blob: 4df674bc6f125973169fd493151304739e04aebd [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 */
33
Joe Perchesa4aee5c2009-12-13 20:06:07 -080034#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/kernel.h>
37#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040044#include <net/ip.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/ip.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040046#include <linux/tcp.h>
47#include <linux/udp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000056#include <linux/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <asm/system.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <asm/dma.h>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000059#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/errno.h>
61#include <linux/netdevice.h>
62#include <linux/inetdevice.h>
Jay Vosburgha816c7c2007-02-28 17:03:37 -080063#include <linux/igmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include <linux/etherdevice.h>
65#include <linux/skbuff.h>
66#include <net/sock.h>
67#include <linux/rtnetlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <linux/smp.h>
69#include <linux/if_ether.h>
70#include <net/arp.h>
71#include <linux/mii.h>
72#include <linux/ethtool.h>
73#include <linux/if_vlan.h>
74#include <linux/if_bonding.h>
David Sterbab63bb732007-12-06 23:40:33 -080075#include <linux/jiffies.h>
Neil Hormane843fa52010-10-13 16:01:50 +000076#include <linux/preempt.h>
Jay Vosburghc3ade5c2005-06-26 17:52:20 -040077#include <net/route.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020078#include <net/net_namespace.h>
Eric W. Biedermanec87fd32009-10-29 14:18:26 +000079#include <net/netns/generic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#include "bonding.h"
81#include "bond_3ad.h"
82#include "bond_alb.h"
83
84/*---------------------------- Module parameters ----------------------------*/
85
86/* monitor all links that often (in milliseconds). <=0 disables monitoring */
87#define BOND_LINK_MON_INTERV 0
88#define BOND_LINK_ARP_INTERV 0
89
90static int max_bonds = BOND_DEFAULT_MAX_BONDS;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +000091static int tx_queues = BOND_DEFAULT_TX_QUEUES;
Moni Shoua7893b242008-05-17 21:10:12 -070092static int num_grat_arp = 1;
Brian Haley305d5522008-11-04 17:51:14 -080093static int num_unsol_na = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094static int miimon = BOND_LINK_MON_INTERV;
Stephen Hemminger3d632c32009-06-12 19:02:48 +000095static int updelay;
96static int downdelay;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097static int use_carrier = 1;
Stephen Hemminger3d632c32009-06-12 19:02:48 +000098static char *mode;
99static char *primary;
Jiri Pirkoa5499522009-09-25 03:28:09 +0000100static char *primary_reselect;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000101static char *lacp_rate;
102static char *ad_select;
103static char *xmit_hash_policy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104static int arp_interval = BOND_LINK_ARP_INTERV;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000105static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106static char *arp_validate;
107static char *fail_over_mac;
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000108static int all_slaves_active = 0;
Stephen Hemmingerd2991f72009-06-12 19:02:44 +0000109static struct bond_params bonding_defaults;
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000110static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
112module_param(max_bonds, int, 0);
113MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
Andy Gospodarekbb1d9122010-06-02 08:40:18 +0000114module_param(tx_queues, int, 0);
115MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
Moni Shoua7893b242008-05-17 21:10:12 -0700116module_param(num_grat_arp, int, 0644);
117MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
Brian Haley305d5522008-11-04 17:51:14 -0800118module_param(num_unsol_na, int, 0644);
119MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120module_param(miimon, int, 0);
121MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
122module_param(updelay, int, 0);
123MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
124module_param(downdelay, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800125MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
126 "in milliseconds");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127module_param(use_carrier, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800128MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
129 "0 for off, 1 for on (default)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130module_param(mode, charp, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800131MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
132 "1 for active-backup, 2 for balance-xor, "
133 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
134 "6 for balance-alb");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135module_param(primary, charp, 0);
136MODULE_PARM_DESC(primary, "Primary network device to use");
Jiri Pirkoa5499522009-09-25 03:28:09 +0000137module_param(primary_reselect, charp, 0);
138MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
139 "once it comes up; "
140 "0 for always (default), "
141 "1 for only if speed of primary is "
142 "better, "
143 "2 for only on active slave "
144 "failure");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145module_param(lacp_rate, charp, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800146MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
147 "(slow/fast)");
Jay Vosburghfd989c82008-11-04 17:51:16 -0800148module_param(ad_select, charp, 0);
149MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400150module_param(xmit_hash_policy, charp, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800151MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
152 ", 1 for layer 3+4");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153module_param(arp_interval, int, 0);
154MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
155module_param_array(arp_ip_target, charp, NULL, 0);
156MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700157module_param(arp_validate, charp, 0);
158MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700159module_param(fail_over_mac, charp, 0);
160MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000161module_param(all_slaves_active, int, 0);
162MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
163 "by setting active flag for all slaves. "
164 "0 for never (default), 1 for always.");
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000165module_param(resend_igmp, int, 0);
166MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
168/*----------------------------- Global variables ----------------------------*/
169
Neil Hormane843fa52010-10-13 16:01:50 +0000170#ifdef CONFIG_NET_POLL_CONTROLLER
Neil Hormanfb4fa762010-12-06 09:05:50 +0000171atomic_t netpoll_block_tx = ATOMIC_INIT(0);
Neil Hormane843fa52010-10-13 16:01:50 +0000172#endif
173
Eric Dumazetf99189b2009-11-17 10:42:49 +0000174int bond_net_id __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000176static __be32 arp_target[BOND_MAX_ARP_TARGETS];
177static int arp_ip_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178static int bond_mode = BOND_MODE_ROUNDROBIN;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000179static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
180static int lacp_fast;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800182const struct bond_parm_tbl bond_lacp_tbl[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{ "slow", AD_LACP_SLOW},
184{ "fast", AD_LACP_FAST},
185{ NULL, -1},
186};
187
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800188const struct bond_parm_tbl bond_mode_tbl[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{ "balance-rr", BOND_MODE_ROUNDROBIN},
190{ "active-backup", BOND_MODE_ACTIVEBACKUP},
191{ "balance-xor", BOND_MODE_XOR},
192{ "broadcast", BOND_MODE_BROADCAST},
193{ "802.3ad", BOND_MODE_8023AD},
194{ "balance-tlb", BOND_MODE_TLB},
195{ "balance-alb", BOND_MODE_ALB},
196{ NULL, -1},
197};
198
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800199const struct bond_parm_tbl xmit_hashtype_tbl[] = {
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400200{ "layer2", BOND_XMIT_POLICY_LAYER2},
201{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
Jay Vosburgh6f6652b2007-12-06 23:40:34 -0800202{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400203{ NULL, -1},
204};
205
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800206const struct bond_parm_tbl arp_validate_tbl[] = {
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700207{ "none", BOND_ARP_VALIDATE_NONE},
208{ "active", BOND_ARP_VALIDATE_ACTIVE},
209{ "backup", BOND_ARP_VALIDATE_BACKUP},
210{ "all", BOND_ARP_VALIDATE_ALL},
211{ NULL, -1},
212};
213
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800214const struct bond_parm_tbl fail_over_mac_tbl[] = {
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700215{ "none", BOND_FOM_NONE},
216{ "active", BOND_FOM_ACTIVE},
217{ "follow", BOND_FOM_FOLLOW},
218{ NULL, -1},
219};
220
Jiri Pirkoa5499522009-09-25 03:28:09 +0000221const struct bond_parm_tbl pri_reselect_tbl[] = {
222{ "always", BOND_PRI_RESELECT_ALWAYS},
223{ "better", BOND_PRI_RESELECT_BETTER},
224{ "failure", BOND_PRI_RESELECT_FAILURE},
225{ NULL, -1},
226};
227
Jay Vosburghfd989c82008-11-04 17:51:16 -0800228struct bond_parm_tbl ad_select_tbl[] = {
229{ "stable", BOND_AD_STABLE},
230{ "bandwidth", BOND_AD_BANDWIDTH},
231{ "count", BOND_AD_COUNT},
232{ NULL, -1},
233};
234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235/*-------------------------- Forward declarations ---------------------------*/
236
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400237static void bond_send_gratuitous_arp(struct bonding *bond);
Stephen Hemminger181470f2009-06-12 19:02:52 +0000238static int bond_init(struct net_device *bond_dev);
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +0000239static void bond_uninit(struct net_device *bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
241/*---------------------------- General routines -----------------------------*/
242
Amerigo Wangbd33acc2011-03-06 21:58:46 +0000243const char *bond_mode_name(int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800245 static const char *names[] = {
246 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
247 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
248 [BOND_MODE_XOR] = "load balancing (xor)",
249 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000250 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800251 [BOND_MODE_TLB] = "transmit load balancing",
252 [BOND_MODE_ALB] = "adaptive load balancing",
253 };
254
255 if (mode < 0 || mode > BOND_MODE_ALB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 return "unknown";
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800257
258 return names[mode];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260
261/*---------------------------------- VLAN -----------------------------------*/
262
263/**
264 * bond_add_vlan - add a new vlan id on bond
265 * @bond: bond that got the notification
266 * @vlan_id: the vlan id to add
267 *
268 * Returns -ENOMEM if allocation failed.
269 */
270static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
271{
272 struct vlan_entry *vlan;
273
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -0800274 pr_debug("bond: %s, vlan id %d\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800275 (bond ? bond->dev->name : "None"), vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
Brian Haley305d5522008-11-04 17:51:14 -0800277 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000278 if (!vlan)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281 INIT_LIST_HEAD(&vlan->vlan_list);
282 vlan->vlan_id = vlan_id;
283
284 write_lock_bh(&bond->lock);
285
286 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
287
288 write_unlock_bh(&bond->lock);
289
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -0800290 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
292 return 0;
293}
294
295/**
296 * bond_del_vlan - delete a vlan id from bond
297 * @bond: bond that got the notification
298 * @vlan_id: the vlan id to delete
299 *
300 * returns -ENODEV if @vlan_id was not found in @bond.
301 */
302static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
303{
Pavel Emelyanov0883bec2008-05-17 21:10:10 -0700304 struct vlan_entry *vlan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 int res = -ENODEV;
306
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -0800307 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Neil Hormane843fa52010-10-13 16:01:50 +0000309 block_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 write_lock_bh(&bond->lock);
311
Pavel Emelyanov0883bec2008-05-17 21:10:10 -0700312 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 if (vlan->vlan_id == vlan_id) {
314 list_del(&vlan->vlan_list);
315
Holger Eitzenberger58402052008-12-09 23:07:13 -0800316 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 bond_alb_clear_vlan(bond, vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800319 pr_debug("removed VLAN ID %d from bond %s\n",
320 vlan_id, bond->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
322 kfree(vlan);
323
324 if (list_empty(&bond->vlan_list) &&
325 (bond->slave_cnt == 0)) {
326 /* Last VLAN removed and no slaves, so
327 * restore block on adding VLANs. This will
328 * be removed once new slaves that are not
329 * VLAN challenged will be added.
330 */
331 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
332 }
333
334 res = 0;
335 goto out;
336 }
337 }
338
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800339 pr_debug("couldn't find VLAN ID %d in bond %s\n",
340 vlan_id, bond->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
342out:
343 write_unlock_bh(&bond->lock);
Neil Hormane843fa52010-10-13 16:01:50 +0000344 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 return res;
346}
347
348/**
349 * bond_has_challenged_slaves
350 * @bond: the bond we're working on
351 *
352 * Searches the slave list. Returns 1 if a vlan challenged slave
353 * was found, 0 otherwise.
354 *
355 * Assumes bond->lock is held.
356 */
357static int bond_has_challenged_slaves(struct bonding *bond)
358{
359 struct slave *slave;
360 int i;
361
362 bond_for_each_slave(bond, slave, i) {
363 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -0800364 pr_debug("found VLAN challenged slave - %s\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800365 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 return 1;
367 }
368 }
369
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -0800370 pr_debug("no VLAN challenged slaves found\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 return 0;
372}
373
374/**
375 * bond_next_vlan - safely skip to the next item in the vlans list.
376 * @bond: the bond we're working on
377 * @curr: item we're advancing from
378 *
379 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
380 * or @curr->next otherwise (even if it is @curr itself again).
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000381 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 * Caller must hold bond->lock
383 */
384struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
385{
386 struct vlan_entry *next, *last;
387
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000388 if (list_empty(&bond->vlan_list))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
391 if (!curr) {
392 next = list_entry(bond->vlan_list.next,
393 struct vlan_entry, vlan_list);
394 } else {
395 last = list_entry(bond->vlan_list.prev,
396 struct vlan_entry, vlan_list);
397 if (last == curr) {
398 next = list_entry(bond->vlan_list.next,
399 struct vlan_entry, vlan_list);
400 } else {
401 next = list_entry(curr->vlan_list.next,
402 struct vlan_entry, vlan_list);
403 }
404 }
405
406 return next;
407}
408
409/**
410 * bond_dev_queue_xmit - Prepare skb for xmit.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000411 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 * @bond: bond device that got this skb for tx.
413 * @skb: hw accel VLAN tagged skb to transmit
414 * @slave_dev: slave that is supposed to xmit this skbuff
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000416int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
417 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
Ben Hutchings83874512010-12-13 08:19:28 +0000419 skb->dev = slave_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 skb->priority = 1;
Amerigo Wang080e4132011-02-17 23:43:33 +0000421 if (unlikely(netpoll_tx_running(slave_dev)))
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000422 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
Amerigo Wang080e4132011-02-17 23:43:33 +0000423 else
WANG Congf6dc31a2010-05-06 00:48:51 -0700424 dev_queue_xmit(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426 return 0;
427}
428
429/*
430 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
431 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
432 * lock because:
433 * a. This operation is performed in IOCTL context,
434 * b. The operation is protected by the RTNL semaphore in the 8021q code,
435 * c. Holding a lock with BH disabled while directly calling a base driver
436 * entry point is generally a BAD idea.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000437 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 * The design of synchronization/protection for this operation in the 8021q
439 * module is good for one or more VLAN devices over a single physical device
440 * and cannot be extended for a teaming solution like bonding, so there is a
441 * potential race condition here where a net device from the vlan group might
442 * be referenced (either by a base driver or the 8021q code) while it is being
443 * removed from the system. However, it turns out we're not making matters
444 * worse, and if it works for regular VLAN usage it will work here too.
445*/
446
447/**
448 * bond_vlan_rx_register - Propagates registration to slaves
449 * @bond_dev: bonding net device that got called
450 * @grp: vlan group being registered
451 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000452static void bond_vlan_rx_register(struct net_device *bond_dev,
453 struct vlan_group *grp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
Wang Chen454d7c92008-11-12 23:37:49 -0800455 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 struct slave *slave;
457 int i;
458
Jarek Poplawskia71fb882010-10-27 07:08:22 +0000459 write_lock_bh(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 bond->vlgrp = grp;
Jarek Poplawskia71fb882010-10-27 07:08:22 +0000461 write_unlock_bh(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463 bond_for_each_slave(bond, slave, i) {
464 struct net_device *slave_dev = slave->dev;
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800465 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
467 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800468 slave_ops->ndo_vlan_rx_register) {
469 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 }
471 }
472}
473
474/**
475 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
476 * @bond_dev: bonding net device that got called
477 * @vid: vlan id being added
478 */
479static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
480{
Wang Chen454d7c92008-11-12 23:37:49 -0800481 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 struct slave *slave;
483 int i, res;
484
485 bond_for_each_slave(bond, slave, i) {
486 struct net_device *slave_dev = slave->dev;
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800487 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
489 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800490 slave_ops->ndo_vlan_rx_add_vid) {
491 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 }
493 }
494
495 res = bond_add_vlan(bond, vid);
496 if (res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800497 pr_err("%s: Error: Failed to add vlan id %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 bond_dev->name, vid);
499 }
500}
501
502/**
503 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
504 * @bond_dev: bonding net device that got called
505 * @vid: vlan id being removed
506 */
507static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
508{
Wang Chen454d7c92008-11-12 23:37:49 -0800509 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 struct slave *slave;
511 struct net_device *vlan_dev;
512 int i, res;
513
514 bond_for_each_slave(bond, slave, i) {
515 struct net_device *slave_dev = slave->dev;
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800516 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
518 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800519 slave_ops->ndo_vlan_rx_kill_vid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 /* Save and then restore vlan_dev in the grp array,
521 * since the slave's driver might clear it.
522 */
Dan Aloni5c15bde2007-03-02 20:44:51 -0800523 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800524 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
Dan Aloni5c15bde2007-03-02 20:44:51 -0800525 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 }
527 }
528
529 res = bond_del_vlan(bond, vid);
530 if (res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800531 pr_err("%s: Error: Failed to remove vlan id %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 bond_dev->name, vid);
533 }
534}
535
536static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
537{
538 struct vlan_entry *vlan;
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800539 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Jay Vosburghf35188f2010-07-21 12:14:47 +0000541 if (!bond->vlgrp)
Jay Vosburgh03dc2f42010-07-21 12:14:48 +0000542 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
544 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800545 slave_ops->ndo_vlan_rx_register)
546 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
548 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800549 !(slave_ops->ndo_vlan_rx_add_vid))
Jay Vosburgh03dc2f42010-07-21 12:14:48 +0000550 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800552 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
553 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554}
555
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000556static void bond_del_vlans_from_slave(struct bonding *bond,
557 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558{
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800559 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 struct vlan_entry *vlan;
561 struct net_device *vlan_dev;
562
Jay Vosburghf35188f2010-07-21 12:14:47 +0000563 if (!bond->vlgrp)
Jay Vosburgh03dc2f42010-07-21 12:14:48 +0000564 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
566 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800567 !(slave_ops->ndo_vlan_rx_kill_vid))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 goto unreg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
570 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Jay Vosburghf35188f2010-07-21 12:14:47 +0000571 if (!vlan->vlan_id)
572 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 /* Save and then restore vlan_dev in the grp array,
574 * since the slave's driver might clear it.
575 */
Dan Aloni5c15bde2007-03-02 20:44:51 -0800576 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800577 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
Dan Aloni5c15bde2007-03-02 20:44:51 -0800578 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
580
581unreg:
582 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800583 slave_ops->ndo_vlan_rx_register)
584 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585}
586
587/*------------------------------- Link status -------------------------------*/
588
589/*
Jay Vosburghff59c452006-03-27 13:27:43 -0800590 * Set the carrier state for the master according to the state of its
591 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
592 * do special 802.3ad magic.
593 *
594 * Returns zero if carrier state does not change, nonzero if it does.
595 */
596static int bond_set_carrier(struct bonding *bond)
597{
598 struct slave *slave;
599 int i;
600
601 if (bond->slave_cnt == 0)
602 goto down;
603
604 if (bond->params.mode == BOND_MODE_8023AD)
605 return bond_3ad_set_carrier(bond);
606
607 bond_for_each_slave(bond, slave, i) {
608 if (slave->link == BOND_LINK_UP) {
609 if (!netif_carrier_ok(bond->dev)) {
610 netif_carrier_on(bond->dev);
611 return 1;
612 }
613 return 0;
614 }
615 }
616
617down:
618 if (netif_carrier_ok(bond->dev)) {
619 netif_carrier_off(bond->dev);
620 return 1;
621 }
622 return 0;
623}
624
625/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 * Get link speed and duplex from the slave's base driver
627 * using ethtool. If for some reason the call fails or the
628 * values are invalid, fake speed and duplex to 100/Full
629 * and return error.
630 */
631static int bond_update_speed_duplex(struct slave *slave)
632{
633 struct net_device *slave_dev = slave->dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 struct ethtool_cmd etool;
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700635 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
637 /* Fake speed and duplex */
638 slave->speed = SPEED_100;
639 slave->duplex = DUPLEX_FULL;
640
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700641 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700644 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
645 if (res < 0)
646 return -1;
647
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 switch (etool.speed) {
649 case SPEED_10:
650 case SPEED_100:
651 case SPEED_1000:
Jay Vosburgh94dbffd2006-09-22 21:52:15 -0700652 case SPEED_10000:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 break;
654 default:
655 return -1;
656 }
657
658 switch (etool.duplex) {
659 case DUPLEX_FULL:
660 case DUPLEX_HALF:
661 break;
662 default:
663 return -1;
664 }
665
666 slave->speed = etool.speed;
667 slave->duplex = etool.duplex;
668
669 return 0;
670}
671
672/*
673 * if <dev> supports MII link status reporting, check its link status.
674 *
675 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000676 * depending upon the setting of the use_carrier parameter.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 *
678 * Return either BMSR_LSTATUS, meaning that the link is up (or we
679 * can't tell and just pretend it is), or 0, meaning that the link is
680 * down.
681 *
682 * If reporting is non-zero, instead of faking link up, return -1 if
683 * both ETHTOOL and MII ioctls fail (meaning the device does not
684 * support them). If use_carrier is set, return whatever it says.
685 * It'd be nice if there was a good way to tell if a driver supports
686 * netif_carrier, but there really isn't.
687 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000688static int bond_check_dev_link(struct bonding *bond,
689 struct net_device *slave_dev, int reporting)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800691 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Jiri Bohacd9d52832009-10-28 22:23:54 -0700692 int (*ioctl)(struct net_device *, struct ifreq *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 struct ifreq ifr;
694 struct mii_ioctl_data *mii;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Petri Gynther6c988852009-08-28 12:05:15 +0000696 if (!reporting && !netif_running(slave_dev))
697 return 0;
698
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800699 if (bond->params.use_carrier)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
Jiri Pirko29112f42009-04-24 01:58:23 +0000702 /* Try to get link status using Ethtool first. */
703 if (slave_dev->ethtool_ops) {
704 if (slave_dev->ethtool_ops->get_link) {
705 u32 link;
706
707 link = slave_dev->ethtool_ops->get_link(slave_dev);
708
709 return link ? BMSR_LSTATUS : 0;
710 }
711 }
712
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000713 /* Ethtool can't be used, fallback to MII ioctls. */
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800714 ioctl = slave_ops->ndo_do_ioctl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 if (ioctl) {
716 /* TODO: set pointer to correct ioctl on a per team member */
717 /* bases to make this more efficient. that is, once */
718 /* we determine the correct ioctl, we will always */
719 /* call it and not the others for that team */
720 /* member. */
721
722 /*
723 * We cannot assume that SIOCGMIIPHY will also read a
724 * register; not all network drivers (e.g., e100)
725 * support that.
726 */
727
728 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
729 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
730 mii = if_mii(&ifr);
731 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
732 mii->reg_num = MII_BMSR;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000733 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
734 return mii->val_out & BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 }
736 }
737
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700738 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 * If reporting, report that either there's no dev->do_ioctl,
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700740 * or both SIOCGMIIREG and get_link failed (meaning that we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 * cannot report link status). If not reporting, pretend
742 * we're ok.
743 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000744 return reporting ? -1 : BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745}
746
747/*----------------------------- Multicast list ------------------------------*/
748
749/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 * Push the promiscuity flag down to appropriate slaves
751 */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700752static int bond_set_promiscuity(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753{
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700754 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 if (USES_PRIMARY(bond->params.mode)) {
756 /* write lock already acquired */
757 if (bond->curr_active_slave) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700758 err = dev_set_promiscuity(bond->curr_active_slave->dev,
759 inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
761 } else {
762 struct slave *slave;
763 int i;
764 bond_for_each_slave(bond, slave, i) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700765 err = dev_set_promiscuity(slave->dev, inc);
766 if (err)
767 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 }
769 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700770 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771}
772
773/*
774 * Push the allmulti flag down to all slaves
775 */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700776static int bond_set_allmulti(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700778 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 if (USES_PRIMARY(bond->params.mode)) {
780 /* write lock already acquired */
781 if (bond->curr_active_slave) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700782 err = dev_set_allmulti(bond->curr_active_slave->dev,
783 inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 }
785 } else {
786 struct slave *slave;
787 int i;
788 bond_for_each_slave(bond, slave, i) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700789 err = dev_set_allmulti(slave->dev, inc);
790 if (err)
791 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 }
793 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700794 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795}
796
797/*
798 * Add a Multicast address to slaves
799 * according to mode
800 */
Jiri Pirko22bedad32010-04-01 21:22:57 +0000801static void bond_mc_add(struct bonding *bond, void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
803 if (USES_PRIMARY(bond->params.mode)) {
804 /* write lock already acquired */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000805 if (bond->curr_active_slave)
Jiri Pirko22bedad32010-04-01 21:22:57 +0000806 dev_mc_add(bond->curr_active_slave->dev, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 } else {
808 struct slave *slave;
809 int i;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000810
811 bond_for_each_slave(bond, slave, i)
Jiri Pirko22bedad32010-04-01 21:22:57 +0000812 dev_mc_add(slave->dev, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 }
814}
815
816/*
817 * Remove a multicast address from slave
818 * according to mode
819 */
Jiri Pirko22bedad32010-04-01 21:22:57 +0000820static void bond_mc_del(struct bonding *bond, void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821{
822 if (USES_PRIMARY(bond->params.mode)) {
823 /* write lock already acquired */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000824 if (bond->curr_active_slave)
Jiri Pirko22bedad32010-04-01 21:22:57 +0000825 dev_mc_del(bond->curr_active_slave->dev, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 } else {
827 struct slave *slave;
828 int i;
829 bond_for_each_slave(bond, slave, i) {
Jiri Pirko22bedad32010-04-01 21:22:57 +0000830 dev_mc_del(slave->dev, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 }
832 }
833}
834
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800835
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000836static void __bond_resend_igmp_join_requests(struct net_device *dev)
837{
838 struct in_device *in_dev;
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000839
840 rcu_read_lock();
841 in_dev = __in_dev_get_rcu(dev);
Eric Dumazet866f3b22010-11-18 09:33:19 -0800842 if (in_dev)
David S. Miller24912422010-11-19 13:13:47 -0800843 ip_mc_rejoin_groups(in_dev);
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000844 rcu_read_unlock();
845}
846
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800847/*
848 * Retrieve the list of registered multicast addresses for the bonding
849 * device and retransmit an IGMP JOIN request to the current active
850 * slave.
851 */
852static void bond_resend_igmp_join_requests(struct bonding *bond)
853{
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000854 struct net_device *vlan_dev;
855 struct vlan_entry *vlan;
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800856
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000857 read_lock(&bond->lock);
858
859 /* rejoin all groups on bond device */
860 __bond_resend_igmp_join_requests(bond->dev);
861
862 /* rejoin all groups on vlan devices */
863 if (bond->vlgrp) {
864 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
865 vlan_dev = vlan_group_get_device(bond->vlgrp,
866 vlan->vlan_id);
867 if (vlan_dev)
868 __bond_resend_igmp_join_requests(vlan_dev);
869 }
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800870 }
871
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000872 if (--bond->igmp_retrans > 0)
873 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
874
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000875 read_unlock(&bond->lock);
876}
877
stephen hemminger379b7382010-10-15 11:02:56 +0000878static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000879{
880 struct bonding *bond = container_of(work, struct bonding,
881 mcast_work.work);
882 bond_resend_igmp_join_requests(bond);
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800883}
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 * flush all members of flush->mc_list from device dev->mc_list
887 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000888static void bond_mc_list_flush(struct net_device *bond_dev,
889 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
Wang Chen454d7c92008-11-12 23:37:49 -0800891 struct bonding *bond = netdev_priv(bond_dev);
Jiri Pirko22bedad32010-04-01 21:22:57 +0000892 struct netdev_hw_addr *ha;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
Jiri Pirko22bedad32010-04-01 21:22:57 +0000894 netdev_for_each_mc_addr(ha, bond_dev)
895 dev_mc_del(slave_dev, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 if (bond->params.mode == BOND_MODE_8023AD) {
898 /* del lacpdu mc addr from mc list */
899 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
900
Jiri Pirko22bedad32010-04-01 21:22:57 +0000901 dev_mc_del(slave_dev, lacpdu_multicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
903}
904
905/*--------------------------- Active slave change ---------------------------*/
906
907/*
908 * Update the mc list and multicast-related flags for the new and
909 * old active slaves (if any) according to the multicast mode, and
910 * promiscuous flags unconditionally.
911 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000912static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
913 struct slave *old_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914{
Jiri Pirko22bedad32010-04-01 21:22:57 +0000915 struct netdev_hw_addr *ha;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000917 if (!USES_PRIMARY(bond->params.mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 /* nothing to do - mc list is already up-to-date on
919 * all slaves
920 */
921 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 if (old_active) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000924 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 dev_set_promiscuity(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000927 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 dev_set_allmulti(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
Jiri Pirko22bedad32010-04-01 21:22:57 +0000930 netdev_for_each_mc_addr(ha, bond->dev)
931 dev_mc_del(old_active->dev, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 }
933
934 if (new_active) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700935 /* FIXME: Signal errors upstream. */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000936 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 dev_set_promiscuity(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000939 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 dev_set_allmulti(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Jiri Pirko22bedad32010-04-01 21:22:57 +0000942 netdev_for_each_mc_addr(ha, bond->dev)
943 dev_mc_add(new_active->dev, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 }
945}
946
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700947/*
948 * bond_do_fail_over_mac
949 *
950 * Perform special MAC address swapping for fail_over_mac settings
951 *
952 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
953 */
954static void bond_do_fail_over_mac(struct bonding *bond,
955 struct slave *new_active,
956 struct slave *old_active)
Hannes Eder1f78d9f2009-02-14 11:15:33 +0000957 __releases(&bond->curr_slave_lock)
958 __releases(&bond->lock)
959 __acquires(&bond->lock)
960 __acquires(&bond->curr_slave_lock)
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700961{
962 u8 tmp_mac[ETH_ALEN];
963 struct sockaddr saddr;
964 int rv;
965
966 switch (bond->params.fail_over_mac) {
967 case BOND_FOM_ACTIVE:
968 if (new_active)
969 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
970 new_active->dev->addr_len);
971 break;
972 case BOND_FOM_FOLLOW:
973 /*
974 * if new_active && old_active, swap them
975 * if just old_active, do nothing (going to no active slave)
976 * if just new_active, set new_active to bond's MAC
977 */
978 if (!new_active)
979 return;
980
981 write_unlock_bh(&bond->curr_slave_lock);
982 read_unlock(&bond->lock);
983
984 if (old_active) {
985 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
986 memcpy(saddr.sa_data, old_active->dev->dev_addr,
987 ETH_ALEN);
988 saddr.sa_family = new_active->dev->type;
989 } else {
990 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
991 saddr.sa_family = bond->dev->type;
992 }
993
994 rv = dev_set_mac_address(new_active->dev, &saddr);
995 if (rv) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800996 pr_err("%s: Error %d setting MAC of slave %s\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700997 bond->dev->name, -rv, new_active->dev->name);
998 goto out;
999 }
1000
1001 if (!old_active)
1002 goto out;
1003
1004 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1005 saddr.sa_family = old_active->dev->type;
1006
1007 rv = dev_set_mac_address(old_active->dev, &saddr);
1008 if (rv)
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001009 pr_err("%s: Error %d setting MAC of slave %s\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001010 bond->dev->name, -rv, new_active->dev->name);
1011out:
1012 read_lock(&bond->lock);
1013 write_lock_bh(&bond->curr_slave_lock);
1014 break;
1015 default:
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001016 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001017 bond->dev->name, bond->params.fail_over_mac);
1018 break;
1019 }
1020
1021}
1022
Jiri Pirkoa5499522009-09-25 03:28:09 +00001023static bool bond_should_change_active(struct bonding *bond)
1024{
1025 struct slave *prim = bond->primary_slave;
1026 struct slave *curr = bond->curr_active_slave;
1027
1028 if (!prim || !curr || curr->link != BOND_LINK_UP)
1029 return true;
1030 if (bond->force_primary) {
1031 bond->force_primary = false;
1032 return true;
1033 }
1034 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1035 (prim->speed < curr->speed ||
1036 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1037 return false;
1038 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1039 return false;
1040 return true;
1041}
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043/**
1044 * find_best_interface - select the best available slave to be the active one
1045 * @bond: our bonding struct
1046 *
1047 * Warning: Caller must hold curr_slave_lock for writing.
1048 */
1049static struct slave *bond_find_best_slave(struct bonding *bond)
1050{
1051 struct slave *new_active, *old_active;
1052 struct slave *bestslave = NULL;
1053 int mintime = bond->params.updelay;
1054 int i;
1055
Nicolas de Pesloüan49b4ad92009-10-07 14:11:00 -07001056 new_active = bond->curr_active_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
1058 if (!new_active) { /* there were no active slaves left */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001059 if (bond->slave_cnt > 0) /* found one slave */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 new_active = bond->first_slave;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001061 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 return NULL; /* still no slave, return NULL */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 }
1064
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 if ((bond->primary_slave) &&
Jiri Pirkoa5499522009-09-25 03:28:09 +00001066 bond->primary_slave->link == BOND_LINK_UP &&
1067 bond_should_change_active(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 new_active = bond->primary_slave;
1069 }
1070
1071 /* remember where to stop iterating over the slaves */
1072 old_active = new_active;
1073
1074 bond_for_each_slave_from(bond, new_active, i, old_active) {
Jiri Pirkob9f60252009-08-31 11:09:38 +00001075 if (new_active->link == BOND_LINK_UP) {
1076 return new_active;
1077 } else if (new_active->link == BOND_LINK_BACK &&
1078 IS_UP(new_active->dev)) {
1079 /* link up, but waiting for stabilization */
1080 if (new_active->delay < mintime) {
1081 mintime = new_active->delay;
1082 bestslave = new_active;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 }
1084 }
1085 }
1086
1087 return bestslave;
1088}
1089
1090/**
1091 * change_active_interface - change the active slave into the specified one
1092 * @bond: our bonding struct
1093 * @new: the new slave to make the active one
1094 *
1095 * Set the new slave to the bond's settings and unset them on the old
1096 * curr_active_slave.
1097 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1098 *
1099 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1100 * because it is apparently the best available slave we have, even though its
1101 * updelay hasn't timed out yet.
1102 *
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001103 * If new_active is not NULL, caller must hold bond->lock for read and
1104 * curr_slave_lock for write_bh.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001106void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107{
1108 struct slave *old_active = bond->curr_active_slave;
1109
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001110 if (old_active == new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113 if (new_active) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07001114 new_active->jiffies = jiffies;
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 if (new_active->link == BOND_LINK_BACK) {
1117 if (USES_PRIMARY(bond->params.mode)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001118 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1119 bond->dev->name, new_active->dev->name,
1120 (bond->params.updelay - new_active->delay) * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 }
1122
1123 new_active->delay = 0;
1124 new_active->link = BOND_LINK_UP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001126 if (bond->params.mode == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Holger Eitzenberger58402052008-12-09 23:07:13 -08001129 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 } else {
1132 if (USES_PRIMARY(bond->params.mode)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001133 pr_info("%s: making interface %s the new active one.\n",
1134 bond->dev->name, new_active->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 }
1136 }
1137 }
1138
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001139 if (USES_PRIMARY(bond->params.mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 bond_mc_swap(bond, new_active, old_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Holger Eitzenberger58402052008-12-09 23:07:13 -08001142 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 bond_alb_handle_active_change(bond, new_active);
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001144 if (old_active)
1145 bond_set_slave_inactive_flags(old_active);
1146 if (new_active)
1147 bond_set_slave_active_flags(new_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 } else {
1149 bond->curr_active_slave = new_active;
1150 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001151
1152 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001153 if (old_active)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001154 bond_set_slave_inactive_flags(old_active);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001155
1156 if (new_active) {
1157 bond_set_slave_active_flags(new_active);
Moni Shoua2ab82852007-10-09 19:43:39 -07001158
Or Gerlitz709f8a42008-06-13 18:12:01 -07001159 if (bond->params.fail_over_mac)
1160 bond_do_fail_over_mac(bond, new_active,
1161 old_active);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001162
Ben Hutchingsffa95ed2010-12-13 08:19:56 +00001163 if (netif_running(bond->dev)) {
1164 bond->send_grat_arp = bond->params.num_grat_arp;
1165 bond_send_gratuitous_arp(bond);
Or Gerlitz01f31092008-06-13 18:12:02 -07001166
Ben Hutchingsffa95ed2010-12-13 08:19:56 +00001167 bond->send_unsol_na = bond->params.num_unsol_na;
1168 bond_send_unsolicited_na(bond);
1169 }
Brian Haley305d5522008-11-04 17:51:14 -08001170
Or Gerlitz01f31092008-06-13 18:12:02 -07001171 write_unlock_bh(&bond->curr_slave_lock);
1172 read_unlock(&bond->lock);
1173
Moni Shoua75c78502009-09-15 02:37:40 -07001174 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
Or Gerlitz01f31092008-06-13 18:12:02 -07001175
1176 read_lock(&bond->lock);
1177 write_lock_bh(&bond->curr_slave_lock);
Moni Shoua7893b242008-05-17 21:10:12 -07001178 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001179 }
Andy Gospodareka2fd9402010-03-25 14:49:05 +00001180
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00001181 /* resend IGMP joins since active slave has changed or
1182 * all were sent on curr_active_slave */
Ben Hutchingsffa95ed2010-12-13 08:19:56 +00001183 if (((USES_PRIMARY(bond->params.mode) && new_active) ||
1184 bond->params.mode == BOND_MODE_ROUNDROBIN) &&
1185 netif_running(bond->dev)) {
Flavio Leitnerc2952c32010-10-05 14:23:59 +00001186 bond->igmp_retrans = bond->params.resend_igmp;
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00001187 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
Andy Gospodareka2fd9402010-03-25 14:49:05 +00001188 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189}
1190
1191/**
1192 * bond_select_active_slave - select a new active slave, if needed
1193 * @bond: our bonding struct
1194 *
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001195 * This functions should be called when one of the following occurs:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 * - The old curr_active_slave has been released or lost its link.
1197 * - The primary_slave has got its link back.
1198 * - A slave has got its link back and there's no old curr_active_slave.
1199 *
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001200 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001202void bond_select_active_slave(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203{
1204 struct slave *best_slave;
Jay Vosburghff59c452006-03-27 13:27:43 -08001205 int rv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207 best_slave = bond_find_best_slave(bond);
1208 if (best_slave != bond->curr_active_slave) {
1209 bond_change_active_slave(bond, best_slave);
Jay Vosburghff59c452006-03-27 13:27:43 -08001210 rv = bond_set_carrier(bond);
1211 if (!rv)
1212 return;
1213
1214 if (netif_carrier_ok(bond->dev)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001215 pr_info("%s: first active interface up!\n",
1216 bond->dev->name);
Jay Vosburghff59c452006-03-27 13:27:43 -08001217 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001218 pr_info("%s: now running without any active interface !\n",
1219 bond->dev->name);
Jay Vosburghff59c452006-03-27 13:27:43 -08001220 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
1222}
1223
1224/*--------------------------- slave list handling ---------------------------*/
1225
1226/*
1227 * This function attaches the slave to the end of list.
1228 *
1229 * bond->lock held for writing by caller.
1230 */
1231static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1232{
1233 if (bond->first_slave == NULL) { /* attaching the first slave */
1234 new_slave->next = new_slave;
1235 new_slave->prev = new_slave;
1236 bond->first_slave = new_slave;
1237 } else {
1238 new_slave->next = bond->first_slave;
1239 new_slave->prev = bond->first_slave->prev;
1240 new_slave->next->prev = new_slave;
1241 new_slave->prev->next = new_slave;
1242 }
1243
1244 bond->slave_cnt++;
1245}
1246
1247/*
1248 * This function detaches the slave from the list.
1249 * WARNING: no check is made to verify if the slave effectively
1250 * belongs to <bond>.
1251 * Nothing is freed on return, structures are just unchained.
1252 * If any slave pointer in bond was pointing to <slave>,
1253 * it should be changed by the calling function.
1254 *
1255 * bond->lock held for writing by caller.
1256 */
1257static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1258{
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001259 if (slave->next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 slave->next->prev = slave->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001262 if (slave->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 slave->prev->next = slave->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
1265 if (bond->first_slave == slave) { /* slave is the first slave */
1266 if (bond->slave_cnt > 1) { /* there are more slave */
1267 bond->first_slave = slave->next;
1268 } else {
1269 bond->first_slave = NULL; /* slave was the last one */
1270 }
1271 }
1272
1273 slave->next = NULL;
1274 slave->prev = NULL;
1275 bond->slave_cnt--;
1276}
1277
WANG Congf6dc31a2010-05-06 00:48:51 -07001278#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001279static inline int slave_enable_netpoll(struct slave *slave)
WANG Congf6dc31a2010-05-06 00:48:51 -07001280{
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001281 struct netpoll *np;
1282 int err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -07001283
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001284 np = kzalloc(sizeof(*np), GFP_KERNEL);
1285 err = -ENOMEM;
1286 if (!np)
1287 goto out;
1288
1289 np->dev = slave->dev;
1290 err = __netpoll_setup(np);
1291 if (err) {
1292 kfree(np);
1293 goto out;
WANG Congf6dc31a2010-05-06 00:48:51 -07001294 }
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001295 slave->np = np;
1296out:
1297 return err;
1298}
1299static inline void slave_disable_netpoll(struct slave *slave)
1300{
1301 struct netpoll *np = slave->np;
1302
1303 if (!np)
1304 return;
1305
1306 slave->np = NULL;
1307 synchronize_rcu_bh();
1308 __netpoll_cleanup(np);
1309 kfree(np);
1310}
1311static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1312{
1313 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1314 return false;
1315 if (!slave_dev->netdev_ops->ndo_poll_controller)
1316 return false;
1317 return true;
WANG Congf6dc31a2010-05-06 00:48:51 -07001318}
1319
1320static void bond_poll_controller(struct net_device *bond_dev)
1321{
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001322}
1323
1324static void __bond_netpoll_cleanup(struct bonding *bond)
1325{
Neil Hormanc2355e12010-10-13 16:01:49 +00001326 struct slave *slave;
1327 int i;
1328
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001329 bond_for_each_slave(bond, slave, i)
1330 if (IS_UP(slave->dev))
1331 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -07001332}
WANG Congf6dc31a2010-05-06 00:48:51 -07001333static void bond_netpoll_cleanup(struct net_device *bond_dev)
1334{
1335 struct bonding *bond = netdev_priv(bond_dev);
WANG Congf6dc31a2010-05-06 00:48:51 -07001336
1337 read_lock(&bond->lock);
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001338 __bond_netpoll_cleanup(bond);
WANG Congf6dc31a2010-05-06 00:48:51 -07001339 read_unlock(&bond->lock);
1340}
1341
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001342static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1343{
1344 struct bonding *bond = netdev_priv(dev);
1345 struct slave *slave;
1346 int i, err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -07001347
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001348 read_lock(&bond->lock);
1349 bond_for_each_slave(bond, slave, i) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001350 err = slave_enable_netpoll(slave);
1351 if (err) {
1352 __bond_netpoll_cleanup(bond);
1353 break;
1354 }
1355 }
1356 read_unlock(&bond->lock);
1357 return err;
1358}
1359
1360static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1361{
1362 return bond->dev->npinfo;
1363}
1364
1365#else
1366static inline int slave_enable_netpoll(struct slave *slave)
1367{
1368 return 0;
1369}
1370static inline void slave_disable_netpoll(struct slave *slave)
1371{
1372}
WANG Congf6dc31a2010-05-06 00:48:51 -07001373static void bond_netpoll_cleanup(struct net_device *bond_dev)
1374{
1375}
WANG Congf6dc31a2010-05-06 00:48:51 -07001376#endif
1377
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378/*---------------------------------- IOCTL ----------------------------------*/
1379
Adrian Bunk4ad072c2007-07-09 11:51:12 -07001380static int bond_sethwaddr(struct net_device *bond_dev,
1381 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382{
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001383 pr_debug("bond_dev=%p\n", bond_dev);
1384 pr_debug("slave_dev=%p\n", slave_dev);
1385 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1387 return 0;
1388}
1389
Herbert Xu7f353bf2007-08-10 15:47:58 -07001390#define BOND_VLAN_FEATURES \
1391 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1392 NETIF_F_HW_VLAN_FILTER)
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001393
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001394/*
Jay Vosburgh8e3babc2005-11-04 18:45:45 -08001395 * Compute the common dev->feature set available to all slaves. Some
Herbert Xu7f353bf2007-08-10 15:47:58 -07001396 * feature bits are managed elsewhere, so preserve those feature bits
1397 * on the master device.
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001398 */
1399static int bond_compute_features(struct bonding *bond)
1400{
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001401 struct slave *slave;
1402 struct net_device *bond_dev = bond->dev;
Michał Mirosław04ed3e72011-01-24 15:32:47 -08001403 u32 features = bond_dev->features;
1404 u32 vlan_features = 0;
Moni Shoua3158bf72007-10-09 19:43:41 -07001405 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1406 bond_dev->hard_header_len);
Jay Vosburgh8e3babc2005-11-04 18:45:45 -08001407 int i;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001408
Herbert Xu7f353bf2007-08-10 15:47:58 -07001409 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
Tom Herbertc6e1a0d2011-04-04 22:30:30 -07001410 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM | NETIF_F_NOCACHE_COPY;
Herbert Xub63365a2008-10-23 01:11:29 -07001411
1412 if (!bond->first_slave)
1413 goto done;
1414
1415 features &= ~NETIF_F_ONE_FOR_ALL;
Herbert Xu7f353bf2007-08-10 15:47:58 -07001416
Jay Vosburgh278339a2009-08-28 12:05:12 +00001417 vlan_features = bond->first_slave->dev->vlan_features;
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001418 bond_for_each_slave(bond, slave, i) {
Herbert Xub63365a2008-10-23 01:11:29 -07001419 features = netdev_increment_features(features,
1420 slave->dev->features,
1421 NETIF_F_ONE_FOR_ALL);
Jay Vosburgh278339a2009-08-28 12:05:12 +00001422 vlan_features = netdev_increment_features(vlan_features,
1423 slave->dev->vlan_features,
1424 NETIF_F_ONE_FOR_ALL);
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001425 if (slave->dev->hard_header_len > max_hard_header_len)
1426 max_hard_header_len = slave->dev->hard_header_len;
1427 }
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001428
Herbert Xub63365a2008-10-23 01:11:29 -07001429done:
Herbert Xu7f353bf2007-08-10 15:47:58 -07001430 features |= (bond_dev->features & BOND_VLAN_FEATURES);
Michał Mirosławacd11302011-01-24 15:45:15 -08001431 bond_dev->features = netdev_fix_features(bond_dev, features);
1432 bond_dev->vlan_features = netdev_fix_features(bond_dev, vlan_features);
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001433 bond_dev->hard_header_len = max_hard_header_len;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001434
1435 return 0;
1436}
1437
Moni Shoua872254d2007-10-09 19:43:38 -07001438static void bond_setup_by_slave(struct net_device *bond_dev,
1439 struct net_device *slave_dev)
1440{
Wang Chen454d7c92008-11-12 23:37:49 -08001441 struct bonding *bond = netdev_priv(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07001442
Stephen Hemminger00829822008-11-20 20:14:53 -08001443 bond_dev->header_ops = slave_dev->header_ops;
Moni Shoua872254d2007-10-09 19:43:38 -07001444
1445 bond_dev->type = slave_dev->type;
1446 bond_dev->hard_header_len = slave_dev->hard_header_len;
1447 bond_dev->addr_len = slave_dev->addr_len;
1448
1449 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1450 slave_dev->addr_len);
Moni Shouad90a1622007-10-09 19:43:43 -07001451 bond->setup_by_slave = 1;
Moni Shoua872254d2007-10-09 19:43:38 -07001452}
1453
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001454/* On bonding slaves other than the currently active slave, suppress
1455 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
1456 * ARP on active-backup slaves with arp_validate enabled.
1457 */
1458static bool bond_should_deliver_exact_match(struct sk_buff *skb,
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001459 struct slave *slave,
1460 struct bonding *bond)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001461{
Jiri Pirko2d7011c2011-03-16 08:46:43 +00001462 if (bond_is_slave_inactive(slave)) {
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001463 if (slave_do_arp_validate(bond, slave) &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001464 skb->protocol == __cpu_to_be16(ETH_P_ARP))
1465 return false;
1466
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001467 if (bond->params.mode == BOND_MODE_ALB &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001468 skb->pkt_type != PACKET_BROADCAST &&
1469 skb->pkt_type != PACKET_MULTICAST)
1470 return false;
1471
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001472 if (bond->params.mode == BOND_MODE_8023AD &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001473 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
1474 return false;
1475
1476 return true;
1477 }
1478 return false;
1479}
1480
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001481static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001482{
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001483 struct sk_buff *skb = *pskb;
Jiri Pirkof1c17752011-03-12 03:14:35 +00001484 struct slave *slave;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001485 struct bonding *bond;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001486
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001487 skb = skb_share_check(skb, GFP_ATOMIC);
1488 if (unlikely(!skb))
1489 return RX_HANDLER_CONSUMED;
1490
1491 *pskb = skb;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001492
Jiri Pirko35d48902011-03-22 02:38:12 +00001493 slave = bond_slave_get_rcu(skb->dev);
1494 bond = slave->bond;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001495
1496 if (bond->params.arp_interval)
Jiri Pirkof1c17752011-03-12 03:14:35 +00001497 slave->dev->last_rx = jiffies;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001498
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001499 if (bond_should_deliver_exact_match(skb, slave, bond)) {
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001500 return RX_HANDLER_EXACT;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001501 }
1502
Jiri Pirko35d48902011-03-22 02:38:12 +00001503 skb->dev = bond->dev;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001504
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001505 if (bond->params.mode == BOND_MODE_ALB &&
Jiri Pirko35d48902011-03-22 02:38:12 +00001506 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001507 skb->pkt_type == PACKET_HOST) {
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001508
Changli Gao541ac7c2011-03-02 21:07:14 +00001509 if (unlikely(skb_cow_head(skb,
1510 skb->data - skb_mac_header(skb)))) {
1511 kfree_skb(skb);
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001512 return RX_HANDLER_CONSUMED;
Changli Gao541ac7c2011-03-02 21:07:14 +00001513 }
Jiri Pirko35d48902011-03-22 02:38:12 +00001514 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001515 }
1516
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001517 return RX_HANDLER_ANOTHER;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001518}
1519
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520/* enslave device <slave> to bond device <master> */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001521int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522{
Wang Chen454d7c92008-11-12 23:37:49 -08001523 struct bonding *bond = netdev_priv(bond_dev);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001524 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 struct slave *new_slave = NULL;
Jiri Pirko22bedad32010-04-01 21:22:57 +00001526 struct netdev_hw_addr *ha;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 struct sockaddr addr;
1528 int link_reporting;
1529 int old_features = bond_dev->features;
1530 int res = 0;
1531
nsxfreddy@gmail.com552709d2005-09-21 14:18:04 -05001532 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001533 slave_ops->ndo_do_ioctl == NULL) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001534 pr_warning("%s: Warning: no link monitoring support for %s\n",
1535 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 }
1537
1538 /* bond must be initialized by bond_open() before enslaving */
1539 if (!(bond_dev->flags & IFF_UP)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001540 pr_warning("%s: master_dev is not up in bond_enslave\n",
1541 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 }
1543
1544 /* already enslaved */
1545 if (slave_dev->flags & IFF_SLAVE) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001546 pr_debug("Error, Device was already enslaved\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 return -EBUSY;
1548 }
1549
1550 /* vlan challenged mutual exclusion */
1551 /* no need to lock since we're protected by rtnl_lock */
1552 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001553 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
Jay Vosburghf35188f2010-07-21 12:14:47 +00001554 if (bond->vlgrp) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001555 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1556 bond_dev->name, slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 return -EPERM;
1558 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001559 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1560 bond_dev->name, slave_dev->name,
1561 slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1563 }
1564 } else {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001565 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 if (bond->slave_cnt == 0) {
1567 /* First slave, and it is not VLAN challenged,
1568 * so remove the block of adding VLANs over the bond.
1569 */
1570 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1571 }
1572 }
1573
Jay Vosburgh217df672005-09-26 16:11:50 -07001574 /*
1575 * Old ifenslave binaries are no longer supported. These can
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001576 * be identified with moderate accuracy by the state of the slave:
Jay Vosburgh217df672005-09-26 16:11:50 -07001577 * the current ifenslave will set the interface down prior to
1578 * enslaving it; the old ifenslave will not.
1579 */
1580 if ((slave_dev->flags & IFF_UP)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001581 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
Jay Vosburgh217df672005-09-26 16:11:50 -07001582 slave_dev->name);
1583 res = -EPERM;
1584 goto err_undo_flags;
1585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
Moni Shoua872254d2007-10-09 19:43:38 -07001587 /* set bonding device ether type by slave - bonding netdevices are
1588 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1589 * there is a need to override some of the type dependent attribs/funcs.
1590 *
1591 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1592 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1593 */
1594 if (bond->slave_cnt == 0) {
Moni Shouae36b9d12009-07-15 04:56:31 +00001595 if (bond_dev->type != slave_dev->type) {
Moni Shouae36b9d12009-07-15 04:56:31 +00001596 pr_debug("%s: change device type from %d to %d\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001597 bond_dev->name,
1598 bond_dev->type, slave_dev->type);
Moni Shoua75c78502009-09-15 02:37:40 -07001599
Jiri Pirko3ca5b402010-03-10 10:29:35 +00001600 res = netdev_bonding_change(bond_dev,
1601 NETDEV_PRE_TYPE_CHANGE);
1602 res = notifier_to_errno(res);
1603 if (res) {
1604 pr_err("%s: refused to change device type\n",
1605 bond_dev->name);
1606 res = -EBUSY;
1607 goto err_undo_flags;
1608 }
Moni Shoua75c78502009-09-15 02:37:40 -07001609
Jiri Pirko32a806c2010-03-19 04:00:23 +00001610 /* Flush unicast and multicast addresses */
Jiri Pirkoa748ee22010-04-01 21:22:09 +00001611 dev_uc_flush(bond_dev);
Jiri Pirko22bedad32010-04-01 21:22:57 +00001612 dev_mc_flush(bond_dev);
Jiri Pirko32a806c2010-03-19 04:00:23 +00001613
Moni Shouae36b9d12009-07-15 04:56:31 +00001614 if (slave_dev->type != ARPHRD_ETHER)
1615 bond_setup_by_slave(bond_dev, slave_dev);
1616 else
1617 ether_setup(bond_dev);
Moni Shoua75c78502009-09-15 02:37:40 -07001618
Jiri Pirko93d9b7d2010-03-10 10:28:56 +00001619 netdev_bonding_change(bond_dev,
1620 NETDEV_POST_TYPE_CHANGE);
Moni Shouae36b9d12009-07-15 04:56:31 +00001621 }
Moni Shoua872254d2007-10-09 19:43:38 -07001622 } else if (bond_dev->type != slave_dev->type) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001623 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1624 slave_dev->name,
1625 slave_dev->type, bond_dev->type);
1626 res = -EINVAL;
1627 goto err_undo_flags;
Moni Shoua872254d2007-10-09 19:43:38 -07001628 }
1629
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001630 if (slave_ops->ndo_set_mac_address == NULL) {
Moni Shoua2ab82852007-10-09 19:43:39 -07001631 if (bond->slave_cnt == 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001632 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1633 bond_dev->name);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001634 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1635 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001636 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1637 bond_dev->name);
Moni Shoua2ab82852007-10-09 19:43:39 -07001638 res = -EOPNOTSUPP;
1639 goto err_undo_flags;
1640 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 }
1642
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001643 /* If this is the first slave, then we need to set the master's hardware
1644 * address to be the same as the slave's. */
David Strandd13a2cb2010-12-01 11:43:08 -08001645 if (is_zero_ether_addr(bond->dev->dev_addr))
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001646 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1647 slave_dev->addr_len);
1648
1649
Joe Jin243cb4e2007-02-06 14:16:40 -08001650 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 if (!new_slave) {
1652 res = -ENOMEM;
1653 goto err_undo_flags;
1654 }
1655
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00001656 /*
1657 * Set the new_slave's queue_id to be zero. Queue ID mapping
1658 * is set via sysfs or module option if desired.
1659 */
1660 new_slave->queue_id = 0;
1661
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001662 /* Save slave's original mtu and then set it to match the bond */
1663 new_slave->original_mtu = slave_dev->mtu;
1664 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1665 if (res) {
1666 pr_debug("Error %d calling dev_set_mtu\n", res);
1667 goto err_free;
1668 }
1669
Jay Vosburgh217df672005-09-26 16:11:50 -07001670 /*
1671 * Save slave's original ("permanent") mac address for modes
1672 * that need it, and for restoring it upon release, and then
1673 * set it to the master's address
1674 */
1675 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Jay Vosburghdd957c52007-10-09 19:57:24 -07001677 if (!bond->params.fail_over_mac) {
Moni Shoua2ab82852007-10-09 19:43:39 -07001678 /*
1679 * Set slave to master's mac address. The application already
1680 * set the master's mac address to that of the first slave
1681 */
1682 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1683 addr.sa_family = slave_dev->type;
1684 res = dev_set_mac_address(slave_dev, &addr);
1685 if (res) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001686 pr_debug("Error %d calling set_mac_address\n", res);
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001687 goto err_restore_mtu;
Moni Shoua2ab82852007-10-09 19:43:39 -07001688 }
Jay Vosburgh217df672005-09-26 16:11:50 -07001689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
Jiri Pirko1765a572011-02-12 06:48:36 +00001691 res = netdev_set_bond_master(slave_dev, bond_dev);
Jay Vosburghc2edacf2007-07-09 10:42:47 -07001692 if (res) {
Jiri Pirko1765a572011-02-12 06:48:36 +00001693 pr_debug("Error %d calling netdev_set_bond_master\n", res);
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001694 goto err_restore_mac;
Jay Vosburghc2edacf2007-07-09 10:42:47 -07001695 }
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001696
Jay Vosburgh217df672005-09-26 16:11:50 -07001697 /* open the slave since the application closed it */
1698 res = dev_open(slave_dev);
1699 if (res) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001700 pr_debug("Opening slave %s failed\n", slave_dev->name);
Jiri Pirko35d48902011-03-22 02:38:12 +00001701 goto err_unset_master;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 }
1703
Jiri Pirko35d48902011-03-22 02:38:12 +00001704 new_slave->bond = bond;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 new_slave->dev = slave_dev;
Jay Vosburgh0b680e72006-09-22 21:54:10 -07001706 slave_dev->priv_flags |= IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
Holger Eitzenberger58402052008-12-09 23:07:13 -08001708 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 /* bond_alb_init_slave() must be called before all other stages since
1710 * it might fail and we do not want to have to undo everything
1711 */
1712 res = bond_alb_init_slave(bond, new_slave);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001713 if (res)
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001714 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 }
1716
1717 /* If the mode USES_PRIMARY, then the new slave gets the
1718 * master's promisc (and mc) settings only if it becomes the
1719 * curr_active_slave, and that is taken care of later when calling
1720 * bond_change_active()
1721 */
1722 if (!USES_PRIMARY(bond->params.mode)) {
1723 /* set promiscuity level to new slave */
1724 if (bond_dev->flags & IFF_PROMISC) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -07001725 res = dev_set_promiscuity(slave_dev, 1);
1726 if (res)
1727 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 }
1729
1730 /* set allmulti level to new slave */
1731 if (bond_dev->flags & IFF_ALLMULTI) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -07001732 res = dev_set_allmulti(slave_dev, 1);
1733 if (res)
1734 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 }
1736
David S. Millerb9e40852008-07-15 00:15:08 -07001737 netif_addr_lock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 /* upload master's mc_list to new slave */
Jiri Pirko22bedad32010-04-01 21:22:57 +00001739 netdev_for_each_mc_addr(ha, bond_dev)
1740 dev_mc_add(slave_dev, ha->addr);
David S. Millerb9e40852008-07-15 00:15:08 -07001741 netif_addr_unlock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 }
1743
1744 if (bond->params.mode == BOND_MODE_8023AD) {
1745 /* add lacpdu mc addr to mc list */
1746 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1747
Jiri Pirko22bedad32010-04-01 21:22:57 +00001748 dev_mc_add(slave_dev, lacpdu_multicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 }
1750
1751 bond_add_vlans_on_slave(bond, slave_dev);
1752
1753 write_lock_bh(&bond->lock);
1754
1755 bond_attach_slave(bond, new_slave);
1756
1757 new_slave->delay = 0;
1758 new_slave->link_failure_count = 0;
1759
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001760 bond_compute_features(bond);
1761
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001762 write_unlock_bh(&bond->lock);
1763
1764 read_lock(&bond->lock);
1765
Jay Vosburghf5b2b962006-09-22 21:54:53 -07001766 new_slave->last_arp_rx = jiffies;
1767
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 if (bond->params.miimon && !bond->params.use_carrier) {
1769 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1770
1771 if ((link_reporting == -1) && !bond->params.arp_interval) {
1772 /*
1773 * miimon is set but a bonded network driver
1774 * does not support ETHTOOL/MII and
1775 * arp_interval is not set. Note: if
1776 * use_carrier is enabled, we will never go
1777 * here (because netif_carrier is always
1778 * supported); thus, we don't need to change
1779 * the messages for netif_carrier.
1780 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001781 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001782 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 } else if (link_reporting == -1) {
1784 /* unable get link status using mii/ethtool */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001785 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1786 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 }
1788 }
1789
1790 /* check for initial state */
1791 if (!bond->params.miimon ||
1792 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1793 if (bond->params.updelay) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001794 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 new_slave->link = BOND_LINK_BACK;
1796 new_slave->delay = bond->params.updelay;
1797 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001798 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 new_slave->link = BOND_LINK_UP;
1800 }
1801 new_slave->jiffies = jiffies;
1802 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001803 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 new_slave->link = BOND_LINK_DOWN;
1805 }
1806
1807 if (bond_update_speed_duplex(new_slave) &&
1808 (new_slave->link != BOND_LINK_DOWN)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001809 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1810 bond_dev->name, new_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
1812 if (bond->params.mode == BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001813 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1814 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 }
1816 }
1817
1818 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1819 /* if there is a primary slave, remember it */
Jiri Pirkoa5499522009-09-25 03:28:09 +00001820 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 bond->primary_slave = new_slave;
Jiri Pirkoa5499522009-09-25 03:28:09 +00001822 bond->force_primary = true;
1823 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 }
1825
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001826 write_lock_bh(&bond->curr_slave_lock);
1827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 switch (bond->params.mode) {
1829 case BOND_MODE_ACTIVEBACKUP:
Jay Vosburgh8a8e4472006-09-22 21:56:15 -07001830 bond_set_slave_inactive_flags(new_slave);
1831 bond_select_active_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 break;
1833 case BOND_MODE_8023AD:
1834 /* in 802.3ad mode, the internal mechanism
1835 * will activate the slaves in the selected
1836 * aggregator
1837 */
1838 bond_set_slave_inactive_flags(new_slave);
1839 /* if this is the first slave */
1840 if (bond->slave_cnt == 1) {
1841 SLAVE_AD_INFO(new_slave).id = 1;
1842 /* Initialize AD with the number of times that the AD timer is called in 1 second
1843 * can be called only after the mac address of the bond is set
1844 */
1845 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1846 bond->params.lacp_fast);
1847 } else {
1848 SLAVE_AD_INFO(new_slave).id =
1849 SLAVE_AD_INFO(new_slave->prev).id + 1;
1850 }
1851
1852 bond_3ad_bind_slave(new_slave);
1853 break;
1854 case BOND_MODE_TLB:
1855 case BOND_MODE_ALB:
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001856 bond_set_active_slave(new_slave);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001857 bond_set_slave_inactive_flags(new_slave);
Jiri Pirko5a29f782009-03-25 17:23:38 -07001858 bond_select_active_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 break;
1860 default:
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001861 pr_debug("This slave is always active in trunk mode\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
1863 /* always active in trunk mode */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001864 bond_set_active_slave(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
1866 /* In trunking mode there is little meaning to curr_active_slave
1867 * anyway (it holds no special properties of the bond device),
1868 * so we can change it without calling change_active_interface()
1869 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001870 if (!bond->curr_active_slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 bond->curr_active_slave = new_slave;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001872
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 break;
1874 } /* switch(bond_mode) */
1875
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001876 write_unlock_bh(&bond->curr_slave_lock);
1877
Jay Vosburghff59c452006-03-27 13:27:43 -08001878 bond_set_carrier(bond);
1879
WANG Congf6dc31a2010-05-06 00:48:51 -07001880#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001881 slave_dev->npinfo = bond_netpoll_info(bond);
1882 if (slave_dev->npinfo) {
1883 if (slave_enable_netpoll(new_slave)) {
1884 read_unlock(&bond->lock);
1885 pr_info("Error, %s: master_dev is using netpoll, "
1886 "but new slave device does not support netpoll.\n",
1887 bond_dev->name);
1888 res = -EBUSY;
1889 goto err_close;
1890 }
WANG Congf6dc31a2010-05-06 00:48:51 -07001891 }
1892#endif
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001893
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001894 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
Mitch Williamsb76cdba2005-11-09 10:36:41 -08001896 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1897 if (res)
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001898 goto err_close;
Mitch Williamsb76cdba2005-11-09 10:36:41 -08001899
Jiri Pirko35d48902011-03-22 02:38:12 +00001900 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1901 new_slave);
1902 if (res) {
1903 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1904 goto err_dest_symlinks;
1905 }
1906
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001907 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1908 bond_dev->name, slave_dev->name,
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001909 bond_is_active_slave(new_slave) ? "n active" : " backup",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001910 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
1912 /* enslave is successful */
1913 return 0;
1914
1915/* Undo stages on error */
Jiri Pirko35d48902011-03-22 02:38:12 +00001916err_dest_symlinks:
1917 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1918
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919err_close:
1920 dev_close(slave_dev);
1921
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001922err_unset_master:
Jiri Pirko1765a572011-02-12 06:48:36 +00001923 netdev_set_bond_master(slave_dev, NULL);
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001924
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925err_restore_mac:
Jay Vosburghdd957c52007-10-09 19:57:24 -07001926 if (!bond->params.fail_over_mac) {
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001927 /* XXX TODO - fom follow mode needs to change master's
1928 * MAC if this slave's MAC is in use by the bond, or at
1929 * least print a warning.
1930 */
Moni Shoua2ab82852007-10-09 19:43:39 -07001931 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1932 addr.sa_family = slave_dev->type;
1933 dev_set_mac_address(slave_dev, &addr);
1934 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001936err_restore_mtu:
1937 dev_set_mtu(slave_dev, new_slave->original_mtu);
1938
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939err_free:
1940 kfree(new_slave);
1941
1942err_undo_flags:
1943 bond_dev->features = old_features;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001944
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 return res;
1946}
1947
1948/*
1949 * Try to release the slave device <slave> from the bond device <master>
1950 * It is legal to access curr_active_slave without a lock because all the function
1951 * is write-locked.
1952 *
1953 * The rules for slave state should be:
1954 * for Active/Backup:
1955 * Active stays on all backups go down
1956 * for Bonded connections:
1957 * The first up interface should be left on and all others downed.
1958 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001959int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960{
Wang Chen454d7c92008-11-12 23:37:49 -08001961 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 struct slave *slave, *oldcurrent;
1963 struct sockaddr addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
1965 /* slave is not a slave or master is not master of this slave */
1966 if (!(slave_dev->flags & IFF_SLAVE) ||
1967 (slave_dev->master != bond_dev)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001968 pr_err("%s: Error: cannot release %s.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 bond_dev->name, slave_dev->name);
1970 return -EINVAL;
1971 }
1972
Neil Hormane843fa52010-10-13 16:01:50 +00001973 block_netpoll_tx();
WANG Congf6dc31a2010-05-06 00:48:51 -07001974 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 write_lock_bh(&bond->lock);
1976
1977 slave = bond_get_slave_by_dev(bond, slave_dev);
1978 if (!slave) {
1979 /* not a slave of this bond */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001980 pr_info("%s: %s not enslaved\n",
1981 bond_dev->name, slave_dev->name);
Jay Vosburghf5e2a7b2006-02-07 21:17:22 -08001982 write_unlock_bh(&bond->lock);
Neil Hormane843fa52010-10-13 16:01:50 +00001983 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 return -EINVAL;
1985 }
1986
Jiri Pirko35d48902011-03-22 02:38:12 +00001987 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1988 * for this slave anymore.
1989 */
1990 netdev_rx_handler_unregister(slave_dev);
1991 write_unlock_bh(&bond->lock);
1992 synchronize_net();
1993 write_lock_bh(&bond->lock);
1994
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001995 if (!bond->params.fail_over_mac) {
Joe Perches8e95a202009-12-03 07:58:21 +00001996 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1997 bond->slave_cnt > 1)
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001998 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1999 bond_dev->name, slave_dev->name,
2000 slave->perm_hwaddr,
2001 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 }
2003
2004 /* Inform AD package of unbinding of slave. */
2005 if (bond->params.mode == BOND_MODE_8023AD) {
2006 /* must be called before the slave is
2007 * detached from the list
2008 */
2009 bond_3ad_unbind_slave(slave);
2010 }
2011
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002012 pr_info("%s: releasing %s interface %s\n",
2013 bond_dev->name,
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002014 bond_is_active_slave(slave) ? "active" : "backup",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002015 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
2017 oldcurrent = bond->curr_active_slave;
2018
2019 bond->current_arp_slave = NULL;
2020
2021 /* release the slave from its bond */
2022 bond_detach_slave(bond, slave);
2023
Arthur Kepner8531c5f2005-08-23 01:34:53 -04002024 bond_compute_features(bond);
2025
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002026 if (bond->primary_slave == slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 bond->primary_slave = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002029 if (oldcurrent == slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 bond_change_active_slave(bond, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
Holger Eitzenberger58402052008-12-09 23:07:13 -08002032 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 /* Must be called only after the slave has been
2034 * detached from the list and the curr_active_slave
2035 * has been cleared (if our_slave == old_current),
2036 * but before a new active slave is selected.
2037 */
Jay Vosburgh25433312008-01-17 16:24:59 -08002038 write_unlock_bh(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 bond_alb_deinit_slave(bond, slave);
Jay Vosburgh25433312008-01-17 16:24:59 -08002040 write_lock_bh(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 }
2042
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002043 if (oldcurrent == slave) {
2044 /*
2045 * Note that we hold RTNL over this sequence, so there
2046 * is no concern that another slave add/remove event
2047 * will interfere.
2048 */
2049 write_unlock_bh(&bond->lock);
2050 read_lock(&bond->lock);
2051 write_lock_bh(&bond->curr_slave_lock);
2052
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 bond_select_active_slave(bond);
2054
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002055 write_unlock_bh(&bond->curr_slave_lock);
2056 read_unlock(&bond->lock);
2057 write_lock_bh(&bond->lock);
2058 }
2059
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 if (bond->slave_cnt == 0) {
Jay Vosburghff59c452006-03-27 13:27:43 -08002061 bond_set_carrier(bond);
2062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 /* if the last slave was removed, zero the mac address
2064 * of the master so it will be set by the application
2065 * to the mac address of the first slave
2066 */
2067 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2068
Jay Vosburghf35188f2010-07-21 12:14:47 +00002069 if (!bond->vlgrp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2071 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002072 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2073 bond_dev->name, bond_dev->name);
2074 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2075 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 }
2077 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2078 !bond_has_challenged_slaves(bond)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002079 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2080 bond_dev->name, slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2082 }
2083
2084 write_unlock_bh(&bond->lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002085 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
Mitch Williamsb76cdba2005-11-09 10:36:41 -08002087 /* must do this from outside any spinlocks */
2088 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2089
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 bond_del_vlans_from_slave(bond, slave_dev);
2091
2092 /* If the mode USES_PRIMARY, then we should only remove its
2093 * promisc and mc settings if it was the curr_active_slave, but that was
2094 * already taken care of above when we detached the slave
2095 */
2096 if (!USES_PRIMARY(bond->params.mode)) {
2097 /* unset promiscuity level from slave */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002098 if (bond_dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 dev_set_promiscuity(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
2101 /* unset allmulti level from slave */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002102 if (bond_dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 dev_set_allmulti(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
2105 /* flush master's mc_list from slave */
David S. Millerb9e40852008-07-15 00:15:08 -07002106 netif_addr_lock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 bond_mc_list_flush(bond_dev, slave_dev);
David S. Millerb9e40852008-07-15 00:15:08 -07002108 netif_addr_unlock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 }
2110
Jiri Pirko1765a572011-02-12 06:48:36 +00002111 netdev_set_bond_master(slave_dev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112
Amerigo Wang8a8efa22011-02-17 23:43:32 +00002113 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -07002114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 /* close slave before restoring its mac address */
2116 dev_close(slave_dev);
2117
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07002118 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
Moni Shoua2ab82852007-10-09 19:43:39 -07002119 /* restore original ("permanent") mac address */
2120 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2121 addr.sa_family = slave_dev->type;
2122 dev_set_mac_address(slave_dev, &addr);
2123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00002125 dev_set_mtu(slave_dev, slave->original_mtu);
2126
Jiri Pirko2d7011c2011-03-16 08:46:43 +00002127 slave_dev->priv_flags &= ~IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
2129 kfree(slave);
2130
2131 return 0; /* deletion OK */
2132}
2133
2134/*
Nicolas de Pesloüandadaa102011-03-19 13:36:18 -07002135* First release a slave and then destroy the bond if no more slaves are left.
Moni Shouad90a1622007-10-09 19:43:43 -07002136* Must be under rtnl_lock when this function is called.
2137*/
stephen hemminger26d8ee72010-10-15 05:09:34 +00002138static int bond_release_and_destroy(struct net_device *bond_dev,
2139 struct net_device *slave_dev)
Moni Shouad90a1622007-10-09 19:43:43 -07002140{
Wang Chen454d7c92008-11-12 23:37:49 -08002141 struct bonding *bond = netdev_priv(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07002142 int ret;
2143
2144 ret = bond_release(bond_dev, slave_dev);
2145 if ((ret == 0) && (bond->slave_cnt == 0)) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +00002146 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002147 pr_info("%s: destroying bond %s.\n",
2148 bond_dev->name, bond_dev->name);
Stephen Hemminger9e716262009-06-12 19:02:47 +00002149 unregister_netdevice(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07002150 }
2151 return ret;
2152}
2153
2154/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 * This function releases all slaves.
2156 */
2157static int bond_release_all(struct net_device *bond_dev)
2158{
Wang Chen454d7c92008-11-12 23:37:49 -08002159 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 struct slave *slave;
2161 struct net_device *slave_dev;
2162 struct sockaddr addr;
2163
2164 write_lock_bh(&bond->lock);
2165
Jay Vosburghff59c452006-03-27 13:27:43 -08002166 netif_carrier_off(bond_dev);
2167
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002168 if (bond->slave_cnt == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
2171 bond->current_arp_slave = NULL;
2172 bond->primary_slave = NULL;
2173 bond_change_active_slave(bond, NULL);
2174
2175 while ((slave = bond->first_slave) != NULL) {
2176 /* Inform AD package of unbinding of slave
2177 * before slave is detached from the list.
2178 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002179 if (bond->params.mode == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 bond_3ad_unbind_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
2182 slave_dev = slave->dev;
2183 bond_detach_slave(bond, slave);
2184
Jay Vosburgh25433312008-01-17 16:24:59 -08002185 /* now that the slave is detached, unlock and perform
2186 * all the undo steps that should not be called from
2187 * within a lock.
2188 */
2189 write_unlock_bh(&bond->lock);
2190
Jiri Pirko35d48902011-03-22 02:38:12 +00002191 /* unregister rx_handler early so bond_handle_frame wouldn't
2192 * be called for this slave anymore.
2193 */
2194 netdev_rx_handler_unregister(slave_dev);
2195 synchronize_net();
2196
Holger Eitzenberger58402052008-12-09 23:07:13 -08002197 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 /* must be called only after the slave
2199 * has been detached from the list
2200 */
2201 bond_alb_deinit_slave(bond, slave);
2202 }
2203
Arthur Kepner8531c5f2005-08-23 01:34:53 -04002204 bond_compute_features(bond);
2205
Mitch Williamsb76cdba2005-11-09 10:36:41 -08002206 bond_destroy_slave_symlinks(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 bond_del_vlans_from_slave(bond, slave_dev);
2208
2209 /* If the mode USES_PRIMARY, then we should only remove its
2210 * promisc and mc settings if it was the curr_active_slave, but that was
2211 * already taken care of above when we detached the slave
2212 */
2213 if (!USES_PRIMARY(bond->params.mode)) {
2214 /* unset promiscuity level from slave */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002215 if (bond_dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 dev_set_promiscuity(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
2218 /* unset allmulti level from slave */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002219 if (bond_dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 dev_set_allmulti(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
2222 /* flush master's mc_list from slave */
David S. Millerb9e40852008-07-15 00:15:08 -07002223 netif_addr_lock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 bond_mc_list_flush(bond_dev, slave_dev);
David S. Millerb9e40852008-07-15 00:15:08 -07002225 netif_addr_unlock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 }
2227
Jiri Pirko1765a572011-02-12 06:48:36 +00002228 netdev_set_bond_master(slave_dev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
Amerigo Wang8a8efa22011-02-17 23:43:32 +00002230 slave_disable_netpoll(slave);
2231
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 /* close slave before restoring its mac address */
2233 dev_close(slave_dev);
2234
Jay Vosburghdd957c52007-10-09 19:57:24 -07002235 if (!bond->params.fail_over_mac) {
Moni Shoua2ab82852007-10-09 19:43:39 -07002236 /* restore original ("permanent") mac address*/
2237 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2238 addr.sa_family = slave_dev->type;
2239 dev_set_mac_address(slave_dev, &addr);
2240 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 kfree(slave);
2243
2244 /* re-acquire the lock before getting the next slave */
2245 write_lock_bh(&bond->lock);
2246 }
2247
2248 /* zero the mac address of the master so it will be
2249 * set by the application to the mac address of the
2250 * first slave
2251 */
2252 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2253
Jay Vosburghf35188f2010-07-21 12:14:47 +00002254 if (!bond->vlgrp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
Jay Vosburghf35188f2010-07-21 12:14:47 +00002256 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002257 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2258 bond_dev->name, bond_dev->name);
2259 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2260 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 }
2262
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002263 pr_info("%s: released all slaves\n", bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
2265out:
2266 write_unlock_bh(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 return 0;
2268}
2269
2270/*
2271 * This function changes the active slave to slave <slave_dev>.
2272 * It returns -EINVAL in the following cases.
2273 * - <slave_dev> is not found in the list.
2274 * - There is not active slave now.
2275 * - <slave_dev> is already active.
2276 * - The link state of <slave_dev> is not BOND_LINK_UP.
2277 * - <slave_dev> is not running.
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002278 * In these cases, this function does nothing.
2279 * In the other cases, current_slave pointer is changed and 0 is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 */
2281static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2282{
Wang Chen454d7c92008-11-12 23:37:49 -08002283 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 struct slave *old_active = NULL;
2285 struct slave *new_active = NULL;
2286 int res = 0;
2287
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002288 if (!USES_PRIMARY(bond->params.mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
2291 /* Verify that master_dev is indeed the master of slave_dev */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002292 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002295 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002297 read_lock(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 old_active = bond->curr_active_slave;
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002299 read_unlock(&bond->curr_slave_lock);
2300
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 new_active = bond_get_slave_by_dev(bond, slave_dev);
2302
2303 /*
2304 * Changing to the current active: do nothing; return success.
2305 */
2306 if (new_active && (new_active == old_active)) {
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002307 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 return 0;
2309 }
2310
2311 if ((new_active) &&
2312 (old_active) &&
2313 (new_active->link == BOND_LINK_UP) &&
2314 IS_UP(new_active->dev)) {
Neil Hormane843fa52010-10-13 16:01:50 +00002315 block_netpoll_tx();
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002316 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 bond_change_active_slave(bond, new_active);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002318 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002319 unblock_netpoll_tx();
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002320 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 res = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002323 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324
2325 return res;
2326}
2327
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2329{
Wang Chen454d7c92008-11-12 23:37:49 -08002330 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
2332 info->bond_mode = bond->params.mode;
2333 info->miimon = bond->params.miimon;
2334
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07002335 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 info->num_slaves = bond->slave_cnt;
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07002337 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
2339 return 0;
2340}
2341
2342static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2343{
Wang Chen454d7c92008-11-12 23:37:49 -08002344 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 struct slave *slave;
Eric Dumazet689c96c2009-04-23 03:39:04 +00002346 int i, res = -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07002348 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349
2350 bond_for_each_slave(bond, slave, i) {
2351 if (i == (int)info->slave_id) {
Eric Dumazet689c96c2009-04-23 03:39:04 +00002352 res = 0;
2353 strcpy(info->slave_name, slave->dev->name);
2354 info->link = slave->link;
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002355 info->state = bond_slave_state(slave);
Eric Dumazet689c96c2009-04-23 03:39:04 +00002356 info->link_failure_count = slave->link_failure_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 break;
2358 }
2359 }
2360
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07002361 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362
Eric Dumazet689c96c2009-04-23 03:39:04 +00002363 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364}
2365
2366/*-------------------------------- Monitoring -------------------------------*/
2367
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002368
2369static int bond_miimon_inspect(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370{
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002371 struct slave *slave;
2372 int i, link_state, commit = 0;
Jiri Pirko41f89102009-04-24 03:57:29 +00002373 bool ignore_updelay;
2374
2375 ignore_updelay = !bond->curr_active_slave ? true : false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376
2377 bond_for_each_slave(bond, slave, i) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002378 slave->new_link = BOND_LINK_NOCHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002380 link_state = bond_check_dev_link(bond, slave->dev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
2382 switch (slave->link) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002383 case BOND_LINK_UP:
2384 if (link_state)
2385 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002387 slave->link = BOND_LINK_FAIL;
2388 slave->delay = bond->params.downdelay;
2389 if (slave->delay) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002390 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2391 bond->dev->name,
2392 (bond->params.mode ==
2393 BOND_MODE_ACTIVEBACKUP) ?
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002394 (bond_is_active_slave(slave) ?
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002395 "active " : "backup ") : "",
2396 slave->dev->name,
2397 bond->params.downdelay * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002399 /*FALLTHRU*/
2400 case BOND_LINK_FAIL:
2401 if (link_state) {
2402 /*
2403 * recovered before downdelay expired
2404 */
2405 slave->link = BOND_LINK_UP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 slave->jiffies = jiffies;
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002407 pr_info("%s: link status up again after %d ms for interface %s.\n",
2408 bond->dev->name,
2409 (bond->params.downdelay - slave->delay) *
2410 bond->params.miimon,
2411 slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002412 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002414
2415 if (slave->delay <= 0) {
2416 slave->new_link = BOND_LINK_DOWN;
2417 commit++;
2418 continue;
2419 }
2420
2421 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002424 case BOND_LINK_DOWN:
2425 if (!link_state)
2426 continue;
2427
2428 slave->link = BOND_LINK_BACK;
2429 slave->delay = bond->params.updelay;
2430
2431 if (slave->delay) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002432 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2433 bond->dev->name, slave->dev->name,
2434 ignore_updelay ? 0 :
2435 bond->params.updelay *
2436 bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002438 /*FALLTHRU*/
2439 case BOND_LINK_BACK:
2440 if (!link_state) {
2441 slave->link = BOND_LINK_DOWN;
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002442 pr_info("%s: link status down again after %d ms for interface %s.\n",
2443 bond->dev->name,
2444 (bond->params.updelay - slave->delay) *
2445 bond->params.miimon,
2446 slave->dev->name);
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002447
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002448 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002450
Jiri Pirko41f89102009-04-24 03:57:29 +00002451 if (ignore_updelay)
2452 slave->delay = 0;
2453
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002454 if (slave->delay <= 0) {
2455 slave->new_link = BOND_LINK_UP;
2456 commit++;
Jiri Pirko41f89102009-04-24 03:57:29 +00002457 ignore_updelay = false;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002458 continue;
2459 }
2460
2461 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 break;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002463 }
2464 }
2465
2466 return commit;
2467}
2468
2469static void bond_miimon_commit(struct bonding *bond)
2470{
2471 struct slave *slave;
2472 int i;
2473
2474 bond_for_each_slave(bond, slave, i) {
2475 switch (slave->new_link) {
2476 case BOND_LINK_NOCHANGE:
2477 continue;
2478
2479 case BOND_LINK_UP:
2480 slave->link = BOND_LINK_UP;
2481 slave->jiffies = jiffies;
2482
2483 if (bond->params.mode == BOND_MODE_8023AD) {
2484 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002485 bond_set_backup_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002486 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2487 /* make it immediately active */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002488 bond_set_active_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002489 } else if (slave != bond->primary_slave) {
2490 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002491 bond_set_backup_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002492 }
2493
Krzysztof Piotr Oledzki546add72010-10-06 14:28:22 -07002494 bond_update_speed_duplex(slave);
2495
Krzysztof Piotr Oledzki700c2a72010-10-06 14:25:06 -07002496 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2497 bond->dev->name, slave->dev->name,
2498 slave->speed, slave->duplex ? "full" : "half");
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002499
2500 /* notify ad that the link status has changed */
2501 if (bond->params.mode == BOND_MODE_8023AD)
2502 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2503
Holger Eitzenberger58402052008-12-09 23:07:13 -08002504 if (bond_is_lb(bond))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002505 bond_alb_handle_link_change(bond, slave,
2506 BOND_LINK_UP);
2507
2508 if (!bond->curr_active_slave ||
2509 (slave == bond->primary_slave))
2510 goto do_failover;
2511
2512 continue;
2513
2514 case BOND_LINK_DOWN:
Jay Vosburghfba4acd2008-10-30 17:41:14 -07002515 if (slave->link_failure_count < UINT_MAX)
2516 slave->link_failure_count++;
2517
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002518 slave->link = BOND_LINK_DOWN;
2519
2520 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2521 bond->params.mode == BOND_MODE_8023AD)
2522 bond_set_slave_inactive_flags(slave);
2523
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002524 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2525 bond->dev->name, slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002526
2527 if (bond->params.mode == BOND_MODE_8023AD)
2528 bond_3ad_handle_link_change(slave,
2529 BOND_LINK_DOWN);
2530
Jiri Pirkoae63e802009-05-27 05:42:36 +00002531 if (bond_is_lb(bond))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002532 bond_alb_handle_link_change(bond, slave,
2533 BOND_LINK_DOWN);
2534
2535 if (slave == bond->curr_active_slave)
2536 goto do_failover;
2537
2538 continue;
2539
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 default:
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002541 pr_err("%s: invalid new link %d on slave %s\n",
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002542 bond->dev->name, slave->new_link,
2543 slave->dev->name);
2544 slave->new_link = BOND_LINK_NOCHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002546 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 }
2548
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002549do_failover:
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002550 ASSERT_RTNL();
Neil Hormane843fa52010-10-13 16:01:50 +00002551 block_netpoll_tx();
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002552 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 bond_select_active_slave(bond);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002554 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002555 unblock_netpoll_tx();
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002556 }
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002557
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002558 bond_set_carrier(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559}
2560
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002561/*
2562 * bond_mii_monitor
2563 *
2564 * Really a wrapper that splits the mii monitor into two phases: an
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002565 * inspection, then (if inspection indicates something needs to be done)
2566 * an acquisition of appropriate locks followed by a commit phase to
2567 * implement whatever link state changes are indicated.
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002568 */
2569void bond_mii_monitor(struct work_struct *work)
2570{
2571 struct bonding *bond = container_of(work, struct bonding,
2572 mii_work.work);
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002573
2574 read_lock(&bond->lock);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002575 if (bond->kill_timers)
2576 goto out;
2577
2578 if (bond->slave_cnt == 0)
2579 goto re_arm;
Jay Vosburghb59f9f72008-06-13 18:12:03 -07002580
2581 if (bond->send_grat_arp) {
2582 read_lock(&bond->curr_slave_lock);
2583 bond_send_gratuitous_arp(bond);
2584 read_unlock(&bond->curr_slave_lock);
2585 }
2586
Brian Haley305d5522008-11-04 17:51:14 -08002587 if (bond->send_unsol_na) {
2588 read_lock(&bond->curr_slave_lock);
2589 bond_send_unsolicited_na(bond);
2590 read_unlock(&bond->curr_slave_lock);
2591 }
2592
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002593 if (bond_miimon_inspect(bond)) {
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002594 read_unlock(&bond->lock);
2595 rtnl_lock();
2596 read_lock(&bond->lock);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002597
2598 bond_miimon_commit(bond);
2599
Jay Vosburgh56556622008-01-17 16:25:03 -08002600 read_unlock(&bond->lock);
2601 rtnl_unlock(); /* might sleep, hold no other locks */
2602 read_lock(&bond->lock);
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002603 }
2604
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002605re_arm:
2606 if (bond->params.miimon)
2607 queue_delayed_work(bond->wq, &bond->mii_work,
2608 msecs_to_jiffies(bond->params.miimon));
2609out:
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002610 read_unlock(&bond->lock);
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002611}
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002612
Al Virod3bb52b2007-08-22 20:06:58 -04002613static __be32 bond_glean_dev_ip(struct net_device *dev)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002614{
2615 struct in_device *idev;
2616 struct in_ifaddr *ifa;
Al Viroa144ea42006-09-28 18:00:55 -07002617 __be32 addr = 0;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002618
2619 if (!dev)
2620 return 0;
2621
2622 rcu_read_lock();
Herbert Xue5ed6392005-10-03 14:35:55 -07002623 idev = __in_dev_get_rcu(dev);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002624 if (!idev)
2625 goto out;
2626
2627 ifa = idev->ifa_list;
2628 if (!ifa)
2629 goto out;
2630
2631 addr = ifa->ifa_local;
2632out:
2633 rcu_read_unlock();
2634 return addr;
2635}
2636
Al Virod3bb52b2007-08-22 20:06:58 -04002637static int bond_has_this_ip(struct bonding *bond, __be32 ip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002638{
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07002639 struct vlan_entry *vlan;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002640
2641 if (ip == bond->master_ip)
2642 return 1;
2643
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07002644 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002645 if (ip == vlan->vlan_ip)
2646 return 1;
2647 }
2648
2649 return 0;
2650}
2651
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002652/*
2653 * We go to the (large) trouble of VLAN tagging ARP frames because
2654 * switches in VLAN mode (especially if ports are configured as
2655 * "native" to a VLAN) might not pass non-tagged frames.
2656 */
Al Virod3bb52b2007-08-22 20:06:58 -04002657static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002658{
2659 struct sk_buff *skb;
2660
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002661 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002662 slave_dev->name, dest_ip, src_ip, vlan_id);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002663
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002664 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2665 NULL, slave_dev->dev_addr, NULL);
2666
2667 if (!skb) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002668 pr_err("ARP packet allocation failed\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002669 return;
2670 }
2671 if (vlan_id) {
2672 skb = vlan_put_tag(skb, vlan_id);
2673 if (!skb) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002674 pr_err("failed to insert VLAN tag\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002675 return;
2676 }
2677 }
2678 arp_xmit(skb);
2679}
2680
2681
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2683{
David S. Millerb23dd4f2011-03-02 14:31:35 -08002684 int i, vlan_id;
Al Virod3bb52b2007-08-22 20:06:58 -04002685 __be32 *targets = bond->params.arp_targets;
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07002686 struct vlan_entry *vlan;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002687 struct net_device *vlan_dev;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002688 struct rtable *rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689
Mitch Williams6b780562005-11-09 10:36:19 -08002690 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2691 if (!targets[i])
Brian Haley5a31bec2009-04-13 00:11:30 -07002692 break;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002693 pr_debug("basa: target %x\n", targets[i]);
Jay Vosburghf35188f2010-07-21 12:14:47 +00002694 if (!bond->vlgrp) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002695 pr_debug("basa: empty vlan: arp_send\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002696 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2697 bond->master_ip, 0);
2698 continue;
2699 }
2700
2701 /*
2702 * If VLANs are configured, we do a route lookup to
2703 * determine which VLAN interface would be used, so we
2704 * can tag the ARP with the proper VLAN tag.
2705 */
David S. Miller78fbfd82011-03-12 00:00:52 -05002706 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2707 RTO_ONLINK, 0);
David S. Millerb23dd4f2011-03-02 14:31:35 -08002708 if (IS_ERR(rt)) {
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002709 if (net_ratelimit()) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002710 pr_warning("%s: no route to arp_ip_target %pI4\n",
David S. Miller78fbfd82011-03-12 00:00:52 -05002711 bond->dev->name, &targets[i]);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002712 }
2713 continue;
2714 }
2715
2716 /*
2717 * This target is not on a VLAN
2718 */
Changli Gaod8d1f302010-06-10 23:31:35 -07002719 if (rt->dst.dev == bond->dev) {
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002720 ip_rt_put(rt);
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002721 pr_debug("basa: rtdev == bond->dev: arp_send\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002722 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2723 bond->master_ip, 0);
2724 continue;
2725 }
2726
2727 vlan_id = 0;
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07002728 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Dan Aloni5c15bde2007-03-02 20:44:51 -08002729 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
Changli Gaod8d1f302010-06-10 23:31:35 -07002730 if (vlan_dev == rt->dst.dev) {
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002731 vlan_id = vlan->vlan_id;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002732 pr_debug("basa: vlan match on %s %d\n",
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002733 vlan_dev->name, vlan_id);
2734 break;
2735 }
2736 }
2737
2738 if (vlan_id) {
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002739 ip_rt_put(rt);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002740 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2741 vlan->vlan_ip, vlan_id);
2742 continue;
2743 }
2744
2745 if (net_ratelimit()) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002746 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
David S. Miller78fbfd82011-03-12 00:00:52 -05002747 bond->dev->name, &targets[i],
Changli Gaod8d1f302010-06-10 23:31:35 -07002748 rt->dst.dev ? rt->dst.dev->name : "NULL");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002749 }
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002750 ip_rt_put(rt);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002751 }
2752}
2753
2754/*
2755 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2756 * for each VLAN above us.
Jay Vosburghb59f9f72008-06-13 18:12:03 -07002757 *
2758 * Caller must hold curr_slave_lock for read or better
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002759 */
2760static void bond_send_gratuitous_arp(struct bonding *bond)
2761{
2762 struct slave *slave = bond->curr_active_slave;
2763 struct vlan_entry *vlan;
2764 struct net_device *vlan_dev;
2765
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002766 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2767 bond->dev->name, slave ? slave->dev->name : "NULL");
Jay Vosburghb59f9f72008-06-13 18:12:03 -07002768
2769 if (!slave || !bond->send_grat_arp ||
2770 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002771 return;
2772
Jay Vosburghb59f9f72008-06-13 18:12:03 -07002773 bond->send_grat_arp--;
2774
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002775 if (bond->master_ip) {
2776 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
Moni Shoua1053f622007-10-09 19:43:42 -07002777 bond->master_ip, 0);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002778 }
2779
Jay Vosburghf35188f2010-07-21 12:14:47 +00002780 if (!bond->vlgrp)
2781 return;
2782
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002783 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Dan Aloni5c15bde2007-03-02 20:44:51 -08002784 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002785 if (vlan->vlan_ip) {
2786 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2787 vlan->vlan_ip, vlan->vlan_id);
2788 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 }
2790}
2791
Al Virod3bb52b2007-08-22 20:06:58 -04002792static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002793{
2794 int i;
Al Virod3bb52b2007-08-22 20:06:58 -04002795 __be32 *targets = bond->params.arp_targets;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002796
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002797 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002798 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002799 &sip, &tip, i, &targets[i],
2800 bond_has_this_ip(bond, tip));
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002801 if (sip == targets[i]) {
2802 if (bond_has_this_ip(bond, tip))
2803 slave->last_arp_rx = jiffies;
2804 return;
2805 }
2806 }
2807}
2808
2809static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2810{
2811 struct arphdr *arp;
2812 struct slave *slave;
2813 struct bonding *bond;
2814 unsigned char *arp_ptr;
Al Virod3bb52b2007-08-22 20:06:58 -04002815 __be32 sip, tip;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002816
Andy Gospodarek1f3c8802009-12-14 10:48:58 +00002817 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2818 /*
2819 * When using VLANS and bonding, dev and oriv_dev may be
2820 * incorrect if the physical interface supports VLAN
2821 * acceleration. With this change ARP validation now
2822 * works for hosts only reachable on the VLAN interface.
2823 */
2824 dev = vlan_dev_real_dev(dev);
2825 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2826 }
2827
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002828 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2829 goto out;
2830
Wang Chen454d7c92008-11-12 23:37:49 -08002831 bond = netdev_priv(dev);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002832 read_lock(&bond->lock);
2833
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002834 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002835 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2836 orig_dev ? orig_dev->name : "NULL");
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002837
2838 slave = bond_get_slave_by_dev(bond, orig_dev);
2839 if (!slave || !slave_do_arp_validate(bond, slave))
2840 goto out_unlock;
2841
Neil Hormanb3053252011-01-20 09:02:31 +00002842 skb = skb_share_check(skb, GFP_ATOMIC);
2843 if (!skb)
2844 goto out_unlock;
2845
Pavel Emelyanov988b7052008-03-03 12:20:57 -08002846 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002847 goto out_unlock;
2848
Arnaldo Carvalho de Melod0a92be2007-03-12 20:56:31 -03002849 arp = arp_hdr(skb);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002850 if (arp->ar_hln != dev->addr_len ||
2851 skb->pkt_type == PACKET_OTHERHOST ||
2852 skb->pkt_type == PACKET_LOOPBACK ||
2853 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2854 arp->ar_pro != htons(ETH_P_IP) ||
2855 arp->ar_pln != 4)
2856 goto out_unlock;
2857
2858 arp_ptr = (unsigned char *)(arp + 1);
2859 arp_ptr += dev->addr_len;
2860 memcpy(&sip, arp_ptr, 4);
2861 arp_ptr += 4 + dev->addr_len;
2862 memcpy(&tip, arp_ptr, 4);
2863
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002864 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002865 bond->dev->name, slave->dev->name, bond_slave_state(slave),
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002866 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2867 &sip, &tip);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002868
2869 /*
2870 * Backup slaves won't see the ARP reply, but do come through
2871 * here for each ARP probe (so we swap the sip/tip to validate
2872 * the probe). In a "redundant switch, common router" type of
2873 * configuration, the ARP probe will (hopefully) travel from
2874 * the active, through one switch, the router, then the other
2875 * switch before reaching the backup.
2876 */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002877 if (bond_is_active_slave(slave))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002878 bond_validate_arp(bond, slave, sip, tip);
2879 else
2880 bond_validate_arp(bond, slave, tip, sip);
2881
2882out_unlock:
2883 read_unlock(&bond->lock);
2884out:
2885 dev_kfree_skb(skb);
2886 return NET_RX_SUCCESS;
2887}
2888
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889/*
2890 * this function is called regularly to monitor each slave's link
2891 * ensuring that traffic is being sent and received when arp monitoring
2892 * is used in load-balancing mode. if the adapter has been dormant, then an
2893 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2894 * arp monitoring in active backup mode.
2895 */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07002896void bond_loadbalance_arp_mon(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897{
Jay Vosburgh1b76b312007-10-17 17:37:45 -07002898 struct bonding *bond = container_of(work, struct bonding,
2899 arp_work.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 struct slave *slave, *oldcurrent;
2901 int do_failover = 0;
2902 int delta_in_ticks;
2903 int i;
2904
2905 read_lock(&bond->lock);
2906
Jay Vosburgh5ce0da82008-05-17 21:10:07 -07002907 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002909 if (bond->kill_timers)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002912 if (bond->slave_cnt == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 goto re_arm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914
2915 read_lock(&bond->curr_slave_lock);
2916 oldcurrent = bond->curr_active_slave;
2917 read_unlock(&bond->curr_slave_lock);
2918
2919 /* see if any of the previous devices are up now (i.e. they have
2920 * xmt and rcv traffic). the curr_active_slave does not come into
2921 * the picture unless it is null. also, slave->jiffies is not needed
2922 * here because we send an arp on each slave and give a slave as
2923 * long as it needs to get the tx/rx within the delta.
2924 * TODO: what about up/down delay in arp mode? it wasn't here before
2925 * so it can wait
2926 */
2927 bond_for_each_slave(bond, slave, i) {
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002928 unsigned long trans_start = dev_trans_start(slave->dev);
2929
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 if (slave->link != BOND_LINK_UP) {
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002931 if (time_in_range(jiffies,
2932 trans_start - delta_in_ticks,
2933 trans_start + delta_in_ticks) &&
2934 time_in_range(jiffies,
2935 slave->dev->last_rx - delta_in_ticks,
2936 slave->dev->last_rx + delta_in_ticks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937
2938 slave->link = BOND_LINK_UP;
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002939 bond_set_active_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940
2941 /* primary_slave has no meaning in round-robin
2942 * mode. the window of a slave being up and
2943 * curr_active_slave being null after enslaving
2944 * is closed.
2945 */
2946 if (!oldcurrent) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002947 pr_info("%s: link status definitely up for interface %s, ",
2948 bond->dev->name,
2949 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 do_failover = 1;
2951 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002952 pr_info("%s: interface %s is now up\n",
2953 bond->dev->name,
2954 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 }
2956 }
2957 } else {
2958 /* slave->link == BOND_LINK_UP */
2959
2960 /* not all switches will respond to an arp request
2961 * when the source ip is 0, so don't take the link down
2962 * if we don't know our ip yet
2963 */
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002964 if (!time_in_range(jiffies,
2965 trans_start - delta_in_ticks,
2966 trans_start + 2 * delta_in_ticks) ||
2967 !time_in_range(jiffies,
2968 slave->dev->last_rx - delta_in_ticks,
2969 slave->dev->last_rx + 2 * delta_in_ticks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970
2971 slave->link = BOND_LINK_DOWN;
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002972 bond_set_backup_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002974 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002977 pr_info("%s: interface %s is now down.\n",
2978 bond->dev->name,
2979 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002981 if (slave == oldcurrent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 do_failover = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 }
2984 }
2985
2986 /* note: if switch is in round-robin mode, all links
2987 * must tx arp to ensure all links rx an arp - otherwise
2988 * links may oscillate or not come up at all; if switch is
2989 * in something like xor mode, there is nothing we can
2990 * do - all replies will be rx'ed on same link causing slaves
2991 * to be unstable during low/no traffic periods
2992 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002993 if (IS_UP(slave->dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 bond_arp_send_all(bond, slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 }
2996
2997 if (do_failover) {
Neil Hormane843fa52010-10-13 16:01:50 +00002998 block_netpoll_tx();
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002999 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000
3001 bond_select_active_slave(bond);
3002
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07003003 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00003004 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 }
3006
3007re_arm:
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003008 if (bond->params.arp_interval)
3009 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010out:
3011 read_unlock(&bond->lock);
3012}
3013
3014/*
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003015 * Called to inspect slaves for active-backup mode ARP monitor link state
3016 * changes. Sets new_link in slaves to specify what action should take
3017 * place for the slave. Returns 0 if no changes are found, >0 if changes
3018 * to link states must be committed.
3019 *
3020 * Called with bond->lock held for read.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 */
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003022static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 struct slave *slave;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003025 int i, commit = 0;
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003026 unsigned long trans_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 bond_for_each_slave(bond, slave, i) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003029 slave->new_link = BOND_LINK_NOCHANGE;
3030
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 if (slave->link != BOND_LINK_UP) {
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003032 if (time_in_range(jiffies,
3033 slave_last_rx(bond, slave) - delta_in_ticks,
3034 slave_last_rx(bond, slave) + delta_in_ticks)) {
3035
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003036 slave->new_link = BOND_LINK_UP;
3037 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003040 continue;
3041 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003043 /*
3044 * Give slaves 2*delta after being enslaved or made
3045 * active. This avoids bouncing, as the last receive
3046 * times need a full ARP monitor cycle to be updated.
3047 */
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003048 if (time_in_range(jiffies,
3049 slave->jiffies - delta_in_ticks,
3050 slave->jiffies + 2 * delta_in_ticks))
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003051 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003053 /*
3054 * Backup slave is down if:
3055 * - No current_arp_slave AND
3056 * - more than 3*delta since last receive AND
3057 * - the bond has an IP address
3058 *
3059 * Note: a non-null current_arp_slave indicates
3060 * the curr_active_slave went down and we are
3061 * searching for a new one; under this condition
3062 * we only take the curr_active_slave down - this
3063 * gives each slave a chance to tx/rx traffic
3064 * before being taken out
3065 */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00003066 if (!bond_is_active_slave(slave) &&
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003067 !bond->current_arp_slave &&
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003068 !time_in_range(jiffies,
3069 slave_last_rx(bond, slave) - delta_in_ticks,
3070 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3071
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003072 slave->new_link = BOND_LINK_DOWN;
3073 commit++;
3074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003076 /*
3077 * Active slave is down if:
3078 * - more than 2*delta since transmitting OR
3079 * - (more than 2*delta since receive AND
3080 * the bond has an IP address)
3081 */
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003082 trans_start = dev_trans_start(slave->dev);
Jiri Pirkoe30bc062011-03-12 03:14:37 +00003083 if (bond_is_active_slave(slave) &&
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003084 (!time_in_range(jiffies,
3085 trans_start - delta_in_ticks,
3086 trans_start + 2 * delta_in_ticks) ||
3087 !time_in_range(jiffies,
3088 slave_last_rx(bond, slave) - delta_in_ticks,
3089 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3090
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003091 slave->new_link = BOND_LINK_DOWN;
3092 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 }
3094 }
3095
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003096 return commit;
3097}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003099/*
3100 * Called to commit link state changes noted by inspection step of
3101 * active-backup mode ARP monitor.
3102 *
3103 * Called with RTNL and bond->lock for read.
3104 */
3105static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3106{
3107 struct slave *slave;
3108 int i;
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003109 unsigned long trans_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003111 bond_for_each_slave(bond, slave, i) {
3112 switch (slave->new_link) {
3113 case BOND_LINK_NOCHANGE:
3114 continue;
3115
3116 case BOND_LINK_UP:
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003117 trans_start = dev_trans_start(slave->dev);
Jiri Pirkob9f60252009-08-31 11:09:38 +00003118 if ((!bond->curr_active_slave &&
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003119 time_in_range(jiffies,
3120 trans_start - delta_in_ticks,
3121 trans_start + delta_in_ticks)) ||
Jiri Pirkob9f60252009-08-31 11:09:38 +00003122 bond->curr_active_slave != slave) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003123 slave->link = BOND_LINK_UP;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003124 bond->current_arp_slave = NULL;
3125
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003126 pr_info("%s: link status definitely up for interface %s.\n",
Jiri Pirkob9f60252009-08-31 11:09:38 +00003127 bond->dev->name, slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003128
Jiri Pirkob9f60252009-08-31 11:09:38 +00003129 if (!bond->curr_active_slave ||
3130 (slave == bond->primary_slave))
3131 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003132
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003133 }
3134
Jiri Pirkob9f60252009-08-31 11:09:38 +00003135 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003136
3137 case BOND_LINK_DOWN:
3138 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003141 slave->link = BOND_LINK_DOWN;
Jiri Pirkob9f60252009-08-31 11:09:38 +00003142 bond_set_slave_inactive_flags(slave);
3143
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003144 pr_info("%s: link status definitely down for interface %s, disabling it\n",
Jiri Pirkob9f60252009-08-31 11:09:38 +00003145 bond->dev->name, slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003146
3147 if (slave == bond->curr_active_slave) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003148 bond->current_arp_slave = NULL;
Jiri Pirkob9f60252009-08-31 11:09:38 +00003149 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003150 }
Jiri Pirkob9f60252009-08-31 11:09:38 +00003151
3152 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003153
3154 default:
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003155 pr_err("%s: impossible: new_link %d on slave %s\n",
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003156 bond->dev->name, slave->new_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157 slave->dev->name);
Jiri Pirkob9f60252009-08-31 11:09:38 +00003158 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160
Jiri Pirkob9f60252009-08-31 11:09:38 +00003161do_failover:
3162 ASSERT_RTNL();
Neil Hormane843fa52010-10-13 16:01:50 +00003163 block_netpoll_tx();
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003164 write_lock_bh(&bond->curr_slave_lock);
Jiri Pirkob9f60252009-08-31 11:09:38 +00003165 bond_select_active_slave(bond);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003166 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00003167 unblock_netpoll_tx();
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003168 }
3169
3170 bond_set_carrier(bond);
3171}
3172
3173/*
3174 * Send ARP probes for active-backup mode ARP monitor.
3175 *
3176 * Called with bond->lock held for read.
3177 */
3178static void bond_ab_arp_probe(struct bonding *bond)
3179{
3180 struct slave *slave;
3181 int i;
3182
3183 read_lock(&bond->curr_slave_lock);
3184
3185 if (bond->current_arp_slave && bond->curr_active_slave)
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003186 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3187 bond->current_arp_slave->dev->name,
3188 bond->curr_active_slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003189
3190 if (bond->curr_active_slave) {
3191 bond_arp_send_all(bond, bond->curr_active_slave);
3192 read_unlock(&bond->curr_slave_lock);
3193 return;
3194 }
3195
3196 read_unlock(&bond->curr_slave_lock);
3197
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198 /* if we don't have a curr_active_slave, search for the next available
3199 * backup slave from the current_arp_slave and make it the candidate
3200 * for becoming the curr_active_slave
3201 */
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003202
3203 if (!bond->current_arp_slave) {
3204 bond->current_arp_slave = bond->first_slave;
3205 if (!bond->current_arp_slave)
3206 return;
3207 }
3208
3209 bond_set_slave_inactive_flags(bond->current_arp_slave);
3210
3211 /* search for next candidate */
3212 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3213 if (IS_UP(slave->dev)) {
3214 slave->link = BOND_LINK_BACK;
3215 bond_set_slave_active_flags(slave);
3216 bond_arp_send_all(bond, slave);
3217 slave->jiffies = jiffies;
3218 bond->current_arp_slave = slave;
3219 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220 }
3221
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003222 /* if the link state is up at this point, we
3223 * mark it down - this can happen if we have
3224 * simultaneous link failures and
3225 * reselect_active_interface doesn't make this
3226 * one the current slave so it is still marked
3227 * up when it is actually down
3228 */
3229 if (slave->link == BOND_LINK_UP) {
3230 slave->link = BOND_LINK_DOWN;
3231 if (slave->link_failure_count < UINT_MAX)
3232 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003234 bond_set_slave_inactive_flags(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003236 pr_info("%s: backup interface %s is now down.\n",
3237 bond->dev->name, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238 }
3239 }
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003240}
3241
3242void bond_activebackup_arp_mon(struct work_struct *work)
3243{
3244 struct bonding *bond = container_of(work, struct bonding,
3245 arp_work.work);
3246 int delta_in_ticks;
3247
3248 read_lock(&bond->lock);
3249
3250 if (bond->kill_timers)
3251 goto out;
3252
3253 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3254
3255 if (bond->slave_cnt == 0)
3256 goto re_arm;
3257
Jay Vosburghb59f9f72008-06-13 18:12:03 -07003258 if (bond->send_grat_arp) {
3259 read_lock(&bond->curr_slave_lock);
3260 bond_send_gratuitous_arp(bond);
3261 read_unlock(&bond->curr_slave_lock);
3262 }
3263
Brian Haley305d5522008-11-04 17:51:14 -08003264 if (bond->send_unsol_na) {
3265 read_lock(&bond->curr_slave_lock);
3266 bond_send_unsolicited_na(bond);
3267 read_unlock(&bond->curr_slave_lock);
3268 }
3269
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003270 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3271 read_unlock(&bond->lock);
3272 rtnl_lock();
3273 read_lock(&bond->lock);
3274
3275 bond_ab_arp_commit(bond, delta_in_ticks);
3276
3277 read_unlock(&bond->lock);
3278 rtnl_unlock();
3279 read_lock(&bond->lock);
3280 }
3281
3282 bond_ab_arp_probe(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283
3284re_arm:
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003285 if (bond->params.arp_interval)
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003286 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287out:
3288 read_unlock(&bond->lock);
3289}
3290
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291/*-------------------------- netdev event handling --------------------------*/
3292
3293/*
3294 * Change device name
3295 */
3296static int bond_event_changename(struct bonding *bond)
3297{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 bond_remove_proc_entry(bond);
3299 bond_create_proc_entry(bond);
Stephen Hemminger7e083842009-06-12 19:02:46 +00003300
Taku Izumif073c7c2010-12-09 15:17:13 +00003301 bond_debug_reregister(bond);
3302
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 return NOTIFY_DONE;
3304}
3305
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003306static int bond_master_netdev_event(unsigned long event,
3307 struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308{
Wang Chen454d7c92008-11-12 23:37:49 -08003309 struct bonding *event_bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310
3311 switch (event) {
3312 case NETDEV_CHANGENAME:
3313 return bond_event_changename(event_bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314 default:
3315 break;
3316 }
3317
3318 return NOTIFY_DONE;
3319}
3320
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003321static int bond_slave_netdev_event(unsigned long event,
3322 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323{
3324 struct net_device *bond_dev = slave_dev->master;
Wang Chen454d7c92008-11-12 23:37:49 -08003325 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326
3327 switch (event) {
3328 case NETDEV_UNREGISTER:
3329 if (bond_dev) {
Jay Vosburgh1284cd32007-10-15 16:44:27 -07003330 if (bond->setup_by_slave)
3331 bond_release_and_destroy(bond_dev, slave_dev);
3332 else
3333 bond_release(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 }
3335 break;
3336 case NETDEV_CHANGE:
Jay Vosburgh17d04502009-03-18 18:38:25 -07003337 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3338 struct slave *slave;
3339
3340 slave = bond_get_slave_by_dev(bond, slave_dev);
3341 if (slave) {
3342 u16 old_speed = slave->speed;
David Decotigny65cce192011-04-13 15:22:30 +00003343 u8 old_duplex = slave->duplex;
Jay Vosburgh17d04502009-03-18 18:38:25 -07003344
3345 bond_update_speed_duplex(slave);
3346
3347 if (bond_is_lb(bond))
3348 break;
3349
3350 if (old_speed != slave->speed)
3351 bond_3ad_adapter_speed_changed(slave);
3352 if (old_duplex != slave->duplex)
3353 bond_3ad_adapter_duplex_changed(slave);
3354 }
3355 }
3356
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357 break;
3358 case NETDEV_DOWN:
3359 /*
3360 * ... Or is it this?
3361 */
3362 break;
3363 case NETDEV_CHANGEMTU:
3364 /*
3365 * TODO: Should slaves be allowed to
3366 * independently alter their MTU? For
3367 * an active-backup bond, slaves need
3368 * not be the same type of device, so
3369 * MTUs may vary. For other modes,
3370 * slaves arguably should have the
3371 * same MTUs. To do this, we'd need to
3372 * take over the slave's change_mtu
3373 * function for the duration of their
3374 * servitude.
3375 */
3376 break;
3377 case NETDEV_CHANGENAME:
3378 /*
3379 * TODO: handle changing the primary's name
3380 */
3381 break;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04003382 case NETDEV_FEAT_CHANGE:
3383 bond_compute_features(bond);
3384 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 default:
3386 break;
3387 }
3388
3389 return NOTIFY_DONE;
3390}
3391
3392/*
3393 * bond_netdev_event: handle netdev notifier chain events.
3394 *
3395 * This function receives events for the netdev chain. The caller (an
Alan Sterne041c682006-03-27 01:16:30 -08003396 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397 * locks for us to safely manipulate the slave devices (RTNL lock,
3398 * dev_probe_lock).
3399 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003400static int bond_netdev_event(struct notifier_block *this,
3401 unsigned long event, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402{
3403 struct net_device *event_dev = (struct net_device *)ptr;
3404
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003405 pr_debug("event_dev: %s, event: %lx\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003406 event_dev ? event_dev->name : "None",
3407 event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408
Jay Vosburgh0b680e72006-09-22 21:54:10 -07003409 if (!(event_dev->priv_flags & IFF_BONDING))
3410 return NOTIFY_DONE;
3411
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412 if (event_dev->flags & IFF_MASTER) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003413 pr_debug("IFF_MASTER\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 return bond_master_netdev_event(event, event_dev);
3415 }
3416
3417 if (event_dev->flags & IFF_SLAVE) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003418 pr_debug("IFF_SLAVE\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419 return bond_slave_netdev_event(event, event_dev);
3420 }
3421
3422 return NOTIFY_DONE;
3423}
3424
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003425/*
3426 * bond_inetaddr_event: handle inetaddr notifier chain events.
3427 *
3428 * We keep track of device IPs primarily to use as source addresses in
3429 * ARP monitor probes (rather than spewing out broadcasts all the time).
3430 *
3431 * We track one IP for the main device (if it has one), plus one per VLAN.
3432 */
3433static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3434{
3435 struct in_ifaddr *ifa = ptr;
3436 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00003437 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07003438 struct bonding *bond;
3439 struct vlan_entry *vlan;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003440
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00003441 list_for_each_entry(bond, &bn->dev_list, bond_list) {
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003442 if (bond->dev == event_dev) {
3443 switch (event) {
3444 case NETDEV_UP:
3445 bond->master_ip = ifa->ifa_local;
3446 return NOTIFY_OK;
3447 case NETDEV_DOWN:
3448 bond->master_ip = bond_glean_dev_ip(bond->dev);
3449 return NOTIFY_OK;
3450 default:
3451 return NOTIFY_DONE;
3452 }
3453 }
3454
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07003455 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Jay Vosburghf35188f2010-07-21 12:14:47 +00003456 if (!bond->vlgrp)
3457 continue;
Dan Aloni5c15bde2007-03-02 20:44:51 -08003458 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003459 if (vlan_dev == event_dev) {
3460 switch (event) {
3461 case NETDEV_UP:
3462 vlan->vlan_ip = ifa->ifa_local;
3463 return NOTIFY_OK;
3464 case NETDEV_DOWN:
3465 vlan->vlan_ip =
3466 bond_glean_dev_ip(vlan_dev);
3467 return NOTIFY_OK;
3468 default:
3469 return NOTIFY_DONE;
3470 }
3471 }
3472 }
3473 }
3474 return NOTIFY_DONE;
3475}
3476
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477static struct notifier_block bond_netdev_notifier = {
3478 .notifier_call = bond_netdev_event,
3479};
3480
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003481static struct notifier_block bond_inetaddr_notifier = {
3482 .notifier_call = bond_inetaddr_event,
3483};
3484
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485/*-------------------------- Packet type handling ---------------------------*/
3486
3487/* register to receive lacpdus on a bond */
3488static void bond_register_lacpdu(struct bonding *bond)
3489{
3490 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3491
3492 /* initialize packet type */
3493 pk_type->type = PKT_TYPE_LACPDU;
3494 pk_type->dev = bond->dev;
3495 pk_type->func = bond_3ad_lacpdu_recv;
3496
3497 dev_add_pack(pk_type);
3498}
3499
3500/* unregister to receive lacpdus on a bond */
3501static void bond_unregister_lacpdu(struct bonding *bond)
3502{
3503 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3504}
3505
Jay Vosburghf5b2b962006-09-22 21:54:53 -07003506void bond_register_arp(struct bonding *bond)
3507{
3508 struct packet_type *pt = &bond->arp_mon_pt;
3509
Jay Vosburghc4f283b2007-02-28 17:03:20 -08003510 if (pt->type)
3511 return;
3512
Jay Vosburghf5b2b962006-09-22 21:54:53 -07003513 pt->type = htons(ETH_P_ARP);
Jay Vosburghe245cb72007-02-28 17:03:27 -08003514 pt->dev = bond->dev;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07003515 pt->func = bond_arp_rcv;
3516 dev_add_pack(pt);
3517}
3518
3519void bond_unregister_arp(struct bonding *bond)
3520{
Jay Vosburghc4f283b2007-02-28 17:03:20 -08003521 struct packet_type *pt = &bond->arp_mon_pt;
3522
3523 dev_remove_pack(pt);
3524 pt->type = 0;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07003525}
3526
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003527/*---------------------------- Hashing Policies -----------------------------*/
3528
3529/*
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003530 * Hash for the output device based upon layer 2 and layer 3 data. If
3531 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3532 */
Jasper Spaansa361c832009-10-23 04:09:24 +00003533static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003534{
3535 struct ethhdr *data = (struct ethhdr *)skb->data;
3536 struct iphdr *iph = ip_hdr(skb);
3537
Brian Haleyf14c4e42008-09-02 10:08:08 -04003538 if (skb->protocol == htons(ETH_P_IP)) {
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003539 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
Jasper Spaansd3da6832009-10-23 04:08:46 +00003540 (data->h_dest[5] ^ data->h_source[5])) % count;
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003541 }
3542
Jasper Spaansd3da6832009-10-23 04:08:46 +00003543 return (data->h_dest[5] ^ data->h_source[5]) % count;
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003544}
3545
3546/*
Michael Opdenacker59c51592007-05-09 08:57:56 +02003547 * Hash for the output device based upon layer 3 and layer 4 data. If
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003548 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3549 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3550 */
Jasper Spaansa361c832009-10-23 04:09:24 +00003551static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003552{
3553 struct ethhdr *data = (struct ethhdr *)skb->data;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07003554 struct iphdr *iph = ip_hdr(skb);
Al Virod3bb52b2007-08-22 20:06:58 -04003555 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003556 int layer4_xor = 0;
3557
Brian Haleyf14c4e42008-09-02 10:08:08 -04003558 if (skb->protocol == htons(ETH_P_IP)) {
3559 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003560 (iph->protocol == IPPROTO_TCP ||
3561 iph->protocol == IPPROTO_UDP)) {
Al Virod3bb52b2007-08-22 20:06:58 -04003562 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003563 }
3564 return (layer4_xor ^
3565 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3566
3567 }
3568
Jasper Spaansd3da6832009-10-23 04:08:46 +00003569 return (data->h_dest[5] ^ data->h_source[5]) % count;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003570}
3571
3572/*
3573 * Hash for the output device based upon layer 2 data
3574 */
Jasper Spaansa361c832009-10-23 04:09:24 +00003575static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003576{
3577 struct ethhdr *data = (struct ethhdr *)skb->data;
3578
Jasper Spaansd3da6832009-10-23 04:08:46 +00003579 return (data->h_dest[5] ^ data->h_source[5]) % count;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003580}
3581
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582/*-------------------------- Device entry points ----------------------------*/
3583
3584static int bond_open(struct net_device *bond_dev)
3585{
Wang Chen454d7c92008-11-12 23:37:49 -08003586 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587
3588 bond->kill_timers = 0;
3589
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00003590 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3591
Holger Eitzenberger58402052008-12-09 23:07:13 -08003592 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 /* bond_alb_initialize must be called before the timer
3594 * is started.
3595 */
3596 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3597 /* something went wrong - fail the open operation */
stephen hemmingerb4739462010-01-25 23:34:15 +00003598 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599 }
3600
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003601 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3602 queue_delayed_work(bond->wq, &bond->alb_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603 }
3604
3605 if (bond->params.miimon) { /* link check interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003606 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3607 queue_delayed_work(bond->wq, &bond->mii_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 }
3609
3610 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003611 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3612 INIT_DELAYED_WORK(&bond->arp_work,
3613 bond_activebackup_arp_mon);
3614 else
3615 INIT_DELAYED_WORK(&bond->arp_work,
3616 bond_loadbalance_arp_mon);
3617
3618 queue_delayed_work(bond->wq, &bond->arp_work, 0);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07003619 if (bond->params.arp_validate)
3620 bond_register_arp(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 }
3622
3623 if (bond->params.mode == BOND_MODE_8023AD) {
Adrian Bunka40745f2007-10-24 18:27:43 +02003624 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003625 queue_delayed_work(bond->wq, &bond->ad_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 /* register to receive LACPDUs */
3627 bond_register_lacpdu(bond);
Jay Vosburghfd989c82008-11-04 17:51:16 -08003628 bond_3ad_initiate_agg_selection(bond, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 }
3630
3631 return 0;
3632}
3633
3634static int bond_close(struct net_device *bond_dev)
3635{
Wang Chen454d7c92008-11-12 23:37:49 -08003636 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003637
3638 if (bond->params.mode == BOND_MODE_8023AD) {
3639 /* Unregister the receive of LACPDUs */
3640 bond_unregister_lacpdu(bond);
3641 }
3642
Jay Vosburghf5b2b962006-09-22 21:54:53 -07003643 if (bond->params.arp_validate)
3644 bond_unregister_arp(bond);
3645
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 write_lock_bh(&bond->lock);
3647
Jay Vosburghb59f9f72008-06-13 18:12:03 -07003648 bond->send_grat_arp = 0;
Brian Haley305d5522008-11-04 17:51:14 -08003649 bond->send_unsol_na = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650
3651 /* signal timers not to re-arm */
3652 bond->kill_timers = 1;
3653
3654 write_unlock_bh(&bond->lock);
3655
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656 if (bond->params.miimon) { /* link check interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003657 cancel_delayed_work(&bond->mii_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658 }
3659
3660 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003661 cancel_delayed_work(&bond->arp_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662 }
3663
3664 switch (bond->params.mode) {
3665 case BOND_MODE_8023AD:
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003666 cancel_delayed_work(&bond->ad_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 break;
3668 case BOND_MODE_TLB:
3669 case BOND_MODE_ALB:
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003670 cancel_delayed_work(&bond->alb_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671 break;
3672 default:
3673 break;
3674 }
3675
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00003676 if (delayed_work_pending(&bond->mcast_work))
3677 cancel_delayed_work(&bond->mcast_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678
Holger Eitzenberger58402052008-12-09 23:07:13 -08003679 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 /* Must be called only after all
3681 * slaves have been released
3682 */
3683 bond_alb_deinitialize(bond);
3684 }
3685
3686 return 0;
3687}
3688
Eric Dumazet28172732010-07-07 14:58:56 -07003689static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3690 struct rtnl_link_stats64 *stats)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691{
Wang Chen454d7c92008-11-12 23:37:49 -08003692 struct bonding *bond = netdev_priv(bond_dev);
Eric Dumazet28172732010-07-07 14:58:56 -07003693 struct rtnl_link_stats64 temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 struct slave *slave;
3695 int i;
3696
Eric Dumazet28172732010-07-07 14:58:56 -07003697 memset(stats, 0, sizeof(*stats));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698
3699 read_lock_bh(&bond->lock);
3700
3701 bond_for_each_slave(bond, slave, i) {
Ben Hutchingsbe1f3c22010-06-08 07:19:54 +00003702 const struct rtnl_link_stats64 *sstats =
Eric Dumazet28172732010-07-07 14:58:56 -07003703 dev_get_stats(slave->dev, &temp);
Stephen Hemmingereeda3fd2008-11-19 21:40:23 -08003704
Eric Dumazet28172732010-07-07 14:58:56 -07003705 stats->rx_packets += sstats->rx_packets;
3706 stats->rx_bytes += sstats->rx_bytes;
3707 stats->rx_errors += sstats->rx_errors;
3708 stats->rx_dropped += sstats->rx_dropped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709
Eric Dumazet28172732010-07-07 14:58:56 -07003710 stats->tx_packets += sstats->tx_packets;
3711 stats->tx_bytes += sstats->tx_bytes;
3712 stats->tx_errors += sstats->tx_errors;
3713 stats->tx_dropped += sstats->tx_dropped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714
Eric Dumazet28172732010-07-07 14:58:56 -07003715 stats->multicast += sstats->multicast;
3716 stats->collisions += sstats->collisions;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717
Eric Dumazet28172732010-07-07 14:58:56 -07003718 stats->rx_length_errors += sstats->rx_length_errors;
3719 stats->rx_over_errors += sstats->rx_over_errors;
3720 stats->rx_crc_errors += sstats->rx_crc_errors;
3721 stats->rx_frame_errors += sstats->rx_frame_errors;
3722 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3723 stats->rx_missed_errors += sstats->rx_missed_errors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724
Eric Dumazet28172732010-07-07 14:58:56 -07003725 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3726 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3727 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3728 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3729 stats->tx_window_errors += sstats->tx_window_errors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730 }
3731
3732 read_unlock_bh(&bond->lock);
3733
3734 return stats;
3735}
3736
3737static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3738{
3739 struct net_device *slave_dev = NULL;
3740 struct ifbond k_binfo;
3741 struct ifbond __user *u_binfo = NULL;
3742 struct ifslave k_sinfo;
3743 struct ifslave __user *u_sinfo = NULL;
3744 struct mii_ioctl_data *mii = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745 int res = 0;
3746
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003747 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748
3749 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003750 case SIOCGMIIPHY:
3751 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003752 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003753 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003754
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755 mii->phy_id = 0;
3756 /* Fall Through */
3757 case SIOCGMIIREG:
3758 /*
3759 * We do this again just in case we were called by SIOCGMIIREG
3760 * instead of SIOCGMIIPHY.
3761 */
3762 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003763 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003765
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766
3767 if (mii->reg_num == 1) {
Wang Chen454d7c92008-11-12 23:37:49 -08003768 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 mii->val_out = 0;
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07003770 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771 read_lock(&bond->curr_slave_lock);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003772 if (netif_carrier_ok(bond->dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 mii->val_out = BMSR_LSTATUS;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003774
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 read_unlock(&bond->curr_slave_lock);
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07003776 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777 }
3778
3779 return 0;
3780 case BOND_INFO_QUERY_OLD:
3781 case SIOCBONDINFOQUERY:
3782 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3783
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003784 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786
3787 res = bond_info_query(bond_dev, &k_binfo);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003788 if (res == 0 &&
3789 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3790 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791
3792 return res;
3793 case BOND_SLAVE_INFO_QUERY_OLD:
3794 case SIOCBONDSLAVEINFOQUERY:
3795 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3796
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003797 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799
3800 res = bond_slave_info_query(bond_dev, &k_sinfo);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003801 if (res == 0 &&
3802 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3803 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804
3805 return res;
3806 default:
3807 /* Go on */
3808 break;
3809 }
3810
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003811 if (!capable(CAP_NET_ADMIN))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00003814 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003816 pr_debug("slave_dev=%p:\n", slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003818 if (!slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 res = -ENODEV;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003820 else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003821 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822 switch (cmd) {
3823 case BOND_ENSLAVE_OLD:
3824 case SIOCBONDENSLAVE:
3825 res = bond_enslave(bond_dev, slave_dev);
3826 break;
3827 case BOND_RELEASE_OLD:
3828 case SIOCBONDRELEASE:
3829 res = bond_release(bond_dev, slave_dev);
3830 break;
3831 case BOND_SETHWADDR_OLD:
3832 case SIOCBONDSETHWADDR:
3833 res = bond_sethwaddr(bond_dev, slave_dev);
3834 break;
3835 case BOND_CHANGE_ACTIVE_OLD:
3836 case SIOCBONDCHANGEACTIVE:
3837 res = bond_ioctl_change_active(bond_dev, slave_dev);
3838 break;
3839 default:
3840 res = -EOPNOTSUPP;
3841 }
3842
3843 dev_put(slave_dev);
3844 }
3845
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846 return res;
3847}
3848
Jiri Pirko22bedad32010-04-01 21:22:57 +00003849static bool bond_addr_in_mc_list(unsigned char *addr,
3850 struct netdev_hw_addr_list *list,
3851 int addrlen)
3852{
3853 struct netdev_hw_addr *ha;
3854
3855 netdev_hw_addr_list_for_each(ha, list)
3856 if (!memcmp(ha->addr, addr, addrlen))
3857 return true;
3858
3859 return false;
3860}
3861
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862static void bond_set_multicast_list(struct net_device *bond_dev)
3863{
Wang Chen454d7c92008-11-12 23:37:49 -08003864 struct bonding *bond = netdev_priv(bond_dev);
Jiri Pirko22bedad32010-04-01 21:22:57 +00003865 struct netdev_hw_addr *ha;
3866 bool found;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868 /*
3869 * Do promisc before checking multicast_mode
3870 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003871 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
Wang Chen7e1a1ac2008-07-14 20:51:36 -07003872 /*
3873 * FIXME: Need to handle the error when one of the multi-slaves
3874 * encounters error.
3875 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876 bond_set_promiscuity(bond, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003878
3879 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 bond_set_promiscuity(bond, -1);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003881
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882
3883 /* set allmulti flag to slaves */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003884 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
Wang Chen7e1a1ac2008-07-14 20:51:36 -07003885 /*
3886 * FIXME: Need to handle the error when one of the multi-slaves
3887 * encounters error.
3888 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889 bond_set_allmulti(bond, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003891
3892 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 bond_set_allmulti(bond, -1);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003894
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895
Jay Vosburgh80ee5ad2008-01-29 18:07:44 -08003896 read_lock(&bond->lock);
3897
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898 bond->flags = bond_dev->flags;
3899
3900 /* looking for addresses to add to slaves' mc list */
Jiri Pirko22bedad32010-04-01 21:22:57 +00003901 netdev_for_each_mc_addr(ha, bond_dev) {
3902 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3903 bond_dev->addr_len);
3904 if (!found)
3905 bond_mc_add(bond, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906 }
3907
3908 /* looking for addresses to delete from slaves' list */
Jiri Pirko22bedad32010-04-01 21:22:57 +00003909 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3910 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3911 bond_dev->addr_len);
3912 if (!found)
3913 bond_mc_del(bond, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914 }
3915
3916 /* save master's multicast list */
Jiri Pirko22bedad32010-04-01 21:22:57 +00003917 __hw_addr_flush(&bond->mc_list);
3918 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3919 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920
Jay Vosburgh80ee5ad2008-01-29 18:07:44 -08003921 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922}
3923
Stephen Hemminger00829822008-11-20 20:14:53 -08003924static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3925{
3926 struct bonding *bond = netdev_priv(dev);
3927 struct slave *slave = bond->first_slave;
3928
3929 if (slave) {
3930 const struct net_device_ops *slave_ops
3931 = slave->dev->netdev_ops;
3932 if (slave_ops->ndo_neigh_setup)
Patrick McHardy72e22402009-03-05 01:57:44 -08003933 return slave_ops->ndo_neigh_setup(slave->dev, parms);
Stephen Hemminger00829822008-11-20 20:14:53 -08003934 }
3935 return 0;
3936}
3937
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938/*
3939 * Change the MTU of all of a master's slaves to match the master
3940 */
3941static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3942{
Wang Chen454d7c92008-11-12 23:37:49 -08003943 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003944 struct slave *slave, *stop_at;
3945 int res = 0;
3946 int i;
3947
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003948 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003949 (bond_dev ? bond_dev->name : "None"), new_mtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950
3951 /* Can't hold bond->lock with bh disabled here since
3952 * some base drivers panic. On the other hand we can't
3953 * hold bond->lock without bh disabled because we'll
3954 * deadlock. The only solution is to rely on the fact
3955 * that we're under rtnl_lock here, and the slaves
3956 * list won't change. This doesn't solve the problem
3957 * of setting the slave's MTU while it is
3958 * transmitting, but the assumption is that the base
3959 * driver can handle that.
3960 *
3961 * TODO: figure out a way to safely iterate the slaves
3962 * list, but without holding a lock around the actual
3963 * call to the base driver.
3964 */
3965
3966 bond_for_each_slave(bond, slave, i) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003967 pr_debug("s %p s->p %p c_m %p\n",
3968 slave,
3969 slave->prev,
3970 slave->dev->netdev_ops->ndo_change_mtu);
Mitch Williamse944ef72005-11-09 10:36:50 -08003971
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 res = dev_set_mtu(slave->dev, new_mtu);
3973
3974 if (res) {
3975 /* If we failed to set the slave's mtu to the new value
3976 * we must abort the operation even in ACTIVE_BACKUP
3977 * mode, because if we allow the backup slaves to have
3978 * different mtu values than the active slave we'll
3979 * need to change their mtu when doing a failover. That
3980 * means changing their mtu from timer context, which
3981 * is probably not a good idea.
3982 */
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003983 pr_debug("err %d %s\n", res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984 goto unwind;
3985 }
3986 }
3987
3988 bond_dev->mtu = new_mtu;
3989
3990 return 0;
3991
3992unwind:
3993 /* unwind from head to the slave that failed */
3994 stop_at = slave;
3995 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3996 int tmp_res;
3997
3998 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3999 if (tmp_res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004000 pr_debug("unwind err %d dev %s\n",
4001 tmp_res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002 }
4003 }
4004
4005 return res;
4006}
4007
4008/*
4009 * Change HW address
4010 *
4011 * Note that many devices must be down to change the HW address, and
4012 * downing the master releases all slaves. We can make bonds full of
4013 * bonding devices to test this, however.
4014 */
4015static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4016{
Wang Chen454d7c92008-11-12 23:37:49 -08004017 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 struct sockaddr *sa = addr, tmp_sa;
4019 struct slave *slave, *stop_at;
4020 int res = 0;
4021 int i;
4022
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004023 if (bond->params.mode == BOND_MODE_ALB)
4024 return bond_alb_set_mac_address(bond_dev, addr);
4025
4026
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004027 pr_debug("bond=%p, name=%s\n",
4028 bond, bond_dev ? bond_dev->name : "None");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029
Jay Vosburghdd957c52007-10-09 19:57:24 -07004030 /*
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004031 * If fail_over_mac is set to active, do nothing and return
4032 * success. Returning an error causes ifenslave to fail.
Jay Vosburghdd957c52007-10-09 19:57:24 -07004033 */
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004034 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
Jay Vosburghdd957c52007-10-09 19:57:24 -07004035 return 0;
Moni Shoua2ab82852007-10-09 19:43:39 -07004036
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004037 if (!is_valid_ether_addr(sa->sa_data))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 return -EADDRNOTAVAIL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039
4040 /* Can't hold bond->lock with bh disabled here since
4041 * some base drivers panic. On the other hand we can't
4042 * hold bond->lock without bh disabled because we'll
4043 * deadlock. The only solution is to rely on the fact
4044 * that we're under rtnl_lock here, and the slaves
4045 * list won't change. This doesn't solve the problem
4046 * of setting the slave's hw address while it is
4047 * transmitting, but the assumption is that the base
4048 * driver can handle that.
4049 *
4050 * TODO: figure out a way to safely iterate the slaves
4051 * list, but without holding a lock around the actual
4052 * call to the base driver.
4053 */
4054
4055 bond_for_each_slave(bond, slave, i) {
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004056 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08004057 pr_debug("slave %p %s\n", slave, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004059 if (slave_ops->ndo_set_mac_address == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060 res = -EOPNOTSUPP;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08004061 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062 goto unwind;
4063 }
4064
4065 res = dev_set_mac_address(slave->dev, addr);
4066 if (res) {
4067 /* TODO: consider downing the slave
4068 * and retry ?
4069 * User should expect communications
4070 * breakage anyway until ARP finish
4071 * updating, so...
4072 */
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08004073 pr_debug("err %d %s\n", res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004074 goto unwind;
4075 }
4076 }
4077
4078 /* success */
4079 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4080 return 0;
4081
4082unwind:
4083 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4084 tmp_sa.sa_family = bond_dev->type;
4085
4086 /* unwind from head to the slave that failed */
4087 stop_at = slave;
4088 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4089 int tmp_res;
4090
4091 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4092 if (tmp_res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004093 pr_debug("unwind err %d dev %s\n",
4094 tmp_res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095 }
4096 }
4097
4098 return res;
4099}
4100
4101static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4102{
Wang Chen454d7c92008-11-12 23:37:49 -08004103 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104 struct slave *slave, *start_at;
Jay Vosburghcf5f9042007-10-17 17:37:47 -07004105 int i, slave_no, res = 1;
Andy Gospodareka2fd9402010-03-25 14:49:05 +00004106 struct iphdr *iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107
4108 read_lock(&bond->lock);
4109
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004110 if (!BOND_IS_OK(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004111 goto out;
Jay Vosburghcf5f9042007-10-17 17:37:47 -07004112 /*
Andy Gospodareka2fd9402010-03-25 14:49:05 +00004113 * Start with the curr_active_slave that joined the bond as the
4114 * default for sending IGMP traffic. For failover purposes one
4115 * needs to maintain some consistency for the interface that will
4116 * send the join/membership reports. The curr_active_slave found
4117 * will send all of this type of traffic.
Jay Vosburghcf5f9042007-10-17 17:37:47 -07004118 */
Eric Dumazet00ae7022010-03-30 23:08:37 +00004119 if ((iph->protocol == IPPROTO_IGMP) &&
Andy Gospodareka2fd9402010-03-25 14:49:05 +00004120 (skb->protocol == htons(ETH_P_IP))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121
Andy Gospodareka2fd9402010-03-25 14:49:05 +00004122 read_lock(&bond->curr_slave_lock);
4123 slave = bond->curr_active_slave;
4124 read_unlock(&bond->curr_slave_lock);
4125
4126 if (!slave)
4127 goto out;
4128 } else {
4129 /*
4130 * Concurrent TX may collide on rr_tx_counter; we accept
4131 * that as being rare enough not to justify using an
4132 * atomic op here.
4133 */
4134 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4135
4136 bond_for_each_slave(bond, slave, i) {
4137 slave_no--;
4138 if (slave_no < 0)
4139 break;
4140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004141 }
4142
Jay Vosburghcf5f9042007-10-17 17:37:47 -07004143 start_at = slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004144 bond_for_each_slave_from(bond, slave, i, start_at) {
4145 if (IS_UP(slave->dev) &&
4146 (slave->link == BOND_LINK_UP) &&
Jiri Pirkoe30bc062011-03-12 03:14:37 +00004147 bond_is_active_slave(slave)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148 res = bond_dev_queue_xmit(bond, skb, slave->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004149 break;
4150 }
4151 }
4152
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153out:
4154 if (res) {
4155 /* no suitable interface, frame not sent */
4156 dev_kfree_skb(skb);
4157 }
4158 read_unlock(&bond->lock);
Patrick McHardyec634fe2009-07-05 19:23:38 -07004159 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004160}
4161
John W. Linville075897c2005-09-28 17:50:53 -04004162
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163/*
4164 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4165 * the bond has a usable interface.
4166 */
4167static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4168{
Wang Chen454d7c92008-11-12 23:37:49 -08004169 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170 int res = 1;
4171
Linus Torvalds1da177e2005-04-16 15:20:36 -07004172 read_lock(&bond->lock);
4173 read_lock(&bond->curr_slave_lock);
4174
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004175 if (!BOND_IS_OK(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004176 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177
John W. Linville075897c2005-09-28 17:50:53 -04004178 if (!bond->curr_active_slave)
4179 goto out;
4180
John W. Linville075897c2005-09-28 17:50:53 -04004181 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4182
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183out:
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004184 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185 /* no suitable interface, frame not sent */
4186 dev_kfree_skb(skb);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004187
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 read_unlock(&bond->curr_slave_lock);
4189 read_unlock(&bond->lock);
Patrick McHardyec634fe2009-07-05 19:23:38 -07004190 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191}
4192
4193/*
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004194 * In bond_xmit_xor() , we determine the output device by using a pre-
4195 * determined xmit_hash_policy(), If the selected device is not enabled,
4196 * find the next active slave.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197 */
4198static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4199{
Wang Chen454d7c92008-11-12 23:37:49 -08004200 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004201 struct slave *slave, *start_at;
4202 int slave_no;
4203 int i;
4204 int res = 1;
4205
4206 read_lock(&bond->lock);
4207
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004208 if (!BOND_IS_OK(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004209 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004210
Jasper Spaansa361c832009-10-23 04:09:24 +00004211 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004212
4213 bond_for_each_slave(bond, slave, i) {
4214 slave_no--;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004215 if (slave_no < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004216 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217 }
4218
4219 start_at = slave;
4220
4221 bond_for_each_slave_from(bond, slave, i, start_at) {
4222 if (IS_UP(slave->dev) &&
4223 (slave->link == BOND_LINK_UP) &&
Jiri Pirkoe30bc062011-03-12 03:14:37 +00004224 bond_is_active_slave(slave)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004225 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4226 break;
4227 }
4228 }
4229
4230out:
4231 if (res) {
4232 /* no suitable interface, frame not sent */
4233 dev_kfree_skb(skb);
4234 }
4235 read_unlock(&bond->lock);
Patrick McHardyec634fe2009-07-05 19:23:38 -07004236 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004237}
4238
4239/*
4240 * in broadcast mode, we send everything to all usable interfaces.
4241 */
4242static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4243{
Wang Chen454d7c92008-11-12 23:37:49 -08004244 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245 struct slave *slave, *start_at;
4246 struct net_device *tx_dev = NULL;
4247 int i;
4248 int res = 1;
4249
4250 read_lock(&bond->lock);
4251
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004252 if (!BOND_IS_OK(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004253 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004254
4255 read_lock(&bond->curr_slave_lock);
4256 start_at = bond->curr_active_slave;
4257 read_unlock(&bond->curr_slave_lock);
4258
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004259 if (!start_at)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004261
4262 bond_for_each_slave_from(bond, slave, i, start_at) {
4263 if (IS_UP(slave->dev) &&
4264 (slave->link == BOND_LINK_UP) &&
Jiri Pirkoe30bc062011-03-12 03:14:37 +00004265 bond_is_active_slave(slave)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004266 if (tx_dev) {
4267 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4268 if (!skb2) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004269 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08004270 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271 continue;
4272 }
4273
4274 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4275 if (res) {
4276 dev_kfree_skb(skb2);
4277 continue;
4278 }
4279 }
4280 tx_dev = slave->dev;
4281 }
4282 }
4283
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004284 if (tx_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004285 res = bond_dev_queue_xmit(bond, skb, tx_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004286
4287out:
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004288 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289 /* no suitable interface, frame not sent */
4290 dev_kfree_skb(skb);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004291
Linus Torvalds1da177e2005-04-16 15:20:36 -07004292 /* frame sent to all suitable interfaces */
4293 read_unlock(&bond->lock);
Patrick McHardyec634fe2009-07-05 19:23:38 -07004294 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295}
4296
4297/*------------------------- Device initialization ---------------------------*/
4298
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08004299static void bond_set_xmit_hash_policy(struct bonding *bond)
4300{
4301 switch (bond->params.xmit_policy) {
4302 case BOND_XMIT_POLICY_LAYER23:
4303 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4304 break;
4305 case BOND_XMIT_POLICY_LAYER34:
4306 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4307 break;
4308 case BOND_XMIT_POLICY_LAYER2:
4309 default:
4310 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4311 break;
4312 }
4313}
4314
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004315/*
4316 * Lookup the slave that corresponds to a qid
4317 */
4318static inline int bond_slave_override(struct bonding *bond,
4319 struct sk_buff *skb)
4320{
4321 int i, res = 1;
4322 struct slave *slave = NULL;
4323 struct slave *check_slave;
4324
4325 read_lock(&bond->lock);
4326
4327 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4328 goto out;
4329
4330 /* Find out if any slaves have the same mapping as this skb. */
4331 bond_for_each_slave(bond, check_slave, i) {
4332 if (check_slave->queue_id == skb->queue_mapping) {
4333 slave = check_slave;
4334 break;
4335 }
4336 }
4337
4338 /* If the slave isn't UP, use default transmit policy. */
4339 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4340 (slave->link == BOND_LINK_UP)) {
4341 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4342 }
4343
4344out:
4345 read_unlock(&bond->lock);
4346 return res;
4347}
4348
4349static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4350{
4351 /*
4352 * This helper function exists to help dev_pick_tx get the correct
Phil Oesterfd0e4352011-03-14 06:22:04 +00004353 * destination queue. Using a helper function skips a call to
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004354 * skb_tx_hash and will put the skbs in the queue we expect on their
4355 * way down to the bonding driver.
4356 */
Phil Oesterfd0e4352011-03-14 06:22:04 +00004357 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4358
4359 if (unlikely(txq >= dev->real_num_tx_queues)) {
David Decotignyd30ee672011-04-13 15:22:29 +00004360 do {
Phil Oesterfd0e4352011-03-14 06:22:04 +00004361 txq -= dev->real_num_tx_queues;
David Decotignyd30ee672011-04-13 15:22:29 +00004362 } while (txq >= dev->real_num_tx_queues);
Phil Oesterfd0e4352011-03-14 06:22:04 +00004363 }
4364 return txq;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004365}
4366
Stephen Hemminger424efe92009-08-31 19:50:51 +00004367static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
Stephen Hemminger00829822008-11-20 20:14:53 -08004368{
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004369 struct bonding *bond = netdev_priv(dev);
4370
Neil Hormane843fa52010-10-13 16:01:50 +00004371 /*
4372 * If we risk deadlock from transmitting this in the
4373 * netpoll path, tell netpoll to queue the frame for later tx
4374 */
4375 if (is_netpoll_tx_blocked(dev))
4376 return NETDEV_TX_BUSY;
4377
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004378 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4379 if (!bond_slave_override(bond, skb))
4380 return NETDEV_TX_OK;
4381 }
Stephen Hemminger00829822008-11-20 20:14:53 -08004382
4383 switch (bond->params.mode) {
4384 case BOND_MODE_ROUNDROBIN:
4385 return bond_xmit_roundrobin(skb, dev);
4386 case BOND_MODE_ACTIVEBACKUP:
4387 return bond_xmit_activebackup(skb, dev);
4388 case BOND_MODE_XOR:
4389 return bond_xmit_xor(skb, dev);
4390 case BOND_MODE_BROADCAST:
4391 return bond_xmit_broadcast(skb, dev);
4392 case BOND_MODE_8023AD:
4393 return bond_3ad_xmit_xor(skb, dev);
4394 case BOND_MODE_ALB:
4395 case BOND_MODE_TLB:
4396 return bond_alb_xmit(skb, dev);
4397 default:
4398 /* Should never happen, mode already checked */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004399 pr_err("%s: Error: Unknown bonding mode %d\n",
4400 dev->name, bond->params.mode);
Stephen Hemminger00829822008-11-20 20:14:53 -08004401 WARN_ON_ONCE(1);
4402 dev_kfree_skb(skb);
4403 return NETDEV_TX_OK;
4404 }
4405}
4406
4407
Linus Torvalds1da177e2005-04-16 15:20:36 -07004408/*
4409 * set bond mode specific net device operations
4410 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08004411void bond_set_mode_ops(struct bonding *bond, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412{
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004413 struct net_device *bond_dev = bond->dev;
4414
Linus Torvalds1da177e2005-04-16 15:20:36 -07004415 switch (mode) {
4416 case BOND_MODE_ROUNDROBIN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417 break;
4418 case BOND_MODE_ACTIVEBACKUP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004419 break;
4420 case BOND_MODE_XOR:
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08004421 bond_set_xmit_hash_policy(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004422 break;
4423 case BOND_MODE_BROADCAST:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004424 break;
4425 case BOND_MODE_8023AD:
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08004426 bond_set_xmit_hash_policy(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004427 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004428 case BOND_MODE_ALB:
Jay Vosburgh8f903c72006-02-21 16:36:44 -08004429 /* FALLTHRU */
4430 case BOND_MODE_TLB:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004431 break;
4432 default:
4433 /* Should never happen, mode already checked */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004434 pr_err("%s: Error: Unknown bonding mode %d\n",
4435 bond_dev->name, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004436 break;
4437 }
4438}
4439
Jay Vosburgh217df672005-09-26 16:11:50 -07004440static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4441 struct ethtool_drvinfo *drvinfo)
4442{
4443 strncpy(drvinfo->driver, DRV_NAME, 32);
4444 strncpy(drvinfo->version, DRV_VERSION, 32);
4445 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4446}
4447
Jeff Garzik7282d492006-09-13 14:30:00 -04004448static const struct ethtool_ops bond_ethtool_ops = {
Jay Vosburgh217df672005-09-26 16:11:50 -07004449 .get_drvinfo = bond_ethtool_get_drvinfo,
Stephen Hemmingerfa53eba2008-09-13 21:17:09 -04004450 .get_link = ethtool_op_get_link,
4451 .get_tx_csum = ethtool_op_get_tx_csum,
4452 .get_sg = ethtool_op_get_sg,
4453 .get_tso = ethtool_op_get_tso,
4454 .get_ufo = ethtool_op_get_ufo,
4455 .get_flags = ethtool_op_get_flags,
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004456};
4457
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004458static const struct net_device_ops bond_netdev_ops = {
Stephen Hemminger181470f2009-06-12 19:02:52 +00004459 .ndo_init = bond_init,
Stephen Hemminger9e716262009-06-12 19:02:47 +00004460 .ndo_uninit = bond_uninit,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004461 .ndo_open = bond_open,
4462 .ndo_stop = bond_close,
Stephen Hemminger00829822008-11-20 20:14:53 -08004463 .ndo_start_xmit = bond_start_xmit,
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004464 .ndo_select_queue = bond_select_queue,
Ben Hutchingsbe1f3c22010-06-08 07:19:54 +00004465 .ndo_get_stats64 = bond_get_stats,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004466 .ndo_do_ioctl = bond_do_ioctl,
4467 .ndo_set_multicast_list = bond_set_multicast_list,
4468 .ndo_change_mtu = bond_change_mtu,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004469 .ndo_set_mac_address = bond_set_mac_address,
Stephen Hemminger00829822008-11-20 20:14:53 -08004470 .ndo_neigh_setup = bond_neigh_setup,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004471 .ndo_vlan_rx_register = bond_vlan_rx_register,
4472 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4473 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
WANG Congf6dc31a2010-05-06 00:48:51 -07004474#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +00004475 .ndo_netpoll_setup = bond_netpoll_setup,
WANG Congf6dc31a2010-05-06 00:48:51 -07004476 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4477 .ndo_poll_controller = bond_poll_controller,
4478#endif
Jiri Pirko9232ecc2011-02-13 09:33:01 +00004479 .ndo_add_slave = bond_enslave,
4480 .ndo_del_slave = bond_release,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004481};
4482
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004483static void bond_destructor(struct net_device *bond_dev)
4484{
4485 struct bonding *bond = netdev_priv(bond_dev);
4486 if (bond->wq)
4487 destroy_workqueue(bond->wq);
4488 free_netdev(bond_dev);
4489}
4490
Stephen Hemminger181470f2009-06-12 19:02:52 +00004491static void bond_setup(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004492{
Wang Chen454d7c92008-11-12 23:37:49 -08004493 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004494
Linus Torvalds1da177e2005-04-16 15:20:36 -07004495 /* initialize rwlocks */
4496 rwlock_init(&bond->lock);
4497 rwlock_init(&bond->curr_slave_lock);
4498
Stephen Hemmingerd2991f72009-06-12 19:02:44 +00004499 bond->params = bonding_defaults;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004500
4501 /* Initialize pointers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004502 bond->dev = bond_dev;
4503 INIT_LIST_HEAD(&bond->vlan_list);
4504
4505 /* Initialize the device entry points */
Stephen Hemminger181470f2009-06-12 19:02:52 +00004506 ether_setup(bond_dev);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004507 bond_dev->netdev_ops = &bond_netdev_ops;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004508 bond_dev->ethtool_ops = &bond_ethtool_ops;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004509 bond_set_mode_ops(bond, bond->params.mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004511 bond_dev->destructor = bond_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004512
4513 /* Initialize the device options */
4514 bond_dev->tx_queue_len = 0;
4515 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
Jay Vosburgh0b680e72006-09-22 21:54:10 -07004516 bond_dev->priv_flags |= IFF_BONDING;
Stephen Hemminger181470f2009-06-12 19:02:52 +00004517 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4518
Linus Torvalds1da177e2005-04-16 15:20:36 -07004519 /* At first, we block adding VLANs. That's the only way to
4520 * prevent problems that occur when adding VLANs over an
4521 * empty bond. The block will be removed once non-challenged
4522 * slaves are enslaved.
4523 */
4524 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4525
Herbert Xu932ff272006-06-09 12:20:56 -07004526 /* don't acquire bond device's netif_tx_lock when
Linus Torvalds1da177e2005-04-16 15:20:36 -07004527 * transmitting */
4528 bond_dev->features |= NETIF_F_LLTX;
4529
4530 /* By default, we declare the bond to be fully
4531 * VLAN hardware accelerated capable. Special
4532 * care is taken in the various xmit functions
4533 * when there are slaves that are not hw accel
4534 * capable
4535 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004536 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4537 NETIF_F_HW_VLAN_RX |
4538 NETIF_F_HW_VLAN_FILTER);
4539
Eric Dumazete6599c22010-09-17 09:25:07 +00004540 /* By default, we enable GRO on bonding devices.
4541 * Actual support requires lowlevel drivers are GRO ready.
4542 */
4543 bond_dev->features |= NETIF_F_GRO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004544}
4545
Jay Vosburghfdaea7a2007-12-06 23:40:35 -08004546static void bond_work_cancel_all(struct bonding *bond)
4547{
4548 write_lock_bh(&bond->lock);
4549 bond->kill_timers = 1;
4550 write_unlock_bh(&bond->lock);
4551
4552 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4553 cancel_delayed_work(&bond->mii_work);
4554
4555 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4556 cancel_delayed_work(&bond->arp_work);
4557
4558 if (bond->params.mode == BOND_MODE_ALB &&
4559 delayed_work_pending(&bond->alb_work))
4560 cancel_delayed_work(&bond->alb_work);
4561
4562 if (bond->params.mode == BOND_MODE_8023AD &&
4563 delayed_work_pending(&bond->ad_work))
4564 cancel_delayed_work(&bond->ad_work);
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00004565
4566 if (delayed_work_pending(&bond->mcast_work))
4567 cancel_delayed_work(&bond->mcast_work);
Jay Vosburghfdaea7a2007-12-06 23:40:35 -08004568}
4569
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004570/*
4571* Destroy a bonding device.
4572* Must be under rtnl_lock when this function is called.
4573*/
4574static void bond_uninit(struct net_device *bond_dev)
Jay Vosburgha434e432008-10-30 17:41:15 -07004575{
Wang Chen454d7c92008-11-12 23:37:49 -08004576 struct bonding *bond = netdev_priv(bond_dev);
Jay Vosburghf35188f2010-07-21 12:14:47 +00004577 struct vlan_entry *vlan, *tmp;
Jay Vosburgha434e432008-10-30 17:41:15 -07004578
WANG Congf6dc31a2010-05-06 00:48:51 -07004579 bond_netpoll_cleanup(bond_dev);
4580
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004581 /* Release the bonded slaves */
4582 bond_release_all(bond_dev);
4583
Jay Vosburgha434e432008-10-30 17:41:15 -07004584 list_del(&bond->bond_list);
4585
4586 bond_work_cancel_all(bond);
4587
Jay Vosburgha434e432008-10-30 17:41:15 -07004588 bond_remove_proc_entry(bond);
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004589
Taku Izumif073c7c2010-12-09 15:17:13 +00004590 bond_debug_unregister(bond);
4591
Jiri Pirko22bedad32010-04-01 21:22:57 +00004592 __hw_addr_flush(&bond->mc_list);
Jay Vosburghf35188f2010-07-21 12:14:47 +00004593
4594 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4595 list_del(&vlan->vlan_list);
4596 kfree(vlan);
4597 }
Jay Vosburgha434e432008-10-30 17:41:15 -07004598}
4599
Linus Torvalds1da177e2005-04-16 15:20:36 -07004600/*------------------------- Module initialization ---------------------------*/
4601
4602/*
4603 * Convert string input module parms. Accept either the
Jay Vosburghece95f72008-01-17 16:25:01 -08004604 * number of the mode or its string name. A bit complicated because
4605 * some mode names are substrings of other names, and calls from sysfs
4606 * may have whitespace in the name (trailing newlines, for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07004607 */
Holger Eitzenberger325dcf72008-12-09 23:10:17 -08004608int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004609{
Hannes Eder54b87322009-02-14 11:15:49 +00004610 int modeint = -1, i, rv;
Jay Vosburgha42e5342008-01-29 18:07:43 -08004611 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
Jay Vosburghece95f72008-01-17 16:25:01 -08004612
Jay Vosburgha42e5342008-01-29 18:07:43 -08004613 for (p = (char *)buf; *p; p++)
4614 if (!(isdigit(*p) || isspace(*p)))
4615 break;
4616
4617 if (*p)
Jay Vosburghece95f72008-01-17 16:25:01 -08004618 rv = sscanf(buf, "%20s", modestr);
Jay Vosburgha42e5342008-01-29 18:07:43 -08004619 else
Hannes Eder54b87322009-02-14 11:15:49 +00004620 rv = sscanf(buf, "%d", &modeint);
Jay Vosburgha42e5342008-01-29 18:07:43 -08004621
4622 if (!rv)
4623 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004624
4625 for (i = 0; tbl[i].modename; i++) {
Hannes Eder54b87322009-02-14 11:15:49 +00004626 if (modeint == tbl[i].mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004627 return tbl[i].mode;
Jay Vosburghece95f72008-01-17 16:25:01 -08004628 if (strcmp(modestr, tbl[i].modename) == 0)
4629 return tbl[i].mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004630 }
4631
4632 return -1;
4633}
4634
4635static int bond_check_params(struct bond_params *params)
4636{
Jiri Pirkoa5499522009-09-25 03:28:09 +00004637 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004638
Linus Torvalds1da177e2005-04-16 15:20:36 -07004639 /*
4640 * Convert string parameters.
4641 */
4642 if (mode) {
4643 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4644 if (bond_mode == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004645 pr_err("Error: Invalid bonding mode \"%s\"\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004646 mode == NULL ? "NULL" : mode);
4647 return -EINVAL;
4648 }
4649 }
4650
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004651 if (xmit_hash_policy) {
4652 if ((bond_mode != BOND_MODE_XOR) &&
4653 (bond_mode != BOND_MODE_8023AD)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004654 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004655 bond_mode_name(bond_mode));
4656 } else {
4657 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4658 xmit_hashtype_tbl);
4659 if (xmit_hashtype == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004660 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004661 xmit_hash_policy == NULL ? "NULL" :
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004662 xmit_hash_policy);
4663 return -EINVAL;
4664 }
4665 }
4666 }
4667
Linus Torvalds1da177e2005-04-16 15:20:36 -07004668 if (lacp_rate) {
4669 if (bond_mode != BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004670 pr_info("lacp_rate param is irrelevant in mode %s\n",
4671 bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004672 } else {
4673 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4674 if (lacp_fast == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004675 pr_err("Error: Invalid lacp rate \"%s\"\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004676 lacp_rate == NULL ? "NULL" : lacp_rate);
4677 return -EINVAL;
4678 }
4679 }
4680 }
4681
Jay Vosburghfd989c82008-11-04 17:51:16 -08004682 if (ad_select) {
4683 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4684 if (params->ad_select == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004685 pr_err("Error: Invalid ad_select \"%s\"\n",
Jay Vosburghfd989c82008-11-04 17:51:16 -08004686 ad_select == NULL ? "NULL" : ad_select);
4687 return -EINVAL;
4688 }
4689
4690 if (bond_mode != BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004691 pr_warning("ad_select param only affects 802.3ad mode\n");
Jay Vosburghfd989c82008-11-04 17:51:16 -08004692 }
4693 } else {
4694 params->ad_select = BOND_AD_STABLE;
4695 }
4696
Nicolas de Pesloüanf5841302009-08-28 13:18:34 +00004697 if (max_bonds < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004698 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4699 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004700 max_bonds = BOND_DEFAULT_MAX_BONDS;
4701 }
4702
4703 if (miimon < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004704 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4705 miimon, INT_MAX, BOND_LINK_MON_INTERV);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004706 miimon = BOND_LINK_MON_INTERV;
4707 }
4708
4709 if (updelay < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004710 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4711 updelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004712 updelay = 0;
4713 }
4714
4715 if (downdelay < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004716 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4717 downdelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004718 downdelay = 0;
4719 }
4720
4721 if ((use_carrier != 0) && (use_carrier != 1)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004722 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4723 use_carrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004724 use_carrier = 1;
4725 }
4726
Moni Shoua7893b242008-05-17 21:10:12 -07004727 if (num_grat_arp < 0 || num_grat_arp > 255) {
Frans Pop2381a552010-03-24 07:57:36 +00004728 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004729 num_grat_arp);
Moni Shoua7893b242008-05-17 21:10:12 -07004730 num_grat_arp = 1;
4731 }
4732
Brian Haley305d5522008-11-04 17:51:14 -08004733 if (num_unsol_na < 0 || num_unsol_na > 255) {
Frans Pop2381a552010-03-24 07:57:36 +00004734 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004735 num_unsol_na);
Brian Haley305d5522008-11-04 17:51:14 -08004736 num_unsol_na = 1;
4737 }
4738
Linus Torvalds1da177e2005-04-16 15:20:36 -07004739 /* reset values for 802.3ad */
4740 if (bond_mode == BOND_MODE_8023AD) {
4741 if (!miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004742 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004743 pr_warning("Forcing miimon to 100msec\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004744 miimon = 100;
4745 }
4746 }
4747
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004748 if (tx_queues < 1 || tx_queues > 255) {
4749 pr_warning("Warning: tx_queues (%d) should be between "
4750 "1 and 255, resetting to %d\n",
4751 tx_queues, BOND_DEFAULT_TX_QUEUES);
4752 tx_queues = BOND_DEFAULT_TX_QUEUES;
4753 }
4754
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004755 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4756 pr_warning("Warning: all_slaves_active module parameter (%d), "
4757 "not of valid value (0/1), so it was set to "
4758 "0\n", all_slaves_active);
4759 all_slaves_active = 0;
4760 }
4761
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004762 if (resend_igmp < 0 || resend_igmp > 255) {
4763 pr_warning("Warning: resend_igmp (%d) should be between "
4764 "0 and 255, resetting to %d\n",
4765 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4766 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4767 }
4768
Linus Torvalds1da177e2005-04-16 15:20:36 -07004769 /* reset values for TLB/ALB */
4770 if ((bond_mode == BOND_MODE_TLB) ||
4771 (bond_mode == BOND_MODE_ALB)) {
4772 if (!miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004773 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004774 pr_warning("Forcing miimon to 100msec\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004775 miimon = 100;
4776 }
4777 }
4778
4779 if (bond_mode == BOND_MODE_ALB) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004780 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",
4781 updelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004782 }
4783
4784 if (!miimon) {
4785 if (updelay || downdelay) {
4786 /* just warn the user the up/down delay will have
4787 * no effect since miimon is zero...
4788 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004789 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4790 updelay, downdelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004791 }
4792 } else {
4793 /* don't allow arp monitoring */
4794 if (arp_interval) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004795 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4796 miimon, arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004797 arp_interval = 0;
4798 }
4799
4800 if ((updelay % miimon) != 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004801 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4802 updelay, miimon,
4803 (updelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004804 }
4805
4806 updelay /= miimon;
4807
4808 if ((downdelay % miimon) != 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004809 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4810 downdelay, miimon,
4811 (downdelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004812 }
4813
4814 downdelay /= miimon;
4815 }
4816
4817 if (arp_interval < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004818 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4819 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004820 arp_interval = BOND_LINK_ARP_INTERV;
4821 }
4822
4823 for (arp_ip_count = 0;
4824 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4825 arp_ip_count++) {
4826 /* not complete check, but should be good enough to
4827 catch mistakes */
4828 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004829 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4830 arp_ip_target[arp_ip_count]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004831 arp_interval = 0;
4832 } else {
Al Virod3bb52b2007-08-22 20:06:58 -04004833 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004834 arp_target[arp_ip_count] = ip;
4835 }
4836 }
4837
4838 if (arp_interval && !arp_ip_count) {
4839 /* don't allow arping if no arp_ip_target given... */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004840 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4841 arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004842 arp_interval = 0;
4843 }
4844
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004845 if (arp_validate) {
4846 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004847 pr_err("arp_validate only supported in active-backup mode\n");
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004848 return -EINVAL;
4849 }
4850 if (!arp_interval) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004851 pr_err("arp_validate requires arp_interval\n");
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004852 return -EINVAL;
4853 }
4854
4855 arp_validate_value = bond_parse_parm(arp_validate,
4856 arp_validate_tbl);
4857 if (arp_validate_value == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004858 pr_err("Error: invalid arp_validate \"%s\"\n",
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004859 arp_validate == NULL ? "NULL" : arp_validate);
4860 return -EINVAL;
4861 }
4862 } else
4863 arp_validate_value = 0;
4864
Linus Torvalds1da177e2005-04-16 15:20:36 -07004865 if (miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004866 pr_info("MII link monitoring set to %d ms\n", miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004867 } else if (arp_interval) {
4868 int i;
4869
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004870 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4871 arp_interval,
4872 arp_validate_tbl[arp_validate_value].modename,
4873 arp_ip_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004874
4875 for (i = 0; i < arp_ip_count; i++)
Jiri Pirkoe5e2a8f2009-08-13 04:11:52 +00004876 pr_info(" %s", arp_ip_target[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004877
Jiri Pirkoe5e2a8f2009-08-13 04:11:52 +00004878 pr_info("\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004879
Jay Vosburghb8a97872008-06-13 18:12:04 -07004880 } else if (max_bonds) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004881 /* miimon and arp_interval not set, we need one so things
4882 * work as expected, see bonding.txt for details
4883 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004884 pr_warning("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 -07004885 }
4886
4887 if (primary && !USES_PRIMARY(bond_mode)) {
4888 /* currently, using a primary only makes sense
4889 * in active backup, TLB or ALB modes
4890 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004891 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4892 primary, bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004893 primary = NULL;
4894 }
4895
Jiri Pirkoa5499522009-09-25 03:28:09 +00004896 if (primary && primary_reselect) {
4897 primary_reselect_value = bond_parse_parm(primary_reselect,
4898 pri_reselect_tbl);
4899 if (primary_reselect_value == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004900 pr_err("Error: Invalid primary_reselect \"%s\"\n",
Jiri Pirkoa5499522009-09-25 03:28:09 +00004901 primary_reselect ==
4902 NULL ? "NULL" : primary_reselect);
4903 return -EINVAL;
4904 }
4905 } else {
4906 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4907 }
4908
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004909 if (fail_over_mac) {
4910 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4911 fail_over_mac_tbl);
4912 if (fail_over_mac_value == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004913 pr_err("Error: invalid fail_over_mac \"%s\"\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004914 arp_validate == NULL ? "NULL" : arp_validate);
4915 return -EINVAL;
4916 }
4917
4918 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004919 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004920 } else {
4921 fail_over_mac_value = BOND_FOM_NONE;
4922 }
Jay Vosburghdd957c52007-10-09 19:57:24 -07004923
Linus Torvalds1da177e2005-04-16 15:20:36 -07004924 /* fill params struct with the proper values */
4925 params->mode = bond_mode;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004926 params->xmit_policy = xmit_hashtype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004927 params->miimon = miimon;
Moni Shoua7893b242008-05-17 21:10:12 -07004928 params->num_grat_arp = num_grat_arp;
Brian Haley305d5522008-11-04 17:51:14 -08004929 params->num_unsol_na = num_unsol_na;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004930 params->arp_interval = arp_interval;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004931 params->arp_validate = arp_validate_value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004932 params->updelay = updelay;
4933 params->downdelay = downdelay;
4934 params->use_carrier = use_carrier;
4935 params->lacp_fast = lacp_fast;
4936 params->primary[0] = 0;
Jiri Pirkoa5499522009-09-25 03:28:09 +00004937 params->primary_reselect = primary_reselect_value;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004938 params->fail_over_mac = fail_over_mac_value;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004939 params->tx_queues = tx_queues;
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004940 params->all_slaves_active = all_slaves_active;
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004941 params->resend_igmp = resend_igmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004942
4943 if (primary) {
4944 strncpy(params->primary, primary, IFNAMSIZ);
4945 params->primary[IFNAMSIZ - 1] = 0;
4946 }
4947
4948 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4949
4950 return 0;
4951}
4952
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004953static struct lock_class_key bonding_netdev_xmit_lock_key;
David S. Millercf508b12008-07-22 14:16:42 -07004954static struct lock_class_key bonding_netdev_addr_lock_key;
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004955
David S. Millere8a04642008-07-17 00:34:19 -07004956static void bond_set_lockdep_class_one(struct net_device *dev,
4957 struct netdev_queue *txq,
4958 void *_unused)
David S. Millerc773e842008-07-08 23:13:53 -07004959{
4960 lockdep_set_class(&txq->_xmit_lock,
4961 &bonding_netdev_xmit_lock_key);
4962}
4963
4964static void bond_set_lockdep_class(struct net_device *dev)
4965{
David S. Millercf508b12008-07-22 14:16:42 -07004966 lockdep_set_class(&dev->addr_list_lock,
4967 &bonding_netdev_addr_lock_key);
David S. Millere8a04642008-07-17 00:34:19 -07004968 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
David S. Millerc773e842008-07-08 23:13:53 -07004969}
4970
Stephen Hemminger181470f2009-06-12 19:02:52 +00004971/*
4972 * Called from registration process
4973 */
4974static int bond_init(struct net_device *bond_dev)
4975{
4976 struct bonding *bond = netdev_priv(bond_dev);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004977 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004978
4979 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4980
4981 bond->wq = create_singlethread_workqueue(bond_dev->name);
4982 if (!bond->wq)
4983 return -ENOMEM;
4984
4985 bond_set_lockdep_class(bond_dev);
4986
Stephen Hemminger181470f2009-06-12 19:02:52 +00004987 bond_create_proc_entry(bond);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004988 list_add_tail(&bond->bond_list, &bn->dev_list);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004989
Eric W. Biederman6151b3d2009-10-29 14:18:22 +00004990 bond_prepare_sysfs_group(bond);
Jiri Pirko22bedad32010-04-01 21:22:57 +00004991
Taku Izumif073c7c2010-12-09 15:17:13 +00004992 bond_debug_register(bond);
4993
Jiri Pirko22bedad32010-04-01 21:22:57 +00004994 __hw_addr_init(&bond->mc_list);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004995 return 0;
4996}
4997
Eric W. Biederman88ead972009-10-29 14:18:25 +00004998static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4999{
5000 if (tb[IFLA_ADDRESS]) {
5001 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5002 return -EINVAL;
5003 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5004 return -EADDRNOTAVAIL;
5005 }
5006 return 0;
5007}
5008
5009static struct rtnl_link_ops bond_link_ops __read_mostly = {
5010 .kind = "bond",
Eric W. Biederman66391042009-10-29 23:58:54 +00005011 .priv_size = sizeof(struct bonding),
Eric W. Biederman88ead972009-10-29 14:18:25 +00005012 .setup = bond_setup,
5013 .validate = bond_validate,
5014};
5015
Mitch Williamsdfe60392005-11-09 10:36:04 -08005016/* Create a new bond based on the specified name and bonding parameters.
Jay Vosburghe4b91c42007-01-19 18:15:31 -08005017 * If name is NULL, obtain a suitable "bond%d" name for us.
Mitch Williamsdfe60392005-11-09 10:36:04 -08005018 * Caller must NOT hold rtnl_lock; we need to release it here before we
5019 * set up our sysfs entries.
5020 */
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005021int bond_create(struct net *net, const char *name)
Mitch Williamsdfe60392005-11-09 10:36:04 -08005022{
5023 struct net_device *bond_dev;
5024 int res;
5025
5026 rtnl_lock();
Jay Vosburgh027ea042008-01-17 16:25:02 -08005027
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00005028 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5029 bond_setup, tx_queues);
Mitch Williamsdfe60392005-11-09 10:36:04 -08005030 if (!bond_dev) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08005031 pr_err("%s: eek! can't alloc netdev!\n", name);
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00005032 rtnl_unlock();
5033 return -ENOMEM;
Mitch Williamsdfe60392005-11-09 10:36:04 -08005034 }
5035
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005036 dev_net_set(bond_dev, net);
Eric W. Biederman88ead972009-10-29 14:18:25 +00005037 bond_dev->rtnl_link_ops = &bond_link_ops;
5038
Jay Vosburghe4b91c42007-01-19 18:15:31 -08005039 if (!name) {
5040 res = dev_alloc_name(bond_dev, "bond%d");
5041 if (res < 0)
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00005042 goto out;
Neil Horman27e6f062010-10-05 03:39:21 +00005043 } else {
5044 /*
5045 * If we're given a name to register
5046 * we need to ensure that its not already
5047 * registered
5048 */
5049 res = -EEXIST;
5050 if (__dev_get_by_name(net, name) != NULL)
5051 goto out;
Jay Vosburghe4b91c42007-01-19 18:15:31 -08005052 }
5053
Mitch Williamsdfe60392005-11-09 10:36:04 -08005054 res = register_netdevice(bond_dev);
Peter Zijlstra0daa23032006-11-08 19:51:01 -08005055
Phil Oestere826eaf2011-03-14 06:22:05 +00005056 netif_carrier_off(bond_dev);
5057
Eric W. Biederman30c15ba2009-10-29 14:18:23 +00005058out:
Mitch Williamsdfe60392005-11-09 10:36:04 -08005059 rtnl_unlock();
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00005060 if (res < 0)
5061 bond_destructor(bond_dev);
Mitch Williamsdfe60392005-11-09 10:36:04 -08005062 return res;
5063}
5064
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00005065static int __net_init bond_net_init(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005066{
Eric W. Biederman15449742009-11-29 15:46:04 +00005067 struct bond_net *bn = net_generic(net, bond_net_id);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005068
5069 bn->net = net;
5070 INIT_LIST_HEAD(&bn->dev_list);
5071
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005072 bond_create_proc_dir(bn);
Eric W. Biederman15449742009-11-29 15:46:04 +00005073
5074 return 0;
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005075}
5076
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00005077static void __net_exit bond_net_exit(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005078{
Eric W. Biederman15449742009-11-29 15:46:04 +00005079 struct bond_net *bn = net_generic(net, bond_net_id);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005080
5081 bond_destroy_proc_dir(bn);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005082}
5083
5084static struct pernet_operations bond_net_ops = {
5085 .init = bond_net_init,
5086 .exit = bond_net_exit,
Eric W. Biederman15449742009-11-29 15:46:04 +00005087 .id = &bond_net_id,
5088 .size = sizeof(struct bond_net),
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005089};
5090
Linus Torvalds1da177e2005-04-16 15:20:36 -07005091static int __init bonding_init(void)
5092{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005093 int i;
5094 int res;
5095
Amerigo Wangbd33acc2011-03-06 21:58:46 +00005096 pr_info("%s", bond_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005097
Mitch Williamsdfe60392005-11-09 10:36:04 -08005098 res = bond_check_params(&bonding_defaults);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00005099 if (res)
Mitch Williamsdfe60392005-11-09 10:36:04 -08005100 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005101
Eric W. Biederman15449742009-11-29 15:46:04 +00005102 res = register_pernet_subsys(&bond_net_ops);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005103 if (res)
5104 goto out;
Jay Vosburgh027ea042008-01-17 16:25:02 -08005105
Eric W. Biederman88ead972009-10-29 14:18:25 +00005106 res = rtnl_link_register(&bond_link_ops);
5107 if (res)
Eric W. Biederman66391042009-10-29 23:58:54 +00005108 goto err_link;
Eric W. Biederman88ead972009-10-29 14:18:25 +00005109
Taku Izumif073c7c2010-12-09 15:17:13 +00005110 bond_create_debugfs();
5111
Linus Torvalds1da177e2005-04-16 15:20:36 -07005112 for (i = 0; i < max_bonds; i++) {
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00005113 res = bond_create(&init_net, NULL);
Mitch Williamsdfe60392005-11-09 10:36:04 -08005114 if (res)
5115 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005116 }
5117
Mitch Williamsb76cdba2005-11-09 10:36:41 -08005118 res = bond_create_sysfs();
5119 if (res)
5120 goto err;
5121
Linus Torvalds1da177e2005-04-16 15:20:36 -07005122 register_netdevice_notifier(&bond_netdev_notifier);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04005123 register_inetaddr_notifier(&bond_inetaddr_notifier);
Brian Haley305d5522008-11-04 17:51:14 -08005124 bond_register_ipv6_notifier();
Mitch Williamsdfe60392005-11-09 10:36:04 -08005125out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07005126 return res;
Eric W. Biederman88ead972009-10-29 14:18:25 +00005127err:
5128 rtnl_link_unregister(&bond_link_ops);
Eric W. Biederman66391042009-10-29 23:58:54 +00005129err_link:
Eric W. Biederman15449742009-11-29 15:46:04 +00005130 unregister_pernet_subsys(&bond_net_ops);
Eric W. Biederman88ead972009-10-29 14:18:25 +00005131 goto out;
Mitch Williamsdfe60392005-11-09 10:36:04 -08005132
Linus Torvalds1da177e2005-04-16 15:20:36 -07005133}
5134
5135static void __exit bonding_exit(void)
5136{
5137 unregister_netdevice_notifier(&bond_netdev_notifier);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04005138 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
Brian Haley305d5522008-11-04 17:51:14 -08005139 bond_unregister_ipv6_notifier();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005140
Pavel Emelyanovae68c392008-05-02 17:49:39 -07005141 bond_destroy_sysfs();
Taku Izumif073c7c2010-12-09 15:17:13 +00005142 bond_destroy_debugfs();
Pavel Emelyanovae68c392008-05-02 17:49:39 -07005143
Eric W. Biederman88ead972009-10-29 14:18:25 +00005144 rtnl_link_unregister(&bond_link_ops);
Eric W. Biederman15449742009-11-29 15:46:04 +00005145 unregister_pernet_subsys(&bond_net_ops);
Neil Hormane843fa52010-10-13 16:01:50 +00005146
5147#ifdef CONFIG_NET_POLL_CONTROLLER
Neil Hormanfb4fa762010-12-06 09:05:50 +00005148 /*
5149 * Make sure we don't have an imbalance on our netpoll blocking
5150 */
5151 WARN_ON(atomic_read(&netpoll_block_tx));
Neil Hormane843fa52010-10-13 16:01:50 +00005152#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005153}
5154
5155module_init(bonding_init);
5156module_exit(bonding_exit);
5157MODULE_LICENSE("GPL");
5158MODULE_VERSION(DRV_VERSION);
5159MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5160MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
Eric W. Biederman88ead972009-10-29 14:18:25 +00005161MODULE_ALIAS_RTNL_LINK("bond");