projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net: rename NETDEV_BONDING_DESLAVE to NETDEV_RELEASE
[deliverable/linux.git] / drivers / net / bonding / bond_main.c
CommitLineData
1da177e4
LT		 1/*
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 *
1da177e4
LT		 32 */
33
a4aee5c8
JP		 34#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
1da177e4
LT		 36#include <linux/kernel.h>
37#include <linux/module.h>
1da177e4
LT		 38#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>
169a3e66 44#include <net/ip.h>
1da177e4 45#include <linux/ip.h>
169a3e66
JV		 46#include <linux/tcp.h>
47#include <linux/udp.h>
1da177e4
LT		 48#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>
3d632c3f 56#include <linux/io.h>
1da177e4 57#include <asm/system.h>
1da177e4 58#include <asm/dma.h>
3d632c3f 59#include <linux/uaccess.h>
1da177e4
LT		 60#include <linux/errno.h>
61#include <linux/netdevice.h>
62#include <linux/inetdevice.h>
a816c7c7 63#include <linux/igmp.h>
1da177e4
LT		 64#include <linux/etherdevice.h>
65#include <linux/skbuff.h>
66#include <net/sock.h>
67#include <linux/rtnetlink.h>
1da177e4
LT		 68#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>
b63bb739 75#include <linux/jiffies.h>
e843fa50 76#include <linux/preempt.h>
c3ade5ca 77#include <net/route.h>
457c4cbc 78#include <net/net_namespace.h>
ec87fd3b 79#include <net/netns/generic.h>
1da177e4
LT		 80#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;
bb1d9123 91static int tx_queues = BOND_DEFAULT_TX_QUEUES;
ad246c99 92static int num_peer_notif = 1;
1da177e4 93static int miimon = BOND_LINK_MON_INTERV;
3d632c3f
SH		 94static int updelay;
95static int downdelay;
1da177e4 96static int use_carrier = 1;
3d632c3f
SH		 97static char *mode;
98static char *primary;
a549952a 99static char *primary_reselect;
3d632c3f
SH		 100static char *lacp_rate;
101static char *ad_select;
102static char *xmit_hash_policy;
1da177e4 103static int arp_interval = BOND_LINK_ARP_INTERV;
3d632c3f
SH		 104static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
105static char *arp_validate;
106static char *fail_over_mac;
ebd8e497 107static int all_slaves_active = 0;
d2991f75 108static struct bond_params bonding_defaults;
c2952c31 109static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
1da177e4
LT		 110
111module_param(max_bonds, int, 0);
112MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
bb1d9123
AG		 113module_param(tx_queues, int, 0);
114MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
ad246c99
BH		 115module_param_named(num_grat_arp, num_peer_notif, int, 0644);
116MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on failover event (alias of num_unsol_na)");
117module_param_named(num_unsol_na, num_peer_notif, int, 0644);
118MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on failover event (alias of num_grat_arp)");
1da177e4
LT		 119module_param(miimon, int, 0);
120MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
121module_param(updelay, int, 0);
122MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
123module_param(downdelay, int, 0);
2ac47660
MW		 124MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
125 "in milliseconds");
1da177e4 126module_param(use_carrier, int, 0);
2ac47660
MW		 127MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
128 "0 for off, 1 for on (default)");
1da177e4 129module_param(mode, charp, 0);
2ac47660
MW		 130MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
131 "1 for active-backup, 2 for balance-xor, "
132 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
133 "6 for balance-alb");
1da177e4
LT		 134module_param(primary, charp, 0);
135MODULE_PARM_DESC(primary, "Primary network device to use");
a549952a
JP		 136module_param(primary_reselect, charp, 0);
137MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
138 "once it comes up; "
139 "0 for always (default), "
140 "1 for only if speed of primary is "
141 "better, "
142 "2 for only on active slave "
143 "failure");
1da177e4 144module_param(lacp_rate, charp, 0);
2ac47660
MW		 145MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
146 "(slow/fast)");
fd989c83
JV		 147module_param(ad_select, charp, 0);
148MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
169a3e66 149module_param(xmit_hash_policy, charp, 0);
2ac47660
MW		 150MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
151 ", 1 for layer 3+4");
1da177e4
LT		 152module_param(arp_interval, int, 0);
153MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
154module_param_array(arp_ip_target, charp, NULL, 0);
155MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
f5b2b966
JV		 156module_param(arp_validate, charp, 0);
157MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
3915c1e8
JV		 158module_param(fail_over_mac, charp, 0);
159MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
ebd8e497
AG		 160module_param(all_slaves_active, int, 0);
161MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
162 "by setting active flag for all slaves. "
163 "0 for never (default), 1 for always.");
c2952c31
FL		 164module_param(resend_igmp, int, 0);
165MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
1da177e4
LT		 166
167/*----------------------------- Global variables ----------------------------*/
168
e843fa50 169#ifdef CONFIG_NET_POLL_CONTROLLER
fb4fa76a 170atomic_t netpoll_block_tx = ATOMIC_INIT(0);
e843fa50
NH		 171#endif
172
f99189b1 173int bond_net_id __read_mostly;
1da177e4 174
3d632c3f
SH		 175static __be32 arp_target[BOND_MAX_ARP_TARGETS];
176static int arp_ip_count;
1da177e4 177static int bond_mode = BOND_MODE_ROUNDROBIN;
3d632c3f
SH		 178static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
179static int lacp_fast;
1da177e4 180
e97fd7c6 181const struct bond_parm_tbl bond_lacp_tbl[] = {
1da177e4
LT		 182{ "slow", AD_LACP_SLOW},
183{ "fast", AD_LACP_FAST},
184{ NULL, -1},
185};
186
e97fd7c6 187const struct bond_parm_tbl bond_mode_tbl[] = {
1da177e4
LT		 188{ "balance-rr", BOND_MODE_ROUNDROBIN},
189{ "active-backup", BOND_MODE_ACTIVEBACKUP},
190{ "balance-xor", BOND_MODE_XOR},
191{ "broadcast", BOND_MODE_BROADCAST},
192{ "802.3ad", BOND_MODE_8023AD},
193{ "balance-tlb", BOND_MODE_TLB},
194{ "balance-alb", BOND_MODE_ALB},
195{ NULL, -1},
196};
197
e97fd7c6 198const struct bond_parm_tbl xmit_hashtype_tbl[] = {
169a3e66
JV		 199{ "layer2", BOND_XMIT_POLICY_LAYER2},
200{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
6f6652be 201{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
169a3e66
JV		 202{ NULL, -1},
203};
204
e97fd7c6 205const struct bond_parm_tbl arp_validate_tbl[] = {
f5b2b966
JV		 206{ "none", BOND_ARP_VALIDATE_NONE},
207{ "active", BOND_ARP_VALIDATE_ACTIVE},
208{ "backup", BOND_ARP_VALIDATE_BACKUP},
209{ "all", BOND_ARP_VALIDATE_ALL},
210{ NULL, -1},
211};
212
e97fd7c6 213const struct bond_parm_tbl fail_over_mac_tbl[] = {
3915c1e8
JV		 214{ "none", BOND_FOM_NONE},
215{ "active", BOND_FOM_ACTIVE},
216{ "follow", BOND_FOM_FOLLOW},
217{ NULL, -1},
218};
219
a549952a
JP		 220const struct bond_parm_tbl pri_reselect_tbl[] = {
221{ "always", BOND_PRI_RESELECT_ALWAYS},
222{ "better", BOND_PRI_RESELECT_BETTER},
223{ "failure", BOND_PRI_RESELECT_FAILURE},
224{ NULL, -1},
225};
226
fd989c83
JV		 227struct bond_parm_tbl ad_select_tbl[] = {
228{ "stable", BOND_AD_STABLE},
229{ "bandwidth", BOND_AD_BANDWIDTH},
230{ "count", BOND_AD_COUNT},
231{ NULL, -1},
232};
233
1da177e4
LT		 234/*-------------------------- Forward declarations ---------------------------*/
235
181470fc 236static int bond_init(struct net_device *bond_dev);
c67dfb29 237static void bond_uninit(struct net_device *bond_dev);
1da177e4
LT		 238
239/*---------------------------- General routines -----------------------------*/
240
bd33acc3 241const char *bond_mode_name(int mode)
1da177e4 242{
77afc92b
HE		 243 static const char *names[] = {
244 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
245 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
246 [BOND_MODE_XOR] = "load balancing (xor)",
247 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
3d632c3f 248 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
77afc92b
HE		 249 [BOND_MODE_TLB] = "transmit load balancing",
250 [BOND_MODE_ALB] = "adaptive load balancing",
251 };
252
253 if (mode < 0 || mode > BOND_MODE_ALB)
1da177e4 254 return "unknown";
77afc92b
HE		 255
256 return names[mode];
1da177e4
LT		 257}
258
259/*---------------------------------- VLAN -----------------------------------*/
260
261/**
262 * bond_add_vlan - add a new vlan id on bond
263 * @bond: bond that got the notification
264 * @vlan_id: the vlan id to add
265 *
266 * Returns -ENOMEM if allocation failed.
267 */
268static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
269{
270 struct vlan_entry *vlan;
271
5a03cdb7 272 pr_debug("bond: %s, vlan id %d\n",
a4aee5c8 273 (bond ? bond->dev->name : "None"), vlan_id);
1da177e4 274
305d552a 275 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
3d632c3f 276 if (!vlan)
1da177e4 277 return -ENOMEM;
1da177e4
LT		 278
279 INIT_LIST_HEAD(&vlan->vlan_list);
280 vlan->vlan_id = vlan_id;
281
282 write_lock_bh(&bond->lock);
283
284 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
285
286 write_unlock_bh(&bond->lock);
287
5a03cdb7 288 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
1da177e4
LT		 289
290 return 0;
291}
292
293/**
294 * bond_del_vlan - delete a vlan id from bond
295 * @bond: bond that got the notification
296 * @vlan_id: the vlan id to delete
297 *
298 * returns -ENODEV if @vlan_id was not found in @bond.
299 */
300static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
301{
0883beca 302 struct vlan_entry *vlan;
1da177e4
LT		 303 int res = -ENODEV;
304
5a03cdb7 305 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
1da177e4 306
e843fa50 307 block_netpoll_tx();
1da177e4
LT		 308 write_lock_bh(&bond->lock);
309
0883beca 310 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
1da177e4
LT		 311 if (vlan->vlan_id == vlan_id) {
312 list_del(&vlan->vlan_list);
313
58402054 314 if (bond_is_lb(bond))
1da177e4 315 bond_alb_clear_vlan(bond, vlan_id);
1da177e4 316
a4aee5c8
JP		 317 pr_debug("removed VLAN ID %d from bond %s\n",
318 vlan_id, bond->dev->name);
1da177e4
LT		 319
320 kfree(vlan);
321
322 if (list_empty(&bond->vlan_list) &&
323 (bond->slave_cnt == 0)) {
324 /* Last VLAN removed and no slaves, so
325 * restore block on adding VLANs. This will
326 * be removed once new slaves that are not
327 * VLAN challenged will be added.
328 */
329 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
330 }
331
332 res = 0;
333 goto out;
334 }
335 }
336
a4aee5c8
JP		 337 pr_debug("couldn't find VLAN ID %d in bond %s\n",
338 vlan_id, bond->dev->name);
1da177e4
LT		 339
340out:
341 write_unlock_bh(&bond->lock);
e843fa50 342 unblock_netpoll_tx();
1da177e4
LT		 343 return res;
344}
345
1da177e4
LT		 346/**
347 * bond_next_vlan - safely skip to the next item in the vlans list.
348 * @bond: the bond we're working on
349 * @curr: item we're advancing from
350 *
351 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
352 * or @curr->next otherwise (even if it is @curr itself again).
3d632c3f 353 *
1da177e4
LT		 354 * Caller must hold bond->lock
355 */
356struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
357{
358 struct vlan_entry *next, *last;
359
3d632c3f 360 if (list_empty(&bond->vlan_list))
1da177e4 361 return NULL;
1da177e4
LT		 362
363 if (!curr) {
364 next = list_entry(bond->vlan_list.next,
365 struct vlan_entry, vlan_list);
366 } else {
367 last = list_entry(bond->vlan_list.prev,
368 struct vlan_entry, vlan_list);
369 if (last == curr) {
370 next = list_entry(bond->vlan_list.next,
371 struct vlan_entry, vlan_list);
372 } else {
373 next = list_entry(curr->vlan_list.next,
374 struct vlan_entry, vlan_list);
375 }
376 }
377
378 return next;
379}
380
381/**
382 * bond_dev_queue_xmit - Prepare skb for xmit.
3d632c3f 383 *
1da177e4
LT		 384 * @bond: bond device that got this skb for tx.
385 * @skb: hw accel VLAN tagged skb to transmit
386 * @slave_dev: slave that is supposed to xmit this skbuff
1da177e4 387 */
3d632c3f
SH		 388int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
389 struct net_device *slave_dev)
1da177e4 390{
8387451e 391 skb->dev = slave_dev;
1da177e4 392 skb->priority = 1;
080e4130 393 if (unlikely(netpoll_tx_running(slave_dev)))
8a8efa22 394 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
080e4130 395 else
f6dc31a8 396 dev_queue_xmit(skb);
1da177e4
LT		 397
398 return 0;
399}
400
401/*
402 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
403 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
404 * lock because:
405 * a. This operation is performed in IOCTL context,
406 * b. The operation is protected by the RTNL semaphore in the 8021q code,
407 * c. Holding a lock with BH disabled while directly calling a base driver
408 * entry point is generally a BAD idea.
3d632c3f 409 *
1da177e4
LT		 410 * The design of synchronization/protection for this operation in the 8021q
411 * module is good for one or more VLAN devices over a single physical device
412 * and cannot be extended for a teaming solution like bonding, so there is a
413 * potential race condition here where a net device from the vlan group might
414 * be referenced (either by a base driver or the 8021q code) while it is being
415 * removed from the system. However, it turns out we're not making matters
416 * worse, and if it works for regular VLAN usage it will work here too.
417*/
418
419/**
420 * bond_vlan_rx_register - Propagates registration to slaves
421 * @bond_dev: bonding net device that got called
422 * @grp: vlan group being registered
423 */
3d632c3f
SH		 424static void bond_vlan_rx_register(struct net_device *bond_dev,
425 struct vlan_group *grp)
1da177e4 426{
454d7c9b 427 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 428 struct slave *slave;
429 int i;
430
a71fb881 431 write_lock_bh(&bond->lock);
1da177e4 432 bond->vlgrp = grp;
a71fb881 433 write_unlock_bh(&bond->lock);
1da177e4
LT		 434
435 bond_for_each_slave(bond, slave, i) {
436 struct net_device *slave_dev = slave->dev;
eb7cc59a 437 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 438
439 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 440 slave_ops->ndo_vlan_rx_register) {
441 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
1da177e4
LT		 442 }
443 }
444}
445
446/**
447 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
448 * @bond_dev: bonding net device that got called
449 * @vid: vlan id being added
450 */
451static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
452{
454d7c9b 453 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 454 struct slave *slave;
455 int i, res;
456
457 bond_for_each_slave(bond, slave, i) {
458 struct net_device *slave_dev = slave->dev;
eb7cc59a 459 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 460
461 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a
SH		 462 slave_ops->ndo_vlan_rx_add_vid) {
463 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
1da177e4
LT		 464 }
465 }
466
467 res = bond_add_vlan(bond, vid);
468 if (res) {
a4aee5c8 469 pr_err("%s: Error: Failed to add vlan id %d\n",
1da177e4
LT		 470 bond_dev->name, vid);
471 }
472}
473
474/**
475 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
476 * @bond_dev: bonding net device that got called
477 * @vid: vlan id being removed
478 */
479static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
480{
454d7c9b 481 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 482 struct slave *slave;
483 struct net_device *vlan_dev;
484 int i, res;
485
486 bond_for_each_slave(bond, slave, i) {
487 struct net_device *slave_dev = slave->dev;
eb7cc59a 488 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 489
490 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
eb7cc59a 491 slave_ops->ndo_vlan_rx_kill_vid) {
1da177e4
LT		 492 /* Save and then restore vlan_dev in the grp array,
493 * since the slave's driver might clear it.
494 */
5c15bdec 495 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
eb7cc59a 496 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
5c15bdec 497 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
1da177e4
LT		 498 }
499 }
500
501 res = bond_del_vlan(bond, vid);
502 if (res) {
a4aee5c8 503 pr_err("%s: Error: Failed to remove vlan id %d\n",
1da177e4
LT		 504 bond_dev->name, vid);
505 }
506}
507
508static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
509{
510 struct vlan_entry *vlan;
eb7cc59a 511 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4 512
f35188fa 513 if (!bond->vlgrp)
03dc2f4c 514 return;
1da177e4
LT		 515
516 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 517 slave_ops->ndo_vlan_rx_register)
518 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
1da177e4
LT		 519
520 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 521 !(slave_ops->ndo_vlan_rx_add_vid))
03dc2f4c 522 return;
1da177e4 523
eb7cc59a
SH		 524 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
525 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
1da177e4
LT		 526}
527
3d632c3f
SH		 528static void bond_del_vlans_from_slave(struct bonding *bond,
529 struct net_device *slave_dev)
1da177e4 530{
eb7cc59a 531 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4
LT		 532 struct vlan_entry *vlan;
533 struct net_device *vlan_dev;
534
f35188fa 535 if (!bond->vlgrp)
03dc2f4c 536 return;
1da177e4
LT		 537
538 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
eb7cc59a 539 !(slave_ops->ndo_vlan_rx_kill_vid))
1da177e4 540 goto unreg;
1da177e4
LT		 541
542 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f35188fa
JV		 543 if (!vlan->vlan_id)
544 continue;
1da177e4
LT		 545 /* Save and then restore vlan_dev in the grp array,
546 * since the slave's driver might clear it.
547 */
5c15bdec 548 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
eb7cc59a 549 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
5c15bdec 550 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
1da177e4
LT		 551 }
552
553unreg:
554 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
eb7cc59a
SH		 555 slave_ops->ndo_vlan_rx_register)
556 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
1da177e4
LT		 557}
558
559/*------------------------------- Link status -------------------------------*/
560
ff59c456
JV		 561/*
562 * Set the carrier state for the master according to the state of its
563 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
564 * do special 802.3ad magic.
565 *
566 * Returns zero if carrier state does not change, nonzero if it does.
567 */
568static int bond_set_carrier(struct bonding *bond)
569{
570 struct slave *slave;
571 int i;
572
573 if (bond->slave_cnt == 0)
574 goto down;
575
576 if (bond->params.mode == BOND_MODE_8023AD)
577 return bond_3ad_set_carrier(bond);
578
579 bond_for_each_slave(bond, slave, i) {
580 if (slave->link == BOND_LINK_UP) {
581 if (!netif_carrier_ok(bond->dev)) {
582 netif_carrier_on(bond->dev);
583 return 1;
584 }
585 return 0;
586 }
587 }
588
589down:
590 if (netif_carrier_ok(bond->dev)) {
591 netif_carrier_off(bond->dev);
592 return 1;
593 }
594 return 0;
595}
596
1da177e4
LT		 597/*
598 * Get link speed and duplex from the slave's base driver
599 * using ethtool. If for some reason the call fails or the
600 * values are invalid, fake speed and duplex to 100/Full
601 * and return error.
602 */
603static int bond_update_speed_duplex(struct slave *slave)
604{
605 struct net_device *slave_dev = slave->dev;
5d30530e
DD		 606 struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
607 u32 slave_speed;
61a44b9c 608 int res;
1da177e4
LT		 609
610 /* Fake speed and duplex */
611 slave->speed = SPEED_100;
612 slave->duplex = DUPLEX_FULL;
613
61a44b9c
MW		 614 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
615 return -1;
1da177e4 616
61a44b9c
MW		 617 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
618 if (res < 0)
1da177e4 619 return -1;
1da177e4 620
5d30530e
DD		 621 slave_speed = ethtool_cmd_speed(&etool);
622 switch (slave_speed) {
1da177e4
LT		 623 case SPEED_10:
624 case SPEED_100:
625 case SPEED_1000:
94dbffd5 626 case SPEED_10000:
1da177e4
LT		 627 break;
628 default:
629 return -1;
630 }
631
632 switch (etool.duplex) {
633 case DUPLEX_FULL:
634 case DUPLEX_HALF:
635 break;
636 default:
637 return -1;
638 }
639
5d30530e 640 slave->speed = slave_speed;
1da177e4
LT		 641 slave->duplex = etool.duplex;
642
643 return 0;
644}
645
646/*
647 * if <dev> supports MII link status reporting, check its link status.
648 *
649 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
3d632c3f 650 * depending upon the setting of the use_carrier parameter.
1da177e4
LT		 651 *
652 * Return either BMSR_LSTATUS, meaning that the link is up (or we
653 * can't tell and just pretend it is), or 0, meaning that the link is
654 * down.
655 *
656 * If reporting is non-zero, instead of faking link up, return -1 if
657 * both ETHTOOL and MII ioctls fail (meaning the device does not
658 * support them). If use_carrier is set, return whatever it says.
659 * It'd be nice if there was a good way to tell if a driver supports
660 * netif_carrier, but there really isn't.
661 */
3d632c3f
SH		 662static int bond_check_dev_link(struct bonding *bond,
663 struct net_device *slave_dev, int reporting)
1da177e4 664{
eb7cc59a 665 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
d9d52832 666 int (*ioctl)(struct net_device *, struct ifreq *, int);
1da177e4
LT		 667 struct ifreq ifr;
668 struct mii_ioctl_data *mii;
1da177e4 669
6c988853
PG		 670 if (!reporting && !netif_running(slave_dev))
671 return 0;
672
eb7cc59a 673 if (bond->params.use_carrier)
1da177e4 674 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1da177e4 675
29112f4e
JP		 676 /* Try to get link status using Ethtool first. */
677 if (slave_dev->ethtool_ops) {
678 if (slave_dev->ethtool_ops->get_link) {
679 u32 link;
680
681 link = slave_dev->ethtool_ops->get_link(slave_dev);
682
683 return link ? BMSR_LSTATUS : 0;
684 }
685 }
686
3d632c3f 687 /* Ethtool can't be used, fallback to MII ioctls. */
eb7cc59a 688 ioctl = slave_ops->ndo_do_ioctl;
1da177e4
LT		 689 if (ioctl) {
690 /* TODO: set pointer to correct ioctl on a per team member */
691 /* bases to make this more efficient. that is, once */
692 /* we determine the correct ioctl, we will always */
693 /* call it and not the others for that team */
694 /* member. */
695
696 /*
697 * We cannot assume that SIOCGMIIPHY will also read a
698 * register; not all network drivers (e.g., e100)
699 * support that.
700 */
701
702 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
703 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
704 mii = if_mii(&ifr);
705 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
706 mii->reg_num = MII_BMSR;
3d632c3f
SH		 707 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
708 return mii->val_out & BMSR_LSTATUS;
1da177e4
LT		 709 }
710 }
711
1da177e4
LT		 712 /*
713 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 714 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 715 * cannot report link status). If not reporting, pretend
716 * we're ok.
717 */
3d632c3f 718 return reporting ? -1 : BMSR_LSTATUS;
1da177e4
LT		 719}
720
721/*----------------------------- Multicast list ------------------------------*/
722
1da177e4
LT		 723/*
724 * Push the promiscuity flag down to appropriate slaves
725 */
7e1a1ac1 726static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 727{
7e1a1ac1 728 int err = 0;
1da177e4
LT		 729 if (USES_PRIMARY(bond->params.mode)) {
730 /* write lock already acquired */
731 if (bond->curr_active_slave) {
7e1a1ac1
WC		 732 err = dev_set_promiscuity(bond->curr_active_slave->dev,
733 inc);
1da177e4
LT		 734 }
735 } else {
736 struct slave *slave;
737 int i;
738 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 739 err = dev_set_promiscuity(slave->dev, inc);
740 if (err)
741 return err;
1da177e4
LT		 742 }
743 }
7e1a1ac1 744 return err;
1da177e4
LT		 745}
746
747/*
748 * Push the allmulti flag down to all slaves
749 */
7e1a1ac1 750static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 751{
7e1a1ac1 752 int err = 0;
1da177e4
LT		 753 if (USES_PRIMARY(bond->params.mode)) {
754 /* write lock already acquired */
755 if (bond->curr_active_slave) {
7e1a1ac1
WC		 756 err = dev_set_allmulti(bond->curr_active_slave->dev,
757 inc);
1da177e4
LT		 758 }
759 } else {
760 struct slave *slave;
761 int i;
762 bond_for_each_slave(bond, slave, i) {
7e1a1ac1
WC		 763 err = dev_set_allmulti(slave->dev, inc);
764 if (err)
765 return err;
1da177e4
LT		 766 }
767 }
7e1a1ac1 768 return err;
1da177e4
LT		 769}
770
771/*
772 * Add a Multicast address to slaves
773 * according to mode
774 */
22bedad3 775static void bond_mc_add(struct bonding *bond, void *addr)
1da177e4
LT		 776{
777 if (USES_PRIMARY(bond->params.mode)) {
778 /* write lock already acquired */
3d632c3f 779 if (bond->curr_active_slave)
22bedad3 780 dev_mc_add(bond->curr_active_slave->dev, addr);
1da177e4
LT		 781 } else {
782 struct slave *slave;
783 int i;
3d632c3f
SH		 784
785 bond_for_each_slave(bond, slave, i)
22bedad3 786 dev_mc_add(slave->dev, addr);
1da177e4
LT		 787 }
788}
789
790/*
791 * Remove a multicast address from slave
792 * according to mode
793 */
22bedad3 794static void bond_mc_del(struct bonding *bond, void *addr)
1da177e4
LT		 795{
796 if (USES_PRIMARY(bond->params.mode)) {
797 /* write lock already acquired */
3d632c3f 798 if (bond->curr_active_slave)
22bedad3 799 dev_mc_del(bond->curr_active_slave->dev, addr);
1da177e4
LT		 800 } else {
801 struct slave *slave;
802 int i;
803 bond_for_each_slave(bond, slave, i) {
22bedad3 804 dev_mc_del(slave->dev, addr);
1da177e4
LT		 805 }
806 }
807}
808
a816c7c7 809
5a37e8ca 810static void __bond_resend_igmp_join_requests(struct net_device *dev)
a816c7c7
JV		 811{
812 struct in_device *in_dev;
a816c7c7
JV		 813
814 rcu_read_lock();
5a37e8ca 815 in_dev = __in_dev_get_rcu(dev);
866f3b25 816 if (in_dev)
24912420 817 ip_mc_rejoin_groups(in_dev);
a816c7c7
JV		 818 rcu_read_unlock();
819}
820
5a37e8ca
FL		 821/*
822 * Retrieve the list of registered multicast addresses for the bonding
823 * device and retransmit an IGMP JOIN request to the current active
824 * slave.
825 */
826static void bond_resend_igmp_join_requests(struct bonding *bond)
827{
828 struct net_device *vlan_dev;
829 struct vlan_entry *vlan;
830
831 read_lock(&bond->lock);
832
833 /* rejoin all groups on bond device */
834 __bond_resend_igmp_join_requests(bond->dev);
835
836 /* rejoin all groups on vlan devices */
837 if (bond->vlgrp) {
838 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
839 vlan_dev = vlan_group_get_device(bond->vlgrp,
840 vlan->vlan_id);
841 if (vlan_dev)
842 __bond_resend_igmp_join_requests(vlan_dev);
843 }
844 }
845
c2952c31
FL		 846 if (--bond->igmp_retrans > 0)
847 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
848
5a37e8ca
FL		 849 read_unlock(&bond->lock);
850}
851
379b7383 852static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
5a37e8ca
FL		 853{
854 struct bonding *bond = container_of(work, struct bonding,
855 mcast_work.work);
856 bond_resend_igmp_join_requests(bond);
857}
858
1da177e4
LT		 859/*
860 * flush all members of flush->mc_list from device dev->mc_list
861 */
3d632c3f
SH		 862static void bond_mc_list_flush(struct net_device *bond_dev,
863 struct net_device *slave_dev)
1da177e4 864{
454d7c9b 865 struct bonding *bond = netdev_priv(bond_dev);
22bedad3 866 struct netdev_hw_addr *ha;
1da177e4 867
22bedad3
JP		 868 netdev_for_each_mc_addr(ha, bond_dev)
869 dev_mc_del(slave_dev, ha->addr);
1da177e4
LT		 870
871 if (bond->params.mode == BOND_MODE_8023AD) {
872 /* del lacpdu mc addr from mc list */
873 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
874
22bedad3 875 dev_mc_del(slave_dev, lacpdu_multicast);
1da177e4
LT		 876 }
877}
878
879/*--------------------------- Active slave change ---------------------------*/
880
881/*
882 * Update the mc list and multicast-related flags for the new and
883 * old active slaves (if any) according to the multicast mode, and
884 * promiscuous flags unconditionally.
885 */
3d632c3f
SH		 886static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
887 struct slave *old_active)
1da177e4 888{
22bedad3 889 struct netdev_hw_addr *ha;
1da177e4 890
3d632c3f 891 if (!USES_PRIMARY(bond->params.mode))
1da177e4
LT		 892 /* nothing to do - mc list is already up-to-date on
893 * all slaves
894 */
895 return;
1da177e4
LT		 896
897 if (old_active) {
3d632c3f 898 if (bond->dev->flags & IFF_PROMISC)
1da177e4 899 dev_set_promiscuity(old_active->dev, -1);
1da177e4 900
3d632c3f 901 if (bond->dev->flags & IFF_ALLMULTI)
1da177e4 902 dev_set_allmulti(old_active->dev, -1);
1da177e4 903
22bedad3
JP		 904 netdev_for_each_mc_addr(ha, bond->dev)
905 dev_mc_del(old_active->dev, ha->addr);
1da177e4
LT		 906 }
907
908 if (new_active) {
7e1a1ac1 909 /* FIXME: Signal errors upstream. */
3d632c3f 910 if (bond->dev->flags & IFF_PROMISC)
1da177e4 911 dev_set_promiscuity(new_active->dev, 1);
1da177e4 912
3d632c3f 913 if (bond->dev->flags & IFF_ALLMULTI)
1da177e4 914 dev_set_allmulti(new_active->dev, 1);
1da177e4 915
22bedad3
JP		 916 netdev_for_each_mc_addr(ha, bond->dev)
917 dev_mc_add(new_active->dev, ha->addr);
1da177e4
LT		 918 }
919}
920
3915c1e8
JV		 921/*
922 * bond_do_fail_over_mac
923 *
924 * Perform special MAC address swapping for fail_over_mac settings
925 *
926 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
927 */
928static void bond_do_fail_over_mac(struct bonding *bond,
929 struct slave *new_active,
930 struct slave *old_active)
1f78d9f9
HE		 931 __releases(&bond->curr_slave_lock)
932 __releases(&bond->lock)
933 __acquires(&bond->lock)
934 __acquires(&bond->curr_slave_lock)
3915c1e8
JV		 935{
936 u8 tmp_mac[ETH_ALEN];
937 struct sockaddr saddr;
938 int rv;
939
940 switch (bond->params.fail_over_mac) {
941 case BOND_FOM_ACTIVE:
942 if (new_active)
943 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
944 new_active->dev->addr_len);
945 break;
946 case BOND_FOM_FOLLOW:
947 /*
948 * if new_active && old_active, swap them
949 * if just old_active, do nothing (going to no active slave)
950 * if just new_active, set new_active to bond's MAC
951 */
952 if (!new_active)
953 return;
954
955 write_unlock_bh(&bond->curr_slave_lock);
956 read_unlock(&bond->lock);
957
958 if (old_active) {
959 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
960 memcpy(saddr.sa_data, old_active->dev->dev_addr,
961 ETH_ALEN);
962 saddr.sa_family = new_active->dev->type;
963 } else {
964 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
965 saddr.sa_family = bond->dev->type;
966 }
967
968 rv = dev_set_mac_address(new_active->dev, &saddr);
969 if (rv) {
a4aee5c8 970 pr_err("%s: Error %d setting MAC of slave %s\n",
3915c1e8
JV		 971 bond->dev->name, -rv, new_active->dev->name);
972 goto out;
973 }
974
975 if (!old_active)
976 goto out;
977
978 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
979 saddr.sa_family = old_active->dev->type;
980
981 rv = dev_set_mac_address(old_active->dev, &saddr);
982 if (rv)
a4aee5c8 983 pr_err("%s: Error %d setting MAC of slave %s\n",
3915c1e8
JV		 984 bond->dev->name, -rv, new_active->dev->name);
985out:
986 read_lock(&bond->lock);
987 write_lock_bh(&bond->curr_slave_lock);
988 break;
989 default:
a4aee5c8 990 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
3915c1e8
JV		 991 bond->dev->name, bond->params.fail_over_mac);
992 break;
993 }
994
995}
996
a549952a
JP		 997static bool bond_should_change_active(struct bonding *bond)
998{
999 struct slave *prim = bond->primary_slave;
1000 struct slave *curr = bond->curr_active_slave;
1001
1002 if (!prim || !curr || curr->link != BOND_LINK_UP)
1003 return true;
1004 if (bond->force_primary) {
1005 bond->force_primary = false;
1006 return true;
1007 }
1008 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1009 (prim->speed < curr->speed ||
1010 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1011 return false;
1012 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1013 return false;
1014 return true;
1015}
3915c1e8 1016
1da177e4
LT		 1017/**
1018 * find_best_interface - select the best available slave to be the active one
1019 * @bond: our bonding struct
1020 *
1021 * Warning: Caller must hold curr_slave_lock for writing.
1022 */
1023static struct slave *bond_find_best_slave(struct bonding *bond)
1024{
1025 struct slave *new_active, *old_active;
1026 struct slave *bestslave = NULL;
1027 int mintime = bond->params.updelay;
1028 int i;
1029
49b4ad92 1030 new_active = bond->curr_active_slave;
1da177e4
LT		 1031
1032 if (!new_active) { /* there were no active slaves left */
3d632c3f 1033 if (bond->slave_cnt > 0) /* found one slave */
1da177e4 1034 new_active = bond->first_slave;
3d632c3f 1035 else
1da177e4 1036 return NULL; /* still no slave, return NULL */
1da177e4
LT		 1037 }
1038
1da177e4 1039 if ((bond->primary_slave) &&
a549952a
JP		 1040 bond->primary_slave->link == BOND_LINK_UP &&
1041 bond_should_change_active(bond)) {
1da177e4
LT		 1042 new_active = bond->primary_slave;
1043 }
1044
1045 /* remember where to stop iterating over the slaves */
1046 old_active = new_active;
1047
1048 bond_for_each_slave_from(bond, new_active, i, old_active) {
b9f60253
JP		 1049 if (new_active->link == BOND_LINK_UP) {
1050 return new_active;
1051 } else if (new_active->link == BOND_LINK_BACK &&
1052 IS_UP(new_active->dev)) {
1053 /* link up, but waiting for stabilization */
1054 if (new_active->delay < mintime) {
1055 mintime = new_active->delay;
1056 bestslave = new_active;
1da177e4
LT		 1057 }
1058 }
1059 }
1060
1061 return bestslave;
1062}
1063
ad246c99
BH		 1064static bool bond_should_notify_peers(struct bonding *bond)
1065{
1066 struct slave *slave = bond->curr_active_slave;
1067
1068 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1069 bond->dev->name, slave ? slave->dev->name : "NULL");
1070
1071 if (!slave || !bond->send_peer_notif ||
1072 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1073 return false;
1074
1075 bond->send_peer_notif--;
1076 return true;
1077}
1078
1da177e4
LT		 1079/**
1080 * change_active_interface - change the active slave into the specified one
1081 * @bond: our bonding struct
1082 * @new: the new slave to make the active one
1083 *
1084 * Set the new slave to the bond's settings and unset them on the old
1085 * curr_active_slave.
1086 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1087 *
1088 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1089 * because it is apparently the best available slave we have, even though its
1090 * updelay hasn't timed out yet.
1091 *
3915c1e8
JV		 1092 * If new_active is not NULL, caller must hold bond->lock for read and
1093 * curr_slave_lock for write_bh.
1da177e4 1094 */
a77b5325 1095void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 1096{
1097 struct slave *old_active = bond->curr_active_slave;
1098
3d632c3f 1099 if (old_active == new_active)
1da177e4 1100 return;
1da177e4
LT		 1101
1102 if (new_active) {
b2220cad
JV		 1103 new_active->jiffies = jiffies;
1104
1da177e4
LT		 1105 if (new_active->link == BOND_LINK_BACK) {
1106 if (USES_PRIMARY(bond->params.mode)) {
a4aee5c8
JP		 1107 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1108 bond->dev->name, new_active->dev->name,
1109 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1da177e4
LT		 1110 }
1111
1112 new_active->delay = 0;
1113 new_active->link = BOND_LINK_UP;
1da177e4 1114
3d632c3f 1115 if (bond->params.mode == BOND_MODE_8023AD)
1da177e4 1116 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1da177e4 1117
58402054 1118 if (bond_is_lb(bond))
1da177e4 1119 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1da177e4
LT		 1120 } else {
1121 if (USES_PRIMARY(bond->params.mode)) {
a4aee5c8
JP		 1122 pr_info("%s: making interface %s the new active one.\n",
1123 bond->dev->name, new_active->dev->name);
1da177e4
LT		 1124 }
1125 }
1126 }
1127
3d632c3f 1128 if (USES_PRIMARY(bond->params.mode))
1da177e4 1129 bond_mc_swap(bond, new_active, old_active);
1da177e4 1130
58402054 1131 if (bond_is_lb(bond)) {
1da177e4 1132 bond_alb_handle_active_change(bond, new_active);
8f903c70
JV		 1133 if (old_active)
1134 bond_set_slave_inactive_flags(old_active);
1135 if (new_active)
1136 bond_set_slave_active_flags(new_active);
1da177e4
LT		 1137 } else {
1138 bond->curr_active_slave = new_active;
1139 }
c3ade5ca
JV		 1140
1141 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3d632c3f 1142 if (old_active)
c3ade5ca 1143 bond_set_slave_inactive_flags(old_active);
c3ade5ca
JV		 1144
1145 if (new_active) {
ad246c99
BH		 1146 bool should_notify_peers = false;
1147
c3ade5ca 1148 bond_set_slave_active_flags(new_active);
2ab82852 1149
709f8a45
OG		 1150 if (bond->params.fail_over_mac)
1151 bond_do_fail_over_mac(bond, new_active,
1152 old_active);
3915c1e8 1153
ad246c99
BH		 1154 if (netif_running(bond->dev)) {
1155 bond->send_peer_notif =
1156 bond->params.num_peer_notif;
1157 should_notify_peers =
1158 bond_should_notify_peers(bond);
1159 }
1160
01f3109d
OG		 1161 write_unlock_bh(&bond->curr_slave_lock);
1162 read_unlock(&bond->lock);
1163
75c78500 1164 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
ad246c99
BH		 1165 if (should_notify_peers)
1166 netdev_bonding_change(bond->dev,
1167 NETDEV_NOTIFY_PEERS);
01f3109d
OG		 1168
1169 read_lock(&bond->lock);
1170 write_lock_bh(&bond->curr_slave_lock);
7893b249 1171 }
c3ade5ca 1172 }
a2fd940f 1173
5a37e8ca
FL		 1174 /* resend IGMP joins since active slave has changed or
1175 * all were sent on curr_active_slave */
ffa95ed5
BH		 1176 if (((USES_PRIMARY(bond->params.mode) && new_active) ||
1177 bond->params.mode == BOND_MODE_ROUNDROBIN) &&
1178 netif_running(bond->dev)) {
c2952c31 1179 bond->igmp_retrans = bond->params.resend_igmp;
5a37e8ca 1180 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
a2fd940f 1181 }
1da177e4
LT		 1182}
1183
1184/**
1185 * bond_select_active_slave - select a new active slave, if needed
1186 * @bond: our bonding struct
1187 *
3d632c3f 1188 * This functions should be called when one of the following occurs:
1da177e4
LT		 1189 * - The old curr_active_slave has been released or lost its link.
1190 * - The primary_slave has got its link back.
1191 * - A slave has got its link back and there's no old curr_active_slave.
1192 *
3915c1e8 1193 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1da177e4 1194 */
a77b5325 1195void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 1196{
1197 struct slave *best_slave;
ff59c456 1198 int rv;
1da177e4
LT		 1199
1200 best_slave = bond_find_best_slave(bond);
1201 if (best_slave != bond->curr_active_slave) {
1202 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 1203 rv = bond_set_carrier(bond);
1204 if (!rv)
1205 return;
1206
1207 if (netif_carrier_ok(bond->dev)) {
a4aee5c8
JP		 1208 pr_info("%s: first active interface up!\n",
1209 bond->dev->name);
ff59c456 1210 } else {
a4aee5c8
JP		 1211 pr_info("%s: now running without any active interface !\n",
1212 bond->dev->name);
ff59c456 1213 }
1da177e4
LT		 1214 }
1215}
1216
1217/*--------------------------- slave list handling ---------------------------*/
1218
1219/*
1220 * This function attaches the slave to the end of list.
1221 *
1222 * bond->lock held for writing by caller.
1223 */
1224static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1225{
1226 if (bond->first_slave == NULL) { /* attaching the first slave */
1227 new_slave->next = new_slave;
1228 new_slave->prev = new_slave;
1229 bond->first_slave = new_slave;
1230 } else {
1231 new_slave->next = bond->first_slave;
1232 new_slave->prev = bond->first_slave->prev;
1233 new_slave->next->prev = new_slave;
1234 new_slave->prev->next = new_slave;
1235 }
1236
1237 bond->slave_cnt++;
1238}
1239
1240/*
1241 * This function detaches the slave from the list.
1242 * WARNING: no check is made to verify if the slave effectively
1243 * belongs to <bond>.
1244 * Nothing is freed on return, structures are just unchained.
1245 * If any slave pointer in bond was pointing to <slave>,
1246 * it should be changed by the calling function.
1247 *
1248 * bond->lock held for writing by caller.
1249 */
1250static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1251{
3d632c3f 1252 if (slave->next)
1da177e4 1253 slave->next->prev = slave->prev;
1da177e4 1254
3d632c3f 1255 if (slave->prev)
1da177e4 1256 slave->prev->next = slave->next;
1da177e4
LT		 1257
1258 if (bond->first_slave == slave) { /* slave is the first slave */
1259 if (bond->slave_cnt > 1) { /* there are more slave */
1260 bond->first_slave = slave->next;
1261 } else {
1262 bond->first_slave = NULL; /* slave was the last one */
1263 }
1264 }
1265
1266 slave->next = NULL;
1267 slave->prev = NULL;
1268 bond->slave_cnt--;
1269}
1270
f6dc31a8 1271#ifdef CONFIG_NET_POLL_CONTROLLER
8a8efa22 1272static inline int slave_enable_netpoll(struct slave *slave)
f6dc31a8 1273{
8a8efa22
AW		 1274 struct netpoll *np;
1275 int err = 0;
f6dc31a8 1276
8a8efa22
AW		 1277 np = kzalloc(sizeof(*np), GFP_KERNEL);
1278 err = -ENOMEM;
1279 if (!np)
1280 goto out;
1281
1282 np->dev = slave->dev;
1283 err = __netpoll_setup(np);
1284 if (err) {
1285 kfree(np);
1286 goto out;
f6dc31a8 1287 }
8a8efa22
AW		 1288 slave->np = np;
1289out:
1290 return err;
1291}
1292static inline void slave_disable_netpoll(struct slave *slave)
1293{
1294 struct netpoll *np = slave->np;
1295
1296 if (!np)
1297 return;
1298
1299 slave->np = NULL;
1300 synchronize_rcu_bh();
1301 __netpoll_cleanup(np);
1302 kfree(np);
1303}
1304static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1305{
1306 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1307 return false;
1308 if (!slave_dev->netdev_ops->ndo_poll_controller)
1309 return false;
1310 return true;
f6dc31a8
WC		 1311}
1312
1313static void bond_poll_controller(struct net_device *bond_dev)
1314{
8a8efa22
AW		 1315}
1316
1317static void __bond_netpoll_cleanup(struct bonding *bond)
1318{
c2355e1a
NH		 1319 struct slave *slave;
1320 int i;
1321
8a8efa22
AW		 1322 bond_for_each_slave(bond, slave, i)
1323 if (IS_UP(slave->dev))
1324 slave_disable_netpoll(slave);
f6dc31a8 1325}
f6dc31a8
WC		 1326static void bond_netpoll_cleanup(struct net_device *bond_dev)
1327{
1328 struct bonding *bond = netdev_priv(bond_dev);
8a8efa22
AW		 1329
1330 read_lock(&bond->lock);
1331 __bond_netpoll_cleanup(bond);
1332 read_unlock(&bond->lock);
1333}
1334
1335static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1336{
1337 struct bonding *bond = netdev_priv(dev);
f6dc31a8 1338 struct slave *slave;
8a8efa22 1339 int i, err = 0;
f6dc31a8
WC		 1340
1341 read_lock(&bond->lock);
f6dc31a8 1342 bond_for_each_slave(bond, slave, i) {
8a8efa22
AW		 1343 err = slave_enable_netpoll(slave);
1344 if (err) {
1345 __bond_netpoll_cleanup(bond);
1346 break;
f6dc31a8
WC		 1347 }
1348 }
1349 read_unlock(&bond->lock);
8a8efa22 1350 return err;
f6dc31a8
WC		 1351}
1352
8a8efa22
AW		 1353static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1354{
1355 return bond->dev->npinfo;
1356}
f6dc31a8 1357
8a8efa22
AW		 1358#else
1359static inline int slave_enable_netpoll(struct slave *slave)
1360{
1361 return 0;
1362}
1363static inline void slave_disable_netpoll(struct slave *slave)
1364{
1365}
f6dc31a8
WC		 1366static void bond_netpoll_cleanup(struct net_device *bond_dev)
1367{
1368}
f6dc31a8
WC		 1369#endif
1370
1da177e4
LT		 1371/*---------------------------------- IOCTL ----------------------------------*/
1372
4ad072c9
AB		 1373static int bond_sethwaddr(struct net_device *bond_dev,
1374 struct net_device *slave_dev)
1da177e4 1375{
5a03cdb7
HE		 1376 pr_debug("bond_dev=%p\n", bond_dev);
1377 pr_debug("slave_dev=%p\n", slave_dev);
1378 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1da177e4
LT		 1379 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1380 return 0;
1381}
1382
b2a103e6 1383static u32 bond_fix_features(struct net_device *dev, u32 features)
8531c5ff 1384{
8531c5ff 1385 struct slave *slave;
b2a103e6
MM		 1386 struct bonding *bond = netdev_priv(dev);
1387 u32 mask;
8e3babcd 1388 int i;
8531c5ff 1389
b2a103e6 1390 read_lock(&bond->lock);
b63365a2 1391
b2a103e6
MM		 1392 if (!bond->first_slave) {
1393 /* Disable adding VLANs to empty bond. But why? --mq */
1394 features |= NETIF_F_VLAN_CHALLENGED;
1395 goto out;
1396 }
b63365a2 1397
b2a103e6 1398 mask = features;
b63365a2 1399 features &= ~NETIF_F_ONE_FOR_ALL;
b2a103e6 1400 features |= NETIF_F_ALL_FOR_ALL;
7f353bf2 1401
54ef3137 1402 bond_for_each_slave(bond, slave, i) {
b63365a2
HX		 1403 features = netdev_increment_features(features,
1404 slave->dev->features,
b2a103e6
MM		 1405 mask);
1406 }
1407
1408out:
1409 read_unlock(&bond->lock);
1410 return features;
1411}
1412
1413#define BOND_VLAN_FEATURES (NETIF_F_ALL_TX_OFFLOADS | \
1414 NETIF_F_SOFT_FEATURES | \
1415 NETIF_F_LRO)
1416
1417static void bond_compute_features(struct bonding *bond)
1418{
1419 struct slave *slave;
1420 struct net_device *bond_dev = bond->dev;
1421 u32 vlan_features = BOND_VLAN_FEATURES;
1422 unsigned short max_hard_header_len = ETH_HLEN;
1423 int i;
1424
1425 read_lock(&bond->lock);
1426
1427 if (!bond->first_slave)
1428 goto done;
1429
1430 bond_for_each_slave(bond, slave, i) {
278339a4 1431 vlan_features = netdev_increment_features(vlan_features,
b2a103e6
MM		 1432 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1433
54ef3137
JV		 1434 if (slave->dev->hard_header_len > max_hard_header_len)
1435 max_hard_header_len = slave->dev->hard_header_len;
1436 }
8531c5ff 1437
b63365a2 1438done:
b2a103e6 1439 bond_dev->vlan_features = vlan_features;
54ef3137 1440 bond_dev->hard_header_len = max_hard_header_len;
8531c5ff 1441
b2a103e6
MM		 1442 read_unlock(&bond->lock);
1443
1444 netdev_change_features(bond_dev);
8531c5ff
AK		 1445}
1446
872254dd
MS		 1447static void bond_setup_by_slave(struct net_device *bond_dev,
1448 struct net_device *slave_dev)
1449{
454d7c9b 1450 struct bonding *bond = netdev_priv(bond_dev);
d90a162a 1451
00829823 1452 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1453
1454 bond_dev->type = slave_dev->type;
1455 bond_dev->hard_header_len = slave_dev->hard_header_len;
1456 bond_dev->addr_len = slave_dev->addr_len;
1457
1458 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1459 slave_dev->addr_len);
d90a162a 1460 bond->setup_by_slave = 1;
872254dd
MS		 1461}
1462
5b2c4dd2 1463/* On bonding slaves other than the currently active slave, suppress
3aba891d 1464 * duplicates except for alb non-mcast/bcast.
5b2c4dd2
JP		 1465 */
1466static bool bond_should_deliver_exact_match(struct sk_buff *skb,
0bd80dad
JP		 1467 struct slave *slave,
1468 struct bonding *bond)
5b2c4dd2 1469{
2d7011ca 1470 if (bond_is_slave_inactive(slave)) {
0bd80dad 1471 if (bond->params.mode == BOND_MODE_ALB &&
5b2c4dd2
JP		 1472 skb->pkt_type != PACKET_BROADCAST &&
1473 skb->pkt_type != PACKET_MULTICAST)
5b2c4dd2 1474 return false;
5b2c4dd2
JP		 1475 return true;
1476 }
1477 return false;
1478}
1479
8a4eb573 1480static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
5b2c4dd2 1481{
8a4eb573 1482 struct sk_buff *skb = *pskb;
f1c1775a 1483 struct slave *slave;
0bd80dad 1484 struct bonding *bond;
5b2c4dd2 1485
8a4eb573
JP		 1486 skb = skb_share_check(skb, GFP_ATOMIC);
1487 if (unlikely(!skb))
1488 return RX_HANDLER_CONSUMED;
1489
1490 *pskb = skb;
5b2c4dd2 1491
35d48903
JP		 1492 slave = bond_slave_get_rcu(skb->dev);
1493 bond = slave->bond;
0bd80dad
JP		 1494
1495 if (bond->params.arp_interval)
f1c1775a 1496 slave->dev->last_rx = jiffies;
5b2c4dd2 1497
3aba891d
JP		 1498 if (bond->recv_probe) {
1499 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1500
1501 if (likely(nskb)) {
1502 bond->recv_probe(nskb, bond, slave);
1503 dev_kfree_skb(nskb);
1504 }
1505 }
1506
0bd80dad 1507 if (bond_should_deliver_exact_match(skb, slave, bond)) {
8a4eb573 1508 return RX_HANDLER_EXACT;
5b2c4dd2
JP		 1509 }
1510
35d48903 1511 skb->dev = bond->dev;
5b2c4dd2 1512
0bd80dad 1513 if (bond->params.mode == BOND_MODE_ALB &&
35d48903 1514 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
5b2c4dd2 1515 skb->pkt_type == PACKET_HOST) {
5b2c4dd2 1516
541ac7c9
CG		 1517 if (unlikely(skb_cow_head(skb,
1518 skb->data - skb_mac_header(skb)))) {
1519 kfree_skb(skb);
8a4eb573 1520 return RX_HANDLER_CONSUMED;
541ac7c9 1521 }
35d48903 1522 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
5b2c4dd2
JP		 1523 }
1524
8a4eb573 1525 return RX_HANDLER_ANOTHER;
5b2c4dd2
JP		 1526}
1527
1da177e4 1528/* enslave device <slave> to bond device <master> */
a77b5325 1529int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1530{
454d7c9b 1531 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1532 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1da177e4 1533 struct slave *new_slave = NULL;
22bedad3 1534 struct netdev_hw_addr *ha;
1da177e4
LT		 1535 struct sockaddr addr;
1536 int link_reporting;
1da177e4
LT		 1537 int res = 0;
1538
552709d5 1539 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
eb7cc59a 1540 slave_ops->ndo_do_ioctl == NULL) {
a4aee5c8
JP		 1541 pr_warning("%s: Warning: no link monitoring support for %s\n",
1542 bond_dev->name, slave_dev->name);
1da177e4
LT		 1543 }
1544
1545 /* bond must be initialized by bond_open() before enslaving */
1546 if (!(bond_dev->flags & IFF_UP)) {
a4aee5c8
JP		 1547 pr_warning("%s: master_dev is not up in bond_enslave\n",
1548 bond_dev->name);
1da177e4
LT		 1549 }
1550
1551 /* already enslaved */
1552 if (slave_dev->flags & IFF_SLAVE) {
5a03cdb7 1553 pr_debug("Error, Device was already enslaved\n");
1da177e4
LT		 1554 return -EBUSY;
1555 }
1556
1557 /* vlan challenged mutual exclusion */
1558 /* no need to lock since we're protected by rtnl_lock */
1559 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
5a03cdb7 1560 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
f35188fa 1561 if (bond->vlgrp) {
a4aee5c8
JP		 1562 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1563 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1564 return -EPERM;
1565 } else {
a4aee5c8
JP		 1566 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1567 bond_dev->name, slave_dev->name,
1568 slave_dev->name, bond_dev->name);
1da177e4
LT		 1569 }
1570 } else {
5a03cdb7 1571 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1da177e4
LT		 1572 }
1573
217df670
JV		 1574 /*
1575 * Old ifenslave binaries are no longer supported. These can
3d632c3f 1576 * be identified with moderate accuracy by the state of the slave:
217df670
JV		 1577 * 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)) {
a4aee5c8 1581 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
217df670
JV		 1582 slave_dev->name);
1583 res = -EPERM;
1584 goto err_undo_flags;
1585 }
1da177e4 1586
872254dd
MS		 1587 /* 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) {
e36b9d16 1595 if (bond_dev->type != slave_dev->type) {
e36b9d16 1596 pr_debug("%s: change device type from %d to %d\n",
a4aee5c8
JP		 1597 bond_dev->name,
1598 bond_dev->type, slave_dev->type);
75c78500 1599
3ca5b404
JP		 1600 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 }
75c78500 1609
32a806c1 1610 /* Flush unicast and multicast addresses */
a748ee24 1611 dev_uc_flush(bond_dev);
22bedad3 1612 dev_mc_flush(bond_dev);
32a806c1 1613
e36b9d16
MS		 1614 if (slave_dev->type != ARPHRD_ETHER)
1615 bond_setup_by_slave(bond_dev, slave_dev);
1616 else
1617 ether_setup(bond_dev);
75c78500 1618
93d9b7d7
JP		 1619 netdev_bonding_change(bond_dev,
1620 NETDEV_POST_TYPE_CHANGE);
e36b9d16 1621 }
872254dd 1622 } else if (bond_dev->type != slave_dev->type) {
a4aee5c8
JP		 1623 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;
872254dd
MS		 1628 }
1629
eb7cc59a 1630 if (slave_ops->ndo_set_mac_address == NULL) {
2ab82852 1631 if (bond->slave_cnt == 0) {
a4aee5c8
JP		 1632 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);
3915c1e8
JV		 1634 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1635 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
a4aee5c8
JP		 1636 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);
2ab82852
MS		 1638 res = -EOPNOTSUPP;
1639 goto err_undo_flags;
1640 }
1da177e4
LT		 1641 }
1642
8d8fc29d
AW		 1643 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1644
c20811a7
JP		 1645 /* If this is the first slave, then we need to set the master's hardware
1646 * address to be the same as the slave's. */
d13a2cb6 1647 if (is_zero_ether_addr(bond->dev->dev_addr))
c20811a7
JP		 1648 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1649 slave_dev->addr_len);
1650
1651
243cb4e5 1652 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1653 if (!new_slave) {
1654 res = -ENOMEM;
1655 goto err_undo_flags;
1656 }
1657
bb1d9123
AG		 1658 /*
1659 * Set the new_slave's queue_id to be zero. Queue ID mapping
1660 * is set via sysfs or module option if desired.
1661 */
1662 new_slave->queue_id = 0;
1663
b15ba0fb
JP		 1664 /* Save slave's original mtu and then set it to match the bond */
1665 new_slave->original_mtu = slave_dev->mtu;
1666 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1667 if (res) {
1668 pr_debug("Error %d calling dev_set_mtu\n", res);
1669 goto err_free;
1670 }
1671
217df670
JV		 1672 /*
1673 * Save slave's original ("permanent") mac address for modes
1674 * that need it, and for restoring it upon release, and then
1675 * set it to the master's address
1676 */
1677 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1678
dd957c57 1679 if (!bond->params.fail_over_mac) {
2ab82852
MS		 1680 /*
1681 * Set slave to master's mac address. The application already
1682 * set the master's mac address to that of the first slave
1683 */
1684 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1685 addr.sa_family = slave_dev->type;
1686 res = dev_set_mac_address(slave_dev, &addr);
1687 if (res) {
5a03cdb7 1688 pr_debug("Error %d calling set_mac_address\n", res);
b15ba0fb 1689 goto err_restore_mtu;
2ab82852 1690 }
217df670 1691 }
1da177e4 1692
1765a575 1693 res = netdev_set_bond_master(slave_dev, bond_dev);
c2edacf8 1694 if (res) {
1765a575 1695 pr_debug("Error %d calling netdev_set_bond_master\n", res);
569f0c4d 1696 goto err_restore_mac;
c2edacf8 1697 }
5b2c4dd2 1698
217df670
JV		 1699 /* open the slave since the application closed it */
1700 res = dev_open(slave_dev);
1701 if (res) {
3d632c3f 1702 pr_debug("Opening slave %s failed\n", slave_dev->name);
35d48903 1703 goto err_unset_master;
1da177e4
LT		 1704 }
1705
35d48903 1706 new_slave->bond = bond;
1da177e4 1707 new_slave->dev = slave_dev;
0b680e75 1708 slave_dev->priv_flags |= IFF_BONDING;
1da177e4 1709
58402054 1710 if (bond_is_lb(bond)) {
1da177e4
LT		 1711 /* bond_alb_init_slave() must be called before all other stages since
1712 * it might fail and we do not want to have to undo everything
1713 */
1714 res = bond_alb_init_slave(bond, new_slave);
3d632c3f 1715 if (res)
569f0c4d 1716 goto err_close;
1da177e4
LT		 1717 }
1718
1719 /* If the mode USES_PRIMARY, then the new slave gets the
1720 * master's promisc (and mc) settings only if it becomes the
1721 * curr_active_slave, and that is taken care of later when calling
1722 * bond_change_active()
1723 */
1724 if (!USES_PRIMARY(bond->params.mode)) {
1725 /* set promiscuity level to new slave */
1726 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1727 res = dev_set_promiscuity(slave_dev, 1);
1728 if (res)
1729 goto err_close;
1da177e4
LT		 1730 }
1731
1732 /* set allmulti level to new slave */
1733 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1734 res = dev_set_allmulti(slave_dev, 1);
1735 if (res)
1736 goto err_close;
1da177e4
LT		 1737 }
1738
b9e40857 1739 netif_addr_lock_bh(bond_dev);
1da177e4 1740 /* upload master's mc_list to new slave */
22bedad3
JP		 1741 netdev_for_each_mc_addr(ha, bond_dev)
1742 dev_mc_add(slave_dev, ha->addr);
b9e40857 1743 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1744 }
1745
1746 if (bond->params.mode == BOND_MODE_8023AD) {
1747 /* add lacpdu mc addr to mc list */
1748 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1749
22bedad3 1750 dev_mc_add(slave_dev, lacpdu_multicast);
1da177e4
LT		 1751 }
1752
1753 bond_add_vlans_on_slave(bond, slave_dev);
1754
1755 write_lock_bh(&bond->lock);
1756
1757 bond_attach_slave(bond, new_slave);
1758
1759 new_slave->delay = 0;
1760 new_slave->link_failure_count = 0;
1761
3915c1e8
JV		 1762 write_unlock_bh(&bond->lock);
1763
b2a103e6
MM		 1764 bond_compute_features(bond);
1765
3915c1e8
JV		 1766 read_lock(&bond->lock);
1767
f5b2b966
JV		 1768 new_slave->last_arp_rx = jiffies;
1769
1da177e4
LT		 1770 if (bond->params.miimon && !bond->params.use_carrier) {
1771 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1772
1773 if ((link_reporting == -1) && !bond->params.arp_interval) {
1774 /*
1775 * miimon is set but a bonded network driver
1776 * does not support ETHTOOL/MII and
1777 * arp_interval is not set. Note: if
1778 * use_carrier is enabled, we will never go
1779 * here (because netif_carrier is always
1780 * supported); thus, we don't need to change
1781 * the messages for netif_carrier.
1782 */
a4aee5c8 1783 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",
4e0952c7 1784 bond_dev->name, slave_dev->name);
1da177e4
LT		 1785 } else if (link_reporting == -1) {
1786 /* unable get link status using mii/ethtool */
a4aee5c8
JP		 1787 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",
1788 bond_dev->name, slave_dev->name);
1da177e4
LT		 1789 }
1790 }
1791
1792 /* check for initial state */
1793 if (!bond->params.miimon ||
1794 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1795 if (bond->params.updelay) {
a4aee5c8 1796 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1da177e4
LT		 1797 new_slave->link = BOND_LINK_BACK;
1798 new_slave->delay = bond->params.updelay;
1799 } else {
a4aee5c8 1800 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1da177e4
LT		 1801 new_slave->link = BOND_LINK_UP;
1802 }
1803 new_slave->jiffies = jiffies;
1804 } else {
a4aee5c8 1805 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1da177e4
LT		 1806 new_slave->link = BOND_LINK_DOWN;
1807 }
1808
1809 if (bond_update_speed_duplex(new_slave) &&
1810 (new_slave->link != BOND_LINK_DOWN)) {
a4aee5c8
JP		 1811 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1812 bond_dev->name, new_slave->dev->name);
1da177e4
LT		 1813
1814 if (bond->params.mode == BOND_MODE_8023AD) {
a4aee5c8
JP		 1815 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1816 bond_dev->name);
1da177e4
LT		 1817 }
1818 }
1819
1820 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1821 /* if there is a primary slave, remember it */
a549952a 1822 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1da177e4 1823 bond->primary_slave = new_slave;
a549952a
JP		 1824 bond->force_primary = true;
1825 }
1da177e4
LT		 1826 }
1827
3915c1e8
JV		 1828 write_lock_bh(&bond->curr_slave_lock);
1829
1da177e4
LT		 1830 switch (bond->params.mode) {
1831 case BOND_MODE_ACTIVEBACKUP:
8a8e447b
JV		 1832 bond_set_slave_inactive_flags(new_slave);
1833 bond_select_active_slave(bond);
1da177e4
LT		 1834 break;
1835 case BOND_MODE_8023AD:
1836 /* in 802.3ad mode, the internal mechanism
1837 * will activate the slaves in the selected
1838 * aggregator
1839 */
1840 bond_set_slave_inactive_flags(new_slave);
1841 /* if this is the first slave */
1842 if (bond->slave_cnt == 1) {
1843 SLAVE_AD_INFO(new_slave).id = 1;
1844 /* Initialize AD with the number of times that the AD timer is called in 1 second
1845 * can be called only after the mac address of the bond is set
1846 */
1847 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1848 bond->params.lacp_fast);
1849 } else {
1850 SLAVE_AD_INFO(new_slave).id =
1851 SLAVE_AD_INFO(new_slave->prev).id + 1;
1852 }
1853
1854 bond_3ad_bind_slave(new_slave);
1855 break;
1856 case BOND_MODE_TLB:
1857 case BOND_MODE_ALB:
e30bc066 1858 bond_set_active_slave(new_slave);
059fe7a5 1859 bond_set_slave_inactive_flags(new_slave);
5a29f789 1860 bond_select_active_slave(bond);
1da177e4
LT		 1861 break;
1862 default:
5a03cdb7 1863 pr_debug("This slave is always active in trunk mode\n");
1da177e4
LT		 1864
1865 /* always active in trunk mode */
e30bc066 1866 bond_set_active_slave(new_slave);
1da177e4
LT		 1867
1868 /* In trunking mode there is little meaning to curr_active_slave
1869 * anyway (it holds no special properties of the bond device),
1870 * so we can change it without calling change_active_interface()
1871 */
3d632c3f 1872 if (!bond->curr_active_slave)
1da177e4 1873 bond->curr_active_slave = new_slave;
3d632c3f 1874
1da177e4
LT		 1875 break;
1876 } /* switch(bond_mode) */
1877
3915c1e8
JV		 1878 write_unlock_bh(&bond->curr_slave_lock);
1879
ff59c456
JV		 1880 bond_set_carrier(bond);
1881
f6dc31a8 1882#ifdef CONFIG_NET_POLL_CONTROLLER
8a8efa22
AW		 1883 slave_dev->npinfo = bond_netpoll_info(bond);
1884 if (slave_dev->npinfo) {
1885 if (slave_enable_netpoll(new_slave)) {
1886 read_unlock(&bond->lock);
1887 pr_info("Error, %s: master_dev is using netpoll, "
1888 "but new slave device does not support netpoll.\n",
1889 bond_dev->name);
1890 res = -EBUSY;
1891 goto err_close;
1892 }
f6dc31a8
WC		 1893 }
1894#endif
8a8efa22 1895
3915c1e8 1896 read_unlock(&bond->lock);
1da177e4 1897
b76cdba9
MW		 1898 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1899 if (res)
569f0c4d 1900 goto err_close;
b76cdba9 1901
35d48903
JP		 1902 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1903 new_slave);
1904 if (res) {
1905 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1906 goto err_dest_symlinks;
1907 }
1908
a4aee5c8
JP		 1909 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1910 bond_dev->name, slave_dev->name,
e30bc066 1911 bond_is_active_slave(new_slave) ? "n active" : " backup",
a4aee5c8 1912 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1da177e4
LT		 1913
1914 /* enslave is successful */
1915 return 0;
1916
1917/* Undo stages on error */
35d48903
JP		 1918err_dest_symlinks:
1919 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1920
1da177e4
LT		 1921err_close:
1922 dev_close(slave_dev);
1923
569f0c4d 1924err_unset_master:
1765a575 1925 netdev_set_bond_master(slave_dev, NULL);
569f0c4d 1926
1da177e4 1927err_restore_mac:
dd957c57 1928 if (!bond->params.fail_over_mac) {
3915c1e8
JV		 1929 /* XXX TODO - fom follow mode needs to change master's
1930 * MAC if this slave's MAC is in use by the bond, or at
1931 * least print a warning.
1932 */
2ab82852
MS		 1933 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1934 addr.sa_family = slave_dev->type;
1935 dev_set_mac_address(slave_dev, &addr);
1936 }
1da177e4 1937
b15ba0fb
JP		 1938err_restore_mtu:
1939 dev_set_mtu(slave_dev, new_slave->original_mtu);
1940
1da177e4
LT		 1941err_free:
1942 kfree(new_slave);
1943
1944err_undo_flags:
b2a103e6 1945 bond_compute_features(bond);
3d632c3f 1946
1da177e4
LT		 1947 return res;
1948}
1949
1950/*
1951 * Try to release the slave device <slave> from the bond device <master>
1952 * It is legal to access curr_active_slave without a lock because all the function
1953 * is write-locked.
1954 *
1955 * The rules for slave state should be:
1956 * for Active/Backup:
1957 * Active stays on all backups go down
1958 * for Bonded connections:
1959 * The first up interface should be left on and all others downed.
1960 */
a77b5325 1961int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1962{
454d7c9b 1963 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1964 struct slave *slave, *oldcurrent;
1965 struct sockaddr addr;
b2a103e6 1966 u32 old_features = bond_dev->features;
1da177e4
LT		 1967
1968 /* slave is not a slave or master is not master of this slave */
1969 if (!(slave_dev->flags & IFF_SLAVE) ||
1970 (slave_dev->master != bond_dev)) {
a4aee5c8 1971 pr_err("%s: Error: cannot release %s.\n",
1da177e4
LT		 1972 bond_dev->name, slave_dev->name);
1973 return -EINVAL;
1974 }
1975
e843fa50 1976 block_netpoll_tx();
daf9209b 1977 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1da177e4
LT		 1978 write_lock_bh(&bond->lock);
1979
1980 slave = bond_get_slave_by_dev(bond, slave_dev);
1981 if (!slave) {
1982 /* not a slave of this bond */
a4aee5c8
JP		 1983 pr_info("%s: %s not enslaved\n",
1984 bond_dev->name, slave_dev->name);
f5e2a7b2 1985 write_unlock_bh(&bond->lock);
e843fa50 1986 unblock_netpoll_tx();
1da177e4
LT		 1987 return -EINVAL;
1988 }
1989
35d48903
JP		 1990 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1991 * for this slave anymore.
1992 */
1993 netdev_rx_handler_unregister(slave_dev);
1994 write_unlock_bh(&bond->lock);
1995 synchronize_net();
1996 write_lock_bh(&bond->lock);
1997
3915c1e8 1998 if (!bond->params.fail_over_mac) {
8e95a202
JP		 1999 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
2000 bond->slave_cnt > 1)
a4aee5c8
JP		 2001 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",
2002 bond_dev->name, slave_dev->name,
2003 slave->perm_hwaddr,
2004 bond_dev->name, slave_dev->name);
1da177e4
LT		 2005 }
2006
2007 /* Inform AD package of unbinding of slave. */
2008 if (bond->params.mode == BOND_MODE_8023AD) {
2009 /* must be called before the slave is
2010 * detached from the list
2011 */
2012 bond_3ad_unbind_slave(slave);
2013 }
2014
a4aee5c8
JP		 2015 pr_info("%s: releasing %s interface %s\n",
2016 bond_dev->name,
e30bc066 2017 bond_is_active_slave(slave) ? "active" : "backup",
a4aee5c8 2018 slave_dev->name);
1da177e4
LT		 2019
2020 oldcurrent = bond->curr_active_slave;
2021
2022 bond->current_arp_slave = NULL;
2023
2024 /* release the slave from its bond */
2025 bond_detach_slave(bond, slave);
2026
3d632c3f 2027 if (bond->primary_slave == slave)
1da177e4 2028 bond->primary_slave = NULL;
1da177e4 2029
3d632c3f 2030 if (oldcurrent == slave)
1da177e4 2031 bond_change_active_slave(bond, NULL);
1da177e4 2032
58402054 2033 if (bond_is_lb(bond)) {
1da177e4
LT		 2034 /* Must be called only after the slave has been
2035 * detached from the list and the curr_active_slave
2036 * has been cleared (if our_slave == old_current),
2037 * but before a new active slave is selected.
2038 */
2543331d 2039 write_unlock_bh(&bond->lock);
1da177e4 2040 bond_alb_deinit_slave(bond, slave);
2543331d 2041 write_lock_bh(&bond->lock);
1da177e4
LT		 2042 }
2043
059fe7a5
JV		 2044 if (oldcurrent == slave) {
2045 /*
2046 * Note that we hold RTNL over this sequence, so there
2047 * is no concern that another slave add/remove event
2048 * will interfere.
2049 */
2050 write_unlock_bh(&bond->lock);
2051 read_lock(&bond->lock);
2052 write_lock_bh(&bond->curr_slave_lock);
2053
1da177e4
LT		 2054 bond_select_active_slave(bond);
2055
059fe7a5
JV		 2056 write_unlock_bh(&bond->curr_slave_lock);
2057 read_unlock(&bond->lock);
2058 write_lock_bh(&bond->lock);
2059 }
2060
1da177e4 2061 if (bond->slave_cnt == 0) {
ff59c456
JV		 2062 bond_set_carrier(bond);
2063
1da177e4
LT		 2064 /* if the last slave was removed, zero the mac address
2065 * of the master so it will be set by the application
2066 * to the mac address of the first slave
2067 */
2068 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2069
b2a103e6 2070 if (bond->vlgrp) {
a4aee5c8
JP		 2071 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2072 bond_dev->name, bond_dev->name);
2073 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2074 bond_dev->name);
1da177e4 2075 }
1da177e4
LT		 2076 }
2077
2078 write_unlock_bh(&bond->lock);
e843fa50 2079 unblock_netpoll_tx();
1da177e4 2080
b2a103e6
MM		 2081 bond_compute_features(bond);
2082 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2083 (old_features & NETIF_F_VLAN_CHALLENGED))
2084 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2085 bond_dev->name, slave_dev->name, bond_dev->name);
2086
b76cdba9
MW		 2087 /* must do this from outside any spinlocks */
2088 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2089
1da177e4
LT		 2090 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 */
3d632c3f 2098 if (bond_dev->flags & IFF_PROMISC)
1da177e4 2099 dev_set_promiscuity(slave_dev, -1);
1da177e4
LT		 2100
2101 /* unset allmulti level from slave */
3d632c3f 2102 if (bond_dev->flags & IFF_ALLMULTI)
1da177e4 2103 dev_set_allmulti(slave_dev, -1);
1da177e4
LT		 2104
2105 /* flush master's mc_list from slave */
b9e40857 2106 netif_addr_lock_bh(bond_dev);
1da177e4 2107 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2108 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2109 }
2110
1765a575 2111 netdev_set_bond_master(slave_dev, NULL);
1da177e4 2112
8a8efa22 2113 slave_disable_netpoll(slave);
f6dc31a8 2114
1da177e4
LT		 2115 /* close slave before restoring its mac address */
2116 dev_close(slave_dev);
2117
3915c1e8 2118 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2ab82852
MS		 2119 /* 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 }
1da177e4 2124
b15ba0fb
JP		 2125 dev_set_mtu(slave_dev, slave->original_mtu);
2126
2d7011ca 2127 slave_dev->priv_flags &= ~IFF_BONDING;
1da177e4
LT		 2128
2129 kfree(slave);
2130
2131 return 0; /* deletion OK */
2132}
2133
d90a162a 2134/*
dadaa10b 2135* First release a slave and then destroy the bond if no more slaves are left.
d90a162a
MS		 2136* Must be under rtnl_lock when this function is called.
2137*/
26d8ee75 2138static int bond_release_and_destroy(struct net_device *bond_dev,
2139 struct net_device *slave_dev)
d90a162a 2140{
454d7c9b 2141 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 2142 int ret;
2143
2144 ret = bond_release(bond_dev, slave_dev);
2145 if ((ret == 0) && (bond->slave_cnt == 0)) {
8a8efa22 2146 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
a4aee5c8
JP		 2147 pr_info("%s: destroying bond %s.\n",
2148 bond_dev->name, bond_dev->name);
9e71626c 2149 unregister_netdevice(bond_dev);
d90a162a
MS		 2150 }
2151 return ret;
2152}
2153
1da177e4
LT		 2154/*
2155 * This function releases all slaves.
2156 */
2157static int bond_release_all(struct net_device *bond_dev)
2158{
454d7c9b 2159 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2160 struct slave *slave;
2161 struct net_device *slave_dev;
2162 struct sockaddr addr;
2163
2164 write_lock_bh(&bond->lock);
2165
ff59c456
JV		 2166 netif_carrier_off(bond_dev);
2167
3d632c3f 2168 if (bond->slave_cnt == 0)
1da177e4 2169 goto out;
1da177e4
LT		 2170
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 */
3d632c3f 2179 if (bond->params.mode == BOND_MODE_8023AD)
1da177e4 2180 bond_3ad_unbind_slave(slave);
1da177e4
LT		 2181
2182 slave_dev = slave->dev;
2183 bond_detach_slave(bond, slave);
2184
2543331d
JV		 2185 /* 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
35d48903
JP		 2191 /* 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
58402054 2197 if (bond_is_lb(bond)) {
1da177e4
LT		 2198 /* must be called only after the slave
2199 * has been detached from the list
2200 */
2201 bond_alb_deinit_slave(bond, slave);
2202 }
2203
b76cdba9 2204 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1da177e4
LT		 2205 bond_del_vlans_from_slave(bond, slave_dev);
2206
2207 /* If the mode USES_PRIMARY, then we should only remove its
2208 * promisc and mc settings if it was the curr_active_slave, but that was
2209 * already taken care of above when we detached the slave
2210 */
2211 if (!USES_PRIMARY(bond->params.mode)) {
2212 /* unset promiscuity level from slave */
3d632c3f 2213 if (bond_dev->flags & IFF_PROMISC)
1da177e4 2214 dev_set_promiscuity(slave_dev, -1);
1da177e4
LT		 2215
2216 /* unset allmulti level from slave */
3d632c3f 2217 if (bond_dev->flags & IFF_ALLMULTI)
1da177e4 2218 dev_set_allmulti(slave_dev, -1);
1da177e4
LT		 2219
2220 /* flush master's mc_list from slave */
b9e40857 2221 netif_addr_lock_bh(bond_dev);
1da177e4 2222 bond_mc_list_flush(bond_dev, slave_dev);
b9e40857 2223 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 2224 }
2225
1765a575 2226 netdev_set_bond_master(slave_dev, NULL);
1da177e4 2227
8a8efa22
AW		 2228 slave_disable_netpoll(slave);
2229
1da177e4
LT		 2230 /* close slave before restoring its mac address */
2231 dev_close(slave_dev);
2232
dd957c57 2233 if (!bond->params.fail_over_mac) {
2ab82852
MS		 2234 /* restore original ("permanent") mac address*/
2235 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2236 addr.sa_family = slave_dev->type;
2237 dev_set_mac_address(slave_dev, &addr);
2238 }
1da177e4 2239
1da177e4
LT		 2240 kfree(slave);
2241
2242 /* re-acquire the lock before getting the next slave */
2243 write_lock_bh(&bond->lock);
2244 }
2245
2246 /* zero the mac address of the master so it will be
2247 * set by the application to the mac address of the
2248 * first slave
2249 */
2250 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2251
b2a103e6 2252 if (bond->vlgrp) {
a4aee5c8
JP		 2253 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2254 bond_dev->name, bond_dev->name);
2255 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2256 bond_dev->name);
1da177e4
LT		 2257 }
2258
a4aee5c8 2259 pr_info("%s: released all slaves\n", bond_dev->name);
1da177e4
LT		 2260
2261out:
2262 write_unlock_bh(&bond->lock);
b2a103e6
MM		 2263
2264 bond_compute_features(bond);
2265
1da177e4
LT		 2266 return 0;
2267}
2268
2269/*
2270 * This function changes the active slave to slave <slave_dev>.
2271 * It returns -EINVAL in the following cases.
2272 * - <slave_dev> is not found in the list.
2273 * - There is not active slave now.
2274 * - <slave_dev> is already active.
2275 * - The link state of <slave_dev> is not BOND_LINK_UP.
2276 * - <slave_dev> is not running.
3d632c3f
SH		 2277 * In these cases, this function does nothing.
2278 * In the other cases, current_slave pointer is changed and 0 is returned.
1da177e4
LT		 2279 */
2280static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2281{
454d7c9b 2282 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2283 struct slave *old_active = NULL;
2284 struct slave *new_active = NULL;
2285 int res = 0;
2286
3d632c3f 2287 if (!USES_PRIMARY(bond->params.mode))
1da177e4 2288 return -EINVAL;
1da177e4
LT		 2289
2290 /* Verify that master_dev is indeed the master of slave_dev */
3d632c3f 2291 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
1da177e4 2292 return -EINVAL;
1da177e4 2293
059fe7a5 2294 read_lock(&bond->lock);
1da177e4 2295
059fe7a5 2296 read_lock(&bond->curr_slave_lock);
1da177e4 2297 old_active = bond->curr_active_slave;
059fe7a5
JV		 2298 read_unlock(&bond->curr_slave_lock);
2299
1da177e4
LT		 2300 new_active = bond_get_slave_by_dev(bond, slave_dev);
2301
2302 /*
2303 * Changing to the current active: do nothing; return success.
2304 */
2305 if (new_active && (new_active == old_active)) {
059fe7a5 2306 read_unlock(&bond->lock);
1da177e4
LT		 2307 return 0;
2308 }
2309
2310 if ((new_active) &&
2311 (old_active) &&
2312 (new_active->link == BOND_LINK_UP) &&
2313 IS_UP(new_active->dev)) {
e843fa50 2314 block_netpoll_tx();
059fe7a5 2315 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2316 bond_change_active_slave(bond, new_active);
059fe7a5 2317 write_unlock_bh(&bond->curr_slave_lock);
e843fa50 2318 unblock_netpoll_tx();
3d632c3f 2319 } else
1da177e4 2320 res = -EINVAL;
1da177e4 2321
059fe7a5 2322 read_unlock(&bond->lock);
1da177e4
LT		 2323
2324 return res;
2325}
2326
1da177e4
LT		 2327static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2328{
454d7c9b 2329 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 2330
2331 info->bond_mode = bond->params.mode;
2332 info->miimon = bond->params.miimon;
2333
6603a6f2 2334 read_lock(&bond->lock);
1da177e4 2335 info->num_slaves = bond->slave_cnt;
6603a6f2 2336 read_unlock(&bond->lock);
1da177e4
LT		 2337
2338 return 0;
2339}
2340
2341static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2342{
454d7c9b 2343 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 2344 struct slave *slave;
689c96cc 2345 int i, res = -ENODEV;
1da177e4 2346
6603a6f2 2347 read_lock(&bond->lock);
1da177e4
LT		 2348
2349 bond_for_each_slave(bond, slave, i) {
2350 if (i == (int)info->slave_id) {
689c96cc
ED		 2351 res = 0;
2352 strcpy(info->slave_name, slave->dev->name);
2353 info->link = slave->link;
e30bc066 2354 info->state = bond_slave_state(slave);
689c96cc 2355 info->link_failure_count = slave->link_failure_count;
1da177e4
LT		 2356 break;
2357 }
2358 }
2359
6603a6f2 2360 read_unlock(&bond->lock);
1da177e4 2361
689c96cc 2362 return res;
1da177e4
LT		 2363}
2364
2365/*-------------------------------- Monitoring -------------------------------*/
2366
1da177e4 2367
f0c76d61
JV		 2368static int bond_miimon_inspect(struct bonding *bond)
2369{
2370 struct slave *slave;
2371 int i, link_state, commit = 0;
41f89100
JP		 2372 bool ignore_updelay;
2373
2374 ignore_updelay = !bond->curr_active_slave ? true : false;
1da177e4
LT		 2375
2376 bond_for_each_slave(bond, slave, i) {
f0c76d61 2377 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2378
f0c76d61 2379 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 2380
2381 switch (slave->link) {
f0c76d61
JV		 2382 case BOND_LINK_UP:
2383 if (link_state)
2384 continue;
1da177e4 2385
f0c76d61
JV		 2386 slave->link = BOND_LINK_FAIL;
2387 slave->delay = bond->params.downdelay;
2388 if (slave->delay) {
a4aee5c8
JP		 2389 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2390 bond->dev->name,
2391 (bond->params.mode ==
2392 BOND_MODE_ACTIVEBACKUP) ?
e30bc066 2393 (bond_is_active_slave(slave) ?
a4aee5c8
JP		 2394 "active " : "backup ") : "",
2395 slave->dev->name,
2396 bond->params.downdelay * bond->params.miimon);
1da177e4 2397 }
f0c76d61
JV		 2398 /*FALLTHRU*/
2399 case BOND_LINK_FAIL:
2400 if (link_state) {
2401 /*
2402 * recovered before downdelay expired
2403 */
2404 slave->link = BOND_LINK_UP;
1da177e4 2405 slave->jiffies = jiffies;
a4aee5c8
JP		 2406 pr_info("%s: link status up again after %d ms for interface %s.\n",
2407 bond->dev->name,
2408 (bond->params.downdelay - slave->delay) *
2409 bond->params.miimon,
2410 slave->dev->name);
f0c76d61 2411 continue;
1da177e4 2412 }
f0c76d61
JV		 2413
2414 if (slave->delay <= 0) {
2415 slave->new_link = BOND_LINK_DOWN;
2416 commit++;
2417 continue;
1da177e4 2418 }
1da177e4 2419
f0c76d61
JV		 2420 slave->delay--;
2421 break;
2422
2423 case BOND_LINK_DOWN:
2424 if (!link_state)
2425 continue;
2426
2427 slave->link = BOND_LINK_BACK;
2428 slave->delay = bond->params.updelay;
2429
2430 if (slave->delay) {
a4aee5c8
JP		 2431 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2432 bond->dev->name, slave->dev->name,
2433 ignore_updelay ? 0 :
2434 bond->params.updelay *
2435 bond->params.miimon);
f0c76d61
JV		 2436 }
2437 /*FALLTHRU*/
2438 case BOND_LINK_BACK:
2439 if (!link_state) {
2440 slave->link = BOND_LINK_DOWN;
a4aee5c8
JP		 2441 pr_info("%s: link status down again after %d ms for interface %s.\n",
2442 bond->dev->name,
2443 (bond->params.updelay - slave->delay) *
2444 bond->params.miimon,
2445 slave->dev->name);
f0c76d61
JV		 2446
2447 continue;
2448 }
2449
41f89100
JP		 2450 if (ignore_updelay)
2451 slave->delay = 0;
2452
f0c76d61
JV		 2453 if (slave->delay <= 0) {
2454 slave->new_link = BOND_LINK_UP;
2455 commit++;
41f89100 2456 ignore_updelay = false;
f0c76d61 2457 continue;
1da177e4 2458 }
f0c76d61
JV		 2459
2460 slave->delay--;
1da177e4 2461 break;
f0c76d61
JV		 2462 }
2463 }
1da177e4 2464
f0c76d61
JV		 2465 return commit;
2466}
1da177e4 2467
f0c76d61
JV		 2468static void bond_miimon_commit(struct bonding *bond)
2469{
2470 struct slave *slave;
2471 int i;
2472
2473 bond_for_each_slave(bond, slave, i) {
2474 switch (slave->new_link) {
2475 case BOND_LINK_NOCHANGE:
2476 continue;
1da177e4 2477
f0c76d61
JV		 2478 case BOND_LINK_UP:
2479 slave->link = BOND_LINK_UP;
2480 slave->jiffies = jiffies;
2481
2482 if (bond->params.mode == BOND_MODE_8023AD) {
2483 /* prevent it from being the active one */
e30bc066 2484 bond_set_backup_slave(slave);
f0c76d61
JV		 2485 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2486 /* make it immediately active */
e30bc066 2487 bond_set_active_slave(slave);
f0c76d61
JV		 2488 } else if (slave != bond->primary_slave) {
2489 /* prevent it from being the active one */
e30bc066 2490 bond_set_backup_slave(slave);
1da177e4 2491 }
1da177e4 2492
546add79
KPO		 2493 bond_update_speed_duplex(slave);
2494
5d30530e 2495 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
700c2a77
KPO		 2496 bond->dev->name, slave->dev->name,
2497 slave->speed, slave->duplex ? "full" : "half");
1da177e4 2498
f0c76d61
JV		 2499 /* notify ad that the link status has changed */
2500 if (bond->params.mode == BOND_MODE_8023AD)
2501 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2502
58402054 2503 if (bond_is_lb(bond))
f0c76d61
JV		 2504 bond_alb_handle_link_change(bond, slave,
2505 BOND_LINK_UP);
1da177e4 2506
f0c76d61
JV		 2507 if (!bond->curr_active_slave ||
2508 (slave == bond->primary_slave))
2509 goto do_failover;
1da177e4 2510
f0c76d61 2511 continue;
059fe7a5 2512
f0c76d61 2513 case BOND_LINK_DOWN:
fba4acda
JV		 2514 if (slave->link_failure_count < UINT_MAX)
2515 slave->link_failure_count++;
2516
f0c76d61 2517 slave->link = BOND_LINK_DOWN;
1da177e4 2518
f0c76d61
JV		 2519 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2520 bond->params.mode == BOND_MODE_8023AD)
2521 bond_set_slave_inactive_flags(slave);
2522
a4aee5c8
JP		 2523 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2524 bond->dev->name, slave->dev->name);
f0c76d61
JV		 2525
2526 if (bond->params.mode == BOND_MODE_8023AD)
2527 bond_3ad_handle_link_change(slave,
2528 BOND_LINK_DOWN);
2529
ae63e808 2530 if (bond_is_lb(bond))
f0c76d61
JV		 2531 bond_alb_handle_link_change(bond, slave,
2532 BOND_LINK_DOWN);
2533
2534 if (slave == bond->curr_active_slave)
2535 goto do_failover;
2536
2537 continue;
2538
2539 default:
a4aee5c8 2540 pr_err("%s: invalid new link %d on slave %s\n",
f0c76d61
JV		 2541 bond->dev->name, slave->new_link,
2542 slave->dev->name);
2543 slave->new_link = BOND_LINK_NOCHANGE;
2544
2545 continue;
2546 }
2547
2548do_failover:
2549 ASSERT_RTNL();
e843fa50 2550 block_netpoll_tx();
f0c76d61
JV		 2551 write_lock_bh(&bond->curr_slave_lock);
2552 bond_select_active_slave(bond);
2553 write_unlock_bh(&bond->curr_slave_lock);
e843fa50 2554 unblock_netpoll_tx();
f0c76d61
JV		 2555 }
2556
2557 bond_set_carrier(bond);
1da177e4
LT		 2558}
2559
0b0eef66
JV		 2560/*
2561 * bond_mii_monitor
2562 *
2563 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2564 * inspection, then (if inspection indicates something needs to be done)
2565 * an acquisition of appropriate locks followed by a commit phase to
2566 * implement whatever link state changes are indicated.
0b0eef66
JV		 2567 */
2568void bond_mii_monitor(struct work_struct *work)
2569{
2570 struct bonding *bond = container_of(work, struct bonding,
2571 mii_work.work);
ad246c99 2572 bool should_notify_peers = false;
0b0eef66
JV		 2573
2574 read_lock(&bond->lock);
f0c76d61
JV		 2575 if (bond->kill_timers)
2576 goto out;
2577
2578 if (bond->slave_cnt == 0)
2579 goto re_arm;
b59f9f74 2580
ad246c99
BH		 2581 should_notify_peers = bond_should_notify_peers(bond);
2582
f0c76d61 2583 if (bond_miimon_inspect(bond)) {
0b0eef66
JV		 2584 read_unlock(&bond->lock);
2585 rtnl_lock();
2586 read_lock(&bond->lock);
f0c76d61
JV		 2587
2588 bond_miimon_commit(bond);
2589
5655662d
JV		 2590 read_unlock(&bond->lock);
2591 rtnl_unlock(); /* might sleep, hold no other locks */
2592 read_lock(&bond->lock);
0b0eef66
JV		 2593 }
2594
f0c76d61
JV		 2595re_arm:
2596 if (bond->params.miimon)
2597 queue_delayed_work(bond->wq, &bond->mii_work,
2598 msecs_to_jiffies(bond->params.miimon));
2599out:
0b0eef66 2600 read_unlock(&bond->lock);
ad246c99
BH		 2601
2602 if (should_notify_peers) {
2603 rtnl_lock();
2604 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2605 rtnl_unlock();
2606 }
0b0eef66 2607}
c3ade5ca 2608
d3bb52b0 2609static __be32 bond_glean_dev_ip(struct net_device *dev)
c3ade5ca
JV		 2610{
2611 struct in_device *idev;
2612 struct in_ifaddr *ifa;
a144ea4b 2613 __be32 addr = 0;
c3ade5ca
JV		 2614
2615 if (!dev)
2616 return 0;
2617
2618 rcu_read_lock();
e5ed6399 2619 idev = __in_dev_get_rcu(dev);
c3ade5ca
JV		 2620 if (!idev)
2621 goto out;
2622
2623 ifa = idev->ifa_list;
2624 if (!ifa)
2625 goto out;
2626
2627 addr = ifa->ifa_local;
2628out:
2629 rcu_read_unlock();
2630 return addr;
2631}
2632
d3bb52b0 2633static int bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2634{
0883beca 2635 struct vlan_entry *vlan;
f5b2b966
JV		 2636
2637 if (ip == bond->master_ip)
2638 return 1;
2639
0883beca 2640 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f5b2b966
JV		 2641 if (ip == vlan->vlan_ip)
2642 return 1;
2643 }
2644
2645 return 0;
2646}
2647
c3ade5ca
JV		 2648/*
2649 * We go to the (large) trouble of VLAN tagging ARP frames because
2650 * switches in VLAN mode (especially if ports are configured as
2651 * "native" to a VLAN) might not pass non-tagged frames.
2652 */
d3bb52b0 2653static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2654{
2655 struct sk_buff *skb;
2656
5a03cdb7 2657 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
a4aee5c8 2658 slave_dev->name, dest_ip, src_ip, vlan_id);
3d632c3f 2659
c3ade5ca
JV		 2660 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2661 NULL, slave_dev->dev_addr, NULL);
2662
2663 if (!skb) {
a4aee5c8 2664 pr_err("ARP packet allocation failed\n");
c3ade5ca
JV		 2665 return;
2666 }
2667 if (vlan_id) {
2668 skb = vlan_put_tag(skb, vlan_id);
2669 if (!skb) {
a4aee5c8 2670 pr_err("failed to insert VLAN tag\n");
c3ade5ca
JV		 2671 return;
2672 }
2673 }
2674 arp_xmit(skb);
2675}
2676
2677
1da177e4
LT		 2678static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2679{
b23dd4fe 2680 int i, vlan_id;
d3bb52b0 2681 __be32 *targets = bond->params.arp_targets;
0883beca 2682 struct vlan_entry *vlan;
c3ade5ca 2683 struct net_device *vlan_dev;
c3ade5ca 2684 struct rtable *rt;
1da177e4 2685
6b780567
MW		 2686 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2687 if (!targets[i])
5a31bec0 2688 break;
5a03cdb7 2689 pr_debug("basa: target %x\n", targets[i]);
f35188fa 2690 if (!bond->vlgrp) {
5a03cdb7 2691 pr_debug("basa: empty vlan: arp_send\n");
c3ade5ca
JV		 2692 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2693 bond->master_ip, 0);
2694 continue;
2695 }
2696
2697 /*
2698 * If VLANs are configured, we do a route lookup to
2699 * determine which VLAN interface would be used, so we
2700 * can tag the ARP with the proper VLAN tag.
2701 */
78fbfd8a
DM		 2702 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2703 RTO_ONLINK, 0);
b23dd4fe 2704 if (IS_ERR(rt)) {
c3ade5ca 2705 if (net_ratelimit()) {
a4aee5c8 2706 pr_warning("%s: no route to arp_ip_target %pI4\n",
78fbfd8a 2707 bond->dev->name, &targets[i]);
c3ade5ca
JV		 2708 }
2709 continue;
2710 }
2711
2712 /*
2713 * This target is not on a VLAN
2714 */
d8d1f30b 2715 if (rt->dst.dev == bond->dev) {
ed4b9f80 2716 ip_rt_put(rt);
5a03cdb7 2717 pr_debug("basa: rtdev == bond->dev: arp_send\n");
c3ade5ca
JV		 2718 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2719 bond->master_ip, 0);
2720 continue;
2721 }
2722
2723 vlan_id = 0;
0883beca 2724 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
5c15bdec 2725 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
d8d1f30b 2726 if (vlan_dev == rt->dst.dev) {
c3ade5ca 2727 vlan_id = vlan->vlan_id;
5a03cdb7 2728 pr_debug("basa: vlan match on %s %d\n",
c3ade5ca
JV		 2729 vlan_dev->name, vlan_id);
2730 break;
2731 }
2732 }
2733
2734 if (vlan_id) {
ed4b9f80 2735 ip_rt_put(rt);
c3ade5ca
JV		 2736 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2737 vlan->vlan_ip, vlan_id);
2738 continue;
2739 }
2740
2741 if (net_ratelimit()) {
a4aee5c8 2742 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
78fbfd8a 2743 bond->dev->name, &targets[i],
d8d1f30b 2744 rt->dst.dev ? rt->dst.dev->name : "NULL");
c3ade5ca 2745 }
ed4b9f80 2746 ip_rt_put(rt);
c3ade5ca
JV		 2747 }
2748}
2749
d3bb52b0 2750static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966
JV		 2751{
2752 int i;
d3bb52b0 2753 __be32 *targets = bond->params.arp_targets;
f5b2b966 2754
f5b2b966 2755 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
5a03cdb7 2756 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
a4aee5c8
JP		 2757 &sip, &tip, i, &targets[i],
2758 bond_has_this_ip(bond, tip));
f5b2b966
JV		 2759 if (sip == targets[i]) {
2760 if (bond_has_this_ip(bond, tip))
2761 slave->last_arp_rx = jiffies;
2762 return;
2763 }
2764 }
2765}
2766
3aba891d
JP		 2767static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2768 struct slave *slave)
f5b2b966
JV		 2769{
2770 struct arphdr *arp;
f5b2b966 2771 unsigned char *arp_ptr;
d3bb52b0 2772 __be32 sip, tip;
f5b2b966 2773
3aba891d
JP		 2774 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2775 return;
f5b2b966 2776
f5b2b966
JV		 2777 read_lock(&bond->lock);
2778
3aba891d
JP		 2779 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2780 bond->dev->name, skb->dev->name);
f5b2b966 2781
3aba891d 2782 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
f5b2b966
JV		 2783 goto out_unlock;
2784
d0a92be0 2785 arp = arp_hdr(skb);
3aba891d 2786 if (arp->ar_hln != bond->dev->addr_len ||
f5b2b966
JV		 2787 skb->pkt_type == PACKET_OTHERHOST ||
2788 skb->pkt_type == PACKET_LOOPBACK ||
2789 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2790 arp->ar_pro != htons(ETH_P_IP) ||
2791 arp->ar_pln != 4)
2792 goto out_unlock;
2793
2794 arp_ptr = (unsigned char *)(arp + 1);
3aba891d 2795 arp_ptr += bond->dev->addr_len;
f5b2b966 2796 memcpy(&sip, arp_ptr, 4);
3aba891d 2797 arp_ptr += 4 + bond->dev->addr_len;
f5b2b966
JV		 2798 memcpy(&tip, arp_ptr, 4);
2799
5a03cdb7 2800 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
e30bc066 2801 bond->dev->name, slave->dev->name, bond_slave_state(slave),
a4aee5c8
JP		 2802 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2803 &sip, &tip);
f5b2b966
JV		 2804
2805 /*
2806 * Backup slaves won't see the ARP reply, but do come through
2807 * here for each ARP probe (so we swap the sip/tip to validate
2808 * the probe). In a "redundant switch, common router" type of
2809 * configuration, the ARP probe will (hopefully) travel from
2810 * the active, through one switch, the router, then the other
2811 * switch before reaching the backup.
2812 */
e30bc066 2813 if (bond_is_active_slave(slave))
f5b2b966
JV		 2814 bond_validate_arp(bond, slave, sip, tip);
2815 else
2816 bond_validate_arp(bond, slave, tip, sip);
2817
2818out_unlock:
2819 read_unlock(&bond->lock);
f5b2b966
JV		 2820}
2821
1da177e4
LT		 2822/*
2823 * this function is called regularly to monitor each slave's link
2824 * ensuring that traffic is being sent and received when arp monitoring
2825 * is used in load-balancing mode. if the adapter has been dormant, then an
2826 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2827 * arp monitoring in active backup mode.
2828 */
1b76b316 2829void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2830{
1b76b316
JV		 2831 struct bonding *bond = container_of(work, struct bonding,
2832 arp_work.work);
1da177e4
LT		 2833 struct slave *slave, *oldcurrent;
2834 int do_failover = 0;
2835 int delta_in_ticks;
2836 int i;
2837
2838 read_lock(&bond->lock);
2839
5ce0da8f 2840 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 2841
3d632c3f 2842 if (bond->kill_timers)
1da177e4 2843 goto out;
1da177e4 2844
3d632c3f 2845 if (bond->slave_cnt == 0)
1da177e4 2846 goto re_arm;
1da177e4
LT		 2847
2848 read_lock(&bond->curr_slave_lock);
2849 oldcurrent = bond->curr_active_slave;
2850 read_unlock(&bond->curr_slave_lock);
2851
2852 /* see if any of the previous devices are up now (i.e. they have
2853 * xmt and rcv traffic). the curr_active_slave does not come into
2854 * the picture unless it is null. also, slave->jiffies is not needed
2855 * here because we send an arp on each slave and give a slave as
2856 * long as it needs to get the tx/rx within the delta.
2857 * TODO: what about up/down delay in arp mode? it wasn't here before
2858 * so it can wait
2859 */
2860 bond_for_each_slave(bond, slave, i) {
cb32f2a0
JB		 2861 unsigned long trans_start = dev_trans_start(slave->dev);
2862
1da177e4 2863 if (slave->link != BOND_LINK_UP) {
cb32f2a0
JB		 2864 if (time_in_range(jiffies,
2865 trans_start - delta_in_ticks,
2866 trans_start + delta_in_ticks) &&
2867 time_in_range(jiffies,
2868 slave->dev->last_rx - delta_in_ticks,
2869 slave->dev->last_rx + delta_in_ticks)) {
1da177e4
LT		 2870
2871 slave->link = BOND_LINK_UP;
e30bc066 2872 bond_set_active_slave(slave);
1da177e4
LT		 2873
2874 /* primary_slave has no meaning in round-robin
2875 * mode. the window of a slave being up and
2876 * curr_active_slave being null after enslaving
2877 * is closed.
2878 */
2879 if (!oldcurrent) {
a4aee5c8
JP		 2880 pr_info("%s: link status definitely up for interface %s, ",
2881 bond->dev->name,
2882 slave->dev->name);
1da177e4
LT		 2883 do_failover = 1;
2884 } else {
a4aee5c8
JP		 2885 pr_info("%s: interface %s is now up\n",
2886 bond->dev->name,
2887 slave->dev->name);
1da177e4
LT		 2888 }
2889 }
2890 } else {
2891 /* slave->link == BOND_LINK_UP */
2892
2893 /* not all switches will respond to an arp request
2894 * when the source ip is 0, so don't take the link down
2895 * if we don't know our ip yet
2896 */
cb32f2a0
JB		 2897 if (!time_in_range(jiffies,
2898 trans_start - delta_in_ticks,
2899 trans_start + 2 * delta_in_ticks) ||
2900 !time_in_range(jiffies,
2901 slave->dev->last_rx - delta_in_ticks,
2902 slave->dev->last_rx + 2 * delta_in_ticks)) {
1da177e4
LT		 2903
2904 slave->link = BOND_LINK_DOWN;
e30bc066 2905 bond_set_backup_slave(slave);
1da177e4 2906
3d632c3f 2907 if (slave->link_failure_count < UINT_MAX)
1da177e4 2908 slave->link_failure_count++;
1da177e4 2909
a4aee5c8
JP		 2910 pr_info("%s: interface %s is now down.\n",
2911 bond->dev->name,
2912 slave->dev->name);
1da177e4 2913
3d632c3f 2914 if (slave == oldcurrent)
1da177e4 2915 do_failover = 1;
1da177e4
LT		 2916 }
2917 }
2918
2919 /* note: if switch is in round-robin mode, all links
2920 * must tx arp to ensure all links rx an arp - otherwise
2921 * links may oscillate or not come up at all; if switch is
2922 * in something like xor mode, there is nothing we can
2923 * do - all replies will be rx'ed on same link causing slaves
2924 * to be unstable during low/no traffic periods
2925 */
3d632c3f 2926 if (IS_UP(slave->dev))
1da177e4 2927 bond_arp_send_all(bond, slave);
1da177e4
LT		 2928 }
2929
2930 if (do_failover) {
e843fa50 2931 block_netpoll_tx();
059fe7a5 2932 write_lock_bh(&bond->curr_slave_lock);
1da177e4
LT		 2933
2934 bond_select_active_slave(bond);
2935
059fe7a5 2936 write_unlock_bh(&bond->curr_slave_lock);
e843fa50 2937 unblock_netpoll_tx();
1da177e4
LT		 2938 }
2939
2940re_arm:
1b76b316
JV		 2941 if (bond->params.arp_interval)
2942 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 2943out:
2944 read_unlock(&bond->lock);
2945}
2946
2947/*
b2220cad
JV		 2948 * Called to inspect slaves for active-backup mode ARP monitor link state
2949 * changes. Sets new_link in slaves to specify what action should take
2950 * place for the slave. Returns 0 if no changes are found, >0 if changes
2951 * to link states must be committed.
2952 *
2953 * Called with bond->lock held for read.
1da177e4 2954 */
b2220cad 2955static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
1da177e4 2956{
1da177e4 2957 struct slave *slave;
b2220cad 2958 int i, commit = 0;
cb32f2a0 2959 unsigned long trans_start;
1da177e4 2960
b2220cad
JV		 2961 bond_for_each_slave(bond, slave, i) {
2962 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 2963
b2220cad 2964 if (slave->link != BOND_LINK_UP) {
cb32f2a0
JB		 2965 if (time_in_range(jiffies,
2966 slave_last_rx(bond, slave) - delta_in_ticks,
2967 slave_last_rx(bond, slave) + delta_in_ticks)) {
2968
b2220cad
JV		 2969 slave->new_link = BOND_LINK_UP;
2970 commit++;
2971 }
1da177e4 2972
b2220cad
JV		 2973 continue;
2974 }
1da177e4 2975
b2220cad
JV		 2976 /*
2977 * Give slaves 2*delta after being enslaved or made
2978 * active. This avoids bouncing, as the last receive
2979 * times need a full ARP monitor cycle to be updated.
2980 */
cb32f2a0
JB		 2981 if (time_in_range(jiffies,
2982 slave->jiffies - delta_in_ticks,
2983 slave->jiffies + 2 * delta_in_ticks))
b2220cad
JV		 2984 continue;
2985
2986 /*
2987 * Backup slave is down if:
2988 * - No current_arp_slave AND
2989 * - more than 3*delta since last receive AND
2990 * - the bond has an IP address
2991 *
2992 * Note: a non-null current_arp_slave indicates
2993 * the curr_active_slave went down and we are
2994 * searching for a new one; under this condition
2995 * we only take the curr_active_slave down - this
2996 * gives each slave a chance to tx/rx traffic
2997 * before being taken out
2998 */
e30bc066 2999 if (!bond_is_active_slave(slave) &&
b2220cad 3000 !bond->current_arp_slave &&
cb32f2a0
JB		 3001 !time_in_range(jiffies,
3002 slave_last_rx(bond, slave) - delta_in_ticks,
3003 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3004
b2220cad
JV		 3005 slave->new_link = BOND_LINK_DOWN;
3006 commit++;
3007 }
3008
3009 /*
3010 * Active slave is down if:
3011 * - more than 2*delta since transmitting OR
3012 * - (more than 2*delta since receive AND
3013 * the bond has an IP address)
3014 */
cb32f2a0 3015 trans_start = dev_trans_start(slave->dev);
e30bc066 3016 if (bond_is_active_slave(slave) &&
cb32f2a0
JB		 3017 (!time_in_range(jiffies,
3018 trans_start - delta_in_ticks,
3019 trans_start + 2 * delta_in_ticks) ||
3020 !time_in_range(jiffies,
3021 slave_last_rx(bond, slave) - delta_in_ticks,
3022 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3023
b2220cad
JV		 3024 slave->new_link = BOND_LINK_DOWN;
3025 commit++;
3026 }
1da177e4
LT		 3027 }
3028
b2220cad
JV		 3029 return commit;
3030}
1da177e4 3031
b2220cad
JV		 3032/*
3033 * Called to commit link state changes noted by inspection step of
3034 * active-backup mode ARP monitor.
3035 *
3036 * Called with RTNL and bond->lock for read.
3037 */
3038static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3039{
3040 struct slave *slave;
3041 int i;
cb32f2a0 3042 unsigned long trans_start;
1da177e4 3043
b2220cad
JV		 3044 bond_for_each_slave(bond, slave, i) {
3045 switch (slave->new_link) {
3046 case BOND_LINK_NOCHANGE:
3047 continue;
ff59c456 3048
b2220cad 3049 case BOND_LINK_UP:
cb32f2a0 3050 trans_start = dev_trans_start(slave->dev);
b9f60253 3051 if ((!bond->curr_active_slave &&
cb32f2a0
JB		 3052 time_in_range(jiffies,
3053 trans_start - delta_in_ticks,
3054 trans_start + delta_in_ticks)) ||
b9f60253 3055 bond->curr_active_slave != slave) {
b2220cad 3056 slave->link = BOND_LINK_UP;
b2220cad
JV		 3057 bond->current_arp_slave = NULL;
3058
a4aee5c8 3059 pr_info("%s: link status definitely up for interface %s.\n",
b9f60253 3060 bond->dev->name, slave->dev->name);
b2220cad 3061
b9f60253
JP		 3062 if (!bond->curr_active_slave ||
3063 (slave == bond->primary_slave))
3064 goto do_failover;
1da177e4 3065
b9f60253 3066 }
1da177e4 3067
b9f60253 3068 continue;
1da177e4 3069
b2220cad
JV		 3070 case BOND_LINK_DOWN:
3071 if (slave->link_failure_count < UINT_MAX)
3072 slave->link_failure_count++;
3073
3074 slave->link = BOND_LINK_DOWN;
b9f60253 3075 bond_set_slave_inactive_flags(slave);
b2220cad 3076
a4aee5c8 3077 pr_info("%s: link status definitely down for interface %s, disabling it\n",
b9f60253 3078 bond->dev->name, slave->dev->name);
b2220cad 3079
b9f60253 3080 if (slave == bond->curr_active_slave) {
b2220cad 3081 bond->current_arp_slave = NULL;
b9f60253 3082 goto do_failover;
1da177e4 3083 }
b9f60253
JP		 3084
3085 continue;
b2220cad
JV		 3086
3087 default:
a4aee5c8 3088 pr_err("%s: impossible: new_link %d on slave %s\n",
b2220cad
JV		 3089 bond->dev->name, slave->new_link,
3090 slave->dev->name);
b9f60253 3091 continue;
1da177e4 3092 }
1da177e4 3093
b9f60253
JP		 3094do_failover:
3095 ASSERT_RTNL();
e843fa50 3096 block_netpoll_tx();
b2220cad 3097 write_lock_bh(&bond->curr_slave_lock);
b9f60253 3098 bond_select_active_slave(bond);
b2220cad 3099 write_unlock_bh(&bond->curr_slave_lock);
e843fa50 3100 unblock_netpoll_tx();
b2220cad 3101 }
1da177e4 3102
b2220cad
JV		 3103 bond_set_carrier(bond);
3104}
1da177e4 3105
b2220cad
JV		 3106/*
3107 * Send ARP probes for active-backup mode ARP monitor.
3108 *
3109 * Called with bond->lock held for read.
3110 */
3111static void bond_ab_arp_probe(struct bonding *bond)
3112{
3113 struct slave *slave;
3114 int i;
1da177e4 3115
b2220cad 3116 read_lock(&bond->curr_slave_lock);
1da177e4 3117
b2220cad 3118 if (bond->current_arp_slave && bond->curr_active_slave)
a4aee5c8
JP		 3119 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3120 bond->current_arp_slave->dev->name,
3121 bond->curr_active_slave->dev->name);
1da177e4 3122
b2220cad
JV		 3123 if (bond->curr_active_slave) {
3124 bond_arp_send_all(bond, bond->curr_active_slave);
3125 read_unlock(&bond->curr_slave_lock);
3126 return;
3127 }
1da177e4 3128
b2220cad 3129 read_unlock(&bond->curr_slave_lock);
059fe7a5 3130
b2220cad
JV		 3131 /* if we don't have a curr_active_slave, search for the next available
3132 * backup slave from the current_arp_slave and make it the candidate
3133 * for becoming the curr_active_slave
3134 */
1da177e4 3135
b2220cad
JV		 3136 if (!bond->current_arp_slave) {
3137 bond->current_arp_slave = bond->first_slave;
3138 if (!bond->current_arp_slave)
3139 return;
3140 }
1da177e4 3141
b2220cad 3142 bond_set_slave_inactive_flags(bond->current_arp_slave);
059fe7a5 3143
b2220cad
JV		 3144 /* search for next candidate */
3145 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3146 if (IS_UP(slave->dev)) {
3147 slave->link = BOND_LINK_BACK;
3148 bond_set_slave_active_flags(slave);
3149 bond_arp_send_all(bond, slave);
1da177e4 3150 slave->jiffies = jiffies;
b2220cad
JV		 3151 bond->current_arp_slave = slave;
3152 break;
1da177e4
LT		 3153 }
3154
b2220cad
JV		 3155 /* if the link state is up at this point, we
3156 * mark it down - this can happen if we have
3157 * simultaneous link failures and
3158 * reselect_active_interface doesn't make this
3159 * one the current slave so it is still marked
3160 * up when it is actually down
1da177e4 3161 */
b2220cad
JV		 3162 if (slave->link == BOND_LINK_UP) {
3163 slave->link = BOND_LINK_DOWN;
3164 if (slave->link_failure_count < UINT_MAX)
3165 slave->link_failure_count++;
1da177e4 3166
b2220cad
JV		 3167 bond_set_slave_inactive_flags(slave);
3168
a4aee5c8
JP		 3169 pr_info("%s: backup interface %s is now down.\n",
3170 bond->dev->name, slave->dev->name);
1da177e4 3171 }
b2220cad
JV		 3172 }
3173}
1da177e4 3174
b2220cad
JV		 3175void bond_activebackup_arp_mon(struct work_struct *work)
3176{
3177 struct bonding *bond = container_of(work, struct bonding,
3178 arp_work.work);
ad246c99 3179 bool should_notify_peers = false;
b2220cad 3180 int delta_in_ticks;
1da177e4 3181
b2220cad 3182 read_lock(&bond->lock);
1da177e4 3183
b2220cad
JV		 3184 if (bond->kill_timers)
3185 goto out;
1da177e4 3186
b2220cad 3187 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
1da177e4 3188
b2220cad
JV		 3189 if (bond->slave_cnt == 0)
3190 goto re_arm;
3191
ad246c99
BH		 3192 should_notify_peers = bond_should_notify_peers(bond);
3193
b2220cad
JV		 3194 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3195 read_unlock(&bond->lock);
3196 rtnl_lock();
3197 read_lock(&bond->lock);
3198
3199 bond_ab_arp_commit(bond, delta_in_ticks);
3200
3201 read_unlock(&bond->lock);
3202 rtnl_unlock();
3203 read_lock(&bond->lock);
1da177e4
LT		 3204 }
3205
b2220cad
JV		 3206 bond_ab_arp_probe(bond);
3207
1da177e4 3208re_arm:
3d632c3f 3209 if (bond->params.arp_interval)
1b76b316 3210 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
1da177e4
LT		 3211out:
3212 read_unlock(&bond->lock);
ad246c99
BH		 3213
3214 if (should_notify_peers) {
3215 rtnl_lock();
3216 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3217 rtnl_unlock();
3218 }
1da177e4
LT		 3219}
3220
1da177e4
LT		 3221/*-------------------------- netdev event handling --------------------------*/
3222
3223/*
3224 * Change device name
3225 */
3226static int bond_event_changename(struct bonding *bond)
3227{
1da177e4
LT		 3228 bond_remove_proc_entry(bond);
3229 bond_create_proc_entry(bond);
7e083840 3230
f073c7ca
TI		 3231 bond_debug_reregister(bond);
3232
1da177e4
LT		 3233 return NOTIFY_DONE;
3234}
3235
3d632c3f
SH		 3236static int bond_master_netdev_event(unsigned long event,
3237 struct net_device *bond_dev)
1da177e4 3238{
454d7c9b 3239 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 3240
3241 switch (event) {
3242 case NETDEV_CHANGENAME:
3243 return bond_event_changename(event_bond);
1da177e4
LT		 3244 default:
3245 break;
3246 }
3247
3248 return NOTIFY_DONE;
3249}
3250
3d632c3f
SH		 3251static int bond_slave_netdev_event(unsigned long event,
3252 struct net_device *slave_dev)
1da177e4
LT		 3253{
3254 struct net_device *bond_dev = slave_dev->master;
454d7c9b 3255 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3256
3257 switch (event) {
3258 case NETDEV_UNREGISTER:
3259 if (bond_dev) {
1284cd3a
JV		 3260 if (bond->setup_by_slave)
3261 bond_release_and_destroy(bond_dev, slave_dev);
3262 else
3263 bond_release(bond_dev, slave_dev);
1da177e4
LT		 3264 }
3265 break;
3266 case NETDEV_CHANGE:
17d04500
JV		 3267 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3268 struct slave *slave;
3269
3270 slave = bond_get_slave_by_dev(bond, slave_dev);
3271 if (slave) {
5d30530e 3272 u32 old_speed = slave->speed;
65cce19c 3273 u8 old_duplex = slave->duplex;
17d04500
JV		 3274
3275 bond_update_speed_duplex(slave);
3276
3277 if (bond_is_lb(bond))
3278 break;
3279
3280 if (old_speed != slave->speed)
3281 bond_3ad_adapter_speed_changed(slave);
3282 if (old_duplex != slave->duplex)
3283 bond_3ad_adapter_duplex_changed(slave);
3284 }
3285 }
3286
1da177e4
LT		 3287 break;
3288 case NETDEV_DOWN:
3289 /*
3290 * ... Or is it this?
3291 */
3292 break;
3293 case NETDEV_CHANGEMTU:
3294 /*
3295 * TODO: Should slaves be allowed to
3296 * independently alter their MTU? For
3297 * an active-backup bond, slaves need
3298 * not be the same type of device, so
3299 * MTUs may vary. For other modes,
3300 * slaves arguably should have the
3301 * same MTUs. To do this, we'd need to
3302 * take over the slave's change_mtu
3303 * function for the duration of their
3304 * servitude.
3305 */
3306 break;
3307 case NETDEV_CHANGENAME:
3308 /*
3309 * TODO: handle changing the primary's name
3310 */
3311 break;
8531c5ff
AK		 3312 case NETDEV_FEAT_CHANGE:
3313 bond_compute_features(bond);
3314 break;
1da177e4
LT		 3315 default:
3316 break;
3317 }
3318
3319 return NOTIFY_DONE;
3320}
3321
3322/*
3323 * bond_netdev_event: handle netdev notifier chain events.
3324 *
3325 * This function receives events for the netdev chain. The caller (an
e041c683 3326 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 3327 * locks for us to safely manipulate the slave devices (RTNL lock,
3328 * dev_probe_lock).
3329 */
3d632c3f
SH		 3330static int bond_netdev_event(struct notifier_block *this,
3331 unsigned long event, void *ptr)
1da177e4
LT		 3332{
3333 struct net_device *event_dev = (struct net_device *)ptr;
3334
5a03cdb7 3335 pr_debug("event_dev: %s, event: %lx\n",
a4aee5c8
JP		 3336 event_dev ? event_dev->name : "None",
3337 event);
1da177e4 3338
0b680e75
JV		 3339 if (!(event_dev->priv_flags & IFF_BONDING))
3340 return NOTIFY_DONE;
3341
1da177e4 3342 if (event_dev->flags & IFF_MASTER) {
5a03cdb7 3343 pr_debug("IFF_MASTER\n");
1da177e4
LT		 3344 return bond_master_netdev_event(event, event_dev);
3345 }
3346
3347 if (event_dev->flags & IFF_SLAVE) {
5a03cdb7 3348 pr_debug("IFF_SLAVE\n");
1da177e4
LT		 3349 return bond_slave_netdev_event(event, event_dev);
3350 }
3351
3352 return NOTIFY_DONE;
3353}
3354
c3ade5ca
JV		 3355/*
3356 * bond_inetaddr_event: handle inetaddr notifier chain events.
3357 *
3358 * We keep track of device IPs primarily to use as source addresses in
3359 * ARP monitor probes (rather than spewing out broadcasts all the time).
3360 *
3361 * We track one IP for the main device (if it has one), plus one per VLAN.
3362 */
3363static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3364{
3365 struct in_ifaddr *ifa = ptr;
3366 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
ec87fd3b 3367 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
0883beca
PE		 3368 struct bonding *bond;
3369 struct vlan_entry *vlan;
c3ade5ca 3370
ec87fd3b 3371 list_for_each_entry(bond, &bn->dev_list, bond_list) {
c3ade5ca
JV		 3372 if (bond->dev == event_dev) {
3373 switch (event) {
3374 case NETDEV_UP:
3375 bond->master_ip = ifa->ifa_local;
3376 return NOTIFY_OK;
3377 case NETDEV_DOWN:
3378 bond->master_ip = bond_glean_dev_ip(bond->dev);
3379 return NOTIFY_OK;
3380 default:
3381 return NOTIFY_DONE;
3382 }
3383 }
3384
0883beca 3385 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
f35188fa
JV		 3386 if (!bond->vlgrp)
3387 continue;
5c15bdec 3388 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
c3ade5ca
JV		 3389 if (vlan_dev == event_dev) {
3390 switch (event) {
3391 case NETDEV_UP:
3392 vlan->vlan_ip = ifa->ifa_local;
3393 return NOTIFY_OK;
3394 case NETDEV_DOWN:
3395 vlan->vlan_ip =
3396 bond_glean_dev_ip(vlan_dev);
3397 return NOTIFY_OK;
3398 default:
3399 return NOTIFY_DONE;
3400 }
3401 }
3402 }
3403 }
3404 return NOTIFY_DONE;
3405}
3406
1da177e4
LT		 3407static struct notifier_block bond_netdev_notifier = {
3408 .notifier_call = bond_netdev_event,
3409};
3410
c3ade5ca
JV		 3411static struct notifier_block bond_inetaddr_notifier = {
3412 .notifier_call = bond_inetaddr_event,
3413};
3414
169a3e66
JV		 3415/*---------------------------- Hashing Policies -----------------------------*/
3416
6f6652be
JV		 3417/*
3418 * Hash for the output device based upon layer 2 and layer 3 data. If
3419 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3420 */
a361c83c 3421static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
6f6652be
JV		 3422{
3423 struct ethhdr *data = (struct ethhdr *)skb->data;
3424 struct iphdr *iph = ip_hdr(skb);
3425
f14c4e4e 3426 if (skb->protocol == htons(ETH_P_IP)) {
6f6652be 3427 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
d3da6831 3428 (data->h_dest[5] ^ data->h_source[5])) % count;
6f6652be
JV		 3429 }
3430
d3da6831 3431 return (data->h_dest[5] ^ data->h_source[5]) % count;
6f6652be
JV		 3432}
3433
169a3e66 3434/*
59c51591 3435 * Hash for the output device based upon layer 3 and layer 4 data. If
169a3e66
JV		 3436 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3437 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3438 */
a361c83c 3439static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
169a3e66
JV		 3440{
3441 struct ethhdr *data = (struct ethhdr *)skb->data;
eddc9ec5 3442 struct iphdr *iph = ip_hdr(skb);
d3bb52b0 3443 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
169a3e66
JV		 3444 int layer4_xor = 0;
3445
f14c4e4e
BH		 3446 if (skb->protocol == htons(ETH_P_IP)) {
3447 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
169a3e66
JV		 3448 (iph->protocol == IPPROTO_TCP ||
3449 iph->protocol == IPPROTO_UDP)) {
d3bb52b0 3450 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
169a3e66
JV		 3451 }
3452 return (layer4_xor ^
3453 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3454
3455 }
3456
d3da6831 3457 return (data->h_dest[5] ^ data->h_source[5]) % count;
169a3e66
JV		 3458}
3459
3460/*
3461 * Hash for the output device based upon layer 2 data
3462 */
a361c83c 3463static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
169a3e66
JV		 3464{
3465 struct ethhdr *data = (struct ethhdr *)skb->data;
3466
d3da6831 3467 return (data->h_dest[5] ^ data->h_source[5]) % count;
169a3e66
JV		 3468}
3469
1da177e4
LT		 3470/*-------------------------- Device entry points ----------------------------*/
3471
3472static int bond_open(struct net_device *bond_dev)
3473{
454d7c9b 3474 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3475
3476 bond->kill_timers = 0;
3477
5a37e8ca
FL		 3478 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3479
58402054 3480 if (bond_is_lb(bond)) {
1da177e4
LT		 3481 /* bond_alb_initialize must be called before the timer
3482 * is started.
3483 */
3484 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3485 /* something went wrong - fail the open operation */
b473946a 3486 return -ENOMEM;
1da177e4
LT		 3487 }
3488
1b76b316
JV		 3489 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3490 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3491 }
3492
3493 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316
JV		 3494 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3495 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3496 }
3497
3498 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316
JV		 3499 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3500 INIT_DELAYED_WORK(&bond->arp_work,
3501 bond_activebackup_arp_mon);
3502 else
3503 INIT_DELAYED_WORK(&bond->arp_work,
3504 bond_loadbalance_arp_mon);
3505
3506 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966 3507 if (bond->params.arp_validate)
3aba891d 3508 bond->recv_probe = bond_arp_rcv;
1da177e4
LT		 3509 }
3510
3511 if (bond->params.mode == BOND_MODE_8023AD) {
a40745f5 3512 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
1b76b316 3513 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4 3514 /* register to receive LACPDUs */
3aba891d 3515 bond->recv_probe = bond_3ad_lacpdu_recv;
fd989c83 3516 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3517 }
3518
3519 return 0;
3520}
3521
3522static int bond_close(struct net_device *bond_dev)
3523{
454d7c9b 3524 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3525
1da177e4
LT		 3526 write_lock_bh(&bond->lock);
3527
ad246c99
BH		 3528 bond->send_peer_notif = 0;
3529
1da177e4
LT		 3530 /* signal timers not to re-arm */
3531 bond->kill_timers = 1;
3532
3533 write_unlock_bh(&bond->lock);
3534
1da177e4 3535 if (bond->params.miimon) { /* link check interval, in milliseconds. */
1b76b316 3536 cancel_delayed_work(&bond->mii_work);
1da177e4
LT		 3537 }
3538
3539 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3540 cancel_delayed_work(&bond->arp_work);
1da177e4
LT		 3541 }
3542
3543 switch (bond->params.mode) {
3544 case BOND_MODE_8023AD:
1b76b316 3545 cancel_delayed_work(&bond->ad_work);
1da177e4
LT		 3546 break;
3547 case BOND_MODE_TLB:
3548 case BOND_MODE_ALB:
1b76b316 3549 cancel_delayed_work(&bond->alb_work);
1da177e4
LT		 3550 break;
3551 default:
3552 break;
3553 }
3554
5a37e8ca
FL		 3555 if (delayed_work_pending(&bond->mcast_work))
3556 cancel_delayed_work(&bond->mcast_work);
1da177e4 3557
58402054 3558 if (bond_is_lb(bond)) {
1da177e4
LT		 3559 /* Must be called only after all
3560 * slaves have been released
3561 */
3562 bond_alb_deinitialize(bond);
3563 }
3aba891d 3564 bond->recv_probe = NULL;
1da177e4
LT		 3565
3566 return 0;
3567}
3568
28172739
ED		 3569static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3570 struct rtnl_link_stats64 *stats)
1da177e4 3571{
454d7c9b 3572 struct bonding *bond = netdev_priv(bond_dev);
28172739 3573 struct rtnl_link_stats64 temp;
1da177e4
LT		 3574 struct slave *slave;
3575 int i;
3576
28172739 3577 memset(stats, 0, sizeof(*stats));
1da177e4
LT		 3578
3579 read_lock_bh(&bond->lock);
3580
3581 bond_for_each_slave(bond, slave, i) {
be1f3c2c 3582 const struct rtnl_link_stats64 *sstats =
28172739
ED		 3583 dev_get_stats(slave->dev, &temp);
3584
3585 stats->rx_packets += sstats->rx_packets;
3586 stats->rx_bytes += sstats->rx_bytes;
3587 stats->rx_errors += sstats->rx_errors;
3588 stats->rx_dropped += sstats->rx_dropped;
3589
3590 stats->tx_packets += sstats->tx_packets;
3591 stats->tx_bytes += sstats->tx_bytes;
3592 stats->tx_errors += sstats->tx_errors;
3593 stats->tx_dropped += sstats->tx_dropped;
3594
3595 stats->multicast += sstats->multicast;
3596 stats->collisions += sstats->collisions;
3597
3598 stats->rx_length_errors += sstats->rx_length_errors;
3599 stats->rx_over_errors += sstats->rx_over_errors;
3600 stats->rx_crc_errors += sstats->rx_crc_errors;
3601 stats->rx_frame_errors += sstats->rx_frame_errors;
3602 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3603 stats->rx_missed_errors += sstats->rx_missed_errors;
3604
3605 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3606 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3607 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3608 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3609 stats->tx_window_errors += sstats->tx_window_errors;
2439f9eb
AG		 3610 }
3611
1da177e4
LT		 3612 read_unlock_bh(&bond->lock);
3613
3614 return stats;
3615}
3616
3617static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3618{
3619 struct net_device *slave_dev = NULL;
3620 struct ifbond k_binfo;
3621 struct ifbond __user *u_binfo = NULL;
3622 struct ifslave k_sinfo;
3623 struct ifslave __user *u_sinfo = NULL;
3624 struct mii_ioctl_data *mii = NULL;
1da177e4
LT		 3625 int res = 0;
3626
a4aee5c8 3627 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
1da177e4
LT		 3628
3629 switch (cmd) {
1da177e4
LT		 3630 case SIOCGMIIPHY:
3631 mii = if_mii(ifr);
3d632c3f 3632 if (!mii)
1da177e4 3633 return -EINVAL;
3d632c3f 3634
1da177e4
LT		 3635 mii->phy_id = 0;
3636 /* Fall Through */
3637 case SIOCGMIIREG:
3638 /*
3639 * We do this again just in case we were called by SIOCGMIIREG
3640 * instead of SIOCGMIIPHY.
3641 */
3642 mii = if_mii(ifr);
3d632c3f 3643 if (!mii)
1da177e4 3644 return -EINVAL;
3d632c3f 3645
1da177e4
LT		 3646
3647 if (mii->reg_num == 1) {
454d7c9b 3648 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3649 mii->val_out = 0;
6603a6f2 3650 read_lock(&bond->lock);
1da177e4 3651 read_lock(&bond->curr_slave_lock);
3d632c3f 3652 if (netif_carrier_ok(bond->dev))
1da177e4 3653 mii->val_out = BMSR_LSTATUS;
3d632c3f 3654
1da177e4 3655 read_unlock(&bond->curr_slave_lock);
6603a6f2 3656 read_unlock(&bond->lock);
1da177e4
LT		 3657 }
3658
3659 return 0;
3660 case BOND_INFO_QUERY_OLD:
3661 case SIOCBONDINFOQUERY:
3662 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3663
3d632c3f 3664 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
1da177e4 3665 return -EFAULT;
1da177e4
LT		 3666
3667 res = bond_info_query(bond_dev, &k_binfo);
3d632c3f
SH		 3668 if (res == 0 &&
3669 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3670 return -EFAULT;
1da177e4
LT		 3671
3672 return res;
3673 case BOND_SLAVE_INFO_QUERY_OLD:
3674 case SIOCBONDSLAVEINFOQUERY:
3675 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3676
3d632c3f 3677 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
1da177e4 3678 return -EFAULT;
1da177e4
LT		 3679
3680 res = bond_slave_info_query(bond_dev, &k_sinfo);
3d632c3f
SH		 3681 if (res == 0 &&
3682 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3683 return -EFAULT;
1da177e4
LT		 3684
3685 return res;
3686 default:
3687 /* Go on */
3688 break;
3689 }
3690
3d632c3f 3691 if (!capable(CAP_NET_ADMIN))
1da177e4 3692 return -EPERM;
1da177e4 3693
ec87fd3b 3694 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
1da177e4 3695
a4aee5c8 3696 pr_debug("slave_dev=%p:\n", slave_dev);
1da177e4 3697
3d632c3f 3698 if (!slave_dev)
1da177e4 3699 res = -ENODEV;
3d632c3f 3700 else {
a4aee5c8 3701 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
1da177e4
LT		 3702 switch (cmd) {
3703 case BOND_ENSLAVE_OLD:
3704 case SIOCBONDENSLAVE:
3705 res = bond_enslave(bond_dev, slave_dev);
3706 break;
3707 case BOND_RELEASE_OLD:
3708 case SIOCBONDRELEASE:
3709 res = bond_release(bond_dev, slave_dev);
3710 break;
3711 case BOND_SETHWADDR_OLD:
3712 case SIOCBONDSETHWADDR:
3713 res = bond_sethwaddr(bond_dev, slave_dev);
3714 break;
3715 case BOND_CHANGE_ACTIVE_OLD:
3716 case SIOCBONDCHANGEACTIVE:
3717 res = bond_ioctl_change_active(bond_dev, slave_dev);
3718 break;
3719 default:
3720 res = -EOPNOTSUPP;
3721 }
3722
3723 dev_put(slave_dev);
3724 }
3725
1da177e4
LT		 3726 return res;
3727}
3728
22bedad3
JP		 3729static bool bond_addr_in_mc_list(unsigned char *addr,
3730 struct netdev_hw_addr_list *list,
3731 int addrlen)
3732{
3733 struct netdev_hw_addr *ha;
3734
3735 netdev_hw_addr_list_for_each(ha, list)
3736 if (!memcmp(ha->addr, addr, addrlen))
3737 return true;
3738
3739 return false;
3740}
3741
1da177e4
LT		 3742static void bond_set_multicast_list(struct net_device *bond_dev)
3743{
454d7c9b 3744 struct bonding *bond = netdev_priv(bond_dev);
22bedad3
JP		 3745 struct netdev_hw_addr *ha;
3746 bool found;
1da177e4 3747
1da177e4
LT		 3748 /*
3749 * Do promisc before checking multicast_mode
3750 */
3d632c3f 3751 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
7e1a1ac1
WC		 3752 /*
3753 * FIXME: Need to handle the error when one of the multi-slaves
3754 * encounters error.
3755 */
1da177e4 3756 bond_set_promiscuity(bond, 1);
1da177e4 3757
3d632c3f
SH		 3758
3759 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
1da177e4 3760 bond_set_promiscuity(bond, -1);
3d632c3f 3761
1da177e4
LT		 3762
3763 /* set allmulti flag to slaves */
3d632c3f 3764 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
7e1a1ac1
WC		 3765 /*
3766 * FIXME: Need to handle the error when one of the multi-slaves
3767 * encounters error.
3768 */
1da177e4 3769 bond_set_allmulti(bond, 1);
1da177e4 3770
3d632c3f
SH		 3771
3772 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
1da177e4 3773 bond_set_allmulti(bond, -1);
3d632c3f 3774
1da177e4 3775
80ee5ad2
JV		 3776 read_lock(&bond->lock);
3777
1da177e4
LT		 3778 bond->flags = bond_dev->flags;
3779
3780 /* looking for addresses to add to slaves' mc list */
22bedad3
JP		 3781 netdev_for_each_mc_addr(ha, bond_dev) {
3782 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3783 bond_dev->addr_len);
3784 if (!found)
3785 bond_mc_add(bond, ha->addr);
1da177e4
LT		 3786 }
3787
3788 /* looking for addresses to delete from slaves' list */
22bedad3
JP		 3789 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3790 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3791 bond_dev->addr_len);
3792 if (!found)
3793 bond_mc_del(bond, ha->addr);
1da177e4
LT		 3794 }
3795
3796 /* save master's multicast list */
22bedad3
JP		 3797 __hw_addr_flush(&bond->mc_list);
3798 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3799 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
1da177e4 3800
80ee5ad2 3801 read_unlock(&bond->lock);
1da177e4
LT		 3802}
3803
00829823
SH		 3804static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3805{
3806 struct bonding *bond = netdev_priv(dev);
3807 struct slave *slave = bond->first_slave;
3808
3809 if (slave) {
3810 const struct net_device_ops *slave_ops
3811 = slave->dev->netdev_ops;
3812 if (slave_ops->ndo_neigh_setup)
72e2240f 3813 return slave_ops->ndo_neigh_setup(slave->dev, parms);
00829823
SH		 3814 }
3815 return 0;
3816}
3817
1da177e4
LT		 3818/*
3819 * Change the MTU of all of a master's slaves to match the master
3820 */
3821static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3822{
454d7c9b 3823 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3824 struct slave *slave, *stop_at;
3825 int res = 0;
3826 int i;
3827
5a03cdb7 3828 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
a4aee5c8 3829 (bond_dev ? bond_dev->name : "None"), new_mtu);
1da177e4
LT		 3830
3831 /* Can't hold bond->lock with bh disabled here since
3832 * some base drivers panic. On the other hand we can't
3833 * hold bond->lock without bh disabled because we'll
3834 * deadlock. The only solution is to rely on the fact
3835 * that we're under rtnl_lock here, and the slaves
3836 * list won't change. This doesn't solve the problem
3837 * of setting the slave's MTU while it is
3838 * transmitting, but the assumption is that the base
3839 * driver can handle that.
3840 *
3841 * TODO: figure out a way to safely iterate the slaves
3842 * list, but without holding a lock around the actual
3843 * call to the base driver.
3844 */
3845
3846 bond_for_each_slave(bond, slave, i) {
a4aee5c8
JP		 3847 pr_debug("s %p s->p %p c_m %p\n",
3848 slave,
3849 slave->prev,
3850 slave->dev->netdev_ops->ndo_change_mtu);
e944ef79 3851
1da177e4
LT		 3852 res = dev_set_mtu(slave->dev, new_mtu);
3853
3854 if (res) {
3855 /* If we failed to set the slave's mtu to the new value
3856 * we must abort the operation even in ACTIVE_BACKUP
3857 * mode, because if we allow the backup slaves to have
3858 * different mtu values than the active slave we'll
3859 * need to change their mtu when doing a failover. That
3860 * means changing their mtu from timer context, which
3861 * is probably not a good idea.
3862 */
5a03cdb7 3863 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 3864 goto unwind;
3865 }
3866 }
3867
3868 bond_dev->mtu = new_mtu;
3869
3870 return 0;
3871
3872unwind:
3873 /* unwind from head to the slave that failed */
3874 stop_at = slave;
3875 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3876 int tmp_res;
3877
3878 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3879 if (tmp_res) {
a4aee5c8
JP		 3880 pr_debug("unwind err %d dev %s\n",
3881 tmp_res, slave->dev->name);
1da177e4
LT		 3882 }
3883 }
3884
3885 return res;
3886}
3887
3888/*
3889 * Change HW address
3890 *
3891 * Note that many devices must be down to change the HW address, and
3892 * downing the master releases all slaves. We can make bonds full of
3893 * bonding devices to test this, however.
3894 */
3895static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3896{
454d7c9b 3897 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3898 struct sockaddr *sa = addr, tmp_sa;
3899 struct slave *slave, *stop_at;
3900 int res = 0;
3901 int i;
3902
eb7cc59a
SH		 3903 if (bond->params.mode == BOND_MODE_ALB)
3904 return bond_alb_set_mac_address(bond_dev, addr);
3905
3906
a4aee5c8
JP		 3907 pr_debug("bond=%p, name=%s\n",
3908 bond, bond_dev ? bond_dev->name : "None");
1da177e4 3909
dd957c57 3910 /*
3915c1e8
JV		 3911 * If fail_over_mac is set to active, do nothing and return
3912 * success. Returning an error causes ifenslave to fail.
dd957c57 3913 */
3915c1e8 3914 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
dd957c57 3915 return 0;
2ab82852 3916
3d632c3f 3917 if (!is_valid_ether_addr(sa->sa_data))
1da177e4 3918 return -EADDRNOTAVAIL;
1da177e4
LT		 3919
3920 /* Can't hold bond->lock with bh disabled here since
3921 * some base drivers panic. On the other hand we can't
3922 * hold bond->lock without bh disabled because we'll
3923 * deadlock. The only solution is to rely on the fact
3924 * that we're under rtnl_lock here, and the slaves
3925 * list won't change. This doesn't solve the problem
3926 * of setting the slave's hw address while it is
3927 * transmitting, but the assumption is that the base
3928 * driver can handle that.
3929 *
3930 * TODO: figure out a way to safely iterate the slaves
3931 * list, but without holding a lock around the actual
3932 * call to the base driver.
3933 */
3934
3935 bond_for_each_slave(bond, slave, i) {
eb7cc59a 3936 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
5a03cdb7 3937 pr_debug("slave %p %s\n", slave, slave->dev->name);
1da177e4 3938
eb7cc59a 3939 if (slave_ops->ndo_set_mac_address == NULL) {
1da177e4 3940 res = -EOPNOTSUPP;
5a03cdb7 3941 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
1da177e4
LT		 3942 goto unwind;
3943 }
3944
3945 res = dev_set_mac_address(slave->dev, addr);
3946 if (res) {
3947 /* TODO: consider downing the slave
3948 * and retry ?
3949 * User should expect communications
3950 * breakage anyway until ARP finish
3951 * updating, so...
3952 */
5a03cdb7 3953 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 3954 goto unwind;
3955 }
3956 }
3957
3958 /* success */
3959 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3960 return 0;
3961
3962unwind:
3963 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3964 tmp_sa.sa_family = bond_dev->type;
3965
3966 /* unwind from head to the slave that failed */
3967 stop_at = slave;
3968 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3969 int tmp_res;
3970
3971 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3972 if (tmp_res) {
a4aee5c8
JP		 3973 pr_debug("unwind err %d dev %s\n",
3974 tmp_res, slave->dev->name);
1da177e4
LT		 3975 }
3976 }
3977
3978 return res;
3979}
3980
3981static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3982{
454d7c9b 3983 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3984 struct slave *slave, *start_at;
cf5f9044 3985 int i, slave_no, res = 1;
a2fd940f 3986 struct iphdr *iph = ip_hdr(skb);
1da177e4 3987
cf5f9044 3988 /*
a2fd940f
AG		 3989 * Start with the curr_active_slave that joined the bond as the
3990 * default for sending IGMP traffic. For failover purposes one
3991 * needs to maintain some consistency for the interface that will
3992 * send the join/membership reports. The curr_active_slave found
3993 * will send all of this type of traffic.
cf5f9044 3994 */
00ae7028 3995 if ((iph->protocol == IPPROTO_IGMP) &&
a2fd940f 3996 (skb->protocol == htons(ETH_P_IP))) {
1da177e4 3997
a2fd940f
AG		 3998 read_lock(&bond->curr_slave_lock);
3999 slave = bond->curr_active_slave;
4000 read_unlock(&bond->curr_slave_lock);
4001
4002 if (!slave)
4003 goto out;
4004 } else {
4005 /*
4006 * Concurrent TX may collide on rr_tx_counter; we accept
4007 * that as being rare enough not to justify using an
4008 * atomic op here.
4009 */
4010 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4011
4012 bond_for_each_slave(bond, slave, i) {
4013 slave_no--;
4014 if (slave_no < 0)
4015 break;
4016 }
1da177e4
LT		 4017 }
4018
cf5f9044 4019 start_at = slave;
1da177e4
LT		 4020 bond_for_each_slave_from(bond, slave, i, start_at) {
4021 if (IS_UP(slave->dev) &&
4022 (slave->link == BOND_LINK_UP) &&
e30bc066 4023 bond_is_active_slave(slave)) {
1da177e4 4024 res = bond_dev_queue_xmit(bond, skb, slave->dev);
1da177e4
LT		 4025 break;
4026 }
4027 }
4028
1da177e4
LT		 4029out:
4030 if (res) {
4031 /* no suitable interface, frame not sent */
4032 dev_kfree_skb(skb);
4033 }
0693e88e 4034
ec634fe3 4035 return NETDEV_TX_OK;
1da177e4
LT		 4036}
4037
075897ce 4038
1da177e4
LT		 4039/*
4040 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4041 * the bond has a usable interface.
4042 */
4043static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4044{
454d7c9b 4045 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4046 int res = 1;
4047
1da177e4
LT		 4048 read_lock(&bond->curr_slave_lock);
4049
0693e88e
MM		 4050 if (bond->curr_active_slave)
4051 res = bond_dev_queue_xmit(bond, skb,
4052 bond->curr_active_slave->dev);
1da177e4 4053
3d632c3f 4054 if (res)
1da177e4
LT		 4055 /* no suitable interface, frame not sent */
4056 dev_kfree_skb(skb);
3d632c3f 4057
1da177e4 4058 read_unlock(&bond->curr_slave_lock);
0693e88e 4059
ec634fe3 4060 return NETDEV_TX_OK;
1da177e4
LT		 4061}
4062
4063/*
169a3e66
JV		 4064 * In bond_xmit_xor() , we determine the output device by using a pre-
4065 * determined xmit_hash_policy(), If the selected device is not enabled,
4066 * find the next active slave.
1da177e4
LT		 4067 */
4068static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4069{
454d7c9b 4070 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4071 struct slave *slave, *start_at;
4072 int slave_no;
4073 int i;
4074 int res = 1;
4075
a361c83c 4076 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
1da177e4
LT		 4077
4078 bond_for_each_slave(bond, slave, i) {
4079 slave_no--;
3d632c3f 4080 if (slave_no < 0)
1da177e4 4081 break;
1da177e4
LT		 4082 }
4083
4084 start_at = slave;
4085
4086 bond_for_each_slave_from(bond, slave, i, start_at) {
4087 if (IS_UP(slave->dev) &&
4088 (slave->link == BOND_LINK_UP) &&
e30bc066 4089 bond_is_active_slave(slave)) {
1da177e4
LT		 4090 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4091 break;
4092 }
4093 }
4094
1da177e4
LT		 4095 if (res) {
4096 /* no suitable interface, frame not sent */
4097 dev_kfree_skb(skb);
4098 }
0693e88e 4099
ec634fe3 4100 return NETDEV_TX_OK;
1da177e4
LT		 4101}
4102
4103/*
4104 * in broadcast mode, we send everything to all usable interfaces.
4105 */
4106static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4107{
454d7c9b 4108 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 4109 struct slave *slave, *start_at;
4110 struct net_device *tx_dev = NULL;
4111 int i;
4112 int res = 1;
4113
1da177e4
LT		 4114 read_lock(&bond->curr_slave_lock);
4115 start_at = bond->curr_active_slave;
4116 read_unlock(&bond->curr_slave_lock);
4117
3d632c3f 4118 if (!start_at)
1da177e4 4119 goto out;
1da177e4
LT		 4120
4121 bond_for_each_slave_from(bond, slave, i, start_at) {
4122 if (IS_UP(slave->dev) &&
4123 (slave->link == BOND_LINK_UP) &&
e30bc066 4124 bond_is_active_slave(slave)) {
1da177e4
LT		 4125 if (tx_dev) {
4126 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4127 if (!skb2) {
a4aee5c8 4128 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4e0952c7 4129 bond_dev->name);
1da177e4
LT		 4130 continue;
4131 }
4132
4133 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4134 if (res) {
4135 dev_kfree_skb(skb2);
4136 continue;
4137 }
4138 }
4139 tx_dev = slave->dev;
4140 }
4141 }
4142
3d632c3f 4143 if (tx_dev)
1da177e4 4144 res = bond_dev_queue_xmit(bond, skb, tx_dev);
1da177e4
LT		 4145
4146out:
3d632c3f 4147 if (res)
1da177e4
LT		 4148 /* no suitable interface, frame not sent */
4149 dev_kfree_skb(skb);
3d632c3f 4150
1da177e4 4151 /* frame sent to all suitable interfaces */
ec634fe3 4152 return NETDEV_TX_OK;
1da177e4
LT		 4153}
4154
4155/*------------------------- Device initialization ---------------------------*/
4156
6f6652be
JV		 4157static void bond_set_xmit_hash_policy(struct bonding *bond)
4158{
4159 switch (bond->params.xmit_policy) {
4160 case BOND_XMIT_POLICY_LAYER23:
4161 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4162 break;
4163 case BOND_XMIT_POLICY_LAYER34:
4164 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4165 break;
4166 case BOND_XMIT_POLICY_LAYER2:
4167 default:
4168 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4169 break;
4170 }
4171}
4172
bb1d9123
AG		 4173/*
4174 * Lookup the slave that corresponds to a qid
4175 */
4176static inline int bond_slave_override(struct bonding *bond,
4177 struct sk_buff *skb)
4178{
4179 int i, res = 1;
4180 struct slave *slave = NULL;
4181 struct slave *check_slave;
4182
0693e88e
MM		 4183 if (!skb->queue_mapping)
4184 return 1;
bb1d9123
AG		 4185
4186 /* Find out if any slaves have the same mapping as this skb. */
4187 bond_for_each_slave(bond, check_slave, i) {
4188 if (check_slave->queue_id == skb->queue_mapping) {
4189 slave = check_slave;
4190 break;
4191 }
4192 }
4193
4194 /* If the slave isn't UP, use default transmit policy. */
4195 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4196 (slave->link == BOND_LINK_UP)) {
4197 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4198 }
4199
bb1d9123
AG		 4200 return res;
4201}
4202
4203static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4204{
4205 /*
4206 * This helper function exists to help dev_pick_tx get the correct
fd0e435b 4207 * destination queue. Using a helper function skips a call to
bb1d9123
AG		 4208 * skb_tx_hash and will put the skbs in the queue we expect on their
4209 * way down to the bonding driver.
4210 */
fd0e435b
PO		 4211 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4212
4213 if (unlikely(txq >= dev->real_num_tx_queues)) {
d30ee670 4214 do {
fd0e435b 4215 txq -= dev->real_num_tx_queues;
d30ee670 4216 } while (txq >= dev->real_num_tx_queues);
fd0e435b
PO		 4217 }
4218 return txq;
bb1d9123
AG		 4219}
4220
0693e88e 4221static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
00829823 4222{
bb1d9123
AG		 4223 struct bonding *bond = netdev_priv(dev);
4224
4225 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4226 if (!bond_slave_override(bond, skb))
4227 return NETDEV_TX_OK;
4228 }
00829823
SH		 4229
4230 switch (bond->params.mode) {
4231 case BOND_MODE_ROUNDROBIN:
4232 return bond_xmit_roundrobin(skb, dev);
4233 case BOND_MODE_ACTIVEBACKUP:
4234 return bond_xmit_activebackup(skb, dev);
4235 case BOND_MODE_XOR:
4236 return bond_xmit_xor(skb, dev);
4237 case BOND_MODE_BROADCAST:
4238 return bond_xmit_broadcast(skb, dev);
4239 case BOND_MODE_8023AD:
4240 return bond_3ad_xmit_xor(skb, dev);
4241 case BOND_MODE_ALB:
4242 case BOND_MODE_TLB:
4243 return bond_alb_xmit(skb, dev);
4244 default:
4245 /* Should never happen, mode already checked */
a4aee5c8
JP		 4246 pr_err("%s: Error: Unknown bonding mode %d\n",
4247 dev->name, bond->params.mode);
00829823
SH		 4248 WARN_ON_ONCE(1);
4249 dev_kfree_skb(skb);
4250 return NETDEV_TX_OK;
4251 }
4252}
4253
0693e88e
MM		 4254static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4255{
4256 struct bonding *bond = netdev_priv(dev);
4257 netdev_tx_t ret = NETDEV_TX_OK;
4258
4259 /*
4260 * If we risk deadlock from transmitting this in the
4261 * netpoll path, tell netpoll to queue the frame for later tx
4262 */
4263 if (is_netpoll_tx_blocked(dev))
4264 return NETDEV_TX_BUSY;
4265
4266 read_lock(&bond->lock);
4267
4268 if (bond->slave_cnt)
4269 ret = __bond_start_xmit(skb, dev);
4270 else
4271 dev_kfree_skb(skb);
4272
4273 read_unlock(&bond->lock);
4274
4275 return ret;
4276}
00829823 4277
1da177e4
LT		 4278/*
4279 * set bond mode specific net device operations
4280 */
a77b5325 4281void bond_set_mode_ops(struct bonding *bond, int mode)
1da177e4 4282{
169a3e66
JV		 4283 struct net_device *bond_dev = bond->dev;
4284
1da177e4
LT		 4285 switch (mode) {
4286 case BOND_MODE_ROUNDROBIN:
1da177e4
LT		 4287 break;
4288 case BOND_MODE_ACTIVEBACKUP:
1da177e4
LT		 4289 break;
4290 case BOND_MODE_XOR:
6f6652be 4291 bond_set_xmit_hash_policy(bond);
1da177e4
LT		 4292 break;
4293 case BOND_MODE_BROADCAST:
1da177e4
LT		 4294 break;
4295 case BOND_MODE_8023AD:
6f6652be 4296 bond_set_xmit_hash_policy(bond);
1da177e4 4297 break;
1da177e4 4298 case BOND_MODE_ALB:
8f903c70
JV		 4299 /* FALLTHRU */
4300 case BOND_MODE_TLB:
1da177e4
LT		 4301 break;
4302 default:
4303 /* Should never happen, mode already checked */
a4aee5c8
JP		 4304 pr_err("%s: Error: Unknown bonding mode %d\n",
4305 bond_dev->name, mode);
1da177e4
LT		 4306 break;
4307 }
4308}
4309
217df670
JV		 4310static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4311 struct ethtool_drvinfo *drvinfo)
4312{
4313 strncpy(drvinfo->driver, DRV_NAME, 32);
4314 strncpy(drvinfo->version, DRV_VERSION, 32);
4315 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4316}
4317
7282d491 4318static const struct ethtool_ops bond_ethtool_ops = {
217df670 4319 .get_drvinfo = bond_ethtool_get_drvinfo,
fa53ebac 4320 .get_link = ethtool_op_get_link,
8531c5ff
AK		 4321};
4322
eb7cc59a 4323static const struct net_device_ops bond_netdev_ops = {
181470fc 4324 .ndo_init = bond_init,
9e71626c 4325 .ndo_uninit = bond_uninit,
eb7cc59a
SH		 4326 .ndo_open = bond_open,
4327 .ndo_stop = bond_close,
00829823 4328 .ndo_start_xmit = bond_start_xmit,
bb1d9123 4329 .ndo_select_queue = bond_select_queue,
be1f3c2c 4330 .ndo_get_stats64 = bond_get_stats,
eb7cc59a
SH		 4331 .ndo_do_ioctl = bond_do_ioctl,
4332 .ndo_set_multicast_list = bond_set_multicast_list,
4333 .ndo_change_mtu = bond_change_mtu,
eb7cc59a 4334 .ndo_set_mac_address = bond_set_mac_address,
00829823 4335 .ndo_neigh_setup = bond_neigh_setup,
eb7cc59a
SH		 4336 .ndo_vlan_rx_register = bond_vlan_rx_register,
4337 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4338 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
f6dc31a8 4339#ifdef CONFIG_NET_POLL_CONTROLLER
8a8efa22 4340 .ndo_netpoll_setup = bond_netpoll_setup,
f6dc31a8
WC		 4341 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4342 .ndo_poll_controller = bond_poll_controller,
4343#endif
9232ecca
JP		 4344 .ndo_add_slave = bond_enslave,
4345 .ndo_del_slave = bond_release,
b2a103e6 4346 .ndo_fix_features = bond_fix_features,
eb7cc59a
SH		 4347};
4348
9e2e61fb
AW		 4349static void bond_destructor(struct net_device *bond_dev)
4350{
4351 struct bonding *bond = netdev_priv(bond_dev);
4352 if (bond->wq)
4353 destroy_workqueue(bond->wq);
4354 free_netdev(bond_dev);
4355}
4356
181470fc 4357static void bond_setup(struct net_device *bond_dev)
1da177e4 4358{
454d7c9b 4359 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 4360
1da177e4
LT		 4361 /* initialize rwlocks */
4362 rwlock_init(&bond->lock);
4363 rwlock_init(&bond->curr_slave_lock);
4364
d2991f75 4365 bond->params = bonding_defaults;
1da177e4
LT		 4366
4367 /* Initialize pointers */
1da177e4
LT		 4368 bond->dev = bond_dev;
4369 INIT_LIST_HEAD(&bond->vlan_list);
4370
4371 /* Initialize the device entry points */
181470fc 4372 ether_setup(bond_dev);
eb7cc59a 4373 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 4374 bond_dev->ethtool_ops = &bond_ethtool_ops;
169a3e66 4375 bond_set_mode_ops(bond, bond->params.mode);
1da177e4 4376
9e2e61fb 4377 bond_dev->destructor = bond_destructor;
1da177e4
LT		 4378
4379 /* Initialize the device options */
4380 bond_dev->tx_queue_len = 0;
4381 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 4382 bond_dev->priv_flags |= IFF_BONDING;
181470fc
SH		 4383 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4384
1da177e4
LT		 4385 /* At first, we block adding VLANs. That's the only way to
4386 * prevent problems that occur when adding VLANs over an
4387 * empty bond. The block will be removed once non-challenged
4388 * slaves are enslaved.
4389 */
4390 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4391
932ff279 4392 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 4393 * transmitting */
4394 bond_dev->features |= NETIF_F_LLTX;
4395
4396 /* By default, we declare the bond to be fully
4397 * VLAN hardware accelerated capable. Special
4398 * care is taken in the various xmit functions
4399 * when there are slaves that are not hw accel
4400 * capable
4401 */
1da177e4 4402
b2a103e6
MM		 4403 bond_dev->hw_features = BOND_VLAN_FEATURES |
4404 NETIF_F_HW_VLAN_TX |
4405 NETIF_F_HW_VLAN_RX |
4406 NETIF_F_HW_VLAN_FILTER;
4407
4408 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4409 bond_dev->features |= bond_dev->hw_features;
1da177e4
LT		 4410}
4411
fdaea7a9
JV		 4412static void bond_work_cancel_all(struct bonding *bond)
4413{
4414 write_lock_bh(&bond->lock);
4415 bond->kill_timers = 1;
4416 write_unlock_bh(&bond->lock);
4417
4418 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4419 cancel_delayed_work(&bond->mii_work);
4420
4421 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4422 cancel_delayed_work(&bond->arp_work);
4423
4424 if (bond->params.mode == BOND_MODE_ALB &&
4425 delayed_work_pending(&bond->alb_work))
4426 cancel_delayed_work(&bond->alb_work);
4427
4428 if (bond->params.mode == BOND_MODE_8023AD &&
4429 delayed_work_pending(&bond->ad_work))
4430 cancel_delayed_work(&bond->ad_work);
5a37e8ca
FL		 4431
4432 if (delayed_work_pending(&bond->mcast_work))
4433 cancel_delayed_work(&bond->mcast_work);
fdaea7a9
JV		 4434}
4435
c67dfb29
EB		 4436/*
4437* Destroy a bonding device.
4438* Must be under rtnl_lock when this function is called.
4439*/
4440static void bond_uninit(struct net_device *bond_dev)
a434e43f 4441{
454d7c9b 4442 struct bonding *bond = netdev_priv(bond_dev);
f35188fa 4443 struct vlan_entry *vlan, *tmp;
a434e43f 4444
f6dc31a8
WC		 4445 bond_netpoll_cleanup(bond_dev);
4446
c67dfb29
EB		 4447 /* Release the bonded slaves */
4448 bond_release_all(bond_dev);
4449
a434e43f
JV		 4450 list_del(&bond->bond_list);
4451
4452 bond_work_cancel_all(bond);
4453
a434e43f 4454 bond_remove_proc_entry(bond);
c67dfb29 4455
f073c7ca
TI		 4456 bond_debug_unregister(bond);
4457
22bedad3 4458 __hw_addr_flush(&bond->mc_list);
f35188fa
JV		 4459
4460 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4461 list_del(&vlan->vlan_list);
4462 kfree(vlan);
4463 }
a434e43f
JV		 4464}
4465
1da177e4
LT		 4466/*------------------------- Module initialization ---------------------------*/
4467
4468/*
4469 * Convert string input module parms. Accept either the
ece95f7f
JV		 4470 * number of the mode or its string name. A bit complicated because
4471 * some mode names are substrings of other names, and calls from sysfs
4472 * may have whitespace in the name (trailing newlines, for example).
1da177e4 4473 */
325dcf7a 4474int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
1da177e4 4475{
54b87323 4476 int modeint = -1, i, rv;
a42e534f 4477 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
ece95f7f 4478
a42e534f
JV		 4479 for (p = (char *)buf; *p; p++)
4480 if (!(isdigit(*p) || isspace(*p)))
4481 break;
4482
4483 if (*p)
ece95f7f 4484 rv = sscanf(buf, "%20s", modestr);
a42e534f 4485 else
54b87323 4486 rv = sscanf(buf, "%d", &modeint);
a42e534f
JV		 4487
4488 if (!rv)
4489 return -1;
1da177e4
LT		 4490
4491 for (i = 0; tbl[i].modename; i++) {
54b87323 4492 if (modeint == tbl[i].mode)
ece95f7f
JV		 4493 return tbl[i].mode;
4494 if (strcmp(modestr, tbl[i].modename) == 0)
1da177e4 4495 return tbl[i].mode;
1da177e4
LT		 4496 }
4497
4498 return -1;
4499}
4500
4501static int bond_check_params(struct bond_params *params)
4502{
a549952a 4503 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
f5b2b966 4504
1da177e4
LT		 4505 /*
4506 * Convert string parameters.
4507 */
4508 if (mode) {
4509 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4510 if (bond_mode == -1) {
a4aee5c8 4511 pr_err("Error: Invalid bonding mode \"%s\"\n",
1da177e4
LT		 4512 mode == NULL ? "NULL" : mode);
4513 return -EINVAL;
4514 }
4515 }
4516
169a3e66
JV		 4517 if (xmit_hash_policy) {
4518 if ((bond_mode != BOND_MODE_XOR) &&
4519 (bond_mode != BOND_MODE_8023AD)) {
a4aee5c8 4520 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
169a3e66
JV		 4521 bond_mode_name(bond_mode));
4522 } else {
4523 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4524 xmit_hashtype_tbl);
4525 if (xmit_hashtype == -1) {
a4aee5c8 4526 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
3d632c3f 4527 xmit_hash_policy == NULL ? "NULL" :
169a3e66
JV		 4528 xmit_hash_policy);
4529 return -EINVAL;
4530 }
4531 }
4532 }
4533
1da177e4
LT		 4534 if (lacp_rate) {
4535 if (bond_mode != BOND_MODE_8023AD) {
a4aee5c8
JP		 4536 pr_info("lacp_rate param is irrelevant in mode %s\n",
4537 bond_mode_name(bond_mode));
1da177e4
LT		 4538 } else {
4539 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4540 if (lacp_fast == -1) {
a4aee5c8 4541 pr_err("Error: Invalid lacp rate \"%s\"\n",
1da177e4
LT		 4542 lacp_rate == NULL ? "NULL" : lacp_rate);
4543 return -EINVAL;
4544 }
4545 }
4546 }
4547
fd989c83
JV		 4548 if (ad_select) {
4549 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4550 if (params->ad_select == -1) {
a4aee5c8 4551 pr_err("Error: Invalid ad_select \"%s\"\n",
fd989c83
JV		 4552 ad_select == NULL ? "NULL" : ad_select);
4553 return -EINVAL;
4554 }
4555
4556 if (bond_mode != BOND_MODE_8023AD) {
a4aee5c8 4557 pr_warning("ad_select param only affects 802.3ad mode\n");
fd989c83
JV		 4558 }
4559 } else {
4560 params->ad_select = BOND_AD_STABLE;
4561 }
4562
f5841306 4563 if (max_bonds < 0) {
a4aee5c8
JP		 4564 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4565 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4566 max_bonds = BOND_DEFAULT_MAX_BONDS;
4567 }
4568
4569 if (miimon < 0) {
a4aee5c8
JP		 4570 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4571 miimon, INT_MAX, BOND_LINK_MON_INTERV);
1da177e4
LT		 4572 miimon = BOND_LINK_MON_INTERV;
4573 }
4574
4575 if (updelay < 0) {
a4aee5c8
JP		 4576 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4577 updelay, INT_MAX);
1da177e4
LT		 4578 updelay = 0;
4579 }
4580
4581 if (downdelay < 0) {
a4aee5c8
JP		 4582 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4583 downdelay, INT_MAX);
1da177e4
LT		 4584 downdelay = 0;
4585 }
4586
4587 if ((use_carrier != 0) && (use_carrier != 1)) {
a4aee5c8
JP		 4588 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4589 use_carrier);
1da177e4
LT		 4590 use_carrier = 1;
4591 }
4592
ad246c99
BH		 4593 if (num_peer_notif < 0 || num_peer_notif > 255) {
4594 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4595 num_peer_notif);
4596 num_peer_notif = 1;
4597 }
4598
1da177e4
LT		 4599 /* reset values for 802.3ad */
4600 if (bond_mode == BOND_MODE_8023AD) {
4601 if (!miimon) {
a4aee5c8 4602 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");
3d632c3f 4603 pr_warning("Forcing miimon to 100msec\n");
1da177e4
LT		 4604 miimon = 100;
4605 }
4606 }
4607
bb1d9123
AG		 4608 if (tx_queues < 1 || tx_queues > 255) {
4609 pr_warning("Warning: tx_queues (%d) should be between "
4610 "1 and 255, resetting to %d\n",
4611 tx_queues, BOND_DEFAULT_TX_QUEUES);
4612 tx_queues = BOND_DEFAULT_TX_QUEUES;
4613 }
4614
ebd8e497
AG		 4615 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4616 pr_warning("Warning: all_slaves_active module parameter (%d), "
4617 "not of valid value (0/1), so it was set to "
4618 "0\n", all_slaves_active);
4619 all_slaves_active = 0;
4620 }
4621
c2952c31
FL		 4622 if (resend_igmp < 0 || resend_igmp > 255) {
4623 pr_warning("Warning: resend_igmp (%d) should be between "
4624 "0 and 255, resetting to %d\n",
4625 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4626 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4627 }
4628
1da177e4
LT		 4629 /* reset values for TLB/ALB */
4630 if ((bond_mode == BOND_MODE_TLB) ||
4631 (bond_mode == BOND_MODE_ALB)) {
4632 if (!miimon) {
a4aee5c8 4633 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");
3d632c3f 4634 pr_warning("Forcing miimon to 100msec\n");
1da177e4
LT		 4635 miimon = 100;
4636 }
4637 }
4638
4639 if (bond_mode == BOND_MODE_ALB) {
a4aee5c8
JP		 4640 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",
4641 updelay);
1da177e4
LT		 4642 }
4643
4644 if (!miimon) {
4645 if (updelay || downdelay) {
4646 /* just warn the user the up/down delay will have
4647 * no effect since miimon is zero...
4648 */
a4aee5c8
JP		 4649 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",
4650 updelay, downdelay);
1da177e4
LT		 4651 }
4652 } else {
4653 /* don't allow arp monitoring */
4654 if (arp_interval) {
a4aee5c8
JP		 4655 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4656 miimon, arp_interval);
1da177e4
LT		 4657 arp_interval = 0;
4658 }
4659
4660 if ((updelay % miimon) != 0) {
a4aee5c8
JP		 4661 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4662 updelay, miimon,
4663 (updelay / miimon) * miimon);
1da177e4
LT		 4664 }
4665
4666 updelay /= miimon;
4667
4668 if ((downdelay % miimon) != 0) {
a4aee5c8
JP		 4669 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4670 downdelay, miimon,
4671 (downdelay / miimon) * miimon);
1da177e4
LT		 4672 }
4673
4674 downdelay /= miimon;
4675 }
4676
4677 if (arp_interval < 0) {
a4aee5c8
JP		 4678 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4679 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
1da177e4
LT		 4680 arp_interval = BOND_LINK_ARP_INTERV;
4681 }
4682
4683 for (arp_ip_count = 0;
4684 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4685 arp_ip_count++) {
4686 /* not complete check, but should be good enough to
4687 catch mistakes */
4688 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
a4aee5c8
JP		 4689 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4690 arp_ip_target[arp_ip_count]);
1da177e4
LT		 4691 arp_interval = 0;
4692 } else {
d3bb52b0 4693 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
1da177e4
LT		 4694 arp_target[arp_ip_count] = ip;
4695 }
4696 }
4697
4698 if (arp_interval && !arp_ip_count) {
4699 /* don't allow arping if no arp_ip_target given... */
a4aee5c8
JP		 4700 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4701 arp_interval);
1da177e4
LT		 4702 arp_interval = 0;
4703 }
4704
f5b2b966
JV		 4705 if (arp_validate) {
4706 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
a4aee5c8 4707 pr_err("arp_validate only supported in active-backup mode\n");
f5b2b966
JV		 4708 return -EINVAL;
4709 }
4710 if (!arp_interval) {
a4aee5c8 4711 pr_err("arp_validate requires arp_interval\n");
f5b2b966
JV		 4712 return -EINVAL;
4713 }
4714
4715 arp_validate_value = bond_parse_parm(arp_validate,
4716 arp_validate_tbl);
4717 if (arp_validate_value == -1) {
a4aee5c8 4718 pr_err("Error: invalid arp_validate \"%s\"\n",
f5b2b966
JV		 4719 arp_validate == NULL ? "NULL" : arp_validate);
4720 return -EINVAL;
4721 }
4722 } else
4723 arp_validate_value = 0;
4724
1da177e4 4725 if (miimon) {
a4aee5c8 4726 pr_info("MII link monitoring set to %d ms\n", miimon);
1da177e4
LT		 4727 } else if (arp_interval) {
4728 int i;
4729
a4aee5c8
JP		 4730 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4731 arp_interval,
4732 arp_validate_tbl[arp_validate_value].modename,
4733 arp_ip_count);
1da177e4
LT		 4734
4735 for (i = 0; i < arp_ip_count; i++)
e5e2a8fd 4736 pr_info(" %s", arp_ip_target[i]);
1da177e4 4737
e5e2a8fd 4738 pr_info("\n");
1da177e4 4739
b8a9787e 4740 } else if (max_bonds) {
1da177e4
LT		 4741 /* miimon and arp_interval not set, we need one so things
4742 * work as expected, see bonding.txt for details
4743 */
a4aee5c8 4744 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");
1da177e4
LT		 4745 }
4746
4747 if (primary && !USES_PRIMARY(bond_mode)) {
4748 /* currently, using a primary only makes sense
4749 * in active backup, TLB or ALB modes
4750 */
a4aee5c8
JP		 4751 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4752 primary, bond_mode_name(bond_mode));
1da177e4
LT		 4753 primary = NULL;
4754 }
4755
a549952a
JP		 4756 if (primary && primary_reselect) {
4757 primary_reselect_value = bond_parse_parm(primary_reselect,
4758 pri_reselect_tbl);
4759 if (primary_reselect_value == -1) {
a4aee5c8 4760 pr_err("Error: Invalid primary_reselect \"%s\"\n",
a549952a
JP		 4761 primary_reselect ==
4762 NULL ? "NULL" : primary_reselect);
4763 return -EINVAL;
4764 }
4765 } else {
4766 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4767 }
4768
3915c1e8
JV		 4769 if (fail_over_mac) {
4770 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4771 fail_over_mac_tbl);
4772 if (fail_over_mac_value == -1) {
a4aee5c8 4773 pr_err("Error: invalid fail_over_mac \"%s\"\n",
3915c1e8
JV		 4774 arp_validate == NULL ? "NULL" : arp_validate);
4775 return -EINVAL;
4776 }
4777
4778 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
a4aee5c8 4779 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
3915c1e8
JV		 4780 } else {
4781 fail_over_mac_value = BOND_FOM_NONE;
4782 }
dd957c57 4783
1da177e4
LT		 4784 /* fill params struct with the proper values */
4785 params->mode = bond_mode;
169a3e66 4786 params->xmit_policy = xmit_hashtype;
1da177e4 4787 params->miimon = miimon;
ad246c99 4788 params->num_peer_notif = num_peer_notif;
1da177e4 4789 params->arp_interval = arp_interval;
f5b2b966 4790 params->arp_validate = arp_validate_value;
1da177e4
LT		 4791 params->updelay = updelay;
4792 params->downdelay = downdelay;
4793 params->use_carrier = use_carrier;
4794 params->lacp_fast = lacp_fast;
4795 params->primary[0] = 0;
a549952a 4796 params->primary_reselect = primary_reselect_value;
3915c1e8 4797 params->fail_over_mac = fail_over_mac_value;
bb1d9123 4798 params->tx_queues = tx_queues;
ebd8e497 4799 params->all_slaves_active = all_slaves_active;
c2952c31 4800 params->resend_igmp = resend_igmp;
1da177e4
LT		 4801
4802 if (primary) {
4803 strncpy(params->primary, primary, IFNAMSIZ);
4804 params->primary[IFNAMSIZ - 1] = 0;
4805 }
4806
4807 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4808
4809 return 0;
4810}
4811
0daa2303 4812static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 4813static struct lock_class_key bonding_netdev_addr_lock_key;
0daa2303 4814
e8a0464c
DM		 4815static void bond_set_lockdep_class_one(struct net_device *dev,
4816 struct netdev_queue *txq,
4817 void *_unused)
c773e847
DM		 4818{
4819 lockdep_set_class(&txq->_xmit_lock,
4820 &bonding_netdev_xmit_lock_key);
4821}
4822
4823static void bond_set_lockdep_class(struct net_device *dev)
4824{
cf508b12
DM		 4825 lockdep_set_class(&dev->addr_list_lock,
4826 &bonding_netdev_addr_lock_key);
e8a0464c 4827 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
c773e847
DM		 4828}
4829
181470fc
SH		 4830/*
4831 * Called from registration process
4832 */
4833static int bond_init(struct net_device *bond_dev)
4834{
4835 struct bonding *bond = netdev_priv(bond_dev);
ec87fd3b 4836 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
181470fc
SH		 4837
4838 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4839
4840 bond->wq = create_singlethread_workqueue(bond_dev->name);
4841 if (!bond->wq)
4842 return -ENOMEM;
4843
4844 bond_set_lockdep_class(bond_dev);
4845
181470fc 4846 bond_create_proc_entry(bond);
ec87fd3b 4847 list_add_tail(&bond->bond_list, &bn->dev_list);
181470fc 4848
6151b3d4 4849 bond_prepare_sysfs_group(bond);
22bedad3 4850
f073c7ca
TI		 4851 bond_debug_register(bond);
4852
22bedad3 4853 __hw_addr_init(&bond->mc_list);
181470fc
SH		 4854 return 0;
4855}
4856
88ead977
EB		 4857static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4858{
4859 if (tb[IFLA_ADDRESS]) {
4860 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4861 return -EINVAL;
4862 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4863 return -EADDRNOTAVAIL;
4864 }
4865 return 0;
4866}
4867
4868static struct rtnl_link_ops bond_link_ops __read_mostly = {
4869 .kind = "bond",
6639104b 4870 .priv_size = sizeof(struct bonding),
88ead977
EB		 4871 .setup = bond_setup,
4872 .validate = bond_validate,
4873};
4874
dfe60397 4875/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 4876 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 4877 * Caller must NOT hold rtnl_lock; we need to release it here before we
4878 * set up our sysfs entries.
4879 */
ec87fd3b 4880int bond_create(struct net *net, const char *name)
dfe60397
MW		 4881{
4882 struct net_device *bond_dev;
4883 int res;
4884
4885 rtnl_lock();
027ea041 4886
1c5cae81
JP		 4887 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4888 name ? name : "bond%d",
4889 bond_setup, tx_queues);
dfe60397 4890 if (!bond_dev) {
a4aee5c8 4891 pr_err("%s: eek! can't alloc netdev!\n", name);
9e2e61fb
AW		 4892 rtnl_unlock();
4893 return -ENOMEM;
dfe60397
MW		 4894 }
4895
ec87fd3b 4896 dev_net_set(bond_dev, net);
88ead977
EB		 4897 bond_dev->rtnl_link_ops = &bond_link_ops;
4898
dfe60397 4899 res = register_netdevice(bond_dev);
0daa2303 4900
e826eafa
PO		 4901 netif_carrier_off(bond_dev);
4902
7e083840 4903 rtnl_unlock();
9e2e61fb
AW		 4904 if (res < 0)
4905 bond_destructor(bond_dev);
30c15ba9 4906 return res;
dfe60397
MW		 4907}
4908
2c8c1e72 4909static int __net_init bond_net_init(struct net *net)
ec87fd3b 4910{
15449745 4911 struct bond_net *bn = net_generic(net, bond_net_id);
ec87fd3b
EB		 4912
4913 bn->net = net;
4914 INIT_LIST_HEAD(&bn->dev_list);
4915
ec87fd3b 4916 bond_create_proc_dir(bn);
15449745
EB		 4917
4918 return 0;
ec87fd3b
EB		 4919}
4920
2c8c1e72 4921static void __net_exit bond_net_exit(struct net *net)
ec87fd3b 4922{
15449745 4923 struct bond_net *bn = net_generic(net, bond_net_id);
ec87fd3b
EB		 4924
4925 bond_destroy_proc_dir(bn);
ec87fd3b
EB		 4926}
4927
4928static struct pernet_operations bond_net_ops = {
4929 .init = bond_net_init,
4930 .exit = bond_net_exit,
15449745
EB		 4931 .id = &bond_net_id,
4932 .size = sizeof(struct bond_net),
ec87fd3b
EB		 4933};
4934
1da177e4
LT		 4935static int __init bonding_init(void)
4936{
1da177e4
LT		 4937 int i;
4938 int res;
4939
bd33acc3 4940 pr_info("%s", bond_version);
1da177e4 4941
dfe60397 4942 res = bond_check_params(&bonding_defaults);
3d632c3f 4943 if (res)
dfe60397 4944 goto out;
1da177e4 4945
15449745 4946 res = register_pernet_subsys(&bond_net_ops);
ec87fd3b
EB		 4947 if (res)
4948 goto out;
027ea041 4949
88ead977
EB		 4950 res = rtnl_link_register(&bond_link_ops);
4951 if (res)
6639104b 4952 goto err_link;
88ead977 4953
f073c7ca
TI		 4954 bond_create_debugfs();
4955
1da177e4 4956 for (i = 0; i < max_bonds; i++) {
ec87fd3b 4957 res = bond_create(&init_net, NULL);
dfe60397
MW		 4958 if (res)
4959 goto err;
1da177e4
LT		 4960 }
4961
b76cdba9
MW		 4962 res = bond_create_sysfs();
4963 if (res)
4964 goto err;
4965
1da177e4 4966 register_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 4967 register_inetaddr_notifier(&bond_inetaddr_notifier);
dfe60397 4968out:
1da177e4 4969 return res;
88ead977
EB		 4970err:
4971 rtnl_link_unregister(&bond_link_ops);
6639104b 4972err_link:
15449745 4973 unregister_pernet_subsys(&bond_net_ops);
88ead977 4974 goto out;
dfe60397 4975
1da177e4
LT		 4976}
4977
4978static void __exit bonding_exit(void)
4979{
4980 unregister_netdevice_notifier(&bond_netdev_notifier);
c3ade5ca 4981 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
1da177e4 4982
ae68c398 4983 bond_destroy_sysfs();
f073c7ca 4984 bond_destroy_debugfs();
ae68c398 4985
88ead977 4986 rtnl_link_unregister(&bond_link_ops);
15449745 4987 unregister_pernet_subsys(&bond_net_ops);
e843fa50
NH		 4988
4989#ifdef CONFIG_NET_POLL_CONTROLLER
fb4fa76a
NH		 4990 /*
4991 * Make sure we don't have an imbalance on our netpoll blocking
4992 */
4993 WARN_ON(atomic_read(&netpoll_block_tx));
e843fa50 4994#endif
1da177e4
LT		 4995}
4996
4997module_init(bonding_init);
4998module_exit(bonding_exit);
4999MODULE_LICENSE("GPL");
5000MODULE_VERSION(DRV_VERSION);
5001MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5002MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
88ead977 5003MODULE_ALIAS_RTNL_LINK("bond");
Linux kernel - Deliverables
This page took 1.854459 seconds and 5 git commands to generate.