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
390 struct net_device *slave_dev = slave->dev;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000391 struct ethtool_cmd ecmd;
David Decotigny5d305302011-04-13 15:22:31 +0000392 u32 slave_speed;
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700393 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
Dan Carpenter589665f2011-11-04 08:21:38 +0000395 slave->speed = SPEED_UNKNOWN;
396 slave->duplex = DUPLEX_UNKNOWN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000398 res = __ethtool_get_settings(slave_dev, &ecmd);
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700399 if (res < 0)
Rick Jones13b95fb2012-04-26 11:20:30 +0000400 return;
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700401
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000402 slave_speed = ethtool_cmd_speed(&ecmd);
Jiri Pirko6f92c662011-06-01 10:36:33 +0000403 if (slave_speed == 0 || slave_speed == ((__u32) -1))
Rick Jones13b95fb2012-04-26 11:20:30 +0000404 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000406 switch (ecmd.duplex) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 case DUPLEX_FULL:
408 case DUPLEX_HALF:
409 break;
410 default:
Rick Jones13b95fb2012-04-26 11:20:30 +0000411 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 }
413
David Decotigny5d305302011-04-13 15:22:31 +0000414 slave->speed = slave_speed;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000415 slave->duplex = ecmd.duplex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
Rick Jones13b95fb2012-04-26 11:20:30 +0000417 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418}
419
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -0800420const char *bond_slave_link_status(s8 link)
421{
422 switch (link) {
423 case BOND_LINK_UP:
424 return "up";
425 case BOND_LINK_FAIL:
426 return "going down";
427 case BOND_LINK_DOWN:
428 return "down";
429 case BOND_LINK_BACK:
430 return "going back";
431 default:
432 return "unknown";
433 }
434}
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436/*
437 * if <dev> supports MII link status reporting, check its link status.
438 *
439 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000440 * depending upon the setting of the use_carrier parameter.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 *
442 * Return either BMSR_LSTATUS, meaning that the link is up (or we
443 * can't tell and just pretend it is), or 0, meaning that the link is
444 * down.
445 *
446 * If reporting is non-zero, instead of faking link up, return -1 if
447 * both ETHTOOL and MII ioctls fail (meaning the device does not
448 * support them). If use_carrier is set, return whatever it says.
449 * It'd be nice if there was a good way to tell if a driver supports
450 * netif_carrier, but there really isn't.
451 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000452static int bond_check_dev_link(struct bonding *bond,
453 struct net_device *slave_dev, int reporting)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800455 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Jiri Bohacd9d52832009-10-28 22:23:54 -0700456 int (*ioctl)(struct net_device *, struct ifreq *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 struct ifreq ifr;
458 struct mii_ioctl_data *mii;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Petri Gynther6c988852009-08-28 12:05:15 +0000460 if (!reporting && !netif_running(slave_dev))
461 return 0;
462
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800463 if (bond->params.use_carrier)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Jiri Pirko29112f42009-04-24 01:58:23 +0000466 /* Try to get link status using Ethtool first. */
Ben Hutchingsc772dde2012-12-07 06:15:32 +0000467 if (slave_dev->ethtool_ops->get_link)
468 return slave_dev->ethtool_ops->get_link(slave_dev) ?
469 BMSR_LSTATUS : 0;
Jiri Pirko29112f42009-04-24 01:58:23 +0000470
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000471 /* Ethtool can't be used, fallback to MII ioctls. */
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800472 ioctl = slave_ops->ndo_do_ioctl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 if (ioctl) {
474 /* TODO: set pointer to correct ioctl on a per team member */
475 /* bases to make this more efficient. that is, once */
476 /* we determine the correct ioctl, we will always */
477 /* call it and not the others for that team */
478 /* member. */
479
480 /*
481 * We cannot assume that SIOCGMIIPHY will also read a
482 * register; not all network drivers (e.g., e100)
483 * support that.
484 */
485
486 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
487 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
488 mii = if_mii(&ifr);
489 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
490 mii->reg_num = MII_BMSR;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000491 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
492 return mii->val_out & BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 }
494 }
495
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700496 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 * If reporting, report that either there's no dev->do_ioctl,
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700498 * or both SIOCGMIIREG and get_link failed (meaning that we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 * cannot report link status). If not reporting, pretend
500 * we're ok.
501 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000502 return reporting ? -1 : BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503}
504
505/*----------------------------- Multicast list ------------------------------*/
506
507/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 * Push the promiscuity flag down to appropriate slaves
509 */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700510static int bond_set_promiscuity(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200512 struct list_head *iter;
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700513 int err = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 if (USES_PRIMARY(bond->params.mode)) {
516 /* write lock already acquired */
517 if (bond->curr_active_slave) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700518 err = dev_set_promiscuity(bond->curr_active_slave->dev,
519 inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 }
521 } else {
522 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200523
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200524 bond_for_each_slave(bond, slave, iter) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700525 err = dev_set_promiscuity(slave->dev, inc);
526 if (err)
527 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 }
529 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700530 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531}
532
533/*
534 * Push the allmulti flag down to all slaves
535 */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700536static int bond_set_allmulti(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200538 struct list_head *iter;
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700539 int err = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 if (USES_PRIMARY(bond->params.mode)) {
542 /* write lock already acquired */
543 if (bond->curr_active_slave) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700544 err = dev_set_allmulti(bond->curr_active_slave->dev,
545 inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 }
547 } else {
548 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200549
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200550 bond_for_each_slave(bond, slave, iter) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700551 err = dev_set_allmulti(slave->dev, inc);
552 if (err)
553 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 }
555 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700556 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800559/*
560 * Retrieve the list of registered multicast addresses for the bonding
561 * device and retransmit an IGMP JOIN request to the current active
562 * slave.
563 */
stephen hemminger379b7382010-10-15 11:02:56 +0000564static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000565{
566 struct bonding *bond = container_of(work, struct bonding,
Flavio Leitner94265cf2011-05-25 08:38:58 +0000567 mcast_work.work);
Veaceslav Falicoad999ee2013-03-26 04:10:02 +0000568
dingtianhongf2369102013-12-13 10:20:26 +0800569 if (!rtnl_trylock()) {
570 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
571 return;
572 }
573 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
574
575 if (bond->igmp_retrans > 1) {
576 bond->igmp_retrans--;
577 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
578 }
579 rtnl_unlock();
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800580}
581
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000582/* Flush bond's hardware addresses from slave
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 */
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000584static void bond_hw_addr_flush(struct net_device *bond_dev,
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000585 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586{
Wang Chen454d7c92008-11-12 23:37:49 -0800587 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000589 dev_uc_unsync(slave_dev, bond_dev);
590 dev_mc_unsync(slave_dev, bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 if (bond->params.mode == BOND_MODE_8023AD) {
593 /* del lacpdu mc addr from mc list */
594 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
595
Jiri Pirko22bedad32010-04-01 21:22:57 +0000596 dev_mc_del(slave_dev, lacpdu_multicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 }
598}
599
600/*--------------------------- Active slave change ---------------------------*/
601
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000602/* Update the hardware address list and promisc/allmulti for the new and
603 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
604 * slaves up date at all times; only the USES_PRIMARY modes need to call
605 * this function to swap these settings during a failover.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 */
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000607static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
608 struct slave *old_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Veaceslav Falico7864a1a2013-08-05 14:56:06 +0200610 ASSERT_RTNL();
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 if (old_active) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000613 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 dev_set_promiscuity(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000616 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 dev_set_allmulti(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000619 bond_hw_addr_flush(bond->dev, old_active->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 }
621
622 if (new_active) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700623 /* FIXME: Signal errors upstream. */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000624 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 dev_set_promiscuity(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000627 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 dev_set_allmulti(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
nikolay@redhat.comd632ce92013-04-18 07:33:38 +0000630 netif_addr_lock_bh(bond->dev);
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000631 dev_uc_sync(new_active->dev, bond->dev);
632 dev_mc_sync(new_active->dev, bond->dev);
nikolay@redhat.comd632ce92013-04-18 07:33:38 +0000633 netif_addr_unlock_bh(bond->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635}
636
nikolay@redhat.comae0d6752013-06-26 17:13:39 +0200637/**
638 * bond_set_dev_addr - clone slave's address to bond
639 * @bond_dev: bond net device
640 * @slave_dev: slave net device
641 *
642 * Should be called with RTNL held.
643 */
644static void bond_set_dev_addr(struct net_device *bond_dev,
645 struct net_device *slave_dev)
646{
Nikolay Aleksandrov008aebd2013-06-29 13:16:59 +0200647 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
648 bond_dev, slave_dev, slave_dev->addr_len);
nikolay@redhat.comae0d6752013-06-26 17:13:39 +0200649 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
650 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
651 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
652}
653
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700654/*
655 * bond_do_fail_over_mac
656 *
657 * Perform special MAC address swapping for fail_over_mac settings
658 *
dingtianhongb2e7ace2013-12-13 10:19:32 +0800659 * Called with RTNL, curr_slave_lock for write_bh.
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700660 */
661static void bond_do_fail_over_mac(struct bonding *bond,
662 struct slave *new_active,
663 struct slave *old_active)
Hannes Eder1f78d9f2009-02-14 11:15:33 +0000664 __releases(&bond->curr_slave_lock)
Hannes Eder1f78d9f2009-02-14 11:15:33 +0000665 __acquires(&bond->curr_slave_lock)
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700666{
667 u8 tmp_mac[ETH_ALEN];
668 struct sockaddr saddr;
669 int rv;
670
671 switch (bond->params.fail_over_mac) {
672 case BOND_FOM_ACTIVE:
Weiping Pan7d26bb12012-03-27 19:18:24 +0000673 if (new_active) {
Weiping Pan7d26bb12012-03-27 19:18:24 +0000674 write_unlock_bh(&bond->curr_slave_lock);
nikolay@redhat.comae0d6752013-06-26 17:13:39 +0200675 bond_set_dev_addr(bond->dev, new_active->dev);
Weiping Pan7d26bb12012-03-27 19:18:24 +0000676 write_lock_bh(&bond->curr_slave_lock);
677 }
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700678 break;
679 case BOND_FOM_FOLLOW:
680 /*
681 * if new_active && old_active, swap them
682 * if just old_active, do nothing (going to no active slave)
683 * if just new_active, set new_active to bond's MAC
684 */
685 if (!new_active)
686 return;
687
688 write_unlock_bh(&bond->curr_slave_lock);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700689
690 if (old_active) {
691 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
692 memcpy(saddr.sa_data, old_active->dev->dev_addr,
693 ETH_ALEN);
694 saddr.sa_family = new_active->dev->type;
695 } else {
696 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
697 saddr.sa_family = bond->dev->type;
698 }
699
700 rv = dev_set_mac_address(new_active->dev, &saddr);
701 if (rv) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800702 pr_err("%s: Error %d setting MAC of slave %s\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700703 bond->dev->name, -rv, new_active->dev->name);
704 goto out;
705 }
706
707 if (!old_active)
708 goto out;
709
710 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
711 saddr.sa_family = old_active->dev->type;
712
713 rv = dev_set_mac_address(old_active->dev, &saddr);
714 if (rv)
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800715 pr_err("%s: Error %d setting MAC of slave %s\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700716 bond->dev->name, -rv, new_active->dev->name);
717out:
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700718 write_lock_bh(&bond->curr_slave_lock);
719 break;
720 default:
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800721 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700722 bond->dev->name, bond->params.fail_over_mac);
723 break;
724 }
725
726}
727
Jiri Pirkoa5499522009-09-25 03:28:09 +0000728static bool bond_should_change_active(struct bonding *bond)
729{
730 struct slave *prim = bond->primary_slave;
731 struct slave *curr = bond->curr_active_slave;
732
733 if (!prim || !curr || curr->link != BOND_LINK_UP)
734 return true;
735 if (bond->force_primary) {
736 bond->force_primary = false;
737 return true;
738 }
739 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
740 (prim->speed < curr->speed ||
741 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
742 return false;
743 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
744 return false;
745 return true;
746}
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748/**
749 * find_best_interface - select the best available slave to be the active one
750 * @bond: our bonding struct
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 */
752static struct slave *bond_find_best_slave(struct bonding *bond)
753{
Veaceslav Falico77140d22013-09-25 09:20:18 +0200754 struct slave *slave, *bestslave = NULL;
755 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 int mintime = bond->params.updelay;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Veaceslav Falico77140d22013-09-25 09:20:18 +0200758 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
759 bond_should_change_active(bond))
760 return bond->primary_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Veaceslav Falico77140d22013-09-25 09:20:18 +0200762 bond_for_each_slave(bond, slave, iter) {
763 if (slave->link == BOND_LINK_UP)
764 return slave;
765 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
766 slave->delay < mintime) {
767 mintime = slave->delay;
768 bestslave = slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 }
770 }
771
772 return bestslave;
773}
774
Ben Hutchingsad246c92011-04-26 15:25:52 +0000775static bool bond_should_notify_peers(struct bonding *bond)
776{
dingtianhong4cb4f972013-12-13 10:19:39 +0800777 struct slave *slave;
778
779 rcu_read_lock();
780 slave = rcu_dereference(bond->curr_active_slave);
781 rcu_read_unlock();
Ben Hutchingsad246c92011-04-26 15:25:52 +0000782
783 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
784 bond->dev->name, slave ? slave->dev->name : "NULL");
785
786 if (!slave || !bond->send_peer_notif ||
787 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
788 return false;
789
Ben Hutchingsad246c92011-04-26 15:25:52 +0000790 return true;
791}
792
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793/**
794 * change_active_interface - change the active slave into the specified one
795 * @bond: our bonding struct
796 * @new: the new slave to make the active one
797 *
798 * Set the new slave to the bond's settings and unset them on the old
799 * curr_active_slave.
800 * Setting include flags, mc-list, promiscuity, allmulti, etc.
801 *
802 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
803 * because it is apparently the best available slave we have, even though its
804 * updelay hasn't timed out yet.
805 *
dingtianhongb2e7ace2013-12-13 10:19:32 +0800806 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 */
Mitch Williamsa77b5322005-11-09 10:35:51 -0800808void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809{
810 struct slave *old_active = bond->curr_active_slave;
811
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000812 if (old_active == new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815 if (new_active) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -0700816 new_active->jiffies = jiffies;
817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if (new_active->link == BOND_LINK_BACK) {
819 if (USES_PRIMARY(bond->params.mode)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800820 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
821 bond->dev->name, new_active->dev->name,
822 (bond->params.updelay - new_active->delay) * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 }
824
825 new_active->delay = 0;
826 new_active->link = BOND_LINK_UP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000828 if (bond->params.mode == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
Holger Eitzenberger58402052008-12-09 23:07:13 -0800831 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 } else {
834 if (USES_PRIMARY(bond->params.mode)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800835 pr_info("%s: making interface %s the new active one.\n",
836 bond->dev->name, new_active->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 }
838 }
839 }
840
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000841 if (USES_PRIMARY(bond->params.mode))
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000842 bond_hw_addr_swap(bond, new_active, old_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Holger Eitzenberger58402052008-12-09 23:07:13 -0800844 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 bond_alb_handle_active_change(bond, new_active);
Jay Vosburgh8f903c72006-02-21 16:36:44 -0800846 if (old_active)
847 bond_set_slave_inactive_flags(old_active);
848 if (new_active)
849 bond_set_slave_active_flags(new_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 } else {
nikolay@redhat.com278b2082013-08-01 16:54:51 +0200851 rcu_assign_pointer(bond->curr_active_slave, new_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400853
854 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000855 if (old_active)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400856 bond_set_slave_inactive_flags(old_active);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400857
858 if (new_active) {
Ben Hutchingsad246c92011-04-26 15:25:52 +0000859 bool should_notify_peers = false;
860
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400861 bond_set_slave_active_flags(new_active);
Moni Shoua2ab82852007-10-09 19:43:39 -0700862
Or Gerlitz709f8a42008-06-13 18:12:01 -0700863 if (bond->params.fail_over_mac)
864 bond_do_fail_over_mac(bond, new_active,
865 old_active);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700866
Ben Hutchingsad246c92011-04-26 15:25:52 +0000867 if (netif_running(bond->dev)) {
868 bond->send_peer_notif =
869 bond->params.num_peer_notif;
870 should_notify_peers =
871 bond_should_notify_peers(bond);
872 }
873
Or Gerlitz01f31092008-06-13 18:12:02 -0700874 write_unlock_bh(&bond->curr_slave_lock);
Or Gerlitz01f31092008-06-13 18:12:02 -0700875
Amerigo Wangb7bc2a52012-08-09 22:14:57 +0000876 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
Ben Hutchingsad246c92011-04-26 15:25:52 +0000877 if (should_notify_peers)
Amerigo Wangb7bc2a52012-08-09 22:14:57 +0000878 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
879 bond->dev);
Or Gerlitz01f31092008-06-13 18:12:02 -0700880
Or Gerlitz01f31092008-06-13 18:12:02 -0700881 write_lock_bh(&bond->curr_slave_lock);
Moni Shoua7893b242008-05-17 21:10:12 -0700882 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400883 }
Andy Gospodareka2fd9402010-03-25 14:49:05 +0000884
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000885 /* resend IGMP joins since active slave has changed or
Flavio Leitner94265cf2011-05-25 08:38:58 +0000886 * all were sent on curr_active_slave.
887 * resend only if bond is brought up with the affected
888 * bonding modes and the retransmission is enabled */
889 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
890 ((USES_PRIMARY(bond->params.mode) && new_active) ||
891 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000892 bond->igmp_retrans = bond->params.resend_igmp;
Nikolay Aleksandrov4beac022013-08-01 11:51:42 +0200893 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
Andy Gospodareka2fd9402010-03-25 14:49:05 +0000894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
896
897/**
898 * bond_select_active_slave - select a new active slave, if needed
899 * @bond: our bonding struct
900 *
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000901 * This functions should be called when one of the following occurs:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 * - The old curr_active_slave has been released or lost its link.
903 * - The primary_slave has got its link back.
904 * - A slave has got its link back and there's no old curr_active_slave.
905 *
dingtianhongb2e7ace2013-12-13 10:19:32 +0800906 * Caller must hold curr_slave_lock for write_bh.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 */
Mitch Williamsa77b5322005-11-09 10:35:51 -0800908void bond_select_active_slave(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909{
910 struct slave *best_slave;
Jay Vosburghff59c452006-03-27 13:27:43 -0800911 int rv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
913 best_slave = bond_find_best_slave(bond);
914 if (best_slave != bond->curr_active_slave) {
915 bond_change_active_slave(bond, best_slave);
Jay Vosburghff59c452006-03-27 13:27:43 -0800916 rv = bond_set_carrier(bond);
917 if (!rv)
918 return;
919
920 if (netif_carrier_ok(bond->dev)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800921 pr_info("%s: first active interface up!\n",
922 bond->dev->name);
Jay Vosburghff59c452006-03-27 13:27:43 -0800923 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800924 pr_info("%s: now running without any active interface !\n",
925 bond->dev->name);
Jay Vosburghff59c452006-03-27 13:27:43 -0800926 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 }
928}
929
WANG Congf6dc31a2010-05-06 00:48:51 -0700930#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000931static inline int slave_enable_netpoll(struct slave *slave)
WANG Congf6dc31a2010-05-06 00:48:51 -0700932{
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000933 struct netpoll *np;
934 int err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -0700935
Amerigo Wang47be03a22012-08-10 01:24:37 +0000936 np = kzalloc(sizeof(*np), GFP_ATOMIC);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000937 err = -ENOMEM;
938 if (!np)
939 goto out;
940
Amerigo Wang47be03a22012-08-10 01:24:37 +0000941 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000942 if (err) {
943 kfree(np);
944 goto out;
WANG Congf6dc31a2010-05-06 00:48:51 -0700945 }
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000946 slave->np = np;
947out:
948 return err;
949}
950static inline void slave_disable_netpoll(struct slave *slave)
951{
952 struct netpoll *np = slave->np;
953
954 if (!np)
955 return;
956
957 slave->np = NULL;
Neil Horman2cde6ac2013-02-11 10:25:30 +0000958 __netpoll_free_async(np);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000959}
960static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
961{
962 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
963 return false;
964 if (!slave_dev->netdev_ops->ndo_poll_controller)
965 return false;
966 return true;
WANG Congf6dc31a2010-05-06 00:48:51 -0700967}
968
969static void bond_poll_controller(struct net_device *bond_dev)
970{
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000971}
972
dingtianhongc4cdef92013-07-23 15:25:27 +0800973static void bond_netpoll_cleanup(struct net_device *bond_dev)
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000974{
dingtianhongc4cdef92013-07-23 15:25:27 +0800975 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200976 struct list_head *iter;
Neil Hormanc2355e12010-10-13 16:01:49 +0000977 struct slave *slave;
Neil Hormanc2355e12010-10-13 16:01:49 +0000978
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200979 bond_for_each_slave(bond, slave, iter)
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000980 if (IS_UP(slave->dev))
981 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -0700982}
WANG Congf6dc31a2010-05-06 00:48:51 -0700983
Amerigo Wang47be03a22012-08-10 01:24:37 +0000984static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000985{
986 struct bonding *bond = netdev_priv(dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200987 struct list_head *iter;
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000988 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200989 int err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -0700990
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200991 bond_for_each_slave(bond, slave, iter) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000992 err = slave_enable_netpoll(slave);
993 if (err) {
dingtianhongc4cdef92013-07-23 15:25:27 +0800994 bond_netpoll_cleanup(dev);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000995 break;
996 }
997 }
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000998 return err;
999}
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001000#else
1001static inline int slave_enable_netpoll(struct slave *slave)
1002{
1003 return 0;
1004}
1005static inline void slave_disable_netpoll(struct slave *slave)
1006{
1007}
WANG Congf6dc31a2010-05-06 00:48:51 -07001008static void bond_netpoll_cleanup(struct net_device *bond_dev)
1009{
1010}
WANG Congf6dc31a2010-05-06 00:48:51 -07001011#endif
1012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013/*---------------------------------- IOCTL ----------------------------------*/
1014
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001015static netdev_features_t bond_fix_features(struct net_device *dev,
nikolay@redhat.com9b7b1652013-09-02 13:51:41 +02001016 netdev_features_t features)
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001017{
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001018 struct bonding *bond = netdev_priv(dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001019 struct list_head *iter;
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001020 netdev_features_t mask;
nikolay@redhat.com9b7b1652013-09-02 13:51:41 +02001021 struct slave *slave;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001022
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001023 if (!bond_has_slaves(bond)) {
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001024 /* Disable adding VLANs to empty bond. But why? --mq */
1025 features |= NETIF_F_VLAN_CHALLENGED;
nikolay@redhat.com9b7b1652013-09-02 13:51:41 +02001026 return features;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001027 }
1028
1029 mask = features;
1030 features &= ~NETIF_F_ONE_FOR_ALL;
1031 features |= NETIF_F_ALL_FOR_ALL;
1032
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001033 bond_for_each_slave(bond, slave, iter) {
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001034 features = netdev_increment_features(features,
1035 slave->dev->features,
1036 mask);
1037 }
Eric Dumazetb0ce3502013-05-16 07:34:53 +00001038 features = netdev_add_tso_features(features, mask);
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001039
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001040 return features;
1041}
1042
Michał Mirosław62f2a3a2011-07-13 14:10:29 +00001043#define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1044 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1045 NETIF_F_HIGHDMA | NETIF_F_LRO)
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001046
1047static void bond_compute_features(struct bonding *bond)
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001048{
nikolay@redhat.comc4826862013-09-02 13:51:42 +02001049 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001050 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001051 struct net_device *bond_dev = bond->dev;
1052 struct list_head *iter;
1053 struct slave *slave;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001054 unsigned short max_hard_header_len = ETH_HLEN;
Sarveshwar Bandi0e376bd2012-11-21 04:35:03 +00001055 unsigned int gso_max_size = GSO_MAX_SIZE;
1056 u16 gso_max_segs = GSO_MAX_SEGS;
Herbert Xub63365a2008-10-23 01:11:29 -07001057
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001058 if (!bond_has_slaves(bond))
Herbert Xub63365a2008-10-23 01:11:29 -07001059 goto done;
1060
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001061 bond_for_each_slave(bond, slave, iter) {
Jay Vosburgh278339a2009-08-28 12:05:12 +00001062 vlan_features = netdev_increment_features(vlan_features,
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001063 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1064
Eric Dumazetb6fe83e2012-07-17 12:19:48 +00001065 dst_release_flag &= slave->dev->priv_flags;
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001066 if (slave->dev->hard_header_len > max_hard_header_len)
1067 max_hard_header_len = slave->dev->hard_header_len;
Sarveshwar Bandi0e376bd2012-11-21 04:35:03 +00001068
1069 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1070 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001071 }
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001072
Herbert Xub63365a2008-10-23 01:11:29 -07001073done:
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001074 bond_dev->vlan_features = vlan_features;
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 Dumazetb6fe83e2012-07-17 12:19:48 +00001079 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1080 bond_dev->priv_flags = flags | dst_release_flag;
1081
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001082 netdev_change_features(bond_dev);
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001083}
1084
Moni Shoua872254d2007-10-09 19:43:38 -07001085static void bond_setup_by_slave(struct net_device *bond_dev,
1086 struct net_device *slave_dev)
1087{
Stephen Hemminger00829822008-11-20 20:14:53 -08001088 bond_dev->header_ops = slave_dev->header_ops;
Moni Shoua872254d2007-10-09 19:43:38 -07001089
1090 bond_dev->type = slave_dev->type;
1091 bond_dev->hard_header_len = slave_dev->hard_header_len;
1092 bond_dev->addr_len = slave_dev->addr_len;
1093
1094 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1095 slave_dev->addr_len);
1096}
1097
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001098/* On bonding slaves other than the currently active slave, suppress
Jiri Pirko3aba8912011-04-19 03:48:16 +00001099 * duplicates except for alb non-mcast/bcast.
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001100 */
1101static bool bond_should_deliver_exact_match(struct sk_buff *skb,
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001102 struct slave *slave,
1103 struct bonding *bond)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001104{
Jiri Pirko2d7011c2011-03-16 08:46:43 +00001105 if (bond_is_slave_inactive(slave)) {
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001106 if (bond->params.mode == BOND_MODE_ALB &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001107 skb->pkt_type != PACKET_BROADCAST &&
1108 skb->pkt_type != PACKET_MULTICAST)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001109 return false;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001110 return true;
1111 }
1112 return false;
1113}
1114
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001115static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001116{
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001117 struct sk_buff *skb = *pskb;
Jiri Pirkof1c17752011-03-12 03:14:35 +00001118 struct slave *slave;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001119 struct bonding *bond;
Eric Dumazetde063b72012-06-11 19:23:07 +00001120 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1121 struct slave *);
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00001122 int ret = RX_HANDLER_ANOTHER;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001123
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001124 skb = skb_share_check(skb, GFP_ATOMIC);
1125 if (unlikely(!skb))
1126 return RX_HANDLER_CONSUMED;
1127
1128 *pskb = skb;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001129
Jiri Pirko35d48902011-03-22 02:38:12 +00001130 slave = bond_slave_get_rcu(skb->dev);
1131 bond = slave->bond;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001132
1133 if (bond->params.arp_interval)
Jiri Pirkof1c17752011-03-12 03:14:35 +00001134 slave->dev->last_rx = jiffies;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001135
Mitsuo Hayasaka4d974802011-10-12 16:04:29 +00001136 recv_probe = ACCESS_ONCE(bond->recv_probe);
1137 if (recv_probe) {
Eric Dumazetde063b72012-06-11 19:23:07 +00001138 ret = recv_probe(skb, bond, slave);
1139 if (ret == RX_HANDLER_CONSUMED) {
1140 consume_skb(skb);
1141 return ret;
Jiri Pirko3aba8912011-04-19 03:48:16 +00001142 }
1143 }
1144
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001145 if (bond_should_deliver_exact_match(skb, slave, bond)) {
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001146 return RX_HANDLER_EXACT;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001147 }
1148
Jiri Pirko35d48902011-03-22 02:38:12 +00001149 skb->dev = bond->dev;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001150
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001151 if (bond->params.mode == BOND_MODE_ALB &&
Jiri Pirko35d48902011-03-22 02:38:12 +00001152 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001153 skb->pkt_type == PACKET_HOST) {
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001154
Changli Gao541ac7c2011-03-02 21:07:14 +00001155 if (unlikely(skb_cow_head(skb,
1156 skb->data - skb_mac_header(skb)))) {
1157 kfree_skb(skb);
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001158 return RX_HANDLER_CONSUMED;
Changli Gao541ac7c2011-03-02 21:07:14 +00001159 }
Jiri Pirko35d48902011-03-22 02:38:12 +00001160 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001161 }
1162
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00001163 return ret;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001164}
1165
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001166static int bond_master_upper_dev_link(struct net_device *bond_dev,
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001167 struct net_device *slave_dev,
1168 struct slave *slave)
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001169{
1170 int err;
1171
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001172 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001173 if (err)
1174 return err;
1175 slave_dev->flags |= IFF_SLAVE;
Alexei Starovoitov7f294052013-10-23 16:02:42 -07001176 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001177 return 0;
1178}
1179
1180static void bond_upper_dev_unlink(struct net_device *bond_dev,
1181 struct net_device *slave_dev)
1182{
1183 netdev_upper_dev_unlink(slave_dev, bond_dev);
1184 slave_dev->flags &= ~IFF_SLAVE;
Alexei Starovoitov7f294052013-10-23 16:02:42 -07001185 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001186}
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188/* enslave device <slave> to bond device <master> */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001189int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
Wang Chen454d7c92008-11-12 23:37:49 -08001191 struct bonding *bond = netdev_priv(bond_dev);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001192 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Veaceslav Falicoc8c23902013-09-25 09:20:25 +02001193 struct slave *new_slave = NULL, *prev_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 struct sockaddr addr;
1195 int link_reporting;
Veaceslav Falico8599b522013-06-24 11:49:34 +02001196 int res = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Ben Hutchingsc772dde2012-12-07 06:15:32 +00001198 if (!bond->params.use_carrier &&
1199 slave_dev->ethtool_ops->get_link == NULL &&
1200 slave_ops->ndo_do_ioctl == NULL) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001201 pr_warning("%s: Warning: no link monitoring support for %s\n",
1202 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 }
1204
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 /* already enslaved */
1206 if (slave_dev->flags & IFF_SLAVE) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001207 pr_debug("Error, Device was already enslaved\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 return -EBUSY;
1209 }
1210
1211 /* vlan challenged mutual exclusion */
1212 /* no need to lock since we're protected by rtnl_lock */
1213 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001214 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
Jiri Pirko55462cf2012-10-14 04:30:56 +00001215 if (vlan_uses_dev(bond_dev)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001216 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1217 bond_dev->name, slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 return -EPERM;
1219 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001220 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1221 bond_dev->name, slave_dev->name,
1222 slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 }
1224 } else {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001225 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 }
1227
Jay Vosburgh217df672005-09-26 16:11:50 -07001228 /*
1229 * Old ifenslave binaries are no longer supported. These can
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001230 * be identified with moderate accuracy by the state of the slave:
Jay Vosburgh217df672005-09-26 16:11:50 -07001231 * the current ifenslave will set the interface down prior to
1232 * enslaving it; the old ifenslave will not.
1233 */
1234 if ((slave_dev->flags & IFF_UP)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001235 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
Jay Vosburgh217df672005-09-26 16:11:50 -07001236 slave_dev->name);
1237 res = -EPERM;
1238 goto err_undo_flags;
1239 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
Moni Shoua872254d2007-10-09 19:43:38 -07001241 /* set bonding device ether type by slave - bonding netdevices are
1242 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1243 * there is a need to override some of the type dependent attribs/funcs.
1244 *
1245 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1246 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1247 */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001248 if (!bond_has_slaves(bond)) {
Moni Shouae36b9d12009-07-15 04:56:31 +00001249 if (bond_dev->type != slave_dev->type) {
Moni Shouae36b9d12009-07-15 04:56:31 +00001250 pr_debug("%s: change device type from %d to %d\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001251 bond_dev->name,
1252 bond_dev->type, slave_dev->type);
Moni Shoua75c78502009-09-15 02:37:40 -07001253
Amerigo Wangb7bc2a52012-08-09 22:14:57 +00001254 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1255 bond_dev);
Jiri Pirko3ca5b402010-03-10 10:29:35 +00001256 res = notifier_to_errno(res);
1257 if (res) {
1258 pr_err("%s: refused to change device type\n",
1259 bond_dev->name);
1260 res = -EBUSY;
1261 goto err_undo_flags;
1262 }
Moni Shoua75c78502009-09-15 02:37:40 -07001263
Jiri Pirko32a806c2010-03-19 04:00:23 +00001264 /* Flush unicast and multicast addresses */
Jiri Pirkoa748ee22010-04-01 21:22:09 +00001265 dev_uc_flush(bond_dev);
Jiri Pirko22bedad32010-04-01 21:22:57 +00001266 dev_mc_flush(bond_dev);
Jiri Pirko32a806c2010-03-19 04:00:23 +00001267
Moni Shouae36b9d12009-07-15 04:56:31 +00001268 if (slave_dev->type != ARPHRD_ETHER)
1269 bond_setup_by_slave(bond_dev, slave_dev);
Neil Horman550fd082011-07-26 06:05:38 +00001270 else {
Moni Shouae36b9d12009-07-15 04:56:31 +00001271 ether_setup(bond_dev);
Neil Horman550fd082011-07-26 06:05:38 +00001272 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1273 }
Moni Shoua75c78502009-09-15 02:37:40 -07001274
Amerigo Wangb7bc2a52012-08-09 22:14:57 +00001275 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1276 bond_dev);
Moni Shouae36b9d12009-07-15 04:56:31 +00001277 }
Moni Shoua872254d2007-10-09 19:43:38 -07001278 } else if (bond_dev->type != slave_dev->type) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001279 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1280 slave_dev->name,
1281 slave_dev->type, bond_dev->type);
1282 res = -EINVAL;
1283 goto err_undo_flags;
Moni Shoua872254d2007-10-09 19:43:38 -07001284 }
1285
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001286 if (slave_ops->ndo_set_mac_address == NULL) {
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001287 if (!bond_has_slaves(bond)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001288 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1289 bond_dev->name);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001290 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1291 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001292 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1293 bond_dev->name);
Moni Shoua2ab82852007-10-09 19:43:39 -07001294 res = -EOPNOTSUPP;
1295 goto err_undo_flags;
1296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 }
1298
Amerigo Wang8d8fc292011-05-19 21:39:10 +00001299 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1300
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001301 /* If this is the first slave, then we need to set the master's hardware
1302 * address to be the same as the slave's. */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001303 if (!bond_has_slaves(bond) &&
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001304 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
Jiri Pirko409cc1f2013-01-30 11:08:11 +01001305 bond_set_dev_addr(bond->dev, slave_dev);
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001306
Joe Jin243cb4e2007-02-06 14:16:40 -08001307 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 if (!new_slave) {
1309 res = -ENOMEM;
1310 goto err_undo_flags;
1311 }
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00001312 /*
1313 * Set the new_slave's queue_id to be zero. Queue ID mapping
1314 * is set via sysfs or module option if desired.
1315 */
1316 new_slave->queue_id = 0;
1317
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001318 /* Save slave's original mtu and then set it to match the bond */
1319 new_slave->original_mtu = slave_dev->mtu;
1320 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1321 if (res) {
1322 pr_debug("Error %d calling dev_set_mtu\n", res);
1323 goto err_free;
1324 }
1325
Jay Vosburgh217df672005-09-26 16:11:50 -07001326 /*
1327 * Save slave's original ("permanent") mac address for modes
1328 * that need it, and for restoring it upon release, and then
1329 * set it to the master's address
1330 */
1331 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
Jay Vosburghdd957c52007-10-09 19:57:24 -07001333 if (!bond->params.fail_over_mac) {
Moni Shoua2ab82852007-10-09 19:43:39 -07001334 /*
1335 * Set slave to master's mac address. The application already
1336 * set the master's mac address to that of the first slave
1337 */
1338 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1339 addr.sa_family = slave_dev->type;
1340 res = dev_set_mac_address(slave_dev, &addr);
1341 if (res) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001342 pr_debug("Error %d calling set_mac_address\n", res);
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001343 goto err_restore_mtu;
Moni Shoua2ab82852007-10-09 19:43:39 -07001344 }
Jay Vosburgh217df672005-09-26 16:11:50 -07001345 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
Jay Vosburgh217df672005-09-26 16:11:50 -07001347 /* open the slave since the application closed it */
1348 res = dev_open(slave_dev);
1349 if (res) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001350 pr_debug("Opening slave %s failed\n", slave_dev->name);
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001351 goto err_restore_mac;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 }
1353
Jiri Pirko35d48902011-03-22 02:38:12 +00001354 new_slave->bond = bond;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 new_slave->dev = slave_dev;
Jay Vosburgh0b680e72006-09-22 21:54:10 -07001356 slave_dev->priv_flags |= IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Holger Eitzenberger58402052008-12-09 23:07:13 -08001358 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 /* bond_alb_init_slave() must be called before all other stages since
1360 * it might fail and we do not want to have to undo everything
1361 */
1362 res = bond_alb_init_slave(bond, new_slave);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001363 if (res)
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001364 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 }
1366
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001367 /* If the mode USES_PRIMARY, then the following is handled by
1368 * bond_change_active_slave().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 */
1370 if (!USES_PRIMARY(bond->params.mode)) {
1371 /* set promiscuity level to new slave */
1372 if (bond_dev->flags & IFF_PROMISC) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -07001373 res = dev_set_promiscuity(slave_dev, 1);
1374 if (res)
1375 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
1377
1378 /* set allmulti level to new slave */
1379 if (bond_dev->flags & IFF_ALLMULTI) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -07001380 res = dev_set_allmulti(slave_dev, 1);
1381 if (res)
1382 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 }
1384
David S. Millerb9e40852008-07-15 00:15:08 -07001385 netif_addr_lock_bh(bond_dev);
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001386
1387 dev_mc_sync_multiple(slave_dev, bond_dev);
1388 dev_uc_sync_multiple(slave_dev, bond_dev);
1389
David S. Millerb9e40852008-07-15 00:15:08 -07001390 netif_addr_unlock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 }
1392
1393 if (bond->params.mode == BOND_MODE_8023AD) {
1394 /* add lacpdu mc addr to mc list */
1395 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1396
Jiri Pirko22bedad32010-04-01 21:22:57 +00001397 dev_mc_add(slave_dev, lacpdu_multicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 }
1399
Wei Yongjunb8e2fde2013-08-23 10:45:07 +08001400 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1401 if (res) {
nikolay@redhat.com1ff412a2013-08-06 12:40:15 +02001402 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1403 bond_dev->name, slave_dev->name);
1404 goto err_close;
1405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
Veaceslav Falicoc8c23902013-09-25 09:20:25 +02001407 prev_slave = bond_last_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409 new_slave->delay = 0;
1410 new_slave->link_failure_count = 0;
1411
Veaceslav Falico876254a2013-03-12 06:31:32 +00001412 bond_update_speed_duplex(new_slave);
1413
Michal Kubečekf31c7932012-04-17 02:02:06 +00001414 new_slave->last_arp_rx = jiffies -
1415 (msecs_to_jiffies(bond->params.arp_interval) + 1);
Veaceslav Falico8599b522013-06-24 11:49:34 +02001416 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1417 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07001418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 if (bond->params.miimon && !bond->params.use_carrier) {
1420 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1421
1422 if ((link_reporting == -1) && !bond->params.arp_interval) {
1423 /*
1424 * miimon is set but a bonded network driver
1425 * does not support ETHTOOL/MII and
1426 * arp_interval is not set. Note: if
1427 * use_carrier is enabled, we will never go
1428 * here (because netif_carrier is always
1429 * supported); thus, we don't need to change
1430 * the messages for netif_carrier.
1431 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001432 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001433 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 } else if (link_reporting == -1) {
1435 /* unable get link status using mii/ethtool */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001436 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1437 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 }
1439 }
1440
1441 /* check for initial state */
Michal Kubečekf31c7932012-04-17 02:02:06 +00001442 if (bond->params.miimon) {
1443 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1444 if (bond->params.updelay) {
1445 new_slave->link = BOND_LINK_BACK;
1446 new_slave->delay = bond->params.updelay;
1447 } else {
1448 new_slave->link = BOND_LINK_UP;
1449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 } else {
Michal Kubečekf31c7932012-04-17 02:02:06 +00001451 new_slave->link = BOND_LINK_DOWN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 }
Michal Kubečekf31c7932012-04-17 02:02:06 +00001453 } else if (bond->params.arp_interval) {
1454 new_slave->link = (netif_carrier_ok(slave_dev) ?
1455 BOND_LINK_UP : BOND_LINK_DOWN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 } else {
Michal Kubečekf31c7932012-04-17 02:02:06 +00001457 new_slave->link = BOND_LINK_UP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 }
1459
Michal Kubečekf31c7932012-04-17 02:02:06 +00001460 if (new_slave->link != BOND_LINK_DOWN)
1461 new_slave->jiffies = jiffies;
1462 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1463 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1464 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1467 /* if there is a primary slave, remember it */
Jiri Pirkoa5499522009-09-25 03:28:09 +00001468 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 bond->primary_slave = new_slave;
Jiri Pirkoa5499522009-09-25 03:28:09 +00001470 bond->force_primary = true;
1471 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 }
1473
1474 switch (bond->params.mode) {
1475 case BOND_MODE_ACTIVEBACKUP:
Jay Vosburgh8a8e4472006-09-22 21:56:15 -07001476 bond_set_slave_inactive_flags(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 break;
1478 case BOND_MODE_8023AD:
1479 /* in 802.3ad mode, the internal mechanism
1480 * will activate the slaves in the selected
1481 * aggregator
1482 */
1483 bond_set_slave_inactive_flags(new_slave);
1484 /* if this is the first slave */
Veaceslav Falico23c147e2013-09-27 15:10:57 +02001485 if (!prev_slave) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 SLAVE_AD_INFO(new_slave).id = 1;
1487 /* Initialize AD with the number of times that the AD timer is called in 1 second
1488 * can be called only after the mac address of the bond is set
1489 */
Peter Pan(潘卫平)56d00c672011-06-08 21:19:02 +00001490 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 } else {
1492 SLAVE_AD_INFO(new_slave).id =
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001493 SLAVE_AD_INFO(prev_slave).id + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 }
1495
1496 bond_3ad_bind_slave(new_slave);
1497 break;
1498 case BOND_MODE_TLB:
1499 case BOND_MODE_ALB:
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001500 bond_set_active_slave(new_slave);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001501 bond_set_slave_inactive_flags(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 break;
1503 default:
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001504 pr_debug("This slave is always active in trunk mode\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
1506 /* always active in trunk mode */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001507 bond_set_active_slave(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
1509 /* In trunking mode there is little meaning to curr_active_slave
1510 * anyway (it holds no special properties of the bond device),
1511 * so we can change it without calling change_active_interface()
1512 */
Michal Kubeček4e591b92012-11-22 02:48:39 +00001513 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
nikolay@redhat.com278b2082013-08-01 16:54:51 +02001514 rcu_assign_pointer(bond->curr_active_slave, new_slave);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 break;
1517 } /* switch(bond_mode) */
1518
WANG Congf6dc31a2010-05-06 00:48:51 -07001519#ifdef CONFIG_NET_POLL_CONTROLLER
stephen hemminger10eccb42013-07-24 11:53:57 -07001520 slave_dev->npinfo = bond->dev->npinfo;
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001521 if (slave_dev->npinfo) {
1522 if (slave_enable_netpoll(new_slave)) {
1523 read_unlock(&bond->lock);
1524 pr_info("Error, %s: master_dev is using netpoll, "
1525 "but new slave device does not support netpoll.\n",
1526 bond_dev->name);
1527 res = -EBUSY;
stephen hemmingerf7d98212011-12-31 13:26:46 +00001528 goto err_detach;
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001529 }
WANG Congf6dc31a2010-05-06 00:48:51 -07001530 }
1531#endif
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001532
Jiri Pirko35d48902011-03-22 02:38:12 +00001533 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1534 new_slave);
1535 if (res) {
1536 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
Veaceslav Falico5831d662013-09-25 09:20:32 +02001537 goto err_detach;
Jiri Pirko35d48902011-03-22 02:38:12 +00001538 }
1539
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001540 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1541 if (res) {
1542 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1543 goto err_unregister;
1544 }
1545
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001546 res = bond_sysfs_slave_add(new_slave);
1547 if (res) {
1548 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1549 goto err_upper_unlink;
1550 }
1551
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001552 bond->slave_cnt++;
1553 bond_compute_features(bond);
1554 bond_set_carrier(bond);
1555
1556 if (USES_PRIMARY(bond->params.mode)) {
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001557 write_lock_bh(&bond->curr_slave_lock);
1558 bond_select_active_slave(bond);
1559 write_unlock_bh(&bond->curr_slave_lock);
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001560 }
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001561
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001562 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1563 bond_dev->name, slave_dev->name,
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001564 bond_is_active_slave(new_slave) ? "n active" : " backup",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001565 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
1567 /* enslave is successful */
1568 return 0;
1569
1570/* Undo stages on error */
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001571err_upper_unlink:
1572 bond_upper_dev_unlink(bond_dev, slave_dev);
1573
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001574err_unregister:
1575 netdev_rx_handler_unregister(slave_dev);
1576
stephen hemmingerf7d98212011-12-31 13:26:46 +00001577err_detach:
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001578 if (!USES_PRIMARY(bond->params.mode))
1579 bond_hw_addr_flush(bond_dev, slave_dev);
1580
nikolay@redhat.com1ff412a2013-08-06 12:40:15 +02001581 vlan_vids_del_by_dev(slave_dev, bond_dev);
nikolay@redhat.com3c5913b532013-04-18 07:33:36 +00001582 if (bond->primary_slave == new_slave)
1583 bond->primary_slave = NULL;
nikolay@redhat.com3c5913b532013-04-18 07:33:36 +00001584 if (bond->curr_active_slave == new_slave) {
nikolay@redhat.com3c5913b532013-04-18 07:33:36 +00001585 write_lock_bh(&bond->curr_slave_lock);
dingtianhongc8517032013-12-13 10:20:07 +08001586 bond_change_active_slave(bond, NULL);
nikolay@redhat.com3c5913b532013-04-18 07:33:36 +00001587 bond_select_active_slave(bond);
1588 write_unlock_bh(&bond->curr_slave_lock);
nikolay@redhat.com3c5913b532013-04-18 07:33:36 +00001589 }
nikolay@redhat.comfc7a72a2013-04-18 07:33:37 +00001590 slave_disable_netpoll(new_slave);
stephen hemmingerf7d98212011-12-31 13:26:46 +00001591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592err_close:
nikolay@redhat.comb6a5a7b2013-04-11 09:18:56 +00001593 slave_dev->priv_flags &= ~IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 dev_close(slave_dev);
1595
1596err_restore_mac:
Jay Vosburghdd957c52007-10-09 19:57:24 -07001597 if (!bond->params.fail_over_mac) {
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001598 /* XXX TODO - fom follow mode needs to change master's
1599 * MAC if this slave's MAC is in use by the bond, or at
1600 * least print a warning.
1601 */
Moni Shoua2ab82852007-10-09 19:43:39 -07001602 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1603 addr.sa_family = slave_dev->type;
1604 dev_set_mac_address(slave_dev, &addr);
1605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001607err_restore_mtu:
1608 dev_set_mtu(slave_dev, new_slave->original_mtu);
1609
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610err_free:
1611 kfree(new_slave);
1612
1613err_undo_flags:
Nikolay Aleksandrovb8fad452013-06-12 00:07:01 +02001614 /* Enslave of first slave has failed and we need to fix master's mac */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001615 if (!bond_has_slaves(bond) &&
dingtianhong844223a2014-01-02 09:13:16 +08001616 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
Nikolay Aleksandrovb8fad452013-06-12 00:07:01 +02001617 eth_hw_addr_random(bond_dev);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001618
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 return res;
1620}
1621
1622/*
1623 * Try to release the slave device <slave> from the bond device <master>
1624 * It is legal to access curr_active_slave without a lock because all the function
nikolay@redhat.com08963412013-02-18 14:09:42 +00001625 * is write-locked. If "all" is true it means that the function is being called
1626 * while destroying a bond interface and all slaves are being released.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 *
1628 * The rules for slave state should be:
1629 * for Active/Backup:
1630 * Active stays on all backups go down
1631 * for Bonded connections:
1632 * The first up interface should be left on and all others downed.
1633 */
nikolay@redhat.com08963412013-02-18 14:09:42 +00001634static int __bond_release_one(struct net_device *bond_dev,
1635 struct net_device *slave_dev,
1636 bool all)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637{
Wang Chen454d7c92008-11-12 23:37:49 -08001638 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 struct slave *slave, *oldcurrent;
1640 struct sockaddr addr;
Neil Horman5a0068d2013-09-27 12:22:15 -04001641 int old_flags = bond_dev->flags;
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001642 netdev_features_t old_features = bond_dev->features;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
1644 /* slave is not a slave or master is not master of this slave */
1645 if (!(slave_dev->flags & IFF_SLAVE) ||
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001646 !netdev_has_upper_dev(slave_dev, bond_dev)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001647 pr_err("%s: Error: cannot release %s.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 bond_dev->name, slave_dev->name);
1649 return -EINVAL;
1650 }
1651
Neil Hormane843fa52010-10-13 16:01:50 +00001652 block_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
1654 slave = bond_get_slave_by_dev(bond, slave_dev);
1655 if (!slave) {
1656 /* not a slave of this bond */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001657 pr_info("%s: %s not enslaved\n",
1658 bond_dev->name, slave_dev->name);
Neil Hormane843fa52010-10-13 16:01:50 +00001659 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 return -EINVAL;
1661 }
1662
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001663 /* release the slave from its bond */
1664 bond->slave_cnt--;
1665
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001666 bond_sysfs_slave_del(slave);
1667
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001668 bond_upper_dev_unlink(bond_dev, slave_dev);
Jiri Pirko35d48902011-03-22 02:38:12 +00001669 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1670 * for this slave anymore.
1671 */
1672 netdev_rx_handler_unregister(slave_dev);
Jiri Pirko35d48902011-03-22 02:38:12 +00001673 write_lock_bh(&bond->lock);
1674
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 /* Inform AD package of unbinding of slave. */
dingtianhongbe79bd02013-12-13 10:20:12 +08001676 if (bond->params.mode == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 bond_3ad_unbind_slave(slave);
dingtianhongbe79bd02013-12-13 10:20:12 +08001678
dingtianhongc8517032013-12-13 10:20:07 +08001679 write_unlock_bh(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001681 pr_info("%s: releasing %s interface %s\n",
1682 bond_dev->name,
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001683 bond_is_active_slave(slave) ? "active" : "backup",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001684 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
1686 oldcurrent = bond->curr_active_slave;
1687
1688 bond->current_arp_slave = NULL;
1689
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001690 if (!all && !bond->params.fail_over_mac) {
dingtianhong844223a2014-01-02 09:13:16 +08001691 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001692 bond_has_slaves(bond))
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001693 pr_warn("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1694 bond_dev->name, slave_dev->name,
1695 slave->perm_hwaddr,
1696 bond_dev->name, slave_dev->name);
1697 }
1698
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001699 if (bond->primary_slave == slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 bond->primary_slave = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
dingtianhongc8517032013-12-13 10:20:07 +08001702 if (oldcurrent == slave) {
1703 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 bond_change_active_slave(bond, NULL);
dingtianhongc8517032013-12-13 10:20:07 +08001705 write_unlock_bh(&bond->curr_slave_lock);
1706 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
Holger Eitzenberger58402052008-12-09 23:07:13 -08001708 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 /* Must be called only after the slave has been
1710 * detached from the list and the curr_active_slave
1711 * has been cleared (if our_slave == old_current),
1712 * but before a new active slave is selected.
1713 */
1714 bond_alb_deinit_slave(bond, slave);
1715 }
1716
nikolay@redhat.com08963412013-02-18 14:09:42 +00001717 if (all) {
nikolay@redhat.com278b2082013-08-01 16:54:51 +02001718 rcu_assign_pointer(bond->curr_active_slave, NULL);
nikolay@redhat.com08963412013-02-18 14:09:42 +00001719 } else if (oldcurrent == slave) {
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001720 /*
1721 * Note that we hold RTNL over this sequence, so there
1722 * is no concern that another slave add/remove event
1723 * will interfere.
1724 */
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001725 write_lock_bh(&bond->curr_slave_lock);
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 bond_select_active_slave(bond);
1728
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001729 write_unlock_bh(&bond->curr_slave_lock);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001730 }
1731
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001732 if (!bond_has_slaves(bond)) {
Jay Vosburghff59c452006-03-27 13:27:43 -08001733 bond_set_carrier(bond);
Jiri Pirko409cc1f2013-01-30 11:08:11 +01001734 eth_hw_addr_random(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
Veaceslav Falicoa59d3d22013-08-28 23:25:12 +02001736 if (vlan_uses_dev(bond_dev)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001737 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1738 bond_dev->name, bond_dev->name);
1739 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1740 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 }
1743
Neil Hormane843fa52010-10-13 16:01:50 +00001744 unblock_netpoll_tx();
nikolay@redhat.com278b2082013-08-01 16:54:51 +02001745 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001747 if (!bond_has_slaves(bond)) {
Shlomo Pongratz2af73d42012-04-03 22:56:19 +00001748 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
Veaceslav Falico80028ea2013-03-06 07:10:32 +00001749 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1750 }
Shlomo Pongratz2af73d42012-04-03 22:56:19 +00001751
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001752 bond_compute_features(bond);
1753 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1754 (old_features & NETIF_F_VLAN_CHALLENGED))
1755 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1756 bond_dev->name, slave_dev->name, bond_dev->name);
1757
Mitch Williamsb76cdba2005-11-09 10:36:41 -08001758 /* must do this from outside any spinlocks */
nikolay@redhat.com1ff412a2013-08-06 12:40:15 +02001759 vlan_vids_del_by_dev(slave_dev, bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001761 /* If the mode USES_PRIMARY, then this cases was handled above by
1762 * bond_change_active_slave(..., NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 */
1764 if (!USES_PRIMARY(bond->params.mode)) {
Neil Horman5a0068d2013-09-27 12:22:15 -04001765 /* unset promiscuity level from slave
1766 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1767 * of the IFF_PROMISC flag in the bond_dev, but we need the
1768 * value of that flag before that change, as that was the value
1769 * when this slave was attached, so we cache at the start of the
1770 * function and use it here. Same goes for ALLMULTI below
1771 */
1772 if (old_flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 dev_set_promiscuity(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
1775 /* unset allmulti level from slave */
Neil Horman5a0068d2013-09-27 12:22:15 -04001776 if (old_flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 dev_set_allmulti(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001779 bond_hw_addr_flush(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 }
1781
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001782 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -07001783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 /* close slave before restoring its mac address */
1785 dev_close(slave_dev);
1786
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001787 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
Moni Shoua2ab82852007-10-09 19:43:39 -07001788 /* restore original ("permanent") mac address */
1789 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1790 addr.sa_family = slave_dev->type;
1791 dev_set_mac_address(slave_dev, &addr);
1792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001794 dev_set_mtu(slave_dev, slave->original_mtu);
1795
Jiri Pirko2d7011c2011-03-16 08:46:43 +00001796 slave_dev->priv_flags &= ~IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
1798 kfree(slave);
1799
1800 return 0; /* deletion OK */
1801}
1802
nikolay@redhat.com08963412013-02-18 14:09:42 +00001803/* A wrapper used because of ndo_del_link */
1804int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1805{
1806 return __bond_release_one(bond_dev, slave_dev, false);
1807}
1808
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809/*
Nicolas de Pesloüandadaa102011-03-19 13:36:18 -07001810* First release a slave and then destroy the bond if no more slaves are left.
Moni Shouad90a1622007-10-09 19:43:43 -07001811* Must be under rtnl_lock when this function is called.
1812*/
stephen hemminger26d8ee72010-10-15 05:09:34 +00001813static int bond_release_and_destroy(struct net_device *bond_dev,
1814 struct net_device *slave_dev)
Moni Shouad90a1622007-10-09 19:43:43 -07001815{
Wang Chen454d7c92008-11-12 23:37:49 -08001816 struct bonding *bond = netdev_priv(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07001817 int ret;
1818
1819 ret = bond_release(bond_dev, slave_dev);
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001820 if (ret == 0 && !bond_has_slaves(bond)) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001821 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001822 pr_info("%s: destroying bond %s.\n",
1823 bond_dev->name, bond_dev->name);
Stephen Hemminger9e716262009-06-12 19:02:47 +00001824 unregister_netdevice(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07001825 }
1826 return ret;
1827}
1828
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1830{
Wang Chen454d7c92008-11-12 23:37:49 -08001831 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
1833 info->bond_mode = bond->params.mode;
1834 info->miimon = bond->params.miimon;
1835
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07001836 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 info->num_slaves = bond->slave_cnt;
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07001838 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
1840 return 0;
1841}
1842
1843static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1844{
Wang Chen454d7c92008-11-12 23:37:49 -08001845 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001846 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001847 int i = 0, res = -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07001850 read_lock(&bond->lock);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001851 bond_for_each_slave(bond, slave, iter) {
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001852 if (i++ == (int)info->slave_id) {
Eric Dumazet689c96c2009-04-23 03:39:04 +00001853 res = 0;
1854 strcpy(info->slave_name, slave->dev->name);
1855 info->link = slave->link;
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001856 info->state = bond_slave_state(slave);
Eric Dumazet689c96c2009-04-23 03:39:04 +00001857 info->link_failure_count = slave->link_failure_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 break;
1859 }
1860 }
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07001861 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
Eric Dumazet689c96c2009-04-23 03:39:04 +00001863 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864}
1865
1866/*-------------------------------- Monitoring -------------------------------*/
1867
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001868
1869static int bond_miimon_inspect(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870{
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001871 int link_state, commit = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001872 struct list_head *iter;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001873 struct slave *slave;
Jiri Pirko41f89102009-04-24 03:57:29 +00001874 bool ignore_updelay;
1875
1876 ignore_updelay = !bond->curr_active_slave ? true : false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
dingtianhong4cb4f972013-12-13 10:19:39 +08001878 bond_for_each_slave_rcu(bond, slave, iter) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001879 slave->new_link = BOND_LINK_NOCHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001881 link_state = bond_check_dev_link(bond, slave->dev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
1883 switch (slave->link) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001884 case BOND_LINK_UP:
1885 if (link_state)
1886 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001888 slave->link = BOND_LINK_FAIL;
1889 slave->delay = bond->params.downdelay;
1890 if (slave->delay) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001891 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1892 bond->dev->name,
1893 (bond->params.mode ==
1894 BOND_MODE_ACTIVEBACKUP) ?
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001895 (bond_is_active_slave(slave) ?
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001896 "active " : "backup ") : "",
1897 slave->dev->name,
1898 bond->params.downdelay * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001900 /*FALLTHRU*/
1901 case BOND_LINK_FAIL:
1902 if (link_state) {
1903 /*
1904 * recovered before downdelay expired
1905 */
1906 slave->link = BOND_LINK_UP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 slave->jiffies = jiffies;
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001908 pr_info("%s: link status up again after %d ms for interface %s.\n",
1909 bond->dev->name,
1910 (bond->params.downdelay - slave->delay) *
1911 bond->params.miimon,
1912 slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001913 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001915
1916 if (slave->delay <= 0) {
1917 slave->new_link = BOND_LINK_DOWN;
1918 commit++;
1919 continue;
1920 }
1921
1922 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001925 case BOND_LINK_DOWN:
1926 if (!link_state)
1927 continue;
1928
1929 slave->link = BOND_LINK_BACK;
1930 slave->delay = bond->params.updelay;
1931
1932 if (slave->delay) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001933 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1934 bond->dev->name, slave->dev->name,
1935 ignore_updelay ? 0 :
1936 bond->params.updelay *
1937 bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001939 /*FALLTHRU*/
1940 case BOND_LINK_BACK:
1941 if (!link_state) {
1942 slave->link = BOND_LINK_DOWN;
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001943 pr_info("%s: link status down again after %d ms for interface %s.\n",
1944 bond->dev->name,
1945 (bond->params.updelay - slave->delay) *
1946 bond->params.miimon,
1947 slave->dev->name);
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07001948
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001949 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001951
Jiri Pirko41f89102009-04-24 03:57:29 +00001952 if (ignore_updelay)
1953 slave->delay = 0;
1954
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001955 if (slave->delay <= 0) {
1956 slave->new_link = BOND_LINK_UP;
1957 commit++;
Jiri Pirko41f89102009-04-24 03:57:29 +00001958 ignore_updelay = false;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001959 continue;
1960 }
1961
1962 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 break;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001964 }
1965 }
1966
1967 return commit;
1968}
1969
1970static void bond_miimon_commit(struct bonding *bond)
1971{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001972 struct list_head *iter;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001973 struct slave *slave;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001974
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001975 bond_for_each_slave(bond, slave, iter) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001976 switch (slave->new_link) {
1977 case BOND_LINK_NOCHANGE:
1978 continue;
1979
1980 case BOND_LINK_UP:
1981 slave->link = BOND_LINK_UP;
1982 slave->jiffies = jiffies;
1983
1984 if (bond->params.mode == BOND_MODE_8023AD) {
1985 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001986 bond_set_backup_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001987 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1988 /* make it immediately active */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001989 bond_set_active_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001990 } else if (slave != bond->primary_slave) {
1991 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001992 bond_set_backup_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001993 }
1994
David Decotigny5d305302011-04-13 15:22:31 +00001995 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
Krzysztof Piotr Oledzki700c2a72010-10-06 14:25:06 -07001996 bond->dev->name, slave->dev->name,
Nikolay Aleksandrovdb4e9b22013-06-20 14:34:13 +02001997 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1998 slave->duplex ? "full" : "half");
Jay Vosburghf0c76d62008-07-02 18:21:58 -07001999
2000 /* notify ad that the link status has changed */
2001 if (bond->params.mode == BOND_MODE_8023AD)
2002 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2003
Holger Eitzenberger58402052008-12-09 23:07:13 -08002004 if (bond_is_lb(bond))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002005 bond_alb_handle_link_change(bond, slave,
2006 BOND_LINK_UP);
2007
2008 if (!bond->curr_active_slave ||
2009 (slave == bond->primary_slave))
2010 goto do_failover;
2011
2012 continue;
2013
2014 case BOND_LINK_DOWN:
Jay Vosburghfba4acd2008-10-30 17:41:14 -07002015 if (slave->link_failure_count < UINT_MAX)
2016 slave->link_failure_count++;
2017
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002018 slave->link = BOND_LINK_DOWN;
2019
2020 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2021 bond->params.mode == BOND_MODE_8023AD)
2022 bond_set_slave_inactive_flags(slave);
2023
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002024 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2025 bond->dev->name, slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002026
2027 if (bond->params.mode == BOND_MODE_8023AD)
2028 bond_3ad_handle_link_change(slave,
2029 BOND_LINK_DOWN);
2030
Jiri Pirkoae63e802009-05-27 05:42:36 +00002031 if (bond_is_lb(bond))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002032 bond_alb_handle_link_change(bond, slave,
2033 BOND_LINK_DOWN);
2034
2035 if (slave == bond->curr_active_slave)
2036 goto do_failover;
2037
2038 continue;
2039
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 default:
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002041 pr_err("%s: invalid new link %d on slave %s\n",
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002042 bond->dev->name, slave->new_link,
2043 slave->dev->name);
2044 slave->new_link = BOND_LINK_NOCHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002046 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 }
2048
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002049do_failover:
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002050 ASSERT_RTNL();
Neil Hormane843fa52010-10-13 16:01:50 +00002051 block_netpoll_tx();
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002052 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 bond_select_active_slave(bond);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002054 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002055 unblock_netpoll_tx();
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002056 }
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002057
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002058 bond_set_carrier(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059}
2060
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002061/*
2062 * bond_mii_monitor
2063 *
2064 * Really a wrapper that splits the mii monitor into two phases: an
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002065 * inspection, then (if inspection indicates something needs to be done)
2066 * an acquisition of appropriate locks followed by a commit phase to
2067 * implement whatever link state changes are indicated.
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002068 */
stephen hemminger6da67d22013-12-30 10:43:41 -08002069static void bond_mii_monitor(struct work_struct *work)
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002070{
2071 struct bonding *bond = container_of(work, struct bonding,
2072 mii_work.work);
Ben Hutchingsad246c92011-04-26 15:25:52 +00002073 bool should_notify_peers = false;
David S. Miller1f2cd842013-10-28 00:11:22 -04002074 unsigned long delay;
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002075
David S. Miller1f2cd842013-10-28 00:11:22 -04002076 delay = msecs_to_jiffies(bond->params.miimon);
2077
2078 if (!bond_has_slaves(bond))
dingtianhong6b6c5262013-10-24 11:09:03 +08002079 goto re_arm;
dingtianhong6b6c5262013-10-24 11:09:03 +08002080
dingtianhong4cb4f972013-12-13 10:19:39 +08002081 rcu_read_lock();
2082
Ben Hutchingsad246c92011-04-26 15:25:52 +00002083 should_notify_peers = bond_should_notify_peers(bond);
2084
David S. Miller1f2cd842013-10-28 00:11:22 -04002085 if (bond_miimon_inspect(bond)) {
dingtianhong4cb4f972013-12-13 10:19:39 +08002086 rcu_read_unlock();
David S. Miller1f2cd842013-10-28 00:11:22 -04002087
2088 /* Race avoidance with bond_close cancel of workqueue */
2089 if (!rtnl_trylock()) {
David S. Miller1f2cd842013-10-28 00:11:22 -04002090 delay = 1;
2091 should_notify_peers = false;
2092 goto re_arm;
2093 }
2094
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002095 bond_miimon_commit(bond);
2096
David S. Miller1f2cd842013-10-28 00:11:22 -04002097 rtnl_unlock(); /* might sleep, hold no other locks */
dingtianhong4cb4f972013-12-13 10:19:39 +08002098 } else
2099 rcu_read_unlock();
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002100
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002101re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00002102 if (bond->params.miimon)
David S. Miller1f2cd842013-10-28 00:11:22 -04002103 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2104
David S. Miller1f2cd842013-10-28 00:11:22 -04002105 if (should_notify_peers) {
2106 if (!rtnl_trylock())
2107 return;
2108 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2109 rtnl_unlock();
2110 }
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002111}
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002112
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002113static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002114{
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002115 struct net_device *upper;
2116 struct list_head *iter;
2117 bool ret = false;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002118
Andy Gospodarekeaddcd72012-03-22 16:14:29 +00002119 if (ip == bond_confirm_addr(bond->dev, 0, ip))
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002120 return true;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002121
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002122 rcu_read_lock();
Veaceslav Falico2f268f12013-09-25 09:20:07 +02002123 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002124 if (ip == bond_confirm_addr(upper, 0, ip)) {
2125 ret = true;
2126 break;
2127 }
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002128 }
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002129 rcu_read_unlock();
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002130
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002131 return ret;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002132}
2133
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002134/*
2135 * We go to the (large) trouble of VLAN tagging ARP frames because
2136 * switches in VLAN mode (especially if ports are configured as
2137 * "native" to a VLAN) might not pass non-tagged frames.
2138 */
Al Virod3bb52b2007-08-22 20:06:58 -04002139static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002140{
2141 struct sk_buff *skb;
2142
nikolay@redhat.comacca2672013-05-18 01:18:29 +00002143 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2144 slave_dev->name, &dest_ip, &src_ip, vlan_id);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002145
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002146 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2147 NULL, slave_dev->dev_addr, NULL);
2148
2149 if (!skb) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002150 pr_err("ARP packet allocation failed\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002151 return;
2152 }
2153 if (vlan_id) {
Patrick McHardy1fd9b1f2013-04-19 02:04:29 +00002154 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002155 if (!skb) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002156 pr_err("failed to insert VLAN tag\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002157 return;
2158 }
2159 }
2160 arp_xmit(skb);
2161}
2162
2163
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2165{
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002166 struct net_device *upper, *vlan_upper;
2167 struct list_head *iter, *vlan_iter;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002168 struct rtable *rt;
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002169 __be32 *targets = bond->params.arp_targets, addr;
2170 int i, vlan_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002172 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
nikolay@redhat.comacca2672013-05-18 01:18:29 +00002173 pr_debug("basa: target %pI4\n", &targets[i]);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002174
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002175 /* Find out through which dev should the packet go */
David S. Miller78fbfd82011-03-12 00:00:52 -05002176 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2177 RTO_ONLINK, 0);
David S. Millerb23dd4f2011-03-02 14:31:35 -08002178 if (IS_ERR(rt)) {
Veaceslav Falico3e325822013-08-28 23:25:16 +02002179 pr_debug("%s: no route to arp_ip_target %pI4\n",
2180 bond->dev->name, &targets[i]);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002181 continue;
2182 }
2183
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002184 vlan_id = 0;
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002185
2186 /* bond device itself */
2187 if (rt->dst.dev == bond->dev)
2188 goto found;
2189
2190 rcu_read_lock();
2191 /* first we search only for vlan devices. for every vlan
2192 * found we verify its upper dev list, searching for the
2193 * rt->dst.dev. If found we save the tag of the vlan and
2194 * proceed to send the packet.
2195 *
2196 * TODO: QinQ?
2197 */
Veaceslav Falico2f268f12013-09-25 09:20:07 +02002198 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2199 vlan_iter) {
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002200 if (!is_vlan_dev(vlan_upper))
2201 continue;
Veaceslav Falico2f268f12013-09-25 09:20:07 +02002202 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2203 iter) {
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002204 if (upper == rt->dst.dev) {
2205 vlan_id = vlan_dev_vlan_id(vlan_upper);
2206 rcu_read_unlock();
2207 goto found;
2208 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002209 }
2210 }
2211
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002212 /* if the device we're looking for is not on top of any of
2213 * our upper vlans, then just search for any dev that
2214 * matches, and in case it's a vlan - save the id
2215 */
Veaceslav Falico2f268f12013-09-25 09:20:07 +02002216 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002217 if (upper == rt->dst.dev) {
2218 /* if it's a vlan - get its VID */
2219 if (is_vlan_dev(upper))
2220 vlan_id = vlan_dev_vlan_id(upper);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002221
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002222 rcu_read_unlock();
2223 goto found;
2224 }
2225 }
2226 rcu_read_unlock();
2227
2228 /* Not our device - skip */
Veaceslav Falico3e325822013-08-28 23:25:16 +02002229 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2230 bond->dev->name, &targets[i],
2231 rt->dst.dev ? rt->dst.dev->name : "NULL");
2232
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002233 ip_rt_put(rt);
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002234 continue;
2235
2236found:
2237 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2238 ip_rt_put(rt);
2239 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2240 addr, vlan_id);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002241 }
2242}
2243
Al Virod3bb52b2007-08-22 20:06:58 -04002244static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002245{
Veaceslav Falico8599b522013-06-24 11:49:34 +02002246 int i;
2247
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002248 if (!sip || !bond_has_this_ip(bond, tip)) {
2249 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2250 return;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002251 }
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002252
Veaceslav Falico8599b522013-06-24 11:49:34 +02002253 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2254 if (i == -1) {
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002255 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2256 return;
2257 }
2258 slave->last_arp_rx = jiffies;
Veaceslav Falico8599b522013-06-24 11:49:34 +02002259 slave->target_last_arp_rx[i] = jiffies;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002260}
2261
nikolay@redhat.com5bb9e0b2013-09-07 00:00:26 +02002262int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2263 struct slave *slave)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002264{
Eric Dumazetde063b72012-06-11 19:23:07 +00002265 struct arphdr *arp = (struct arphdr *)skb->data;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002266 unsigned char *arp_ptr;
Al Virod3bb52b2007-08-22 20:06:58 -04002267 __be32 sip, tip;
Eric Dumazetde063b72012-06-11 19:23:07 +00002268 int alen;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002269
Jiri Pirko3aba8912011-04-19 03:48:16 +00002270 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00002271 return RX_HANDLER_ANOTHER;
Andy Gospodarek1f3c8802009-12-14 10:48:58 +00002272
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002273 read_lock(&bond->lock);
Veaceslav Falico2c146102013-06-24 11:49:31 +02002274
2275 if (!slave_do_arp_validate(bond, slave))
2276 goto out_unlock;
2277
Eric Dumazetde063b72012-06-11 19:23:07 +00002278 alen = arp_hdr_len(bond->dev);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002279
Jiri Pirko3aba8912011-04-19 03:48:16 +00002280 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2281 bond->dev->name, skb->dev->name);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002282
Eric Dumazetde063b72012-06-11 19:23:07 +00002283 if (alen > skb_headlen(skb)) {
2284 arp = kmalloc(alen, GFP_ATOMIC);
2285 if (!arp)
2286 goto out_unlock;
2287 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2288 goto out_unlock;
2289 }
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002290
Jiri Pirko3aba8912011-04-19 03:48:16 +00002291 if (arp->ar_hln != bond->dev->addr_len ||
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002292 skb->pkt_type == PACKET_OTHERHOST ||
2293 skb->pkt_type == PACKET_LOOPBACK ||
2294 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2295 arp->ar_pro != htons(ETH_P_IP) ||
2296 arp->ar_pln != 4)
2297 goto out_unlock;
2298
2299 arp_ptr = (unsigned char *)(arp + 1);
Jiri Pirko3aba8912011-04-19 03:48:16 +00002300 arp_ptr += bond->dev->addr_len;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002301 memcpy(&sip, arp_ptr, 4);
Jiri Pirko3aba8912011-04-19 03:48:16 +00002302 arp_ptr += 4 + bond->dev->addr_len;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002303 memcpy(&tip, arp_ptr, 4);
2304
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002305 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002306 bond->dev->name, slave->dev->name, bond_slave_state(slave),
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002307 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2308 &sip, &tip);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002309
2310 /*
2311 * Backup slaves won't see the ARP reply, but do come through
2312 * here for each ARP probe (so we swap the sip/tip to validate
2313 * the probe). In a "redundant switch, common router" type of
2314 * configuration, the ARP probe will (hopefully) travel from
2315 * the active, through one switch, the router, then the other
2316 * switch before reaching the backup.
Veaceslav Falicoaeea64a2013-06-24 11:49:32 +02002317 *
2318 * We 'trust' the arp requests if there is an active slave and
2319 * it received valid arp reply(s) after it became active. This
2320 * is done to avoid endless looping when we can't reach the
2321 * arp_ip_target and fool ourselves with our own arp requests.
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002322 */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002323 if (bond_is_active_slave(slave))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002324 bond_validate_arp(bond, slave, sip, tip);
Veaceslav Falicoaeea64a2013-06-24 11:49:32 +02002325 else if (bond->curr_active_slave &&
2326 time_after(slave_last_rx(bond, bond->curr_active_slave),
2327 bond->curr_active_slave->jiffies))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002328 bond_validate_arp(bond, slave, tip, sip);
2329
2330out_unlock:
2331 read_unlock(&bond->lock);
Eric Dumazetde063b72012-06-11 19:23:07 +00002332 if (arp != (struct arphdr *)skb->data)
2333 kfree(arp);
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00002334 return RX_HANDLER_ANOTHER;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002335}
2336
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002337/* function to verify if we're in the arp_interval timeslice, returns true if
2338 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2339 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2340 */
2341static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2342 int mod)
2343{
2344 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2345
2346 return time_in_range(jiffies,
2347 last_act - delta_in_ticks,
2348 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2349}
2350
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351/*
2352 * this function is called regularly to monitor each slave's link
2353 * ensuring that traffic is being sent and received when arp monitoring
2354 * is used in load-balancing mode. if the adapter has been dormant, then an
2355 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2356 * arp monitoring in active backup mode.
2357 */
stephen hemminger6da67d22013-12-30 10:43:41 -08002358static void bond_loadbalance_arp_mon(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359{
Jay Vosburgh1b76b312007-10-17 17:37:45 -07002360 struct bonding *bond = container_of(work, struct bonding,
2361 arp_work.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 struct slave *slave, *oldcurrent;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002363 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 int do_failover = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
David S. Miller1f2cd842013-10-28 00:11:22 -04002366 if (!bond_has_slaves(bond))
dingtianhong7f1bb572013-10-24 11:09:17 +08002367 goto re_arm;
dingtianhong7f1bb572013-10-24 11:09:17 +08002368
dingtianhong2e52f4f2013-12-13 10:19:50 +08002369 rcu_read_lock();
2370
2371 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 /* see if any of the previous devices are up now (i.e. they have
2373 * xmt and rcv traffic). the curr_active_slave does not come into
2374 * the picture unless it is null. also, slave->jiffies is not needed
2375 * here because we send an arp on each slave and give a slave as
2376 * long as it needs to get the tx/rx within the delta.
2377 * TODO: what about up/down delay in arp mode? it wasn't here before
2378 * so it can wait
2379 */
dingtianhong2e52f4f2013-12-13 10:19:50 +08002380 bond_for_each_slave_rcu(bond, slave, iter) {
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002381 unsigned long trans_start = dev_trans_start(slave->dev);
2382
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 if (slave->link != BOND_LINK_UP) {
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002384 if (bond_time_in_interval(bond, trans_start, 1) &&
2385 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386
2387 slave->link = BOND_LINK_UP;
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002388 bond_set_active_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
2390 /* primary_slave has no meaning in round-robin
2391 * mode. the window of a slave being up and
2392 * curr_active_slave being null after enslaving
2393 * is closed.
2394 */
2395 if (!oldcurrent) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002396 pr_info("%s: link status definitely up for interface %s, ",
2397 bond->dev->name,
2398 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 do_failover = 1;
2400 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002401 pr_info("%s: interface %s is now up\n",
2402 bond->dev->name,
2403 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 }
2405 }
2406 } else {
2407 /* slave->link == BOND_LINK_UP */
2408
2409 /* not all switches will respond to an arp request
2410 * when the source ip is 0, so don't take the link down
2411 * if we don't know our ip yet
2412 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002413 if (!bond_time_in_interval(bond, trans_start, 2) ||
2414 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
2416 slave->link = BOND_LINK_DOWN;
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002417 bond_set_backup_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002419 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002422 pr_info("%s: interface %s is now down.\n",
2423 bond->dev->name,
2424 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002426 if (slave == oldcurrent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 do_failover = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 }
2429 }
2430
2431 /* note: if switch is in round-robin mode, all links
2432 * must tx arp to ensure all links rx an arp - otherwise
2433 * links may oscillate or not come up at all; if switch is
2434 * in something like xor mode, there is nothing we can
2435 * do - all replies will be rx'ed on same link causing slaves
2436 * to be unstable during low/no traffic periods
2437 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002438 if (IS_UP(slave->dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 bond_arp_send_all(bond, slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 }
2441
dingtianhong2e52f4f2013-12-13 10:19:50 +08002442 rcu_read_unlock();
2443
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 if (do_failover) {
dingtianhong2e52f4f2013-12-13 10:19:50 +08002445 /* the bond_select_active_slave must hold RTNL
2446 * and curr_slave_lock for write.
2447 */
2448 if (!rtnl_trylock())
2449 goto re_arm;
Neil Hormane843fa52010-10-13 16:01:50 +00002450 block_netpoll_tx();
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002451 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
2453 bond_select_active_slave(bond);
2454
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002455 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002456 unblock_netpoll_tx();
dingtianhong2e52f4f2013-12-13 10:19:50 +08002457 rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 }
2459
2460re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00002461 if (bond->params.arp_interval)
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002462 queue_delayed_work(bond->wq, &bond->arp_work,
2463 msecs_to_jiffies(bond->params.arp_interval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464}
2465
2466/*
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002467 * Called to inspect slaves for active-backup mode ARP monitor link state
2468 * changes. Sets new_link in slaves to specify what action should take
2469 * place for the slave. Returns 0 if no changes are found, >0 if changes
2470 * to link states must be committed.
2471 *
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002472 * Called with rcu_read_lock hold.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002474static int bond_ab_arp_inspect(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475{
Veaceslav Falicodef44602013-08-03 03:50:35 +02002476 unsigned long trans_start, last_rx;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002477 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002478 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002479 int commit = 0;
Jiri Bohacda210f52012-08-30 12:02:47 +00002480
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002481 bond_for_each_slave_rcu(bond, slave, iter) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002482 slave->new_link = BOND_LINK_NOCHANGE;
Veaceslav Falicodef44602013-08-03 03:50:35 +02002483 last_rx = slave_last_rx(bond, slave);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002484
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 if (slave->link != BOND_LINK_UP) {
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002486 if (bond_time_in_interval(bond, last_rx, 1)) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002487 slave->new_link = BOND_LINK_UP;
2488 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 }
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002490 continue;
2491 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002493 /*
2494 * Give slaves 2*delta after being enslaved or made
2495 * active. This avoids bouncing, as the last receive
2496 * times need a full ARP monitor cycle to be updated.
2497 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002498 if (bond_time_in_interval(bond, slave->jiffies, 2))
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002499 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002501 /*
2502 * Backup slave is down if:
2503 * - No current_arp_slave AND
2504 * - more than 3*delta since last receive AND
2505 * - the bond has an IP address
2506 *
2507 * Note: a non-null current_arp_slave indicates
2508 * the curr_active_slave went down and we are
2509 * searching for a new one; under this condition
2510 * we only take the curr_active_slave down - this
2511 * gives each slave a chance to tx/rx traffic
2512 * before being taken out
2513 */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002514 if (!bond_is_active_slave(slave) &&
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002515 !bond->current_arp_slave &&
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002516 !bond_time_in_interval(bond, last_rx, 3)) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002517 slave->new_link = BOND_LINK_DOWN;
2518 commit++;
2519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002521 /*
2522 * Active slave is down if:
2523 * - more than 2*delta since transmitting OR
2524 * - (more than 2*delta since receive AND
2525 * the bond has an IP address)
2526 */
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002527 trans_start = dev_trans_start(slave->dev);
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002528 if (bond_is_active_slave(slave) &&
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002529 (!bond_time_in_interval(bond, trans_start, 2) ||
2530 !bond_time_in_interval(bond, last_rx, 2))) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002531 slave->new_link = BOND_LINK_DOWN;
2532 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 }
2534 }
2535
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002536 return commit;
2537}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002539/*
2540 * Called to commit link state changes noted by inspection step of
2541 * active-backup mode ARP monitor.
2542 *
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002543 * Called with RTNL hold.
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002544 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002545static void bond_ab_arp_commit(struct bonding *bond)
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002546{
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002547 unsigned long trans_start;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002548 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002549 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002551 bond_for_each_slave(bond, slave, iter) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002552 switch (slave->new_link) {
2553 case BOND_LINK_NOCHANGE:
2554 continue;
2555
2556 case BOND_LINK_UP:
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002557 trans_start = dev_trans_start(slave->dev);
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002558 if (bond->curr_active_slave != slave ||
2559 (!bond->curr_active_slave &&
2560 bond_time_in_interval(bond, trans_start, 1))) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002561 slave->link = BOND_LINK_UP;
Veaceslav Falico5a430972012-04-05 03:47:43 +00002562 if (bond->current_arp_slave) {
2563 bond_set_slave_inactive_flags(
2564 bond->current_arp_slave);
2565 bond->current_arp_slave = NULL;
2566 }
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002567
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002568 pr_info("%s: link status definitely up for interface %s.\n",
Jiri Pirkob9f60252009-08-31 11:09:38 +00002569 bond->dev->name, slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002570
Jiri Pirkob9f60252009-08-31 11:09:38 +00002571 if (!bond->curr_active_slave ||
2572 (slave == bond->primary_slave))
2573 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002574
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002575 }
2576
Jiri Pirkob9f60252009-08-31 11:09:38 +00002577 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002578
2579 case BOND_LINK_DOWN:
2580 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002583 slave->link = BOND_LINK_DOWN;
Jiri Pirkob9f60252009-08-31 11:09:38 +00002584 bond_set_slave_inactive_flags(slave);
2585
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002586 pr_info("%s: link status definitely down for interface %s, disabling it\n",
Jiri Pirkob9f60252009-08-31 11:09:38 +00002587 bond->dev->name, slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002588
2589 if (slave == bond->curr_active_slave) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002590 bond->current_arp_slave = NULL;
Jiri Pirkob9f60252009-08-31 11:09:38 +00002591 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002592 }
Jiri Pirkob9f60252009-08-31 11:09:38 +00002593
2594 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002595
2596 default:
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002597 pr_err("%s: impossible: new_link %d on slave %s\n",
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002598 bond->dev->name, slave->new_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 slave->dev->name);
Jiri Pirkob9f60252009-08-31 11:09:38 +00002600 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602
Jiri Pirkob9f60252009-08-31 11:09:38 +00002603do_failover:
2604 ASSERT_RTNL();
Neil Hormane843fa52010-10-13 16:01:50 +00002605 block_netpoll_tx();
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002606 write_lock_bh(&bond->curr_slave_lock);
Jiri Pirkob9f60252009-08-31 11:09:38 +00002607 bond_select_active_slave(bond);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002608 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002609 unblock_netpoll_tx();
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002610 }
2611
2612 bond_set_carrier(bond);
2613}
2614
2615/*
2616 * Send ARP probes for active-backup mode ARP monitor.
2617 *
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002618 * Called with rcu_read_lock hold.
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002619 */
2620static void bond_ab_arp_probe(struct bonding *bond)
2621{
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002622 struct slave *slave, *before = NULL, *new_slave = NULL,
2623 *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
Veaceslav Falico4087df872013-09-25 09:20:19 +02002624 struct list_head *iter;
2625 bool found = false;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002626
2627 read_lock(&bond->curr_slave_lock);
2628
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002629 if (curr_arp_slave && bond->curr_active_slave)
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002630 pr_info("PROBE: c_arp %s && cas %s BAD\n",
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002631 curr_arp_slave->dev->name,
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002632 bond->curr_active_slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002633
2634 if (bond->curr_active_slave) {
2635 bond_arp_send_all(bond, bond->curr_active_slave);
2636 read_unlock(&bond->curr_slave_lock);
2637 return;
2638 }
2639
2640 read_unlock(&bond->curr_slave_lock);
2641
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 /* if we don't have a curr_active_slave, search for the next available
2643 * backup slave from the current_arp_slave and make it the candidate
2644 * for becoming the curr_active_slave
2645 */
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002646
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002647 if (!curr_arp_slave) {
2648 curr_arp_slave = bond_first_slave_rcu(bond);
2649 if (!curr_arp_slave)
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002650 return;
2651 }
2652
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002653 bond_set_slave_inactive_flags(curr_arp_slave);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002654
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002655 bond_for_each_slave_rcu(bond, slave, iter) {
Veaceslav Falico4087df872013-09-25 09:20:19 +02002656 if (!found && !before && IS_UP(slave->dev))
2657 before = slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658
Veaceslav Falico4087df872013-09-25 09:20:19 +02002659 if (found && !new_slave && IS_UP(slave->dev))
2660 new_slave = slave;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002661 /* if the link state is up at this point, we
2662 * mark it down - this can happen if we have
2663 * simultaneous link failures and
2664 * reselect_active_interface doesn't make this
2665 * one the current slave so it is still marked
2666 * up when it is actually down
2667 */
Veaceslav Falico4087df872013-09-25 09:20:19 +02002668 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002669 slave->link = BOND_LINK_DOWN;
2670 if (slave->link_failure_count < UINT_MAX)
2671 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002673 bond_set_slave_inactive_flags(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002675 pr_info("%s: backup interface %s is now down.\n",
2676 bond->dev->name, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 }
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002678 if (slave == curr_arp_slave)
Veaceslav Falico4087df872013-09-25 09:20:19 +02002679 found = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 }
Veaceslav Falico4087df872013-09-25 09:20:19 +02002681
2682 if (!new_slave && before)
2683 new_slave = before;
2684
2685 if (!new_slave)
2686 return;
2687
2688 new_slave->link = BOND_LINK_BACK;
2689 bond_set_slave_active_flags(new_slave);
2690 bond_arp_send_all(bond, new_slave);
2691 new_slave->jiffies = jiffies;
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002692 rcu_assign_pointer(bond->current_arp_slave, new_slave);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002693}
2694
stephen hemminger6da67d22013-12-30 10:43:41 -08002695static void bond_activebackup_arp_mon(struct work_struct *work)
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002696{
2697 struct bonding *bond = container_of(work, struct bonding,
2698 arp_work.work);
Ben Hutchingsad246c92011-04-26 15:25:52 +00002699 bool should_notify_peers = false;
David S. Miller1f2cd842013-10-28 00:11:22 -04002700 int delta_in_ticks;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002701
David S. Miller1f2cd842013-10-28 00:11:22 -04002702 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2703
2704 if (!bond_has_slaves(bond))
dingtianhong80b9d232013-10-24 11:09:25 +08002705 goto re_arm;
dingtianhong80b9d232013-10-24 11:09:25 +08002706
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002707 rcu_read_lock();
2708
Ben Hutchingsad246c92011-04-26 15:25:52 +00002709 should_notify_peers = bond_should_notify_peers(bond);
2710
David S. Miller1f2cd842013-10-28 00:11:22 -04002711 if (bond_ab_arp_inspect(bond)) {
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002712 rcu_read_unlock();
David S. Miller1f2cd842013-10-28 00:11:22 -04002713
2714 /* Race avoidance with bond_close flush of workqueue */
2715 if (!rtnl_trylock()) {
David S. Miller1f2cd842013-10-28 00:11:22 -04002716 delta_in_ticks = 1;
2717 should_notify_peers = false;
2718 goto re_arm;
2719 }
2720
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002721 bond_ab_arp_commit(bond);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002722
David S. Miller1f2cd842013-10-28 00:11:22 -04002723 rtnl_unlock();
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002724 rcu_read_lock();
David S. Miller1f2cd842013-10-28 00:11:22 -04002725 }
2726
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002727 bond_ab_arp_probe(bond);
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002728 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729
2730re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00002731 if (bond->params.arp_interval)
David S. Miller1f2cd842013-10-28 00:11:22 -04002732 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2733
David S. Miller1f2cd842013-10-28 00:11:22 -04002734 if (should_notify_peers) {
2735 if (!rtnl_trylock())
2736 return;
2737 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2738 rtnl_unlock();
2739 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740}
2741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742/*-------------------------- netdev event handling --------------------------*/
2743
2744/*
2745 * Change device name
2746 */
2747static int bond_event_changename(struct bonding *bond)
2748{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 bond_remove_proc_entry(bond);
2750 bond_create_proc_entry(bond);
Stephen Hemminger7e083842009-06-12 19:02:46 +00002751
Taku Izumif073c7c2010-12-09 15:17:13 +00002752 bond_debug_reregister(bond);
2753
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 return NOTIFY_DONE;
2755}
2756
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002757static int bond_master_netdev_event(unsigned long event,
2758 struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759{
Wang Chen454d7c92008-11-12 23:37:49 -08002760 struct bonding *event_bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761
2762 switch (event) {
2763 case NETDEV_CHANGENAME:
2764 return bond_event_changename(event_bond);
Eric W. Biedermana64d49c2012-07-09 10:51:45 +00002765 case NETDEV_UNREGISTER:
2766 bond_remove_proc_entry(event_bond);
2767 break;
2768 case NETDEV_REGISTER:
2769 bond_create_proc_entry(event_bond);
2770 break;
nikolay@redhat.com6c8d23f2013-09-02 13:51:38 +02002771 case NETDEV_NOTIFY_PEERS:
2772 if (event_bond->send_peer_notif)
2773 event_bond->send_peer_notif--;
2774 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 default:
2776 break;
2777 }
2778
2779 return NOTIFY_DONE;
2780}
2781
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002782static int bond_slave_netdev_event(unsigned long event,
2783 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784{
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002785 struct slave *slave = bond_slave_get_rtnl(slave_dev);
nikolay@redhat.com61013912013-04-11 09:18:55 +00002786 struct bonding *bond;
2787 struct net_device *bond_dev;
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002788 u32 old_speed;
2789 u8 old_duplex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790
nikolay@redhat.com61013912013-04-11 09:18:55 +00002791 /* A netdev event can be generated while enslaving a device
2792 * before netdev_rx_handler_register is called in which case
2793 * slave will be NULL
2794 */
2795 if (!slave)
2796 return NOTIFY_DONE;
2797 bond_dev = slave->bond->dev;
2798 bond = slave->bond;
2799
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 switch (event) {
2801 case NETDEV_UNREGISTER:
nikolay@redhat.com8d2ada72013-06-26 17:13:37 +02002802 if (bond_dev->type != ARPHRD_ETHER)
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002803 bond_release_and_destroy(bond_dev, slave_dev);
2804 else
2805 bond_release(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 break;
Weiping Pan98f41f62011-10-31 17:20:48 +00002807 case NETDEV_UP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 case NETDEV_CHANGE:
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002809 old_speed = slave->speed;
2810 old_duplex = slave->duplex;
Jay Vosburgh17d04502009-03-18 18:38:25 -07002811
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002812 bond_update_speed_duplex(slave);
Jay Vosburgh17d04502009-03-18 18:38:25 -07002813
Jiri Pirko471cb5a2013-01-03 22:49:01 +00002814 if (bond->params.mode == BOND_MODE_8023AD) {
2815 if (old_speed != slave->speed)
2816 bond_3ad_adapter_speed_changed(slave);
2817 if (old_duplex != slave->duplex)
2818 bond_3ad_adapter_duplex_changed(slave);
Jay Vosburgh17d04502009-03-18 18:38:25 -07002819 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 break;
2821 case NETDEV_DOWN:
2822 /*
2823 * ... Or is it this?
2824 */
2825 break;
2826 case NETDEV_CHANGEMTU:
2827 /*
2828 * TODO: Should slaves be allowed to
2829 * independently alter their MTU? For
2830 * an active-backup bond, slaves need
2831 * not be the same type of device, so
2832 * MTUs may vary. For other modes,
2833 * slaves arguably should have the
2834 * same MTUs. To do this, we'd need to
2835 * take over the slave's change_mtu
2836 * function for the duration of their
2837 * servitude.
2838 */
2839 break;
2840 case NETDEV_CHANGENAME:
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01002841 /* we don't care if we don't have primary set */
2842 if (!USES_PRIMARY(bond->params.mode) ||
2843 !bond->params.primary[0])
2844 break;
2845
2846 if (slave == bond->primary_slave) {
2847 /* slave's name changed - he's no longer primary */
2848 bond->primary_slave = NULL;
2849 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2850 /* we have a new primary slave */
2851 bond->primary_slave = slave;
2852 } else { /* we didn't change primary - exit */
2853 break;
2854 }
2855
2856 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2857 bond->dev->name, bond->primary_slave ? slave_dev->name :
2858 "none");
2859 write_lock_bh(&bond->curr_slave_lock);
2860 bond_select_active_slave(bond);
2861 write_unlock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 break;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04002863 case NETDEV_FEAT_CHANGE:
2864 bond_compute_features(bond);
2865 break;
Jiri Pirko4aa5dee2013-07-20 12:13:53 +02002866 case NETDEV_RESEND_IGMP:
2867 /* Propagate to master device */
2868 call_netdevice_notifiers(event, slave->bond->dev);
2869 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 default:
2871 break;
2872 }
2873
2874 return NOTIFY_DONE;
2875}
2876
2877/*
2878 * bond_netdev_event: handle netdev notifier chain events.
2879 *
2880 * This function receives events for the netdev chain. The caller (an
Alan Sterne041c682006-03-27 01:16:30 -08002881 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 * locks for us to safely manipulate the slave devices (RTNL lock,
2883 * dev_probe_lock).
2884 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002885static int bond_netdev_event(struct notifier_block *this,
2886 unsigned long event, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887{
Jiri Pirko351638e2013-05-28 01:30:21 +00002888 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002890 pr_debug("event_dev: %s, event: %lx\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002891 event_dev ? event_dev->name : "None",
2892 event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893
Jay Vosburgh0b680e72006-09-22 21:54:10 -07002894 if (!(event_dev->priv_flags & IFF_BONDING))
2895 return NOTIFY_DONE;
2896
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 if (event_dev->flags & IFF_MASTER) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002898 pr_debug("IFF_MASTER\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 return bond_master_netdev_event(event, event_dev);
2900 }
2901
2902 if (event_dev->flags & IFF_SLAVE) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002903 pr_debug("IFF_SLAVE\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 return bond_slave_netdev_event(event, event_dev);
2905 }
2906
2907 return NOTIFY_DONE;
2908}
2909
2910static struct notifier_block bond_netdev_notifier = {
2911 .notifier_call = bond_netdev_event,
2912};
2913
Jay Vosburgh169a3e62005-06-26 17:54:11 -04002914/*---------------------------- Hashing Policies -----------------------------*/
2915
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002916/* L2 hash helper */
2917static inline u32 bond_eth_hash(struct sk_buff *skb)
Jay Vosburgh169a3e62005-06-26 17:54:11 -04002918{
2919 struct ethhdr *data = (struct ethhdr *)skb->data;
2920
John Eaglesham6b923cb2012-08-21 20:43:35 +00002921 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002922 return data->h_dest[5] ^ data->h_source[5];
John Eaglesham6b923cb2012-08-21 20:43:35 +00002923
2924 return 0;
2925}
2926
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002927/* Extract the appropriate headers based on bond's xmit policy */
2928static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2929 struct flow_keys *fk)
John Eaglesham6b923cb2012-08-21 20:43:35 +00002930{
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002931 const struct ipv6hdr *iph6;
Eric Dumazet43945422013-04-15 17:03:24 +00002932 const struct iphdr *iph;
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002933 int noff, proto = -1;
John Eaglesham6b923cb2012-08-21 20:43:35 +00002934
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002935 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2936 return skb_flow_dissect(skb, fk);
2937
2938 fk->ports = 0;
2939 noff = skb_network_offset(skb);
2940 if (skb->protocol == htons(ETH_P_IP)) {
2941 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2942 return false;
John Eaglesham6b923cb2012-08-21 20:43:35 +00002943 iph = ip_hdr(skb);
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002944 fk->src = iph->saddr;
2945 fk->dst = iph->daddr;
2946 noff += iph->ihl << 2;
2947 if (!ip_is_fragment(iph))
2948 proto = iph->protocol;
2949 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2950 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2951 return false;
2952 iph6 = ipv6_hdr(skb);
2953 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2954 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2955 noff += sizeof(*iph6);
2956 proto = iph6->nexthdr;
2957 } else {
2958 return false;
John Eaglesham6b923cb2012-08-21 20:43:35 +00002959 }
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002960 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2961 fk->ports = skb_flow_get_ports(skb, noff, proto);
John Eaglesham6b923cb2012-08-21 20:43:35 +00002962
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002963 return true;
John Eaglesham6b923cb2012-08-21 20:43:35 +00002964}
2965
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002966/**
2967 * bond_xmit_hash - generate a hash value based on the xmit policy
2968 * @bond: bonding device
2969 * @skb: buffer to use for headers
2970 * @count: modulo value
2971 *
2972 * This function will extract the necessary headers from the skb buffer and use
2973 * them to generate a hash based on the xmit_policy set in the bonding device
2974 * which will be reduced modulo count before returning.
John Eaglesham6b923cb2012-08-21 20:43:35 +00002975 */
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002976int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
John Eaglesham6b923cb2012-08-21 20:43:35 +00002977{
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002978 struct flow_keys flow;
2979 u32 hash;
John Eaglesham6b923cb2012-08-21 20:43:35 +00002980
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002981 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2982 !bond_flow_dissect(bond, skb, &flow))
2983 return bond_eth_hash(skb) % count;
Eric Dumazet43945422013-04-15 17:03:24 +00002984
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002985 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
2986 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
2987 hash = bond_eth_hash(skb);
2988 else
2989 hash = (__force u32)flow.ports;
2990 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
2991 hash ^= (hash >> 16);
2992 hash ^= (hash >> 8);
John Eaglesham6b923cb2012-08-21 20:43:35 +00002993
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02002994 return hash % count;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04002995}
2996
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997/*-------------------------- Device entry points ----------------------------*/
2998
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00002999static void bond_work_init_all(struct bonding *bond)
3000{
3001 INIT_DELAYED_WORK(&bond->mcast_work,
3002 bond_resend_igmp_join_requests_delayed);
3003 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3004 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3005 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3006 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3007 else
3008 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3009 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3010}
3011
3012static void bond_work_cancel_all(struct bonding *bond)
3013{
3014 cancel_delayed_work_sync(&bond->mii_work);
3015 cancel_delayed_work_sync(&bond->arp_work);
3016 cancel_delayed_work_sync(&bond->alb_work);
3017 cancel_delayed_work_sync(&bond->ad_work);
3018 cancel_delayed_work_sync(&bond->mcast_work);
3019}
3020
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021static int bond_open(struct net_device *bond_dev)
3022{
Wang Chen454d7c92008-11-12 23:37:49 -08003023 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003024 struct list_head *iter;
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003025 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003027 /* reset slave->backup and slave->inactive */
3028 read_lock(&bond->lock);
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02003029 if (bond_has_slaves(bond)) {
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003030 read_lock(&bond->curr_slave_lock);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003031 bond_for_each_slave(bond, slave, iter) {
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003032 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3033 && (slave != bond->curr_active_slave)) {
3034 bond_set_slave_inactive_flags(slave);
3035 } else {
3036 bond_set_slave_active_flags(slave);
3037 }
3038 }
3039 read_unlock(&bond->curr_slave_lock);
3040 }
3041 read_unlock(&bond->lock);
3042
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003043 bond_work_init_all(bond);
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00003044
Holger Eitzenberger58402052008-12-09 23:07:13 -08003045 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 /* bond_alb_initialize must be called before the timer
3047 * is started.
3048 */
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003049 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
stephen hemmingerb4739462010-01-25 23:34:15 +00003050 return -ENOMEM;
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003051 queue_delayed_work(bond->wq, &bond->alb_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 }
3053
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003054 if (bond->params.miimon) /* link check interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003055 queue_delayed_work(bond->wq, &bond->mii_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056
3057 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003058 queue_delayed_work(bond->wq, &bond->arp_work, 0);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07003059 if (bond->params.arp_validate)
Jiri Pirko3aba8912011-04-19 03:48:16 +00003060 bond->recv_probe = bond_arp_rcv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 }
3062
3063 if (bond->params.mode == BOND_MODE_8023AD) {
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003064 queue_delayed_work(bond->wq, &bond->ad_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 /* register to receive LACPDUs */
Jiri Pirko3aba8912011-04-19 03:48:16 +00003066 bond->recv_probe = bond_3ad_lacpdu_recv;
Jay Vosburghfd989c82008-11-04 17:51:16 -08003067 bond_3ad_initiate_agg_selection(bond, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 }
3069
3070 return 0;
3071}
3072
3073static int bond_close(struct net_device *bond_dev)
3074{
Wang Chen454d7c92008-11-12 23:37:49 -08003075 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003077 bond_work_cancel_all(bond);
nikolay@redhat.com6c8d23f2013-09-02 13:51:38 +02003078 bond->send_peer_notif = 0;
nikolay@redhat.comee8487c2013-09-02 13:51:39 +02003079 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080 bond_alb_deinitialize(bond);
Jiri Pirko3aba8912011-04-19 03:48:16 +00003081 bond->recv_probe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082
3083 return 0;
3084}
3085
Eric Dumazet28172732010-07-07 14:58:56 -07003086static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3087 struct rtnl_link_stats64 *stats)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088{
Wang Chen454d7c92008-11-12 23:37:49 -08003089 struct bonding *bond = netdev_priv(bond_dev);
Eric Dumazet28172732010-07-07 14:58:56 -07003090 struct rtnl_link_stats64 temp;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003091 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093
Eric Dumazet28172732010-07-07 14:58:56 -07003094 memset(stats, 0, sizeof(*stats));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095
3096 read_lock_bh(&bond->lock);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003097 bond_for_each_slave(bond, slave, iter) {
Ben Hutchingsbe1f3c22010-06-08 07:19:54 +00003098 const struct rtnl_link_stats64 *sstats =
Eric Dumazet28172732010-07-07 14:58:56 -07003099 dev_get_stats(slave->dev, &temp);
Stephen Hemmingereeda3fd2008-11-19 21:40:23 -08003100
Eric Dumazet28172732010-07-07 14:58:56 -07003101 stats->rx_packets += sstats->rx_packets;
3102 stats->rx_bytes += sstats->rx_bytes;
3103 stats->rx_errors += sstats->rx_errors;
3104 stats->rx_dropped += sstats->rx_dropped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105
Eric Dumazet28172732010-07-07 14:58:56 -07003106 stats->tx_packets += sstats->tx_packets;
3107 stats->tx_bytes += sstats->tx_bytes;
3108 stats->tx_errors += sstats->tx_errors;
3109 stats->tx_dropped += sstats->tx_dropped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
Eric Dumazet28172732010-07-07 14:58:56 -07003111 stats->multicast += sstats->multicast;
3112 stats->collisions += sstats->collisions;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113
Eric Dumazet28172732010-07-07 14:58:56 -07003114 stats->rx_length_errors += sstats->rx_length_errors;
3115 stats->rx_over_errors += sstats->rx_over_errors;
3116 stats->rx_crc_errors += sstats->rx_crc_errors;
3117 stats->rx_frame_errors += sstats->rx_frame_errors;
3118 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3119 stats->rx_missed_errors += sstats->rx_missed_errors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120
Eric Dumazet28172732010-07-07 14:58:56 -07003121 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3122 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3123 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3124 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3125 stats->tx_window_errors += sstats->tx_window_errors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 read_unlock_bh(&bond->lock);
3128
3129 return stats;
3130}
3131
3132static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3133{
Jiri Pirko080a06e2013-10-18 17:43:36 +02003134 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 struct net_device *slave_dev = NULL;
3136 struct ifbond k_binfo;
3137 struct ifbond __user *u_binfo = NULL;
3138 struct ifslave k_sinfo;
3139 struct ifslave __user *u_sinfo = NULL;
3140 struct mii_ioctl_data *mii = NULL;
Gao feng387ff9112013-01-31 16:31:00 +00003141 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 int res = 0;
3143
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003144 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145
3146 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 case SIOCGMIIPHY:
3148 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003149 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003151
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 mii->phy_id = 0;
3153 /* Fall Through */
3154 case SIOCGMIIREG:
3155 /*
3156 * We do this again just in case we were called by SIOCGMIIREG
3157 * instead of SIOCGMIIPHY.
3158 */
3159 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003160 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003162
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163
3164 if (mii->reg_num == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 mii->val_out = 0;
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07003166 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 read_lock(&bond->curr_slave_lock);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003168 if (netif_carrier_ok(bond->dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169 mii->val_out = BMSR_LSTATUS;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003170
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 read_unlock(&bond->curr_slave_lock);
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07003172 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 }
3174
3175 return 0;
3176 case BOND_INFO_QUERY_OLD:
3177 case SIOCBONDINFOQUERY:
3178 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3179
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003180 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182
3183 res = bond_info_query(bond_dev, &k_binfo);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003184 if (res == 0 &&
3185 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3186 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187
3188 return res;
3189 case BOND_SLAVE_INFO_QUERY_OLD:
3190 case SIOCBONDSLAVEINFOQUERY:
3191 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3192
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003193 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195
3196 res = bond_slave_info_query(bond_dev, &k_sinfo);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003197 if (res == 0 &&
3198 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3199 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200
3201 return res;
3202 default:
3203 /* Go on */
3204 break;
3205 }
3206
Gao feng387ff9112013-01-31 16:31:00 +00003207 net = dev_net(bond_dev);
3208
3209 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211
Ying Xue0917b932014-01-15 10:23:37 +08003212 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003214 pr_debug("slave_dev=%p:\n", slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003216 if (!slave_dev)
Ying Xue0917b932014-01-15 10:23:37 +08003217 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218
Ying Xue0917b932014-01-15 10:23:37 +08003219 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3220 switch (cmd) {
3221 case BOND_ENSLAVE_OLD:
3222 case SIOCBONDENSLAVE:
3223 res = bond_enslave(bond_dev, slave_dev);
3224 break;
3225 case BOND_RELEASE_OLD:
3226 case SIOCBONDRELEASE:
3227 res = bond_release(bond_dev, slave_dev);
3228 break;
3229 case BOND_SETHWADDR_OLD:
3230 case SIOCBONDSETHWADDR:
3231 bond_set_dev_addr(bond_dev, slave_dev);
3232 res = 0;
3233 break;
3234 case BOND_CHANGE_ACTIVE_OLD:
3235 case SIOCBONDCHANGEACTIVE:
3236 res = bond_option_active_slave_set(bond, slave_dev);
3237 break;
3238 default:
3239 res = -EOPNOTSUPP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240 }
3241
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 return res;
3243}
3244
Jiri Pirkod03462b2011-08-16 03:15:04 +00003245static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3246{
3247 struct bonding *bond = netdev_priv(bond_dev);
3248
3249 if (change & IFF_PROMISC)
3250 bond_set_promiscuity(bond,
3251 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3252
3253 if (change & IFF_ALLMULTI)
3254 bond_set_allmulti(bond,
3255 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3256}
3257
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003258static void bond_set_rx_mode(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259{
Wang Chen454d7c92008-11-12 23:37:49 -08003260 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003261 struct list_head *iter;
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003262 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263
Jay Vosburgh80ee5ad2008-01-29 18:07:44 -08003264
Veaceslav Falicob3241872013-09-28 21:18:56 +02003265 rcu_read_lock();
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003266 if (USES_PRIMARY(bond->params.mode)) {
Veaceslav Falicob3241872013-09-28 21:18:56 +02003267 slave = rcu_dereference(bond->curr_active_slave);
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003268 if (slave) {
3269 dev_uc_sync(slave->dev, bond_dev);
3270 dev_mc_sync(slave->dev, bond_dev);
3271 }
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003272 } else {
Veaceslav Falicob3241872013-09-28 21:18:56 +02003273 bond_for_each_slave_rcu(bond, slave, iter) {
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003274 dev_uc_sync_multiple(slave->dev, bond_dev);
3275 dev_mc_sync_multiple(slave->dev, bond_dev);
3276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277 }
Veaceslav Falicob3241872013-09-28 21:18:56 +02003278 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279}
3280
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003281static int bond_neigh_init(struct neighbour *n)
Stephen Hemminger00829822008-11-20 20:14:53 -08003282{
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003283 struct bonding *bond = netdev_priv(n->dev);
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003284 const struct net_device_ops *slave_ops;
3285 struct neigh_parms parms;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003286 struct slave *slave;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003287 int ret;
Stephen Hemminger00829822008-11-20 20:14:53 -08003288
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003289 slave = bond_first_slave(bond);
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003290 if (!slave)
3291 return 0;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003292 slave_ops = slave->dev->netdev_ops;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003293 if (!slave_ops->ndo_neigh_setup)
3294 return 0;
3295
3296 parms.neigh_setup = NULL;
3297 parms.neigh_cleanup = NULL;
3298 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3299 if (ret)
3300 return ret;
3301
3302 /*
3303 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3304 * after the last slave has been detached. Assumes that all slaves
3305 * utilize the same neigh_cleanup (true at this writing as only user
3306 * is ipoib).
3307 */
3308 n->parms->neigh_cleanup = parms.neigh_cleanup;
3309
3310 if (!parms.neigh_setup)
3311 return 0;
3312
3313 return parms.neigh_setup(n);
3314}
3315
3316/*
3317 * The bonding ndo_neigh_setup is called at init time beofre any
3318 * slave exists. So we must declare proxy setup function which will
3319 * be used at run time to resolve the actual slave neigh param setup.
Veaceslav Falico9918d5b2013-08-02 19:07:39 +02003320 *
3321 * It's also called by master devices (such as vlans) to setup their
3322 * underlying devices. In that case - do nothing, we're already set up from
3323 * our init.
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003324 */
3325static int bond_neigh_setup(struct net_device *dev,
3326 struct neigh_parms *parms)
3327{
Veaceslav Falico9918d5b2013-08-02 19:07:39 +02003328 /* modify only our neigh_parms */
3329 if (parms->dev == dev)
3330 parms->neigh_setup = bond_neigh_init;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003331
Stephen Hemminger00829822008-11-20 20:14:53 -08003332 return 0;
3333}
3334
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335/*
3336 * Change the MTU of all of a master's slaves to match the master
3337 */
3338static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3339{
Wang Chen454d7c92008-11-12 23:37:49 -08003340 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003341 struct slave *slave, *rollback_slave;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003342 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343 int res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003345 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003346 (bond_dev ? bond_dev->name : "None"), new_mtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347
3348 /* Can't hold bond->lock with bh disabled here since
3349 * some base drivers panic. On the other hand we can't
3350 * hold bond->lock without bh disabled because we'll
3351 * deadlock. The only solution is to rely on the fact
3352 * that we're under rtnl_lock here, and the slaves
3353 * list won't change. This doesn't solve the problem
3354 * of setting the slave's MTU while it is
3355 * transmitting, but the assumption is that the base
3356 * driver can handle that.
3357 *
3358 * TODO: figure out a way to safely iterate the slaves
3359 * list, but without holding a lock around the actual
3360 * call to the base driver.
3361 */
3362
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003363 bond_for_each_slave(bond, slave, iter) {
Veaceslav Falicoc8c23902013-09-25 09:20:25 +02003364 pr_debug("s %p c_m %p\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003365 slave,
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003366 slave->dev->netdev_ops->ndo_change_mtu);
Mitch Williamse944ef72005-11-09 10:36:50 -08003367
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 res = dev_set_mtu(slave->dev, new_mtu);
3369
3370 if (res) {
3371 /* If we failed to set the slave's mtu to the new value
3372 * we must abort the operation even in ACTIVE_BACKUP
3373 * mode, because if we allow the backup slaves to have
3374 * different mtu values than the active slave we'll
3375 * need to change their mtu when doing a failover. That
3376 * means changing their mtu from timer context, which
3377 * is probably not a good idea.
3378 */
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003379 pr_debug("err %d %s\n", res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380 goto unwind;
3381 }
3382 }
3383
3384 bond_dev->mtu = new_mtu;
3385
3386 return 0;
3387
3388unwind:
3389 /* unwind from head to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003390 bond_for_each_slave(bond, rollback_slave, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 int tmp_res;
3392
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003393 if (rollback_slave == slave)
3394 break;
3395
3396 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397 if (tmp_res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003398 pr_debug("unwind err %d dev %s\n",
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003399 tmp_res, rollback_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 }
3401 }
3402
3403 return res;
3404}
3405
3406/*
3407 * Change HW address
3408 *
3409 * Note that many devices must be down to change the HW address, and
3410 * downing the master releases all slaves. We can make bonds full of
3411 * bonding devices to test this, however.
3412 */
3413static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3414{
Wang Chen454d7c92008-11-12 23:37:49 -08003415 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003416 struct slave *slave, *rollback_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417 struct sockaddr *sa = addr, tmp_sa;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003418 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419 int res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003421 if (bond->params.mode == BOND_MODE_ALB)
3422 return bond_alb_set_mac_address(bond_dev, addr);
3423
3424
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003425 pr_debug("bond=%p, name=%s\n",
3426 bond, bond_dev ? bond_dev->name : "None");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427
Jay Vosburgh1b5acd22013-05-31 11:57:31 +00003428 /* If fail_over_mac is enabled, do nothing and return success.
3429 * Returning an error causes ifenslave to fail.
Jay Vosburghdd957c52007-10-09 19:57:24 -07003430 */
Jay Vosburgh1b5acd22013-05-31 11:57:31 +00003431 if (bond->params.fail_over_mac)
Jay Vosburghdd957c52007-10-09 19:57:24 -07003432 return 0;
Moni Shoua2ab82852007-10-09 19:43:39 -07003433
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003434 if (!is_valid_ether_addr(sa->sa_data))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435 return -EADDRNOTAVAIL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436
3437 /* Can't hold bond->lock with bh disabled here since
3438 * some base drivers panic. On the other hand we can't
3439 * hold bond->lock without bh disabled because we'll
3440 * deadlock. The only solution is to rely on the fact
3441 * that we're under rtnl_lock here, and the slaves
3442 * list won't change. This doesn't solve the problem
3443 * of setting the slave's hw address while it is
3444 * transmitting, but the assumption is that the base
3445 * driver can handle that.
3446 *
3447 * TODO: figure out a way to safely iterate the slaves
3448 * list, but without holding a lock around the actual
3449 * call to the base driver.
3450 */
3451
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003452 bond_for_each_slave(bond, slave, iter) {
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003453 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003454 pr_debug("slave %p %s\n", slave, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003455
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003456 if (slave_ops->ndo_set_mac_address == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 res = -EOPNOTSUPP;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003458 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 goto unwind;
3460 }
3461
3462 res = dev_set_mac_address(slave->dev, addr);
3463 if (res) {
3464 /* TODO: consider downing the slave
3465 * and retry ?
3466 * User should expect communications
3467 * breakage anyway until ARP finish
3468 * updating, so...
3469 */
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003470 pr_debug("err %d %s\n", res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471 goto unwind;
3472 }
3473 }
3474
3475 /* success */
3476 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3477 return 0;
3478
3479unwind:
3480 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3481 tmp_sa.sa_family = bond_dev->type;
3482
3483 /* unwind from head to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003484 bond_for_each_slave(bond, rollback_slave, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485 int tmp_res;
3486
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003487 if (rollback_slave == slave)
3488 break;
3489
3490 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 if (tmp_res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003492 pr_debug("unwind err %d dev %s\n",
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003493 tmp_res, rollback_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494 }
3495 }
3496
3497 return res;
3498}
3499
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003500/**
3501 * bond_xmit_slave_id - transmit skb through slave with slave_id
3502 * @bond: bonding device that is transmitting
3503 * @skb: buffer to transmit
3504 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3505 *
3506 * This function tries to transmit through slave with slave_id but in case
3507 * it fails, it tries to find the first available slave for transmission.
3508 * The skb is consumed in all cases, thus the function is void.
3509 */
stephen hemmingerda131dd2013-12-29 11:41:25 -08003510static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003511{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003512 struct list_head *iter;
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003513 struct slave *slave;
3514 int i = slave_id;
3515
3516 /* Here we start from the slave with slave_id */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003517 bond_for_each_slave_rcu(bond, slave, iter) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003518 if (--i < 0) {
3519 if (slave_can_tx(slave)) {
3520 bond_dev_queue_xmit(bond, skb, slave->dev);
3521 return;
3522 }
3523 }
3524 }
3525
3526 /* Here we start from the first slave up to slave_id */
3527 i = slave_id;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003528 bond_for_each_slave_rcu(bond, slave, iter) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003529 if (--i < 0)
3530 break;
3531 if (slave_can_tx(slave)) {
3532 bond_dev_queue_xmit(bond, skb, slave->dev);
3533 return;
3534 }
3535 }
3536 /* no slave that can tx has been found */
3537 kfree_skb(skb);
3538}
3539
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003540/**
3541 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3542 * @bond: bonding device to use
3543 *
3544 * Based on the value of the bonding device's packets_per_slave parameter
3545 * this function generates a slave id, which is usually used as the next
3546 * slave to transmit through.
3547 */
3548static u32 bond_rr_gen_slave_id(struct bonding *bond)
3549{
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003550 u32 slave_id;
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01003551 struct reciprocal_value reciprocal_packets_per_slave;
3552 int packets_per_slave = bond->params.packets_per_slave;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003553
3554 switch (packets_per_slave) {
3555 case 0:
3556 slave_id = prandom_u32();
3557 break;
3558 case 1:
3559 slave_id = bond->rr_tx_counter;
3560 break;
3561 default:
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01003562 reciprocal_packets_per_slave =
3563 bond->params.reciprocal_packets_per_slave;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003564 slave_id = reciprocal_divide(bond->rr_tx_counter,
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01003565 reciprocal_packets_per_slave);
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003566 break;
3567 }
3568 bond->rr_tx_counter++;
3569
3570 return slave_id;
3571}
3572
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3574{
Wang Chen454d7c92008-11-12 23:37:49 -08003575 struct bonding *bond = netdev_priv(bond_dev);
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003576 struct iphdr *iph = ip_hdr(skb);
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003577 struct slave *slave;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003578 u32 slave_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003580 /* Start with the curr_active_slave that joined the bond as the
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003581 * default for sending IGMP traffic. For failover purposes one
3582 * needs to maintain some consistency for the interface that will
3583 * send the join/membership reports. The curr_active_slave found
3584 * will send all of this type of traffic.
Jay Vosburghcf5f9042007-10-17 17:37:47 -07003585 */
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003586 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003587 slave = rcu_dereference(bond->curr_active_slave);
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003588 if (slave && slave_can_tx(slave))
3589 bond_dev_queue_xmit(bond, skb, slave->dev);
3590 else
3591 bond_xmit_slave_id(bond, skb, 0);
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003592 } else {
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003593 slave_id = bond_rr_gen_slave_id(bond);
3594 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595 }
Michał Mirosław0693e882011-05-07 01:48:02 +00003596
Patrick McHardyec634fe2009-07-05 19:23:38 -07003597 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598}
3599
3600/*
3601 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3602 * the bond has a usable interface.
3603 */
3604static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3605{
Wang Chen454d7c92008-11-12 23:37:49 -08003606 struct bonding *bond = netdev_priv(bond_dev);
nikolay@redhat.com71bc3b22013-08-01 16:54:48 +02003607 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003609 slave = rcu_dereference(bond->curr_active_slave);
nikolay@redhat.com71bc3b22013-08-01 16:54:48 +02003610 if (slave)
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003611 bond_dev_queue_xmit(bond, skb, slave->dev);
3612 else
Eric Dumazet04502432012-06-13 05:30:07 +00003613 kfree_skb(skb);
Michał Mirosław0693e882011-05-07 01:48:02 +00003614
Patrick McHardyec634fe2009-07-05 19:23:38 -07003615 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616}
3617
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003618/* In bond_xmit_xor() , we determine the output device by using a pre-
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003619 * determined xmit_hash_policy(), If the selected device is not enabled,
3620 * find the next active slave.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 */
3622static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3623{
Wang Chen454d7c92008-11-12 23:37:49 -08003624 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003626 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
Michał Mirosław0693e882011-05-07 01:48:02 +00003627
Patrick McHardyec634fe2009-07-05 19:23:38 -07003628 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629}
3630
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003631/* in broadcast mode, we send everything to all usable interfaces. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003632static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3633{
Wang Chen454d7c92008-11-12 23:37:49 -08003634 struct bonding *bond = netdev_priv(bond_dev);
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003635 struct slave *slave = NULL;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003636 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003637
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003638 bond_for_each_slave_rcu(bond, slave, iter) {
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003639 if (bond_is_last_slave(bond, slave))
3640 break;
3641 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3642 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003644 if (!skb2) {
3645 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3646 bond_dev->name);
3647 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003648 }
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003649 /* bond_dev_queue_xmit always returns 0 */
3650 bond_dev_queue_xmit(bond, skb2, slave->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651 }
3652 }
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02003653 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3654 bond_dev_queue_xmit(bond, skb, slave->dev);
3655 else
Eric Dumazet04502432012-06-13 05:30:07 +00003656 kfree_skb(skb);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003657
Patrick McHardyec634fe2009-07-05 19:23:38 -07003658 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659}
3660
3661/*------------------------- Device initialization ---------------------------*/
3662
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003663/*
3664 * Lookup the slave that corresponds to a qid
3665 */
3666static inline int bond_slave_override(struct bonding *bond,
3667 struct sk_buff *skb)
3668{
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003669 struct slave *slave = NULL;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003670 struct list_head *iter;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003671
Michał Mirosław0693e882011-05-07 01:48:02 +00003672 if (!skb->queue_mapping)
3673 return 1;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003674
3675 /* Find out if any slaves have the same mapping as this skb. */
dingtianhong3900f292014-01-02 09:13:06 +08003676 bond_for_each_slave_rcu(bond, slave, iter) {
3677 if (slave->queue_id == skb->queue_mapping) {
3678 if (slave_can_tx(slave)) {
3679 bond_dev_queue_xmit(bond, skb, slave->dev);
3680 return 0;
3681 }
3682 /* If the slave isn't UP, use default transmit policy. */
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003683 break;
3684 }
3685 }
3686
dingtianhong3900f292014-01-02 09:13:06 +08003687 return 1;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003688}
3689
Neil Horman374eeb52011-06-03 10:35:52 +00003690
Jason Wangf663dd92014-01-10 16:18:26 +08003691static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3692 void *accel_priv)
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003693{
3694 /*
3695 * This helper function exists to help dev_pick_tx get the correct
Phil Oesterfd0e4352011-03-14 06:22:04 +00003696 * destination queue. Using a helper function skips a call to
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003697 * skb_tx_hash and will put the skbs in the queue we expect on their
3698 * way down to the bonding driver.
3699 */
Phil Oesterfd0e4352011-03-14 06:22:04 +00003700 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3701
Neil Horman374eeb52011-06-03 10:35:52 +00003702 /*
3703 * Save the original txq to restore before passing to the driver
3704 */
Jiri Pirkodf4ab5b2012-07-20 02:28:49 +00003705 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
Neil Horman374eeb52011-06-03 10:35:52 +00003706
Phil Oesterfd0e4352011-03-14 06:22:04 +00003707 if (unlikely(txq >= dev->real_num_tx_queues)) {
David Decotignyd30ee672011-04-13 15:22:29 +00003708 do {
Phil Oesterfd0e4352011-03-14 06:22:04 +00003709 txq -= dev->real_num_tx_queues;
David Decotignyd30ee672011-04-13 15:22:29 +00003710 } while (txq >= dev->real_num_tx_queues);
Phil Oesterfd0e4352011-03-14 06:22:04 +00003711 }
3712 return txq;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003713}
3714
Michał Mirosław0693e882011-05-07 01:48:02 +00003715static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
Stephen Hemminger00829822008-11-20 20:14:53 -08003716{
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003717 struct bonding *bond = netdev_priv(dev);
3718
3719 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3720 if (!bond_slave_override(bond, skb))
3721 return NETDEV_TX_OK;
3722 }
Stephen Hemminger00829822008-11-20 20:14:53 -08003723
3724 switch (bond->params.mode) {
3725 case BOND_MODE_ROUNDROBIN:
3726 return bond_xmit_roundrobin(skb, dev);
3727 case BOND_MODE_ACTIVEBACKUP:
3728 return bond_xmit_activebackup(skb, dev);
3729 case BOND_MODE_XOR:
3730 return bond_xmit_xor(skb, dev);
3731 case BOND_MODE_BROADCAST:
3732 return bond_xmit_broadcast(skb, dev);
3733 case BOND_MODE_8023AD:
3734 return bond_3ad_xmit_xor(skb, dev);
3735 case BOND_MODE_ALB:
3736 case BOND_MODE_TLB:
3737 return bond_alb_xmit(skb, dev);
3738 default:
3739 /* Should never happen, mode already checked */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003740 pr_err("%s: Error: Unknown bonding mode %d\n",
3741 dev->name, bond->params.mode);
Stephen Hemminger00829822008-11-20 20:14:53 -08003742 WARN_ON_ONCE(1);
Eric Dumazet04502432012-06-13 05:30:07 +00003743 kfree_skb(skb);
Stephen Hemminger00829822008-11-20 20:14:53 -08003744 return NETDEV_TX_OK;
3745 }
3746}
3747
Michał Mirosław0693e882011-05-07 01:48:02 +00003748static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3749{
3750 struct bonding *bond = netdev_priv(dev);
3751 netdev_tx_t ret = NETDEV_TX_OK;
3752
3753 /*
3754 * If we risk deadlock from transmitting this in the
3755 * netpoll path, tell netpoll to queue the frame for later tx
3756 */
3757 if (is_netpoll_tx_blocked(dev))
3758 return NETDEV_TX_BUSY;
3759
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003760 rcu_read_lock();
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02003761 if (bond_has_slaves(bond))
Michał Mirosław0693e882011-05-07 01:48:02 +00003762 ret = __bond_start_xmit(skb, dev);
3763 else
Eric Dumazet04502432012-06-13 05:30:07 +00003764 kfree_skb(skb);
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003765 rcu_read_unlock();
Michał Mirosław0693e882011-05-07 01:48:02 +00003766
3767 return ret;
3768}
Stephen Hemminger00829822008-11-20 20:14:53 -08003769
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003770static int bond_ethtool_get_settings(struct net_device *bond_dev,
3771 struct ethtool_cmd *ecmd)
3772{
3773 struct bonding *bond = netdev_priv(bond_dev);
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003774 unsigned long speed = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003775 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003776 struct slave *slave;
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003777
3778 ecmd->duplex = DUPLEX_UNKNOWN;
3779 ecmd->port = PORT_OTHER;
3780
3781 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3782 * do not need to check mode. Though link speed might not represent
3783 * the true receive or transmit bandwidth (not all modes are symmetric)
3784 * this is an accurate maximum.
3785 */
3786 read_lock(&bond->lock);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003787 bond_for_each_slave(bond, slave, iter) {
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003788 if (SLAVE_IS_OK(slave)) {
3789 if (slave->speed != SPEED_UNKNOWN)
3790 speed += slave->speed;
3791 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3792 slave->duplex != DUPLEX_UNKNOWN)
3793 ecmd->duplex = slave->duplex;
3794 }
3795 }
3796 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3797 read_unlock(&bond->lock);
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003798
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003799 return 0;
3800}
3801
Jay Vosburgh217df672005-09-26 16:11:50 -07003802static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
Jiri Pirko7826d432013-01-06 00:44:26 +00003803 struct ethtool_drvinfo *drvinfo)
Jay Vosburgh217df672005-09-26 16:11:50 -07003804{
Jiri Pirko7826d432013-01-06 00:44:26 +00003805 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3806 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3807 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3808 BOND_ABI_VERSION);
Jay Vosburgh217df672005-09-26 16:11:50 -07003809}
3810
Jeff Garzik7282d492006-09-13 14:30:00 -04003811static const struct ethtool_ops bond_ethtool_ops = {
Jay Vosburgh217df672005-09-26 16:11:50 -07003812 .get_drvinfo = bond_ethtool_get_drvinfo,
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00003813 .get_settings = bond_ethtool_get_settings,
Stephen Hemmingerfa53eba2008-09-13 21:17:09 -04003814 .get_link = ethtool_op_get_link,
Arthur Kepner8531c5f2005-08-23 01:34:53 -04003815};
3816
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003817static const struct net_device_ops bond_netdev_ops = {
Stephen Hemminger181470f2009-06-12 19:02:52 +00003818 .ndo_init = bond_init,
Stephen Hemminger9e716262009-06-12 19:02:47 +00003819 .ndo_uninit = bond_uninit,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003820 .ndo_open = bond_open,
3821 .ndo_stop = bond_close,
Stephen Hemminger00829822008-11-20 20:14:53 -08003822 .ndo_start_xmit = bond_start_xmit,
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00003823 .ndo_select_queue = bond_select_queue,
Ben Hutchingsbe1f3c22010-06-08 07:19:54 +00003824 .ndo_get_stats64 = bond_get_stats,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003825 .ndo_do_ioctl = bond_do_ioctl,
Jiri Pirkod03462b2011-08-16 03:15:04 +00003826 .ndo_change_rx_flags = bond_change_rx_flags,
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003827 .ndo_set_rx_mode = bond_set_rx_mode,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003828 .ndo_change_mtu = bond_change_mtu,
Jiri Pirkocc0e4072011-07-20 04:54:46 +00003829 .ndo_set_mac_address = bond_set_mac_address,
Stephen Hemminger00829822008-11-20 20:14:53 -08003830 .ndo_neigh_setup = bond_neigh_setup,
Jiri Pirkocc0e4072011-07-20 04:54:46 +00003831 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003832 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
WANG Congf6dc31a2010-05-06 00:48:51 -07003833#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +00003834 .ndo_netpoll_setup = bond_netpoll_setup,
WANG Congf6dc31a2010-05-06 00:48:51 -07003835 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3836 .ndo_poll_controller = bond_poll_controller,
3837#endif
Jiri Pirko9232ecc2011-02-13 09:33:01 +00003838 .ndo_add_slave = bond_enslave,
3839 .ndo_del_slave = bond_release,
sfeldma@cumulusnetworks.com1d3ee882014-01-16 22:57:56 -08003840 .ndo_get_slave = bond_get_slave,
Michał Mirosławb2a103e2011-05-07 03:22:17 +00003841 .ndo_fix_features = bond_fix_features,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003842};
3843
Doug Goldsteinb3f92b62013-02-18 14:59:23 +00003844static const struct device_type bond_type = {
3845 .name = "bond",
3846};
3847
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00003848static void bond_destructor(struct net_device *bond_dev)
3849{
3850 struct bonding *bond = netdev_priv(bond_dev);
3851 if (bond->wq)
3852 destroy_workqueue(bond->wq);
3853 free_netdev(bond_dev);
3854}
3855
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02003856void bond_setup(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857{
Wang Chen454d7c92008-11-12 23:37:49 -08003858 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859
Linus Torvalds1da177e2005-04-16 15:20:36 -07003860 /* initialize rwlocks */
3861 rwlock_init(&bond->lock);
3862 rwlock_init(&bond->curr_slave_lock);
Stephen Hemmingerd2991f72009-06-12 19:02:44 +00003863 bond->params = bonding_defaults;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864
3865 /* Initialize pointers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866 bond->dev = bond_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867
3868 /* Initialize the device entry points */
Stephen Hemminger181470f2009-06-12 19:02:52 +00003869 ether_setup(bond_dev);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003870 bond_dev->netdev_ops = &bond_netdev_ops;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04003871 bond_dev->ethtool_ops = &bond_ethtool_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00003873 bond_dev->destructor = bond_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874
Doug Goldsteinb3f92b62013-02-18 14:59:23 +00003875 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3876
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877 /* Initialize the device options */
3878 bond_dev->tx_queue_len = 0;
3879 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
Jay Vosburgh0b680e72006-09-22 21:54:10 -07003880 bond_dev->priv_flags |= IFF_BONDING;
Neil Horman550fd082011-07-26 06:05:38 +00003881 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
Stephen Hemminger181470f2009-06-12 19:02:52 +00003882
Linus Torvalds1da177e2005-04-16 15:20:36 -07003883 /* At first, we block adding VLANs. That's the only way to
3884 * prevent problems that occur when adding VLANs over an
3885 * empty bond. The block will be removed once non-challenged
3886 * slaves are enslaved.
3887 */
3888 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3889
Herbert Xu932ff272006-06-09 12:20:56 -07003890 /* don't acquire bond device's netif_tx_lock when
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 * transmitting */
3892 bond_dev->features |= NETIF_F_LLTX;
3893
3894 /* By default, we declare the bond to be fully
3895 * VLAN hardware accelerated capable. Special
3896 * care is taken in the various xmit functions
3897 * when there are slaves that are not hw accel
3898 * capable
3899 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900
Michał Mirosławb2a103e2011-05-07 03:22:17 +00003901 bond_dev->hw_features = BOND_VLAN_FEATURES |
Patrick McHardyf6469682013-04-19 02:04:27 +00003902 NETIF_F_HW_VLAN_CTAG_TX |
3903 NETIF_F_HW_VLAN_CTAG_RX |
3904 NETIF_F_HW_VLAN_CTAG_FILTER;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00003905
Michał Mirosław34324dc2011-11-15 15:29:55 +00003906 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
Michał Mirosławb2a103e2011-05-07 03:22:17 +00003907 bond_dev->features |= bond_dev->hw_features;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003908}
3909
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00003910/*
3911* Destroy a bonding device.
3912* Must be under rtnl_lock when this function is called.
3913*/
3914static void bond_uninit(struct net_device *bond_dev)
Jay Vosburgha434e432008-10-30 17:41:15 -07003915{
Wang Chen454d7c92008-11-12 23:37:49 -08003916 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico544a0282013-09-25 09:20:15 +02003917 struct list_head *iter;
3918 struct slave *slave;
Jay Vosburgha434e432008-10-30 17:41:15 -07003919
WANG Congf6dc31a2010-05-06 00:48:51 -07003920 bond_netpoll_cleanup(bond_dev);
3921
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00003922 /* Release the bonded slaves */
Veaceslav Falico544a0282013-09-25 09:20:15 +02003923 bond_for_each_slave(bond, slave, iter)
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003924 __bond_release_one(bond_dev, slave->dev, true);
nikolay@redhat.com08963412013-02-18 14:09:42 +00003925 pr_info("%s: released all slaves\n", bond_dev->name);
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00003926
Jay Vosburgha434e432008-10-30 17:41:15 -07003927 list_del(&bond->bond_list);
3928
Taku Izumif073c7c2010-12-09 15:17:13 +00003929 bond_debug_unregister(bond);
Jay Vosburgha434e432008-10-30 17:41:15 -07003930}
3931
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932/*------------------------- Module initialization ---------------------------*/
3933
sfeldma@cumulusnetworks.comec029fa2014-01-03 14:18:49 -08003934int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3935{
3936 int i;
3937
3938 for (i = 0; tbl[i].modename; i++)
3939 if (mode == tbl[i].mode)
3940 return tbl[i].mode;
3941
3942 return -1;
3943}
3944
3945static int bond_parm_tbl_lookup_name(const char *modename,
3946 const struct bond_parm_tbl *tbl)
3947{
3948 int i;
3949
3950 for (i = 0; tbl[i].modename; i++)
3951 if (strcmp(modename, tbl[i].modename) == 0)
3952 return tbl[i].mode;
3953
3954 return -1;
3955}
3956
Linus Torvalds1da177e2005-04-16 15:20:36 -07003957/*
3958 * Convert string input module parms. Accept either the
Jay Vosburghece95f72008-01-17 16:25:01 -08003959 * number of the mode or its string name. A bit complicated because
3960 * some mode names are substrings of other names, and calls from sysfs
3961 * may have whitespace in the name (trailing newlines, for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003962 */
Holger Eitzenberger325dcf72008-12-09 23:10:17 -08003963int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964{
sfeldma@cumulusnetworks.comec029fa2014-01-03 14:18:49 -08003965 int modeint;
3966 char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
Jay Vosburghece95f72008-01-17 16:25:01 -08003967
Jay Vosburgha42e5342008-01-29 18:07:43 -08003968 for (p = (char *)buf; *p; p++)
3969 if (!(isdigit(*p) || isspace(*p)))
3970 break;
3971
sfeldma@cumulusnetworks.comec029fa2014-01-03 14:18:49 -08003972 if (*p && sscanf(buf, "%20s", modestr) != 0)
3973 return bond_parm_tbl_lookup_name(modestr, tbl);
3974 else if (sscanf(buf, "%d", &modeint) != 0)
3975 return bond_parm_tbl_lookup(modeint, tbl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976
3977 return -1;
3978}
3979
3980static int bond_check_params(struct bond_params *params)
3981{
nikolay@redhat.com5a5c5fd2013-05-18 01:18:30 +00003982 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
Nikolay Aleksandrov2b3798d2014-01-22 14:53:17 +01003983 struct bond_opt_value newval, *valptr;
Veaceslav Falico8599b522013-06-24 11:49:34 +02003984 int arp_all_targets_value;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07003985
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986 /*
3987 * Convert string parameters.
3988 */
3989 if (mode) {
Nikolay Aleksandrov2b3798d2014-01-22 14:53:17 +01003990 bond_opt_initstr(&newval, mode);
3991 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
3992 if (!valptr) {
3993 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994 return -EINVAL;
3995 }
Nikolay Aleksandrov2b3798d2014-01-22 14:53:17 +01003996 bond_mode = valptr->value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003997 }
3998
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003999 if (xmit_hash_policy) {
4000 if ((bond_mode != BOND_MODE_XOR) &&
4001 (bond_mode != BOND_MODE_8023AD)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004002 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004003 bond_mode_name(bond_mode));
4004 } else {
Nikolay Aleksandrova4b32ce2014-01-22 14:53:19 +01004005 bond_opt_initstr(&newval, xmit_hash_policy);
4006 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4007 &newval);
4008 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004009 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004010 xmit_hash_policy);
4011 return -EINVAL;
4012 }
Nikolay Aleksandrova4b32ce2014-01-22 14:53:19 +01004013 xmit_hashtype = valptr->value;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004014 }
4015 }
4016
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017 if (lacp_rate) {
4018 if (bond_mode != BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004019 pr_info("lacp_rate param is irrelevant in mode %s\n",
4020 bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004021 } else {
Nikolay Aleksandrovd3131de2014-01-22 14:53:27 +01004022 bond_opt_initstr(&newval, lacp_rate);
4023 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4024 &newval);
4025 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004026 pr_err("Error: Invalid lacp rate \"%s\"\n",
Nikolay Aleksandrovd3131de2014-01-22 14:53:27 +01004027 lacp_rate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 return -EINVAL;
4029 }
Nikolay Aleksandrovd3131de2014-01-22 14:53:27 +01004030 lacp_fast = valptr->value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 }
4032 }
4033
Jay Vosburghfd989c82008-11-04 17:51:16 -08004034 if (ad_select) {
4035 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4036 if (params->ad_select == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004037 pr_err("Error: Invalid ad_select \"%s\"\n",
Jay Vosburghfd989c82008-11-04 17:51:16 -08004038 ad_select == NULL ? "NULL" : ad_select);
4039 return -EINVAL;
4040 }
4041
4042 if (bond_mode != BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004043 pr_warning("ad_select param only affects 802.3ad mode\n");
Jay Vosburghfd989c82008-11-04 17:51:16 -08004044 }
4045 } else {
4046 params->ad_select = BOND_AD_STABLE;
4047 }
4048
Nicolas de Pesloüanf5841302009-08-28 13:18:34 +00004049 if (max_bonds < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004050 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4051 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052 max_bonds = BOND_DEFAULT_MAX_BONDS;
4053 }
4054
4055 if (miimon < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004056 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4057 miimon, INT_MAX, BOND_LINK_MON_INTERV);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058 miimon = BOND_LINK_MON_INTERV;
4059 }
4060
4061 if (updelay < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004062 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4063 updelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004064 updelay = 0;
4065 }
4066
4067 if (downdelay < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004068 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4069 downdelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070 downdelay = 0;
4071 }
4072
4073 if ((use_carrier != 0) && (use_carrier != 1)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004074 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4075 use_carrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076 use_carrier = 1;
4077 }
4078
Ben Hutchingsad246c92011-04-26 15:25:52 +00004079 if (num_peer_notif < 0 || num_peer_notif > 255) {
4080 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4081 num_peer_notif);
4082 num_peer_notif = 1;
4083 }
4084
dingtianhong834db4b2013-12-21 14:40:17 +08004085 /* reset values for 802.3ad/TLB/ALB */
4086 if (BOND_NO_USES_ARP(bond_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004087 if (!miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004088 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004089 pr_warning("Forcing miimon to 100msec\n");
dingtianhongfe9d04a2013-11-22 22:28:43 +08004090 miimon = BOND_DEFAULT_MIIMON;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 }
4092 }
4093
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004094 if (tx_queues < 1 || tx_queues > 255) {
4095 pr_warning("Warning: tx_queues (%d) should be between "
4096 "1 and 255, resetting to %d\n",
4097 tx_queues, BOND_DEFAULT_TX_QUEUES);
4098 tx_queues = BOND_DEFAULT_TX_QUEUES;
4099 }
4100
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004101 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4102 pr_warning("Warning: all_slaves_active module parameter (%d), "
4103 "not of valid value (0/1), so it was set to "
4104 "0\n", all_slaves_active);
4105 all_slaves_active = 0;
4106 }
4107
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004108 if (resend_igmp < 0 || resend_igmp > 255) {
4109 pr_warning("Warning: resend_igmp (%d) should be between "
4110 "0 and 255, resetting to %d\n",
4111 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4112 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4113 }
4114
Nikolay Aleksandrovaa59d852014-01-22 14:53:18 +01004115 bond_opt_initval(&newval, packets_per_slave);
4116 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01004117 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4118 packets_per_slave, USHRT_MAX);
4119 packets_per_slave = 1;
4120 }
4121
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122 if (bond_mode == BOND_MODE_ALB) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004123 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",
4124 updelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125 }
4126
4127 if (!miimon) {
4128 if (updelay || downdelay) {
4129 /* just warn the user the up/down delay will have
4130 * no effect since miimon is zero...
4131 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004132 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4133 updelay, downdelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134 }
4135 } else {
4136 /* don't allow arp monitoring */
4137 if (arp_interval) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004138 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4139 miimon, arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004140 arp_interval = 0;
4141 }
4142
4143 if ((updelay % miimon) != 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004144 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4145 updelay, miimon,
4146 (updelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004147 }
4148
4149 updelay /= miimon;
4150
4151 if ((downdelay % miimon) != 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004152 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4153 downdelay, miimon,
4154 (downdelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155 }
4156
4157 downdelay /= miimon;
4158 }
4159
4160 if (arp_interval < 0) {
Nikolay Aleksandrov7bdb04e2014-01-22 14:53:23 +01004161 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to 0\n",
4162 arp_interval, INT_MAX);
4163 arp_interval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004164 }
4165
nikolay@redhat.com5a5c5fd2013-05-18 01:18:30 +00004166 for (arp_ip_count = 0, i = 0;
4167 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004168 /* not complete check, but should be good enough to
4169 catch mistakes */
dingtianhong89015c12013-12-04 18:59:31 +08004170 __be32 ip;
4171 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4172 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004173 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
nikolay@redhat.com5a5c5fd2013-05-18 01:18:30 +00004174 arp_ip_target[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004175 arp_interval = 0;
4176 } else {
Veaceslav Falico0afee4e2013-06-24 11:49:30 +02004177 if (bond_get_targets_ip(arp_target, ip) == -1)
4178 arp_target[arp_ip_count++] = ip;
4179 else
4180 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4181 &ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182 }
4183 }
4184
4185 if (arp_interval && !arp_ip_count) {
4186 /* don't allow arping if no arp_ip_target given... */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004187 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4188 arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189 arp_interval = 0;
4190 }
4191
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004192 if (arp_validate) {
4193 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004194 pr_err("arp_validate only supported in active-backup mode\n");
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004195 return -EINVAL;
4196 }
4197 if (!arp_interval) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004198 pr_err("arp_validate requires arp_interval\n");
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004199 return -EINVAL;
4200 }
4201
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004202 bond_opt_initstr(&newval, arp_validate);
4203 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4204 &newval);
4205 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004206 pr_err("Error: invalid arp_validate \"%s\"\n",
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004207 arp_validate);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004208 return -EINVAL;
4209 }
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004210 arp_validate_value = valptr->value;
4211 } else {
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004212 arp_validate_value = 0;
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004213 }
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004214
Veaceslav Falico8599b522013-06-24 11:49:34 +02004215 arp_all_targets_value = 0;
4216 if (arp_all_targets) {
Nikolay Aleksandrovedf36b22014-01-22 14:53:21 +01004217 bond_opt_initstr(&newval, arp_all_targets);
4218 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4219 &newval);
4220 if (!valptr) {
Veaceslav Falico8599b522013-06-24 11:49:34 +02004221 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4222 arp_all_targets);
4223 arp_all_targets_value = 0;
Nikolay Aleksandrovedf36b22014-01-22 14:53:21 +01004224 } else {
4225 arp_all_targets_value = valptr->value;
Veaceslav Falico8599b522013-06-24 11:49:34 +02004226 }
4227 }
4228
Linus Torvalds1da177e2005-04-16 15:20:36 -07004229 if (miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004230 pr_info("MII link monitoring set to %d ms\n", miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231 } else if (arp_interval) {
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004232 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4233 arp_validate_value);
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004234 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004235 arp_interval, valptr->string, arp_ip_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004236
4237 for (i = 0; i < arp_ip_count; i++)
Jiri Pirkoe5e2a8f2009-08-13 04:11:52 +00004238 pr_info(" %s", arp_ip_target[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239
Jiri Pirkoe5e2a8f2009-08-13 04:11:52 +00004240 pr_info("\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241
Jay Vosburghb8a97872008-06-13 18:12:04 -07004242 } else if (max_bonds) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243 /* miimon and arp_interval not set, we need one so things
4244 * work as expected, see bonding.txt for details
4245 */
Andy Gospodarekb2730f42011-07-27 10:09:26 +00004246 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 -07004247 }
4248
4249 if (primary && !USES_PRIMARY(bond_mode)) {
4250 /* currently, using a primary only makes sense
4251 * in active backup, TLB or ALB modes
4252 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004253 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4254 primary, bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004255 primary = NULL;
4256 }
4257
Jiri Pirkoa5499522009-09-25 03:28:09 +00004258 if (primary && primary_reselect) {
4259 primary_reselect_value = bond_parse_parm(primary_reselect,
4260 pri_reselect_tbl);
4261 if (primary_reselect_value == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004262 pr_err("Error: Invalid primary_reselect \"%s\"\n",
Jiri Pirkoa5499522009-09-25 03:28:09 +00004263 primary_reselect ==
4264 NULL ? "NULL" : primary_reselect);
4265 return -EINVAL;
4266 }
4267 } else {
4268 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4269 }
4270
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004271 if (fail_over_mac) {
Nikolay Aleksandrov1df6b6a2014-01-22 14:53:22 +01004272 bond_opt_initstr(&newval, fail_over_mac);
4273 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4274 &newval);
4275 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004276 pr_err("Error: invalid fail_over_mac \"%s\"\n",
Nikolay Aleksandrov1df6b6a2014-01-22 14:53:22 +01004277 fail_over_mac);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004278 return -EINVAL;
4279 }
Nikolay Aleksandrov1df6b6a2014-01-22 14:53:22 +01004280 fail_over_mac_value = valptr->value;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004281 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004282 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004283 } else {
4284 fail_over_mac_value = BOND_FOM_NONE;
4285 }
Jay Vosburghdd957c52007-10-09 19:57:24 -07004286
dingtianhong3a7129e2013-12-21 14:40:12 +08004287 if (lp_interval == 0) {
4288 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4289 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4290 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4291 }
4292
Linus Torvalds1da177e2005-04-16 15:20:36 -07004293 /* fill params struct with the proper values */
4294 params->mode = bond_mode;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004295 params->xmit_policy = xmit_hashtype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004296 params->miimon = miimon;
Ben Hutchingsad246c92011-04-26 15:25:52 +00004297 params->num_peer_notif = num_peer_notif;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004298 params->arp_interval = arp_interval;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004299 params->arp_validate = arp_validate_value;
Veaceslav Falico8599b522013-06-24 11:49:34 +02004300 params->arp_all_targets = arp_all_targets_value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004301 params->updelay = updelay;
4302 params->downdelay = downdelay;
4303 params->use_carrier = use_carrier;
4304 params->lacp_fast = lacp_fast;
4305 params->primary[0] = 0;
Jiri Pirkoa5499522009-09-25 03:28:09 +00004306 params->primary_reselect = primary_reselect_value;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004307 params->fail_over_mac = fail_over_mac_value;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004308 params->tx_queues = tx_queues;
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004309 params->all_slaves_active = all_slaves_active;
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004310 params->resend_igmp = resend_igmp;
stephen hemminger655f8912011-06-22 09:54:39 +00004311 params->min_links = min_links;
dingtianhong3a7129e2013-12-21 14:40:12 +08004312 params->lp_interval = lp_interval;
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01004313 params->packets_per_slave = packets_per_slave;
4314 if (packets_per_slave > 0) {
4315 params->reciprocal_packets_per_slave =
4316 reciprocal_value(packets_per_slave);
4317 } else {
4318 /* reciprocal_packets_per_slave is unused if
4319 * packets_per_slave is 0 or 1, just initialize it
4320 */
4321 params->reciprocal_packets_per_slave =
4322 (struct reciprocal_value) { 0 };
4323 }
4324
Linus Torvalds1da177e2005-04-16 15:20:36 -07004325 if (primary) {
4326 strncpy(params->primary, primary, IFNAMSIZ);
4327 params->primary[IFNAMSIZ - 1] = 0;
4328 }
4329
4330 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4331
4332 return 0;
4333}
4334
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004335static struct lock_class_key bonding_netdev_xmit_lock_key;
David S. Millercf508b12008-07-22 14:16:42 -07004336static struct lock_class_key bonding_netdev_addr_lock_key;
Eric Dumazet49ee4922012-10-03 23:05:26 +00004337static struct lock_class_key bonding_tx_busylock_key;
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004338
David S. Millere8a04642008-07-17 00:34:19 -07004339static void bond_set_lockdep_class_one(struct net_device *dev,
4340 struct netdev_queue *txq,
4341 void *_unused)
David S. Millerc773e842008-07-08 23:13:53 -07004342{
4343 lockdep_set_class(&txq->_xmit_lock,
4344 &bonding_netdev_xmit_lock_key);
4345}
4346
4347static void bond_set_lockdep_class(struct net_device *dev)
4348{
David S. Millercf508b12008-07-22 14:16:42 -07004349 lockdep_set_class(&dev->addr_list_lock,
4350 &bonding_netdev_addr_lock_key);
David S. Millere8a04642008-07-17 00:34:19 -07004351 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
Eric Dumazet49ee4922012-10-03 23:05:26 +00004352 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
David S. Millerc773e842008-07-08 23:13:53 -07004353}
4354
Stephen Hemminger181470f2009-06-12 19:02:52 +00004355/*
4356 * Called from registration process
4357 */
4358static int bond_init(struct net_device *bond_dev)
4359{
4360 struct bonding *bond = netdev_priv(bond_dev);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004361 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
Neil Horman9fe06172011-05-25 08:13:01 +00004362 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Stephen Hemminger181470f2009-06-12 19:02:52 +00004363
4364 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4365
Neil Horman9fe06172011-05-25 08:13:01 +00004366 /*
4367 * Initialize locks that may be required during
4368 * en/deslave operations. All of the bond_open work
4369 * (of which this is part) should really be moved to
4370 * a phase prior to dev_open
4371 */
4372 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4373 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4374
Stephen Hemminger181470f2009-06-12 19:02:52 +00004375 bond->wq = create_singlethread_workqueue(bond_dev->name);
4376 if (!bond->wq)
4377 return -ENOMEM;
4378
4379 bond_set_lockdep_class(bond_dev);
4380
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004381 list_add_tail(&bond->bond_list, &bn->dev_list);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004382
Eric W. Biederman6151b3d2009-10-29 14:18:22 +00004383 bond_prepare_sysfs_group(bond);
Jiri Pirko22bedad32010-04-01 21:22:57 +00004384
Taku Izumif073c7c2010-12-09 15:17:13 +00004385 bond_debug_register(bond);
4386
Jiri Pirko409cc1f2013-01-30 11:08:11 +01004387 /* Ensure valid dev_addr */
4388 if (is_zero_ether_addr(bond_dev->dev_addr) &&
nikolay@redhat.com97a1e632013-06-26 17:13:38 +02004389 bond_dev->addr_assign_type == NET_ADDR_PERM)
Jiri Pirko409cc1f2013-01-30 11:08:11 +01004390 eth_hw_addr_random(bond_dev);
Jiri Pirko409cc1f2013-01-30 11:08:11 +01004391
Stephen Hemminger181470f2009-06-12 19:02:52 +00004392 return 0;
4393}
4394
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004395unsigned int bond_get_num_tx_queues(void)
Jiri Pirkod5da4512011-08-10 06:09:44 +00004396{
stephen hemmingerefacb302012-04-10 18:34:43 +00004397 return tx_queues;
Jiri Pirkod5da4512011-08-10 06:09:44 +00004398}
4399
Mitch Williamsdfe60392005-11-09 10:36:04 -08004400/* Create a new bond based on the specified name and bonding parameters.
Jay Vosburghe4b91c42007-01-19 18:15:31 -08004401 * If name is NULL, obtain a suitable "bond%d" name for us.
Mitch Williamsdfe60392005-11-09 10:36:04 -08004402 * Caller must NOT hold rtnl_lock; we need to release it here before we
4403 * set up our sysfs entries.
4404 */
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004405int bond_create(struct net *net, const char *name)
Mitch Williamsdfe60392005-11-09 10:36:04 -08004406{
4407 struct net_device *bond_dev;
4408 int res;
4409
4410 rtnl_lock();
Jay Vosburgh027ea042008-01-17 16:25:02 -08004411
Jiri Pirko1c5cae82011-04-30 01:21:32 +00004412 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4413 name ? name : "bond%d",
4414 bond_setup, tx_queues);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004415 if (!bond_dev) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004416 pr_err("%s: eek! can't alloc netdev!\n", name);
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004417 rtnl_unlock();
4418 return -ENOMEM;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004419 }
4420
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004421 dev_net_set(bond_dev, net);
Eric W. Biederman88ead972009-10-29 14:18:25 +00004422 bond_dev->rtnl_link_ops = &bond_link_ops;
4423
Mitch Williamsdfe60392005-11-09 10:36:04 -08004424 res = register_netdevice(bond_dev);
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004425
Phil Oestere826eaf2011-03-14 06:22:05 +00004426 netif_carrier_off(bond_dev);
4427
Mitch Williamsdfe60392005-11-09 10:36:04 -08004428 rtnl_unlock();
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004429 if (res < 0)
4430 bond_destructor(bond_dev);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004431 return res;
4432}
4433
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004434static int __net_init bond_net_init(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004435{
Eric W. Biederman15449742009-11-29 15:46:04 +00004436 struct bond_net *bn = net_generic(net, bond_net_id);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004437
4438 bn->net = net;
4439 INIT_LIST_HEAD(&bn->dev_list);
4440
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004441 bond_create_proc_dir(bn);
Eric W. Biederman4c224002011-10-12 21:56:25 +00004442 bond_create_sysfs(bn);
Veaceslav Falico87a7b842013-06-24 11:49:29 +02004443
Eric W. Biederman15449742009-11-29 15:46:04 +00004444 return 0;
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004445}
4446
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004447static void __net_exit bond_net_exit(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004448{
Eric W. Biederman15449742009-11-29 15:46:04 +00004449 struct bond_net *bn = net_generic(net, bond_net_id);
nikolay@redhat.com69b02162013-04-06 00:54:38 +00004450 struct bonding *bond, *tmp_bond;
4451 LIST_HEAD(list);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004452
Eric W. Biederman4c224002011-10-12 21:56:25 +00004453 bond_destroy_sysfs(bn);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004454 bond_destroy_proc_dir(bn);
nikolay@redhat.com69b02162013-04-06 00:54:38 +00004455
4456 /* Kill off any bonds created after unregistering bond rtnl ops */
4457 rtnl_lock();
4458 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4459 unregister_netdevice_queue(bond->dev, &list);
4460 unregister_netdevice_many(&list);
4461 rtnl_unlock();
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004462}
4463
4464static struct pernet_operations bond_net_ops = {
4465 .init = bond_net_init,
4466 .exit = bond_net_exit,
Eric W. Biederman15449742009-11-29 15:46:04 +00004467 .id = &bond_net_id,
4468 .size = sizeof(struct bond_net),
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004469};
4470
Linus Torvalds1da177e2005-04-16 15:20:36 -07004471static int __init bonding_init(void)
4472{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004473 int i;
4474 int res;
4475
Amerigo Wangbd33acc2011-03-06 21:58:46 +00004476 pr_info("%s", bond_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004477
Mitch Williamsdfe60392005-11-09 10:36:04 -08004478 res = bond_check_params(&bonding_defaults);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004479 if (res)
Mitch Williamsdfe60392005-11-09 10:36:04 -08004480 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004481
Eric W. Biederman15449742009-11-29 15:46:04 +00004482 res = register_pernet_subsys(&bond_net_ops);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004483 if (res)
4484 goto out;
Jay Vosburgh027ea042008-01-17 16:25:02 -08004485
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004486 res = bond_netlink_init();
Eric W. Biederman88ead972009-10-29 14:18:25 +00004487 if (res)
Eric W. Biederman66391042009-10-29 23:58:54 +00004488 goto err_link;
Eric W. Biederman88ead972009-10-29 14:18:25 +00004489
Taku Izumif073c7c2010-12-09 15:17:13 +00004490 bond_create_debugfs();
4491
Linus Torvalds1da177e2005-04-16 15:20:36 -07004492 for (i = 0; i < max_bonds; i++) {
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004493 res = bond_create(&init_net, NULL);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004494 if (res)
4495 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004496 }
4497
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498 register_netdevice_notifier(&bond_netdev_notifier);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004499out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004500 return res;
Eric W. Biederman88ead972009-10-29 14:18:25 +00004501err:
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004502 bond_netlink_fini();
Eric W. Biederman66391042009-10-29 23:58:54 +00004503err_link:
Eric W. Biederman15449742009-11-29 15:46:04 +00004504 unregister_pernet_subsys(&bond_net_ops);
Eric W. Biederman88ead972009-10-29 14:18:25 +00004505 goto out;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004506
Linus Torvalds1da177e2005-04-16 15:20:36 -07004507}
4508
4509static void __exit bonding_exit(void)
4510{
4511 unregister_netdevice_notifier(&bond_netdev_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004512
Taku Izumif073c7c2010-12-09 15:17:13 +00004513 bond_destroy_debugfs();
Pavel Emelyanovae68c392008-05-02 17:49:39 -07004514
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004515 bond_netlink_fini();
nikolay@redhat.comffcdedb2013-04-06 00:54:37 +00004516 unregister_pernet_subsys(&bond_net_ops);
Neil Hormane843fa52010-10-13 16:01:50 +00004517
4518#ifdef CONFIG_NET_POLL_CONTROLLER
Neil Hormanfb4fa762010-12-06 09:05:50 +00004519 /*
4520 * Make sure we don't have an imbalance on our netpoll blocking
4521 */
4522 WARN_ON(atomic_read(&netpoll_block_tx));
Neil Hormane843fa52010-10-13 16:01:50 +00004523#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004524}
4525
4526module_init(bonding_init);
4527module_exit(bonding_exit);
4528MODULE_LICENSE("GPL");
4529MODULE_VERSION(DRV_VERSION);
4530MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4531MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");