blob: ee28ec9e0abaddd13053da9fda5c0cefc722555d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 */
33
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/kernel.h>
35#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/types.h>
37#include <linux/fcntl.h>
38#include <linux/interrupt.h>
39#include <linux/ptrace.h>
40#include <linux/ioport.h>
41#include <linux/in.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040042#include <net/ip.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/ip.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040044#include <linux/tcp.h>
45#include <linux/udp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/slab.h>
47#include <linux/string.h>
48#include <linux/init.h>
49#include <linux/timer.h>
50#include <linux/socket.h>
51#include <linux/ctype.h>
52#include <linux/inet.h>
53#include <linux/bitops.h>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000054#include <linux/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <asm/dma.h>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000056#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <linux/errno.h>
58#include <linux/netdevice.h>
59#include <linux/inetdevice.h>
Jay Vosburgha816c7c2007-02-28 17:03:37 -080060#include <linux/igmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#include <linux/etherdevice.h>
62#include <linux/skbuff.h>
63#include <net/sock.h>
64#include <linux/rtnetlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#include <linux/smp.h>
66#include <linux/if_ether.h>
67#include <net/arp.h>
68#include <linux/mii.h>
69#include <linux/ethtool.h>
70#include <linux/if_vlan.h>
71#include <linux/if_bonding.h>
David Sterbab63bb732007-12-06 23:40:33 -080072#include <linux/jiffies.h>
Neil Hormane843fa52010-10-13 16:01:50 +000073#include <linux/preempt.h>
Jay Vosburghc3ade5c2005-06-26 17:52:20 -040074#include <net/route.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020075#include <net/net_namespace.h>
Eric W. Biedermanec87fd32009-10-29 14:18:26 +000076#include <net/netns/generic.h>
Eric Dumazet5ee31c682012-06-12 06:03:51 +000077#include <net/pkt_sched.h>
nikolay@redhat.com278b2082013-08-01 16:54:51 +020078#include <linux/rculist.h>
Jiri Pirko1bd758e2015-05-12 14:56:07 +020079#include <net/flow_dissector.h>
Roopa Prabhuc158cba2015-01-29 22:40:16 -080080#include <net/switchdev.h>
David S. Miller1ef80192014-11-10 13:27:49 -050081#include <net/bonding.h>
82#include <net/bond_3ad.h>
83#include <net/bond_alb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Matan Barak73b5a6f22015-04-26 15:55:57 +030085#include "bonding_priv.h"
86
Linus Torvalds1da177e2005-04-16 15:20:36 -070087/*---------------------------- Module parameters ----------------------------*/
88
89/* monitor all links that often (in milliseconds). <=0 disables monitoring */
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
91static int max_bonds = BOND_DEFAULT_MAX_BONDS;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +000092static int tx_queues = BOND_DEFAULT_TX_QUEUES;
Ben Hutchingsad246c92011-04-26 15:25:52 +000093static int num_peer_notif = 1;
Nikolay Aleksandrovb98d9c62014-01-22 14:53:31 +010094static int miimon;
Stephen Hemminger3d632c32009-06-12 19:02:48 +000095static int updelay;
96static int downdelay;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097static int use_carrier = 1;
Stephen Hemminger3d632c32009-06-12 19:02:48 +000098static char *mode;
99static char *primary;
Jiri Pirkoa5499522009-09-25 03:28:09 +0000100static char *primary_reselect;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000101static char *lacp_rate;
stephen hemminger655f8912011-06-22 09:54:39 +0000102static int min_links;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000103static char *ad_select;
104static char *xmit_hash_policy;
Nikolay Aleksandrov7bdb04e2014-01-22 14:53:23 +0100105static int arp_interval;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000106static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107static char *arp_validate;
Veaceslav Falico8599b522013-06-24 11:49:34 +0200108static char *arp_all_targets;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000109static char *fail_over_mac;
dingtianhongb07ea072013-07-23 15:25:47 +0800110static int all_slaves_active;
Stephen Hemmingerd2991f72009-06-12 19:02:44 +0000111static struct bond_params bonding_defaults;
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000112static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +0100113static int packets_per_slave = 1;
dingtianhong3a7129e2013-12-21 14:40:12 +0800114static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116module_param(max_bonds, int, 0);
117MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
Andy Gospodarekbb1d9122010-06-02 08:40:18 +0000118module_param(tx_queues, int, 0);
119MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
Ben Hutchingsad246c92011-04-26 15:25:52 +0000120module_param_named(num_grat_arp, num_peer_notif, int, 0644);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000121MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
Ben Hutchingsad246c92011-04-26 15:25:52 +0000123module_param_named(num_unsol_na, num_peer_notif, int, 0644);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000124MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126module_param(miimon, int, 0);
127MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128module_param(updelay, int, 0);
129MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130module_param(downdelay, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800131MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 "in milliseconds");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133module_param(use_carrier, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800134MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
Debabrata Banerjeeb3c898e2018-05-16 14:02:13 -0400135 "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);
Zhu Yanjun0d039f32016-08-09 21:36:04 +0800155MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
Andy Gospodarek90e62472011-05-25 04:41:59 +0000156 "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);
Debabrata Banerjeee79c1052018-05-14 14:48:09 -0400162MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
Andy Gospodarek90e62472011-05-25 04:41:59 +0000163 "0 for layer 2 (default), 1 for layer 3+4, "
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +0200164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166module_param(arp_interval, int, 0);
167MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168module_param_array(arp_ip_target, charp, NULL, 0);
169MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700170module_param(arp_validate, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000171MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
Veaceslav Falico8599b522013-06-24 11:49:34 +0200174module_param(arp_all_targets, charp, 0);
175MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700176module_param(fail_over_mac, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000177MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000180module_param(all_slaves_active, int, 0);
Masanari Iida37b70212014-09-09 18:07:55 +0900181MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
Andy Gospodarek90e62472011-05-25 04:41:59 +0000182 "by setting active flag for all slaves; "
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000183 "0 for never (default), 1 for always.");
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000184module_param(resend_igmp, int, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000185MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 "link failure");
Nikolay Aleksandrov73958322013-11-05 13:51:41 +0100187module_param(packets_per_slave, int, 0);
188MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
dingtianhong3a7129e2013-12-21 14:40:12 +0800191module_param(lp_interval, uint, 0);
192MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196/*----------------------------- Global variables ----------------------------*/
197
Neil Hormane843fa52010-10-13 16:01:50 +0000198#ifdef CONFIG_NET_POLL_CONTROLLER
Neil Hormanfb4fa762010-12-06 09:05:50 +0000199atomic_t netpoll_block_tx = ATOMIC_INIT(0);
Neil Hormane843fa52010-10-13 16:01:50 +0000200#endif
201
Alexey Dobriyanc7d03a02016-11-17 04:58:21 +0300202unsigned int bond_net_id __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204/*-------------------------- Forward declarations ---------------------------*/
205
Stephen Hemminger181470f2009-06-12 19:02:52 +0000206static int bond_init(struct net_device *bond_dev);
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +0000207static void bond_uninit(struct net_device *bond_dev);
stephen hemmingerbc1f4472017-01-06 19:12:52 -0800208static void bond_get_stats(struct net_device *bond_dev,
209 struct rtnl_link_stats64 *stats);
Mahesh Bandewaree637712014-10-04 17:45:01 -0700210static void bond_slave_arr_handler(struct work_struct *work);
Jay Vosburgh21a75f02016-02-02 13:35:56 -0800211static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
212 int mod);
Mahesh Bandeward4859d72018-09-24 14:40:11 -0700213static void bond_netdev_notify_work(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
215/*---------------------------- General routines -----------------------------*/
216
Amerigo Wangbd33acc2011-03-06 21:58:46 +0000217const char *bond_mode_name(int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800219 static const char *names[] = {
220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000224 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800225 [BOND_MODE_TLB] = "transmit load balancing",
226 [BOND_MODE_ALB] = "adaptive load balancing",
227 };
228
Wang Sheng-Huif5280942013-07-24 14:53:26 +0800229 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return "unknown";
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800231
232 return names[mode];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
235/*---------------------------------- VLAN -----------------------------------*/
236
237/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 * bond_dev_queue_xmit - Prepare skb for xmit.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000239 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 * @bond: bond device that got this skb for tx.
241 * @skb: hw accel VLAN tagged skb to transmit
242 * @slave_dev: slave that is supposed to xmit this skbuff
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 */
dingtianhongd316ded2014-01-02 09:13:09 +0800244void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000245 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
Ben Hutchings83874512010-12-13 08:19:28 +0000247 skb->dev = slave_dev;
Neil Horman374eeb52011-06-03 10:35:52 +0000248
Eric Dumazet5ee31c682012-06-12 06:03:51 +0000249 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
Jiri Pirkodf4ab5b2012-07-20 02:28:49 +0000250 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
Tonghao Zhangae35c6f2018-05-11 02:53:11 -0700251 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
Neil Horman374eeb52011-06-03 10:35:52 +0000252
Amerigo Wange15c3c222012-08-10 01:24:45 +0000253 if (unlikely(netpoll_tx_running(bond->dev)))
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000254 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
Amerigo Wang080e4132011-02-17 23:43:33 +0000255 else
WANG Congf6dc31a2010-05-06 00:48:51 -0700256 dev_queue_xmit(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200259/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
Jiri Pirkocc0e4072011-07-20 04:54:46 +0000260 * We don't protect the slave list iteration with a lock because:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000265 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
273*/
274
275/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
279 */
Patrick McHardy80d5c362013-04-19 02:04:28 +0000280static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
281 __be16 proto, u16 vid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
Wang Chen454d7c92008-11-12 23:37:49 -0800283 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico81f23b12013-09-25 09:20:13 +0200284 struct slave *slave, *rollback_slave;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200285 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200286 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200288 bond_for_each_slave(bond, slave, iter) {
Patrick McHardy80d5c362013-04-19 02:04:28 +0000289 res = vlan_vid_add(slave->dev, proto, vid);
Jiri Pirko87002b02011-12-08 04:11:17 +0000290 if (res)
291 goto unwind;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 }
293
Jiri Pirko8e586132011-12-08 19:52:37 -0500294 return 0;
Jiri Pirko87002b02011-12-08 04:11:17 +0000295
296unwind:
Veaceslav Falico81f23b12013-09-25 09:20:13 +0200297 /* unwind to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200298 bond_for_each_slave(bond, rollback_slave, iter) {
Veaceslav Falico81f23b12013-09-25 09:20:13 +0200299 if (rollback_slave == slave)
300 break;
301
302 vlan_vid_del(rollback_slave->dev, proto, vid);
303 }
Jiri Pirko87002b02011-12-08 04:11:17 +0000304
305 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
308/**
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
312 */
Patrick McHardy80d5c362013-04-19 02:04:28 +0000313static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
314 __be16 proto, u16 vid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
Wang Chen454d7c92008-11-12 23:37:49 -0800316 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200317 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200320 bond_for_each_slave(bond, slave, iter)
Patrick McHardy80d5c362013-04-19 02:04:28 +0000321 vlan_vid_del(slave->dev, proto, vid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Veaceslav Falicoe868b0c2013-08-28 23:25:15 +0200323 if (bond_is_lb(bond))
324 bond_alb_clear_vlan(bond, vid);
Jiri Pirko8e586132011-12-08 19:52:37 -0500325
326 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329/*------------------------------- Link status -------------------------------*/
330
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200331/* Set the carrier state for the master according to the state of its
Jay Vosburghff59c452006-03-27 13:27:43 -0800332 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
333 * do special 802.3ad magic.
334 *
335 * Returns zero if carrier state does not change, nonzero if it does.
336 */
Jonathan Toppins2477bc92015-01-26 01:16:57 -0500337int bond_set_carrier(struct bonding *bond)
Jay Vosburghff59c452006-03-27 13:27:43 -0800338{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200339 struct list_head *iter;
Jay Vosburghff59c452006-03-27 13:27:43 -0800340 struct slave *slave;
Jay Vosburghff59c452006-03-27 13:27:43 -0800341
Veaceslav Falico0965a1f2013-09-25 09:20:21 +0200342 if (!bond_has_slaves(bond))
Jay Vosburghff59c452006-03-27 13:27:43 -0800343 goto down;
344
Veaceslav Falico01844092014-05-15 21:39:55 +0200345 if (BOND_MODE(bond) == BOND_MODE_8023AD)
Jay Vosburghff59c452006-03-27 13:27:43 -0800346 return bond_3ad_set_carrier(bond);
347
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200348 bond_for_each_slave(bond, slave, iter) {
Jay Vosburghff59c452006-03-27 13:27:43 -0800349 if (slave->link == BOND_LINK_UP) {
350 if (!netif_carrier_ok(bond->dev)) {
351 netif_carrier_on(bond->dev);
352 return 1;
353 }
354 return 0;
355 }
356 }
357
358down:
359 if (netif_carrier_ok(bond->dev)) {
360 netif_carrier_off(bond->dev);
361 return 1;
362 }
363 return 0;
364}
365
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200366/* Get link speed and duplex from the slave's base driver
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 * using ethtool. If for some reason the call fails or the
Weiping Pan98f41f62011-10-31 17:20:48 +0000368 * values are invalid, set speed and duplex to -1,
Mahesh Bandewarc4adfc82017-03-27 11:37:35 -0700369 * and return. Return 1 if speed or duplex settings are
370 * UNKNOWN; 0 otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 */
Mahesh Bandewarc4adfc82017-03-27 11:37:35 -0700372static int bond_update_speed_duplex(struct slave *slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
374 struct net_device *slave_dev = slave->dev;
David Decotigny98569092016-02-24 10:58:02 -0800375 struct ethtool_link_ksettings ecmd;
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700376 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Dan Carpenter589665f2011-11-04 08:21:38 +0000378 slave->speed = SPEED_UNKNOWN;
379 slave->duplex = DUPLEX_UNKNOWN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
David Decotigny98569092016-02-24 10:58:02 -0800381 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700382 if (res < 0)
Mahesh Bandewarc4adfc82017-03-27 11:37:35 -0700383 return 1;
David Decotigny98569092016-02-24 10:58:02 -0800384 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
Mahesh Bandewarc4adfc82017-03-27 11:37:35 -0700385 return 1;
David Decotigny98569092016-02-24 10:58:02 -0800386 switch (ecmd.base.duplex) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 case DUPLEX_FULL:
388 case DUPLEX_HALF:
389 break;
390 default:
Mahesh Bandewarc4adfc82017-03-27 11:37:35 -0700391 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 }
393
David Decotigny98569092016-02-24 10:58:02 -0800394 slave->speed = ecmd.base.speed;
395 slave->duplex = ecmd.base.duplex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
Mahesh Bandewarc4adfc82017-03-27 11:37:35 -0700397 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
399
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -0800400const char *bond_slave_link_status(s8 link)
401{
402 switch (link) {
403 case BOND_LINK_UP:
404 return "up";
405 case BOND_LINK_FAIL:
406 return "going down";
407 case BOND_LINK_DOWN:
408 return "down";
409 case BOND_LINK_BACK:
410 return "going back";
411 default:
412 return "unknown";
413 }
414}
415
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200416/* if <dev> supports MII link status reporting, check its link status.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 *
418 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000419 * depending upon the setting of the use_carrier parameter.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 *
421 * Return either BMSR_LSTATUS, meaning that the link is up (or we
422 * can't tell and just pretend it is), or 0, meaning that the link is
423 * down.
424 *
425 * If reporting is non-zero, instead of faking link up, return -1 if
426 * both ETHTOOL and MII ioctls fail (meaning the device does not
427 * support them). If use_carrier is set, return whatever it says.
428 * It'd be nice if there was a good way to tell if a driver supports
429 * netif_carrier, but there really isn't.
430 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000431static int bond_check_dev_link(struct bonding *bond,
432 struct net_device *slave_dev, int reporting)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800434 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Jiri Bohacd9d52832009-10-28 22:23:54 -0700435 int (*ioctl)(struct net_device *, struct ifreq *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 struct ifreq ifr;
437 struct mii_ioctl_data *mii;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Petri Gynther6c988852009-08-28 12:05:15 +0000439 if (!reporting && !netif_running(slave_dev))
440 return 0;
441
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800442 if (bond->params.use_carrier)
Debabrata Banerjeeb3c898e2018-05-16 14:02:13 -0400443 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Jiri Pirko29112f42009-04-24 01:58:23 +0000445 /* Try to get link status using Ethtool first. */
Ben Hutchingsc772dde2012-12-07 06:15:32 +0000446 if (slave_dev->ethtool_ops->get_link)
447 return slave_dev->ethtool_ops->get_link(slave_dev) ?
448 BMSR_LSTATUS : 0;
Jiri Pirko29112f42009-04-24 01:58:23 +0000449
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000450 /* Ethtool can't be used, fallback to MII ioctls. */
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800451 ioctl = slave_ops->ndo_do_ioctl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 if (ioctl) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200453 /* TODO: set pointer to correct ioctl on a per team member
454 * bases to make this more efficient. that is, once
455 * we determine the correct ioctl, we will always
456 * call it and not the others for that team
457 * member.
458 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200460 /* We cannot assume that SIOCGMIIPHY will also read a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 * register; not all network drivers (e.g., e100)
462 * support that.
463 */
464
465 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
466 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
467 mii = if_mii(&ifr);
Al Viro4ad41c12016-09-03 19:37:25 -0400468 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 mii->reg_num = MII_BMSR;
Al Viro4ad41c12016-09-03 19:37:25 -0400470 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000471 return mii->val_out & BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 }
473 }
474
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200475 /* If reporting, report that either there's no dev->do_ioctl,
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700476 * or both SIOCGMIIREG and get_link failed (meaning that we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 * cannot report link status). If not reporting, pretend
478 * we're ok.
479 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000480 return reporting ? -1 : BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481}
482
483/*----------------------------- Multicast list ------------------------------*/
484
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200485/* Push the promiscuity flag down to appropriate slaves */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700486static int bond_set_promiscuity(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200488 struct list_head *iter;
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700489 int err = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200490
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200491 if (bond_uses_primary(bond)) {
Veaceslav Falico14056e72014-07-16 18:32:01 +0200492 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
Eric Dumazet4740d632014-07-15 06:56:55 -0700493
Eric Dumazet4740d632014-07-15 06:56:55 -0700494 if (curr_active)
495 err = dev_set_promiscuity(curr_active->dev, inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 } else {
497 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200498
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200499 bond_for_each_slave(bond, slave, iter) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700500 err = dev_set_promiscuity(slave->dev, inc);
501 if (err)
502 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 }
504 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700505 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506}
507
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200508/* Push the allmulti flag down to all slaves */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700509static int bond_set_allmulti(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200511 struct list_head *iter;
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700512 int err = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200513
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200514 if (bond_uses_primary(bond)) {
Veaceslav Falico14056e72014-07-16 18:32:01 +0200515 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
Eric Dumazet4740d632014-07-15 06:56:55 -0700516
Eric Dumazet4740d632014-07-15 06:56:55 -0700517 if (curr_active)
518 err = dev_set_allmulti(curr_active->dev, inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 } else {
520 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +0200521
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200522 bond_for_each_slave(bond, slave, iter) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700523 err = dev_set_allmulti(slave->dev, inc);
524 if (err)
525 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 }
527 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700528 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529}
530
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200531/* Retrieve the list of registered multicast addresses for the bonding
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800532 * device and retransmit an IGMP JOIN request to the current active
533 * slave.
534 */
stephen hemminger379b7382010-10-15 11:02:56 +0000535static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000536{
537 struct bonding *bond = container_of(work, struct bonding,
Flavio Leitner94265cf2011-05-25 08:38:58 +0000538 mcast_work.work);
Veaceslav Falicoad999ee2013-03-26 04:10:02 +0000539
dingtianhongf2369102013-12-13 10:20:26 +0800540 if (!rtnl_trylock()) {
541 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
542 return;
543 }
544 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
545
546 if (bond->igmp_retrans > 1) {
547 bond->igmp_retrans--;
548 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
549 }
550 rtnl_unlock();
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800551}
552
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200553/* Flush bond's hardware addresses from slave */
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000554static void bond_hw_addr_flush(struct net_device *bond_dev,
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000555 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Wang Chen454d7c92008-11-12 23:37:49 -0800557 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000559 dev_uc_unsync(slave_dev, bond_dev);
560 dev_mc_unsync(slave_dev, bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Veaceslav Falico01844092014-05-15 21:39:55 +0200562 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 /* del lacpdu mc addr from mc list */
564 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
565
Jiri Pirko22bedad32010-04-01 21:22:57 +0000566 dev_mc_del(slave_dev, lacpdu_multicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
568}
569
570/*--------------------------- Active slave change ---------------------------*/
571
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000572/* Update the hardware address list and promisc/allmulti for the new and
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200573 * old active slaves (if any). Modes that are not using primary keep all
574 * slaves up date at all times; only the modes that use primary need to call
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000575 * this function to swap these settings during a failover.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 */
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000577static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
578 struct slave *old_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 if (old_active) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000581 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 dev_set_promiscuity(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000584 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 dev_set_allmulti(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000587 bond_hw_addr_flush(bond->dev, old_active->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 }
589
590 if (new_active) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700591 /* FIXME: Signal errors upstream. */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000592 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 dev_set_promiscuity(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000595 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 dev_set_allmulti(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
nikolay@redhat.comd632ce92013-04-18 07:33:38 +0000598 netif_addr_lock_bh(bond->dev);
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000599 dev_uc_sync(new_active->dev, bond->dev);
600 dev_mc_sync(new_active->dev, bond->dev);
nikolay@redhat.comd632ce92013-04-18 07:33:38 +0000601 netif_addr_unlock_bh(bond->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 }
603}
604
nikolay@redhat.comae0d6752013-06-26 17:13:39 +0200605/**
606 * bond_set_dev_addr - clone slave's address to bond
607 * @bond_dev: bond net device
608 * @slave_dev: slave net device
609 *
610 * Should be called with RTNL held.
611 */
612static void bond_set_dev_addr(struct net_device *bond_dev,
613 struct net_device *slave_dev)
614{
Zhang Shengjuc6140a22016-02-02 08:32:55 +0000615 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
616 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
nikolay@redhat.comae0d6752013-06-26 17:13:39 +0200617 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
618 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
619 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
620}
621
dingtianhonga951bc12015-07-16 16:30:02 +0800622static struct slave *bond_get_old_active(struct bonding *bond,
623 struct slave *new_active)
624{
625 struct slave *slave;
626 struct list_head *iter;
627
628 bond_for_each_slave(bond, slave, iter) {
629 if (slave == new_active)
630 continue;
631
632 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
633 return slave;
634 }
635
636 return NULL;
637}
638
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200639/* bond_do_fail_over_mac
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700640 *
641 * Perform special MAC address swapping for fail_over_mac settings
642 *
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200643 * Called with RTNL
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700644 */
645static void bond_do_fail_over_mac(struct bonding *bond,
646 struct slave *new_active,
647 struct slave *old_active)
648{
Jarod Wilsonfaeeb312017-04-04 17:32:42 -0400649 u8 tmp_mac[MAX_ADDR_LEN];
650 struct sockaddr_storage ss;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700651 int rv;
652
653 switch (bond->params.fail_over_mac) {
654 case BOND_FOM_ACTIVE:
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200655 if (new_active)
nikolay@redhat.comae0d6752013-06-26 17:13:39 +0200656 bond_set_dev_addr(bond->dev, new_active->dev);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700657 break;
658 case BOND_FOM_FOLLOW:
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200659 /* if new_active && old_active, swap them
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700660 * if just old_active, do nothing (going to no active slave)
661 * if just new_active, set new_active to bond's MAC
662 */
663 if (!new_active)
664 return;
665
dingtianhonga951bc12015-07-16 16:30:02 +0800666 if (!old_active)
667 old_active = bond_get_old_active(bond, new_active);
668
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700669 if (old_active) {
Jarod Wilsonfaeeb312017-04-04 17:32:42 -0400670 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
671 new_active->dev->addr_len);
672 bond_hw_addr_copy(ss.__data,
673 old_active->dev->dev_addr,
674 old_active->dev->addr_len);
675 ss.ss_family = new_active->dev->type;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700676 } else {
Jarod Wilsonfaeeb312017-04-04 17:32:42 -0400677 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
678 bond->dev->addr_len);
679 ss.ss_family = bond->dev->type;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700680 }
681
Jarod Wilsonfaeeb312017-04-04 17:32:42 -0400682 rv = dev_set_mac_address(new_active->dev,
683 (struct sockaddr *)&ss);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700684 if (rv) {
Veaceslav Falico76444f52014-07-15 19:35:58 +0200685 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
686 -rv, new_active->dev->name);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700687 goto out;
688 }
689
690 if (!old_active)
691 goto out;
692
Jarod Wilsonfaeeb312017-04-04 17:32:42 -0400693 bond_hw_addr_copy(ss.__data, tmp_mac,
694 new_active->dev->addr_len);
695 ss.ss_family = old_active->dev->type;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700696
Jarod Wilsonfaeeb312017-04-04 17:32:42 -0400697 rv = dev_set_mac_address(old_active->dev,
698 (struct sockaddr *)&ss);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700699 if (rv)
Veaceslav Falico76444f52014-07-15 19:35:58 +0200700 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
701 -rv, new_active->dev->name);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700702out:
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700703 break;
704 default:
Veaceslav Falico76444f52014-07-15 19:35:58 +0200705 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
706 bond->params.fail_over_mac);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700707 break;
708 }
709
710}
711
Mazhar Ranab5a983f2015-07-07 15:04:50 +0530712static struct slave *bond_choose_primary_or_current(struct bonding *bond)
Jiri Pirkoa5499522009-09-25 03:28:09 +0000713{
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +0200714 struct slave *prim = rtnl_dereference(bond->primary_slave);
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200715 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
Jiri Pirkoa5499522009-09-25 03:28:09 +0000716
Mazhar Ranab5a983f2015-07-07 15:04:50 +0530717 if (!prim || prim->link != BOND_LINK_UP) {
718 if (!curr || curr->link != BOND_LINK_UP)
719 return NULL;
720 return curr;
721 }
722
Jiri Pirkoa5499522009-09-25 03:28:09 +0000723 if (bond->force_primary) {
724 bond->force_primary = false;
Mazhar Ranab5a983f2015-07-07 15:04:50 +0530725 return prim;
Jiri Pirkoa5499522009-09-25 03:28:09 +0000726 }
Mazhar Ranab5a983f2015-07-07 15:04:50 +0530727
728 if (!curr || curr->link != BOND_LINK_UP)
729 return prim;
730
731 /* At this point, prim and curr are both up */
732 switch (bond->params.primary_reselect) {
733 case BOND_PRI_RESELECT_ALWAYS:
734 return prim;
735 case BOND_PRI_RESELECT_BETTER:
736 if (prim->speed < curr->speed)
737 return curr;
738 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
739 return curr;
740 return prim;
741 case BOND_PRI_RESELECT_FAILURE:
742 return curr;
743 default:
744 netdev_err(bond->dev, "impossible primary_reselect %d\n",
745 bond->params.primary_reselect);
746 return curr;
747 }
Jiri Pirkoa5499522009-09-25 03:28:09 +0000748}
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750/**
Mazhar Ranab5a983f2015-07-07 15:04:50 +0530751 * bond_find_best_slave - select the best available slave to be the active one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 * @bond: our bonding struct
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 */
754static struct slave *bond_find_best_slave(struct bonding *bond)
755{
Mazhar Ranab5a983f2015-07-07 15:04:50 +0530756 struct slave *slave, *bestslave = NULL;
Veaceslav Falico77140d22013-09-25 09:20:18 +0200757 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 int mintime = bond->params.updelay;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Mazhar Ranab5a983f2015-07-07 15:04:50 +0530760 slave = bond_choose_primary_or_current(bond);
761 if (slave)
762 return slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Veaceslav Falico77140d22013-09-25 09:20:18 +0200764 bond_for_each_slave(bond, slave, iter) {
765 if (slave->link == BOND_LINK_UP)
766 return slave;
Veaceslav Falicob6adc612014-05-15 21:39:57 +0200767 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
Veaceslav Falico77140d22013-09-25 09:20:18 +0200768 slave->delay < mintime) {
769 mintime = slave->delay;
770 bestslave = slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 }
772 }
773
774 return bestslave;
775}
776
Ben Hutchingsad246c92011-04-26 15:25:52 +0000777static bool bond_should_notify_peers(struct bonding *bond)
778{
dingtianhong4cb4f972013-12-13 10:19:39 +0800779 struct slave *slave;
780
781 rcu_read_lock();
782 slave = rcu_dereference(bond->curr_active_slave);
783 rcu_read_unlock();
Ben Hutchingsad246c92011-04-26 15:25:52 +0000784
Veaceslav Falico76444f52014-07-15 19:35:58 +0200785 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
786 slave ? slave->dev->name : "NULL");
Ben Hutchingsad246c92011-04-26 15:25:52 +0000787
788 if (!slave || !bond->send_peer_notif ||
Venkat Venkatsubrab02e3e92015-08-11 07:57:23 -0700789 !netif_carrier_ok(bond->dev) ||
Ben Hutchingsad246c92011-04-26 15:25:52 +0000790 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
791 return false;
792
Ben Hutchingsad246c92011-04-26 15:25:52 +0000793 return true;
794}
795
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796/**
797 * change_active_interface - change the active slave into the specified one
798 * @bond: our bonding struct
799 * @new: the new slave to make the active one
800 *
801 * Set the new slave to the bond's settings and unset them on the old
802 * curr_active_slave.
803 * Setting include flags, mc-list, promiscuity, allmulti, etc.
804 *
805 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
806 * because it is apparently the best available slave we have, even though its
807 * updelay hasn't timed out yet.
808 *
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200809 * Caller must hold RTNL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 */
Mitch Williamsa77b5322005-11-09 10:35:51 -0800811void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
Eric Dumazet4740d632014-07-15 06:56:55 -0700813 struct slave *old_active;
814
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200815 ASSERT_RTNL();
816
817 old_active = rtnl_dereference(bond->curr_active_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000819 if (old_active == new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
822 if (new_active) {
Veaceslav Falico8e603462014-02-18 07:48:46 +0100823 new_active->last_link_up = jiffies;
Jay Vosburghb2220ca2008-05-17 21:10:13 -0700824
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 if (new_active->link == BOND_LINK_BACK) {
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200826 if (bond_uses_primary(bond)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +0200827 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
828 new_active->dev->name,
829 (bond->params.updelay - new_active->delay) * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
831
832 new_active->delay = 0;
Jiri Pirko5d397062015-12-03 12:12:19 +0100833 bond_set_slave_link_state(new_active, BOND_LINK_UP,
834 BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Veaceslav Falico01844092014-05-15 21:39:55 +0200836 if (BOND_MODE(bond) == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Holger Eitzenberger58402052008-12-09 23:07:13 -0800839 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 } else {
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200842 if (bond_uses_primary(bond)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +0200843 netdev_info(bond->dev, "making interface %s the new active one\n",
844 new_active->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 }
846 }
847 }
848
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200849 if (bond_uses_primary(bond))
Jay Vosburgh303d1cb2013-05-31 11:57:30 +0000850 bond_hw_addr_swap(bond, new_active, old_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Holger Eitzenberger58402052008-12-09 23:07:13 -0800852 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 bond_alb_handle_active_change(bond, new_active);
Jay Vosburgh8f903c72006-02-21 16:36:44 -0800854 if (old_active)
dingtianhong5e5b0662014-02-26 11:05:22 +0800855 bond_set_slave_inactive_flags(old_active,
856 BOND_SLAVE_NOTIFY_NOW);
Jay Vosburgh8f903c72006-02-21 16:36:44 -0800857 if (new_active)
dingtianhong5e5b0662014-02-26 11:05:22 +0800858 bond_set_slave_active_flags(new_active,
859 BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 } else {
nikolay@redhat.com278b2082013-08-01 16:54:51 +0200861 rcu_assign_pointer(bond->curr_active_slave, new_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400863
Veaceslav Falico01844092014-05-15 21:39:55 +0200864 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000865 if (old_active)
dingtianhong5e5b0662014-02-26 11:05:22 +0800866 bond_set_slave_inactive_flags(old_active,
867 BOND_SLAVE_NOTIFY_NOW);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400868
869 if (new_active) {
Ben Hutchingsad246c92011-04-26 15:25:52 +0000870 bool should_notify_peers = false;
871
dingtianhong5e5b0662014-02-26 11:05:22 +0800872 bond_set_slave_active_flags(new_active,
873 BOND_SLAVE_NOTIFY_NOW);
Moni Shoua2ab82852007-10-09 19:43:39 -0700874
Or Gerlitz709f8a42008-06-13 18:12:01 -0700875 if (bond->params.fail_over_mac)
876 bond_do_fail_over_mac(bond, new_active,
877 old_active);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700878
Ben Hutchingsad246c92011-04-26 15:25:52 +0000879 if (netif_running(bond->dev)) {
880 bond->send_peer_notif =
881 bond->params.num_peer_notif;
882 should_notify_peers =
883 bond_should_notify_peers(bond);
884 }
885
Amerigo Wangb7bc2a52012-08-09 22:14:57 +0000886 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
Ben Hutchingsad246c92011-04-26 15:25:52 +0000887 if (should_notify_peers)
Amerigo Wangb7bc2a52012-08-09 22:14:57 +0000888 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
889 bond->dev);
Moni Shoua7893b242008-05-17 21:10:12 -0700890 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400891 }
Andy Gospodareka2fd9402010-03-25 14:49:05 +0000892
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000893 /* resend IGMP joins since active slave has changed or
Flavio Leitner94265cf2011-05-25 08:38:58 +0000894 * all were sent on curr_active_slave.
895 * resend only if bond is brought up with the affected
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +0200896 * bonding modes and the retransmission is enabled
897 */
Flavio Leitner94265cf2011-05-25 08:38:58 +0000898 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
Veaceslav Falicoec0865a2014-05-15 21:39:54 +0200899 ((bond_uses_primary(bond) && new_active) ||
Veaceslav Falico01844092014-05-15 21:39:55 +0200900 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000901 bond->igmp_retrans = bond->params.resend_igmp;
Nikolay Aleksandrov4beac022013-08-01 11:51:42 +0200902 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
Andy Gospodareka2fd9402010-03-25 14:49:05 +0000903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904}
905
906/**
907 * bond_select_active_slave - select a new active slave, if needed
908 * @bond: our bonding struct
909 *
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000910 * This functions should be called when one of the following occurs:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 * - The old curr_active_slave has been released or lost its link.
912 * - The primary_slave has got its link back.
913 * - A slave has got its link back and there's no old curr_active_slave.
914 *
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200915 * Caller must hold RTNL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 */
Mitch Williamsa77b5322005-11-09 10:35:51 -0800917void bond_select_active_slave(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918{
919 struct slave *best_slave;
Jay Vosburghff59c452006-03-27 13:27:43 -0800920 int rv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
Nikolay Aleksandrove0974582014-09-15 17:19:35 +0200922 ASSERT_RTNL();
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 best_slave = bond_find_best_slave(bond);
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +0200925 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 bond_change_active_slave(bond, best_slave);
Jay Vosburghff59c452006-03-27 13:27:43 -0800927 rv = bond_set_carrier(bond);
928 if (!rv)
929 return;
930
Zhang Shengjud66bd902016-02-03 02:02:32 +0000931 if (netif_carrier_ok(bond->dev))
Veaceslav Falico76444f52014-07-15 19:35:58 +0200932 netdev_info(bond->dev, "first active interface up!\n");
Zhang Shengjud66bd902016-02-03 02:02:32 +0000933 else
Veaceslav Falico76444f52014-07-15 19:35:58 +0200934 netdev_info(bond->dev, "now running without any active interface!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 }
936}
937
WANG Congf6dc31a2010-05-06 00:48:51 -0700938#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000939static inline int slave_enable_netpoll(struct slave *slave)
WANG Congf6dc31a2010-05-06 00:48:51 -0700940{
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000941 struct netpoll *np;
942 int err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -0700943
Eric W. Biedermana8779ec2014-03-27 15:36:38 -0700944 np = kzalloc(sizeof(*np), GFP_KERNEL);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000945 err = -ENOMEM;
946 if (!np)
947 goto out;
948
Eric W. Biedermana8779ec2014-03-27 15:36:38 -0700949 err = __netpoll_setup(np, slave->dev);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000950 if (err) {
951 kfree(np);
952 goto out;
WANG Congf6dc31a2010-05-06 00:48:51 -0700953 }
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000954 slave->np = np;
955out:
956 return err;
957}
958static inline void slave_disable_netpoll(struct slave *slave)
959{
960 struct netpoll *np = slave->np;
961
962 if (!np)
963 return;
964
965 slave->np = NULL;
Neil Horman2cde6ac2013-02-11 10:25:30 +0000966 __netpoll_free_async(np);
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000967}
WANG Congf6dc31a2010-05-06 00:48:51 -0700968
969static void bond_poll_controller(struct net_device *bond_dev)
970{
Mahesh Bandewar616f4542015-03-04 21:57:52 -0800971 struct bonding *bond = netdev_priv(bond_dev);
972 struct slave *slave = NULL;
973 struct list_head *iter;
974 struct ad_info ad_info;
Mahesh Bandewar616f4542015-03-04 21:57:52 -0800975
976 if (BOND_MODE(bond) == BOND_MODE_8023AD)
977 if (bond_3ad_get_active_agg_info(bond, &ad_info))
978 return;
979
Mahesh Bandewar616f4542015-03-04 21:57:52 -0800980 bond_for_each_slave_rcu(bond, slave, iter) {
Eric Dumazet93f62ad2018-09-21 15:27:39 -0700981 if (!bond_slave_is_up(slave))
Mahesh Bandewar616f4542015-03-04 21:57:52 -0800982 continue;
983
984 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
985 struct aggregator *agg =
986 SLAVE_AD_INFO(slave)->port.aggregator;
987
988 if (agg &&
989 agg->aggregator_identifier != ad_info.aggregator_id)
990 continue;
991 }
992
Eric Dumazet93f62ad2018-09-21 15:27:39 -0700993 netpoll_poll_dev(slave->dev);
Mahesh Bandewar616f4542015-03-04 21:57:52 -0800994 }
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000995}
996
dingtianhongc4cdef92013-07-23 15:25:27 +0800997static void bond_netpoll_cleanup(struct net_device *bond_dev)
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000998{
dingtianhongc4cdef92013-07-23 15:25:27 +0800999 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001000 struct list_head *iter;
Neil Hormanc2355e12010-10-13 16:01:49 +00001001 struct slave *slave;
Neil Hormanc2355e12010-10-13 16:01:49 +00001002
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001003 bond_for_each_slave(bond, slave, iter)
Veaceslav Falicob6adc612014-05-15 21:39:57 +02001004 if (bond_slave_is_up(slave))
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001005 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -07001006}
WANG Congf6dc31a2010-05-06 00:48:51 -07001007
Eric W. Biedermana8779ec2014-03-27 15:36:38 -07001008static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001009{
1010 struct bonding *bond = netdev_priv(dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001011 struct list_head *iter;
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001012 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001013 int err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -07001014
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001015 bond_for_each_slave(bond, slave, iter) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001016 err = slave_enable_netpoll(slave);
1017 if (err) {
dingtianhongc4cdef92013-07-23 15:25:27 +08001018 bond_netpoll_cleanup(dev);
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001019 break;
1020 }
1021 }
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001022 return err;
1023}
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001024#else
1025static inline int slave_enable_netpoll(struct slave *slave)
1026{
1027 return 0;
1028}
1029static inline void slave_disable_netpoll(struct slave *slave)
1030{
1031}
WANG Congf6dc31a2010-05-06 00:48:51 -07001032static void bond_netpoll_cleanup(struct net_device *bond_dev)
1033{
1034}
WANG Congf6dc31a2010-05-06 00:48:51 -07001035#endif
1036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037/*---------------------------------- IOCTL ----------------------------------*/
1038
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001039static netdev_features_t bond_fix_features(struct net_device *dev,
nikolay@redhat.com9b7b1652013-09-02 13:51:41 +02001040 netdev_features_t features)
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001041{
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001042 struct bonding *bond = netdev_priv(dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001043 struct list_head *iter;
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001044 netdev_features_t mask;
nikolay@redhat.com9b7b1652013-09-02 13:51:41 +02001045 struct slave *slave;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001046
Scott Feldman7889cbe2015-05-10 09:48:07 -07001047 mask = features;
Roopa Prabhuc158cba2015-01-29 22:40:16 -08001048
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001049 features &= ~NETIF_F_ONE_FOR_ALL;
1050 features |= NETIF_F_ALL_FOR_ALL;
1051
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001052 bond_for_each_slave(bond, slave, iter) {
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001053 features = netdev_increment_features(features,
1054 slave->dev->features,
1055 mask);
1056 }
Eric Dumazetb0ce3502013-05-16 07:34:53 +00001057 features = netdev_add_tso_features(features, mask);
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001058
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001059 return features;
1060}
1061
Tom Herberta1882222015-12-14 11:19:43 -08001062#define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
Michał Mirosław62f2a3a2011-07-13 14:10:29 +00001063 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1064 NETIF_F_HIGHDMA | NETIF_F_LRO)
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001065
Tom Herberta1882222015-12-14 11:19:43 -08001066#define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1067 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
Or Gerlitz5a7baa72014-06-17 16:11:09 +03001068
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001069static void bond_compute_features(struct bonding *bond)
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001070{
Eric Dumazet02875872014-10-05 18:38:35 -07001071 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1072 IFF_XMIT_DST_RELEASE_PERM;
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001073 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
Or Gerlitz5a7baa72014-06-17 16:11:09 +03001074 netdev_features_t enc_features = BOND_ENC_FEATURES;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001075 struct net_device *bond_dev = bond->dev;
1076 struct list_head *iter;
1077 struct slave *slave;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001078 unsigned short max_hard_header_len = ETH_HLEN;
Sarveshwar Bandi0e376bd2012-11-21 04:35:03 +00001079 unsigned int gso_max_size = GSO_MAX_SIZE;
1080 u16 gso_max_segs = GSO_MAX_SEGS;
Herbert Xub63365a2008-10-23 01:11:29 -07001081
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001082 if (!bond_has_slaves(bond))
Herbert Xub63365a2008-10-23 01:11:29 -07001083 goto done;
Michal Kubečeka9b3ace2014-05-20 08:29:35 +02001084 vlan_features &= NETIF_F_ALL_FOR_ALL;
Herbert Xub63365a2008-10-23 01:11:29 -07001085
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001086 bond_for_each_slave(bond, slave, iter) {
Jay Vosburgh278339a2009-08-28 12:05:12 +00001087 vlan_features = netdev_increment_features(vlan_features,
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001088 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1089
Or Gerlitz5a7baa72014-06-17 16:11:09 +03001090 enc_features = netdev_increment_features(enc_features,
1091 slave->dev->hw_enc_features,
1092 BOND_ENC_FEATURES);
Eric Dumazetb6fe83e2012-07-17 12:19:48 +00001093 dst_release_flag &= slave->dev->priv_flags;
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001094 if (slave->dev->hard_header_len > max_hard_header_len)
1095 max_hard_header_len = slave->dev->hard_header_len;
Sarveshwar Bandi0e376bd2012-11-21 04:35:03 +00001096
1097 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1098 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001099 }
Paolo Abeni19cdead2017-04-27 19:29:34 +02001100 bond_dev->hard_header_len = max_hard_header_len;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001101
Herbert Xub63365a2008-10-23 01:11:29 -07001102done:
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001103 bond_dev->vlan_features = vlan_features;
Willem de Bruijn8eea1ca2018-05-22 11:34:40 -04001104 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1105 NETIF_F_GSO_UDP_L4;
Sarveshwar Bandi0e376bd2012-11-21 04:35:03 +00001106 bond_dev->gso_max_segs = gso_max_segs;
1107 netif_set_gso_max_size(bond_dev, gso_max_size);
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001108
Eric Dumazet02875872014-10-05 18:38:35 -07001109 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1110 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1111 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1112 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
Eric Dumazetb6fe83e2012-07-17 12:19:48 +00001113
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001114 netdev_change_features(bond_dev);
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001115}
1116
Moni Shoua872254d2007-10-09 19:43:38 -07001117static void bond_setup_by_slave(struct net_device *bond_dev,
1118 struct net_device *slave_dev)
1119{
Stephen Hemminger00829822008-11-20 20:14:53 -08001120 bond_dev->header_ops = slave_dev->header_ops;
Moni Shoua872254d2007-10-09 19:43:38 -07001121
1122 bond_dev->type = slave_dev->type;
1123 bond_dev->hard_header_len = slave_dev->hard_header_len;
1124 bond_dev->addr_len = slave_dev->addr_len;
1125
1126 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1127 slave_dev->addr_len);
1128}
1129
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001130/* On bonding slaves other than the currently active slave, suppress
Jiri Pirko3aba8912011-04-19 03:48:16 +00001131 * duplicates except for alb non-mcast/bcast.
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001132 */
1133static bool bond_should_deliver_exact_match(struct sk_buff *skb,
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001134 struct slave *slave,
1135 struct bonding *bond)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001136{
Jiri Pirko2d7011c2011-03-16 08:46:43 +00001137 if (bond_is_slave_inactive(slave)) {
Veaceslav Falico01844092014-05-15 21:39:55 +02001138 if (BOND_MODE(bond) == BOND_MODE_ALB &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001139 skb->pkt_type != PACKET_BROADCAST &&
1140 skb->pkt_type != PACKET_MULTICAST)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001141 return false;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001142 return true;
1143 }
1144 return false;
1145}
1146
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001147static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001148{
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001149 struct sk_buff *skb = *pskb;
Jiri Pirkof1c17752011-03-12 03:14:35 +00001150 struct slave *slave;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001151 struct bonding *bond;
Eric Dumazetde063b72012-06-11 19:23:07 +00001152 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1153 struct slave *);
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00001154 int ret = RX_HANDLER_ANOTHER;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001155
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001156 skb = skb_share_check(skb, GFP_ATOMIC);
1157 if (unlikely(!skb))
1158 return RX_HANDLER_CONSUMED;
1159
1160 *pskb = skb;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001161
Jiri Pirko35d48902011-03-22 02:38:12 +00001162 slave = bond_slave_get_rcu(skb->dev);
1163 bond = slave->bond;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001164
Mark Rutland6aa7de02017-10-23 14:07:29 -07001165 recv_probe = READ_ONCE(bond->recv_probe);
Mitsuo Hayasaka4d974802011-10-12 16:04:29 +00001166 if (recv_probe) {
Eric Dumazetde063b72012-06-11 19:23:07 +00001167 ret = recv_probe(skb, bond, slave);
1168 if (ret == RX_HANDLER_CONSUMED) {
1169 consume_skb(skb);
1170 return ret;
Jiri Pirko3aba8912011-04-19 03:48:16 +00001171 }
1172 }
1173
Mahesh Bandewar6a9e4612018-09-24 14:39:42 -07001174 /* Link-local multicast packets should be passed to the
1175 * stack on the link they arrive as well as pass them to the
1176 * bond-master device. These packets are mostly usable when
1177 * stack receives it with the link on which they arrive
1178 * (e.g. LLDP) they also must be available on master. Some of
1179 * the use cases include (but are not limited to): LLDP agents
1180 * that must be able to operate both on enslaved interfaces as
1181 * well as on bonds themselves; linux bridges that must be able
1182 * to process/pass BPDUs from attached bonds when any kind of
1183 * STP version is enabled on the network.
1184 */
1185 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) {
1186 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1187
1188 if (nskb) {
1189 nskb->dev = bond->dev;
Mahesh Bandewar0f3b9142018-10-02 12:14:34 -07001190 nskb->queue_mapping = 0;
Mahesh Bandewar6a9e4612018-09-24 14:39:42 -07001191 netif_rx(nskb);
1192 }
Chonggang Lib89f04c2017-04-16 12:02:18 -07001193 return RX_HANDLER_PASS;
Mahesh Bandewar6a9e4612018-09-24 14:39:42 -07001194 }
Zhang Shengjud66bd902016-02-03 02:02:32 +00001195 if (bond_should_deliver_exact_match(skb, slave, bond))
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001196 return RX_HANDLER_EXACT;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001197
Jiri Pirko35d48902011-03-22 02:38:12 +00001198 skb->dev = bond->dev;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001199
Veaceslav Falico01844092014-05-15 21:39:55 +02001200 if (BOND_MODE(bond) == BOND_MODE_ALB &&
Jiri Pirko35d48902011-03-22 02:38:12 +00001201 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001202 skb->pkt_type == PACKET_HOST) {
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001203
Changli Gao541ac7c2011-03-02 21:07:14 +00001204 if (unlikely(skb_cow_head(skb,
1205 skb->data - skb_mac_header(skb)))) {
1206 kfree_skb(skb);
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001207 return RX_HANDLER_CONSUMED;
Changli Gao541ac7c2011-03-02 21:07:14 +00001208 }
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04001209 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1210 bond->dev->addr_len);
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001211 }
1212
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00001213 return ret;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001214}
1215
Jiri Pirko41f0b042015-12-03 12:12:14 +01001216static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001217{
Jiri Pirko41f0b042015-12-03 12:12:14 +01001218 switch (BOND_MODE(bond)) {
1219 case BOND_MODE_ROUNDROBIN:
1220 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1221 case BOND_MODE_ACTIVEBACKUP:
1222 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1223 case BOND_MODE_BROADCAST:
1224 return NETDEV_LAG_TX_TYPE_BROADCAST;
1225 case BOND_MODE_XOR:
1226 case BOND_MODE_8023AD:
1227 return NETDEV_LAG_TX_TYPE_HASH;
1228 default:
1229 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1230 }
1231}
1232
John Hurleyf44aa9e2018-05-23 19:22:52 -07001233static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1234 enum netdev_lag_tx_type type)
1235{
1236 if (type != NETDEV_LAG_TX_TYPE_HASH)
1237 return NETDEV_LAG_HASH_NONE;
1238
1239 switch (bond->params.xmit_policy) {
1240 case BOND_XMIT_POLICY_LAYER2:
1241 return NETDEV_LAG_HASH_L2;
1242 case BOND_XMIT_POLICY_LAYER34:
1243 return NETDEV_LAG_HASH_L34;
1244 case BOND_XMIT_POLICY_LAYER23:
1245 return NETDEV_LAG_HASH_L23;
1246 case BOND_XMIT_POLICY_ENCAP23:
1247 return NETDEV_LAG_HASH_E23;
1248 case BOND_XMIT_POLICY_ENCAP34:
1249 return NETDEV_LAG_HASH_E34;
1250 default:
1251 return NETDEV_LAG_HASH_UNKNOWN;
1252 }
1253}
1254
David Ahern42ab19e2017-10-04 17:48:47 -07001255static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1256 struct netlink_ext_ack *extack)
Jiri Pirko41f0b042015-12-03 12:12:14 +01001257{
1258 struct netdev_lag_upper_info lag_upper_info;
John Hurleyf44aa9e2018-05-23 19:22:52 -07001259 enum netdev_lag_tx_type type;
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001260
John Hurleyf44aa9e2018-05-23 19:22:52 -07001261 type = bond_lag_tx_type(bond);
1262 lag_upper_info.tx_type = type;
1263 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
Xin Long4597efe2017-10-24 13:54:18 +08001264
1265 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1266 &lag_upper_info, extack);
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001267}
1268
Jiri Pirko41f0b042015-12-03 12:12:14 +01001269static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001270{
Jiri Pirko41f0b042015-12-03 12:12:14 +01001271 netdev_upper_dev_unlink(slave->dev, bond->dev);
1272 slave->dev->flags &= ~IFF_SLAVE;
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001273}
1274
dingtianhong3fdddd82014-05-12 15:08:43 +08001275static struct slave *bond_alloc_slave(struct bonding *bond)
1276{
1277 struct slave *slave = NULL;
1278
Zhang Shengjud66bd902016-02-03 02:02:32 +00001279 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
dingtianhong3fdddd82014-05-12 15:08:43 +08001280 if (!slave)
1281 return NULL;
1282
Veaceslav Falico01844092014-05-15 21:39:55 +02001283 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
dingtianhong3fdddd82014-05-12 15:08:43 +08001284 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1285 GFP_KERNEL);
1286 if (!SLAVE_AD_INFO(slave)) {
1287 kfree(slave);
1288 return NULL;
1289 }
1290 }
Mahesh Bandeward4859d72018-09-24 14:40:11 -07001291 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1292
dingtianhong3fdddd82014-05-12 15:08:43 +08001293 return slave;
1294}
1295
1296static void bond_free_slave(struct slave *slave)
1297{
1298 struct bonding *bond = bond_get_bond_by_slave(slave);
1299
Mahesh Bandeward4859d72018-09-24 14:40:11 -07001300 cancel_delayed_work_sync(&slave->notify_work);
Veaceslav Falico01844092014-05-15 21:39:55 +02001301 if (BOND_MODE(bond) == BOND_MODE_8023AD)
dingtianhong3fdddd82014-05-12 15:08:43 +08001302 kfree(SLAVE_AD_INFO(slave));
1303
1304 kfree(slave);
1305}
1306
Moni Shoua69a23382015-02-03 16:48:30 +02001307static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1308{
1309 info->bond_mode = BOND_MODE(bond);
1310 info->miimon = bond->params.miimon;
1311 info->num_slaves = bond->slave_cnt;
1312}
1313
1314static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1315{
1316 strcpy(info->slave_name, slave->dev->name);
1317 info->link = slave->link;
1318 info->state = bond_slave_state(slave);
1319 info->link_failure_count = slave->link_failure_count;
1320}
1321
Moni Shoua69e61132015-02-03 16:48:31 +02001322static void bond_netdev_notify_work(struct work_struct *_work)
1323{
Mahesh Bandeward4859d72018-09-24 14:40:11 -07001324 struct slave *slave = container_of(_work, struct slave,
1325 notify_work.work);
Moni Shoua69e61132015-02-03 16:48:31 +02001326
Mahesh Bandeward4859d72018-09-24 14:40:11 -07001327 if (rtnl_trylock()) {
1328 struct netdev_bonding_info binfo;
1329
1330 bond_fill_ifslave(slave, &binfo.slave);
1331 bond_fill_ifbond(slave->bond, &binfo.master);
1332 netdev_bonding_info_change(slave->dev, &binfo);
1333 rtnl_unlock();
1334 } else {
1335 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1336 }
Moni Shoua69e61132015-02-03 16:48:31 +02001337}
1338
1339void bond_queue_slave_event(struct slave *slave)
1340{
Mahesh Bandeward4859d72018-09-24 14:40:11 -07001341 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
Moni Shoua69e61132015-02-03 16:48:31 +02001342}
1343
Jiri Pirkof7c7eb72015-12-03 12:12:20 +01001344void bond_lower_state_changed(struct slave *slave)
1345{
1346 struct netdev_lag_lower_state_info info;
1347
1348 info.link_up = slave->link == BOND_LINK_UP ||
1349 slave->link == BOND_LINK_FAIL;
1350 info.tx_enabled = bond_is_active_slave(slave);
1351 netdev_lower_state_changed(slave->dev, &info);
1352}
1353
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354/* enslave device <slave> to bond device <master> */
David Ahern33eaf2a2017-10-04 17:48:46 -07001355int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1356 struct netlink_ext_ack *extack)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357{
Wang Chen454d7c92008-11-12 23:37:49 -08001358 struct bonding *bond = netdev_priv(bond_dev);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001359 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Veaceslav Falicoc8c23902013-09-25 09:20:25 +02001360 struct slave *new_slave = NULL, *prev_slave;
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04001361 struct sockaddr_storage ss;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 int link_reporting;
Veaceslav Falico8599b522013-06-24 11:49:34 +02001363 int res = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Ben Hutchingsc772dde2012-12-07 06:15:32 +00001365 if (!bond->params.use_carrier &&
1366 slave_dev->ethtool_ops->get_link == NULL &&
1367 slave_ops->ndo_do_ioctl == NULL) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001368 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1369 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
1371
Mahesh Bandewar24b27fc2016-09-01 22:18:34 -07001372 /* already in-use? */
1373 if (netdev_is_rx_handler_busy(slave_dev)) {
David Ahern759088b2017-10-04 17:48:49 -07001374 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
Mahesh Bandewar24b27fc2016-09-01 22:18:34 -07001375 netdev_err(bond_dev,
1376 "Error: Device is in use and cannot be enslaved\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 return -EBUSY;
1378 }
1379
Jiri Bohac09a89c22014-02-26 18:20:13 +01001380 if (bond_dev == slave_dev) {
David Ahern759088b2017-10-04 17:48:49 -07001381 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
Veaceslav Falico76444f52014-07-15 19:35:58 +02001382 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
Jiri Bohac09a89c22014-02-26 18:20:13 +01001383 return -EPERM;
1384 }
1385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 /* vlan challenged mutual exclusion */
1387 /* no need to lock since we're protected by rtnl_lock */
1388 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001389 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1390 slave_dev->name);
Jiri Pirko55462cf2012-10-14 04:30:56 +00001391 if (vlan_uses_dev(bond_dev)) {
David Ahern759088b2017-10-04 17:48:49 -07001392 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
Veaceslav Falico76444f52014-07-15 19:35:58 +02001393 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1394 slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 return -EPERM;
1396 } else {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001397 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1398 slave_dev->name, slave_dev->name,
1399 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 }
1401 } else {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001402 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1403 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 }
1405
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001406 /* Old ifenslave binaries are no longer supported. These can
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001407 * be identified with moderate accuracy by the state of the slave:
Jay Vosburgh217df672005-09-26 16:11:50 -07001408 * the current ifenslave will set the interface down prior to
1409 * enslaving it; the old ifenslave will not.
1410 */
yzhu1ce3ea1c2015-12-03 18:00:55 +08001411 if (slave_dev->flags & IFF_UP) {
David Ahern759088b2017-10-04 17:48:49 -07001412 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
Veaceslav Falico76444f52014-07-15 19:35:58 +02001413 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1414 slave_dev->name);
Zhang Shengju1e2a8862016-02-09 10:37:46 +00001415 return -EPERM;
Jay Vosburgh217df672005-09-26 16:11:50 -07001416 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Moni Shoua872254d2007-10-09 19:43:38 -07001418 /* set bonding device ether type by slave - bonding netdevices are
1419 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1420 * there is a need to override some of the type dependent attribs/funcs.
1421 *
1422 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1423 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1424 */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001425 if (!bond_has_slaves(bond)) {
Moni Shouae36b9d12009-07-15 04:56:31 +00001426 if (bond_dev->type != slave_dev->type) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001427 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1428 bond_dev->type, slave_dev->type);
Moni Shoua75c78502009-09-15 02:37:40 -07001429
Amerigo Wangb7bc2a52012-08-09 22:14:57 +00001430 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1431 bond_dev);
Jiri Pirko3ca5b402010-03-10 10:29:35 +00001432 res = notifier_to_errno(res);
1433 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001434 netdev_err(bond_dev, "refused to change device type\n");
Zhang Shengju1e2a8862016-02-09 10:37:46 +00001435 return -EBUSY;
Jiri Pirko3ca5b402010-03-10 10:29:35 +00001436 }
Moni Shoua75c78502009-09-15 02:37:40 -07001437
Jiri Pirko32a806c2010-03-19 04:00:23 +00001438 /* Flush unicast and multicast addresses */
Jiri Pirkoa748ee22010-04-01 21:22:09 +00001439 dev_uc_flush(bond_dev);
Jiri Pirko22bedad32010-04-01 21:22:57 +00001440 dev_mc_flush(bond_dev);
Jiri Pirko32a806c2010-03-19 04:00:23 +00001441
Moni Shouae36b9d12009-07-15 04:56:31 +00001442 if (slave_dev->type != ARPHRD_ETHER)
1443 bond_setup_by_slave(bond_dev, slave_dev);
Neil Horman550fd082011-07-26 06:05:38 +00001444 else {
Moni Shouae36b9d12009-07-15 04:56:31 +00001445 ether_setup(bond_dev);
Neil Horman550fd082011-07-26 06:05:38 +00001446 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1447 }
Moni Shoua75c78502009-09-15 02:37:40 -07001448
Amerigo Wangb7bc2a52012-08-09 22:14:57 +00001449 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1450 bond_dev);
Moni Shouae36b9d12009-07-15 04:56:31 +00001451 }
Moni Shoua872254d2007-10-09 19:43:38 -07001452 } else if (bond_dev->type != slave_dev->type) {
David Ahern759088b2017-10-04 17:48:49 -07001453 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
Veaceslav Falico76444f52014-07-15 19:35:58 +02001454 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1455 slave_dev->name, slave_dev->type, bond_dev->type);
Zhang Shengju1e2a8862016-02-09 10:37:46 +00001456 return -EINVAL;
Moni Shoua872254d2007-10-09 19:43:38 -07001457 }
1458
Mark Bloch1533e772016-07-21 11:52:55 +03001459 if (slave_dev->type == ARPHRD_INFINIBAND &&
1460 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
David Ahern759088b2017-10-04 17:48:49 -07001461 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
Mark Bloch1533e772016-07-21 11:52:55 +03001462 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1463 slave_dev->type);
1464 res = -EOPNOTSUPP;
1465 goto err_undo_flags;
1466 }
1467
1468 if (!slave_ops->ndo_set_mac_address ||
1469 slave_dev->type == ARPHRD_INFINIBAND) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001470 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
Veaceslav Falicof5442442014-07-15 13:26:01 +02001471 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1472 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1473 if (!bond_has_slaves(bond)) {
dingtianhong00503b62014-01-25 13:00:29 +08001474 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
Veaceslav Falico76444f52014-07-15 19:35:58 +02001475 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
Veaceslav Falicof5442442014-07-15 13:26:01 +02001476 } else {
David Ahern759088b2017-10-04 17:48:49 -07001477 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
Veaceslav Falico76444f52014-07-15 19:35:58 +02001478 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
Veaceslav Falicof5442442014-07-15 13:26:01 +02001479 res = -EOPNOTSUPP;
1480 goto err_undo_flags;
dingtianhong00503b62014-01-25 13:00:29 +08001481 }
Moni Shoua2ab82852007-10-09 19:43:39 -07001482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 }
1484
Amerigo Wang8d8fc292011-05-19 21:39:10 +00001485 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1486
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001487 /* If this is the first slave, then we need to set the master's hardware
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001488 * address to be the same as the slave's.
1489 */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001490 if (!bond_has_slaves(bond) &&
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001491 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
Jiri Pirko409cc1f2013-01-30 11:08:11 +01001492 bond_set_dev_addr(bond->dev, slave_dev);
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001493
dingtianhong3fdddd82014-05-12 15:08:43 +08001494 new_slave = bond_alloc_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 if (!new_slave) {
1496 res = -ENOMEM;
1497 goto err_undo_flags;
1498 }
dingtianhong3fdddd82014-05-12 15:08:43 +08001499
Veaceslav Falicodc73c412014-05-21 17:42:00 +02001500 new_slave->bond = bond;
1501 new_slave->dev = slave_dev;
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001502 /* Set the new_slave's queue_id to be zero. Queue ID mapping
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00001503 * is set via sysfs or module option if desired.
1504 */
1505 new_slave->queue_id = 0;
1506
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001507 /* Save slave's original mtu and then set it to match the bond */
1508 new_slave->original_mtu = slave_dev->mtu;
1509 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1510 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001511 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001512 goto err_free;
1513 }
1514
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001515 /* Save slave's original ("permanent") mac address for modes
Jay Vosburgh217df672005-09-26 16:11:50 -07001516 * that need it, and for restoring it upon release, and then
1517 * set it to the master's address
1518 */
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04001519 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1520 slave_dev->addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
dingtianhong00503b62014-01-25 13:00:29 +08001522 if (!bond->params.fail_over_mac ||
Veaceslav Falico01844092014-05-15 21:39:55 +02001523 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001524 /* Set slave to master's mac address. The application already
Moni Shoua2ab82852007-10-09 19:43:39 -07001525 * set the master's mac address to that of the first slave
1526 */
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04001527 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1528 ss.ss_family = slave_dev->type;
1529 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
Moni Shoua2ab82852007-10-09 19:43:39 -07001530 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001531 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001532 goto err_restore_mtu;
Moni Shoua2ab82852007-10-09 19:43:39 -07001533 }
Jay Vosburgh217df672005-09-26 16:11:50 -07001534 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
Karl Heiss03d84a52016-01-11 08:28:43 -05001536 /* set slave flag before open to prevent IPv6 addrconf */
1537 slave_dev->flags |= IFF_SLAVE;
1538
Jay Vosburgh217df672005-09-26 16:11:50 -07001539 /* open the slave since the application closed it */
1540 res = dev_open(slave_dev);
1541 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001542 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001543 goto err_restore_mac;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 }
1545
Jay Vosburgh0b680e72006-09-22 21:54:10 -07001546 slave_dev->priv_flags |= IFF_BONDING;
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04001547 /* initialize slave stats */
1548 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
Holger Eitzenberger58402052008-12-09 23:07:13 -08001550 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 /* bond_alb_init_slave() must be called before all other stages since
1552 * it might fail and we do not want to have to undo everything
1553 */
1554 res = bond_alb_init_slave(bond, new_slave);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001555 if (res)
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001556 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 }
1558
Wei Yongjunb8e2fde2013-08-23 10:45:07 +08001559 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1560 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001561 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1562 slave_dev->name);
Xin Longae42cc62018-03-26 01:16:46 +08001563 goto err_close;
nikolay@redhat.com1ff412a2013-08-06 12:40:15 +02001564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
Veaceslav Falicoc8c23902013-09-25 09:20:25 +02001566 prev_slave = bond_last_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
1568 new_slave->delay = 0;
1569 new_slave->link_failure_count = 0;
1570
Andreas Bornad729bc92017-08-10 06:41:44 +02001571 if (bond_update_speed_duplex(new_slave) &&
1572 bond_needs_speed_duplex(bond))
Mahesh Bandewar3f3c2782017-04-03 18:38:39 -07001573 new_slave->link = BOND_LINK_DOWN;
Veaceslav Falico876254a2013-03-12 06:31:32 +00001574
Veaceslav Falico49f17de2014-02-18 07:48:47 +01001575 new_slave->last_rx = jiffies -
Michal Kubečekf31c7932012-04-17 02:02:06 +00001576 (msecs_to_jiffies(bond->params.arp_interval) + 1);
Veaceslav Falico8599b522013-06-24 11:49:34 +02001577 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
Veaceslav Falico49f17de2014-02-18 07:48:47 +01001578 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07001579
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 if (bond->params.miimon && !bond->params.use_carrier) {
1581 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1582
1583 if ((link_reporting == -1) && !bond->params.arp_interval) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001584 /* miimon is set but a bonded network driver
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 * does not support ETHTOOL/MII and
1586 * arp_interval is not set. Note: if
1587 * use_carrier is enabled, we will never go
1588 * here (because netif_carrier is always
1589 * supported); thus, we don't need to change
1590 * the messages for netif_carrier.
1591 */
Veaceslav Falico76444f52014-07-15 19:35:58 +02001592 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1593 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 } else if (link_reporting == -1) {
1595 /* unable get link status using mii/ethtool */
Veaceslav Falico76444f52014-07-15 19:35:58 +02001596 netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1597 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 }
1599 }
1600
1601 /* check for initial state */
Aviv Hellera30b0162016-07-05 12:09:47 +03001602 new_slave->link = BOND_LINK_NOCHANGE;
Michal Kubečekf31c7932012-04-17 02:02:06 +00001603 if (bond->params.miimon) {
1604 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1605 if (bond->params.updelay) {
Moni Shoua69a23382015-02-03 16:48:30 +02001606 bond_set_slave_link_state(new_slave,
Jiri Pirko5d397062015-12-03 12:12:19 +01001607 BOND_LINK_BACK,
1608 BOND_SLAVE_NOTIFY_NOW);
Michal Kubečekf31c7932012-04-17 02:02:06 +00001609 new_slave->delay = bond->params.updelay;
1610 } else {
Moni Shoua69a23382015-02-03 16:48:30 +02001611 bond_set_slave_link_state(new_slave,
Jiri Pirko5d397062015-12-03 12:12:19 +01001612 BOND_LINK_UP,
1613 BOND_SLAVE_NOTIFY_NOW);
Michal Kubečekf31c7932012-04-17 02:02:06 +00001614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 } else {
Jiri Pirko5d397062015-12-03 12:12:19 +01001616 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1617 BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 }
Michal Kubečekf31c7932012-04-17 02:02:06 +00001619 } else if (bond->params.arp_interval) {
Moni Shoua69a23382015-02-03 16:48:30 +02001620 bond_set_slave_link_state(new_slave,
1621 (netif_carrier_ok(slave_dev) ?
Jiri Pirko5d397062015-12-03 12:12:19 +01001622 BOND_LINK_UP : BOND_LINK_DOWN),
1623 BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 } else {
Jiri Pirko5d397062015-12-03 12:12:19 +01001625 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1626 BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 }
1628
Michal Kubečekf31c7932012-04-17 02:02:06 +00001629 if (new_slave->link != BOND_LINK_DOWN)
Veaceslav Falico8e603462014-02-18 07:48:46 +01001630 new_slave->last_link_up = jiffies;
Veaceslav Falico76444f52014-07-15 19:35:58 +02001631 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1632 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1633 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
Michal Kubečekf31c7932012-04-17 02:02:06 +00001634
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02001635 if (bond_uses_primary(bond) && bond->params.primary[0]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 /* if there is a primary slave, remember it */
Jiri Pirkoa5499522009-09-25 03:28:09 +00001637 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02001638 rcu_assign_pointer(bond->primary_slave, new_slave);
Jiri Pirkoa5499522009-09-25 03:28:09 +00001639 bond->force_primary = true;
1640 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 }
1642
Veaceslav Falico01844092014-05-15 21:39:55 +02001643 switch (BOND_MODE(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 case BOND_MODE_ACTIVEBACKUP:
dingtianhong5e5b0662014-02-26 11:05:22 +08001645 bond_set_slave_inactive_flags(new_slave,
1646 BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 break;
1648 case BOND_MODE_8023AD:
1649 /* in 802.3ad mode, the internal mechanism
1650 * will activate the slaves in the selected
1651 * aggregator
1652 */
dingtianhong5e5b0662014-02-26 11:05:22 +08001653 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 /* if this is the first slave */
Veaceslav Falico23c147e2013-09-27 15:10:57 +02001655 if (!prev_slave) {
dingtianhong3fdddd82014-05-12 15:08:43 +08001656 SLAVE_AD_INFO(new_slave)->id = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 /* Initialize AD with the number of times that the AD timer is called in 1 second
1658 * can be called only after the mac address of the bond is set
1659 */
Peter Pan(潘卫平)56d00c672011-06-08 21:19:02 +00001660 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 } else {
dingtianhong3fdddd82014-05-12 15:08:43 +08001662 SLAVE_AD_INFO(new_slave)->id =
1663 SLAVE_AD_INFO(prev_slave)->id + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 }
1665
1666 bond_3ad_bind_slave(new_slave);
1667 break;
1668 case BOND_MODE_TLB:
1669 case BOND_MODE_ALB:
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001670 bond_set_active_slave(new_slave);
dingtianhong5e5b0662014-02-26 11:05:22 +08001671 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 break;
1673 default:
Veaceslav Falico76444f52014-07-15 19:35:58 +02001674 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676 /* always active in trunk mode */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001677 bond_set_active_slave(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
1679 /* In trunking mode there is little meaning to curr_active_slave
1680 * anyway (it holds no special properties of the bond device),
1681 * so we can change it without calling change_active_interface()
1682 */
Eric Dumazet4740d632014-07-15 06:56:55 -07001683 if (!rcu_access_pointer(bond->curr_active_slave) &&
1684 new_slave->link == BOND_LINK_UP)
nikolay@redhat.com278b2082013-08-01 16:54:51 +02001685 rcu_assign_pointer(bond->curr_active_slave, new_slave);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001686
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 break;
1688 } /* switch(bond_mode) */
1689
WANG Congf6dc31a2010-05-06 00:48:51 -07001690#ifdef CONFIG_NET_POLL_CONTROLLER
Xin Longddea7882018-04-22 19:11:50 +08001691 if (bond->dev->npinfo) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001692 if (slave_enable_netpoll(new_slave)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001693 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001694 res = -EBUSY;
stephen hemmingerf7d98212011-12-31 13:26:46 +00001695 goto err_detach;
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001696 }
WANG Congf6dc31a2010-05-06 00:48:51 -07001697 }
1698#endif
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001699
Michal Kubečekfbe168b2014-11-13 07:54:50 +01001700 if (!(bond_dev->features & NETIF_F_LRO))
1701 dev_disable_lro(slave_dev);
1702
Jiri Pirko35d48902011-03-22 02:38:12 +00001703 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1704 new_slave);
1705 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001706 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
Veaceslav Falico5831d662013-09-25 09:20:32 +02001707 goto err_detach;
Jiri Pirko35d48902011-03-22 02:38:12 +00001708 }
1709
David Ahern42ab19e2017-10-04 17:48:47 -07001710 res = bond_master_upper_dev_link(bond, new_slave, extack);
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001711 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001712 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001713 goto err_unregister;
1714 }
1715
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001716 res = bond_sysfs_slave_add(new_slave);
1717 if (res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02001718 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001719 goto err_upper_unlink;
1720 }
1721
Eric Dumazet7e2556e2018-07-31 06:30:54 -07001722 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1723
Xin Longae42cc62018-03-26 01:16:46 +08001724 /* If the mode uses primary, then the following is handled by
1725 * bond_change_active_slave().
1726 */
1727 if (!bond_uses_primary(bond)) {
1728 /* set promiscuity level to new slave */
1729 if (bond_dev->flags & IFF_PROMISC) {
1730 res = dev_set_promiscuity(slave_dev, 1);
1731 if (res)
1732 goto err_sysfs_del;
1733 }
1734
1735 /* set allmulti level to new slave */
1736 if (bond_dev->flags & IFF_ALLMULTI) {
1737 res = dev_set_allmulti(slave_dev, 1);
Xin Long9f5a90c2018-03-26 01:16:47 +08001738 if (res) {
1739 if (bond_dev->flags & IFF_PROMISC)
1740 dev_set_promiscuity(slave_dev, -1);
Xin Longae42cc62018-03-26 01:16:46 +08001741 goto err_sysfs_del;
Xin Long9f5a90c2018-03-26 01:16:47 +08001742 }
Xin Longae42cc62018-03-26 01:16:46 +08001743 }
1744
1745 netif_addr_lock_bh(bond_dev);
1746 dev_mc_sync_multiple(slave_dev, bond_dev);
1747 dev_uc_sync_multiple(slave_dev, bond_dev);
1748 netif_addr_unlock_bh(bond_dev);
1749
1750 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1751 /* add lacpdu mc addr to mc list */
1752 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1753
1754 dev_mc_add(slave_dev, lacpdu_multicast);
1755 }
1756 }
1757
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001758 bond->slave_cnt++;
1759 bond_compute_features(bond);
1760 bond_set_carrier(bond);
1761
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02001762 if (bond_uses_primary(bond)) {
dingtianhongf80889a2014-02-12 12:06:40 +08001763 block_netpoll_tx();
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001764 bond_select_active_slave(bond);
dingtianhongf80889a2014-02-12 12:06:40 +08001765 unblock_netpoll_tx();
Veaceslav Falico5378c2e2013-10-21 11:48:30 +02001766 }
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001767
Debabrata Banerjeee79c1052018-05-14 14:48:09 -04001768 if (bond_mode_can_use_xmit_hash(bond))
Mahesh Bandewaree637712014-10-04 17:45:01 -07001769 bond_update_slave_arr(bond, NULL);
1770
Debabrata Banerjee21706ee2018-05-09 19:32:11 -04001771
Veaceslav Falico76444f52014-07-15 19:35:58 +02001772 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1773 slave_dev->name,
1774 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1775 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776
1777 /* enslave is successful */
Moni Shoua69e61132015-02-03 16:48:31 +02001778 bond_queue_slave_event(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 return 0;
1780
1781/* Undo stages on error */
Xin Longae42cc62018-03-26 01:16:46 +08001782err_sysfs_del:
1783 bond_sysfs_slave_del(new_slave);
1784
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001785err_upper_unlink:
Jiri Pirko41f0b042015-12-03 12:12:14 +01001786 bond_upper_dev_unlink(bond, new_slave);
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001787
Veaceslav Falico1f718f02013-09-25 09:20:10 +02001788err_unregister:
1789 netdev_rx_handler_unregister(slave_dev);
1790
stephen hemmingerf7d98212011-12-31 13:26:46 +00001791err_detach:
nikolay@redhat.com1ff412a2013-08-06 12:40:15 +02001792 vlan_vids_del_by_dev(slave_dev, bond_dev);
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02001793 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1794 RCU_INIT_POINTER(bond->primary_slave, NULL);
Eric Dumazet4740d632014-07-15 06:56:55 -07001795 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
dingtianhongf80889a2014-02-12 12:06:40 +08001796 block_netpoll_tx();
dingtianhongc8517032013-12-13 10:20:07 +08001797 bond_change_active_slave(bond, NULL);
nikolay@redhat.com3c5913b532013-04-18 07:33:36 +00001798 bond_select_active_slave(bond);
dingtianhongf80889a2014-02-12 12:06:40 +08001799 unblock_netpoll_tx();
nikolay@redhat.com3c5913b532013-04-18 07:33:36 +00001800 }
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02001801 /* either primary_slave or curr_active_slave might've changed */
1802 synchronize_rcu();
nikolay@redhat.comfc7a72a2013-04-18 07:33:37 +00001803 slave_disable_netpoll(new_slave);
stephen hemmingerf7d98212011-12-31 13:26:46 +00001804
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805err_close:
nikolay@redhat.comb6a5a7b2013-04-11 09:18:56 +00001806 slave_dev->priv_flags &= ~IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 dev_close(slave_dev);
1808
1809err_restore_mac:
Karl Heiss03d84a52016-01-11 08:28:43 -05001810 slave_dev->flags &= ~IFF_SLAVE;
dingtianhong00503b62014-01-25 13:00:29 +08001811 if (!bond->params.fail_over_mac ||
Veaceslav Falico01844092014-05-15 21:39:55 +02001812 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001813 /* XXX TODO - fom follow mode needs to change master's
1814 * MAC if this slave's MAC is in use by the bond, or at
1815 * least print a warning.
1816 */
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04001817 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1818 new_slave->dev->addr_len);
1819 ss.ss_family = slave_dev->type;
1820 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
Moni Shoua2ab82852007-10-09 19:43:39 -07001821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001823err_restore_mtu:
1824 dev_set_mtu(slave_dev, new_slave->original_mtu);
1825
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826err_free:
dingtianhong3fdddd82014-05-12 15:08:43 +08001827 bond_free_slave(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
1829err_undo_flags:
Nikolay Aleksandrovb8fad452013-06-12 00:07:01 +02001830 /* Enslave of first slave has failed and we need to fix master's mac */
Nikolay Aleksandrov7d5cd2c2015-07-15 22:57:01 +02001831 if (!bond_has_slaves(bond)) {
1832 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1833 slave_dev->dev_addr))
1834 eth_hw_addr_random(bond_dev);
1835 if (bond_dev->type != ARPHRD_ETHER) {
Jay Vosburgh40baec22015-11-06 17:23:23 -08001836 dev_close(bond_dev);
Nikolay Aleksandrov7d5cd2c2015-07-15 22:57:01 +02001837 ether_setup(bond_dev);
1838 bond_dev->flags |= IFF_MASTER;
1839 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1840 }
1841 }
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 return res;
1844}
1845
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001846/* Try to release the slave device <slave> from the bond device <master>
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 * It is legal to access curr_active_slave without a lock because all the function
Nikolay Aleksandrov8c0bc552014-09-11 22:49:28 +02001848 * is RTNL-locked. If "all" is true it means that the function is being called
nikolay@redhat.com08963412013-02-18 14:09:42 +00001849 * while destroying a bond interface and all slaves are being released.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 *
1851 * The rules for slave state should be:
1852 * for Active/Backup:
1853 * Active stays on all backups go down
1854 * for Bonded connections:
1855 * The first up interface should be left on and all others downed.
1856 */
nikolay@redhat.com08963412013-02-18 14:09:42 +00001857static int __bond_release_one(struct net_device *bond_dev,
1858 struct net_device *slave_dev,
WANG Congf51048c2017-07-06 15:01:57 -07001859 bool all, bool unregister)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860{
Wang Chen454d7c92008-11-12 23:37:49 -08001861 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 struct slave *slave, *oldcurrent;
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04001863 struct sockaddr_storage ss;
Neil Horman5a0068d2013-09-27 12:22:15 -04001864 int old_flags = bond_dev->flags;
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001865 netdev_features_t old_features = bond_dev->features;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
1867 /* slave is not a slave or master is not master of this slave */
1868 if (!(slave_dev->flags & IFF_SLAVE) ||
Jiri Pirko471cb5a2013-01-03 22:49:01 +00001869 !netdev_has_upper_dev(slave_dev, bond_dev)) {
Wengang Wanga22a9e42014-12-23 09:24:36 +08001870 netdev_dbg(bond_dev, "cannot release %s\n",
Veaceslav Falico76444f52014-07-15 19:35:58 +02001871 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 return -EINVAL;
1873 }
1874
Neil Hormane843fa52010-10-13 16:01:50 +00001875 block_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
1877 slave = bond_get_slave_by_dev(bond, slave_dev);
1878 if (!slave) {
1879 /* not a slave of this bond */
Veaceslav Falico76444f52014-07-15 19:35:58 +02001880 netdev_info(bond_dev, "%s not enslaved\n",
1881 slave_dev->name);
Neil Hormane843fa52010-10-13 16:01:50 +00001882 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 return -EINVAL;
1884 }
1885
Jiri Pirko57beaca2015-12-03 12:12:21 +01001886 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1887
sfeldma@cumulusnetworks.com07699f92014-01-16 22:57:49 -08001888 bond_sysfs_slave_del(slave);
1889
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04001890 /* recompute stats just before removing the slave */
1891 bond_get_stats(bond->dev, &bond->bond_stats);
1892
Jiri Pirko41f0b042015-12-03 12:12:14 +01001893 bond_upper_dev_unlink(bond, slave);
Jiri Pirko35d48902011-03-22 02:38:12 +00001894 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1895 * for this slave anymore.
1896 */
1897 netdev_rx_handler_unregister(slave_dev);
Jiri Pirko35d48902011-03-22 02:38:12 +00001898
Nikolay Aleksandrove4702592014-09-11 22:49:27 +02001899 if (BOND_MODE(bond) == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 bond_3ad_unbind_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901
Debabrata Banerjeee79c1052018-05-14 14:48:09 -04001902 if (bond_mode_can_use_xmit_hash(bond))
Mahesh Bandewaree637712014-10-04 17:45:01 -07001903 bond_update_slave_arr(bond, slave);
1904
Veaceslav Falico76444f52014-07-15 19:35:58 +02001905 netdev_info(bond_dev, "Releasing %s interface %s\n",
1906 bond_is_active_slave(slave) ? "active" : "backup",
1907 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Eric Dumazet4740d632014-07-15 06:56:55 -07001909 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
Eric Dumazet85741712014-07-15 06:56:56 -07001911 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
dingtianhong00503b62014-01-25 13:00:29 +08001913 if (!all && (!bond->params.fail_over_mac ||
Veaceslav Falico01844092014-05-15 21:39:55 +02001914 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
dingtianhong844223a2014-01-02 09:13:16 +08001915 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001916 bond_has_slaves(bond))
Veaceslav Falico76444f52014-07-15 19:35:58 +02001917 netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1918 slave_dev->name, slave->perm_hwaddr,
1919 bond_dev->name, slave_dev->name);
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001920 }
1921
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02001922 if (rtnl_dereference(bond->primary_slave) == slave)
1923 RCU_INIT_POINTER(bond->primary_slave, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
Nikolay Aleksandrov1c72cfdc92014-09-11 22:49:24 +02001925 if (oldcurrent == slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 bond_change_active_slave(bond, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
Holger Eitzenberger58402052008-12-09 23:07:13 -08001928 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 /* Must be called only after the slave has been
1930 * detached from the list and the curr_active_slave
1931 * has been cleared (if our_slave == old_current),
1932 * but before a new active slave is selected.
1933 */
1934 bond_alb_deinit_slave(bond, slave);
1935 }
1936
nikolay@redhat.com08963412013-02-18 14:09:42 +00001937 if (all) {
Paul E. McKenney36708b82013-12-09 15:19:53 -08001938 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
nikolay@redhat.com08963412013-02-18 14:09:42 +00001939 } else if (oldcurrent == slave) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001940 /* Note that we hold RTNL over this sequence, so there
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001941 * is no concern that another slave add/remove event
1942 * will interfere.
1943 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 bond_select_active_slave(bond);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001945 }
1946
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001947 if (!bond_has_slaves(bond)) {
Jay Vosburghff59c452006-03-27 13:27:43 -08001948 bond_set_carrier(bond);
Jiri Pirko409cc1f2013-01-30 11:08:11 +01001949 eth_hw_addr_random(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 }
1951
Neil Hormane843fa52010-10-13 16:01:50 +00001952 unblock_netpoll_tx();
nikolay@redhat.com278b2082013-08-01 16:54:51 +02001953 synchronize_rcu();
Nikolay Aleksandrovee6154e2014-02-26 14:20:30 +01001954 bond->slave_cnt--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001956 if (!bond_has_slaves(bond)) {
Shlomo Pongratz2af73d42012-04-03 22:56:19 +00001957 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
Veaceslav Falico80028ea2013-03-06 07:10:32 +00001958 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1959 }
Shlomo Pongratz2af73d42012-04-03 22:56:19 +00001960
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001961 bond_compute_features(bond);
1962 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1963 (old_features & NETIF_F_VLAN_CHALLENGED))
Veaceslav Falico76444f52014-07-15 19:35:58 +02001964 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1965 slave_dev->name, bond_dev->name);
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001966
nikolay@redhat.com1ff412a2013-08-06 12:40:15 +02001967 vlan_vids_del_by_dev(slave_dev, bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02001969 /* If the mode uses primary, then this case was handled above by
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001970 * bond_change_active_slave(..., NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 */
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02001972 if (!bond_uses_primary(bond)) {
Neil Horman5a0068d2013-09-27 12:22:15 -04001973 /* unset promiscuity level from slave
1974 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1975 * of the IFF_PROMISC flag in the bond_dev, but we need the
1976 * value of that flag before that change, as that was the value
1977 * when this slave was attached, so we cache at the start of the
1978 * function and use it here. Same goes for ALLMULTI below
1979 */
1980 if (old_flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 dev_set_promiscuity(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
1983 /* unset allmulti level from slave */
Neil Horman5a0068d2013-09-27 12:22:15 -04001984 if (old_flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 dev_set_allmulti(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00001987 bond_hw_addr_flush(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 }
1989
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001990 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -07001991
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 /* close slave before restoring its mac address */
1993 dev_close(slave_dev);
1994
dingtianhong00503b62014-01-25 13:00:29 +08001995 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
Veaceslav Falico01844092014-05-15 21:39:55 +02001996 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
Moni Shoua2ab82852007-10-09 19:43:39 -07001997 /* restore original ("permanent") mac address */
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04001998 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
1999 slave->dev->addr_len);
2000 ss.ss_family = slave_dev->type;
2001 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
Moni Shoua2ab82852007-10-09 19:43:39 -07002002 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003
WANG Congf51048c2017-07-06 15:01:57 -07002004 if (unregister)
2005 __dev_set_mtu(slave_dev, slave->original_mtu);
2006 else
2007 dev_set_mtu(slave_dev, slave->original_mtu);
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00002008
Jiri Pirko2d7011c2011-03-16 08:46:43 +00002009 slave_dev->priv_flags &= ~IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
dingtianhong3fdddd82014-05-12 15:08:43 +08002011 bond_free_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002013 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014}
2015
nikolay@redhat.com08963412013-02-18 14:09:42 +00002016/* A wrapper used because of ndo_del_link */
2017int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2018{
WANG Congf51048c2017-07-06 15:01:57 -07002019 return __bond_release_one(bond_dev, slave_dev, false, false);
nikolay@redhat.com08963412013-02-18 14:09:42 +00002020}
2021
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002022/* First release a slave and then destroy the bond if no more slaves are left.
2023 * Must be under rtnl_lock when this function is called.
2024 */
stephen hemminger26d8ee72010-10-15 05:09:34 +00002025static int bond_release_and_destroy(struct net_device *bond_dev,
2026 struct net_device *slave_dev)
Moni Shouad90a1622007-10-09 19:43:43 -07002027{
Wang Chen454d7c92008-11-12 23:37:49 -08002028 struct bonding *bond = netdev_priv(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07002029 int ret;
2030
WANG Congf51048c2017-07-06 15:01:57 -07002031 ret = __bond_release_one(bond_dev, slave_dev, false, true);
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02002032 if (ret == 0 && !bond_has_slaves(bond)) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +00002033 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
Veaceslav Falico76444f52014-07-15 19:35:58 +02002034 netdev_info(bond_dev, "Destroying bond %s\n",
2035 bond_dev->name);
Nikolay Aleksandrov06f6d102015-07-15 21:52:51 +02002036 bond_remove_proc_entry(bond);
Stephen Hemminger9e716262009-06-12 19:02:47 +00002037 unregister_netdevice(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07002038 }
2039 return ret;
2040}
2041
Zhu Yanjun3d675762017-02-02 23:46:21 -05002042static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043{
Wang Chen454d7c92008-11-12 23:37:49 -08002044 struct bonding *bond = netdev_priv(bond_dev);
Moni Shoua69a23382015-02-03 16:48:30 +02002045 bond_fill_ifbond(bond, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046}
2047
2048static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2049{
Wang Chen454d7c92008-11-12 23:37:49 -08002050 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002051 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002052 int i = 0, res = -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002055 bond_for_each_slave(bond, slave, iter) {
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002056 if (i++ == (int)info->slave_id) {
Eric Dumazet689c96c2009-04-23 03:39:04 +00002057 res = 0;
Moni Shoua69a23382015-02-03 16:48:30 +02002058 bond_fill_ifslave(slave, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 break;
2060 }
2061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
Eric Dumazet689c96c2009-04-23 03:39:04 +00002063 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064}
2065
2066/*-------------------------------- Monitoring -------------------------------*/
2067
Eric Dumazet4740d632014-07-15 06:56:55 -07002068/* called with rcu_read_lock() */
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002069static int bond_miimon_inspect(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070{
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002071 int link_state, commit = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002072 struct list_head *iter;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002073 struct slave *slave;
Jiri Pirko41f89102009-04-24 03:57:29 +00002074 bool ignore_updelay;
2075
Eric Dumazet4740d632014-07-15 06:56:55 -07002076 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
dingtianhong4cb4f972013-12-13 10:19:39 +08002078 bond_for_each_slave_rcu(bond, slave, iter) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002079 slave->new_link = BOND_LINK_NOCHANGE;
Jay Vosburgh055db692017-11-07 19:50:07 +09002080 slave->link_new_state = slave->link;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002082 link_state = bond_check_dev_link(bond, slave->dev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
2084 switch (slave->link) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002085 case BOND_LINK_UP:
2086 if (link_state)
2087 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
Mahesh Bandewarde77ecd2017-03-27 11:37:33 -07002089 bond_propose_link_state(slave, BOND_LINK_FAIL);
WANG Congd94708a2017-07-25 09:44:25 -07002090 commit++;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002091 slave->delay = bond->params.downdelay;
2092 if (slave->delay) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002093 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2094 (BOND_MODE(bond) ==
2095 BOND_MODE_ACTIVEBACKUP) ?
2096 (bond_is_active_slave(slave) ?
2097 "active " : "backup ") : "",
2098 slave->dev->name,
2099 bond->params.downdelay * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002101 /*FALLTHRU*/
2102 case BOND_LINK_FAIL:
2103 if (link_state) {
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002104 /* recovered before downdelay expired */
Mahesh Bandewarde77ecd2017-03-27 11:37:33 -07002105 bond_propose_link_state(slave, BOND_LINK_UP);
Veaceslav Falico8e603462014-02-18 07:48:46 +01002106 slave->last_link_up = jiffies;
Veaceslav Falico76444f52014-07-15 19:35:58 +02002107 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2108 (bond->params.downdelay - slave->delay) *
2109 bond->params.miimon,
2110 slave->dev->name);
Mahesh Bandewarfb9eb892017-04-11 22:36:00 -07002111 commit++;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002112 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002114
2115 if (slave->delay <= 0) {
2116 slave->new_link = BOND_LINK_DOWN;
2117 commit++;
2118 continue;
2119 }
2120
2121 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002124 case BOND_LINK_DOWN:
2125 if (!link_state)
2126 continue;
2127
Mahesh Bandewarde77ecd2017-03-27 11:37:33 -07002128 bond_propose_link_state(slave, BOND_LINK_BACK);
WANG Congd94708a2017-07-25 09:44:25 -07002129 commit++;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002130 slave->delay = bond->params.updelay;
2131
2132 if (slave->delay) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002133 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2134 slave->dev->name,
2135 ignore_updelay ? 0 :
2136 bond->params.updelay *
2137 bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002139 /*FALLTHRU*/
2140 case BOND_LINK_BACK:
2141 if (!link_state) {
Mahesh Bandewarde77ecd2017-03-27 11:37:33 -07002142 bond_propose_link_state(slave, BOND_LINK_DOWN);
Veaceslav Falico76444f52014-07-15 19:35:58 +02002143 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2144 (bond->params.updelay - slave->delay) *
2145 bond->params.miimon,
2146 slave->dev->name);
Mahesh Bandewarfb9eb892017-04-11 22:36:00 -07002147 commit++;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002148 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002150
Jiri Pirko41f89102009-04-24 03:57:29 +00002151 if (ignore_updelay)
2152 slave->delay = 0;
2153
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002154 if (slave->delay <= 0) {
2155 slave->new_link = BOND_LINK_UP;
2156 commit++;
Jiri Pirko41f89102009-04-24 03:57:29 +00002157 ignore_updelay = false;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002158 continue;
2159 }
2160
2161 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 break;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002163 }
2164 }
2165
2166 return commit;
2167}
2168
Tonghao Zhang7e878b62018-05-16 19:09:23 -07002169static void bond_miimon_link_change(struct bonding *bond,
2170 struct slave *slave,
2171 char link)
2172{
2173 switch (BOND_MODE(bond)) {
2174 case BOND_MODE_8023AD:
2175 bond_3ad_handle_link_change(slave, link);
2176 break;
2177 case BOND_MODE_TLB:
2178 case BOND_MODE_ALB:
2179 bond_alb_handle_link_change(bond, slave, link);
2180 break;
2181 case BOND_MODE_XOR:
2182 bond_update_slave_arr(bond, NULL);
2183 break;
2184 }
2185}
2186
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002187static void bond_miimon_commit(struct bonding *bond)
2188{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002189 struct list_head *iter;
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002190 struct slave *slave, *primary;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002191
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002192 bond_for_each_slave(bond, slave, iter) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002193 switch (slave->new_link) {
2194 case BOND_LINK_NOCHANGE:
2195 continue;
2196
2197 case BOND_LINK_UP:
Andreas Bornad729bc92017-08-10 06:41:44 +02002198 if (bond_update_speed_duplex(slave) &&
2199 bond_needs_speed_duplex(bond)) {
Mahesh Bandewar3f3c2782017-04-03 18:38:39 -07002200 slave->link = BOND_LINK_DOWN;
Andreas Born11e9d782017-08-12 00:36:55 +02002201 if (net_ratelimit())
2202 netdev_warn(bond->dev,
2203 "failed to get link speed/duplex for %s\n",
2204 slave->dev->name);
Mahesh Bandewarb5bf0f52017-03-27 11:37:37 -07002205 continue;
2206 }
Jiri Pirko5d397062015-12-03 12:12:19 +01002207 bond_set_slave_link_state(slave, BOND_LINK_UP,
2208 BOND_SLAVE_NOTIFY_NOW);
Veaceslav Falico8e603462014-02-18 07:48:46 +01002209 slave->last_link_up = jiffies;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002210
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002211 primary = rtnl_dereference(bond->primary_slave);
Veaceslav Falico01844092014-05-15 21:39:55 +02002212 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002213 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002214 bond_set_backup_slave(slave);
Veaceslav Falico01844092014-05-15 21:39:55 +02002215 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002216 /* make it immediately active */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002217 bond_set_active_slave(slave);
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002218 } else if (slave != primary) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002219 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002220 bond_set_backup_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002221 }
2222
Veaceslav Falico76444f52014-07-15 19:35:58 +02002223 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2224 slave->dev->name,
2225 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2226 slave->duplex ? "full" : "half");
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002227
Tonghao Zhang7e878b62018-05-16 19:09:23 -07002228 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
Mahesh Bandewaree637712014-10-04 17:45:01 -07002229
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002230 if (!bond->curr_active_slave || slave == primary)
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002231 goto do_failover;
2232
2233 continue;
2234
2235 case BOND_LINK_DOWN:
Jay Vosburghfba4acd2008-10-30 17:41:14 -07002236 if (slave->link_failure_count < UINT_MAX)
2237 slave->link_failure_count++;
2238
Jiri Pirko5d397062015-12-03 12:12:19 +01002239 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2240 BOND_SLAVE_NOTIFY_NOW);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002241
Veaceslav Falico01844092014-05-15 21:39:55 +02002242 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2243 BOND_MODE(bond) == BOND_MODE_8023AD)
dingtianhong5e5b0662014-02-26 11:05:22 +08002244 bond_set_slave_inactive_flags(slave,
2245 BOND_SLAVE_NOTIFY_NOW);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002246
Veaceslav Falico76444f52014-07-15 19:35:58 +02002247 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2248 slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002249
Tonghao Zhang7e878b62018-05-16 19:09:23 -07002250 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
Mahesh Bandewaree637712014-10-04 17:45:01 -07002251
Eric Dumazet4740d632014-07-15 06:56:55 -07002252 if (slave == rcu_access_pointer(bond->curr_active_slave))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002253 goto do_failover;
2254
2255 continue;
2256
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 default:
Veaceslav Falico76444f52014-07-15 19:35:58 +02002258 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2259 slave->new_link, slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002260 slave->new_link = BOND_LINK_NOCHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002262 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 }
2264
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002265do_failover:
Neil Hormane843fa52010-10-13 16:01:50 +00002266 block_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 bond_select_active_slave(bond);
Neil Hormane843fa52010-10-13 16:01:50 +00002268 unblock_netpoll_tx();
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002269 }
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002270
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002271 bond_set_carrier(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272}
2273
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002274/* bond_mii_monitor
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002275 *
2276 * Really a wrapper that splits the mii monitor into two phases: an
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002277 * inspection, then (if inspection indicates something needs to be done)
2278 * an acquisition of appropriate locks followed by a commit phase to
2279 * implement whatever link state changes are indicated.
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002280 */
stephen hemminger6da67d22013-12-30 10:43:41 -08002281static void bond_mii_monitor(struct work_struct *work)
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002282{
2283 struct bonding *bond = container_of(work, struct bonding,
2284 mii_work.work);
Ben Hutchingsad246c92011-04-26 15:25:52 +00002285 bool should_notify_peers = false;
David S. Miller1f2cd842013-10-28 00:11:22 -04002286 unsigned long delay;
Mahesh Bandewarde77ecd2017-03-27 11:37:33 -07002287 struct slave *slave;
2288 struct list_head *iter;
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002289
David S. Miller1f2cd842013-10-28 00:11:22 -04002290 delay = msecs_to_jiffies(bond->params.miimon);
2291
2292 if (!bond_has_slaves(bond))
dingtianhong6b6c5262013-10-24 11:09:03 +08002293 goto re_arm;
dingtianhong6b6c5262013-10-24 11:09:03 +08002294
dingtianhong4cb4f972013-12-13 10:19:39 +08002295 rcu_read_lock();
2296
Ben Hutchingsad246c92011-04-26 15:25:52 +00002297 should_notify_peers = bond_should_notify_peers(bond);
2298
David S. Miller1f2cd842013-10-28 00:11:22 -04002299 if (bond_miimon_inspect(bond)) {
dingtianhong4cb4f972013-12-13 10:19:39 +08002300 rcu_read_unlock();
David S. Miller1f2cd842013-10-28 00:11:22 -04002301
2302 /* Race avoidance with bond_close cancel of workqueue */
2303 if (!rtnl_trylock()) {
David S. Miller1f2cd842013-10-28 00:11:22 -04002304 delay = 1;
2305 should_notify_peers = false;
2306 goto re_arm;
2307 }
2308
Mahesh Bandewarde77ecd2017-03-27 11:37:33 -07002309 bond_for_each_slave(bond, slave, iter) {
2310 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2311 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002312 bond_miimon_commit(bond);
2313
David S. Miller1f2cd842013-10-28 00:11:22 -04002314 rtnl_unlock(); /* might sleep, hold no other locks */
dingtianhong4cb4f972013-12-13 10:19:39 +08002315 } else
2316 rcu_read_unlock();
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002317
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002318re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00002319 if (bond->params.miimon)
David S. Miller1f2cd842013-10-28 00:11:22 -04002320 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2321
David S. Miller1f2cd842013-10-28 00:11:22 -04002322 if (should_notify_peers) {
2323 if (!rtnl_trylock())
2324 return;
2325 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2326 rtnl_unlock();
2327 }
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002328}
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002329
David Ahernb3208b22016-10-17 19:15:45 -07002330static int bond_upper_dev_walk(struct net_device *upper, void *data)
2331{
2332 __be32 ip = *((__be32 *)data);
2333
2334 return ip == bond_confirm_addr(upper, 0, ip);
2335}
2336
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002337static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002338{
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002339 bool ret = false;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002340
Andy Gospodarekeaddcd72012-03-22 16:14:29 +00002341 if (ip == bond_confirm_addr(bond->dev, 0, ip))
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002342 return true;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002343
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002344 rcu_read_lock();
David Ahernb3208b22016-10-17 19:15:45 -07002345 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2346 ret = true;
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002347 rcu_read_unlock();
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002348
Veaceslav Falico50223ce2013-08-28 23:25:11 +02002349 return ret;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002350}
2351
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002352/* We go to the (large) trouble of VLAN tagging ARP frames because
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002353 * switches in VLAN mode (especially if ports are configured as
2354 * "native" to a VLAN) might not pass non-tagged frames.
2355 */
dingtianhongfbd929f2014-03-25 17:44:43 +08002356static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2357 __be32 dest_ip, __be32 src_ip,
Vlad Yasevich44a40852014-05-16 17:20:38 -04002358 struct bond_vlan_tag *tags)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002359{
2360 struct sk_buff *skb;
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002361 struct bond_vlan_tag *outer_tag = tags;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002362
Veaceslav Falico76444f52014-07-15 19:35:58 +02002363 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2364 arp_op, slave_dev->name, &dest_ip, &src_ip);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002365
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002366 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2367 NULL, slave_dev->dev_addr, NULL);
2368
2369 if (!skb) {
dingtianhong4873ac32014-03-25 17:44:44 +08002370 net_err_ratelimited("ARP packet allocation failed\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002371 return;
2372 }
dingtianhongfbd929f2014-03-25 17:44:43 +08002373
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002374 if (!tags || tags->vlan_proto == VLAN_N_VID)
2375 goto xmit;
2376
2377 tags++;
2378
Vlad Yasevich44a40852014-05-16 17:20:38 -04002379 /* Go through all the tags backwards and add them to the packet */
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002380 while (tags->vlan_proto != VLAN_N_VID) {
2381 if (!tags->vlan_id) {
2382 tags++;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002383 continue;
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002384 }
Vlad Yasevich44a40852014-05-16 17:20:38 -04002385
Veaceslav Falico76444f52014-07-15 19:35:58 +02002386 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002387 ntohs(outer_tag->vlan_proto), tags->vlan_id);
Jiri Pirko62749e22014-11-19 14:04:58 +01002388 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2389 tags->vlan_id);
Vlad Yasevich44a40852014-05-16 17:20:38 -04002390 if (!skb) {
2391 net_err_ratelimited("failed to insert inner VLAN tag\n");
2392 return;
2393 }
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002394
2395 tags++;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002396 }
2397 /* Set the outer tag */
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002398 if (outer_tag->vlan_id) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002399 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002400 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
Jiri Pirkob4bef1b2014-11-19 14:04:57 +01002401 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2402 outer_tag->vlan_id);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002403 }
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002404
2405xmit:
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002406 arp_xmit(skb);
2407}
2408
Vlad Yasevich44a40852014-05-16 17:20:38 -04002409/* Validate the device path between the @start_dev and the @end_dev.
2410 * The path is valid if the @end_dev is reachable through device
2411 * stacking.
2412 * When the path is validated, collect any vlan information in the
2413 * path.
2414 */
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002415struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2416 struct net_device *end_dev,
2417 int level)
Vlad Yasevich44a40852014-05-16 17:20:38 -04002418{
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002419 struct bond_vlan_tag *tags;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002420 struct net_device *upper;
2421 struct list_head *iter;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002422
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002423 if (start_dev == end_dev) {
Kees Cook6396bb22018-06-12 14:03:40 -07002424 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002425 if (!tags)
2426 return ERR_PTR(-ENOMEM);
2427 tags[level].vlan_proto = VLAN_N_VID;
2428 return tags;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002429 }
2430
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002431 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2432 tags = bond_verify_device_path(upper, end_dev, level + 1);
2433 if (IS_ERR_OR_NULL(tags)) {
2434 if (IS_ERR(tags))
2435 return tags;
2436 continue;
2437 }
2438 if (is_vlan_dev(upper)) {
2439 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2440 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2441 }
2442
2443 return tags;
2444 }
2445
2446 return NULL;
Vlad Yasevich44a40852014-05-16 17:20:38 -04002447}
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002448
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2450{
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002451 struct rtable *rt;
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002452 struct bond_vlan_tag *tags;
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002453 __be32 *targets = bond->params.arp_targets, addr;
dingtianhongfbd929f2014-03-25 17:44:43 +08002454 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002456 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002457 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002458 tags = NULL;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002459
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002460 /* Find out through which dev should the packet go */
David S. Miller78fbfd82011-03-12 00:00:52 -05002461 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2462 RTO_ONLINK, 0);
David S. Millerb23dd4f2011-03-02 14:31:35 -08002463 if (IS_ERR(rt)) {
Veaceslav Falico28572762014-02-28 12:39:19 +01002464 /* there's no route to target - try to send arp
2465 * probe to generate any traffic (arp_validate=0)
2466 */
dingtianhong4873ac32014-03-25 17:44:44 +08002467 if (bond->params.arp_validate)
2468 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2469 bond->dev->name,
2470 &targets[i]);
Vlad Yasevich44a40852014-05-16 17:20:38 -04002471 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2472 0, tags);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002473 continue;
2474 }
2475
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002476 /* bond device itself */
2477 if (rt->dst.dev == bond->dev)
2478 goto found;
2479
2480 rcu_read_lock();
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002481 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002482 rcu_read_unlock();
2483
Veaceslav Falico3e403a72014-07-17 17:02:23 +02002484 if (!IS_ERR_OR_NULL(tags))
Vlad Yasevich44a40852014-05-16 17:20:38 -04002485 goto found;
2486
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002487 /* Not our device - skip */
Veaceslav Falico76444f52014-07-15 19:35:58 +02002488 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2489 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
Veaceslav Falico3e325822013-08-28 23:25:16 +02002490
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002491 ip_rt_put(rt);
Veaceslav Falico27bc11e2013-08-28 23:25:10 +02002492 continue;
2493
2494found:
2495 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2496 ip_rt_put(rt);
2497 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
Vlad Yasevich44a40852014-05-16 17:20:38 -04002498 addr, tags);
Dan Carpentera67eed52014-07-25 15:21:21 +03002499 kfree(tags);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002500 }
2501}
2502
Al Virod3bb52b2007-08-22 20:06:58 -04002503static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002504{
Veaceslav Falico8599b522013-06-24 11:49:34 +02002505 int i;
2506
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002507 if (!sip || !bond_has_this_ip(bond, tip)) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002508 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2509 &sip, &tip);
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002510 return;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002511 }
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002512
Veaceslav Falico8599b522013-06-24 11:49:34 +02002513 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2514 if (i == -1) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002515 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2516 &sip);
Veaceslav Falico87a7b842013-06-24 11:49:29 +02002517 return;
2518 }
Veaceslav Falico49f17de2014-02-18 07:48:47 +01002519 slave->last_rx = jiffies;
Veaceslav Falico8599b522013-06-24 11:49:34 +02002520 slave->target_last_arp_rx[i] = jiffies;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002521}
2522
nikolay@redhat.com5bb9e0b2013-09-07 00:00:26 +02002523int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2524 struct slave *slave)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002525{
Eric Dumazetde063b72012-06-11 19:23:07 +00002526 struct arphdr *arp = (struct arphdr *)skb->data;
Jay Vosburgh21a75f02016-02-02 13:35:56 -08002527 struct slave *curr_active_slave, *curr_arp_slave;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002528 unsigned char *arp_ptr;
Al Virod3bb52b2007-08-22 20:06:58 -04002529 __be32 sip, tip;
Alexey Dobriyan6ade97d2017-09-26 23:12:28 +03002530 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2531 unsigned int alen;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002532
Veaceslav Falicof2cb6912014-02-18 07:48:42 +01002533 if (!slave_do_arp_validate(bond, slave)) {
dingtianhongbedabf92014-05-07 22:10:20 +08002534 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2535 !slave_do_arp_validate_only(bond))
Veaceslav Falico49f17de2014-02-18 07:48:47 +01002536 slave->last_rx = jiffies;
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00002537 return RX_HANDLER_ANOTHER;
Veaceslav Falicof2cb6912014-02-18 07:48:42 +01002538 } else if (!is_arp) {
2539 return RX_HANDLER_ANOTHER;
2540 }
Veaceslav Falico2c146102013-06-24 11:49:31 +02002541
Eric Dumazetde063b72012-06-11 19:23:07 +00002542 alen = arp_hdr_len(bond->dev);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002543
Veaceslav Falico76444f52014-07-15 19:35:58 +02002544 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2545 skb->dev->name);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002546
Eric Dumazetde063b72012-06-11 19:23:07 +00002547 if (alen > skb_headlen(skb)) {
2548 arp = kmalloc(alen, GFP_ATOMIC);
2549 if (!arp)
2550 goto out_unlock;
2551 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2552 goto out_unlock;
2553 }
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002554
Jiri Pirko3aba8912011-04-19 03:48:16 +00002555 if (arp->ar_hln != bond->dev->addr_len ||
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002556 skb->pkt_type == PACKET_OTHERHOST ||
2557 skb->pkt_type == PACKET_LOOPBACK ||
2558 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2559 arp->ar_pro != htons(ETH_P_IP) ||
2560 arp->ar_pln != 4)
2561 goto out_unlock;
2562
2563 arp_ptr = (unsigned char *)(arp + 1);
Jiri Pirko3aba8912011-04-19 03:48:16 +00002564 arp_ptr += bond->dev->addr_len;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002565 memcpy(&sip, arp_ptr, 4);
Jiri Pirko3aba8912011-04-19 03:48:16 +00002566 arp_ptr += 4 + bond->dev->addr_len;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002567 memcpy(&tip, arp_ptr, 4);
2568
Veaceslav Falico76444f52014-07-15 19:35:58 +02002569 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2570 slave->dev->name, bond_slave_state(slave),
2571 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2572 &sip, &tip);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002573
Veaceslav Falico010d3c32014-02-20 12:07:57 +01002574 curr_active_slave = rcu_dereference(bond->curr_active_slave);
Jay Vosburgh21a75f02016-02-02 13:35:56 -08002575 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
Veaceslav Falico010d3c32014-02-20 12:07:57 +01002576
Jay Vosburgh21a75f02016-02-02 13:35:56 -08002577 /* We 'trust' the received ARP enough to validate it if:
Veaceslav Falicoaeea64a2013-06-24 11:49:32 +02002578 *
Jay Vosburgh21a75f02016-02-02 13:35:56 -08002579 * (a) the slave receiving the ARP is active (which includes the
2580 * current ARP slave, if any), or
2581 *
2582 * (b) the receiving slave isn't active, but there is a currently
2583 * active slave and it received valid arp reply(s) after it became
2584 * the currently active slave, or
2585 *
2586 * (c) there is an ARP slave that sent an ARP during the prior ARP
2587 * interval, and we receive an ARP reply on any slave. We accept
2588 * these because switch FDB update delays may deliver the ARP
2589 * reply to a slave other than the sender of the ARP request.
2590 *
2591 * Note: for (b), backup slaves are receiving the broadcast ARP
2592 * request, not a reply. This request passes from the sending
2593 * slave through the L2 switch(es) to the receiving slave. Since
2594 * this is checking the request, sip/tip are swapped for
2595 * validation.
2596 *
2597 * This is done to avoid endless looping when we can't reach the
Veaceslav Falicoaeea64a2013-06-24 11:49:32 +02002598 * arp_ip_target and fool ourselves with our own arp requests.
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002599 */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002600 if (bond_is_active_slave(slave))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002601 bond_validate_arp(bond, slave, sip, tip);
Veaceslav Falico010d3c32014-02-20 12:07:57 +01002602 else if (curr_active_slave &&
2603 time_after(slave_last_rx(bond, curr_active_slave),
2604 curr_active_slave->last_link_up))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002605 bond_validate_arp(bond, slave, tip, sip);
Jay Vosburgh21a75f02016-02-02 13:35:56 -08002606 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2607 bond_time_in_interval(bond,
2608 dev_trans_start(curr_arp_slave->dev), 1))
2609 bond_validate_arp(bond, slave, sip, tip);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002610
2611out_unlock:
Eric Dumazetde063b72012-06-11 19:23:07 +00002612 if (arp != (struct arphdr *)skb->data)
2613 kfree(arp);
Jiri Bohac13a8e0c2012-05-09 01:01:40 +00002614 return RX_HANDLER_ANOTHER;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002615}
2616
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002617/* function to verify if we're in the arp_interval timeslice, returns true if
2618 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2619 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2620 */
2621static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2622 int mod)
2623{
2624 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2625
2626 return time_in_range(jiffies,
2627 last_act - delta_in_ticks,
2628 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2629}
2630
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002631/* This function is called regularly to monitor each slave's link
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 * ensuring that traffic is being sent and received when arp monitoring
2633 * is used in load-balancing mode. if the adapter has been dormant, then an
2634 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2635 * arp monitoring in active backup mode.
2636 */
Mahesh Bandeward5e73f72017-03-08 10:55:51 -08002637static void bond_loadbalance_arp_mon(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 struct slave *slave, *oldcurrent;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002640 struct list_head *iter;
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002641 int do_failover = 0, slave_state_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642
David S. Miller1f2cd842013-10-28 00:11:22 -04002643 if (!bond_has_slaves(bond))
dingtianhong7f1bb572013-10-24 11:09:17 +08002644 goto re_arm;
dingtianhong7f1bb572013-10-24 11:09:17 +08002645
dingtianhong2e52f4f2013-12-13 10:19:50 +08002646 rcu_read_lock();
2647
Eric Dumazet4740d632014-07-15 06:56:55 -07002648 oldcurrent = rcu_dereference(bond->curr_active_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 /* see if any of the previous devices are up now (i.e. they have
2650 * xmt and rcv traffic). the curr_active_slave does not come into
Veaceslav Falico8e603462014-02-18 07:48:46 +01002651 * the picture unless it is null. also, slave->last_link_up is not
2652 * needed here because we send an arp on each slave and give a slave
2653 * as long as it needs to get the tx/rx within the delta.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 * TODO: what about up/down delay in arp mode? it wasn't here before
2655 * so it can wait
2656 */
dingtianhong2e52f4f2013-12-13 10:19:50 +08002657 bond_for_each_slave_rcu(bond, slave, iter) {
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002658 unsigned long trans_start = dev_trans_start(slave->dev);
2659
Nithin Sujir797a9362017-05-24 19:45:17 -07002660 slave->new_link = BOND_LINK_NOCHANGE;
2661
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 if (slave->link != BOND_LINK_UP) {
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002663 if (bond_time_in_interval(bond, trans_start, 1) &&
Veaceslav Falico49f17de2014-02-18 07:48:47 +01002664 bond_time_in_interval(bond, slave->last_rx, 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665
Nithin Sujir797a9362017-05-24 19:45:17 -07002666 slave->new_link = BOND_LINK_UP;
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002667 slave_state_changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668
2669 /* primary_slave has no meaning in round-robin
2670 * mode. the window of a slave being up and
2671 * curr_active_slave being null after enslaving
2672 * is closed.
2673 */
2674 if (!oldcurrent) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002675 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2676 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 do_failover = 1;
2678 } else {
Veaceslav Falico76444f52014-07-15 19:35:58 +02002679 netdev_info(bond->dev, "interface %s is now up\n",
2680 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 }
2682 }
2683 } else {
2684 /* slave->link == BOND_LINK_UP */
2685
2686 /* not all switches will respond to an arp request
2687 * when the source ip is 0, so don't take the link down
2688 * if we don't know our ip yet
2689 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002690 if (!bond_time_in_interval(bond, trans_start, 2) ||
Veaceslav Falico49f17de2014-02-18 07:48:47 +01002691 !bond_time_in_interval(bond, slave->last_rx, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
Nithin Sujir797a9362017-05-24 19:45:17 -07002693 slave->new_link = BOND_LINK_DOWN;
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002694 slave_state_changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002696 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698
Veaceslav Falico76444f52014-07-15 19:35:58 +02002699 netdev_info(bond->dev, "interface %s is now down\n",
2700 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002702 if (slave == oldcurrent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 do_failover = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 }
2705 }
2706
2707 /* note: if switch is in round-robin mode, all links
2708 * must tx arp to ensure all links rx an arp - otherwise
2709 * links may oscillate or not come up at all; if switch is
2710 * in something like xor mode, there is nothing we can
2711 * do - all replies will be rx'ed on same link causing slaves
2712 * to be unstable during low/no traffic periods
2713 */
Veaceslav Falicob6adc612014-05-15 21:39:57 +02002714 if (bond_slave_is_up(slave))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 bond_arp_send_all(bond, slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 }
2717
dingtianhong2e52f4f2013-12-13 10:19:50 +08002718 rcu_read_unlock();
2719
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002720 if (do_failover || slave_state_changed) {
dingtianhong2e52f4f2013-12-13 10:19:50 +08002721 if (!rtnl_trylock())
2722 goto re_arm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723
Nithin Sujir797a9362017-05-24 19:45:17 -07002724 bond_for_each_slave(bond, slave, iter) {
2725 if (slave->new_link != BOND_LINK_NOCHANGE)
2726 slave->link = slave->new_link;
2727 }
2728
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002729 if (slave_state_changed) {
2730 bond_slave_state_change(bond);
Mahesh Bandewaree637712014-10-04 17:45:01 -07002731 if (BOND_MODE(bond) == BOND_MODE_XOR)
2732 bond_update_slave_arr(bond, NULL);
Nikolay Aleksandrovb8e45002014-11-18 15:14:44 +01002733 }
2734 if (do_failover) {
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002735 block_netpoll_tx();
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002736 bond_select_active_slave(bond);
Ding Tianhong6fde8f02014-01-28 11:48:53 +08002737 unblock_netpoll_tx();
2738 }
dingtianhong2e52f4f2013-12-13 10:19:50 +08002739 rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 }
2741
2742re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00002743 if (bond->params.arp_interval)
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002744 queue_delayed_work(bond->wq, &bond->arp_work,
2745 msecs_to_jiffies(bond->params.arp_interval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746}
2747
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002748/* Called to inspect slaves for active-backup mode ARP monitor link state
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002749 * changes. Sets new_link in slaves to specify what action should take
2750 * place for the slave. Returns 0 if no changes are found, >0 if changes
2751 * to link states must be committed.
2752 *
Nikolay Aleksandrov8c0bc552014-09-11 22:49:28 +02002753 * Called with rcu_read_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002755static int bond_ab_arp_inspect(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756{
Veaceslav Falicodef44602013-08-03 03:50:35 +02002757 unsigned long trans_start, last_rx;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002758 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002759 struct slave *slave;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002760 int commit = 0;
Jiri Bohacda210f52012-08-30 12:02:47 +00002761
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002762 bond_for_each_slave_rcu(bond, slave, iter) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002763 slave->new_link = BOND_LINK_NOCHANGE;
Veaceslav Falicodef44602013-08-03 03:50:35 +02002764 last_rx = slave_last_rx(bond, slave);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002765
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 if (slave->link != BOND_LINK_UP) {
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002767 if (bond_time_in_interval(bond, last_rx, 1)) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002768 slave->new_link = BOND_LINK_UP;
2769 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 }
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002771 continue;
2772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002774 /* Give slaves 2*delta after being enslaved or made
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002775 * active. This avoids bouncing, as the last receive
2776 * times need a full ARP monitor cycle to be updated.
2777 */
Veaceslav Falico8e603462014-02-18 07:48:46 +01002778 if (bond_time_in_interval(bond, slave->last_link_up, 2))
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002779 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002781 /* Backup slave is down if:
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002782 * - No current_arp_slave AND
2783 * - more than 3*delta since last receive AND
2784 * - the bond has an IP address
2785 *
2786 * Note: a non-null current_arp_slave indicates
2787 * the curr_active_slave went down and we are
2788 * searching for a new one; under this condition
2789 * we only take the curr_active_slave down - this
2790 * gives each slave a chance to tx/rx traffic
2791 * before being taken out
2792 */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002793 if (!bond_is_active_slave(slave) &&
Eric Dumazet85741712014-07-15 06:56:56 -07002794 !rcu_access_pointer(bond->current_arp_slave) &&
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002795 !bond_time_in_interval(bond, last_rx, 3)) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002796 slave->new_link = BOND_LINK_DOWN;
2797 commit++;
2798 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002800 /* Active slave is down if:
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002801 * - more than 2*delta since transmitting OR
2802 * - (more than 2*delta since receive AND
2803 * the bond has an IP address)
2804 */
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002805 trans_start = dev_trans_start(slave->dev);
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002806 if (bond_is_active_slave(slave) &&
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002807 (!bond_time_in_interval(bond, trans_start, 2) ||
2808 !bond_time_in_interval(bond, last_rx, 2))) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002809 slave->new_link = BOND_LINK_DOWN;
2810 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 }
2812 }
2813
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002814 return commit;
2815}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002817/* Called to commit link state changes noted by inspection step of
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002818 * active-backup mode ARP monitor.
2819 *
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002820 * Called with RTNL hold.
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002821 */
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002822static void bond_ab_arp_commit(struct bonding *bond)
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002823{
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002824 unsigned long trans_start;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002825 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02002826 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02002828 bond_for_each_slave(bond, slave, iter) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002829 switch (slave->new_link) {
2830 case BOND_LINK_NOCHANGE:
2831 continue;
2832
2833 case BOND_LINK_UP:
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002834 trans_start = dev_trans_start(slave->dev);
Eric Dumazet4740d632014-07-15 06:56:55 -07002835 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2836 (!rtnl_dereference(bond->curr_active_slave) &&
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02002837 bond_time_in_interval(bond, trans_start, 1))) {
Eric Dumazet85741712014-07-15 06:56:56 -07002838 struct slave *current_arp_slave;
2839
2840 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
Jiri Pirko5d397062015-12-03 12:12:19 +01002841 bond_set_slave_link_state(slave, BOND_LINK_UP,
2842 BOND_SLAVE_NOTIFY_NOW);
Eric Dumazet85741712014-07-15 06:56:56 -07002843 if (current_arp_slave) {
Veaceslav Falico5a4309742012-04-05 03:47:43 +00002844 bond_set_slave_inactive_flags(
Eric Dumazet85741712014-07-15 06:56:56 -07002845 current_arp_slave,
dingtianhong5e5b0662014-02-26 11:05:22 +08002846 BOND_SLAVE_NOTIFY_NOW);
Eric Dumazet85741712014-07-15 06:56:56 -07002847 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
Veaceslav Falico5a4309742012-04-05 03:47:43 +00002848 }
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002849
Veaceslav Falico76444f52014-07-15 19:35:58 +02002850 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2851 slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002852
Eric Dumazet4740d632014-07-15 06:56:55 -07002853 if (!rtnl_dereference(bond->curr_active_slave) ||
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02002854 slave == rtnl_dereference(bond->primary_slave))
Jiri Pirkob9f60252009-08-31 11:09:38 +00002855 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002856
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002857 }
2858
Jiri Pirkob9f60252009-08-31 11:09:38 +00002859 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002860
2861 case BOND_LINK_DOWN:
2862 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864
Jiri Pirko5d397062015-12-03 12:12:19 +01002865 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2866 BOND_SLAVE_NOTIFY_NOW);
dingtianhong5e5b0662014-02-26 11:05:22 +08002867 bond_set_slave_inactive_flags(slave,
2868 BOND_SLAVE_NOTIFY_NOW);
Jiri Pirkob9f60252009-08-31 11:09:38 +00002869
Veaceslav Falico76444f52014-07-15 19:35:58 +02002870 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2871 slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002872
Eric Dumazet4740d632014-07-15 06:56:55 -07002873 if (slave == rtnl_dereference(bond->curr_active_slave)) {
Eric Dumazet85741712014-07-15 06:56:56 -07002874 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
Jiri Pirkob9f60252009-08-31 11:09:38 +00002875 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002876 }
Jiri Pirkob9f60252009-08-31 11:09:38 +00002877
2878 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002879
2880 default:
Veaceslav Falico76444f52014-07-15 19:35:58 +02002881 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2882 slave->new_link, slave->dev->name);
Jiri Pirkob9f60252009-08-31 11:09:38 +00002883 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885
Jiri Pirkob9f60252009-08-31 11:09:38 +00002886do_failover:
Neil Hormane843fa52010-10-13 16:01:50 +00002887 block_netpoll_tx();
Jiri Pirkob9f60252009-08-31 11:09:38 +00002888 bond_select_active_slave(bond);
Neil Hormane843fa52010-10-13 16:01:50 +00002889 unblock_netpoll_tx();
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002890 }
2891
2892 bond_set_carrier(bond);
2893}
2894
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02002895/* Send ARP probes for active-backup mode ARP monitor.
dingtianhongb0929912014-02-26 11:05:23 +08002896 *
Nikolay Aleksandrov8c0bc552014-09-11 22:49:28 +02002897 * Called with rcu_read_lock held.
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002898 */
Veaceslav Falicof2ebd472014-01-27 14:37:32 +01002899static bool bond_ab_arp_probe(struct bonding *bond)
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002900{
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002901 struct slave *slave, *before = NULL, *new_slave = NULL,
dingtianhongb0929912014-02-26 11:05:23 +08002902 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2903 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
Veaceslav Falico4087df872013-09-25 09:20:19 +02002904 struct list_head *iter;
2905 bool found = false;
dingtianhongb0929912014-02-26 11:05:23 +08002906 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
Veaceslav Falicof2ebd472014-01-27 14:37:32 +01002907
Veaceslav Falico98b90f22014-01-27 14:37:31 +01002908 if (curr_arp_slave && curr_active_slave)
Veaceslav Falico76444f52014-07-15 19:35:58 +02002909 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2910 curr_arp_slave->dev->name,
2911 curr_active_slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002912
Veaceslav Falico98b90f22014-01-27 14:37:31 +01002913 if (curr_active_slave) {
2914 bond_arp_send_all(bond, curr_active_slave);
dingtianhongb0929912014-02-26 11:05:23 +08002915 return should_notify_rtnl;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002916 }
2917
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 /* if we don't have a curr_active_slave, search for the next available
2919 * backup slave from the current_arp_slave and make it the candidate
2920 * for becoming the curr_active_slave
2921 */
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002922
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002923 if (!curr_arp_slave) {
dingtianhongb0929912014-02-26 11:05:23 +08002924 curr_arp_slave = bond_first_slave_rcu(bond);
2925 if (!curr_arp_slave)
2926 return should_notify_rtnl;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002927 }
2928
dingtianhongb0929912014-02-26 11:05:23 +08002929 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002930
dingtianhongb0929912014-02-26 11:05:23 +08002931 bond_for_each_slave_rcu(bond, slave, iter) {
Veaceslav Falicob6adc612014-05-15 21:39:57 +02002932 if (!found && !before && bond_slave_is_up(slave))
Veaceslav Falico4087df872013-09-25 09:20:19 +02002933 before = slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934
Veaceslav Falicob6adc612014-05-15 21:39:57 +02002935 if (found && !new_slave && bond_slave_is_up(slave))
Veaceslav Falico4087df872013-09-25 09:20:19 +02002936 new_slave = slave;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002937 /* if the link state is up at this point, we
2938 * mark it down - this can happen if we have
2939 * simultaneous link failures and
2940 * reselect_active_interface doesn't make this
2941 * one the current slave so it is still marked
2942 * up when it is actually down
2943 */
Veaceslav Falicob6adc612014-05-15 21:39:57 +02002944 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
Jiri Pirko5d397062015-12-03 12:12:19 +01002945 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2946 BOND_SLAVE_NOTIFY_LATER);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002947 if (slave->link_failure_count < UINT_MAX)
2948 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949
dingtianhong5e5b0662014-02-26 11:05:22 +08002950 bond_set_slave_inactive_flags(slave,
dingtianhongb0929912014-02-26 11:05:23 +08002951 BOND_SLAVE_NOTIFY_LATER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952
Veaceslav Falico76444f52014-07-15 19:35:58 +02002953 netdev_info(bond->dev, "backup interface %s is now down\n",
2954 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 }
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002956 if (slave == curr_arp_slave)
Veaceslav Falico4087df872013-09-25 09:20:19 +02002957 found = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 }
Veaceslav Falico4087df872013-09-25 09:20:19 +02002959
2960 if (!new_slave && before)
2961 new_slave = before;
2962
dingtianhongb0929912014-02-26 11:05:23 +08002963 if (!new_slave)
2964 goto check_state;
Veaceslav Falico4087df872013-09-25 09:20:19 +02002965
Jiri Pirko5d397062015-12-03 12:12:19 +01002966 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2967 BOND_SLAVE_NOTIFY_LATER);
dingtianhongb0929912014-02-26 11:05:23 +08002968 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
Veaceslav Falico4087df872013-09-25 09:20:19 +02002969 bond_arp_send_all(bond, new_slave);
Veaceslav Falico8e603462014-02-18 07:48:46 +01002970 new_slave->last_link_up = jiffies;
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002971 rcu_assign_pointer(bond->current_arp_slave, new_slave);
Veaceslav Falicof2ebd472014-01-27 14:37:32 +01002972
dingtianhongb0929912014-02-26 11:05:23 +08002973check_state:
2974 bond_for_each_slave_rcu(bond, slave, iter) {
Jiri Pirko5d397062015-12-03 12:12:19 +01002975 if (slave->should_notify || slave->should_notify_link) {
dingtianhongb0929912014-02-26 11:05:23 +08002976 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2977 break;
2978 }
2979 }
2980 return should_notify_rtnl;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002981}
2982
Mahesh Bandeward5e73f72017-03-08 10:55:51 -08002983static void bond_activebackup_arp_mon(struct bonding *bond)
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002984{
dingtianhongb0929912014-02-26 11:05:23 +08002985 bool should_notify_peers = false;
2986 bool should_notify_rtnl = false;
David S. Miller1f2cd842013-10-28 00:11:22 -04002987 int delta_in_ticks;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002988
David S. Miller1f2cd842013-10-28 00:11:22 -04002989 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2990
2991 if (!bond_has_slaves(bond))
dingtianhong80b9d232013-10-24 11:09:25 +08002992 goto re_arm;
dingtianhong80b9d232013-10-24 11:09:25 +08002993
dingtianhongeb9fa4b2013-12-13 10:20:02 +08002994 rcu_read_lock();
Ben Hutchingsad246c92011-04-26 15:25:52 +00002995
dingtianhongb0929912014-02-26 11:05:23 +08002996 should_notify_peers = bond_should_notify_peers(bond);
2997
2998 if (bond_ab_arp_inspect(bond)) {
2999 rcu_read_unlock();
3000
David S. Miller1f2cd842013-10-28 00:11:22 -04003001 /* Race avoidance with bond_close flush of workqueue */
3002 if (!rtnl_trylock()) {
David S. Miller1f2cd842013-10-28 00:11:22 -04003003 delta_in_ticks = 1;
3004 should_notify_peers = false;
3005 goto re_arm;
3006 }
3007
Veaceslav Falicoe7f63f12013-08-03 03:50:36 +02003008 bond_ab_arp_commit(bond);
dingtianhongb0929912014-02-26 11:05:23 +08003009
David S. Miller1f2cd842013-10-28 00:11:22 -04003010 rtnl_unlock();
dingtianhongb0929912014-02-26 11:05:23 +08003011 rcu_read_lock();
David S. Miller1f2cd842013-10-28 00:11:22 -04003012 }
3013
dingtianhongb0929912014-02-26 11:05:23 +08003014 should_notify_rtnl = bond_ab_arp_probe(bond);
3015 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016
3017re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00003018 if (bond->params.arp_interval)
David S. Miller1f2cd842013-10-28 00:11:22 -04003019 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3020
dingtianhongb0929912014-02-26 11:05:23 +08003021 if (should_notify_peers || should_notify_rtnl) {
David S. Miller1f2cd842013-10-28 00:11:22 -04003022 if (!rtnl_trylock())
3023 return;
dingtianhongb0929912014-02-26 11:05:23 +08003024
3025 if (should_notify_peers)
3026 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3027 bond->dev);
Jiri Pirko5d397062015-12-03 12:12:19 +01003028 if (should_notify_rtnl) {
dingtianhongb0929912014-02-26 11:05:23 +08003029 bond_slave_state_notify(bond);
Jiri Pirko5d397062015-12-03 12:12:19 +01003030 bond_slave_link_notify(bond);
3031 }
dingtianhongb0929912014-02-26 11:05:23 +08003032
David S. Miller1f2cd842013-10-28 00:11:22 -04003033 rtnl_unlock();
3034 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035}
3036
Mahesh Bandeward5e73f72017-03-08 10:55:51 -08003037static void bond_arp_monitor(struct work_struct *work)
3038{
3039 struct bonding *bond = container_of(work, struct bonding,
3040 arp_work.work);
3041
3042 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3043 bond_activebackup_arp_mon(bond);
3044 else
3045 bond_loadbalance_arp_mon(bond);
3046}
3047
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048/*-------------------------- netdev event handling --------------------------*/
3049
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003050/* Change device name */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051static int bond_event_changename(struct bonding *bond)
3052{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053 bond_remove_proc_entry(bond);
3054 bond_create_proc_entry(bond);
Stephen Hemminger7e083842009-06-12 19:02:46 +00003055
Taku Izumif073c7c2010-12-09 15:17:13 +00003056 bond_debug_reregister(bond);
3057
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 return NOTIFY_DONE;
3059}
3060
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003061static int bond_master_netdev_event(unsigned long event,
3062 struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063{
Wang Chen454d7c92008-11-12 23:37:49 -08003064 struct bonding *event_bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065
3066 switch (event) {
3067 case NETDEV_CHANGENAME:
3068 return bond_event_changename(event_bond);
Eric W. Biedermana64d49c2012-07-09 10:51:45 +00003069 case NETDEV_UNREGISTER:
3070 bond_remove_proc_entry(event_bond);
3071 break;
3072 case NETDEV_REGISTER:
3073 bond_create_proc_entry(event_bond);
3074 break;
nikolay@redhat.com6c8d23f2013-09-02 13:51:38 +02003075 case NETDEV_NOTIFY_PEERS:
3076 if (event_bond->send_peer_notif)
3077 event_bond->send_peer_notif--;
3078 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 default:
3080 break;
3081 }
3082
3083 return NOTIFY_DONE;
3084}
3085
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003086static int bond_slave_netdev_event(unsigned long event,
3087 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088{
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02003089 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
nikolay@redhat.com61013912013-04-11 09:18:55 +00003090 struct bonding *bond;
3091 struct net_device *bond_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092
nikolay@redhat.com61013912013-04-11 09:18:55 +00003093 /* A netdev event can be generated while enslaving a device
3094 * before netdev_rx_handler_register is called in which case
3095 * slave will be NULL
3096 */
3097 if (!slave)
3098 return NOTIFY_DONE;
3099 bond_dev = slave->bond->dev;
3100 bond = slave->bond;
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02003101 primary = rtnl_dereference(bond->primary_slave);
nikolay@redhat.com61013912013-04-11 09:18:55 +00003102
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 switch (event) {
3104 case NETDEV_UNREGISTER:
nikolay@redhat.com8d2ada72013-06-26 17:13:37 +02003105 if (bond_dev->type != ARPHRD_ETHER)
Jiri Pirko471cb5a2013-01-03 22:49:01 +00003106 bond_release_and_destroy(bond_dev, slave_dev);
3107 else
WANG Congf51048c2017-07-06 15:01:57 -07003108 __bond_release_one(bond_dev, slave_dev, false, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 break;
Weiping Pan98f41f62011-10-31 17:20:48 +00003110 case NETDEV_UP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 case NETDEV_CHANGE:
Mahesh Bandewar4d2c0cd2017-09-27 18:03:49 -07003112 /* For 802.3ad mode only:
3113 * Getting invalid Speed/Duplex values here will put slave
3114 * in weird state. So mark it as link-down for the time
3115 * being and let link-monitoring (miimon) set it right when
3116 * correct speeds/duplex are available.
3117 */
3118 if (bond_update_speed_duplex(slave) &&
3119 BOND_MODE(bond) == BOND_MODE_8023AD)
3120 slave->link = BOND_LINK_DOWN;
3121
Mahesh Bandewar52bc6712015-10-31 12:45:11 -07003122 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3123 bond_3ad_adapter_speed_duplex_changed(slave);
Mahesh Bandewar950ddcb2015-02-19 10:13:25 -08003124 /* Fallthrough */
3125 case NETDEV_DOWN:
Mahesh Bandewaree637712014-10-04 17:45:01 -07003126 /* Refresh slave-array if applicable!
3127 * If the setup does not use miimon or arpmon (mode-specific!),
3128 * then these events will not cause the slave-array to be
3129 * refreshed. This will cause xmit to use a slave that is not
3130 * usable. Avoid such situation by refeshing the array at these
3131 * events. If these (miimon/arpmon) parameters are configured
3132 * then array gets refreshed twice and that should be fine!
3133 */
Debabrata Banerjeee79c1052018-05-14 14:48:09 -04003134 if (bond_mode_can_use_xmit_hash(bond))
Mahesh Bandewaree637712014-10-04 17:45:01 -07003135 bond_update_slave_arr(bond, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137 case NETDEV_CHANGEMTU:
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003138 /* TODO: Should slaves be allowed to
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 * independently alter their MTU? For
3140 * an active-backup bond, slaves need
3141 * not be the same type of device, so
3142 * MTUs may vary. For other modes,
3143 * slaves arguably should have the
3144 * same MTUs. To do this, we'd need to
3145 * take over the slave's change_mtu
3146 * function for the duration of their
3147 * servitude.
3148 */
3149 break;
3150 case NETDEV_CHANGENAME:
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01003151 /* we don't care if we don't have primary set */
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02003152 if (!bond_uses_primary(bond) ||
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01003153 !bond->params.primary[0])
3154 break;
3155
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02003156 if (slave == primary) {
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01003157 /* slave's name changed - he's no longer primary */
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02003158 RCU_INIT_POINTER(bond->primary_slave, NULL);
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01003159 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3160 /* we have a new primary slave */
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02003161 rcu_assign_pointer(bond->primary_slave, slave);
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01003162 } else { /* we didn't change primary - exit */
3163 break;
3164 }
3165
Veaceslav Falico76444f52014-07-15 19:35:58 +02003166 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
Nikolay Aleksandrov059b47e2014-09-09 23:17:00 +02003167 primary ? slave_dev->name : "none");
dingtianhongf80889a2014-02-12 12:06:40 +08003168
3169 block_netpoll_tx();
Veaceslav Falico3ec775b2014-01-16 02:04:29 +01003170 bond_select_active_slave(bond);
dingtianhongf80889a2014-02-12 12:06:40 +08003171 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 break;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04003173 case NETDEV_FEAT_CHANGE:
3174 bond_compute_features(bond);
3175 break;
Jiri Pirko4aa5dee2013-07-20 12:13:53 +02003176 case NETDEV_RESEND_IGMP:
3177 /* Propagate to master device */
3178 call_netdevice_notifiers(event, slave->bond->dev);
3179 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180 default:
3181 break;
3182 }
3183
3184 return NOTIFY_DONE;
3185}
3186
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003187/* bond_netdev_event: handle netdev notifier chain events.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 *
3189 * This function receives events for the netdev chain. The caller (an
Alan Sterne041c682006-03-27 01:16:30 -08003190 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 * locks for us to safely manipulate the slave devices (RTNL lock,
3192 * dev_probe_lock).
3193 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003194static int bond_netdev_event(struct notifier_block *this,
3195 unsigned long event, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196{
Jiri Pirko351638e2013-05-28 01:30:21 +00003197 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198
Veaceslav Falico76444f52014-07-15 19:35:58 +02003199 netdev_dbg(event_dev, "event: %lx\n", event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200
Jay Vosburgh0b680e72006-09-22 21:54:10 -07003201 if (!(event_dev->priv_flags & IFF_BONDING))
3202 return NOTIFY_DONE;
3203
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 if (event_dev->flags & IFF_MASTER) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003205 netdev_dbg(event_dev, "IFF_MASTER\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 return bond_master_netdev_event(event, event_dev);
3207 }
3208
3209 if (event_dev->flags & IFF_SLAVE) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003210 netdev_dbg(event_dev, "IFF_SLAVE\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 return bond_slave_netdev_event(event, event_dev);
3212 }
3213
3214 return NOTIFY_DONE;
3215}
3216
3217static struct notifier_block bond_netdev_notifier = {
3218 .notifier_call = bond_netdev_event,
3219};
3220
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003221/*---------------------------- Hashing Policies -----------------------------*/
3222
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003223/* L2 hash helper */
3224static inline u32 bond_eth_hash(struct sk_buff *skb)
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003225{
Jianhua Xiece04d632014-07-17 14:16:25 +08003226 struct ethhdr *ep, hdr_tmp;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003227
Jianhua Xiece04d632014-07-17 14:16:25 +08003228 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3229 if (ep)
3230 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003231 return 0;
3232}
3233
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003234/* Extract the appropriate headers based on bond's xmit policy */
3235static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3236 struct flow_keys *fk)
John Eaglesham6b923cb2012-08-21 20:43:35 +00003237{
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003238 const struct ipv6hdr *iph6;
Eric Dumazet43945422013-04-15 17:03:24 +00003239 const struct iphdr *iph;
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003240 int noff, proto = -1;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003241
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003242 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
Tom Herbertcd79a232015-09-01 09:24:27 -07003243 return skb_flow_dissect_flow_keys(skb, fk, 0);
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003244
Jiri Pirko06635a32015-05-12 14:56:16 +02003245 fk->ports.ports = 0;
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003246 noff = skb_network_offset(skb);
3247 if (skb->protocol == htons(ETH_P_IP)) {
dingtianhong054bb882014-03-25 17:00:09 +08003248 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003249 return false;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003250 iph = ip_hdr(skb);
Tom Herbertc3f83242015-06-04 09:16:40 -07003251 iph_to_flow_copy_v4addrs(fk, iph);
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003252 noff += iph->ihl << 2;
3253 if (!ip_is_fragment(iph))
3254 proto = iph->protocol;
3255 } else if (skb->protocol == htons(ETH_P_IPV6)) {
dingtianhong054bb882014-03-25 17:00:09 +08003256 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003257 return false;
3258 iph6 = ipv6_hdr(skb);
Tom Herbertc3f83242015-06-04 09:16:40 -07003259 iph_to_flow_copy_v6addrs(fk, iph6);
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003260 noff += sizeof(*iph6);
3261 proto = iph6->nexthdr;
3262 } else {
3263 return false;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003264 }
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003265 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
Jiri Pirko06635a32015-05-12 14:56:16 +02003266 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
John Eaglesham6b923cb2012-08-21 20:43:35 +00003267
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003268 return true;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003269}
3270
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003271/**
3272 * bond_xmit_hash - generate a hash value based on the xmit policy
3273 * @bond: bonding device
3274 * @skb: buffer to use for headers
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003275 *
3276 * This function will extract the necessary headers from the skb buffer and use
3277 * them to generate a hash based on the xmit_policy set in the bonding device
John Eaglesham6b923cb2012-08-21 20:43:35 +00003278 */
Mahesh Bandewaree62e862014-04-22 16:30:15 -07003279u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
John Eaglesham6b923cb2012-08-21 20:43:35 +00003280{
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003281 struct flow_keys flow;
3282 u32 hash;
John Eaglesham6b923cb2012-08-21 20:43:35 +00003283
Eric Dumazet4b1b8652015-09-15 15:24:28 -07003284 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3285 skb->l4_hash)
3286 return skb->hash;
3287
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003288 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3289 !bond_flow_dissect(bond, skb, &flow))
Mahesh Bandewaree62e862014-04-22 16:30:15 -07003290 return bond_eth_hash(skb);
Eric Dumazet43945422013-04-15 17:03:24 +00003291
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003292 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3293 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3294 hash = bond_eth_hash(skb);
3295 else
Jiri Pirko06635a32015-05-12 14:56:16 +02003296 hash = (__force u32)flow.ports.ports;
Tom Herbertc3f83242015-06-04 09:16:40 -07003297 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3298 (__force u32)flow_get_u32_src(&flow);
Nikolay Aleksandrov32819dc2013-10-02 13:39:25 +02003299 hash ^= (hash >> 16);
3300 hash ^= (hash >> 8);
John Eaglesham6b923cb2012-08-21 20:43:35 +00003301
Hangbin Liub5f86212017-11-06 09:01:57 +08003302 return hash >> 1;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003303}
3304
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305/*-------------------------- Device entry points ----------------------------*/
3306
Mahesh Bandewarea8ffc02017-04-20 12:49:24 -07003307void bond_work_init_all(struct bonding *bond)
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003308{
3309 INIT_DELAYED_WORK(&bond->mcast_work,
3310 bond_resend_igmp_join_requests_delayed);
3311 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3312 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
Mahesh Bandeward5e73f72017-03-08 10:55:51 -08003313 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003314 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
Mahesh Bandewaree637712014-10-04 17:45:01 -07003315 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003316}
3317
3318static void bond_work_cancel_all(struct bonding *bond)
3319{
3320 cancel_delayed_work_sync(&bond->mii_work);
3321 cancel_delayed_work_sync(&bond->arp_work);
3322 cancel_delayed_work_sync(&bond->alb_work);
3323 cancel_delayed_work_sync(&bond->ad_work);
3324 cancel_delayed_work_sync(&bond->mcast_work);
Mahesh Bandewaree637712014-10-04 17:45:01 -07003325 cancel_delayed_work_sync(&bond->slave_arr_work);
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003326}
3327
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328static int bond_open(struct net_device *bond_dev)
3329{
Wang Chen454d7c92008-11-12 23:37:49 -08003330 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003331 struct list_head *iter;
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003332 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003334 /* reset slave->backup and slave->inactive */
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02003335 if (bond_has_slaves(bond)) {
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003336 bond_for_each_slave(bond, slave, iter) {
Eric Dumazet4740d632014-07-15 06:56:55 -07003337 if (bond_uses_primary(bond) &&
3338 slave != rcu_access_pointer(bond->curr_active_slave)) {
dingtianhong5e5b0662014-02-26 11:05:22 +08003339 bond_set_slave_inactive_flags(slave,
3340 BOND_SLAVE_NOTIFY_NOW);
Wilson Kok8bbe71a2015-01-26 01:16:58 -05003341 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
dingtianhong5e5b0662014-02-26 11:05:22 +08003342 bond_set_slave_active_flags(slave,
3343 BOND_SLAVE_NOTIFY_NOW);
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003344 }
3345 }
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003346 }
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003347
Holger Eitzenberger58402052008-12-09 23:07:13 -08003348 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 /* bond_alb_initialize must be called before the timer
3350 * is started.
3351 */
Veaceslav Falico01844092014-05-15 21:39:55 +02003352 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
stephen hemmingerb4739462010-01-25 23:34:15 +00003353 return -ENOMEM;
Debabrata Banerjeee79c1052018-05-14 14:48:09 -04003354 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
Mahesh Bandeware9f0fb82014-04-22 16:30:22 -07003355 queue_delayed_work(bond->wq, &bond->alb_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356 }
3357
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003358 if (bond->params.miimon) /* link check interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003359 queue_delayed_work(bond->wq, &bond->mii_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360
3361 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003362 queue_delayed_work(bond->wq, &bond->arp_work, 0);
Veaceslav Falico3fe68df2014-02-18 07:48:39 +01003363 bond->recv_probe = bond_arp_rcv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 }
3365
Veaceslav Falico01844092014-05-15 21:39:55 +02003366 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003367 queue_delayed_work(bond->wq, &bond->ad_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 /* register to receive LACPDUs */
Jiri Pirko3aba8912011-04-19 03:48:16 +00003369 bond->recv_probe = bond_3ad_lacpdu_recv;
Jay Vosburghfd989c82008-11-04 17:51:16 -08003370 bond_3ad_initiate_agg_selection(bond, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 }
3372
Debabrata Banerjeee79c1052018-05-14 14:48:09 -04003373 if (bond_mode_can_use_xmit_hash(bond))
Mahesh Bandewaree637712014-10-04 17:45:01 -07003374 bond_update_slave_arr(bond, NULL);
3375
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 return 0;
3377}
3378
3379static int bond_close(struct net_device *bond_dev)
3380{
Wang Chen454d7c92008-11-12 23:37:49 -08003381 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382
nikolay@redhat.comfbb0c412012-11-29 01:31:31 +00003383 bond_work_cancel_all(bond);
nikolay@redhat.com6c8d23f2013-09-02 13:51:38 +02003384 bond->send_peer_notif = 0;
nikolay@redhat.comee8487c2013-09-02 13:51:39 +02003385 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386 bond_alb_deinitialize(bond);
Jiri Pirko3aba8912011-04-19 03:48:16 +00003387 bond->recv_probe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388
3389 return 0;
3390}
3391
Eric Dumazetfe309372016-03-17 17:23:36 -07003392/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3393 * that some drivers can provide 32bit values only.
3394 */
3395static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3396 const struct rtnl_link_stats64 *_new,
3397 const struct rtnl_link_stats64 *_old)
3398{
3399 const u64 *new = (const u64 *)_new;
3400 const u64 *old = (const u64 *)_old;
3401 u64 *res = (u64 *)_res;
3402 int i;
3403
3404 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3405 u64 nv = new[i];
3406 u64 ov = old[i];
Eric Dumazet142c6592017-03-29 10:45:44 -07003407 s64 delta = nv - ov;
Eric Dumazetfe309372016-03-17 17:23:36 -07003408
3409 /* detects if this particular field is 32bit only */
3410 if (((nv | ov) >> 32) == 0)
Eric Dumazet142c6592017-03-29 10:45:44 -07003411 delta = (s64)(s32)((u32)nv - (u32)ov);
3412
3413 /* filter anomalies, some drivers reset their stats
3414 * at down/up events.
3415 */
3416 if (delta > 0)
3417 res[i] += delta;
Eric Dumazetfe309372016-03-17 17:23:36 -07003418 }
3419}
3420
Eric Dumazet7e2556e2018-07-31 06:30:54 -07003421static int bond_get_nest_level(struct net_device *bond_dev)
3422{
3423 struct bonding *bond = netdev_priv(bond_dev);
3424
3425 return bond->nest_level;
3426}
3427
stephen hemmingerbc1f4472017-01-06 19:12:52 -08003428static void bond_get_stats(struct net_device *bond_dev,
3429 struct rtnl_link_stats64 *stats)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430{
Wang Chen454d7c92008-11-12 23:37:49 -08003431 struct bonding *bond = netdev_priv(bond_dev);
Eric Dumazet28172732010-07-07 14:58:56 -07003432 struct rtnl_link_stats64 temp;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003433 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435
Eric Dumazet7e2556e2018-07-31 06:30:54 -07003436 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003437 memcpy(stats, &bond->bond_stats, sizeof(*stats));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438
Eric Dumazetfe309372016-03-17 17:23:36 -07003439 rcu_read_lock();
3440 bond_for_each_slave_rcu(bond, slave, iter) {
3441 const struct rtnl_link_stats64 *new =
Eric Dumazet28172732010-07-07 14:58:56 -07003442 dev_get_stats(slave->dev, &temp);
Stephen Hemmingereeda3fd2008-11-19 21:40:23 -08003443
Eric Dumazetfe309372016-03-17 17:23:36 -07003444 bond_fold_stats(stats, new, &slave->slave_stats);
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003445
3446 /* save off the slave stats for the next run */
Eric Dumazetfe309372016-03-17 17:23:36 -07003447 memcpy(&slave->slave_stats, new, sizeof(*new));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448 }
Eric Dumazetfe309372016-03-17 17:23:36 -07003449 rcu_read_unlock();
3450
Andy Gospodarek5f0c5f72014-09-28 22:34:37 -04003451 memcpy(&bond->bond_stats, stats, sizeof(*stats));
Eric Dumazetfe309372016-03-17 17:23:36 -07003452 spin_unlock(&bond->stats_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453}
3454
3455static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3456{
Jiri Pirko080a06e2013-10-18 17:43:36 +02003457 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 struct net_device *slave_dev = NULL;
3459 struct ifbond k_binfo;
3460 struct ifbond __user *u_binfo = NULL;
3461 struct ifslave k_sinfo;
3462 struct ifslave __user *u_sinfo = NULL;
3463 struct mii_ioctl_data *mii = NULL;
Nikolay Aleksandrovd1fbd3ed2014-01-22 14:53:35 +01003464 struct bond_opt_value newval;
Gao feng387ff9112013-01-31 16:31:00 +00003465 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466 int res = 0;
3467
Veaceslav Falico76444f52014-07-15 19:35:58 +02003468 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469
3470 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471 case SIOCGMIIPHY:
3472 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003473 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003475
Linus Torvalds1da177e2005-04-16 15:20:36 -07003476 mii->phy_id = 0;
3477 /* Fall Through */
3478 case SIOCGMIIREG:
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003479 /* We do this again just in case we were called by SIOCGMIIREG
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 * instead of SIOCGMIIPHY.
3481 */
3482 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003483 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003485
Linus Torvalds1da177e2005-04-16 15:20:36 -07003486 if (mii->reg_num == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 mii->val_out = 0;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003488 if (netif_carrier_ok(bond->dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 mii->val_out = BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490 }
3491
3492 return 0;
3493 case BOND_INFO_QUERY_OLD:
3494 case SIOCBONDINFOQUERY:
3495 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3496
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003497 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499
Zhu Yanjun3d675762017-02-02 23:46:21 -05003500 bond_info_query(bond_dev, &k_binfo);
3501 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003502 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503
Zhu Yanjun3d675762017-02-02 23:46:21 -05003504 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505 case BOND_SLAVE_INFO_QUERY_OLD:
3506 case SIOCBONDSLAVEINFOQUERY:
3507 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3508
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003509 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511
3512 res = bond_slave_info_query(bond_dev, &k_sinfo);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003513 if (res == 0 &&
3514 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3515 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516
3517 return res;
3518 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519 break;
3520 }
3521
Gao feng387ff9112013-01-31 16:31:00 +00003522 net = dev_net(bond_dev);
3523
3524 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526
Ying Xue0917b932014-01-15 10:23:37 +08003527 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003528
Veaceslav Falico76444f52014-07-15 19:35:58 +02003529 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003531 if (!slave_dev)
Ying Xue0917b932014-01-15 10:23:37 +08003532 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533
Veaceslav Falico76444f52014-07-15 19:35:58 +02003534 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
Ying Xue0917b932014-01-15 10:23:37 +08003535 switch (cmd) {
3536 case BOND_ENSLAVE_OLD:
3537 case SIOCBONDENSLAVE:
David Ahern33eaf2a2017-10-04 17:48:46 -07003538 res = bond_enslave(bond_dev, slave_dev, NULL);
Ying Xue0917b932014-01-15 10:23:37 +08003539 break;
3540 case BOND_RELEASE_OLD:
3541 case SIOCBONDRELEASE:
3542 res = bond_release(bond_dev, slave_dev);
3543 break;
3544 case BOND_SETHWADDR_OLD:
3545 case SIOCBONDSETHWADDR:
3546 bond_set_dev_addr(bond_dev, slave_dev);
3547 res = 0;
3548 break;
3549 case BOND_CHANGE_ACTIVE_OLD:
3550 case SIOCBONDCHANGEACTIVE:
Nikolay Aleksandrovd1fbd3ed2014-01-22 14:53:35 +01003551 bond_opt_initstr(&newval, slave_dev->name);
Vlad Yasevich7a7e96e2017-05-27 10:14:35 -04003552 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3553 &newval);
Ying Xue0917b932014-01-15 10:23:37 +08003554 break;
3555 default:
3556 res = -EOPNOTSUPP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557 }
3558
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559 return res;
3560}
3561
Jiri Pirkod03462b2011-08-16 03:15:04 +00003562static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3563{
3564 struct bonding *bond = netdev_priv(bond_dev);
3565
3566 if (change & IFF_PROMISC)
3567 bond_set_promiscuity(bond,
3568 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3569
3570 if (change & IFF_ALLMULTI)
3571 bond_set_allmulti(bond,
3572 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3573}
3574
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003575static void bond_set_rx_mode(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576{
Wang Chen454d7c92008-11-12 23:37:49 -08003577 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003578 struct list_head *iter;
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003579 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580
Veaceslav Falicob3241872013-09-28 21:18:56 +02003581 rcu_read_lock();
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02003582 if (bond_uses_primary(bond)) {
Veaceslav Falicob3241872013-09-28 21:18:56 +02003583 slave = rcu_dereference(bond->curr_active_slave);
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003584 if (slave) {
3585 dev_uc_sync(slave->dev, bond_dev);
3586 dev_mc_sync(slave->dev, bond_dev);
3587 }
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003588 } else {
Veaceslav Falicob3241872013-09-28 21:18:56 +02003589 bond_for_each_slave_rcu(bond, slave, iter) {
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00003590 dev_uc_sync_multiple(slave->dev, bond_dev);
3591 dev_mc_sync_multiple(slave->dev, bond_dev);
3592 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 }
Veaceslav Falicob3241872013-09-28 21:18:56 +02003594 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595}
3596
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003597static int bond_neigh_init(struct neighbour *n)
Stephen Hemminger00829822008-11-20 20:14:53 -08003598{
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003599 struct bonding *bond = netdev_priv(n->dev);
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003600 const struct net_device_ops *slave_ops;
3601 struct neigh_parms parms;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003602 struct slave *slave;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003603 int ret;
Stephen Hemminger00829822008-11-20 20:14:53 -08003604
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02003605 slave = bond_first_slave(bond);
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003606 if (!slave)
3607 return 0;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003608 slave_ops = slave->dev->netdev_ops;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003609 if (!slave_ops->ndo_neigh_setup)
3610 return 0;
3611
3612 parms.neigh_setup = NULL;
3613 parms.neigh_cleanup = NULL;
3614 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3615 if (ret)
3616 return ret;
3617
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003618 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003619 * after the last slave has been detached. Assumes that all slaves
3620 * utilize the same neigh_cleanup (true at this writing as only user
3621 * is ipoib).
3622 */
3623 n->parms->neigh_cleanup = parms.neigh_cleanup;
3624
3625 if (!parms.neigh_setup)
3626 return 0;
3627
3628 return parms.neigh_setup(n);
3629}
3630
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003631/* The bonding ndo_neigh_setup is called at init time beofre any
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003632 * slave exists. So we must declare proxy setup function which will
3633 * be used at run time to resolve the actual slave neigh param setup.
Veaceslav Falico9918d5b2013-08-02 19:07:39 +02003634 *
3635 * It's also called by master devices (such as vlans) to setup their
3636 * underlying devices. In that case - do nothing, we're already set up from
3637 * our init.
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003638 */
3639static int bond_neigh_setup(struct net_device *dev,
3640 struct neigh_parms *parms)
3641{
Veaceslav Falico9918d5b2013-08-02 19:07:39 +02003642 /* modify only our neigh_parms */
3643 if (parms->dev == dev)
3644 parms->neigh_setup = bond_neigh_init;
Shlomo Pongratz234bcf82012-04-03 22:56:20 +00003645
Stephen Hemminger00829822008-11-20 20:14:53 -08003646 return 0;
3647}
3648
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003649/* Change the MTU of all of a master's slaves to match the master */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3651{
Wang Chen454d7c92008-11-12 23:37:49 -08003652 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003653 struct slave *slave, *rollback_slave;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003654 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 int res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656
Veaceslav Falico76444f52014-07-15 19:35:58 +02003657 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003659 bond_for_each_slave(bond, slave, iter) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003660 netdev_dbg(bond_dev, "s %p c_m %p\n",
3661 slave, slave->dev->netdev_ops->ndo_change_mtu);
Mitch Williamse944ef72005-11-09 10:36:50 -08003662
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 res = dev_set_mtu(slave->dev, new_mtu);
3664
3665 if (res) {
3666 /* If we failed to set the slave's mtu to the new value
3667 * we must abort the operation even in ACTIVE_BACKUP
3668 * mode, because if we allow the backup slaves to have
3669 * different mtu values than the active slave we'll
3670 * need to change their mtu when doing a failover. That
3671 * means changing their mtu from timer context, which
3672 * is probably not a good idea.
3673 */
Veaceslav Falico76444f52014-07-15 19:35:58 +02003674 netdev_dbg(bond_dev, "err %d %s\n", res,
3675 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676 goto unwind;
3677 }
3678 }
3679
3680 bond_dev->mtu = new_mtu;
3681
3682 return 0;
3683
3684unwind:
3685 /* unwind from head to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003686 bond_for_each_slave(bond, rollback_slave, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 int tmp_res;
3688
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003689 if (rollback_slave == slave)
3690 break;
3691
3692 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003693 if (tmp_res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003694 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3695 tmp_res, rollback_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 }
3697 }
3698
3699 return res;
3700}
3701
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003702/* Change HW address
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703 *
3704 * Note that many devices must be down to change the HW address, and
3705 * downing the master releases all slaves. We can make bonds full of
3706 * bonding devices to test this, however.
3707 */
3708static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3709{
Wang Chen454d7c92008-11-12 23:37:49 -08003710 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003711 struct slave *slave, *rollback_slave;
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04003712 struct sockaddr_storage *ss = addr, tmp_ss;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003713 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714 int res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715
Veaceslav Falico01844092014-05-15 21:39:55 +02003716 if (BOND_MODE(bond) == BOND_MODE_ALB)
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003717 return bond_alb_set_mac_address(bond_dev, addr);
3718
3719
Veaceslav Falico76444f52014-07-15 19:35:58 +02003720 netdev_dbg(bond_dev, "bond=%p\n", bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003721
Jay Vosburgh1b5acd22013-05-31 11:57:31 +00003722 /* If fail_over_mac is enabled, do nothing and return success.
3723 * Returning an error causes ifenslave to fail.
Jay Vosburghdd957c52007-10-09 19:57:24 -07003724 */
dingtianhongcc689aa2014-01-25 13:00:57 +08003725 if (bond->params.fail_over_mac &&
Veaceslav Falico01844092014-05-15 21:39:55 +02003726 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
Jay Vosburghdd957c52007-10-09 19:57:24 -07003727 return 0;
Moni Shoua2ab82852007-10-09 19:43:39 -07003728
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04003729 if (!is_valid_ether_addr(ss->__data))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730 return -EADDRNOTAVAIL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003732 bond_for_each_slave(bond, slave, iter) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003733 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734 res = dev_set_mac_address(slave->dev, addr);
3735 if (res) {
3736 /* TODO: consider downing the slave
3737 * and retry ?
3738 * User should expect communications
3739 * breakage anyway until ARP finish
3740 * updating, so...
3741 */
Veaceslav Falico76444f52014-07-15 19:35:58 +02003742 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743 goto unwind;
3744 }
3745 }
3746
3747 /* success */
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04003748 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 return 0;
3750
3751unwind:
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04003752 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3753 tmp_ss.ss_family = bond_dev->type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754
3755 /* unwind from head to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003756 bond_for_each_slave(bond, rollback_slave, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757 int tmp_res;
3758
Veaceslav Falico81f23b12013-09-25 09:20:13 +02003759 if (rollback_slave == slave)
3760 break;
3761
Jarod Wilsonfaeeb312017-04-04 17:32:42 -04003762 tmp_res = dev_set_mac_address(rollback_slave->dev,
3763 (struct sockaddr *)&tmp_ss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764 if (tmp_res) {
Veaceslav Falico76444f52014-07-15 19:35:58 +02003765 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3766 tmp_res, rollback_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 }
3768 }
3769
3770 return res;
3771}
3772
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003773/**
3774 * bond_xmit_slave_id - transmit skb through slave with slave_id
3775 * @bond: bonding device that is transmitting
3776 * @skb: buffer to transmit
3777 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3778 *
3779 * This function tries to transmit through slave with slave_id but in case
3780 * it fails, it tries to find the first available slave for transmission.
3781 * The skb is consumed in all cases, thus the function is void.
3782 */
stephen hemmingerda131dd2013-12-29 11:41:25 -08003783static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003784{
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003785 struct list_head *iter;
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003786 struct slave *slave;
3787 int i = slave_id;
3788
3789 /* Here we start from the slave with slave_id */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003790 bond_for_each_slave_rcu(bond, slave, iter) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003791 if (--i < 0) {
Veaceslav Falico891ab542014-05-15 21:39:58 +02003792 if (bond_slave_can_tx(slave)) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003793 bond_dev_queue_xmit(bond, skb, slave->dev);
3794 return;
3795 }
3796 }
3797 }
3798
3799 /* Here we start from the first slave up to slave_id */
3800 i = slave_id;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02003801 bond_for_each_slave_rcu(bond, slave, iter) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003802 if (--i < 0)
3803 break;
Veaceslav Falico891ab542014-05-15 21:39:58 +02003804 if (bond_slave_can_tx(slave)) {
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003805 bond_dev_queue_xmit(bond, skb, slave->dev);
3806 return;
3807 }
3808 }
3809 /* no slave that can tx has been found */
Eric Dumazet31aa8602014-10-31 11:47:54 -07003810 bond_tx_drop(bond->dev, skb);
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003811}
3812
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003813/**
3814 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3815 * @bond: bonding device to use
3816 *
3817 * Based on the value of the bonding device's packets_per_slave parameter
3818 * this function generates a slave id, which is usually used as the next
3819 * slave to transmit through.
3820 */
3821static u32 bond_rr_gen_slave_id(struct bonding *bond)
3822{
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003823 u32 slave_id;
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01003824 struct reciprocal_value reciprocal_packets_per_slave;
3825 int packets_per_slave = bond->params.packets_per_slave;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003826
3827 switch (packets_per_slave) {
3828 case 0:
3829 slave_id = prandom_u32();
3830 break;
3831 case 1:
3832 slave_id = bond->rr_tx_counter;
3833 break;
3834 default:
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01003835 reciprocal_packets_per_slave =
3836 bond->params.reciprocal_packets_per_slave;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003837 slave_id = reciprocal_divide(bond->rr_tx_counter,
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01003838 reciprocal_packets_per_slave);
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003839 break;
3840 }
3841 bond->rr_tx_counter++;
3842
3843 return slave_id;
3844}
3845
Tonghao Zhangdbdc8a22018-05-11 02:53:10 -07003846static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3847 struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848{
Wang Chen454d7c92008-11-12 23:37:49 -08003849 struct bonding *bond = netdev_priv(bond_dev);
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003850 struct iphdr *iph = ip_hdr(skb);
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003851 struct slave *slave;
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003852 u32 slave_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01003854 /* Start with the curr_active_slave that joined the bond as the
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003855 * default for sending IGMP traffic. For failover purposes one
3856 * needs to maintain some consistency for the interface that will
3857 * send the join/membership reports. The curr_active_slave found
3858 * will send all of this type of traffic.
Jay Vosburghcf5f9042007-10-17 17:37:47 -07003859 */
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003860 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003861 slave = rcu_dereference(bond->curr_active_slave);
dingtianhonga64d0442014-09-19 21:04:57 +08003862 if (slave)
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003863 bond_dev_queue_xmit(bond, skb, slave->dev);
3864 else
3865 bond_xmit_slave_id(bond, skb, 0);
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003866 } else {
Mark Rutland6aa7de02017-10-23 14:07:29 -07003867 int slave_cnt = READ_ONCE(bond->slave_cnt);
Nikolay Aleksandrov9a72c2d2014-09-12 17:38:18 +02003868
3869 if (likely(slave_cnt)) {
3870 slave_id = bond_rr_gen_slave_id(bond);
3871 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3872 } else {
Eric Dumazet31aa8602014-10-31 11:47:54 -07003873 bond_tx_drop(bond_dev, skb);
Nikolay Aleksandrov9a72c2d2014-09-12 17:38:18 +02003874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 }
Michał Mirosław0693e882011-05-07 01:48:02 +00003876
Patrick McHardyec634fe2009-07-05 19:23:38 -07003877 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878}
3879
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02003880/* In active-backup mode, we know that bond->curr_active_slave is always valid if
Linus Torvalds1da177e2005-04-16 15:20:36 -07003881 * the bond has a usable interface.
3882 */
Tonghao Zhangdbdc8a22018-05-11 02:53:10 -07003883static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3884 struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885{
Wang Chen454d7c92008-11-12 23:37:49 -08003886 struct bonding *bond = netdev_priv(bond_dev);
nikolay@redhat.com71bc3b22013-08-01 16:54:48 +02003887 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888
nikolay@redhat.com278b2082013-08-01 16:54:51 +02003889 slave = rcu_dereference(bond->curr_active_slave);
nikolay@redhat.com71bc3b22013-08-01 16:54:48 +02003890 if (slave)
Nikolay Aleksandrov15077222013-08-01 16:54:50 +02003891 bond_dev_queue_xmit(bond, skb, slave->dev);
3892 else
Eric Dumazet31aa8602014-10-31 11:47:54 -07003893 bond_tx_drop(bond_dev, skb);
Michał Mirosław0693e882011-05-07 01:48:02 +00003894
Patrick McHardyec634fe2009-07-05 19:23:38 -07003895 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896}
3897
Mahesh Bandewaree637712014-10-04 17:45:01 -07003898/* Use this to update slave_array when (a) it's not appropriate to update
3899 * slave_array right away (note that update_slave_array() may sleep)
3900 * and / or (b) RTNL is not held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901 */
Mahesh Bandewaree637712014-10-04 17:45:01 -07003902void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903{
Mahesh Bandewaree637712014-10-04 17:45:01 -07003904 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3905}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906
Mahesh Bandewaree637712014-10-04 17:45:01 -07003907/* Slave array work handler. Holds only RTNL */
3908static void bond_slave_arr_handler(struct work_struct *work)
3909{
3910 struct bonding *bond = container_of(work, struct bonding,
3911 slave_arr_work.work);
3912 int ret;
3913
3914 if (!rtnl_trylock())
3915 goto err;
3916
3917 ret = bond_update_slave_arr(bond, NULL);
3918 rtnl_unlock();
3919 if (ret) {
3920 pr_warn_ratelimited("Failed to update slave array from WT\n");
3921 goto err;
3922 }
3923 return;
3924
3925err:
3926 bond_slave_arr_work_rearm(bond, 1);
3927}
3928
3929/* Build the usable slaves array in control path for modes that use xmit-hash
3930 * to determine the slave interface -
3931 * (a) BOND_MODE_8023AD
3932 * (b) BOND_MODE_XOR
Debabrata Banerjeee79c1052018-05-14 14:48:09 -04003933 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
Mahesh Bandewaree637712014-10-04 17:45:01 -07003934 *
3935 * The caller is expected to hold RTNL only and NO other lock!
3936 */
3937int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3938{
3939 struct slave *slave;
3940 struct list_head *iter;
3941 struct bond_up_slave *new_arr, *old_arr;
Mahesh Bandewaree637712014-10-04 17:45:01 -07003942 int agg_id = 0;
3943 int ret = 0;
3944
3945#ifdef CONFIG_LOCKDEP
3946 WARN_ON(lockdep_is_held(&bond->mode_lock));
3947#endif
3948
3949 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3950 GFP_KERNEL);
3951 if (!new_arr) {
3952 ret = -ENOMEM;
3953 pr_err("Failed to build slave-array.\n");
3954 goto out;
3955 }
3956 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3957 struct ad_info ad_info;
3958
3959 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3960 pr_debug("bond_3ad_get_active_agg_info failed\n");
3961 kfree_rcu(new_arr, rcu);
3962 /* No active aggragator means it's not safe to use
3963 * the previous array.
3964 */
3965 old_arr = rtnl_dereference(bond->slave_arr);
3966 if (old_arr) {
3967 RCU_INIT_POINTER(bond->slave_arr, NULL);
3968 kfree_rcu(old_arr, rcu);
3969 }
3970 goto out;
3971 }
Mahesh Bandewaree637712014-10-04 17:45:01 -07003972 agg_id = ad_info.aggregator_id;
3973 }
3974 bond_for_each_slave(bond, slave, iter) {
3975 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3976 struct aggregator *agg;
3977
3978 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3979 if (!agg || agg->aggregator_identifier != agg_id)
3980 continue;
3981 }
3982 if (!bond_slave_can_tx(slave))
3983 continue;
3984 if (skipslave == slave)
3985 continue;
Debabrata Banerjeee79c1052018-05-14 14:48:09 -04003986
3987 netdev_dbg(bond->dev,
3988 "Adding slave dev %s to tx hash array[%d]\n",
3989 slave->dev->name, new_arr->count);
3990
Mahesh Bandewaree637712014-10-04 17:45:01 -07003991 new_arr->arr[new_arr->count++] = slave;
3992 }
3993
3994 old_arr = rtnl_dereference(bond->slave_arr);
3995 rcu_assign_pointer(bond->slave_arr, new_arr);
3996 if (old_arr)
3997 kfree_rcu(old_arr, rcu);
3998out:
3999 if (ret != 0 && skipslave) {
4000 int idx;
4001
4002 /* Rare situation where caller has asked to skip a specific
4003 * slave but allocation failed (most likely!). BTW this is
4004 * only possible when the call is initiated from
4005 * __bond_release_one(). In this situation; overwrite the
4006 * skipslave entry in the array with the last entry from the
4007 * array to avoid a situation where the xmit path may choose
4008 * this to-be-skipped slave to send a packet out.
4009 */
4010 old_arr = rtnl_dereference(bond->slave_arr);
4011 for (idx = 0; idx < old_arr->count; idx++) {
4012 if (skipslave == old_arr->arr[idx]) {
4013 old_arr->arr[idx] =
4014 old_arr->arr[old_arr->count-1];
4015 old_arr->count--;
4016 break;
4017 }
4018 }
4019 }
4020 return ret;
4021}
4022
4023/* Use this Xmit function for 3AD as well as XOR modes. The current
4024 * usable slave array is formed in the control path. The xmit function
4025 * just calculates hash and sends the packet out.
4026 */
Tonghao Zhangdbdc8a22018-05-11 02:53:10 -07004027static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4028 struct net_device *dev)
Mahesh Bandewaree637712014-10-04 17:45:01 -07004029{
4030 struct bonding *bond = netdev_priv(dev);
4031 struct slave *slave;
4032 struct bond_up_slave *slaves;
4033 unsigned int count;
4034
4035 slaves = rcu_dereference(bond->slave_arr);
Mark Rutland6aa7de02017-10-23 14:07:29 -07004036 count = slaves ? READ_ONCE(slaves->count) : 0;
Mahesh Bandewaree637712014-10-04 17:45:01 -07004037 if (likely(count)) {
4038 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4039 bond_dev_queue_xmit(bond, skb, slave->dev);
4040 } else {
Eric Dumazet31aa8602014-10-31 11:47:54 -07004041 bond_tx_drop(dev, skb);
Mahesh Bandewaree637712014-10-04 17:45:01 -07004042 }
Michał Mirosław0693e882011-05-07 01:48:02 +00004043
Patrick McHardyec634fe2009-07-05 19:23:38 -07004044 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045}
4046
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02004047/* in broadcast mode, we send everything to all usable interfaces. */
Tonghao Zhangdbdc8a22018-05-11 02:53:10 -07004048static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4049 struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050{
Wang Chen454d7c92008-11-12 23:37:49 -08004051 struct bonding *bond = netdev_priv(bond_dev);
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02004052 struct slave *slave = NULL;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02004053 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004054
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02004055 bond_for_each_slave_rcu(bond, slave, iter) {
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02004056 if (bond_is_last_slave(bond, slave))
4057 break;
Veaceslav Falicob6adc612014-05-15 21:39:57 +02004058 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02004059 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02004061 if (!skb2) {
dingtianhong9152e262014-03-25 17:00:10 +08004062 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4063 bond_dev->name, __func__);
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02004064 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065 }
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02004066 bond_dev_queue_xmit(bond, skb2, slave->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067 }
4068 }
Veaceslav Falicob6adc612014-05-15 21:39:57 +02004069 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
Nikolay Aleksandrov78a646c2013-08-01 16:54:49 +02004070 bond_dev_queue_xmit(bond, skb, slave->dev);
4071 else
Eric Dumazet31aa8602014-10-31 11:47:54 -07004072 bond_tx_drop(bond_dev, skb);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004073
Patrick McHardyec634fe2009-07-05 19:23:38 -07004074 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075}
4076
4077/*------------------------- Device initialization ---------------------------*/
4078
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004079/* Lookup the slave that corresponds to a qid */
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004080static inline int bond_slave_override(struct bonding *bond,
4081 struct sk_buff *skb)
4082{
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004083 struct slave *slave = NULL;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02004084 struct list_head *iter;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004085
Tonghao Zhangae35c6f2018-05-11 02:53:11 -07004086 if (!skb_rx_queue_recorded(skb))
Michał Mirosław0693e882011-05-07 01:48:02 +00004087 return 1;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004088
4089 /* Find out if any slaves have the same mapping as this skb. */
dingtianhong3900f292014-01-02 09:13:06 +08004090 bond_for_each_slave_rcu(bond, slave, iter) {
Tonghao Zhangae35c6f2018-05-11 02:53:11 -07004091 if (slave->queue_id == skb_get_queue_mapping(skb)) {
Anton Nayshtutf5e2dc52015-03-29 14:20:25 +03004092 if (bond_slave_is_up(slave) &&
4093 slave->link == BOND_LINK_UP) {
dingtianhong3900f292014-01-02 09:13:06 +08004094 bond_dev_queue_xmit(bond, skb, slave->dev);
4095 return 0;
4096 }
4097 /* If the slave isn't UP, use default transmit policy. */
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004098 break;
4099 }
4100 }
4101
dingtianhong3900f292014-01-02 09:13:06 +08004102 return 1;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004103}
4104
Neil Horman374eeb52011-06-03 10:35:52 +00004105
Jason Wangf663dd92014-01-10 16:18:26 +08004106static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
Alexander Duyck4f49dec2018-07-09 12:19:59 -04004107 struct net_device *sb_dev,
4108 select_queue_fallback_t fallback)
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004109{
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004110 /* This helper function exists to help dev_pick_tx get the correct
Phil Oesterfd0e4352011-03-14 06:22:04 +00004111 * destination queue. Using a helper function skips a call to
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004112 * skb_tx_hash and will put the skbs in the queue we expect on their
4113 * way down to the bonding driver.
4114 */
Phil Oesterfd0e4352011-03-14 06:22:04 +00004115 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4116
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004117 /* Save the original txq to restore before passing to the driver */
Tonghao Zhangae35c6f2018-05-11 02:53:11 -07004118 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
Neil Horman374eeb52011-06-03 10:35:52 +00004119
Phil Oesterfd0e4352011-03-14 06:22:04 +00004120 if (unlikely(txq >= dev->real_num_tx_queues)) {
David Decotignyd30ee672011-04-13 15:22:29 +00004121 do {
Phil Oesterfd0e4352011-03-14 06:22:04 +00004122 txq -= dev->real_num_tx_queues;
David Decotignyd30ee672011-04-13 15:22:29 +00004123 } while (txq >= dev->real_num_tx_queues);
Phil Oesterfd0e4352011-03-14 06:22:04 +00004124 }
4125 return txq;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004126}
4127
Michał Mirosław0693e882011-05-07 01:48:02 +00004128static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
Stephen Hemminger00829822008-11-20 20:14:53 -08004129{
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004130 struct bonding *bond = netdev_priv(dev);
4131
Veaceslav Falicod1e2e5c2014-05-15 21:39:52 +02004132 if (bond_should_override_tx_queue(bond) &&
4133 !bond_slave_override(bond, skb))
4134 return NETDEV_TX_OK;
Stephen Hemminger00829822008-11-20 20:14:53 -08004135
Veaceslav Falico01844092014-05-15 21:39:55 +02004136 switch (BOND_MODE(bond)) {
Stephen Hemminger00829822008-11-20 20:14:53 -08004137 case BOND_MODE_ROUNDROBIN:
4138 return bond_xmit_roundrobin(skb, dev);
4139 case BOND_MODE_ACTIVEBACKUP:
4140 return bond_xmit_activebackup(skb, dev);
Mahesh Bandewaree637712014-10-04 17:45:01 -07004141 case BOND_MODE_8023AD:
Stephen Hemminger00829822008-11-20 20:14:53 -08004142 case BOND_MODE_XOR:
Mahesh Bandewaree637712014-10-04 17:45:01 -07004143 return bond_3ad_xor_xmit(skb, dev);
Stephen Hemminger00829822008-11-20 20:14:53 -08004144 case BOND_MODE_BROADCAST:
4145 return bond_xmit_broadcast(skb, dev);
Stephen Hemminger00829822008-11-20 20:14:53 -08004146 case BOND_MODE_ALB:
Stephen Hemminger00829822008-11-20 20:14:53 -08004147 return bond_alb_xmit(skb, dev);
Mahesh Bandewarf05b42e2014-04-22 16:30:20 -07004148 case BOND_MODE_TLB:
4149 return bond_tlb_xmit(skb, dev);
Stephen Hemminger00829822008-11-20 20:14:53 -08004150 default:
4151 /* Should never happen, mode already checked */
Veaceslav Falico76444f52014-07-15 19:35:58 +02004152 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
Stephen Hemminger00829822008-11-20 20:14:53 -08004153 WARN_ON_ONCE(1);
Eric Dumazet31aa8602014-10-31 11:47:54 -07004154 bond_tx_drop(dev, skb);
Stephen Hemminger00829822008-11-20 20:14:53 -08004155 return NETDEV_TX_OK;
4156 }
4157}
4158
Michał Mirosław0693e882011-05-07 01:48:02 +00004159static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4160{
4161 struct bonding *bond = netdev_priv(dev);
4162 netdev_tx_t ret = NETDEV_TX_OK;
4163
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004164 /* If we risk deadlock from transmitting this in the
Michał Mirosław0693e882011-05-07 01:48:02 +00004165 * netpoll path, tell netpoll to queue the frame for later tx
4166 */
dingtianhong054bb882014-03-25 17:00:09 +08004167 if (unlikely(is_netpoll_tx_blocked(dev)))
Michał Mirosław0693e882011-05-07 01:48:02 +00004168 return NETDEV_TX_BUSY;
4169
nikolay@redhat.com278b2082013-08-01 16:54:51 +02004170 rcu_read_lock();
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02004171 if (bond_has_slaves(bond))
Michał Mirosław0693e882011-05-07 01:48:02 +00004172 ret = __bond_start_xmit(skb, dev);
4173 else
Eric Dumazet31aa8602014-10-31 11:47:54 -07004174 bond_tx_drop(dev, skb);
nikolay@redhat.com278b2082013-08-01 16:54:51 +02004175 rcu_read_unlock();
Michał Mirosław0693e882011-05-07 01:48:02 +00004176
4177 return ret;
4178}
Stephen Hemminger00829822008-11-20 20:14:53 -08004179
Philippe Reynesd46b6342016-10-25 18:41:31 +02004180static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4181 struct ethtool_link_ksettings *cmd)
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004182{
4183 struct bonding *bond = netdev_priv(bond_dev);
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004184 unsigned long speed = 0;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02004185 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02004186 struct slave *slave;
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004187
Philippe Reynesd46b6342016-10-25 18:41:31 +02004188 cmd->base.duplex = DUPLEX_UNKNOWN;
4189 cmd->base.port = PORT_OTHER;
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004190
Veaceslav Falico8557cd72014-05-15 21:39:59 +02004191 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004192 * do not need to check mode. Though link speed might not represent
4193 * the true receive or transmit bandwidth (not all modes are symmetric)
4194 * this is an accurate maximum.
4195 */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02004196 bond_for_each_slave(bond, slave, iter) {
Veaceslav Falico8557cd72014-05-15 21:39:59 +02004197 if (bond_slave_can_tx(slave)) {
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004198 if (slave->speed != SPEED_UNKNOWN)
4199 speed += slave->speed;
Philippe Reynesd46b6342016-10-25 18:41:31 +02004200 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004201 slave->duplex != DUPLEX_UNKNOWN)
Philippe Reynesd46b6342016-10-25 18:41:31 +02004202 cmd->base.duplex = slave->duplex;
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004203 }
4204 }
Philippe Reynesd46b6342016-10-25 18:41:31 +02004205 cmd->base.speed = speed ? : SPEED_UNKNOWN;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02004206
Andy Gospodarekbb5b0522013-04-16 14:46:00 +00004207 return 0;
4208}
4209
Jay Vosburgh217df672005-09-26 16:11:50 -07004210static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
Jiri Pirko7826d432013-01-06 00:44:26 +00004211 struct ethtool_drvinfo *drvinfo)
Jay Vosburgh217df672005-09-26 16:11:50 -07004212{
Jiri Pirko7826d432013-01-06 00:44:26 +00004213 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4214 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4215 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4216 BOND_ABI_VERSION);
Jay Vosburgh217df672005-09-26 16:11:50 -07004217}
4218
Jeff Garzik7282d492006-09-13 14:30:00 -04004219static const struct ethtool_ops bond_ethtool_ops = {
Jay Vosburgh217df672005-09-26 16:11:50 -07004220 .get_drvinfo = bond_ethtool_get_drvinfo,
Stephen Hemmingerfa53eba2008-09-13 21:17:09 -04004221 .get_link = ethtool_op_get_link,
Philippe Reynesd46b6342016-10-25 18:41:31 +02004222 .get_link_ksettings = bond_ethtool_get_link_ksettings,
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004223};
4224
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004225static const struct net_device_ops bond_netdev_ops = {
Stephen Hemminger181470f2009-06-12 19:02:52 +00004226 .ndo_init = bond_init,
Stephen Hemminger9e716262009-06-12 19:02:47 +00004227 .ndo_uninit = bond_uninit,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004228 .ndo_open = bond_open,
4229 .ndo_stop = bond_close,
Stephen Hemminger00829822008-11-20 20:14:53 -08004230 .ndo_start_xmit = bond_start_xmit,
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004231 .ndo_select_queue = bond_select_queue,
Ben Hutchingsbe1f3c22010-06-08 07:19:54 +00004232 .ndo_get_stats64 = bond_get_stats,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004233 .ndo_do_ioctl = bond_do_ioctl,
Jiri Pirkod03462b2011-08-16 03:15:04 +00004234 .ndo_change_rx_flags = bond_change_rx_flags,
Jay Vosburgh303d1cb2013-05-31 11:57:30 +00004235 .ndo_set_rx_mode = bond_set_rx_mode,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004236 .ndo_change_mtu = bond_change_mtu,
Jiri Pirkocc0e4072011-07-20 04:54:46 +00004237 .ndo_set_mac_address = bond_set_mac_address,
Stephen Hemminger00829822008-11-20 20:14:53 -08004238 .ndo_neigh_setup = bond_neigh_setup,
Jiri Pirkocc0e4072011-07-20 04:54:46 +00004239 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004240 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
Eric Dumazet7e2556e2018-07-31 06:30:54 -07004241 .ndo_get_lock_subclass = bond_get_nest_level,
WANG Congf6dc31a2010-05-06 00:48:51 -07004242#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +00004243 .ndo_netpoll_setup = bond_netpoll_setup,
WANG Congf6dc31a2010-05-06 00:48:51 -07004244 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4245 .ndo_poll_controller = bond_poll_controller,
4246#endif
Jiri Pirko9232ecc2011-02-13 09:33:01 +00004247 .ndo_add_slave = bond_enslave,
4248 .ndo_del_slave = bond_release,
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004249 .ndo_fix_features = bond_fix_features,
Toshiaki Makita4847f042015-03-27 14:31:14 +09004250 .ndo_features_check = passthru_features_check,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004251};
4252
Doug Goldsteinb3f92b62013-02-18 14:59:23 +00004253static const struct device_type bond_type = {
4254 .name = "bond",
4255};
4256
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004257static void bond_destructor(struct net_device *bond_dev)
4258{
4259 struct bonding *bond = netdev_priv(bond_dev);
4260 if (bond->wq)
4261 destroy_workqueue(bond->wq);
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004262}
4263
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004264void bond_setup(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004265{
Wang Chen454d7c92008-11-12 23:37:49 -08004266 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004267
Nikolay Aleksandrovb7435622014-09-11 22:49:25 +02004268 spin_lock_init(&bond->mode_lock);
Eric Dumazetfe309372016-03-17 17:23:36 -07004269 spin_lock_init(&bond->stats_lock);
Stephen Hemmingerd2991f72009-06-12 19:02:44 +00004270 bond->params = bonding_defaults;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271
4272 /* Initialize pointers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004273 bond->dev = bond_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274
4275 /* Initialize the device entry points */
Stephen Hemminger181470f2009-06-12 19:02:52 +00004276 ether_setup(bond_dev);
WANG Cong31c05412017-03-02 12:24:36 -08004277 bond_dev->max_mtu = ETH_MAX_MTU;
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004278 bond_dev->netdev_ops = &bond_netdev_ops;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004279 bond_dev->ethtool_ops = &bond_ethtool_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004280
David S. Millercf124db2017-05-08 12:52:56 -04004281 bond_dev->needs_free_netdev = true;
4282 bond_dev->priv_destructor = bond_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004283
Doug Goldsteinb3f92b62013-02-18 14:59:23 +00004284 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4285
Linus Torvalds1da177e2005-04-16 15:20:36 -07004286 /* Initialize the device options */
Zhang Shengju1098cee2016-03-16 09:59:15 +00004287 bond_dev->flags |= IFF_MASTER;
Phil Sutter1e6f20c2015-08-18 10:30:39 +02004288 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
Neil Horman550fd082011-07-26 06:05:38 +00004289 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004290
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004291 /* don't acquire bond device's netif_tx_lock when transmitting */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004292 bond_dev->features |= NETIF_F_LLTX;
4293
4294 /* By default, we declare the bond to be fully
4295 * VLAN hardware accelerated capable. Special
4296 * care is taken in the various xmit functions
4297 * when there are slaves that are not hw accel
4298 * capable
4299 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004300
Weilong Chenf9399812014-01-22 17:16:30 +08004301 /* Don't allow bond devices to change network namespaces. */
4302 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4303
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004304 bond_dev->hw_features = BOND_VLAN_FEATURES |
Patrick McHardyf6469682013-04-19 02:04:27 +00004305 NETIF_F_HW_VLAN_CTAG_TX |
4306 NETIF_F_HW_VLAN_CTAG_RX |
4307 NETIF_F_HW_VLAN_CTAG_FILTER;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004308
Willem de Bruijn8eea1ca2018-05-22 11:34:40 -04004309 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004310 bond_dev->features |= bond_dev->hw_features;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311}
4312
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004313/* Destroy a bonding device.
4314 * Must be under rtnl_lock when this function is called.
4315 */
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004316static void bond_uninit(struct net_device *bond_dev)
Jay Vosburgha434e432008-10-30 17:41:15 -07004317{
Wang Chen454d7c92008-11-12 23:37:49 -08004318 struct bonding *bond = netdev_priv(bond_dev);
Veaceslav Falico544a0282013-09-25 09:20:15 +02004319 struct list_head *iter;
4320 struct slave *slave;
Mahesh Bandewaree637712014-10-04 17:45:01 -07004321 struct bond_up_slave *arr;
Jay Vosburgha434e432008-10-30 17:41:15 -07004322
WANG Congf6dc31a2010-05-06 00:48:51 -07004323 bond_netpoll_cleanup(bond_dev);
4324
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004325 /* Release the bonded slaves */
Veaceslav Falico544a0282013-09-25 09:20:15 +02004326 bond_for_each_slave(bond, slave, iter)
WANG Congf51048c2017-07-06 15:01:57 -07004327 __bond_release_one(bond_dev, slave->dev, true, true);
Veaceslav Falico76444f52014-07-15 19:35:58 +02004328 netdev_info(bond_dev, "Released all slaves\n");
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004329
Mahesh Bandewaree637712014-10-04 17:45:01 -07004330 arr = rtnl_dereference(bond->slave_arr);
4331 if (arr) {
4332 RCU_INIT_POINTER(bond->slave_arr, NULL);
4333 kfree_rcu(arr, rcu);
4334 }
4335
Jay Vosburgha434e432008-10-30 17:41:15 -07004336 list_del(&bond->bond_list);
4337
Taku Izumif073c7c2010-12-09 15:17:13 +00004338 bond_debug_unregister(bond);
Jay Vosburgha434e432008-10-30 17:41:15 -07004339}
4340
Linus Torvalds1da177e2005-04-16 15:20:36 -07004341/*------------------------- Module initialization ---------------------------*/
4342
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343static int bond_check_params(struct bond_params *params)
4344{
nikolay@redhat.com5a5c5fd2013-05-18 01:18:30 +00004345 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
stephen hemmingerf3253332014-03-04 16:36:44 -08004346 struct bond_opt_value newval;
4347 const struct bond_opt_value *valptr;
Jarod Wilson72ccc472017-05-19 14:46:46 -04004348 int arp_all_targets_value = 0;
Mahesh Bandewar6791e462015-05-09 00:01:55 -07004349 u16 ad_actor_sys_prio = 0;
Mahesh Bandeward22a5fc2015-05-09 00:01:57 -07004350 u16 ad_user_port_key = 0;
Jarod Wilson72ccc472017-05-19 14:46:46 -04004351 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
Mahesh Bandewardc9c4d02017-03-08 10:56:02 -08004352 int arp_ip_count;
4353 int bond_mode = BOND_MODE_ROUNDROBIN;
4354 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4355 int lacp_fast = 0;
Nikolay Aleksandrovf13ad102017-09-12 15:10:05 +03004356 int tlb_dynamic_lb;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004357
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004358 /* Convert string parameters. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004359 if (mode) {
Nikolay Aleksandrov2b3798d2014-01-22 14:53:17 +01004360 bond_opt_initstr(&newval, mode);
4361 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4362 if (!valptr) {
4363 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004364 return -EINVAL;
4365 }
Nikolay Aleksandrov2b3798d2014-01-22 14:53:17 +01004366 bond_mode = valptr->value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367 }
4368
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004369 if (xmit_hash_policy) {
Debabrata Banerjeee79c1052018-05-14 14:48:09 -04004370 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4371 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4372 bond_mode == BOND_MODE_BROADCAST) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004373 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
Joe Perches90194262014-02-15 16:01:45 -08004374 bond_mode_name(bond_mode));
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004375 } else {
Nikolay Aleksandrova4b32ce2014-01-22 14:53:19 +01004376 bond_opt_initstr(&newval, xmit_hash_policy);
4377 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4378 &newval);
4379 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004380 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004381 xmit_hash_policy);
4382 return -EINVAL;
4383 }
Nikolay Aleksandrova4b32ce2014-01-22 14:53:19 +01004384 xmit_hashtype = valptr->value;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004385 }
4386 }
4387
Linus Torvalds1da177e2005-04-16 15:20:36 -07004388 if (lacp_rate) {
4389 if (bond_mode != BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004390 pr_info("lacp_rate param is irrelevant in mode %s\n",
4391 bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392 } else {
Nikolay Aleksandrovd3131de2014-01-22 14:53:27 +01004393 bond_opt_initstr(&newval, lacp_rate);
4394 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4395 &newval);
4396 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004397 pr_err("Error: Invalid lacp rate \"%s\"\n",
Nikolay Aleksandrovd3131de2014-01-22 14:53:27 +01004398 lacp_rate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004399 return -EINVAL;
4400 }
Nikolay Aleksandrovd3131de2014-01-22 14:53:27 +01004401 lacp_fast = valptr->value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004402 }
4403 }
4404
Jay Vosburghfd989c82008-11-04 17:51:16 -08004405 if (ad_select) {
Nikolay Aleksandrov548d28b2014-07-13 09:47:47 +02004406 bond_opt_initstr(&newval, ad_select);
Nikolay Aleksandrov9e5f5ee2014-01-22 14:53:29 +01004407 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4408 &newval);
4409 if (!valptr) {
4410 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
Jay Vosburghfd989c82008-11-04 17:51:16 -08004411 return -EINVAL;
4412 }
Nikolay Aleksandrov9e5f5ee2014-01-22 14:53:29 +01004413 params->ad_select = valptr->value;
4414 if (bond_mode != BOND_MODE_8023AD)
Joe Perches91565eb2014-02-15 15:57:04 -08004415 pr_warn("ad_select param only affects 802.3ad mode\n");
Jay Vosburghfd989c82008-11-04 17:51:16 -08004416 } else {
4417 params->ad_select = BOND_AD_STABLE;
4418 }
4419
Nicolas de Pesloüanf5841302009-08-28 13:18:34 +00004420 if (max_bonds < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004421 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4422 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004423 max_bonds = BOND_DEFAULT_MAX_BONDS;
4424 }
4425
4426 if (miimon < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004427 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4428 miimon, INT_MAX);
Nikolay Aleksandrovb98d9c62014-01-22 14:53:31 +01004429 miimon = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004430 }
4431
4432 if (updelay < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004433 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4434 updelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004435 updelay = 0;
4436 }
4437
4438 if (downdelay < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004439 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4440 downdelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441 downdelay = 0;
4442 }
4443
Debabrata Banerjeeb3c898e2018-05-16 14:02:13 -04004444 if ((use_carrier != 0) && (use_carrier != 1)) {
4445 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
Joe Perches91565eb2014-02-15 15:57:04 -08004446 use_carrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447 use_carrier = 1;
4448 }
4449
Ben Hutchingsad246c92011-04-26 15:25:52 +00004450 if (num_peer_notif < 0 || num_peer_notif > 255) {
Joe Perches91565eb2014-02-15 15:57:04 -08004451 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4452 num_peer_notif);
Ben Hutchingsad246c92011-04-26 15:25:52 +00004453 num_peer_notif = 1;
4454 }
4455
dingtianhong834db4b2013-12-21 14:40:17 +08004456 /* reset values for 802.3ad/TLB/ALB */
Veaceslav Falico267bed72014-05-15 21:39:53 +02004457 if (!bond_mode_uses_arp(bond_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004458 if (!miimon) {
Joe Perches91565eb2014-02-15 15:57:04 -08004459 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4460 pr_warn("Forcing miimon to 100msec\n");
dingtianhongfe9d04a2013-11-22 22:28:43 +08004461 miimon = BOND_DEFAULT_MIIMON;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004462 }
4463 }
4464
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004465 if (tx_queues < 1 || tx_queues > 255) {
Joe Perches91565eb2014-02-15 15:57:04 -08004466 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4467 tx_queues, BOND_DEFAULT_TX_QUEUES);
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004468 tx_queues = BOND_DEFAULT_TX_QUEUES;
4469 }
4470
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004471 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
Joe Perches91565eb2014-02-15 15:57:04 -08004472 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4473 all_slaves_active);
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004474 all_slaves_active = 0;
4475 }
4476
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004477 if (resend_igmp < 0 || resend_igmp > 255) {
Joe Perches91565eb2014-02-15 15:57:04 -08004478 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4479 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004480 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4481 }
4482
Nikolay Aleksandrovaa59d852014-01-22 14:53:18 +01004483 bond_opt_initval(&newval, packets_per_slave);
4484 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
Nikolay Aleksandrov73958322013-11-05 13:51:41 +01004485 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4486 packets_per_slave, USHRT_MAX);
4487 packets_per_slave = 1;
4488 }
4489
Linus Torvalds1da177e2005-04-16 15:20:36 -07004490 if (bond_mode == BOND_MODE_ALB) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004491 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",
4492 updelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004493 }
4494
4495 if (!miimon) {
4496 if (updelay || downdelay) {
4497 /* just warn the user the up/down delay will have
4498 * no effect since miimon is zero...
4499 */
Joe Perches91565eb2014-02-15 15:57:04 -08004500 pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4501 updelay, downdelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004502 }
4503 } else {
4504 /* don't allow arp monitoring */
4505 if (arp_interval) {
Joe Perches91565eb2014-02-15 15:57:04 -08004506 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4507 miimon, arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004508 arp_interval = 0;
4509 }
4510
4511 if ((updelay % miimon) != 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004512 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4513 updelay, miimon, (updelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004514 }
4515
4516 updelay /= miimon;
4517
4518 if ((downdelay % miimon) != 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004519 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4520 downdelay, miimon,
4521 (downdelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004522 }
4523
4524 downdelay /= miimon;
4525 }
4526
4527 if (arp_interval < 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004528 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4529 arp_interval, INT_MAX);
Nikolay Aleksandrov7bdb04e2014-01-22 14:53:23 +01004530 arp_interval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004531 }
4532
nikolay@redhat.com5a5c5fd2013-05-18 01:18:30 +00004533 for (arp_ip_count = 0, i = 0;
4534 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
dingtianhong89015c12013-12-04 18:59:31 +08004535 __be32 ip;
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004536
4537 /* not a complete check, but good enough to catch mistakes */
dingtianhong89015c12013-12-04 18:59:31 +08004538 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
Veaceslav Falico2807a9f2014-05-15 21:39:56 +02004539 !bond_is_ip_target_ok(ip)) {
Joe Perches91565eb2014-02-15 15:57:04 -08004540 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4541 arp_ip_target[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004542 arp_interval = 0;
4543 } else {
Veaceslav Falico0afee4e2013-06-24 11:49:30 +02004544 if (bond_get_targets_ip(arp_target, ip) == -1)
4545 arp_target[arp_ip_count++] = ip;
4546 else
Joe Perches91565eb2014-02-15 15:57:04 -08004547 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4548 &ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004549 }
4550 }
4551
4552 if (arp_interval && !arp_ip_count) {
4553 /* don't allow arping if no arp_ip_target given... */
Joe Perches91565eb2014-02-15 15:57:04 -08004554 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4555 arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004556 arp_interval = 0;
4557 }
4558
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004559 if (arp_validate) {
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004560 if (!arp_interval) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004561 pr_err("arp_validate requires arp_interval\n");
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004562 return -EINVAL;
4563 }
4564
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004565 bond_opt_initstr(&newval, arp_validate);
4566 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4567 &newval);
4568 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004569 pr_err("Error: invalid arp_validate \"%s\"\n",
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004570 arp_validate);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004571 return -EINVAL;
4572 }
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004573 arp_validate_value = valptr->value;
4574 } else {
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004575 arp_validate_value = 0;
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004576 }
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004577
Veaceslav Falico8599b522013-06-24 11:49:34 +02004578 if (arp_all_targets) {
Nikolay Aleksandrovedf36b22014-01-22 14:53:21 +01004579 bond_opt_initstr(&newval, arp_all_targets);
4580 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4581 &newval);
4582 if (!valptr) {
Veaceslav Falico8599b522013-06-24 11:49:34 +02004583 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4584 arp_all_targets);
4585 arp_all_targets_value = 0;
Nikolay Aleksandrovedf36b22014-01-22 14:53:21 +01004586 } else {
4587 arp_all_targets_value = valptr->value;
Veaceslav Falico8599b522013-06-24 11:49:34 +02004588 }
4589 }
4590
Linus Torvalds1da177e2005-04-16 15:20:36 -07004591 if (miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004592 pr_info("MII link monitoring set to %d ms\n", miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004593 } else if (arp_interval) {
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004594 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4595 arp_validate_value);
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004596 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
Nikolay Aleksandrov16228882014-01-22 14:53:20 +01004597 arp_interval, valptr->string, arp_ip_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004598
4599 for (i = 0; i < arp_ip_count; i++)
Joe Perches90194262014-02-15 16:01:45 -08004600 pr_cont(" %s", arp_ip_target[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004601
Joe Perches90194262014-02-15 16:01:45 -08004602 pr_cont("\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004603
Jay Vosburghb8a97872008-06-13 18:12:04 -07004604 } else if (max_bonds) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004605 /* miimon and arp_interval not set, we need one so things
4606 * work as expected, see bonding.txt for details
4607 */
Joe Perches90194262014-02-15 16:01:45 -08004608 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 -07004609 }
4610
Veaceslav Falicoec0865a2014-05-15 21:39:54 +02004611 if (primary && !bond_mode_uses_primary(bond_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004612 /* currently, using a primary only makes sense
4613 * in active backup, TLB or ALB modes
4614 */
Joe Perches91565eb2014-02-15 15:57:04 -08004615 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4616 primary, bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004617 primary = NULL;
4618 }
4619
Jiri Pirkoa5499522009-09-25 03:28:09 +00004620 if (primary && primary_reselect) {
Nikolay Aleksandrov388d3a62014-01-22 14:53:33 +01004621 bond_opt_initstr(&newval, primary_reselect);
4622 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4623 &newval);
4624 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004625 pr_err("Error: Invalid primary_reselect \"%s\"\n",
Nikolay Aleksandrov388d3a62014-01-22 14:53:33 +01004626 primary_reselect);
Jiri Pirkoa5499522009-09-25 03:28:09 +00004627 return -EINVAL;
4628 }
Nikolay Aleksandrov388d3a62014-01-22 14:53:33 +01004629 primary_reselect_value = valptr->value;
Jiri Pirkoa5499522009-09-25 03:28:09 +00004630 } else {
4631 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4632 }
4633
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004634 if (fail_over_mac) {
Nikolay Aleksandrov1df6b6a2014-01-22 14:53:22 +01004635 bond_opt_initstr(&newval, fail_over_mac);
4636 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4637 &newval);
4638 if (!valptr) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004639 pr_err("Error: invalid fail_over_mac \"%s\"\n",
Nikolay Aleksandrov1df6b6a2014-01-22 14:53:22 +01004640 fail_over_mac);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004641 return -EINVAL;
4642 }
Nikolay Aleksandrov1df6b6a2014-01-22 14:53:22 +01004643 fail_over_mac_value = valptr->value;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004644 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
Joe Perches91565eb2014-02-15 15:57:04 -08004645 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004646 } else {
4647 fail_over_mac_value = BOND_FOM_NONE;
4648 }
Jay Vosburghdd957c52007-10-09 19:57:24 -07004649
Mahesh Bandewar6791e462015-05-09 00:01:55 -07004650 bond_opt_initstr(&newval, "default");
4651 valptr = bond_opt_parse(
4652 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4653 &newval);
4654 if (!valptr) {
4655 pr_err("Error: No ad_actor_sys_prio default value");
4656 return -EINVAL;
4657 }
4658 ad_actor_sys_prio = valptr->value;
4659
Mahesh Bandeward22a5fc2015-05-09 00:01:57 -07004660 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4661 &newval);
4662 if (!valptr) {
4663 pr_err("Error: No ad_user_port_key default value");
4664 return -EINVAL;
4665 }
4666 ad_user_port_key = valptr->value;
4667
Nikolay Aleksandrovf13ad102017-09-12 15:10:05 +03004668 bond_opt_initstr(&newval, "default");
4669 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4670 if (!valptr) {
4671 pr_err("Error: No tlb_dynamic_lb default value");
4672 return -EINVAL;
Mahesh Bandewar8b426dc2017-03-08 10:55:56 -08004673 }
Nikolay Aleksandrovf13ad102017-09-12 15:10:05 +03004674 tlb_dynamic_lb = valptr->value;
Mahesh Bandewar8b426dc2017-03-08 10:55:56 -08004675
dingtianhong3a7129e2013-12-21 14:40:12 +08004676 if (lp_interval == 0) {
Joe Perches91565eb2014-02-15 15:57:04 -08004677 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4678 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
dingtianhong3a7129e2013-12-21 14:40:12 +08004679 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4680 }
4681
Linus Torvalds1da177e2005-04-16 15:20:36 -07004682 /* fill params struct with the proper values */
4683 params->mode = bond_mode;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004684 params->xmit_policy = xmit_hashtype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004685 params->miimon = miimon;
Ben Hutchingsad246c92011-04-26 15:25:52 +00004686 params->num_peer_notif = num_peer_notif;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004687 params->arp_interval = arp_interval;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004688 params->arp_validate = arp_validate_value;
Veaceslav Falico8599b522013-06-24 11:49:34 +02004689 params->arp_all_targets = arp_all_targets_value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004690 params->updelay = updelay;
4691 params->downdelay = downdelay;
4692 params->use_carrier = use_carrier;
4693 params->lacp_fast = lacp_fast;
4694 params->primary[0] = 0;
Jiri Pirkoa5499522009-09-25 03:28:09 +00004695 params->primary_reselect = primary_reselect_value;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004696 params->fail_over_mac = fail_over_mac_value;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004697 params->tx_queues = tx_queues;
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004698 params->all_slaves_active = all_slaves_active;
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004699 params->resend_igmp = resend_igmp;
stephen hemminger655f8912011-06-22 09:54:39 +00004700 params->min_links = min_links;
dingtianhong3a7129e2013-12-21 14:40:12 +08004701 params->lp_interval = lp_interval;
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01004702 params->packets_per_slave = packets_per_slave;
Mahesh Bandewar8b426dc2017-03-08 10:55:56 -08004703 params->tlb_dynamic_lb = tlb_dynamic_lb;
Mahesh Bandewar6791e462015-05-09 00:01:55 -07004704 params->ad_actor_sys_prio = ad_actor_sys_prio;
Mahesh Bandewar74514952015-05-09 00:01:56 -07004705 eth_zero_addr(params->ad_actor_system);
Mahesh Bandeward22a5fc2015-05-09 00:01:57 -07004706 params->ad_user_port_key = ad_user_port_key;
Hannes Frederic Sowa809fa972014-01-22 02:29:41 +01004707 if (packets_per_slave > 0) {
4708 params->reciprocal_packets_per_slave =
4709 reciprocal_value(packets_per_slave);
4710 } else {
4711 /* reciprocal_packets_per_slave is unused if
4712 * packets_per_slave is 0 or 1, just initialize it
4713 */
4714 params->reciprocal_packets_per_slave =
4715 (struct reciprocal_value) { 0 };
4716 }
4717
Linus Torvalds1da177e2005-04-16 15:20:36 -07004718 if (primary) {
4719 strncpy(params->primary, primary, IFNAMSIZ);
4720 params->primary[IFNAMSIZ - 1] = 0;
4721 }
4722
4723 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4724
4725 return 0;
4726}
4727
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004728/* Called from registration process */
Stephen Hemminger181470f2009-06-12 19:02:52 +00004729static int bond_init(struct net_device *bond_dev)
4730{
4731 struct bonding *bond = netdev_priv(bond_dev);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004732 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004733
Veaceslav Falico76444f52014-07-15 19:35:58 +02004734 netdev_dbg(bond_dev, "Begin bond_init\n");
Stephen Hemminger181470f2009-06-12 19:02:52 +00004735
Bhaktipriya Shridharf9f225e2016-08-30 22:02:01 +05304736 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004737 if (!bond->wq)
4738 return -ENOMEM;
4739
Eric Dumazet7e2556e2018-07-31 06:30:54 -07004740 bond->nest_level = SINGLE_DEPTH_NESTING;
Eric Dumazetd3fff6c2016-06-09 07:45:12 -07004741 netdev_lockdep_set_classes(bond_dev);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004742
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004743 list_add_tail(&bond->bond_list, &bn->dev_list);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004744
Eric W. Biederman6151b3d2009-10-29 14:18:22 +00004745 bond_prepare_sysfs_group(bond);
Jiri Pirko22bedad32010-04-01 21:22:57 +00004746
Taku Izumif073c7c2010-12-09 15:17:13 +00004747 bond_debug_register(bond);
4748
Jiri Pirko409cc1f2013-01-30 11:08:11 +01004749 /* Ensure valid dev_addr */
4750 if (is_zero_ether_addr(bond_dev->dev_addr) &&
nikolay@redhat.com97a1e632013-06-26 17:13:38 +02004751 bond_dev->addr_assign_type == NET_ADDR_PERM)
Jiri Pirko409cc1f2013-01-30 11:08:11 +01004752 eth_hw_addr_random(bond_dev);
Jiri Pirko409cc1f2013-01-30 11:08:11 +01004753
Stephen Hemminger181470f2009-06-12 19:02:52 +00004754 return 0;
4755}
4756
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004757unsigned int bond_get_num_tx_queues(void)
Jiri Pirkod5da4512011-08-10 06:09:44 +00004758{
stephen hemmingerefacb302012-04-10 18:34:43 +00004759 return tx_queues;
Jiri Pirkod5da4512011-08-10 06:09:44 +00004760}
4761
Mitch Williamsdfe60392005-11-09 10:36:04 -08004762/* Create a new bond based on the specified name and bonding parameters.
Jay Vosburghe4b91c42007-01-19 18:15:31 -08004763 * If name is NULL, obtain a suitable "bond%d" name for us.
Mitch Williamsdfe60392005-11-09 10:36:04 -08004764 * Caller must NOT hold rtnl_lock; we need to release it here before we
4765 * set up our sysfs entries.
4766 */
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004767int bond_create(struct net *net, const char *name)
Mitch Williamsdfe60392005-11-09 10:36:04 -08004768{
4769 struct net_device *bond_dev;
Paie913fb22015-04-29 14:24:23 -04004770 struct bonding *bond;
4771 struct alb_bond_info *bond_info;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004772 int res;
4773
4774 rtnl_lock();
Jay Vosburgh027ea042008-01-17 16:25:02 -08004775
Jiri Pirko1c5cae82011-04-30 01:21:32 +00004776 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
Tom Gundersenc835a672014-07-14 16:37:24 +02004777 name ? name : "bond%d", NET_NAME_UNKNOWN,
Jiri Pirko1c5cae82011-04-30 01:21:32 +00004778 bond_setup, tx_queues);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004779 if (!bond_dev) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004780 pr_err("%s: eek! can't alloc netdev!\n", name);
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004781 rtnl_unlock();
4782 return -ENOMEM;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004783 }
4784
Paie913fb22015-04-29 14:24:23 -04004785 /*
4786 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4787 * It is set to 0 by default which is wrong.
4788 */
4789 bond = netdev_priv(bond_dev);
4790 bond_info = &(BOND_ALB_INFO(bond));
4791 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4792
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004793 dev_net_set(bond_dev, net);
Eric W. Biederman88ead972009-10-29 14:18:25 +00004794 bond_dev->rtnl_link_ops = &bond_link_ops;
4795
Mitch Williamsdfe60392005-11-09 10:36:04 -08004796 res = register_netdevice(bond_dev);
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004797
Phil Oestere826eaf2011-03-14 06:22:05 +00004798 netif_carrier_off(bond_dev);
4799
Mahesh Bandewar4493b812017-03-08 10:55:54 -08004800 bond_work_init_all(bond);
4801
Mitch Williamsdfe60392005-11-09 10:36:04 -08004802 rtnl_unlock();
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004803 if (res < 0)
David S. Millercf124db2017-05-08 12:52:56 -04004804 free_netdev(bond_dev);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004805 return res;
4806}
4807
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004808static int __net_init bond_net_init(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004809{
Eric W. Biederman15449742009-11-29 15:46:04 +00004810 struct bond_net *bn = net_generic(net, bond_net_id);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004811
4812 bn->net = net;
4813 INIT_LIST_HEAD(&bn->dev_list);
4814
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004815 bond_create_proc_dir(bn);
Eric W. Biederman4c224002011-10-12 21:56:25 +00004816 bond_create_sysfs(bn);
Veaceslav Falico87a7b842013-06-24 11:49:29 +02004817
Eric W. Biederman15449742009-11-29 15:46:04 +00004818 return 0;
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004819}
4820
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004821static void __net_exit bond_net_exit(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004822{
Eric W. Biederman15449742009-11-29 15:46:04 +00004823 struct bond_net *bn = net_generic(net, bond_net_id);
nikolay@redhat.com69b02162013-04-06 00:54:38 +00004824 struct bonding *bond, *tmp_bond;
4825 LIST_HEAD(list);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004826
Eric W. Biederman4c224002011-10-12 21:56:25 +00004827 bond_destroy_sysfs(bn);
nikolay@redhat.com69b02162013-04-06 00:54:38 +00004828
4829 /* Kill off any bonds created after unregistering bond rtnl ops */
4830 rtnl_lock();
4831 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4832 unregister_netdevice_queue(bond->dev, &list);
4833 unregister_netdevice_many(&list);
4834 rtnl_unlock();
Veaceslav Falico23fa5c22014-07-17 12:04:08 +02004835
4836 bond_destroy_proc_dir(bn);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004837}
4838
4839static struct pernet_operations bond_net_ops = {
4840 .init = bond_net_init,
4841 .exit = bond_net_exit,
Eric W. Biederman15449742009-11-29 15:46:04 +00004842 .id = &bond_net_id,
4843 .size = sizeof(struct bond_net),
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004844};
4845
Linus Torvalds1da177e2005-04-16 15:20:36 -07004846static int __init bonding_init(void)
4847{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004848 int i;
4849 int res;
4850
Amerigo Wangbd33acc2011-03-06 21:58:46 +00004851 pr_info("%s", bond_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004852
Mitch Williamsdfe60392005-11-09 10:36:04 -08004853 res = bond_check_params(&bonding_defaults);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004854 if (res)
Mitch Williamsdfe60392005-11-09 10:36:04 -08004855 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004856
Eric W. Biederman15449742009-11-29 15:46:04 +00004857 res = register_pernet_subsys(&bond_net_ops);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004858 if (res)
4859 goto out;
Jay Vosburgh027ea042008-01-17 16:25:02 -08004860
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004861 res = bond_netlink_init();
Eric W. Biederman88ead972009-10-29 14:18:25 +00004862 if (res)
Eric W. Biederman66391042009-10-29 23:58:54 +00004863 goto err_link;
Eric W. Biederman88ead972009-10-29 14:18:25 +00004864
Taku Izumif073c7c2010-12-09 15:17:13 +00004865 bond_create_debugfs();
4866
Linus Torvalds1da177e2005-04-16 15:20:36 -07004867 for (i = 0; i < max_bonds; i++) {
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004868 res = bond_create(&init_net, NULL);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004869 if (res)
4870 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004871 }
4872
Linus Torvalds1da177e2005-04-16 15:20:36 -07004873 register_netdevice_notifier(&bond_netdev_notifier);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004874out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004875 return res;
Eric W. Biederman88ead972009-10-29 14:18:25 +00004876err:
Thomas Richterdb298682014-04-09 12:52:59 +02004877 bond_destroy_debugfs();
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004878 bond_netlink_fini();
Eric W. Biederman66391042009-10-29 23:58:54 +00004879err_link:
Eric W. Biederman15449742009-11-29 15:46:04 +00004880 unregister_pernet_subsys(&bond_net_ops);
Eric W. Biederman88ead972009-10-29 14:18:25 +00004881 goto out;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004882
Linus Torvalds1da177e2005-04-16 15:20:36 -07004883}
4884
4885static void __exit bonding_exit(void)
4886{
4887 unregister_netdevice_notifier(&bond_netdev_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004888
Taku Izumif073c7c2010-12-09 15:17:13 +00004889 bond_destroy_debugfs();
Pavel Emelyanovae68c392008-05-02 17:49:39 -07004890
Jiri Pirko0a2a78c2013-10-18 17:43:33 +02004891 bond_netlink_fini();
nikolay@redhat.comffcdedb2013-04-06 00:54:37 +00004892 unregister_pernet_subsys(&bond_net_ops);
Neil Hormane843fa52010-10-13 16:01:50 +00004893
4894#ifdef CONFIG_NET_POLL_CONTROLLER
Nikolay Aleksandrov547942c2014-09-15 17:19:34 +02004895 /* Make sure we don't have an imbalance on our netpoll blocking */
Neil Hormanfb4fa762010-12-06 09:05:50 +00004896 WARN_ON(atomic_read(&netpoll_block_tx));
Neil Hormane843fa52010-10-13 16:01:50 +00004897#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004898}
4899
4900module_init(bonding_init);
4901module_exit(bonding_exit);
4902MODULE_LICENSE("GPL");
4903MODULE_VERSION(DRV_VERSION);
4904MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4905MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");