projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
vlan: kill vlan_put_tag helper
[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
1da177e4
LT		 34#include <linux/kernel.h>
35#include <linux/module.h>
1da177e4
LT		 36#include <linux/types.h>
37#include <linux/fcntl.h>
38#include <linux/interrupt.h>
39#include <linux/ptrace.h>
40#include <linux/ioport.h>
41#include <linux/in.h>
169a3e66 42#include <net/ip.h>
1da177e4 43#include <linux/ip.h>
169a3e66
JV		 44#include <linux/tcp.h>
45#include <linux/udp.h>
1da177e4
LT		 46#include <linux/slab.h>
47#include <linux/string.h>
48#include <linux/init.h>
49#include <linux/timer.h>
50#include <linux/socket.h>
51#include <linux/ctype.h>
52#include <linux/inet.h>
53#include <linux/bitops.h>
3d632c3f 54#include <linux/io.h>
1da177e4 55#include <asm/dma.h>
3d632c3f 56#include <linux/uaccess.h>
1da177e4
LT		 57#include <linux/errno.h>
58#include <linux/netdevice.h>
59#include <linux/inetdevice.h>
a816c7c7 60#include <linux/igmp.h>
1da177e4
LT		 61#include <linux/etherdevice.h>
62#include <linux/skbuff.h>
63#include <net/sock.h>
64#include <linux/rtnetlink.h>
1da177e4
LT		 65#include <linux/smp.h>
66#include <linux/if_ether.h>
67#include <net/arp.h>
68#include <linux/mii.h>
69#include <linux/ethtool.h>
70#include <linux/if_vlan.h>
71#include <linux/if_bonding.h>
b63bb739 72#include <linux/jiffies.h>
e843fa50 73#include <linux/preempt.h>
c3ade5ca 74#include <net/route.h>
457c4cbc 75#include <net/net_namespace.h>
ec87fd3b 76#include <net/netns/generic.h>
5ee31c68 77#include <net/pkt_sched.h>
278b2083 78#include <linux/rculist.h>
32819dc1 79#include <net/flow_keys.h>
1ef8019b
DM		 80#include <net/bonding.h>
81#include <net/bond_3ad.h>
82#include <net/bond_alb.h>
1da177e4
LT		 83
84/*---------------------------- Module parameters ----------------------------*/
85
86/* monitor all links that often (in milliseconds). <=0 disables monitoring */
1da177e4
LT		 87
88static int max_bonds = BOND_DEFAULT_MAX_BONDS;
bb1d9123 89static int tx_queues = BOND_DEFAULT_TX_QUEUES;
ad246c99 90static int num_peer_notif = 1;
b98d9c66 91static int miimon;
3d632c3f
SH		 92static int updelay;
93static int downdelay;
1da177e4 94static int use_carrier = 1;
3d632c3f
SH		 95static char *mode;
96static char *primary;
a549952a 97static char *primary_reselect;
3d632c3f 98static char *lacp_rate;
655f8919 99static int min_links;
3d632c3f
SH		 100static char *ad_select;
101static char *xmit_hash_policy;
7bdb04ed 102static int arp_interval;
3d632c3f
SH		 103static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
104static char *arp_validate;
8599b52e 105static char *arp_all_targets;
3d632c3f 106static char *fail_over_mac;
b07ea07b 107static int all_slaves_active;
d2991f75 108static struct bond_params bonding_defaults;
c2952c31 109static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
73958329 110static int packets_per_slave = 1;
3a7129e5 111static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
1da177e4
LT		 112
113module_param(max_bonds, int, 0);
114MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
bb1d9123
AG		 115module_param(tx_queues, int, 0);
116MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
ad246c99 117module_param_named(num_grat_arp, num_peer_notif, int, 0644);
90e62474
AG		 118MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
119 "failover event (alias of num_unsol_na)");
ad246c99 120module_param_named(num_unsol_na, num_peer_notif, int, 0644);
90e62474
AG		 121MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
122 "failover event (alias of num_grat_arp)");
1da177e4
LT		 123module_param(miimon, int, 0);
124MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
125module_param(updelay, int, 0);
126MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
127module_param(downdelay, int, 0);
2ac47660
MW		 128MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
129 "in milliseconds");
1da177e4 130module_param(use_carrier, int, 0);
2ac47660
MW		 131MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132 "0 for off, 1 for on (default)");
1da177e4 133module_param(mode, charp, 0);
90e62474 134MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
2ac47660
MW		 135 "1 for active-backup, 2 for balance-xor, "
136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137 "6 for balance-alb");
1da177e4
LT		 138module_param(primary, charp, 0);
139MODULE_PARM_DESC(primary, "Primary network device to use");
a549952a
JP		 140module_param(primary_reselect, charp, 0);
141MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
142 "once it comes up; "
143 "0 for always (default), "
144 "1 for only if speed of primary is "
145 "better, "
146 "2 for only on active slave "
147 "failure");
1da177e4 148module_param(lacp_rate, charp, 0);
90e62474
AG		 149MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
150 "0 for slow, 1 for fast");
fd989c83 151module_param(ad_select, charp, 0);
90e62474
AG		 152MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
153 "0 for stable (default), 1 for bandwidth, "
154 "2 for count");
655f8919 155module_param(min_links, int, 0);
156MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
157
169a3e66 158module_param(xmit_hash_policy, charp, 0);
90e62474
AG		 159MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
160 "0 for layer 2 (default), 1 for layer 3+4, "
32819dc1
NA		 161 "2 for layer 2+3, 3 for encap layer 2+3, "
162 "4 for encap layer 3+4");
1da177e4
LT		 163module_param(arp_interval, int, 0);
164MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
165module_param_array(arp_ip_target, charp, NULL, 0);
166MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
f5b2b966 167module_param(arp_validate, charp, 0);
90e62474
AG		 168MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
169 "0 for none (default), 1 for active, "
170 "2 for backup, 3 for all");
8599b52e
VF		 171module_param(arp_all_targets, charp, 0);
172MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
3915c1e8 173module_param(fail_over_mac, charp, 0);
90e62474
AG		 174MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
175 "the same MAC; 0 for none (default), "
176 "1 for active, 2 for follow");
ebd8e497 177module_param(all_slaves_active, int, 0);
37b7021d 178MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
90e62474 179 "by setting active flag for all slaves; "
ebd8e497 180 "0 for never (default), 1 for always.");
c2952c31 181module_param(resend_igmp, int, 0);
90e62474
AG		 182MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
183 "link failure");
73958329
NA		 184module_param(packets_per_slave, int, 0);
185MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
186 "mode; 0 for a random slave, 1 packet per "
187 "slave (default), >1 packets per slave.");
3a7129e5 188module_param(lp_interval, uint, 0);
189MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
190 "the bonding driver sends learning packets to "
191 "each slaves peer switch. The default is 1.");
1da177e4
LT		 192
193/*----------------------------- Global variables ----------------------------*/
194
e843fa50 195#ifdef CONFIG_NET_POLL_CONTROLLER
fb4fa76a 196atomic_t netpoll_block_tx = ATOMIC_INIT(0);
e843fa50
NH		 197#endif
198
f99189b1 199int bond_net_id __read_mostly;
1da177e4 200
3d632c3f
SH		 201static __be32 arp_target[BOND_MAX_ARP_TARGETS];
202static int arp_ip_count;
1da177e4 203static int bond_mode = BOND_MODE_ROUNDROBIN;
3d632c3f
SH		 204static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
205static int lacp_fast;
1da177e4 206
1da177e4
LT		 207/*-------------------------- Forward declarations ---------------------------*/
208
181470fc 209static int bond_init(struct net_device *bond_dev);
c67dfb29 210static void bond_uninit(struct net_device *bond_dev);
5f0c5f73
AG		 211static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
212 struct rtnl_link_stats64 *stats);
ee637714 213static void bond_slave_arr_handler(struct work_struct *work);
1da177e4
LT		 214
215/*---------------------------- General routines -----------------------------*/
216
bd33acc3 217const char *bond_mode_name(int mode)
1da177e4 218{
77afc92b
HE		 219 static const char *names[] = {
220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
3d632c3f 224 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
77afc92b
HE		 225 [BOND_MODE_TLB] = "transmit load balancing",
226 [BOND_MODE_ALB] = "adaptive load balancing",
227 };
228
f5280948 229 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
1da177e4 230 return "unknown";
77afc92b
HE		 231
232 return names[mode];
1da177e4
LT		 233}
234
235/*---------------------------------- VLAN -----------------------------------*/
236
1da177e4
LT		 237/**
238 * bond_dev_queue_xmit - Prepare skb for xmit.
3d632c3f 239 *
1da177e4
LT		 240 * @bond: bond device that got this skb for tx.
241 * @skb: hw accel VLAN tagged skb to transmit
242 * @slave_dev: slave that is supposed to xmit this skbuff
1da177e4 243 */
d316dedd 244void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
3d632c3f 245 struct net_device *slave_dev)
1da177e4 246{
8387451e 247 skb->dev = slave_dev;
374eeb5a 248
5ee31c68 249 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
df4ab5b3
JP		 250 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
251 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
374eeb5a 252
e15c3c22 253 if (unlikely(netpoll_tx_running(bond->dev)))
8a8efa22 254 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
080e4130 255 else
f6dc31a8 256 dev_queue_xmit(skb);
1da177e4
LT		 257}
258
547942ca 259/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
cc0e4070 260 * We don't protect the slave list iteration with a lock because:
1da177e4
LT		 261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
3d632c3f 265 *
1da177e4
LT		 266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
273*/
274
1da177e4
LT		 275/**
276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
279 */
80d5c368
PM		 280static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
281 __be16 proto, u16 vid)
1da177e4 282{
454d7c9b 283 struct bonding *bond = netdev_priv(bond_dev);
81f23b13 284 struct slave *slave, *rollback_slave;
9caff1e7 285 struct list_head *iter;
dec1e90e 286 int res;
1da177e4 287
9caff1e7 288 bond_for_each_slave(bond, slave, iter) {
80d5c368 289 res = vlan_vid_add(slave->dev, proto, vid);
87002b03
JP		 290 if (res)
291 goto unwind;
1da177e4
LT		 292 }
293
8e586137 294 return 0;
87002b03
JP		 295
296unwind:
81f23b13 297 /* unwind to the slave that failed */
9caff1e7 298 bond_for_each_slave(bond, rollback_slave, iter) {
81f23b13
VF		 299 if (rollback_slave == slave)
300 break;
301
302 vlan_vid_del(rollback_slave->dev, proto, vid);
303 }
87002b03
JP		 304
305 return res;
1da177e4
LT		 306}
307
308/**
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
312 */
80d5c368
PM		 313static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
314 __be16 proto, u16 vid)
1da177e4 315{
454d7c9b 316 struct bonding *bond = netdev_priv(bond_dev);
9caff1e7 317 struct list_head *iter;
1da177e4 318 struct slave *slave;
1da177e4 319
9caff1e7 320 bond_for_each_slave(bond, slave, iter)
80d5c368 321 vlan_vid_del(slave->dev, proto, vid);
1da177e4 322
e868b0c9
VF		 323 if (bond_is_lb(bond))
324 bond_alb_clear_vlan(bond, vid);
8e586137
JP		 325
326 return 0;
1da177e4
LT		 327}
328
1da177e4
LT		 329/*------------------------------- Link status -------------------------------*/
330
547942ca 331/* Set the carrier state for the master according to the state of its
ff59c456
JV		 332 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
333 * do special 802.3ad magic.
334 *
335 * Returns zero if carrier state does not change, nonzero if it does.
336 */
337static int bond_set_carrier(struct bonding *bond)
338{
9caff1e7 339 struct list_head *iter;
ff59c456 340 struct slave *slave;
ff59c456 341
0965a1f3 342 if (!bond_has_slaves(bond))
ff59c456
JV		 343 goto down;
344
01844098 345 if (BOND_MODE(bond) == BOND_MODE_8023AD)
ff59c456
JV		 346 return bond_3ad_set_carrier(bond);
347
9caff1e7 348 bond_for_each_slave(bond, slave, iter) {
ff59c456
JV		 349 if (slave->link == BOND_LINK_UP) {
350 if (!netif_carrier_ok(bond->dev)) {
351 netif_carrier_on(bond->dev);
352 return 1;
353 }
354 return 0;
355 }
356 }
357
358down:
359 if (netif_carrier_ok(bond->dev)) {
360 netif_carrier_off(bond->dev);
361 return 1;
362 }
363 return 0;
364}
365
547942ca 366/* Get link speed and duplex from the slave's base driver
1da177e4 367 * using ethtool. If for some reason the call fails or the
98f41f69 368 * values are invalid, set speed and duplex to -1,
13b95fb7 369 * and return.
1da177e4 370 */
13b95fb7 371static void bond_update_speed_duplex(struct slave *slave)
1da177e4
LT		 372{
373 struct net_device *slave_dev = slave->dev;
4bc71cb9 374 struct ethtool_cmd ecmd;
5d30530e 375 u32 slave_speed;
61a44b9c 376 int res;
1da177e4 377
589665f5
DC		 378 slave->speed = SPEED_UNKNOWN;
379 slave->duplex = DUPLEX_UNKNOWN;
1da177e4 380
4bc71cb9 381 res = __ethtool_get_settings(slave_dev, &ecmd);
61a44b9c 382 if (res < 0)
13b95fb7 383 return;
1da177e4 384
4bc71cb9 385 slave_speed = ethtool_cmd_speed(&ecmd);
6f92c66f 386 if (slave_speed == 0 || slave_speed == ((__u32) -1))
13b95fb7 387 return;
1da177e4 388
4bc71cb9 389 switch (ecmd.duplex) {
1da177e4
LT		 390 case DUPLEX_FULL:
391 case DUPLEX_HALF:
392 break;
393 default:
13b95fb7 394 return;
1da177e4
LT		 395 }
396
5d30530e 397 slave->speed = slave_speed;
4bc71cb9 398 slave->duplex = ecmd.duplex;
1da177e4 399
13b95fb7 400 return;
1da177e4
LT		 401}
402
07699f9a 403const char *bond_slave_link_status(s8 link)
404{
405 switch (link) {
406 case BOND_LINK_UP:
407 return "up";
408 case BOND_LINK_FAIL:
409 return "going down";
410 case BOND_LINK_DOWN:
411 return "down";
412 case BOND_LINK_BACK:
413 return "going back";
414 default:
415 return "unknown";
416 }
417}
418
547942ca 419/* if <dev> supports MII link status reporting, check its link status.
1da177e4
LT		 420 *
421 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
3d632c3f 422 * depending upon the setting of the use_carrier parameter.
1da177e4
LT		 423 *
424 * Return either BMSR_LSTATUS, meaning that the link is up (or we
425 * can't tell and just pretend it is), or 0, meaning that the link is
426 * down.
427 *
428 * If reporting is non-zero, instead of faking link up, return -1 if
429 * both ETHTOOL and MII ioctls fail (meaning the device does not
430 * support them). If use_carrier is set, return whatever it says.
431 * It'd be nice if there was a good way to tell if a driver supports
432 * netif_carrier, but there really isn't.
433 */
3d632c3f
SH		 434static int bond_check_dev_link(struct bonding *bond,
435 struct net_device *slave_dev, int reporting)
1da177e4 436{
eb7cc59a 437 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
d9d52832 438 int (*ioctl)(struct net_device *, struct ifreq *, int);
1da177e4
LT		 439 struct ifreq ifr;
440 struct mii_ioctl_data *mii;
1da177e4 441
6c988853
PG		 442 if (!reporting && !netif_running(slave_dev))
443 return 0;
444
eb7cc59a 445 if (bond->params.use_carrier)
1da177e4 446 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1da177e4 447
29112f4e 448 /* Try to get link status using Ethtool first. */
c772dde3
BH		 449 if (slave_dev->ethtool_ops->get_link)
450 return slave_dev->ethtool_ops->get_link(slave_dev) ?
451 BMSR_LSTATUS : 0;
29112f4e 452
3d632c3f 453 /* Ethtool can't be used, fallback to MII ioctls. */
eb7cc59a 454 ioctl = slave_ops->ndo_do_ioctl;
1da177e4 455 if (ioctl) {
547942ca
NA		 456 /* TODO: set pointer to correct ioctl on a per team member
457 * bases to make this more efficient. that is, once
458 * we determine the correct ioctl, we will always
459 * call it and not the others for that team
460 * member.
461 */
462
463 /* We cannot assume that SIOCGMIIPHY will also read a
1da177e4
LT		 464 * register; not all network drivers (e.g., e100)
465 * support that.
466 */
467
468 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
469 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
470 mii = if_mii(&ifr);
471 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
472 mii->reg_num = MII_BMSR;
3d632c3f
SH		 473 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
474 return mii->val_out & BMSR_LSTATUS;
1da177e4
LT		 475 }
476 }
477
547942ca 478 /* If reporting, report that either there's no dev->do_ioctl,
61a44b9c 479 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 480 * cannot report link status). If not reporting, pretend
481 * we're ok.
482 */
3d632c3f 483 return reporting ? -1 : BMSR_LSTATUS;
1da177e4
LT		 484}
485
486/*----------------------------- Multicast list ------------------------------*/
487
547942ca 488/* Push the promiscuity flag down to appropriate slaves */
7e1a1ac1 489static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 490{
9caff1e7 491 struct list_head *iter;
7e1a1ac1 492 int err = 0;
9caff1e7 493
ec0865a9 494 if (bond_uses_primary(bond)) {
14056e79 495 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
4740d638 496
4740d638
ED		 497 if (curr_active)
498 err = dev_set_promiscuity(curr_active->dev, inc);
1da177e4
LT		 499 } else {
500 struct slave *slave;
dec1e90e 501
9caff1e7 502 bond_for_each_slave(bond, slave, iter) {
7e1a1ac1
WC		 503 err = dev_set_promiscuity(slave->dev, inc);
504 if (err)
505 return err;
1da177e4
LT		 506 }
507 }
7e1a1ac1 508 return err;
1da177e4
LT		 509}
510
547942ca 511/* Push the allmulti flag down to all slaves */
7e1a1ac1 512static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 513{
9caff1e7 514 struct list_head *iter;
7e1a1ac1 515 int err = 0;
9caff1e7 516
ec0865a9 517 if (bond_uses_primary(bond)) {
14056e79 518 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
4740d638 519
4740d638
ED		 520 if (curr_active)
521 err = dev_set_allmulti(curr_active->dev, inc);
1da177e4
LT		 522 } else {
523 struct slave *slave;
dec1e90e 524
9caff1e7 525 bond_for_each_slave(bond, slave, iter) {
7e1a1ac1
WC		 526 err = dev_set_allmulti(slave->dev, inc);
527 if (err)
528 return err;
1da177e4
LT		 529 }
530 }
7e1a1ac1 531 return err;
1da177e4
LT		 532}
533
547942ca 534/* Retrieve the list of registered multicast addresses for the bonding
5a37e8ca
FL		 535 * device and retransmit an IGMP JOIN request to the current active
536 * slave.
537 */
f2369109 538static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
5a37e8ca 539{
f2369109 540 struct bonding *bond = container_of(work, struct bonding,
541 mcast_work.work);
542
4aa5dee4 543 if (!rtnl_trylock()) {
4beac029 544 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
4aa5dee4 545 return;
5a37e8ca 546 }
4aa5dee4 547 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
5a37e8ca 548
4f5474e7
NA		 549 if (bond->igmp_retrans > 1) {
550 bond->igmp_retrans--;
c2952c31 551 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
4f5474e7 552 }
f2369109 553 rtnl_unlock();
5a37e8ca
FL		 554}
555
547942ca 556/* Flush bond's hardware addresses from slave */
303d1cbf 557static void bond_hw_addr_flush(struct net_device *bond_dev,
3d632c3f 558 struct net_device *slave_dev)
1da177e4 559{
454d7c9b 560 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 561
303d1cbf
JV		 562 dev_uc_unsync(slave_dev, bond_dev);
563 dev_mc_unsync(slave_dev, bond_dev);
1da177e4 564
01844098 565 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1da177e4
LT		 566 /* del lacpdu mc addr from mc list */
567 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
568
22bedad3 569 dev_mc_del(slave_dev, lacpdu_multicast);
1da177e4
LT		 570 }
571}
572
573/*--------------------------- Active slave change ---------------------------*/
574
303d1cbf 575/* Update the hardware address list and promisc/allmulti for the new and
ec0865a9
VF		 576 * old active slaves (if any). Modes that are not using primary keep all
577 * slaves up date at all times; only the modes that use primary need to call
303d1cbf 578 * this function to swap these settings during a failover.
1da177e4 579 */
303d1cbf
JV		 580static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
581 struct slave *old_active)
1da177e4 582{
1da177e4 583 if (old_active) {
3d632c3f 584 if (bond->dev->flags & IFF_PROMISC)
1da177e4 585 dev_set_promiscuity(old_active->dev, -1);
1da177e4 586
3d632c3f 587 if (bond->dev->flags & IFF_ALLMULTI)
1da177e4 588 dev_set_allmulti(old_active->dev, -1);
1da177e4 589
303d1cbf 590 bond_hw_addr_flush(bond->dev, old_active->dev);
1da177e4
LT		 591 }
592
593 if (new_active) {
7e1a1ac1 594 /* FIXME: Signal errors upstream. */
3d632c3f 595 if (bond->dev->flags & IFF_PROMISC)
1da177e4 596 dev_set_promiscuity(new_active->dev, 1);
1da177e4 597
3d632c3f 598 if (bond->dev->flags & IFF_ALLMULTI)
1da177e4 599 dev_set_allmulti(new_active->dev, 1);
1da177e4 600
d632ce98 601 netif_addr_lock_bh(bond->dev);
303d1cbf
JV		 602 dev_uc_sync(new_active->dev, bond->dev);
603 dev_mc_sync(new_active->dev, bond->dev);
d632ce98 604 netif_addr_unlock_bh(bond->dev);
1da177e4
LT		 605 }
606}
607
ae0d6750 608/**
609 * bond_set_dev_addr - clone slave's address to bond
610 * @bond_dev: bond net device
611 * @slave_dev: slave net device
612 *
613 * Should be called with RTNL held.
614 */
615static void bond_set_dev_addr(struct net_device *bond_dev,
616 struct net_device *slave_dev)
617{
76444f50
VF		 618 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
619 bond_dev, slave_dev, slave_dev->addr_len);
ae0d6750 620 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
621 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
622 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
623}
624
547942ca 625/* bond_do_fail_over_mac
3915c1e8
JV		 626 *
627 * Perform special MAC address swapping for fail_over_mac settings
628 *
1c72cfdc 629 * Called with RTNL
3915c1e8
JV		 630 */
631static void bond_do_fail_over_mac(struct bonding *bond,
632 struct slave *new_active,
633 struct slave *old_active)
634{
635 u8 tmp_mac[ETH_ALEN];
636 struct sockaddr saddr;
637 int rv;
638
639 switch (bond->params.fail_over_mac) {
640 case BOND_FOM_ACTIVE:
1c72cfdc 641 if (new_active)
ae0d6750 642 bond_set_dev_addr(bond->dev, new_active->dev);
3915c1e8
JV		 643 break;
644 case BOND_FOM_FOLLOW:
547942ca 645 /* if new_active && old_active, swap them
3915c1e8
JV		 646 * if just old_active, do nothing (going to no active slave)
647 * if just new_active, set new_active to bond's MAC
648 */
649 if (!new_active)
650 return;
651
3915c1e8 652 if (old_active) {
ada0f863 653 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
2a7c183b
JP		 654 ether_addr_copy(saddr.sa_data,
655 old_active->dev->dev_addr);
3915c1e8
JV		 656 saddr.sa_family = new_active->dev->type;
657 } else {
ada0f863 658 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
3915c1e8
JV		 659 saddr.sa_family = bond->dev->type;
660 }
661
662 rv = dev_set_mac_address(new_active->dev, &saddr);
663 if (rv) {
76444f50
VF		 664 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
665 -rv, new_active->dev->name);
3915c1e8
JV		 666 goto out;
667 }
668
669 if (!old_active)
670 goto out;
671
ada0f863 672 ether_addr_copy(saddr.sa_data, tmp_mac);
3915c1e8
JV		 673 saddr.sa_family = old_active->dev->type;
674
675 rv = dev_set_mac_address(old_active->dev, &saddr);
676 if (rv)
76444f50
VF		 677 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
678 -rv, new_active->dev->name);
3915c1e8 679out:
3915c1e8
JV		 680 break;
681 default:
76444f50
VF		 682 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
683 bond->params.fail_over_mac);
3915c1e8
JV		 684 break;
685 }
686
687}
688
a549952a
JP		 689static bool bond_should_change_active(struct bonding *bond)
690{
059b47e8 691 struct slave *prim = rtnl_dereference(bond->primary_slave);
1c72cfdc 692 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
a549952a
JP		 693
694 if (!prim || !curr || curr->link != BOND_LINK_UP)
695 return true;
696 if (bond->force_primary) {
697 bond->force_primary = false;
698 return true;
699 }
700 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
701 (prim->speed < curr->speed ||
702 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
703 return false;
704 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
705 return false;
706 return true;
707}
3915c1e8 708
1da177e4
LT		 709/**
710 * find_best_interface - select the best available slave to be the active one
711 * @bond: our bonding struct
1da177e4
LT		 712 */
713static struct slave *bond_find_best_slave(struct bonding *bond)
714{
059b47e8 715 struct slave *slave, *bestslave = NULL, *primary;
77140d29 716 struct list_head *iter;
1da177e4 717 int mintime = bond->params.updelay;
1da177e4 718
059b47e8
NA		 719 primary = rtnl_dereference(bond->primary_slave);
720 if (primary && primary->link == BOND_LINK_UP &&
77140d29 721 bond_should_change_active(bond))
059b47e8 722 return primary;
1da177e4 723
77140d29
VF		 724 bond_for_each_slave(bond, slave, iter) {
725 if (slave->link == BOND_LINK_UP)
726 return slave;
b6adc610 727 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
77140d29
VF		 728 slave->delay < mintime) {
729 mintime = slave->delay;
730 bestslave = slave;
1da177e4
LT		 731 }
732 }
733
734 return bestslave;
735}
736
ad246c99
BH		 737static bool bond_should_notify_peers(struct bonding *bond)
738{
4cb4f97b 739 struct slave *slave;
740
741 rcu_read_lock();
742 slave = rcu_dereference(bond->curr_active_slave);
743 rcu_read_unlock();
ad246c99 744
76444f50
VF		 745 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
746 slave ? slave->dev->name : "NULL");
ad246c99
BH		 747
748 if (!slave || !bond->send_peer_notif ||
749 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
750 return false;
751
ad246c99
BH		 752 return true;
753}
754
1da177e4
LT		 755/**
756 * change_active_interface - change the active slave into the specified one
757 * @bond: our bonding struct
758 * @new: the new slave to make the active one
759 *
760 * Set the new slave to the bond's settings and unset them on the old
761 * curr_active_slave.
762 * Setting include flags, mc-list, promiscuity, allmulti, etc.
763 *
764 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
765 * because it is apparently the best available slave we have, even though its
766 * updelay hasn't timed out yet.
767 *
1c72cfdc 768 * Caller must hold RTNL.
1da177e4 769 */
a77b5325 770void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4 771{
4740d638
ED		 772 struct slave *old_active;
773
1c72cfdc
NA		 774 ASSERT_RTNL();
775
776 old_active = rtnl_dereference(bond->curr_active_slave);
1da177e4 777
3d632c3f 778 if (old_active == new_active)
1da177e4 779 return;
1da177e4
LT		 780
781 if (new_active) {
8e603460 782 new_active->last_link_up = jiffies;
b2220cad 783
1da177e4 784 if (new_active->link == BOND_LINK_BACK) {
ec0865a9 785 if (bond_uses_primary(bond)) {
76444f50
VF		 786 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
787 new_active->dev->name,
788 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1da177e4
LT		 789 }
790
791 new_active->delay = 0;
792 new_active->link = BOND_LINK_UP;
1da177e4 793
01844098 794 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1da177e4 795 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1da177e4 796
58402054 797 if (bond_is_lb(bond))
1da177e4 798 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1da177e4 799 } else {
ec0865a9 800 if (bond_uses_primary(bond)) {
76444f50
VF		 801 netdev_info(bond->dev, "making interface %s the new active one\n",
802 new_active->dev->name);
1da177e4
LT		 803 }
804 }
805 }
806
ec0865a9 807 if (bond_uses_primary(bond))
303d1cbf 808 bond_hw_addr_swap(bond, new_active, old_active);
1da177e4 809
58402054 810 if (bond_is_lb(bond)) {
1da177e4 811 bond_alb_handle_active_change(bond, new_active);
8f903c70 812 if (old_active)
5e5b0665 813 bond_set_slave_inactive_flags(old_active,
814 BOND_SLAVE_NOTIFY_NOW);
8f903c70 815 if (new_active)
5e5b0665 816 bond_set_slave_active_flags(new_active,
817 BOND_SLAVE_NOTIFY_NOW);
1da177e4 818 } else {
278b2083 819 rcu_assign_pointer(bond->curr_active_slave, new_active);
1da177e4 820 }
c3ade5ca 821
01844098 822 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
3d632c3f 823 if (old_active)
5e5b0665 824 bond_set_slave_inactive_flags(old_active,
825 BOND_SLAVE_NOTIFY_NOW);
c3ade5ca
JV		 826
827 if (new_active) {
ad246c99
BH		 828 bool should_notify_peers = false;
829
5e5b0665 830 bond_set_slave_active_flags(new_active,
831 BOND_SLAVE_NOTIFY_NOW);
2ab82852 832
709f8a45
OG		 833 if (bond->params.fail_over_mac)
834 bond_do_fail_over_mac(bond, new_active,
835 old_active);
3915c1e8 836
ad246c99
BH		 837 if (netif_running(bond->dev)) {
838 bond->send_peer_notif =
839 bond->params.num_peer_notif;
840 should_notify_peers =
841 bond_should_notify_peers(bond);
842 }
843
b7bc2a5b 844 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
ad246c99 845 if (should_notify_peers)
b7bc2a5b
AW		 846 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
847 bond->dev);
7893b249 848 }
c3ade5ca 849 }
a2fd940f 850
5a37e8ca 851 /* resend IGMP joins since active slave has changed or
94265cf5
FL		 852 * all were sent on curr_active_slave.
853 * resend only if bond is brought up with the affected
547942ca
NA		 854 * bonding modes and the retransmission is enabled
855 */
94265cf5 856 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
ec0865a9 857 ((bond_uses_primary(bond) && new_active) ||
01844098 858 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
c2952c31 859 bond->igmp_retrans = bond->params.resend_igmp;
4beac029 860 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
a2fd940f 861 }
1da177e4
LT		 862}
863
864/**
865 * bond_select_active_slave - select a new active slave, if needed
866 * @bond: our bonding struct
867 *
3d632c3f 868 * This functions should be called when one of the following occurs:
1da177e4
LT		 869 * - The old curr_active_slave has been released or lost its link.
870 * - The primary_slave has got its link back.
871 * - A slave has got its link back and there's no old curr_active_slave.
872 *
1c72cfdc 873 * Caller must hold RTNL.
1da177e4 874 */
a77b5325 875void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 876{
877 struct slave *best_slave;
ff59c456 878 int rv;
1da177e4 879
e0974585
NA		 880 ASSERT_RTNL();
881
1da177e4 882 best_slave = bond_find_best_slave(bond);
1c72cfdc 883 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1da177e4 884 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 885 rv = bond_set_carrier(bond);
886 if (!rv)
887 return;
888
889 if (netif_carrier_ok(bond->dev)) {
76444f50 890 netdev_info(bond->dev, "first active interface up!\n");
ff59c456 891 } else {
76444f50 892 netdev_info(bond->dev, "now running without any active interface!\n");
ff59c456 893 }
1da177e4
LT		 894 }
895}
896
f6dc31a8 897#ifdef CONFIG_NET_POLL_CONTROLLER
8a8efa22 898static inline int slave_enable_netpoll(struct slave *slave)
f6dc31a8 899{
8a8efa22
AW		 900 struct netpoll *np;
901 int err = 0;
f6dc31a8 902
a8779ec1 903 np = kzalloc(sizeof(*np), GFP_KERNEL);
8a8efa22
AW		 904 err = -ENOMEM;
905 if (!np)
906 goto out;
907
a8779ec1 908 err = __netpoll_setup(np, slave->dev);
8a8efa22
AW		 909 if (err) {
910 kfree(np);
911 goto out;
f6dc31a8 912 }
8a8efa22
AW		 913 slave->np = np;
914out:
915 return err;
916}
917static inline void slave_disable_netpoll(struct slave *slave)
918{
919 struct netpoll *np = slave->np;
920
921 if (!np)
922 return;
923
924 slave->np = NULL;
2cde6acd 925 __netpoll_free_async(np);
8a8efa22 926}
f6dc31a8
WC		 927
928static void bond_poll_controller(struct net_device *bond_dev)
929{
8a8efa22
AW		 930}
931
c4cdef9b 932static void bond_netpoll_cleanup(struct net_device *bond_dev)
8a8efa22 933{
c4cdef9b 934 struct bonding *bond = netdev_priv(bond_dev);
9caff1e7 935 struct list_head *iter;
c2355e1a 936 struct slave *slave;
c2355e1a 937
9caff1e7 938 bond_for_each_slave(bond, slave, iter)
b6adc610 939 if (bond_slave_is_up(slave))
8a8efa22 940 slave_disable_netpoll(slave);
f6dc31a8 941}
8a8efa22 942
a8779ec1 943static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
8a8efa22
AW		 944{
945 struct bonding *bond = netdev_priv(dev);
9caff1e7 946 struct list_head *iter;
f6dc31a8 947 struct slave *slave;
dec1e90e 948 int err = 0;
f6dc31a8 949
9caff1e7 950 bond_for_each_slave(bond, slave, iter) {
8a8efa22
AW		 951 err = slave_enable_netpoll(slave);
952 if (err) {
c4cdef9b 953 bond_netpoll_cleanup(dev);
8a8efa22 954 break;
f6dc31a8
WC		 955 }
956 }
8a8efa22 957 return err;
f6dc31a8 958}
8a8efa22
AW		 959#else
960static inline int slave_enable_netpoll(struct slave *slave)
961{
962 return 0;
963}
964static inline void slave_disable_netpoll(struct slave *slave)
965{
966}
f6dc31a8
WC		 967static void bond_netpoll_cleanup(struct net_device *bond_dev)
968{
969}
f6dc31a8
WC		 970#endif
971
1da177e4
LT		 972/*---------------------------------- IOCTL ----------------------------------*/
973
c8f44aff 974static netdev_features_t bond_fix_features(struct net_device *dev,
9b7b165a 975 netdev_features_t features)
8531c5ff 976{
b2a103e6 977 struct bonding *bond = netdev_priv(dev);
9caff1e7 978 struct list_head *iter;
c8f44aff 979 netdev_features_t mask;
9b7b165a 980 struct slave *slave;
b63365a2 981
b2a103e6 982 mask = features;
b63365a2 983 features &= ~NETIF_F_ONE_FOR_ALL;
b2a103e6 984 features |= NETIF_F_ALL_FOR_ALL;
7f353bf2 985
9caff1e7 986 bond_for_each_slave(bond, slave, iter) {
b63365a2
HX		 987 features = netdev_increment_features(features,
988 slave->dev->features,
b2a103e6
MM		 989 mask);
990 }
b0ce3508 991 features = netdev_add_tso_features(features, mask);
b2a103e6 992
b2a103e6
MM		 993 return features;
994}
995
62f2a3a4
MM		 996#define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
997 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
998 NETIF_F_HIGHDMA | NETIF_F_LRO)
b2a103e6 999
5a7baa78
OG		 1000#define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1001 NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL)
1002
b2a103e6
MM		 1003static void bond_compute_features(struct bonding *bond)
1004{
02875878
ED		 1005 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1006 IFF_XMIT_DST_RELEASE_PERM;
c8f44aff 1007 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
5a7baa78 1008 netdev_features_t enc_features = BOND_ENC_FEATURES;
9caff1e7
VF		 1009 struct net_device *bond_dev = bond->dev;
1010 struct list_head *iter;
1011 struct slave *slave;
b2a103e6 1012 unsigned short max_hard_header_len = ETH_HLEN;
0e376bd0
SB		 1013 unsigned int gso_max_size = GSO_MAX_SIZE;
1014 u16 gso_max_segs = GSO_MAX_SEGS;
b2a103e6 1015
0965a1f3 1016 if (!bond_has_slaves(bond))
b2a103e6 1017 goto done;
a9b3ace4 1018 vlan_features &= NETIF_F_ALL_FOR_ALL;
b2a103e6 1019
9caff1e7 1020 bond_for_each_slave(bond, slave, iter) {
278339a4 1021 vlan_features = netdev_increment_features(vlan_features,
b2a103e6
MM		 1022 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1023
5a7baa78
OG		 1024 enc_features = netdev_increment_features(enc_features,
1025 slave->dev->hw_enc_features,
1026 BOND_ENC_FEATURES);
b6fe83e9 1027 dst_release_flag &= slave->dev->priv_flags;
54ef3137
JV		 1028 if (slave->dev->hard_header_len > max_hard_header_len)
1029 max_hard_header_len = slave->dev->hard_header_len;
0e376bd0
SB		 1030
1031 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1032 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
54ef3137 1033 }
8531c5ff 1034
b63365a2 1035done:
b2a103e6 1036 bond_dev->vlan_features = vlan_features;
5a7baa78 1037 bond_dev->hw_enc_features = enc_features;
54ef3137 1038 bond_dev->hard_header_len = max_hard_header_len;
0e376bd0
SB		 1039 bond_dev->gso_max_segs = gso_max_segs;
1040 netif_set_gso_max_size(bond_dev, gso_max_size);
8531c5ff 1041
02875878
ED		 1042 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1043 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1044 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1045 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
b6fe83e9 1046
b2a103e6 1047 netdev_change_features(bond_dev);
8531c5ff
AK		 1048}
1049
872254dd
MS		 1050static void bond_setup_by_slave(struct net_device *bond_dev,
1051 struct net_device *slave_dev)
1052{
00829823 1053 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1054
1055 bond_dev->type = slave_dev->type;
1056 bond_dev->hard_header_len = slave_dev->hard_header_len;
1057 bond_dev->addr_len = slave_dev->addr_len;
1058
1059 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1060 slave_dev->addr_len);
1061}
1062
5b2c4dd2 1063/* On bonding slaves other than the currently active slave, suppress
3aba891d 1064 * duplicates except for alb non-mcast/bcast.
5b2c4dd2
JP		 1065 */
1066static bool bond_should_deliver_exact_match(struct sk_buff *skb,
0bd80dad
JP		 1067 struct slave *slave,
1068 struct bonding *bond)
5b2c4dd2 1069{
2d7011ca 1070 if (bond_is_slave_inactive(slave)) {
01844098 1071 if (BOND_MODE(bond) == BOND_MODE_ALB &&
5b2c4dd2
JP		 1072 skb->pkt_type != PACKET_BROADCAST &&
1073 skb->pkt_type != PACKET_MULTICAST)
5b2c4dd2 1074 return false;
5b2c4dd2
JP		 1075 return true;
1076 }
1077 return false;
1078}
1079
8a4eb573 1080static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
5b2c4dd2 1081{
8a4eb573 1082 struct sk_buff *skb = *pskb;
f1c1775a 1083 struct slave *slave;
0bd80dad 1084 struct bonding *bond;
de063b70
ED		 1085 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1086 struct slave *);
13a8e0c8 1087 int ret = RX_HANDLER_ANOTHER;
5b2c4dd2 1088
8a4eb573
JP		 1089 skb = skb_share_check(skb, GFP_ATOMIC);
1090 if (unlikely(!skb))
1091 return RX_HANDLER_CONSUMED;
1092
1093 *pskb = skb;
5b2c4dd2 1094
35d48903
JP		 1095 slave = bond_slave_get_rcu(skb->dev);
1096 bond = slave->bond;
0bd80dad 1097
4d97480b
MH		 1098 recv_probe = ACCESS_ONCE(bond->recv_probe);
1099 if (recv_probe) {
de063b70
ED		 1100 ret = recv_probe(skb, bond, slave);
1101 if (ret == RX_HANDLER_CONSUMED) {
1102 consume_skb(skb);
1103 return ret;
3aba891d
JP		 1104 }
1105 }
1106
0bd80dad 1107 if (bond_should_deliver_exact_match(skb, slave, bond)) {
8a4eb573 1108 return RX_HANDLER_EXACT;
5b2c4dd2
JP		 1109 }
1110
35d48903 1111 skb->dev = bond->dev;
5b2c4dd2 1112
01844098 1113 if (BOND_MODE(bond) == BOND_MODE_ALB &&
35d48903 1114 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
5b2c4dd2 1115 skb->pkt_type == PACKET_HOST) {
5b2c4dd2 1116
541ac7c9
CG		 1117 if (unlikely(skb_cow_head(skb,
1118 skb->data - skb_mac_header(skb)))) {
1119 kfree_skb(skb);
8a4eb573 1120 return RX_HANDLER_CONSUMED;
541ac7c9 1121 }
2a7c183b 1122 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
5b2c4dd2
JP		 1123 }
1124
13a8e0c8 1125 return ret;
5b2c4dd2
JP		 1126}
1127
471cb5a3 1128static int bond_master_upper_dev_link(struct net_device *bond_dev,
1f718f0f
VF		 1129 struct net_device *slave_dev,
1130 struct slave *slave)
471cb5a3
JP		 1131{
1132 int err;
1133
1f718f0f 1134 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
471cb5a3
JP		 1135 if (err)
1136 return err;
1137 slave_dev->flags |= IFF_SLAVE;
7f294054 1138 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
471cb5a3
JP		 1139 return 0;
1140}
1141
1142static void bond_upper_dev_unlink(struct net_device *bond_dev,
1143 struct net_device *slave_dev)
1144{
1145 netdev_upper_dev_unlink(slave_dev, bond_dev);
1146 slave_dev->flags &= ~IFF_SLAVE;
7f294054 1147 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
471cb5a3
JP		 1148}
1149
3fdddd85 1150static struct slave *bond_alloc_slave(struct bonding *bond)
1151{
1152 struct slave *slave = NULL;
1153
1154 slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1155 if (!slave)
1156 return NULL;
1157
01844098 1158 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3fdddd85 1159 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1160 GFP_KERNEL);
1161 if (!SLAVE_AD_INFO(slave)) {
1162 kfree(slave);
1163 return NULL;
1164 }
1165 }
1166 return slave;
1167}
1168
1169static void bond_free_slave(struct slave *slave)
1170{
1171 struct bonding *bond = bond_get_bond_by_slave(slave);
1172
01844098 1173 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3fdddd85 1174 kfree(SLAVE_AD_INFO(slave));
1175
1176 kfree(slave);
1177}
1178
1da177e4 1179/* enslave device <slave> to bond device <master> */
a77b5325 1180int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1181{
454d7c9b 1182 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1183 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
c8c23903 1184 struct slave *new_slave = NULL, *prev_slave;
1da177e4
LT		 1185 struct sockaddr addr;
1186 int link_reporting;
8599b52e 1187 int res = 0, i;
1da177e4 1188
c772dde3
BH		 1189 if (!bond->params.use_carrier &&
1190 slave_dev->ethtool_ops->get_link == NULL &&
1191 slave_ops->ndo_do_ioctl == NULL) {
76444f50
VF		 1192 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1193 slave_dev->name);
1da177e4
LT		 1194 }
1195
1da177e4
LT		 1196 /* already enslaved */
1197 if (slave_dev->flags & IFF_SLAVE) {
76444f50 1198 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1da177e4
LT		 1199 return -EBUSY;
1200 }
1201
09a89c21 1202 if (bond_dev == slave_dev) {
76444f50 1203 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
09a89c21
JB		 1204 return -EPERM;
1205 }
1206
1da177e4
LT		 1207 /* vlan challenged mutual exclusion */
1208 /* no need to lock since we're protected by rtnl_lock */
1209 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
76444f50
VF		 1210 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1211 slave_dev->name);
55462cf3 1212 if (vlan_uses_dev(bond_dev)) {
76444f50
VF		 1213 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1214 slave_dev->name, bond_dev->name);
1da177e4
LT		 1215 return -EPERM;
1216 } else {
76444f50
VF		 1217 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1218 slave_dev->name, slave_dev->name,
1219 bond_dev->name);
1da177e4
LT		 1220 }
1221 } else {
76444f50
VF		 1222 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1223 slave_dev->name);
1da177e4
LT		 1224 }
1225
547942ca 1226 /* Old ifenslave binaries are no longer supported. These can
3d632c3f 1227 * be identified with moderate accuracy by the state of the slave:
217df670
JV		 1228 * the current ifenslave will set the interface down prior to
1229 * enslaving it; the old ifenslave will not.
1230 */
1231 if ((slave_dev->flags & IFF_UP)) {
76444f50
VF		 1232 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1233 slave_dev->name);
217df670
JV		 1234 res = -EPERM;
1235 goto err_undo_flags;
1236 }
1da177e4 1237
872254dd
MS		 1238 /* set bonding device ether type by slave - bonding netdevices are
1239 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1240 * there is a need to override some of the type dependent attribs/funcs.
1241 *
1242 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1243 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1244 */
0965a1f3 1245 if (!bond_has_slaves(bond)) {
e36b9d16 1246 if (bond_dev->type != slave_dev->type) {
76444f50
VF		 1247 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1248 bond_dev->type, slave_dev->type);
75c78500 1249
b7bc2a5b
AW		 1250 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1251 bond_dev);
3ca5b404
JP		 1252 res = notifier_to_errno(res);
1253 if (res) {
76444f50 1254 netdev_err(bond_dev, "refused to change device type\n");
3ca5b404
JP		 1255 res = -EBUSY;
1256 goto err_undo_flags;
1257 }
75c78500 1258
32a806c1 1259 /* Flush unicast and multicast addresses */
a748ee24 1260 dev_uc_flush(bond_dev);
22bedad3 1261 dev_mc_flush(bond_dev);
32a806c1 1262
e36b9d16
MS		 1263 if (slave_dev->type != ARPHRD_ETHER)
1264 bond_setup_by_slave(bond_dev, slave_dev);
550fd08c 1265 else {
e36b9d16 1266 ether_setup(bond_dev);
550fd08c
NH		 1267 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1268 }
75c78500 1269
b7bc2a5b
AW		 1270 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1271 bond_dev);
e36b9d16 1272 }
872254dd 1273 } else if (bond_dev->type != slave_dev->type) {
76444f50
VF		 1274 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1275 slave_dev->name, slave_dev->type, bond_dev->type);
a4aee5c8
JP		 1276 res = -EINVAL;
1277 goto err_undo_flags;
872254dd
MS		 1278 }
1279
eb7cc59a 1280 if (slave_ops->ndo_set_mac_address == NULL) {
76444f50 1281 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
f5442441
VF		 1282 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1283 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1284 if (!bond_has_slaves(bond)) {
00503b6f 1285 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
76444f50 1286 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
f5442441 1287 } else {
76444f50 1288 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
f5442441
VF		 1289 res = -EOPNOTSUPP;
1290 goto err_undo_flags;
00503b6f 1291 }
2ab82852 1292 }
1da177e4
LT		 1293 }
1294
8d8fc29d
AW		 1295 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1296
c20811a7 1297 /* If this is the first slave, then we need to set the master's hardware
547942ca
NA		 1298 * address to be the same as the slave's.
1299 */
0965a1f3 1300 if (!bond_has_slaves(bond) &&
dec1e90e 1301 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
409cc1f8 1302 bond_set_dev_addr(bond->dev, slave_dev);
c20811a7 1303
3fdddd85 1304 new_slave = bond_alloc_slave(bond);
1da177e4
LT		 1305 if (!new_slave) {
1306 res = -ENOMEM;
1307 goto err_undo_flags;
1308 }
3fdddd85 1309
dc73c41f
VF		 1310 new_slave->bond = bond;
1311 new_slave->dev = slave_dev;
547942ca 1312 /* Set the new_slave's queue_id to be zero. Queue ID mapping
bb1d9123
AG		 1313 * is set via sysfs or module option if desired.
1314 */
1315 new_slave->queue_id = 0;
1316
b15ba0fb
JP		 1317 /* Save slave's original mtu and then set it to match the bond */
1318 new_slave->original_mtu = slave_dev->mtu;
1319 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1320 if (res) {
76444f50 1321 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
b15ba0fb
JP		 1322 goto err_free;
1323 }
1324
547942ca 1325 /* Save slave's original ("permanent") mac address for modes
217df670
JV		 1326 * that need it, and for restoring it upon release, and then
1327 * set it to the master's address
1328 */
ada0f863 1329 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1da177e4 1330
00503b6f 1331 if (!bond->params.fail_over_mac ||
01844098 1332 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
547942ca 1333 /* Set slave to master's mac address. The application already
2ab82852
MS		 1334 * set the master's mac address to that of the first slave
1335 */
1336 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1337 addr.sa_family = slave_dev->type;
1338 res = dev_set_mac_address(slave_dev, &addr);
1339 if (res) {
76444f50 1340 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
b15ba0fb 1341 goto err_restore_mtu;
2ab82852 1342 }
217df670 1343 }
1da177e4 1344
217df670
JV		 1345 /* open the slave since the application closed it */
1346 res = dev_open(slave_dev);
1347 if (res) {
76444f50 1348 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1f718f0f 1349 goto err_restore_mac;
1da177e4
LT		 1350 }
1351
0b680e75 1352 slave_dev->priv_flags |= IFF_BONDING;
5f0c5f73
AG		 1353 /* initialize slave stats */
1354 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1da177e4 1355
58402054 1356 if (bond_is_lb(bond)) {
1da177e4
LT		 1357 /* bond_alb_init_slave() must be called before all other stages since
1358 * it might fail and we do not want to have to undo everything
1359 */
1360 res = bond_alb_init_slave(bond, new_slave);
3d632c3f 1361 if (res)
569f0c4d 1362 goto err_close;
1da177e4
LT		 1363 }
1364
ec0865a9 1365 /* If the mode uses primary, then the following is handled by
303d1cbf 1366 * bond_change_active_slave().
1da177e4 1367 */
ec0865a9 1368 if (!bond_uses_primary(bond)) {
1da177e4
LT		 1369 /* set promiscuity level to new slave */
1370 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1371 res = dev_set_promiscuity(slave_dev, 1);
1372 if (res)
1373 goto err_close;
1da177e4
LT		 1374 }
1375
1376 /* set allmulti level to new slave */
1377 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1378 res = dev_set_allmulti(slave_dev, 1);
1379 if (res)
1380 goto err_close;
1da177e4
LT		 1381 }
1382
b9e40857 1383 netif_addr_lock_bh(bond_dev);
303d1cbf
JV		 1384
1385 dev_mc_sync_multiple(slave_dev, bond_dev);
1386 dev_uc_sync_multiple(slave_dev, bond_dev);
1387
b9e40857 1388 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1389 }
1390
01844098 1391 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1da177e4
LT		 1392 /* add lacpdu mc addr to mc list */
1393 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1394
22bedad3 1395 dev_mc_add(slave_dev, lacpdu_multicast);
1da177e4
LT		 1396 }
1397
b8e2fde4
WY		 1398 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1399 if (res) {
76444f50
VF		 1400 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1401 slave_dev->name);
1ff412ad 1402 goto err_close;
1403 }
1da177e4 1404
c8c23903 1405 prev_slave = bond_last_slave(bond);
1da177e4
LT		 1406
1407 new_slave->delay = 0;
1408 new_slave->link_failure_count = 0;
1409
876254ae
VF		 1410 bond_update_speed_duplex(new_slave);
1411
49f17de7 1412 new_slave->last_rx = jiffies -
f31c7937 1413 (msecs_to_jiffies(bond->params.arp_interval) + 1);
8599b52e 1414 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
49f17de7 1415 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
f5b2b966 1416
1da177e4
LT		 1417 if (bond->params.miimon && !bond->params.use_carrier) {
1418 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1419
1420 if ((link_reporting == -1) && !bond->params.arp_interval) {
547942ca 1421 /* miimon is set but a bonded network driver
1da177e4
LT		 1422 * does not support ETHTOOL/MII and
1423 * arp_interval is not set. Note: if
1424 * use_carrier is enabled, we will never go
1425 * here (because netif_carrier is always
1426 * supported); thus, we don't need to change
1427 * the messages for netif_carrier.
1428 */
76444f50
VF		 1429 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1430 slave_dev->name);
1da177e4
LT		 1431 } else if (link_reporting == -1) {
1432 /* unable get link status using mii/ethtool */
76444f50
VF		 1433 netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1434 slave_dev->name);
1da177e4
LT		 1435 }
1436 }
1437
1438 /* check for initial state */
f31c7937
MK		 1439 if (bond->params.miimon) {
1440 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1441 if (bond->params.updelay) {
1442 new_slave->link = BOND_LINK_BACK;
1443 new_slave->delay = bond->params.updelay;
1444 } else {
1445 new_slave->link = BOND_LINK_UP;
1446 }
1da177e4 1447 } else {
f31c7937 1448 new_slave->link = BOND_LINK_DOWN;
1da177e4 1449 }
f31c7937
MK		 1450 } else if (bond->params.arp_interval) {
1451 new_slave->link = (netif_carrier_ok(slave_dev) ?
1452 BOND_LINK_UP : BOND_LINK_DOWN);
1da177e4 1453 } else {
f31c7937 1454 new_slave->link = BOND_LINK_UP;
1da177e4
LT		 1455 }
1456
f31c7937 1457 if (new_slave->link != BOND_LINK_DOWN)
8e603460 1458 new_slave->last_link_up = jiffies;
76444f50
VF		 1459 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1460 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1461 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
f31c7937 1462
ec0865a9 1463 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1da177e4 1464 /* if there is a primary slave, remember it */
a549952a 1465 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
059b47e8 1466 rcu_assign_pointer(bond->primary_slave, new_slave);
a549952a
JP		 1467 bond->force_primary = true;
1468 }
1da177e4
LT		 1469 }
1470
01844098 1471 switch (BOND_MODE(bond)) {
1da177e4 1472 case BOND_MODE_ACTIVEBACKUP:
5e5b0665 1473 bond_set_slave_inactive_flags(new_slave,
1474 BOND_SLAVE_NOTIFY_NOW);
1da177e4
LT		 1475 break;
1476 case BOND_MODE_8023AD:
1477 /* in 802.3ad mode, the internal mechanism
1478 * will activate the slaves in the selected
1479 * aggregator
1480 */
5e5b0665 1481 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1da177e4 1482 /* if this is the first slave */
23c147e0 1483 if (!prev_slave) {
3fdddd85 1484 SLAVE_AD_INFO(new_slave)->id = 1;
1da177e4
LT		 1485 /* Initialize AD with the number of times that the AD timer is called in 1 second
1486 * can be called only after the mac address of the bond is set
1487 */
56d00c67 1488 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1da177e4 1489 } else {
3fdddd85 1490 SLAVE_AD_INFO(new_slave)->id =
1491 SLAVE_AD_INFO(prev_slave)->id + 1;
1da177e4
LT		 1492 }
1493
1494 bond_3ad_bind_slave(new_slave);
1495 break;
1496 case BOND_MODE_TLB:
1497 case BOND_MODE_ALB:
e30bc066 1498 bond_set_active_slave(new_slave);
5e5b0665 1499 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1da177e4
LT		 1500 break;
1501 default:
76444f50 1502 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1da177e4
LT		 1503
1504 /* always active in trunk mode */
e30bc066 1505 bond_set_active_slave(new_slave);
1da177e4
LT		 1506
1507 /* In trunking mode there is little meaning to curr_active_slave
1508 * anyway (it holds no special properties of the bond device),
1509 * so we can change it without calling change_active_interface()
1510 */
4740d638
ED		 1511 if (!rcu_access_pointer(bond->curr_active_slave) &&
1512 new_slave->link == BOND_LINK_UP)
278b2083 1513 rcu_assign_pointer(bond->curr_active_slave, new_slave);
3d632c3f 1514
1da177e4
LT		 1515 break;
1516 } /* switch(bond_mode) */
1517
f6dc31a8 1518#ifdef CONFIG_NET_POLL_CONTROLLER
10eccb46 1519 slave_dev->npinfo = bond->dev->npinfo;
8a8efa22
AW		 1520 if (slave_dev->npinfo) {
1521 if (slave_enable_netpoll(new_slave)) {
76444f50 1522 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
8a8efa22 1523 res = -EBUSY;
f7d9821a 1524 goto err_detach;
8a8efa22 1525 }
f6dc31a8
WC		 1526 }
1527#endif
8a8efa22 1528
fbe168ba
MK		 1529 if (!(bond_dev->features & NETIF_F_LRO))
1530 dev_disable_lro(slave_dev);
1531
35d48903
JP		 1532 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1533 new_slave);
1534 if (res) {
76444f50 1535 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
5831d66e 1536 goto err_detach;
35d48903
JP		 1537 }
1538
1f718f0f
VF		 1539 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1540 if (res) {
76444f50 1541 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1f718f0f
VF		 1542 goto err_unregister;
1543 }
1544
07699f9a 1545 res = bond_sysfs_slave_add(new_slave);
1546 if (res) {
76444f50 1547 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
07699f9a 1548 goto err_upper_unlink;
1549 }
1550
5378c2e6
VF		 1551 bond->slave_cnt++;
1552 bond_compute_features(bond);
1553 bond_set_carrier(bond);
1554
ec0865a9 1555 if (bond_uses_primary(bond)) {
f80889a5 1556 block_netpoll_tx();
5378c2e6 1557 bond_select_active_slave(bond);
f80889a5 1558 unblock_netpoll_tx();
5378c2e6 1559 }
1f718f0f 1560
ee637714
MB		 1561 if (bond_mode_uses_xmit_hash(bond))
1562 bond_update_slave_arr(bond, NULL);
1563
76444f50
VF		 1564 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1565 slave_dev->name,
1566 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1567 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1da177e4
LT		 1568
1569 /* enslave is successful */
1570 return 0;
1571
1572/* Undo stages on error */
07699f9a 1573err_upper_unlink:
1574 bond_upper_dev_unlink(bond_dev, slave_dev);
1575
1f718f0f
VF		 1576err_unregister:
1577 netdev_rx_handler_unregister(slave_dev);
1578
f7d9821a 1579err_detach:
ec0865a9 1580 if (!bond_uses_primary(bond))
303d1cbf
JV		 1581 bond_hw_addr_flush(bond_dev, slave_dev);
1582
1ff412ad 1583 vlan_vids_del_by_dev(slave_dev, bond_dev);
059b47e8
NA		 1584 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1585 RCU_INIT_POINTER(bond->primary_slave, NULL);
4740d638 1586 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
f80889a5 1587 block_netpoll_tx();
c8517035 1588 bond_change_active_slave(bond, NULL);
3c5913b5 1589 bond_select_active_slave(bond);
f80889a5 1590 unblock_netpoll_tx();
3c5913b5 1591 }
059b47e8
NA		 1592 /* either primary_slave or curr_active_slave might've changed */
1593 synchronize_rcu();
fc7a72ac 1594 slave_disable_netpoll(new_slave);
f7d9821a 1595
1da177e4 1596err_close:
b6a5a7b9 1597 slave_dev->priv_flags &= ~IFF_BONDING;
1da177e4
LT		 1598 dev_close(slave_dev);
1599
1600err_restore_mac:
00503b6f 1601 if (!bond->params.fail_over_mac ||
01844098 1602 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
3915c1e8
JV		 1603 /* XXX TODO - fom follow mode needs to change master's
1604 * MAC if this slave's MAC is in use by the bond, or at
1605 * least print a warning.
1606 */
ada0f863 1607 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
2ab82852
MS		 1608 addr.sa_family = slave_dev->type;
1609 dev_set_mac_address(slave_dev, &addr);
1610 }
1da177e4 1611
b15ba0fb
JP		 1612err_restore_mtu:
1613 dev_set_mtu(slave_dev, new_slave->original_mtu);
1614
1da177e4 1615err_free:
3fdddd85 1616 bond_free_slave(new_slave);
1da177e4
LT		 1617
1618err_undo_flags:
b8fad459 1619 /* Enslave of first slave has failed and we need to fix master's mac */
0965a1f3 1620 if (!bond_has_slaves(bond) &&
844223ab 1621 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
b8fad459 1622 eth_hw_addr_random(bond_dev);
3d632c3f 1623
1da177e4
LT		 1624 return res;
1625}
1626
547942ca 1627/* Try to release the slave device <slave> from the bond device <master>
1da177e4 1628 * It is legal to access curr_active_slave without a lock because all the function
8c0bc550 1629 * is RTNL-locked. If "all" is true it means that the function is being called
0896341a 1630 * while destroying a bond interface and all slaves are being released.
1da177e4
LT		 1631 *
1632 * The rules for slave state should be:
1633 * for Active/Backup:
1634 * Active stays on all backups go down
1635 * for Bonded connections:
1636 * The first up interface should be left on and all others downed.
1637 */
0896341a 1638static int __bond_release_one(struct net_device *bond_dev,
1639 struct net_device *slave_dev,
1640 bool all)
1da177e4 1641{
454d7c9b 1642 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1643 struct slave *slave, *oldcurrent;
1644 struct sockaddr addr;
5a0068de 1645 int old_flags = bond_dev->flags;
c8f44aff 1646 netdev_features_t old_features = bond_dev->features;
1da177e4
LT		 1647
1648 /* slave is not a slave or master is not master of this slave */
1649 if (!(slave_dev->flags & IFF_SLAVE) ||
471cb5a3 1650 !netdev_has_upper_dev(slave_dev, bond_dev)) {
76444f50
VF		 1651 netdev_err(bond_dev, "cannot release %s\n",
1652 slave_dev->name);
1da177e4
LT		 1653 return -EINVAL;
1654 }
1655
e843fa50 1656 block_netpoll_tx();
1da177e4
LT		 1657
1658 slave = bond_get_slave_by_dev(bond, slave_dev);
1659 if (!slave) {
1660 /* not a slave of this bond */
76444f50
VF		 1661 netdev_info(bond_dev, "%s not enslaved\n",
1662 slave_dev->name);
e843fa50 1663 unblock_netpoll_tx();
1da177e4
LT		 1664 return -EINVAL;
1665 }
1666
07699f9a 1667 bond_sysfs_slave_del(slave);
1668
5f0c5f73
AG		 1669 /* recompute stats just before removing the slave */
1670 bond_get_stats(bond->dev, &bond->bond_stats);
1671
1f718f0f 1672 bond_upper_dev_unlink(bond_dev, slave_dev);
35d48903
JP		 1673 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1674 * for this slave anymore.
1675 */
1676 netdev_rx_handler_unregister(slave_dev);
35d48903 1677
e470259f 1678 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1da177e4 1679 bond_3ad_unbind_slave(slave);
1da177e4 1680
ee637714
MB		 1681 if (bond_mode_uses_xmit_hash(bond))
1682 bond_update_slave_arr(bond, slave);
1683
76444f50
VF		 1684 netdev_info(bond_dev, "Releasing %s interface %s\n",
1685 bond_is_active_slave(slave) ? "active" : "backup",
1686 slave_dev->name);
1da177e4 1687
4740d638 1688 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1da177e4 1689
85741718 1690 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1da177e4 1691
00503b6f 1692 if (!all && (!bond->params.fail_over_mac ||
01844098 1693 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
844223ab 1694 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
0965a1f3 1695 bond_has_slaves(bond))
76444f50
VF		 1696 netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1697 slave_dev->name, slave->perm_hwaddr,
1698 bond_dev->name, slave_dev->name);
dec1e90e 1699 }
1700
059b47e8
NA		 1701 if (rtnl_dereference(bond->primary_slave) == slave)
1702 RCU_INIT_POINTER(bond->primary_slave, NULL);
1da177e4 1703
1c72cfdc 1704 if (oldcurrent == slave)
1da177e4 1705 bond_change_active_slave(bond, NULL);
1da177e4 1706
58402054 1707 if (bond_is_lb(bond)) {
1da177e4
LT		 1708 /* Must be called only after the slave has been
1709 * detached from the list and the curr_active_slave
1710 * has been cleared (if our_slave == old_current),
1711 * but before a new active slave is selected.
1712 */
1713 bond_alb_deinit_slave(bond, slave);
1714 }
1715
0896341a 1716 if (all) {
36708b89 1717 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
0896341a 1718 } else if (oldcurrent == slave) {
547942ca 1719 /* Note that we hold RTNL over this sequence, so there
059fe7a5
JV		 1720 * is no concern that another slave add/remove event
1721 * will interfere.
1722 */
1da177e4 1723 bond_select_active_slave(bond);
059fe7a5
JV		 1724 }
1725
0965a1f3 1726 if (!bond_has_slaves(bond)) {
ff59c456 1727 bond_set_carrier(bond);
409cc1f8 1728 eth_hw_addr_random(bond_dev);
1da177e4
LT		 1729 }
1730
e843fa50 1731 unblock_netpoll_tx();
278b2083 1732 synchronize_rcu();
ee6154e1 1733 bond->slave_cnt--;
1da177e4 1734
0965a1f3 1735 if (!bond_has_slaves(bond)) {
2af73d4b 1736 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
80028ea1
VF		 1737 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1738 }
2af73d4b 1739
b2a103e6
MM		 1740 bond_compute_features(bond);
1741 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1742 (old_features & NETIF_F_VLAN_CHALLENGED))
76444f50
VF		 1743 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1744 slave_dev->name, bond_dev->name);
b2a103e6 1745
1ff412ad 1746 vlan_vids_del_by_dev(slave_dev, bond_dev);
1da177e4 1747
547942ca 1748 /* If the mode uses primary, then this case was handled above by
303d1cbf 1749 * bond_change_active_slave(..., NULL)
1da177e4 1750 */
ec0865a9 1751 if (!bond_uses_primary(bond)) {
5a0068de
NH		 1752 /* unset promiscuity level from slave
1753 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1754 * of the IFF_PROMISC flag in the bond_dev, but we need the
1755 * value of that flag before that change, as that was the value
1756 * when this slave was attached, so we cache at the start of the
1757 * function and use it here. Same goes for ALLMULTI below
1758 */
1759 if (old_flags & IFF_PROMISC)
1da177e4 1760 dev_set_promiscuity(slave_dev, -1);
1da177e4
LT		 1761
1762 /* unset allmulti level from slave */
5a0068de 1763 if (old_flags & IFF_ALLMULTI)
1da177e4 1764 dev_set_allmulti(slave_dev, -1);
1da177e4 1765
303d1cbf 1766 bond_hw_addr_flush(bond_dev, slave_dev);
1da177e4
LT		 1767 }
1768
8a8efa22 1769 slave_disable_netpoll(slave);
f6dc31a8 1770
1da177e4
LT		 1771 /* close slave before restoring its mac address */
1772 dev_close(slave_dev);
1773
00503b6f 1774 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
01844098 1775 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2ab82852 1776 /* restore original ("permanent") mac address */
ada0f863 1777 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
2ab82852
MS		 1778 addr.sa_family = slave_dev->type;
1779 dev_set_mac_address(slave_dev, &addr);
1780 }
1da177e4 1781
b15ba0fb
JP		 1782 dev_set_mtu(slave_dev, slave->original_mtu);
1783
2d7011ca 1784 slave_dev->priv_flags &= ~IFF_BONDING;
1da177e4 1785
3fdddd85 1786 bond_free_slave(slave);
1da177e4 1787
547942ca 1788 return 0;
1da177e4
LT		 1789}
1790
0896341a 1791/* A wrapper used because of ndo_del_link */
1792int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1793{
1794 return __bond_release_one(bond_dev, slave_dev, false);
1795}
1796
547942ca
NA		 1797/* First release a slave and then destroy the bond if no more slaves are left.
1798 * Must be under rtnl_lock when this function is called.
1799 */
26d8ee75 1800static int bond_release_and_destroy(struct net_device *bond_dev,
1801 struct net_device *slave_dev)
d90a162a 1802{
454d7c9b 1803 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 1804 int ret;
1805
1806 ret = bond_release(bond_dev, slave_dev);
0965a1f3 1807 if (ret == 0 && !bond_has_slaves(bond)) {
8a8efa22 1808 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
76444f50
VF		 1809 netdev_info(bond_dev, "Destroying bond %s\n",
1810 bond_dev->name);
9e71626c 1811 unregister_netdevice(bond_dev);
d90a162a
MS		 1812 }
1813 return ret;
1814}
1815
1da177e4
LT		 1816static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1817{
454d7c9b 1818 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 1819
01844098 1820 info->bond_mode = BOND_MODE(bond);
1da177e4 1821 info->miimon = bond->params.miimon;
1da177e4 1822 info->num_slaves = bond->slave_cnt;
1da177e4
LT		 1823
1824 return 0;
1825}
1826
1827static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1828{
454d7c9b 1829 struct bonding *bond = netdev_priv(bond_dev);
9caff1e7 1830 struct list_head *iter;
dec1e90e 1831 int i = 0, res = -ENODEV;
1da177e4 1832 struct slave *slave;
1da177e4 1833
9caff1e7 1834 bond_for_each_slave(bond, slave, iter) {
dec1e90e 1835 if (i++ == (int)info->slave_id) {
689c96cc
ED		 1836 res = 0;
1837 strcpy(info->slave_name, slave->dev->name);
1838 info->link = slave->link;
e30bc066 1839 info->state = bond_slave_state(slave);
689c96cc 1840 info->link_failure_count = slave->link_failure_count;
1da177e4
LT		 1841 break;
1842 }
1843 }
1da177e4 1844
689c96cc 1845 return res;
1da177e4
LT		 1846}
1847
1848/*-------------------------------- Monitoring -------------------------------*/
1849
4740d638 1850/* called with rcu_read_lock() */
f0c76d61
JV		 1851static int bond_miimon_inspect(struct bonding *bond)
1852{
dec1e90e 1853 int link_state, commit = 0;
9caff1e7 1854 struct list_head *iter;
f0c76d61 1855 struct slave *slave;
41f89100
JP		 1856 bool ignore_updelay;
1857
4740d638 1858 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1da177e4 1859
4cb4f97b 1860 bond_for_each_slave_rcu(bond, slave, iter) {
f0c76d61 1861 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 1862
f0c76d61 1863 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 1864
1865 switch (slave->link) {
f0c76d61
JV		 1866 case BOND_LINK_UP:
1867 if (link_state)
1868 continue;
1da177e4 1869
f0c76d61
JV		 1870 slave->link = BOND_LINK_FAIL;
1871 slave->delay = bond->params.downdelay;
1872 if (slave->delay) {
76444f50
VF		 1873 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1874 (BOND_MODE(bond) ==
1875 BOND_MODE_ACTIVEBACKUP) ?
1876 (bond_is_active_slave(slave) ?
1877 "active " : "backup ") : "",
1878 slave->dev->name,
1879 bond->params.downdelay * bond->params.miimon);
1da177e4 1880 }
f0c76d61
JV		 1881 /*FALLTHRU*/
1882 case BOND_LINK_FAIL:
1883 if (link_state) {
547942ca 1884 /* recovered before downdelay expired */
f0c76d61 1885 slave->link = BOND_LINK_UP;
8e603460 1886 slave->last_link_up = jiffies;
76444f50
VF		 1887 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1888 (bond->params.downdelay - slave->delay) *
1889 bond->params.miimon,
1890 slave->dev->name);
f0c76d61 1891 continue;
1da177e4 1892 }
f0c76d61
JV		 1893
1894 if (slave->delay <= 0) {
1895 slave->new_link = BOND_LINK_DOWN;
1896 commit++;
1897 continue;
1da177e4 1898 }
1da177e4 1899
f0c76d61
JV		 1900 slave->delay--;
1901 break;
1902
1903 case BOND_LINK_DOWN:
1904 if (!link_state)
1905 continue;
1906
1907 slave->link = BOND_LINK_BACK;
1908 slave->delay = bond->params.updelay;
1909
1910 if (slave->delay) {
76444f50
VF		 1911 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1912 slave->dev->name,
1913 ignore_updelay ? 0 :
1914 bond->params.updelay *
1915 bond->params.miimon);
f0c76d61
JV		 1916 }
1917 /*FALLTHRU*/
1918 case BOND_LINK_BACK:
1919 if (!link_state) {
1920 slave->link = BOND_LINK_DOWN;
76444f50
VF		 1921 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
1922 (bond->params.updelay - slave->delay) *
1923 bond->params.miimon,
1924 slave->dev->name);
f0c76d61
JV		 1925
1926 continue;
1927 }
1928
41f89100
JP		 1929 if (ignore_updelay)
1930 slave->delay = 0;
1931
f0c76d61
JV		 1932 if (slave->delay <= 0) {
1933 slave->new_link = BOND_LINK_UP;
1934 commit++;
41f89100 1935 ignore_updelay = false;
f0c76d61 1936 continue;
1da177e4 1937 }
f0c76d61
JV		 1938
1939 slave->delay--;
1da177e4 1940 break;
f0c76d61
JV		 1941 }
1942 }
1da177e4 1943
f0c76d61
JV		 1944 return commit;
1945}
1da177e4 1946
f0c76d61
JV		 1947static void bond_miimon_commit(struct bonding *bond)
1948{
9caff1e7 1949 struct list_head *iter;
059b47e8 1950 struct slave *slave, *primary;
f0c76d61 1951
9caff1e7 1952 bond_for_each_slave(bond, slave, iter) {
f0c76d61
JV		 1953 switch (slave->new_link) {
1954 case BOND_LINK_NOCHANGE:
1955 continue;
1da177e4 1956
f0c76d61
JV		 1957 case BOND_LINK_UP:
1958 slave->link = BOND_LINK_UP;
8e603460 1959 slave->last_link_up = jiffies;
f0c76d61 1960
059b47e8 1961 primary = rtnl_dereference(bond->primary_slave);
01844098 1962 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
f0c76d61 1963 /* prevent it from being the active one */
e30bc066 1964 bond_set_backup_slave(slave);
01844098 1965 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
f0c76d61 1966 /* make it immediately active */
e30bc066 1967 bond_set_active_slave(slave);
059b47e8 1968 } else if (slave != primary) {
f0c76d61 1969 /* prevent it from being the active one */
e30bc066 1970 bond_set_backup_slave(slave);
1da177e4 1971 }
1da177e4 1972
76444f50
VF		 1973 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
1974 slave->dev->name,
1975 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1976 slave->duplex ? "full" : "half");
1da177e4 1977
f0c76d61 1978 /* notify ad that the link status has changed */
01844098 1979 if (BOND_MODE(bond) == BOND_MODE_8023AD)
f0c76d61 1980 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 1981
58402054 1982 if (bond_is_lb(bond))
f0c76d61
JV		 1983 bond_alb_handle_link_change(bond, slave,
1984 BOND_LINK_UP);
1da177e4 1985
ee637714
MB		 1986 if (BOND_MODE(bond) == BOND_MODE_XOR)
1987 bond_update_slave_arr(bond, NULL);
1988
059b47e8 1989 if (!bond->curr_active_slave || slave == primary)
f0c76d61 1990 goto do_failover;
1da177e4 1991
f0c76d61 1992 continue;
059fe7a5 1993
f0c76d61 1994 case BOND_LINK_DOWN:
fba4acda
JV		 1995 if (slave->link_failure_count < UINT_MAX)
1996 slave->link_failure_count++;
1997
f0c76d61 1998 slave->link = BOND_LINK_DOWN;
1da177e4 1999
01844098
VF		 2000 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2001 BOND_MODE(bond) == BOND_MODE_8023AD)
5e5b0665 2002 bond_set_slave_inactive_flags(slave,
2003 BOND_SLAVE_NOTIFY_NOW);
f0c76d61 2004
76444f50
VF		 2005 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2006 slave->dev->name);
f0c76d61 2007
01844098 2008 if (BOND_MODE(bond) == BOND_MODE_8023AD)
f0c76d61
JV		 2009 bond_3ad_handle_link_change(slave,
2010 BOND_LINK_DOWN);
2011
ae63e808 2012 if (bond_is_lb(bond))
f0c76d61
JV		 2013 bond_alb_handle_link_change(bond, slave,
2014 BOND_LINK_DOWN);
2015
ee637714
MB		 2016 if (BOND_MODE(bond) == BOND_MODE_XOR)
2017 bond_update_slave_arr(bond, NULL);
2018
4740d638 2019 if (slave == rcu_access_pointer(bond->curr_active_slave))
f0c76d61
JV		 2020 goto do_failover;
2021
2022 continue;
2023
2024 default:
76444f50
VF		 2025 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2026 slave->new_link, slave->dev->name);
f0c76d61
JV		 2027 slave->new_link = BOND_LINK_NOCHANGE;
2028
2029 continue;
2030 }
2031
2032do_failover:
e843fa50 2033 block_netpoll_tx();
f0c76d61 2034 bond_select_active_slave(bond);
e843fa50 2035 unblock_netpoll_tx();
f0c76d61
JV		 2036 }
2037
2038 bond_set_carrier(bond);
1da177e4
LT		 2039}
2040
547942ca 2041/* bond_mii_monitor
0b0eef66
JV		 2042 *
2043 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2044 * inspection, then (if inspection indicates something needs to be done)
2045 * an acquisition of appropriate locks followed by a commit phase to
2046 * implement whatever link state changes are indicated.
0b0eef66 2047 */
6da67d26 2048static void bond_mii_monitor(struct work_struct *work)
0b0eef66
JV		 2049{
2050 struct bonding *bond = container_of(work, struct bonding,
2051 mii_work.work);
ad246c99 2052 bool should_notify_peers = false;
1f2cd845 2053 unsigned long delay;
0b0eef66 2054
1f2cd845
DM		 2055 delay = msecs_to_jiffies(bond->params.miimon);
2056
2057 if (!bond_has_slaves(bond))
f0c76d61 2058 goto re_arm;
b59f9f74 2059
4cb4f97b 2060 rcu_read_lock();
2061
ad246c99
BH		 2062 should_notify_peers = bond_should_notify_peers(bond);
2063
1f2cd845 2064 if (bond_miimon_inspect(bond)) {
4cb4f97b 2065 rcu_read_unlock();
f0c76d61 2066
1f2cd845
DM		 2067 /* Race avoidance with bond_close cancel of workqueue */
2068 if (!rtnl_trylock()) {
1f2cd845
DM		 2069 delay = 1;
2070 should_notify_peers = false;
2071 goto re_arm;
2072 }
6b6c5261 2073
1f2cd845
DM		 2074 bond_miimon_commit(bond);
2075
1f2cd845 2076 rtnl_unlock(); /* might sleep, hold no other locks */
4cb4f97b 2077 } else
2078 rcu_read_unlock();
0b0eef66 2079
f0c76d61 2080re_arm:
e6d265e8 2081 if (bond->params.miimon)
1f2cd845
DM		 2082 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2083
1f2cd845
DM		 2084 if (should_notify_peers) {
2085 if (!rtnl_trylock())
2086 return;
2087 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2088 rtnl_unlock();
2089 }
0b0eef66 2090}
c3ade5ca 2091
50223ce4 2092static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2093{
50223ce4
VF		 2094 struct net_device *upper;
2095 struct list_head *iter;
2096 bool ret = false;
f5b2b966 2097
eaddcd76 2098 if (ip == bond_confirm_addr(bond->dev, 0, ip))
50223ce4 2099 return true;
f5b2b966 2100
50223ce4 2101 rcu_read_lock();
2f268f12 2102 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
50223ce4
VF		 2103 if (ip == bond_confirm_addr(upper, 0, ip)) {
2104 ret = true;
2105 break;
2106 }
f5b2b966 2107 }
50223ce4 2108 rcu_read_unlock();
f5b2b966 2109
50223ce4 2110 return ret;
f5b2b966
JV		 2111}
2112
547942ca 2113/* We go to the (large) trouble of VLAN tagging ARP frames because
c3ade5ca
JV		 2114 * switches in VLAN mode (especially if ports are configured as
2115 * "native" to a VLAN) might not pass non-tagged frames.
2116 */
fbd929f2 2117static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2118 __be32 dest_ip, __be32 src_ip,
44a40855 2119 struct bond_vlan_tag *tags)
c3ade5ca
JV		 2120{
2121 struct sk_buff *skb;
3e403a77 2122 struct bond_vlan_tag *outer_tag = tags;
c3ade5ca 2123
76444f50
VF		 2124 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2125 arp_op, slave_dev->name, &dest_ip, &src_ip);
3d632c3f 2126
c3ade5ca
JV		 2127 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2128 NULL, slave_dev->dev_addr, NULL);
2129
2130 if (!skb) {
4873ac3c 2131 net_err_ratelimited("ARP packet allocation failed\n");
c3ade5ca
JV		 2132 return;
2133 }
fbd929f2 2134
3e403a77
VF		 2135 if (!tags || tags->vlan_proto == VLAN_N_VID)
2136 goto xmit;
2137
2138 tags++;
2139
44a40855 2140 /* Go through all the tags backwards and add them to the packet */
3e403a77
VF		 2141 while (tags->vlan_proto != VLAN_N_VID) {
2142 if (!tags->vlan_id) {
2143 tags++;
44a40855 2144 continue;
3e403a77 2145 }
44a40855 2146
76444f50 2147 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
3e403a77
VF		 2148 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2149 skb = __vlan_put_tag(skb, tags->vlan_proto,
2150 tags->vlan_id);
44a40855
VY		 2151 if (!skb) {
2152 net_err_ratelimited("failed to insert inner VLAN tag\n");
2153 return;
2154 }
3e403a77
VF		 2155
2156 tags++;
44a40855
VY		 2157 }
2158 /* Set the outer tag */
3e403a77 2159 if (outer_tag->vlan_id) {
76444f50 2160 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
3e403a77 2161 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
b4bef1b5
JP		 2162 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2163 outer_tag->vlan_id);
c3ade5ca 2164 }
3e403a77
VF		 2165
2166xmit:
c3ade5ca
JV		 2167 arp_xmit(skb);
2168}
2169
44a40855
VY		 2170/* Validate the device path between the @start_dev and the @end_dev.
2171 * The path is valid if the @end_dev is reachable through device
2172 * stacking.
2173 * When the path is validated, collect any vlan information in the
2174 * path.
2175 */
3e403a77
VF		 2176struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2177 struct net_device *end_dev,
2178 int level)
44a40855 2179{
3e403a77 2180 struct bond_vlan_tag *tags;
44a40855
VY		 2181 struct net_device *upper;
2182 struct list_head *iter;
44a40855 2183
3e403a77
VF		 2184 if (start_dev == end_dev) {
2185 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2186 if (!tags)
2187 return ERR_PTR(-ENOMEM);
2188 tags[level].vlan_proto = VLAN_N_VID;
2189 return tags;
2190 }
44a40855
VY		 2191
2192 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
3e403a77
VF		 2193 tags = bond_verify_device_path(upper, end_dev, level + 1);
2194 if (IS_ERR_OR_NULL(tags)) {
2195 if (IS_ERR(tags))
2196 return tags;
2197 continue;
44a40855 2198 }
3e403a77
VF		 2199 if (is_vlan_dev(upper)) {
2200 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2201 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2202 }
2203
2204 return tags;
44a40855
VY		 2205 }
2206
3e403a77 2207 return NULL;
44a40855 2208}
c3ade5ca 2209
1da177e4
LT		 2210static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2211{
c3ade5ca 2212 struct rtable *rt;
3e403a77 2213 struct bond_vlan_tag *tags;
27bc11e6 2214 __be32 *targets = bond->params.arp_targets, addr;
fbd929f2 2215 int i;
1da177e4 2216
27bc11e6 2217 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
76444f50 2218 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
3e403a77 2219 tags = NULL;
c3ade5ca 2220
27bc11e6 2221 /* Find out through which dev should the packet go */
78fbfd8a
DM		 2222 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2223 RTO_ONLINK, 0);
b23dd4fe 2224 if (IS_ERR(rt)) {
28572760
VF		 2225 /* there's no route to target - try to send arp
2226 * probe to generate any traffic (arp_validate=0)
2227 */
4873ac3c 2228 if (bond->params.arp_validate)
2229 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2230 bond->dev->name,
2231 &targets[i]);
44a40855
VY		 2232 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2233 0, tags);
c3ade5ca
JV		 2234 continue;
2235 }
2236
27bc11e6
VF		 2237 /* bond device itself */
2238 if (rt->dst.dev == bond->dev)
2239 goto found;
2240
2241 rcu_read_lock();
3e403a77 2242 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
27bc11e6 2243 rcu_read_unlock();
c3ade5ca 2244
3e403a77 2245 if (!IS_ERR_OR_NULL(tags))
44a40855
VY		 2246 goto found;
2247
27bc11e6 2248 /* Not our device - skip */
76444f50
VF		 2249 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2250 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3e32582f 2251
ed4b9f80 2252 ip_rt_put(rt);
27bc11e6
VF		 2253 continue;
2254
2255found:
2256 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2257 ip_rt_put(rt);
2258 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
44a40855 2259 addr, tags);
a67eed57 2260 kfree(tags);
c3ade5ca
JV		 2261 }
2262}
2263
d3bb52b0 2264static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966 2265{
8599b52e
VF		 2266 int i;
2267
87a7b84b 2268 if (!sip || !bond_has_this_ip(bond, tip)) {
76444f50
VF		 2269 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2270 &sip, &tip);
87a7b84b
VF		 2271 return;
2272 }
f5b2b966 2273
8599b52e
VF		 2274 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2275 if (i == -1) {
76444f50
VF		 2276 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2277 &sip);
87a7b84b 2278 return;
f5b2b966 2279 }
49f17de7 2280 slave->last_rx = jiffies;
8599b52e 2281 slave->target_last_arp_rx[i] = jiffies;
f5b2b966
JV		 2282}
2283
5bb9e0b5 2284int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2285 struct slave *slave)
f5b2b966 2286{
de063b70 2287 struct arphdr *arp = (struct arphdr *)skb->data;
010d3c39 2288 struct slave *curr_active_slave;
f5b2b966 2289 unsigned char *arp_ptr;
d3bb52b0 2290 __be32 sip, tip;
f2cb691a 2291 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
f5b2b966 2292
f2cb691a 2293 if (!slave_do_arp_validate(bond, slave)) {
bedabf90 2294 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2295 !slave_do_arp_validate_only(bond))
49f17de7 2296 slave->last_rx = jiffies;
13a8e0c8 2297 return RX_HANDLER_ANOTHER;
f2cb691a
VF		 2298 } else if (!is_arp) {
2299 return RX_HANDLER_ANOTHER;
2300 }
2c146102 2301
de063b70 2302 alen = arp_hdr_len(bond->dev);
f5b2b966 2303
76444f50
VF		 2304 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2305 skb->dev->name);
f5b2b966 2306
de063b70
ED		 2307 if (alen > skb_headlen(skb)) {
2308 arp = kmalloc(alen, GFP_ATOMIC);
2309 if (!arp)
2310 goto out_unlock;
2311 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2312 goto out_unlock;
2313 }
f5b2b966 2314
3aba891d 2315 if (arp->ar_hln != bond->dev->addr_len ||
f5b2b966
JV		 2316 skb->pkt_type == PACKET_OTHERHOST ||
2317 skb->pkt_type == PACKET_LOOPBACK ||
2318 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2319 arp->ar_pro != htons(ETH_P_IP) ||
2320 arp->ar_pln != 4)
2321 goto out_unlock;
2322
2323 arp_ptr = (unsigned char *)(arp + 1);
3aba891d 2324 arp_ptr += bond->dev->addr_len;
f5b2b966 2325 memcpy(&sip, arp_ptr, 4);
3aba891d 2326 arp_ptr += 4 + bond->dev->addr_len;
f5b2b966
JV		 2327 memcpy(&tip, arp_ptr, 4);
2328
76444f50
VF		 2329 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2330 slave->dev->name, bond_slave_state(slave),
2331 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2332 &sip, &tip);
f5b2b966 2333
010d3c39
VF		 2334 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2335
547942ca 2336 /* Backup slaves won't see the ARP reply, but do come through
f5b2b966
JV		 2337 * here for each ARP probe (so we swap the sip/tip to validate
2338 * the probe). In a "redundant switch, common router" type of
2339 * configuration, the ARP probe will (hopefully) travel from
2340 * the active, through one switch, the router, then the other
2341 * switch before reaching the backup.
aeea64ac
VF		 2342 *
2343 * We 'trust' the arp requests if there is an active slave and
2344 * it received valid arp reply(s) after it became active. This
2345 * is done to avoid endless looping when we can't reach the
2346 * arp_ip_target and fool ourselves with our own arp requests.
f5b2b966 2347 */
010d3c39 2348
e30bc066 2349 if (bond_is_active_slave(slave))
f5b2b966 2350 bond_validate_arp(bond, slave, sip, tip);
010d3c39
VF		 2351 else if (curr_active_slave &&
2352 time_after(slave_last_rx(bond, curr_active_slave),
2353 curr_active_slave->last_link_up))
f5b2b966
JV		 2354 bond_validate_arp(bond, slave, tip, sip);
2355
2356out_unlock:
de063b70
ED		 2357 if (arp != (struct arphdr *)skb->data)
2358 kfree(arp);
13a8e0c8 2359 return RX_HANDLER_ANOTHER;
f5b2b966
JV		 2360}
2361
e7f63f1d
VF		 2362/* function to verify if we're in the arp_interval timeslice, returns true if
2363 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2364 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2365 */
2366static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2367 int mod)
2368{
2369 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2370
2371 return time_in_range(jiffies,
2372 last_act - delta_in_ticks,
2373 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2374}
2375
547942ca 2376/* This function is called regularly to monitor each slave's link
1da177e4
LT		 2377 * ensuring that traffic is being sent and received when arp monitoring
2378 * is used in load-balancing mode. if the adapter has been dormant, then an
2379 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2380 * arp monitoring in active backup mode.
2381 */
6da67d26 2382static void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2383{
1b76b316
JV		 2384 struct bonding *bond = container_of(work, struct bonding,
2385 arp_work.work);
1da177e4 2386 struct slave *slave, *oldcurrent;
9caff1e7 2387 struct list_head *iter;
6fde8f03 2388 int do_failover = 0, slave_state_changed = 0;
1da177e4 2389
1f2cd845 2390 if (!bond_has_slaves(bond))
1da177e4 2391 goto re_arm;
1da177e4 2392
2e52f4fe 2393 rcu_read_lock();
2394
4740d638 2395 oldcurrent = rcu_dereference(bond->curr_active_slave);
1da177e4
LT		 2396 /* see if any of the previous devices are up now (i.e. they have
2397 * xmt and rcv traffic). the curr_active_slave does not come into
8e603460
VF		 2398 * the picture unless it is null. also, slave->last_link_up is not
2399 * needed here because we send an arp on each slave and give a slave
2400 * as long as it needs to get the tx/rx within the delta.
1da177e4
LT		 2401 * TODO: what about up/down delay in arp mode? it wasn't here before
2402 * so it can wait
2403 */
2e52f4fe 2404 bond_for_each_slave_rcu(bond, slave, iter) {
cb32f2a0
JB		 2405 unsigned long trans_start = dev_trans_start(slave->dev);
2406
1da177e4 2407 if (slave->link != BOND_LINK_UP) {
e7f63f1d 2408 if (bond_time_in_interval(bond, trans_start, 1) &&
49f17de7 2409 bond_time_in_interval(bond, slave->last_rx, 1)) {
1da177e4
LT		 2410
2411 slave->link = BOND_LINK_UP;
6fde8f03 2412 slave_state_changed = 1;
1da177e4
LT		 2413
2414 /* primary_slave has no meaning in round-robin
2415 * mode. the window of a slave being up and
2416 * curr_active_slave being null after enslaving
2417 * is closed.
2418 */
2419 if (!oldcurrent) {
76444f50
VF		 2420 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2421 slave->dev->name);
1da177e4
LT		 2422 do_failover = 1;
2423 } else {
76444f50
VF		 2424 netdev_info(bond->dev, "interface %s is now up\n",
2425 slave->dev->name);
1da177e4
LT		 2426 }
2427 }
2428 } else {
2429 /* slave->link == BOND_LINK_UP */
2430
2431 /* not all switches will respond to an arp request
2432 * when the source ip is 0, so don't take the link down
2433 * if we don't know our ip yet
2434 */
e7f63f1d 2435 if (!bond_time_in_interval(bond, trans_start, 2) ||
49f17de7 2436 !bond_time_in_interval(bond, slave->last_rx, 2)) {
1da177e4
LT		 2437
2438 slave->link = BOND_LINK_DOWN;
6fde8f03 2439 slave_state_changed = 1;
1da177e4 2440
3d632c3f 2441 if (slave->link_failure_count < UINT_MAX)
1da177e4 2442 slave->link_failure_count++;
1da177e4 2443
76444f50
VF		 2444 netdev_info(bond->dev, "interface %s is now down\n",
2445 slave->dev->name);
1da177e4 2446
3d632c3f 2447 if (slave == oldcurrent)
1da177e4 2448 do_failover = 1;
1da177e4
LT		 2449 }
2450 }
2451
2452 /* note: if switch is in round-robin mode, all links
2453 * must tx arp to ensure all links rx an arp - otherwise
2454 * links may oscillate or not come up at all; if switch is
2455 * in something like xor mode, there is nothing we can
2456 * do - all replies will be rx'ed on same link causing slaves
2457 * to be unstable during low/no traffic periods
2458 */
b6adc610 2459 if (bond_slave_is_up(slave))
1da177e4 2460 bond_arp_send_all(bond, slave);
1da177e4
LT		 2461 }
2462
2e52f4fe 2463 rcu_read_unlock();
2464
6fde8f03 2465 if (do_failover || slave_state_changed) {
2e52f4fe 2466 if (!rtnl_trylock())
2467 goto re_arm;
1da177e4 2468
6fde8f03
DT		 2469 if (slave_state_changed) {
2470 bond_slave_state_change(bond);
ee637714
MB		 2471 if (BOND_MODE(bond) == BOND_MODE_XOR)
2472 bond_update_slave_arr(bond, NULL);
6fde8f03 2473 } else if (do_failover) {
6fde8f03 2474 block_netpoll_tx();
6fde8f03 2475 bond_select_active_slave(bond);
6fde8f03
DT		 2476 unblock_netpoll_tx();
2477 }
2e52f4fe 2478 rtnl_unlock();
1da177e4
LT		 2479 }
2480
2481re_arm:
e6d265e8 2482 if (bond->params.arp_interval)
e7f63f1d
VF		 2483 queue_delayed_work(bond->wq, &bond->arp_work,
2484 msecs_to_jiffies(bond->params.arp_interval));
1da177e4
LT		 2485}
2486
547942ca 2487/* Called to inspect slaves for active-backup mode ARP monitor link state
b2220cad
JV		 2488 * changes. Sets new_link in slaves to specify what action should take
2489 * place for the slave. Returns 0 if no changes are found, >0 if changes
2490 * to link states must be committed.
2491 *
8c0bc550 2492 * Called with rcu_read_lock held.
1da177e4 2493 */
e7f63f1d 2494static int bond_ab_arp_inspect(struct bonding *bond)
1da177e4 2495{
def4460c 2496 unsigned long trans_start, last_rx;
9caff1e7 2497 struct list_head *iter;
dec1e90e 2498 struct slave *slave;
dec1e90e 2499 int commit = 0;
da210f55 2500
eb9fa4b0 2501 bond_for_each_slave_rcu(bond, slave, iter) {
b2220cad 2502 slave->new_link = BOND_LINK_NOCHANGE;
def4460c 2503 last_rx = slave_last_rx(bond, slave);
1da177e4 2504
b2220cad 2505 if (slave->link != BOND_LINK_UP) {
e7f63f1d 2506 if (bond_time_in_interval(bond, last_rx, 1)) {
b2220cad
JV		 2507 slave->new_link = BOND_LINK_UP;
2508 commit++;
2509 }
b2220cad
JV		 2510 continue;
2511 }
1da177e4 2512
547942ca 2513 /* Give slaves 2*delta after being enslaved or made
b2220cad
JV		 2514 * active. This avoids bouncing, as the last receive
2515 * times need a full ARP monitor cycle to be updated.
2516 */
8e603460 2517 if (bond_time_in_interval(bond, slave->last_link_up, 2))
b2220cad
JV		 2518 continue;
2519
547942ca 2520 /* Backup slave is down if:
b2220cad
JV		 2521 * - No current_arp_slave AND
2522 * - more than 3*delta since last receive AND
2523 * - the bond has an IP address
2524 *
2525 * Note: a non-null current_arp_slave indicates
2526 * the curr_active_slave went down and we are
2527 * searching for a new one; under this condition
2528 * we only take the curr_active_slave down - this
2529 * gives each slave a chance to tx/rx traffic
2530 * before being taken out
2531 */
e30bc066 2532 if (!bond_is_active_slave(slave) &&
85741718 2533 !rcu_access_pointer(bond->current_arp_slave) &&
e7f63f1d 2534 !bond_time_in_interval(bond, last_rx, 3)) {
b2220cad
JV		 2535 slave->new_link = BOND_LINK_DOWN;
2536 commit++;
2537 }
2538
547942ca 2539 /* Active slave is down if:
b2220cad
JV		 2540 * - more than 2*delta since transmitting OR
2541 * - (more than 2*delta since receive AND
2542 * the bond has an IP address)
2543 */
cb32f2a0 2544 trans_start = dev_trans_start(slave->dev);
e30bc066 2545 if (bond_is_active_slave(slave) &&
e7f63f1d
VF		 2546 (!bond_time_in_interval(bond, trans_start, 2) ||
2547 !bond_time_in_interval(bond, last_rx, 2))) {
b2220cad
JV		 2548 slave->new_link = BOND_LINK_DOWN;
2549 commit++;
2550 }
1da177e4
LT		 2551 }
2552
b2220cad
JV		 2553 return commit;
2554}
1da177e4 2555
547942ca 2556/* Called to commit link state changes noted by inspection step of
b2220cad
JV		 2557 * active-backup mode ARP monitor.
2558 *
eb9fa4b0 2559 * Called with RTNL hold.
b2220cad 2560 */
e7f63f1d 2561static void bond_ab_arp_commit(struct bonding *bond)
b2220cad 2562{
cb32f2a0 2563 unsigned long trans_start;
9caff1e7 2564 struct list_head *iter;
dec1e90e 2565 struct slave *slave;
1da177e4 2566
9caff1e7 2567 bond_for_each_slave(bond, slave, iter) {
b2220cad
JV		 2568 switch (slave->new_link) {
2569 case BOND_LINK_NOCHANGE:
2570 continue;
ff59c456 2571
b2220cad 2572 case BOND_LINK_UP:
cb32f2a0 2573 trans_start = dev_trans_start(slave->dev);
4740d638
ED		 2574 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2575 (!rtnl_dereference(bond->curr_active_slave) &&
e7f63f1d 2576 bond_time_in_interval(bond, trans_start, 1))) {
85741718
ED		 2577 struct slave *current_arp_slave;
2578
2579 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
b2220cad 2580 slave->link = BOND_LINK_UP;
85741718 2581 if (current_arp_slave) {
5a430974 2582 bond_set_slave_inactive_flags(
85741718 2583 current_arp_slave,
5e5b0665 2584 BOND_SLAVE_NOTIFY_NOW);
85741718 2585 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
5a430974 2586 }
b2220cad 2587
76444f50
VF		 2588 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2589 slave->dev->name);
b2220cad 2590
4740d638 2591 if (!rtnl_dereference(bond->curr_active_slave) ||
059b47e8 2592 slave == rtnl_dereference(bond->primary_slave))
b9f60253 2593 goto do_failover;
1da177e4 2594
b9f60253 2595 }
1da177e4 2596
b9f60253 2597 continue;
1da177e4 2598
b2220cad
JV		 2599 case BOND_LINK_DOWN:
2600 if (slave->link_failure_count < UINT_MAX)
2601 slave->link_failure_count++;
2602
2603 slave->link = BOND_LINK_DOWN;
5e5b0665 2604 bond_set_slave_inactive_flags(slave,
2605 BOND_SLAVE_NOTIFY_NOW);
b2220cad 2606
76444f50
VF		 2607 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2608 slave->dev->name);
b2220cad 2609
4740d638 2610 if (slave == rtnl_dereference(bond->curr_active_slave)) {
85741718 2611 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
b9f60253 2612 goto do_failover;
1da177e4 2613 }
b9f60253
JP		 2614
2615 continue;
b2220cad
JV		 2616
2617 default:
76444f50
VF		 2618 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2619 slave->new_link, slave->dev->name);
b9f60253 2620 continue;
1da177e4 2621 }
1da177e4 2622
b9f60253 2623do_failover:
e843fa50 2624 block_netpoll_tx();
b9f60253 2625 bond_select_active_slave(bond);
e843fa50 2626 unblock_netpoll_tx();
b2220cad 2627 }
1da177e4 2628
b2220cad
JV		 2629 bond_set_carrier(bond);
2630}
1da177e4 2631
547942ca 2632/* Send ARP probes for active-backup mode ARP monitor.
b0929915 2633 *
8c0bc550 2634 * Called with rcu_read_lock held.
b2220cad 2635 */
f2ebd477 2636static bool bond_ab_arp_probe(struct bonding *bond)
b2220cad 2637{
eb9fa4b0 2638 struct slave *slave, *before = NULL, *new_slave = NULL,
b0929915 2639 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2640 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
4087df87
VF		 2641 struct list_head *iter;
2642 bool found = false;
b0929915 2643 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
f2ebd477 2644
98b90f26 2645 if (curr_arp_slave && curr_active_slave)
76444f50
VF		 2646 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2647 curr_arp_slave->dev->name,
2648 curr_active_slave->dev->name);
1da177e4 2649
98b90f26
VF		 2650 if (curr_active_slave) {
2651 bond_arp_send_all(bond, curr_active_slave);
b0929915 2652 return should_notify_rtnl;
b2220cad 2653 }
1da177e4 2654
b2220cad
JV		 2655 /* if we don't have a curr_active_slave, search for the next available
2656 * backup slave from the current_arp_slave and make it the candidate
2657 * for becoming the curr_active_slave
2658 */
1da177e4 2659
eb9fa4b0 2660 if (!curr_arp_slave) {
b0929915 2661 curr_arp_slave = bond_first_slave_rcu(bond);
2662 if (!curr_arp_slave)
2663 return should_notify_rtnl;
b2220cad 2664 }
1da177e4 2665
b0929915 2666 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
059fe7a5 2667
b0929915 2668 bond_for_each_slave_rcu(bond, slave, iter) {
b6adc610 2669 if (!found && !before && bond_slave_is_up(slave))
4087df87 2670 before = slave;
1da177e4 2671
b6adc610 2672 if (found && !new_slave && bond_slave_is_up(slave))
4087df87 2673 new_slave = slave;
b2220cad
JV		 2674 /* if the link state is up at this point, we
2675 * mark it down - this can happen if we have
2676 * simultaneous link failures and
2677 * reselect_active_interface doesn't make this
2678 * one the current slave so it is still marked
2679 * up when it is actually down
1da177e4 2680 */
b6adc610 2681 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
b2220cad
JV		 2682 slave->link = BOND_LINK_DOWN;
2683 if (slave->link_failure_count < UINT_MAX)
2684 slave->link_failure_count++;
1da177e4 2685
5e5b0665 2686 bond_set_slave_inactive_flags(slave,
b0929915 2687 BOND_SLAVE_NOTIFY_LATER);
b2220cad 2688
76444f50
VF		 2689 netdev_info(bond->dev, "backup interface %s is now down\n",
2690 slave->dev->name);
1da177e4 2691 }
eb9fa4b0 2692 if (slave == curr_arp_slave)
4087df87 2693 found = true;
b2220cad 2694 }
4087df87
VF		 2695
2696 if (!new_slave && before)
2697 new_slave = before;
2698
b0929915 2699 if (!new_slave)
2700 goto check_state;
4087df87
VF		 2701
2702 new_slave->link = BOND_LINK_BACK;
b0929915 2703 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
4087df87 2704 bond_arp_send_all(bond, new_slave);
8e603460 2705 new_slave->last_link_up = jiffies;
eb9fa4b0 2706 rcu_assign_pointer(bond->current_arp_slave, new_slave);
f2ebd477 2707
b0929915 2708check_state:
2709 bond_for_each_slave_rcu(bond, slave, iter) {
2710 if (slave->should_notify) {
2711 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2712 break;
2713 }
2714 }
2715 return should_notify_rtnl;
b2220cad 2716}
1da177e4 2717
6da67d26 2718static void bond_activebackup_arp_mon(struct work_struct *work)
b2220cad
JV		 2719{
2720 struct bonding *bond = container_of(work, struct bonding,
2721 arp_work.work);
b0929915 2722 bool should_notify_peers = false;
2723 bool should_notify_rtnl = false;
1f2cd845 2724 int delta_in_ticks;
1da177e4 2725
1f2cd845
DM		 2726 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2727
2728 if (!bond_has_slaves(bond))
b2220cad
JV		 2729 goto re_arm;
2730
eb9fa4b0 2731 rcu_read_lock();
b0929915 2732
ad246c99
BH		 2733 should_notify_peers = bond_should_notify_peers(bond);
2734
b0929915 2735 if (bond_ab_arp_inspect(bond)) {
2736 rcu_read_unlock();
2737
1f2cd845
DM		 2738 /* Race avoidance with bond_close flush of workqueue */
2739 if (!rtnl_trylock()) {
1f2cd845
DM		 2740 delta_in_ticks = 1;
2741 should_notify_peers = false;
2742 goto re_arm;
2743 }
b2220cad 2744
1f2cd845 2745 bond_ab_arp_commit(bond);
b0929915 2746
1f2cd845 2747 rtnl_unlock();
b0929915 2748 rcu_read_lock();
1f2cd845
DM		 2749 }
2750
b0929915 2751 should_notify_rtnl = bond_ab_arp_probe(bond);
2752 rcu_read_unlock();
ad246c99 2753
80b9d236 2754re_arm:
2755 if (bond->params.arp_interval)
1f2cd845
DM		 2756 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2757
b0929915 2758 if (should_notify_peers || should_notify_rtnl) {
1f2cd845
DM		 2759 if (!rtnl_trylock())
2760 return;
b0929915 2761
2762 if (should_notify_peers)
2763 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2764 bond->dev);
2765 if (should_notify_rtnl)
2766 bond_slave_state_notify(bond);
2767
1f2cd845
DM		 2768 rtnl_unlock();
2769 }
1da177e4
LT		 2770}
2771
1da177e4
LT		 2772/*-------------------------- netdev event handling --------------------------*/
2773
547942ca 2774/* Change device name */
1da177e4
LT		 2775static int bond_event_changename(struct bonding *bond)
2776{
1da177e4
LT		 2777 bond_remove_proc_entry(bond);
2778 bond_create_proc_entry(bond);
7e083840 2779
f073c7ca
TI		 2780 bond_debug_reregister(bond);
2781
1da177e4
LT		 2782 return NOTIFY_DONE;
2783}
2784
3d632c3f
SH		 2785static int bond_master_netdev_event(unsigned long event,
2786 struct net_device *bond_dev)
1da177e4 2787{
454d7c9b 2788 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 2789
2790 switch (event) {
2791 case NETDEV_CHANGENAME:
2792 return bond_event_changename(event_bond);
a64d49c3
EB		 2793 case NETDEV_UNREGISTER:
2794 bond_remove_proc_entry(event_bond);
2795 break;
2796 case NETDEV_REGISTER:
2797 bond_create_proc_entry(event_bond);
2798 break;
6c8d23f7 2799 case NETDEV_NOTIFY_PEERS:
2800 if (event_bond->send_peer_notif)
2801 event_bond->send_peer_notif--;
2802 break;
1da177e4
LT		 2803 default:
2804 break;
2805 }
2806
2807 return NOTIFY_DONE;
2808}
2809
3d632c3f
SH		 2810static int bond_slave_netdev_event(unsigned long event,
2811 struct net_device *slave_dev)
1da177e4 2812{
059b47e8 2813 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
6101391d 2814 struct bonding *bond;
2815 struct net_device *bond_dev;
471cb5a3
JP		 2816 u32 old_speed;
2817 u8 old_duplex;
1da177e4 2818
6101391d 2819 /* A netdev event can be generated while enslaving a device
2820 * before netdev_rx_handler_register is called in which case
2821 * slave will be NULL
2822 */
2823 if (!slave)
2824 return NOTIFY_DONE;
2825 bond_dev = slave->bond->dev;
2826 bond = slave->bond;
059b47e8 2827 primary = rtnl_dereference(bond->primary_slave);
6101391d 2828
1da177e4
LT		 2829 switch (event) {
2830 case NETDEV_UNREGISTER:
8d2ada77 2831 if (bond_dev->type != ARPHRD_ETHER)
471cb5a3
JP		 2832 bond_release_and_destroy(bond_dev, slave_dev);
2833 else
2834 bond_release(bond_dev, slave_dev);
1da177e4 2835 break;
98f41f69 2836 case NETDEV_UP:
1da177e4 2837 case NETDEV_CHANGE:
471cb5a3
JP		 2838 old_speed = slave->speed;
2839 old_duplex = slave->duplex;
17d04500 2840
471cb5a3 2841 bond_update_speed_duplex(slave);
17d04500 2842
01844098 2843 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
471cb5a3
JP		 2844 if (old_speed != slave->speed)
2845 bond_3ad_adapter_speed_changed(slave);
2846 if (old_duplex != slave->duplex)
2847 bond_3ad_adapter_duplex_changed(slave);
17d04500 2848 }
ee637714
MB		 2849 /* Refresh slave-array if applicable!
2850 * If the setup does not use miimon or arpmon (mode-specific!),
2851 * then these events will not cause the slave-array to be
2852 * refreshed. This will cause xmit to use a slave that is not
2853 * usable. Avoid such situation by refeshing the array at these
2854 * events. If these (miimon/arpmon) parameters are configured
2855 * then array gets refreshed twice and that should be fine!
2856 */
2857 if (bond_mode_uses_xmit_hash(bond))
2858 bond_update_slave_arr(bond, NULL);
1da177e4
LT		 2859 break;
2860 case NETDEV_DOWN:
ee637714
MB		 2861 if (bond_mode_uses_xmit_hash(bond))
2862 bond_update_slave_arr(bond, NULL);
1da177e4
LT		 2863 break;
2864 case NETDEV_CHANGEMTU:
547942ca 2865 /* TODO: Should slaves be allowed to
1da177e4
LT		 2866 * independently alter their MTU? For
2867 * an active-backup bond, slaves need
2868 * not be the same type of device, so
2869 * MTUs may vary. For other modes,
2870 * slaves arguably should have the
2871 * same MTUs. To do this, we'd need to
2872 * take over the slave's change_mtu
2873 * function for the duration of their
2874 * servitude.
2875 */
2876 break;
2877 case NETDEV_CHANGENAME:
3ec775b9 2878 /* we don't care if we don't have primary set */
ec0865a9 2879 if (!bond_uses_primary(bond) ||
3ec775b9
VF		 2880 !bond->params.primary[0])
2881 break;
2882
059b47e8 2883 if (slave == primary) {
3ec775b9 2884 /* slave's name changed - he's no longer primary */
059b47e8 2885 RCU_INIT_POINTER(bond->primary_slave, NULL);
3ec775b9
VF		 2886 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2887 /* we have a new primary slave */
059b47e8 2888 rcu_assign_pointer(bond->primary_slave, slave);
3ec775b9
VF		 2889 } else { /* we didn't change primary - exit */
2890 break;
2891 }
2892
76444f50 2893 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
059b47e8 2894 primary ? slave_dev->name : "none");
f80889a5 2895
2896 block_netpoll_tx();
3ec775b9 2897 bond_select_active_slave(bond);
f80889a5 2898 unblock_netpoll_tx();
1da177e4 2899 break;
8531c5ff
AK		 2900 case NETDEV_FEAT_CHANGE:
2901 bond_compute_features(bond);
2902 break;
4aa5dee4
JP		 2903 case NETDEV_RESEND_IGMP:
2904 /* Propagate to master device */
2905 call_netdevice_notifiers(event, slave->bond->dev);
2906 break;
1da177e4
LT		 2907 default:
2908 break;
2909 }
2910
2911 return NOTIFY_DONE;
2912}
2913
547942ca 2914/* bond_netdev_event: handle netdev notifier chain events.
1da177e4
LT		 2915 *
2916 * This function receives events for the netdev chain. The caller (an
e041c683 2917 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 2918 * locks for us to safely manipulate the slave devices (RTNL lock,
2919 * dev_probe_lock).
2920 */
3d632c3f
SH		 2921static int bond_netdev_event(struct notifier_block *this,
2922 unsigned long event, void *ptr)
1da177e4 2923{
351638e7 2924 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1da177e4 2925
76444f50 2926 netdev_dbg(event_dev, "event: %lx\n", event);
1da177e4 2927
0b680e75
JV		 2928 if (!(event_dev->priv_flags & IFF_BONDING))
2929 return NOTIFY_DONE;
2930
1da177e4 2931 if (event_dev->flags & IFF_MASTER) {
76444f50 2932 netdev_dbg(event_dev, "IFF_MASTER\n");
1da177e4
LT		 2933 return bond_master_netdev_event(event, event_dev);
2934 }
2935
2936 if (event_dev->flags & IFF_SLAVE) {
76444f50 2937 netdev_dbg(event_dev, "IFF_SLAVE\n");
1da177e4
LT		 2938 return bond_slave_netdev_event(event, event_dev);
2939 }
2940
2941 return NOTIFY_DONE;
2942}
2943
2944static struct notifier_block bond_netdev_notifier = {
2945 .notifier_call = bond_netdev_event,
2946};
2947
169a3e66
JV		 2948/*---------------------------- Hashing Policies -----------------------------*/
2949
32819dc1
NA		 2950/* L2 hash helper */
2951static inline u32 bond_eth_hash(struct sk_buff *skb)
6b923cb7 2952{
ce04d635 2953 struct ethhdr *ep, hdr_tmp;
6b923cb7 2954
ce04d635
JX		 2955 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
2956 if (ep)
2957 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
6b923cb7
JE		 2958 return 0;
2959}
2960
32819dc1
NA		 2961/* Extract the appropriate headers based on bond's xmit policy */
2962static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2963 struct flow_keys *fk)
6f6652be 2964{
32819dc1 2965 const struct ipv6hdr *iph6;
4394542c 2966 const struct iphdr *iph;
32819dc1 2967 int noff, proto = -1;
6b923cb7 2968
32819dc1
NA		 2969 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2970 return skb_flow_dissect(skb, fk);
2971
2972 fk->ports = 0;
2973 noff = skb_network_offset(skb);
2974 if (skb->protocol == htons(ETH_P_IP)) {
054bb880 2975 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
32819dc1 2976 return false;
6b923cb7 2977 iph = ip_hdr(skb);
32819dc1
NA		 2978 fk->src = iph->saddr;
2979 fk->dst = iph->daddr;
2980 noff += iph->ihl << 2;
2981 if (!ip_is_fragment(iph))
2982 proto = iph->protocol;
2983 } else if (skb->protocol == htons(ETH_P_IPV6)) {
054bb880 2984 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
32819dc1
NA		 2985 return false;
2986 iph6 = ipv6_hdr(skb);
2987 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2988 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2989 noff += sizeof(*iph6);
2990 proto = iph6->nexthdr;
2991 } else {
2992 return false;
6f6652be 2993 }
32819dc1
NA		 2994 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2995 fk->ports = skb_flow_get_ports(skb, noff, proto);
6f6652be 2996
32819dc1 2997 return true;
6f6652be
JV		 2998}
2999
32819dc1
NA		 3000/**
3001 * bond_xmit_hash - generate a hash value based on the xmit policy
3002 * @bond: bonding device
3003 * @skb: buffer to use for headers
32819dc1
NA		 3004 *
3005 * This function will extract the necessary headers from the skb buffer and use
3006 * them to generate a hash based on the xmit_policy set in the bonding device
169a3e66 3007 */
ee62e868 3008u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
169a3e66 3009{
32819dc1
NA		 3010 struct flow_keys flow;
3011 u32 hash;
4394542c 3012
32819dc1
NA		 3013 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3014 !bond_flow_dissect(bond, skb, &flow))
ee62e868 3015 return bond_eth_hash(skb);
169a3e66 3016
32819dc1
NA		 3017 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3018 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3019 hash = bond_eth_hash(skb);
3020 else
3021 hash = (__force u32)flow.ports;
3022 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3023 hash ^= (hash >> 16);
3024 hash ^= (hash >> 8);
3025
ee62e868 3026 return hash;
169a3e66
JV		 3027}
3028
1da177e4
LT		 3029/*-------------------------- Device entry points ----------------------------*/
3030
fbb0c41b 3031static void bond_work_init_all(struct bonding *bond)
3032{
3033 INIT_DELAYED_WORK(&bond->mcast_work,
3034 bond_resend_igmp_join_requests_delayed);
3035 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3036 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
01844098 3037 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
fbb0c41b 3038 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3039 else
3040 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3041 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
ee637714 3042 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
fbb0c41b 3043}
3044
3045static void bond_work_cancel_all(struct bonding *bond)
3046{
3047 cancel_delayed_work_sync(&bond->mii_work);
3048 cancel_delayed_work_sync(&bond->arp_work);
3049 cancel_delayed_work_sync(&bond->alb_work);
3050 cancel_delayed_work_sync(&bond->ad_work);
3051 cancel_delayed_work_sync(&bond->mcast_work);
ee637714 3052 cancel_delayed_work_sync(&bond->slave_arr_work);
fbb0c41b 3053}
3054
1da177e4
LT		 3055static int bond_open(struct net_device *bond_dev)
3056{
454d7c9b 3057 struct bonding *bond = netdev_priv(bond_dev);
9caff1e7 3058 struct list_head *iter;
ba3211cc 3059 struct slave *slave;
1da177e4 3060
ba3211cc 3061 /* reset slave->backup and slave->inactive */
0965a1f3 3062 if (bond_has_slaves(bond)) {
9caff1e7 3063 bond_for_each_slave(bond, slave, iter) {
4740d638
ED		 3064 if (bond_uses_primary(bond) &&
3065 slave != rcu_access_pointer(bond->curr_active_slave)) {
5e5b0665 3066 bond_set_slave_inactive_flags(slave,
3067 BOND_SLAVE_NOTIFY_NOW);
ba3211cc 3068 } else {
5e5b0665 3069 bond_set_slave_active_flags(slave,
3070 BOND_SLAVE_NOTIFY_NOW);
ba3211cc
PP		 3071 }
3072 }
ba3211cc 3073 }
ba3211cc 3074
fbb0c41b 3075 bond_work_init_all(bond);
5a37e8ca 3076
58402054 3077 if (bond_is_lb(bond)) {
1da177e4
LT		 3078 /* bond_alb_initialize must be called before the timer
3079 * is started.
3080 */
01844098 3081 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
b473946a 3082 return -ENOMEM;
e9f0fb88
MB		 3083 if (bond->params.tlb_dynamic_lb)
3084 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3085 }
3086
fbb0c41b 3087 if (bond->params.miimon) /* link check interval, in milliseconds. */
1b76b316 3088 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3089
3090 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3091 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3fe68df9 3092 bond->recv_probe = bond_arp_rcv;
1da177e4
LT		 3093 }
3094
01844098 3095 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1b76b316 3096 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4 3097 /* register to receive LACPDUs */
3aba891d 3098 bond->recv_probe = bond_3ad_lacpdu_recv;
fd989c83 3099 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3100 }
3101
ee637714
MB		 3102 if (bond_mode_uses_xmit_hash(bond))
3103 bond_update_slave_arr(bond, NULL);
3104
1da177e4
LT		 3105 return 0;
3106}
3107
3108static int bond_close(struct net_device *bond_dev)
3109{
454d7c9b 3110 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3111
fbb0c41b 3112 bond_work_cancel_all(bond);
6c8d23f7 3113 bond->send_peer_notif = 0;
ee8487c0 3114 if (bond_is_lb(bond))
1da177e4 3115 bond_alb_deinitialize(bond);
3aba891d 3116 bond->recv_probe = NULL;
1da177e4
LT		 3117
3118 return 0;
3119}
3120
28172739
ED		 3121static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3122 struct rtnl_link_stats64 *stats)
1da177e4 3123{
454d7c9b 3124 struct bonding *bond = netdev_priv(bond_dev);
28172739 3125 struct rtnl_link_stats64 temp;
9caff1e7 3126 struct list_head *iter;
1da177e4 3127 struct slave *slave;
1da177e4 3128
5f0c5f73 3129 memcpy(stats, &bond->bond_stats, sizeof(*stats));
1da177e4 3130
9caff1e7 3131 bond_for_each_slave(bond, slave, iter) {
be1f3c2c 3132 const struct rtnl_link_stats64 *sstats =
28172739 3133 dev_get_stats(slave->dev, &temp);
5f0c5f73
AG		 3134 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3135
3136 stats->rx_packets += sstats->rx_packets - pstats->rx_packets;
3137 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3138 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3139 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3140
3141 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3142 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3143 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3144 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3145
3146 stats->multicast += sstats->multicast - pstats->multicast;
3147 stats->collisions += sstats->collisions - pstats->collisions;
3148
3149 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3150 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3151 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3152 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3153 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3154 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3155
3156 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3157 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3158 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3159 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3160 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3161
3162 /* save off the slave stats for the next run */
3163 memcpy(pstats, sstats, sizeof(*sstats));
3164 }
3165 memcpy(&bond->bond_stats, stats, sizeof(*stats));
1da177e4
LT		 3166
3167 return stats;
3168}
3169
3170static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3171{
080a06e1 3172 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3173 struct net_device *slave_dev = NULL;
3174 struct ifbond k_binfo;
3175 struct ifbond __user *u_binfo = NULL;
3176 struct ifslave k_sinfo;
3177 struct ifslave __user *u_sinfo = NULL;
3178 struct mii_ioctl_data *mii = NULL;
d1fbd3ed 3179 struct bond_opt_value newval;
387ff911 3180 struct net *net;
1da177e4
LT		 3181 int res = 0;
3182
76444f50 3183 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
1da177e4
LT		 3184
3185 switch (cmd) {
1da177e4
LT		 3186 case SIOCGMIIPHY:
3187 mii = if_mii(ifr);
3d632c3f 3188 if (!mii)
1da177e4 3189 return -EINVAL;
3d632c3f 3190
1da177e4
LT		 3191 mii->phy_id = 0;
3192 /* Fall Through */
3193 case SIOCGMIIREG:
547942ca 3194 /* We do this again just in case we were called by SIOCGMIIREG
1da177e4
LT		 3195 * instead of SIOCGMIIPHY.
3196 */
3197 mii = if_mii(ifr);
3d632c3f 3198 if (!mii)
1da177e4 3199 return -EINVAL;
3d632c3f 3200
1da177e4 3201 if (mii->reg_num == 1) {
1da177e4 3202 mii->val_out = 0;
3d632c3f 3203 if (netif_carrier_ok(bond->dev))
1da177e4 3204 mii->val_out = BMSR_LSTATUS;
1da177e4
LT		 3205 }
3206
3207 return 0;
3208 case BOND_INFO_QUERY_OLD:
3209 case SIOCBONDINFOQUERY:
3210 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3211
3d632c3f 3212 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
1da177e4 3213 return -EFAULT;
1da177e4
LT		 3214
3215 res = bond_info_query(bond_dev, &k_binfo);
3d632c3f
SH		 3216 if (res == 0 &&
3217 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3218 return -EFAULT;
1da177e4
LT		 3219
3220 return res;
3221 case BOND_SLAVE_INFO_QUERY_OLD:
3222 case SIOCBONDSLAVEINFOQUERY:
3223 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3224
3d632c3f 3225 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
1da177e4 3226 return -EFAULT;
1da177e4
LT		 3227
3228 res = bond_slave_info_query(bond_dev, &k_sinfo);
3d632c3f
SH		 3229 if (res == 0 &&
3230 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3231 return -EFAULT;
1da177e4
LT		 3232
3233 return res;
3234 default:
1da177e4
LT		 3235 break;
3236 }
3237
387ff911
G		 3238 net = dev_net(bond_dev);
3239
3240 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1da177e4 3241 return -EPERM;
1da177e4 3242
0917b933 3243 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
1da177e4 3244
76444f50 3245 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
1da177e4 3246
3d632c3f 3247 if (!slave_dev)
0917b933 3248 return -ENODEV;
1da177e4 3249
76444f50 3250 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
0917b933
YX		 3251 switch (cmd) {
3252 case BOND_ENSLAVE_OLD:
3253 case SIOCBONDENSLAVE:
3254 res = bond_enslave(bond_dev, slave_dev);
3255 break;
3256 case BOND_RELEASE_OLD:
3257 case SIOCBONDRELEASE:
3258 res = bond_release(bond_dev, slave_dev);
3259 break;
3260 case BOND_SETHWADDR_OLD:
3261 case SIOCBONDSETHWADDR:
3262 bond_set_dev_addr(bond_dev, slave_dev);
3263 res = 0;
3264 break;
3265 case BOND_CHANGE_ACTIVE_OLD:
3266 case SIOCBONDCHANGEACTIVE:
d1fbd3ed
NA		 3267 bond_opt_initstr(&newval, slave_dev->name);
3268 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
0917b933
YX		 3269 break;
3270 default:
3271 res = -EOPNOTSUPP;
1da177e4
LT		 3272 }
3273
1da177e4
LT		 3274 return res;
3275}
3276
d03462b9 3277static void bond_change_rx_flags(struct net_device *bond_dev, int change)
1da177e4 3278{
454d7c9b 3279 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3280
d03462b9
JP		 3281 if (change & IFF_PROMISC)
3282 bond_set_promiscuity(bond,
3283 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3d632c3f 3284
d03462b9
JP		 3285 if (change & IFF_ALLMULTI)
3286 bond_set_allmulti(bond,
3287 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3288}
1da177e4 3289
303d1cbf 3290static void bond_set_rx_mode(struct net_device *bond_dev)
d03462b9
JP		 3291{
3292 struct bonding *bond = netdev_priv(bond_dev);
9caff1e7 3293 struct list_head *iter;
303d1cbf 3294 struct slave *slave;
1da177e4 3295
b3241870 3296 rcu_read_lock();
ec0865a9 3297 if (bond_uses_primary(bond)) {
b3241870 3298 slave = rcu_dereference(bond->curr_active_slave);
303d1cbf
JV		 3299 if (slave) {
3300 dev_uc_sync(slave->dev, bond_dev);
3301 dev_mc_sync(slave->dev, bond_dev);
3302 }
303d1cbf 3303 } else {
b3241870 3304 bond_for_each_slave_rcu(bond, slave, iter) {
303d1cbf
JV		 3305 dev_uc_sync_multiple(slave->dev, bond_dev);
3306 dev_mc_sync_multiple(slave->dev, bond_dev);
3307 }
1da177e4 3308 }
b3241870 3309 rcu_read_unlock();
1da177e4
LT		 3310}
3311
234bcf8a 3312static int bond_neigh_init(struct neighbour *n)
00829823 3313{
234bcf8a 3314 struct bonding *bond = netdev_priv(n->dev);
234bcf8a
SP		 3315 const struct net_device_ops *slave_ops;
3316 struct neigh_parms parms;
dec1e90e 3317 struct slave *slave;
234bcf8a
SP		 3318 int ret;
3319
dec1e90e 3320 slave = bond_first_slave(bond);
234bcf8a
SP		 3321 if (!slave)
3322 return 0;
234bcf8a 3323 slave_ops = slave->dev->netdev_ops;
234bcf8a
SP		 3324 if (!slave_ops->ndo_neigh_setup)
3325 return 0;
3326
3327 parms.neigh_setup = NULL;
3328 parms.neigh_cleanup = NULL;
3329 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3330 if (ret)
3331 return ret;
3332
547942ca 3333 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
234bcf8a
SP		 3334 * after the last slave has been detached. Assumes that all slaves
3335 * utilize the same neigh_cleanup (true at this writing as only user
3336 * is ipoib).
3337 */
3338 n->parms->neigh_cleanup = parms.neigh_cleanup;
3339
3340 if (!parms.neigh_setup)
3341 return 0;
3342
3343 return parms.neigh_setup(n);
3344}
3345
547942ca 3346/* The bonding ndo_neigh_setup is called at init time beofre any
234bcf8a
SP		 3347 * slave exists. So we must declare proxy setup function which will
3348 * be used at run time to resolve the actual slave neigh param setup.
9918d5bf
VF		 3349 *
3350 * It's also called by master devices (such as vlans) to setup their
3351 * underlying devices. In that case - do nothing, we're already set up from
3352 * our init.
234bcf8a
SP		 3353 */
3354static int bond_neigh_setup(struct net_device *dev,
3355 struct neigh_parms *parms)
3356{
9918d5bf
VF		 3357 /* modify only our neigh_parms */
3358 if (parms->dev == dev)
3359 parms->neigh_setup = bond_neigh_init;
00829823 3360
00829823
SH		 3361 return 0;
3362}
3363
547942ca 3364/* Change the MTU of all of a master's slaves to match the master */
1da177e4
LT		 3365static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3366{
454d7c9b 3367 struct bonding *bond = netdev_priv(bond_dev);
81f23b13 3368 struct slave *slave, *rollback_slave;
9caff1e7 3369 struct list_head *iter;
1da177e4 3370 int res = 0;
1da177e4 3371
76444f50 3372 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
1da177e4 3373
9caff1e7 3374 bond_for_each_slave(bond, slave, iter) {
76444f50
VF		 3375 netdev_dbg(bond_dev, "s %p c_m %p\n",
3376 slave, slave->dev->netdev_ops->ndo_change_mtu);
e944ef79 3377
1da177e4
LT		 3378 res = dev_set_mtu(slave->dev, new_mtu);
3379
3380 if (res) {
3381 /* If we failed to set the slave's mtu to the new value
3382 * we must abort the operation even in ACTIVE_BACKUP
3383 * mode, because if we allow the backup slaves to have
3384 * different mtu values than the active slave we'll
3385 * need to change their mtu when doing a failover. That
3386 * means changing their mtu from timer context, which
3387 * is probably not a good idea.
3388 */
76444f50
VF		 3389 netdev_dbg(bond_dev, "err %d %s\n", res,
3390 slave->dev->name);
1da177e4
LT		 3391 goto unwind;
3392 }
3393 }
3394
3395 bond_dev->mtu = new_mtu;
3396
3397 return 0;
3398
3399unwind:
3400 /* unwind from head to the slave that failed */
9caff1e7 3401 bond_for_each_slave(bond, rollback_slave, iter) {
1da177e4
LT		 3402 int tmp_res;
3403
81f23b13
VF		 3404 if (rollback_slave == slave)
3405 break;
3406
3407 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
1da177e4 3408 if (tmp_res) {
76444f50
VF		 3409 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3410 tmp_res, rollback_slave->dev->name);
1da177e4
LT		 3411 }
3412 }
3413
3414 return res;
3415}
3416
547942ca 3417/* Change HW address
1da177e4
LT		 3418 *
3419 * Note that many devices must be down to change the HW address, and
3420 * downing the master releases all slaves. We can make bonds full of
3421 * bonding devices to test this, however.
3422 */
3423static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3424{
454d7c9b 3425 struct bonding *bond = netdev_priv(bond_dev);
81f23b13 3426 struct slave *slave, *rollback_slave;
1da177e4 3427 struct sockaddr *sa = addr, tmp_sa;
9caff1e7 3428 struct list_head *iter;
1da177e4 3429 int res = 0;
1da177e4 3430
01844098 3431 if (BOND_MODE(bond) == BOND_MODE_ALB)
eb7cc59a
SH		 3432 return bond_alb_set_mac_address(bond_dev, addr);
3433
3434
76444f50 3435 netdev_dbg(bond_dev, "bond=%p\n", bond);
1da177e4 3436
1b5acd29
JV		 3437 /* If fail_over_mac is enabled, do nothing and return success.
3438 * Returning an error causes ifenslave to fail.
dd957c57 3439 */
cc689aaa 3440 if (bond->params.fail_over_mac &&
01844098 3441 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
dd957c57 3442 return 0;
2ab82852 3443
3d632c3f 3444 if (!is_valid_ether_addr(sa->sa_data))
1da177e4 3445 return -EADDRNOTAVAIL;
1da177e4 3446
9caff1e7 3447 bond_for_each_slave(bond, slave, iter) {
76444f50 3448 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
1da177e4
LT		 3449 res = dev_set_mac_address(slave->dev, addr);
3450 if (res) {
3451 /* TODO: consider downing the slave
3452 * and retry ?
3453 * User should expect communications
3454 * breakage anyway until ARP finish
3455 * updating, so...
3456 */
76444f50 3457 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
1da177e4
LT		 3458 goto unwind;
3459 }
3460 }
3461
3462 /* success */
3463 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3464 return 0;
3465
3466unwind:
3467 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3468 tmp_sa.sa_family = bond_dev->type;
3469
3470 /* unwind from head to the slave that failed */
9caff1e7 3471 bond_for_each_slave(bond, rollback_slave, iter) {
1da177e4
LT		 3472 int tmp_res;
3473
81f23b13
VF		 3474 if (rollback_slave == slave)
3475 break;
3476
3477 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
1da177e4 3478 if (tmp_res) {
76444f50
VF		 3479 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3480 tmp_res, rollback_slave->dev->name);
1da177e4
LT		 3481 }
3482 }
3483
3484 return res;
3485}
3486
15077228
NA		 3487/**
3488 * bond_xmit_slave_id - transmit skb through slave with slave_id
3489 * @bond: bonding device that is transmitting
3490 * @skb: buffer to transmit
3491 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3492 *
3493 * This function tries to transmit through slave with slave_id but in case
3494 * it fails, it tries to find the first available slave for transmission.
3495 * The skb is consumed in all cases, thus the function is void.
3496 */
da131ddb 3497static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
15077228 3498{
9caff1e7 3499 struct list_head *iter;
15077228
NA		 3500 struct slave *slave;
3501 int i = slave_id;
3502
3503 /* Here we start from the slave with slave_id */
9caff1e7 3504 bond_for_each_slave_rcu(bond, slave, iter) {
15077228 3505 if (--i < 0) {
891ab54d 3506 if (bond_slave_can_tx(slave)) {
15077228
NA		 3507 bond_dev_queue_xmit(bond, skb, slave->dev);
3508 return;
3509 }
3510 }
3511 }
3512
3513 /* Here we start from the first slave up to slave_id */
3514 i = slave_id;
9caff1e7 3515 bond_for_each_slave_rcu(bond, slave, iter) {
15077228
NA		 3516 if (--i < 0)
3517 break;
891ab54d 3518 if (bond_slave_can_tx(slave)) {
15077228
NA		 3519 bond_dev_queue_xmit(bond, skb, slave->dev);
3520 return;
3521 }
3522 }
3523 /* no slave that can tx has been found */
31aa860e 3524 bond_tx_drop(bond->dev, skb);
15077228
NA		 3525}
3526
73958329
NA		 3527/**
3528 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3529 * @bond: bonding device to use
3530 *
3531 * Based on the value of the bonding device's packets_per_slave parameter
3532 * this function generates a slave id, which is usually used as the next
3533 * slave to transmit through.
3534 */
3535static u32 bond_rr_gen_slave_id(struct bonding *bond)
3536{
73958329 3537 u32 slave_id;
809fa972
HFS		 3538 struct reciprocal_value reciprocal_packets_per_slave;
3539 int packets_per_slave = bond->params.packets_per_slave;
73958329
NA		 3540
3541 switch (packets_per_slave) {
3542 case 0:
3543 slave_id = prandom_u32();
3544 break;
3545 case 1:
3546 slave_id = bond->rr_tx_counter;
3547 break;
3548 default:
809fa972
HFS		 3549 reciprocal_packets_per_slave =
3550 bond->params.reciprocal_packets_per_slave;
73958329 3551 slave_id = reciprocal_divide(bond->rr_tx_counter,
809fa972 3552 reciprocal_packets_per_slave);
73958329
NA		 3553 break;
3554 }
3555 bond->rr_tx_counter++;
3556
3557 return slave_id;
3558}
3559
1da177e4
LT		 3560static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3561{
454d7c9b 3562 struct bonding *bond = netdev_priv(bond_dev);
a2fd940f 3563 struct iphdr *iph = ip_hdr(skb);
15077228 3564 struct slave *slave;
73958329 3565 u32 slave_id;
1da177e4 3566
73958329 3567 /* Start with the curr_active_slave that joined the bond as the
a2fd940f
AG		 3568 * default for sending IGMP traffic. For failover purposes one
3569 * needs to maintain some consistency for the interface that will
3570 * send the join/membership reports. The curr_active_slave found
3571 * will send all of this type of traffic.
cf5f9044 3572 */
15077228 3573 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
278b2083 3574 slave = rcu_dereference(bond->curr_active_slave);
a64d044e 3575 if (slave)
15077228
NA		 3576 bond_dev_queue_xmit(bond, skb, slave->dev);
3577 else
3578 bond_xmit_slave_id(bond, skb, 0);
a2fd940f 3579 } else {
9a72c2da
NA		 3580 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3581
3582 if (likely(slave_cnt)) {
3583 slave_id = bond_rr_gen_slave_id(bond);
3584 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3585 } else {
31aa860e 3586 bond_tx_drop(bond_dev, skb);
9a72c2da 3587 }
1da177e4 3588 }
0693e88e 3589
ec634fe3 3590 return NETDEV_TX_OK;
1da177e4
LT		 3591}
3592
547942ca 3593/* In active-backup mode, we know that bond->curr_active_slave is always valid if
1da177e4
LT		 3594 * the bond has a usable interface.
3595 */
3596static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3597{
454d7c9b 3598 struct bonding *bond = netdev_priv(bond_dev);
71bc3b2d 3599 struct slave *slave;
1da177e4 3600
278b2083 3601 slave = rcu_dereference(bond->curr_active_slave);
71bc3b2d 3602 if (slave)
15077228
NA		 3603 bond_dev_queue_xmit(bond, skb, slave->dev);
3604 else
31aa860e 3605 bond_tx_drop(bond_dev, skb);
0693e88e 3606
ec634fe3 3607 return NETDEV_TX_OK;
1da177e4
LT		 3608}
3609
ee637714
MB		 3610/* Use this to update slave_array when (a) it's not appropriate to update
3611 * slave_array right away (note that update_slave_array() may sleep)
3612 * and / or (b) RTNL is not held.
1da177e4 3613 */
ee637714 3614void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
1da177e4 3615{
ee637714
MB		 3616 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3617}
1da177e4 3618
ee637714
MB		 3619/* Slave array work handler. Holds only RTNL */
3620static void bond_slave_arr_handler(struct work_struct *work)
3621{
3622 struct bonding *bond = container_of(work, struct bonding,
3623 slave_arr_work.work);
3624 int ret;
3625
3626 if (!rtnl_trylock())
3627 goto err;
3628
3629 ret = bond_update_slave_arr(bond, NULL);
3630 rtnl_unlock();
3631 if (ret) {
3632 pr_warn_ratelimited("Failed to update slave array from WT\n");
3633 goto err;
3634 }
3635 return;
3636
3637err:
3638 bond_slave_arr_work_rearm(bond, 1);
3639}
3640
3641/* Build the usable slaves array in control path for modes that use xmit-hash
3642 * to determine the slave interface -
3643 * (a) BOND_MODE_8023AD
3644 * (b) BOND_MODE_XOR
3645 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3646 *
3647 * The caller is expected to hold RTNL only and NO other lock!
3648 */
3649int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3650{
3651 struct slave *slave;
3652 struct list_head *iter;
3653 struct bond_up_slave *new_arr, *old_arr;
3654 int slaves_in_agg;
3655 int agg_id = 0;
3656 int ret = 0;
3657
3658#ifdef CONFIG_LOCKDEP
3659 WARN_ON(lockdep_is_held(&bond->mode_lock));
3660#endif
3661
3662 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3663 GFP_KERNEL);
3664 if (!new_arr) {
3665 ret = -ENOMEM;
3666 pr_err("Failed to build slave-array.\n");
3667 goto out;
3668 }
3669 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3670 struct ad_info ad_info;
3671
3672 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3673 pr_debug("bond_3ad_get_active_agg_info failed\n");
3674 kfree_rcu(new_arr, rcu);
3675 /* No active aggragator means it's not safe to use
3676 * the previous array.
3677 */
3678 old_arr = rtnl_dereference(bond->slave_arr);
3679 if (old_arr) {
3680 RCU_INIT_POINTER(bond->slave_arr, NULL);
3681 kfree_rcu(old_arr, rcu);
3682 }
3683 goto out;
3684 }
3685 slaves_in_agg = ad_info.ports;
3686 agg_id = ad_info.aggregator_id;
3687 }
3688 bond_for_each_slave(bond, slave, iter) {
3689 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3690 struct aggregator *agg;
3691
3692 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3693 if (!agg || agg->aggregator_identifier != agg_id)
3694 continue;
3695 }
3696 if (!bond_slave_can_tx(slave))
3697 continue;
3698 if (skipslave == slave)
3699 continue;
3700 new_arr->arr[new_arr->count++] = slave;
3701 }
3702
3703 old_arr = rtnl_dereference(bond->slave_arr);
3704 rcu_assign_pointer(bond->slave_arr, new_arr);
3705 if (old_arr)
3706 kfree_rcu(old_arr, rcu);
3707out:
3708 if (ret != 0 && skipslave) {
3709 int idx;
3710
3711 /* Rare situation where caller has asked to skip a specific
3712 * slave but allocation failed (most likely!). BTW this is
3713 * only possible when the call is initiated from
3714 * __bond_release_one(). In this situation; overwrite the
3715 * skipslave entry in the array with the last entry from the
3716 * array to avoid a situation where the xmit path may choose
3717 * this to-be-skipped slave to send a packet out.
3718 */
3719 old_arr = rtnl_dereference(bond->slave_arr);
3720 for (idx = 0; idx < old_arr->count; idx++) {
3721 if (skipslave == old_arr->arr[idx]) {
3722 old_arr->arr[idx] =
3723 old_arr->arr[old_arr->count-1];
3724 old_arr->count--;
3725 break;
3726 }
3727 }
3728 }
3729 return ret;
3730}
3731
3732/* Use this Xmit function for 3AD as well as XOR modes. The current
3733 * usable slave array is formed in the control path. The xmit function
3734 * just calculates hash and sends the packet out.
3735 */
3736int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3737{
3738 struct bonding *bond = netdev_priv(dev);
3739 struct slave *slave;
3740 struct bond_up_slave *slaves;
3741 unsigned int count;
3742
3743 slaves = rcu_dereference(bond->slave_arr);
3744 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3745 if (likely(count)) {
3746 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3747 bond_dev_queue_xmit(bond, skb, slave->dev);
3748 } else {
31aa860e 3749 bond_tx_drop(dev, skb);
ee637714 3750 }
0693e88e 3751
ec634fe3 3752 return NETDEV_TX_OK;
1da177e4
LT		 3753}
3754
78a646ce 3755/* in broadcast mode, we send everything to all usable interfaces. */
1da177e4
LT		 3756static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3757{
454d7c9b 3758 struct bonding *bond = netdev_priv(bond_dev);
78a646ce 3759 struct slave *slave = NULL;
9caff1e7 3760 struct list_head *iter;
1da177e4 3761
9caff1e7 3762 bond_for_each_slave_rcu(bond, slave, iter) {
78a646ce
NA		 3763 if (bond_is_last_slave(bond, slave))
3764 break;
b6adc610 3765 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
78a646ce 3766 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1da177e4 3767
78a646ce 3768 if (!skb2) {
9152e26d 3769 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3770 bond_dev->name, __func__);
78a646ce 3771 continue;
1da177e4 3772 }
78a646ce 3773 bond_dev_queue_xmit(bond, skb2, slave->dev);
1da177e4
LT		 3774 }
3775 }
b6adc610 3776 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
78a646ce
NA		 3777 bond_dev_queue_xmit(bond, skb, slave->dev);
3778 else
31aa860e 3779 bond_tx_drop(bond_dev, skb);
3d632c3f 3780
ec634fe3 3781 return NETDEV_TX_OK;
1da177e4
LT		 3782}
3783
3784/*------------------------- Device initialization ---------------------------*/
3785
547942ca 3786/* Lookup the slave that corresponds to a qid */
bb1d9123
AG		 3787static inline int bond_slave_override(struct bonding *bond,
3788 struct sk_buff *skb)
3789{
bb1d9123 3790 struct slave *slave = NULL;
9caff1e7 3791 struct list_head *iter;
bb1d9123 3792
0693e88e
MM		 3793 if (!skb->queue_mapping)
3794 return 1;
bb1d9123
AG		 3795
3796 /* Find out if any slaves have the same mapping as this skb. */
3900f290 3797 bond_for_each_slave_rcu(bond, slave, iter) {
3798 if (slave->queue_id == skb->queue_mapping) {
891ab54d 3799 if (bond_slave_can_tx(slave)) {
3900f290 3800 bond_dev_queue_xmit(bond, skb, slave->dev);
3801 return 0;
3802 }
3803 /* If the slave isn't UP, use default transmit policy. */
bb1d9123
AG		 3804 break;
3805 }
3806 }
3807
3900f290 3808 return 1;
bb1d9123
AG		 3809}
3810
374eeb5a 3811
f663dd9a 3812static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
99932d4f 3813 void *accel_priv, select_queue_fallback_t fallback)
bb1d9123 3814{
547942ca 3815 /* This helper function exists to help dev_pick_tx get the correct
fd0e435b 3816 * destination queue. Using a helper function skips a call to
bb1d9123
AG		 3817 * skb_tx_hash and will put the skbs in the queue we expect on their
3818 * way down to the bonding driver.
3819 */
fd0e435b
PO		 3820 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3821
547942ca 3822 /* Save the original txq to restore before passing to the driver */
df4ab5b3 3823 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
374eeb5a 3824
fd0e435b 3825 if (unlikely(txq >= dev->real_num_tx_queues)) {
d30ee670 3826 do {
fd0e435b 3827 txq -= dev->real_num_tx_queues;
d30ee670 3828 } while (txq >= dev->real_num_tx_queues);
fd0e435b
PO		 3829 }
3830 return txq;
bb1d9123
AG		 3831}
3832
0693e88e 3833static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
00829823 3834{
bb1d9123
AG		 3835 struct bonding *bond = netdev_priv(dev);
3836
d1e2e5cd
VF		 3837 if (bond_should_override_tx_queue(bond) &&
3838 !bond_slave_override(bond, skb))
3839 return NETDEV_TX_OK;
00829823 3840
01844098 3841 switch (BOND_MODE(bond)) {
00829823
SH		 3842 case BOND_MODE_ROUNDROBIN:
3843 return bond_xmit_roundrobin(skb, dev);
3844 case BOND_MODE_ACTIVEBACKUP:
3845 return bond_xmit_activebackup(skb, dev);
ee637714 3846 case BOND_MODE_8023AD:
00829823 3847 case BOND_MODE_XOR:
ee637714 3848 return bond_3ad_xor_xmit(skb, dev);
00829823
SH		 3849 case BOND_MODE_BROADCAST:
3850 return bond_xmit_broadcast(skb, dev);
00829823 3851 case BOND_MODE_ALB:
00829823 3852 return bond_alb_xmit(skb, dev);
f05b42ea
MB		 3853 case BOND_MODE_TLB:
3854 return bond_tlb_xmit(skb, dev);
00829823
SH		 3855 default:
3856 /* Should never happen, mode already checked */
76444f50 3857 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
00829823 3858 WARN_ON_ONCE(1);
31aa860e 3859 bond_tx_drop(dev, skb);
00829823
SH		 3860 return NETDEV_TX_OK;
3861 }
3862}
3863
0693e88e
MM		 3864static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3865{
3866 struct bonding *bond = netdev_priv(dev);
3867 netdev_tx_t ret = NETDEV_TX_OK;
3868
547942ca 3869 /* If we risk deadlock from transmitting this in the
0693e88e
MM		 3870 * netpoll path, tell netpoll to queue the frame for later tx
3871 */
054bb880 3872 if (unlikely(is_netpoll_tx_blocked(dev)))
0693e88e
MM		 3873 return NETDEV_TX_BUSY;
3874
278b2083 3875 rcu_read_lock();
0965a1f3 3876 if (bond_has_slaves(bond))
0693e88e
MM		 3877 ret = __bond_start_xmit(skb, dev);
3878 else
31aa860e 3879 bond_tx_drop(dev, skb);
278b2083 3880 rcu_read_unlock();
0693e88e
MM		 3881
3882 return ret;
3883}
00829823 3884
bb5b052f
AG		 3885static int bond_ethtool_get_settings(struct net_device *bond_dev,
3886 struct ethtool_cmd *ecmd)
3887{
3888 struct bonding *bond = netdev_priv(bond_dev);
bb5b052f 3889 unsigned long speed = 0;
9caff1e7 3890 struct list_head *iter;
dec1e90e 3891 struct slave *slave;
bb5b052f
AG		 3892
3893 ecmd->duplex = DUPLEX_UNKNOWN;
3894 ecmd->port = PORT_OTHER;
3895
8557cd74 3896 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
bb5b052f
AG		 3897 * do not need to check mode. Though link speed might not represent
3898 * the true receive or transmit bandwidth (not all modes are symmetric)
3899 * this is an accurate maximum.
3900 */
9caff1e7 3901 bond_for_each_slave(bond, slave, iter) {
8557cd74 3902 if (bond_slave_can_tx(slave)) {
bb5b052f
AG		 3903 if (slave->speed != SPEED_UNKNOWN)
3904 speed += slave->speed;
3905 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3906 slave->duplex != DUPLEX_UNKNOWN)
3907 ecmd->duplex = slave->duplex;
3908 }
3909 }
3910 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
dec1e90e 3911
bb5b052f
AG		 3912 return 0;
3913}
3914
217df670 3915static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
7826d43f 3916 struct ethtool_drvinfo *drvinfo)
217df670 3917{
7826d43f
JP		 3918 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3919 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3920 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3921 BOND_ABI_VERSION);
217df670
JV		 3922}
3923
7282d491 3924static const struct ethtool_ops bond_ethtool_ops = {
217df670 3925 .get_drvinfo = bond_ethtool_get_drvinfo,
bb5b052f 3926 .get_settings = bond_ethtool_get_settings,
fa53ebac 3927 .get_link = ethtool_op_get_link,
8531c5ff
AK		 3928};
3929
eb7cc59a 3930static const struct net_device_ops bond_netdev_ops = {
181470fc 3931 .ndo_init = bond_init,
9e71626c 3932 .ndo_uninit = bond_uninit,
eb7cc59a
SH		 3933 .ndo_open = bond_open,
3934 .ndo_stop = bond_close,
00829823 3935 .ndo_start_xmit = bond_start_xmit,
bb1d9123 3936 .ndo_select_queue = bond_select_queue,
be1f3c2c 3937 .ndo_get_stats64 = bond_get_stats,
eb7cc59a 3938 .ndo_do_ioctl = bond_do_ioctl,
d03462b9 3939 .ndo_change_rx_flags = bond_change_rx_flags,
303d1cbf 3940 .ndo_set_rx_mode = bond_set_rx_mode,
eb7cc59a 3941 .ndo_change_mtu = bond_change_mtu,
cc0e4070 3942 .ndo_set_mac_address = bond_set_mac_address,
00829823 3943 .ndo_neigh_setup = bond_neigh_setup,
cc0e4070 3944 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
eb7cc59a 3945 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
f6dc31a8 3946#ifdef CONFIG_NET_POLL_CONTROLLER
8a8efa22 3947 .ndo_netpoll_setup = bond_netpoll_setup,
f6dc31a8
WC		 3948 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3949 .ndo_poll_controller = bond_poll_controller,
3950#endif
9232ecca
JP		 3951 .ndo_add_slave = bond_enslave,
3952 .ndo_del_slave = bond_release,
b2a103e6 3953 .ndo_fix_features = bond_fix_features,
eb7cc59a
SH		 3954};
3955
b3f92b63
DG		 3956static const struct device_type bond_type = {
3957 .name = "bond",
3958};
3959
9e2e61fb
AW		 3960static void bond_destructor(struct net_device *bond_dev)
3961{
3962 struct bonding *bond = netdev_priv(bond_dev);
3963 if (bond->wq)
3964 destroy_workqueue(bond->wq);
3965 free_netdev(bond_dev);
3966}
3967
0a2a78c4 3968void bond_setup(struct net_device *bond_dev)
1da177e4 3969{
454d7c9b 3970 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3971
b7435628 3972 spin_lock_init(&bond->mode_lock);
d2991f75 3973 bond->params = bonding_defaults;
1da177e4
LT		 3974
3975 /* Initialize pointers */
1da177e4 3976 bond->dev = bond_dev;
1da177e4
LT		 3977
3978 /* Initialize the device entry points */
181470fc 3979 ether_setup(bond_dev);
eb7cc59a 3980 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 3981 bond_dev->ethtool_ops = &bond_ethtool_ops;
1da177e4 3982
9e2e61fb 3983 bond_dev->destructor = bond_destructor;
1da177e4 3984
b3f92b63
DG		 3985 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3986
1da177e4
LT		 3987 /* Initialize the device options */
3988 bond_dev->tx_queue_len = 0;
3989 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
c565b488 3990 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
550fd08c 3991 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
181470fc 3992
547942ca 3993 /* don't acquire bond device's netif_tx_lock when transmitting */
1da177e4
LT		 3994 bond_dev->features |= NETIF_F_LLTX;
3995
3996 /* By default, we declare the bond to be fully
3997 * VLAN hardware accelerated capable. Special
3998 * care is taken in the various xmit functions
3999 * when there are slaves that are not hw accel
4000 * capable
4001 */
1da177e4 4002
f9399814
WC		 4003 /* Don't allow bond devices to change network namespaces. */
4004 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4005
b2a103e6 4006 bond_dev->hw_features = BOND_VLAN_FEATURES |
f646968f
PM		 4007 NETIF_F_HW_VLAN_CTAG_TX |
4008 NETIF_F_HW_VLAN_CTAG_RX |
4009 NETIF_F_HW_VLAN_CTAG_FILTER;
b2a103e6 4010
34324dc2 4011 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
5a7baa78 4012 bond_dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
b2a103e6 4013 bond_dev->features |= bond_dev->hw_features;
1da177e4
LT		 4014}
4015
547942ca
NA		 4016/* Destroy a bonding device.
4017 * Must be under rtnl_lock when this function is called.
4018 */
c67dfb29 4019static void bond_uninit(struct net_device *bond_dev)
a434e43f 4020{
454d7c9b 4021 struct bonding *bond = netdev_priv(bond_dev);
544a028e
VF		 4022 struct list_head *iter;
4023 struct slave *slave;
ee637714 4024 struct bond_up_slave *arr;
a434e43f 4025
f6dc31a8
WC		 4026 bond_netpoll_cleanup(bond_dev);
4027
c67dfb29 4028 /* Release the bonded slaves */
544a028e 4029 bond_for_each_slave(bond, slave, iter)
dec1e90e 4030 __bond_release_one(bond_dev, slave->dev, true);
76444f50 4031 netdev_info(bond_dev, "Released all slaves\n");
c67dfb29 4032
ee637714
MB		 4033 arr = rtnl_dereference(bond->slave_arr);
4034 if (arr) {
4035 RCU_INIT_POINTER(bond->slave_arr, NULL);
4036 kfree_rcu(arr, rcu);
4037 }
4038
a434e43f
JV		 4039 list_del(&bond->bond_list);
4040
f073c7ca 4041 bond_debug_unregister(bond);
a434e43f
JV		 4042}
4043
1da177e4
LT		 4044/*------------------------- Module initialization ---------------------------*/
4045
1da177e4
LT		 4046static int bond_check_params(struct bond_params *params)
4047{
5a5c5fd4 4048 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
f3253339 4049 struct bond_opt_value newval;
4050 const struct bond_opt_value *valptr;
8599b52e 4051 int arp_all_targets_value;
f5b2b966 4052
547942ca 4053 /* Convert string parameters. */
1da177e4 4054 if (mode) {
2b3798d5
NA		 4055 bond_opt_initstr(&newval, mode);
4056 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4057 if (!valptr) {
4058 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
1da177e4
LT		 4059 return -EINVAL;
4060 }
2b3798d5 4061 bond_mode = valptr->value;
1da177e4
LT		 4062 }
4063
169a3e66
JV		 4064 if (xmit_hash_policy) {
4065 if ((bond_mode != BOND_MODE_XOR) &&
f05b42ea
MB		 4066 (bond_mode != BOND_MODE_8023AD) &&
4067 (bond_mode != BOND_MODE_TLB)) {
a4aee5c8 4068 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
90194264 4069 bond_mode_name(bond_mode));
169a3e66 4070 } else {
a4b32ce7
NA		 4071 bond_opt_initstr(&newval, xmit_hash_policy);
4072 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4073 &newval);
4074 if (!valptr) {
a4aee5c8 4075 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
169a3e66
JV		 4076 xmit_hash_policy);
4077 return -EINVAL;
4078 }
a4b32ce7 4079 xmit_hashtype = valptr->value;
169a3e66
JV		 4080 }
4081 }
4082
1da177e4
LT		 4083 if (lacp_rate) {
4084 if (bond_mode != BOND_MODE_8023AD) {
a4aee5c8
JP		 4085 pr_info("lacp_rate param is irrelevant in mode %s\n",
4086 bond_mode_name(bond_mode));
1da177e4 4087 } else {
d3131de7
NA		 4088 bond_opt_initstr(&newval, lacp_rate);
4089 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4090 &newval);
4091 if (!valptr) {
a4aee5c8 4092 pr_err("Error: Invalid lacp rate \"%s\"\n",
d3131de7 4093 lacp_rate);
1da177e4
LT		 4094 return -EINVAL;
4095 }
d3131de7 4096 lacp_fast = valptr->value;
1da177e4
LT		 4097 }
4098 }
4099
fd989c83 4100 if (ad_select) {
548d28bd 4101 bond_opt_initstr(&newval, ad_select);
9e5f5eeb
NA		 4102 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4103 &newval);
4104 if (!valptr) {
4105 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
fd989c83
JV		 4106 return -EINVAL;
4107 }
9e5f5eeb
NA		 4108 params->ad_select = valptr->value;
4109 if (bond_mode != BOND_MODE_8023AD)
91565ebb 4110 pr_warn("ad_select param only affects 802.3ad mode\n");
fd989c83
JV		 4111 } else {
4112 params->ad_select = BOND_AD_STABLE;
4113 }
4114
f5841306 4115 if (max_bonds < 0) {
91565ebb
JP		 4116 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4117 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4118 max_bonds = BOND_DEFAULT_MAX_BONDS;
4119 }
4120
4121 if (miimon < 0) {
91565ebb
JP		 4122 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4123 miimon, INT_MAX);
b98d9c66 4124 miimon = 0;
1da177e4
LT		 4125 }
4126
4127 if (updelay < 0) {
91565ebb
JP		 4128 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4129 updelay, INT_MAX);
1da177e4
LT		 4130 updelay = 0;
4131 }
4132
4133 if (downdelay < 0) {
91565ebb
JP		 4134 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4135 downdelay, INT_MAX);
1da177e4
LT		 4136 downdelay = 0;
4137 }
4138
4139 if ((use_carrier != 0) && (use_carrier != 1)) {
91565ebb
JP		 4140 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4141 use_carrier);
1da177e4
LT		 4142 use_carrier = 1;
4143 }
4144
ad246c99 4145 if (num_peer_notif < 0 || num_peer_notif > 255) {
91565ebb
JP		 4146 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4147 num_peer_notif);
ad246c99
BH		 4148 num_peer_notif = 1;
4149 }
4150
834db4bc 4151 /* reset values for 802.3ad/TLB/ALB */
267bed77 4152 if (!bond_mode_uses_arp(bond_mode)) {
1da177e4 4153 if (!miimon) {
91565ebb
JP		 4154 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4155 pr_warn("Forcing miimon to 100msec\n");
fe9d04af 4156 miimon = BOND_DEFAULT_MIIMON;
1da177e4
LT		 4157 }
4158 }
4159
bb1d9123 4160 if (tx_queues < 1 || tx_queues > 255) {
91565ebb
JP		 4161 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4162 tx_queues, BOND_DEFAULT_TX_QUEUES);
bb1d9123
AG		 4163 tx_queues = BOND_DEFAULT_TX_QUEUES;
4164 }
4165
ebd8e497 4166 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
91565ebb
JP		 4167 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4168 all_slaves_active);
ebd8e497
AG		 4169 all_slaves_active = 0;
4170 }
4171
c2952c31 4172 if (resend_igmp < 0 || resend_igmp > 255) {
91565ebb
JP		 4173 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4174 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
c2952c31
FL		 4175 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4176 }
4177
aa59d851
NA		 4178 bond_opt_initval(&newval, packets_per_slave);
4179 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
73958329
NA		 4180 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4181 packets_per_slave, USHRT_MAX);
4182 packets_per_slave = 1;
4183 }
4184
1da177e4 4185 if (bond_mode == BOND_MODE_ALB) {
a4aee5c8
JP		 4186 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",
4187 updelay);
1da177e4
LT		 4188 }
4189
4190 if (!miimon) {
4191 if (updelay || downdelay) {
4192 /* just warn the user the up/down delay will have
4193 * no effect since miimon is zero...
4194 */
91565ebb
JP		 4195 pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4196 updelay, downdelay);
1da177e4
LT		 4197 }
4198 } else {
4199 /* don't allow arp monitoring */
4200 if (arp_interval) {
91565ebb
JP		 4201 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4202 miimon, arp_interval);
1da177e4
LT		 4203 arp_interval = 0;
4204 }
4205
4206 if ((updelay % miimon) != 0) {
91565ebb
JP		 4207 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4208 updelay, miimon, (updelay / miimon) * miimon);
1da177e4
LT		 4209 }
4210
4211 updelay /= miimon;
4212
4213 if ((downdelay % miimon) != 0) {
91565ebb
JP		 4214 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4215 downdelay, miimon,
4216 (downdelay / miimon) * miimon);
1da177e4
LT		 4217 }
4218
4219 downdelay /= miimon;
4220 }
4221
4222 if (arp_interval < 0) {
91565ebb
JP		 4223 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4224 arp_interval, INT_MAX);
7bdb04ed 4225 arp_interval = 0;
1da177e4
LT		 4226 }
4227
5a5c5fd4 4228 for (arp_ip_count = 0, i = 0;
4229 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
89015c18 4230 __be32 ip;
547942ca
NA		 4231
4232 /* not a complete check, but good enough to catch mistakes */
89015c18 4233 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
2807a9fe 4234 !bond_is_ip_target_ok(ip)) {
91565ebb
JP		 4235 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4236 arp_ip_target[i]);
1da177e4
LT		 4237 arp_interval = 0;
4238 } else {
0afee4e8
VF		 4239 if (bond_get_targets_ip(arp_target, ip) == -1)
4240 arp_target[arp_ip_count++] = ip;
4241 else
91565ebb
JP		 4242 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4243 &ip);
1da177e4
LT		 4244 }
4245 }
4246
4247 if (arp_interval && !arp_ip_count) {
4248 /* don't allow arping if no arp_ip_target given... */
91565ebb
JP		 4249 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4250 arp_interval);
1da177e4
LT		 4251 arp_interval = 0;
4252 }
4253
f5b2b966 4254 if (arp_validate) {
f5b2b966 4255 if (!arp_interval) {
a4aee5c8 4256 pr_err("arp_validate requires arp_interval\n");
f5b2b966
JV		 4257 return -EINVAL;
4258 }
4259
16228881
NA		 4260 bond_opt_initstr(&newval, arp_validate);
4261 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4262 &newval);
4263 if (!valptr) {
a4aee5c8 4264 pr_err("Error: invalid arp_validate \"%s\"\n",
16228881 4265 arp_validate);
f5b2b966
JV		 4266 return -EINVAL;
4267 }
16228881
NA		 4268 arp_validate_value = valptr->value;
4269 } else {
f5b2b966 4270 arp_validate_value = 0;
16228881 4271 }
f5b2b966 4272
8599b52e
VF		 4273 arp_all_targets_value = 0;
4274 if (arp_all_targets) {
edf36b24
NA		 4275 bond_opt_initstr(&newval, arp_all_targets);
4276 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4277 &newval);
4278 if (!valptr) {
8599b52e
VF		 4279 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4280 arp_all_targets);
4281 arp_all_targets_value = 0;
edf36b24
NA		 4282 } else {
4283 arp_all_targets_value = valptr->value;
8599b52e
VF		 4284 }
4285 }
4286
1da177e4 4287 if (miimon) {
a4aee5c8 4288 pr_info("MII link monitoring set to %d ms\n", miimon);
1da177e4 4289 } else if (arp_interval) {
16228881
NA		 4290 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4291 arp_validate_value);
a4aee5c8 4292 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
16228881 4293 arp_interval, valptr->string, arp_ip_count);
1da177e4
LT		 4294
4295 for (i = 0; i < arp_ip_count; i++)
90194264 4296 pr_cont(" %s", arp_ip_target[i]);
1da177e4 4297
90194264 4298 pr_cont("\n");
1da177e4 4299
b8a9787e 4300 } else if (max_bonds) {
1da177e4
LT		 4301 /* miimon and arp_interval not set, we need one so things
4302 * work as expected, see bonding.txt for details
4303 */
90194264 4304 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details\n");
1da177e4
LT		 4305 }
4306
ec0865a9 4307 if (primary && !bond_mode_uses_primary(bond_mode)) {
1da177e4
LT		 4308 /* currently, using a primary only makes sense
4309 * in active backup, TLB or ALB modes
4310 */
91565ebb
JP		 4311 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4312 primary, bond_mode_name(bond_mode));
1da177e4
LT		 4313 primary = NULL;
4314 }
4315
a549952a 4316 if (primary && primary_reselect) {
388d3a6d
NA		 4317 bond_opt_initstr(&newval, primary_reselect);
4318 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4319 &newval);
4320 if (!valptr) {
a4aee5c8 4321 pr_err("Error: Invalid primary_reselect \"%s\"\n",
388d3a6d 4322 primary_reselect);
a549952a
JP		 4323 return -EINVAL;
4324 }
388d3a6d 4325 primary_reselect_value = valptr->value;
a549952a
JP		 4326 } else {
4327 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4328 }
4329
3915c1e8 4330 if (fail_over_mac) {
1df6b6aa
NA		 4331 bond_opt_initstr(&newval, fail_over_mac);
4332 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4333 &newval);
4334 if (!valptr) {
a4aee5c8 4335 pr_err("Error: invalid fail_over_mac \"%s\"\n",
1df6b6aa 4336 fail_over_mac);
3915c1e8
JV		 4337 return -EINVAL;
4338 }
1df6b6aa 4339 fail_over_mac_value = valptr->value;
3915c1e8 4340 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
91565ebb 4341 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
3915c1e8
JV		 4342 } else {
4343 fail_over_mac_value = BOND_FOM_NONE;
4344 }
dd957c57 4345
3a7129e5 4346 if (lp_interval == 0) {
91565ebb
JP		 4347 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4348 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
3a7129e5 4349 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4350 }
4351
1da177e4
LT		 4352 /* fill params struct with the proper values */
4353 params->mode = bond_mode;
169a3e66 4354 params->xmit_policy = xmit_hashtype;
1da177e4 4355 params->miimon = miimon;
ad246c99 4356 params->num_peer_notif = num_peer_notif;
1da177e4 4357 params->arp_interval = arp_interval;
f5b2b966 4358 params->arp_validate = arp_validate_value;
8599b52e 4359 params->arp_all_targets = arp_all_targets_value;
1da177e4
LT		 4360 params->updelay = updelay;
4361 params->downdelay = downdelay;
4362 params->use_carrier = use_carrier;
4363 params->lacp_fast = lacp_fast;
4364 params->primary[0] = 0;
a549952a 4365 params->primary_reselect = primary_reselect_value;
3915c1e8 4366 params->fail_over_mac = fail_over_mac_value;
bb1d9123 4367 params->tx_queues = tx_queues;
ebd8e497 4368 params->all_slaves_active = all_slaves_active;
c2952c31 4369 params->resend_igmp = resend_igmp;
655f8919 4370 params->min_links = min_links;
3a7129e5 4371 params->lp_interval = lp_interval;
809fa972 4372 params->packets_per_slave = packets_per_slave;
e9f0fb88 4373 params->tlb_dynamic_lb = 1; /* Default value */
809fa972
HFS		 4374 if (packets_per_slave > 0) {
4375 params->reciprocal_packets_per_slave =
4376 reciprocal_value(packets_per_slave);
4377 } else {
4378 /* reciprocal_packets_per_slave is unused if
4379 * packets_per_slave is 0 or 1, just initialize it
4380 */
4381 params->reciprocal_packets_per_slave =
4382 (struct reciprocal_value) { 0 };
4383 }
4384
1da177e4
LT		 4385 if (primary) {
4386 strncpy(params->primary, primary, IFNAMSIZ);
4387 params->primary[IFNAMSIZ - 1] = 0;
4388 }
4389
4390 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4391
4392 return 0;
4393}
4394
0daa2303 4395static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 4396static struct lock_class_key bonding_netdev_addr_lock_key;
49ee4920 4397static struct lock_class_key bonding_tx_busylock_key;
0daa2303 4398
e8a0464c
DM		 4399static void bond_set_lockdep_class_one(struct net_device *dev,
4400 struct netdev_queue *txq,
4401 void *_unused)
c773e847
DM		 4402{
4403 lockdep_set_class(&txq->_xmit_lock,
4404 &bonding_netdev_xmit_lock_key);
4405}
4406
4407static void bond_set_lockdep_class(struct net_device *dev)
4408{
cf508b12
DM		 4409 lockdep_set_class(&dev->addr_list_lock,
4410 &bonding_netdev_addr_lock_key);
e8a0464c 4411 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
49ee4920 4412 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
c773e847
DM		 4413}
4414
547942ca 4415/* Called from registration process */
181470fc
SH		 4416static int bond_init(struct net_device *bond_dev)
4417{
4418 struct bonding *bond = netdev_priv(bond_dev);
ec87fd3b 4419 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
181470fc 4420
76444f50 4421 netdev_dbg(bond_dev, "Begin bond_init\n");
181470fc
SH		 4422
4423 bond->wq = create_singlethread_workqueue(bond_dev->name);
4424 if (!bond->wq)
4425 return -ENOMEM;
4426
4427 bond_set_lockdep_class(bond_dev);
4428
ec87fd3b 4429 list_add_tail(&bond->bond_list, &bn->dev_list);
181470fc 4430
6151b3d4 4431 bond_prepare_sysfs_group(bond);
22bedad3 4432
f073c7ca
TI		 4433 bond_debug_register(bond);
4434
409cc1f8
JP		 4435 /* Ensure valid dev_addr */
4436 if (is_zero_ether_addr(bond_dev->dev_addr) &&
97a1e639 4437 bond_dev->addr_assign_type == NET_ADDR_PERM)
409cc1f8 4438 eth_hw_addr_random(bond_dev);
409cc1f8 4439
181470fc
SH		 4440 return 0;
4441}
4442
0a2a78c4 4443unsigned int bond_get_num_tx_queues(void)
d5da4510 4444{
efacb309 4445 return tx_queues;
d5da4510
JP		 4446}
4447
dfe60397 4448/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 4449 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 4450 * Caller must NOT hold rtnl_lock; we need to release it here before we
4451 * set up our sysfs entries.
4452 */
ec87fd3b 4453int bond_create(struct net *net, const char *name)
dfe60397
MW		 4454{
4455 struct net_device *bond_dev;
4456 int res;
4457
4458 rtnl_lock();
027ea041 4459
1c5cae81 4460 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
c835a677 4461 name ? name : "bond%d", NET_NAME_UNKNOWN,
1c5cae81 4462 bond_setup, tx_queues);
dfe60397 4463 if (!bond_dev) {
a4aee5c8 4464 pr_err("%s: eek! can't alloc netdev!\n", name);
9e2e61fb
AW		 4465 rtnl_unlock();
4466 return -ENOMEM;
dfe60397
MW		 4467 }
4468
ec87fd3b 4469 dev_net_set(bond_dev, net);
88ead977
EB		 4470 bond_dev->rtnl_link_ops = &bond_link_ops;
4471
dfe60397 4472 res = register_netdevice(bond_dev);
0daa2303 4473
e826eafa
PO		 4474 netif_carrier_off(bond_dev);
4475
7e083840 4476 rtnl_unlock();
9e2e61fb
AW		 4477 if (res < 0)
4478 bond_destructor(bond_dev);
30c15ba9 4479 return res;
dfe60397
MW		 4480}
4481
2c8c1e72 4482static int __net_init bond_net_init(struct net *net)
ec87fd3b 4483{
15449745 4484 struct bond_net *bn = net_generic(net, bond_net_id);
ec87fd3b
EB		 4485
4486 bn->net = net;
4487 INIT_LIST_HEAD(&bn->dev_list);
4488
ec87fd3b 4489 bond_create_proc_dir(bn);
4c22400a 4490 bond_create_sysfs(bn);
87a7b84b 4491
15449745 4492 return 0;
ec87fd3b
EB		 4493}
4494
2c8c1e72 4495static void __net_exit bond_net_exit(struct net *net)
ec87fd3b 4496{
15449745 4497 struct bond_net *bn = net_generic(net, bond_net_id);
69b0216a 4498 struct bonding *bond, *tmp_bond;
4499 LIST_HEAD(list);
ec87fd3b 4500
4c22400a 4501 bond_destroy_sysfs(bn);
69b0216a 4502
4503 /* Kill off any bonds created after unregistering bond rtnl ops */
4504 rtnl_lock();
4505 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4506 unregister_netdevice_queue(bond->dev, &list);
4507 unregister_netdevice_many(&list);
4508 rtnl_unlock();
23fa5c2c
VF		 4509
4510 bond_destroy_proc_dir(bn);
ec87fd3b
EB		 4511}
4512
4513static struct pernet_operations bond_net_ops = {
4514 .init = bond_net_init,
4515 .exit = bond_net_exit,
15449745
EB		 4516 .id = &bond_net_id,
4517 .size = sizeof(struct bond_net),
ec87fd3b
EB		 4518};
4519
1da177e4
LT		 4520static int __init bonding_init(void)
4521{
1da177e4
LT		 4522 int i;
4523 int res;
4524
bd33acc3 4525 pr_info("%s", bond_version);
1da177e4 4526
dfe60397 4527 res = bond_check_params(&bonding_defaults);
3d632c3f 4528 if (res)
dfe60397 4529 goto out;
1da177e4 4530
15449745 4531 res = register_pernet_subsys(&bond_net_ops);
ec87fd3b
EB		 4532 if (res)
4533 goto out;
027ea041 4534
0a2a78c4 4535 res = bond_netlink_init();
88ead977 4536 if (res)
6639104b 4537 goto err_link;
88ead977 4538
f073c7ca
TI		 4539 bond_create_debugfs();
4540
1da177e4 4541 for (i = 0; i < max_bonds; i++) {
ec87fd3b 4542 res = bond_create(&init_net, NULL);
dfe60397
MW		 4543 if (res)
4544 goto err;
1da177e4
LT		 4545 }
4546
1da177e4 4547 register_netdevice_notifier(&bond_netdev_notifier);
dfe60397 4548out:
1da177e4 4549 return res;
88ead977 4550err:
db298686 4551 bond_destroy_debugfs();
0a2a78c4 4552 bond_netlink_fini();
6639104b 4553err_link:
15449745 4554 unregister_pernet_subsys(&bond_net_ops);
88ead977 4555 goto out;
dfe60397 4556
1da177e4
LT		 4557}
4558
4559static void __exit bonding_exit(void)
4560{
4561 unregister_netdevice_notifier(&bond_netdev_notifier);
4562
f073c7ca 4563 bond_destroy_debugfs();
ae68c398 4564
0a2a78c4 4565 bond_netlink_fini();
ffcdedb6 4566 unregister_pernet_subsys(&bond_net_ops);
e843fa50
NH		 4567
4568#ifdef CONFIG_NET_POLL_CONTROLLER
547942ca 4569 /* Make sure we don't have an imbalance on our netpoll blocking */
fb4fa76a 4570 WARN_ON(atomic_read(&netpoll_block_tx));
e843fa50 4571#endif
1da177e4
LT		 4572}
4573
4574module_init(bonding_init);
4575module_exit(bonding_exit);
4576MODULE_LICENSE("GPL");
4577MODULE_VERSION(DRV_VERSION);
4578MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4579MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
Linux kernel - Deliverables
This page took 1.849942 seconds and 5 git commands to generate.