]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - drivers/net/bonding/bond_main.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge remote-tracking branches 'spi/topic/s3c64xx', 'spi/topic/sc18is602', 'spi/topic...
[mirror_ubuntu-jammy-kernel.git] / drivers / net / bonding / bond_main.c
CommitLineData
1da177e4
LT		 1/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
1da177e4
LT		 32 */
33
a4aee5c8
JP		 34#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
1da177e4
LT		 36#include <linux/kernel.h>
37#include <linux/module.h>
1da177e4
LT		 38#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
169a3e66 44#include <net/ip.h>
1da177e4 45#include <linux/ip.h>
169a3e66
JV		 46#include <linux/tcp.h>
47#include <linux/udp.h>
1da177e4
LT		 48#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
3d632c3f 56#include <linux/io.h>
1da177e4 57#include <asm/dma.h>
3d632c3f 58#include <linux/uaccess.h>
1da177e4
LT		 59#include <linux/errno.h>
60#include <linux/netdevice.h>
61#include <linux/inetdevice.h>
a816c7c7 62#include <linux/igmp.h>
1da177e4
LT		 63#include <linux/etherdevice.h>
64#include <linux/skbuff.h>
65#include <net/sock.h>
66#include <linux/rtnetlink.h>
1da177e4
LT		 67#include <linux/smp.h>
68#include <linux/if_ether.h>
69#include <net/arp.h>
70#include <linux/mii.h>
71#include <linux/ethtool.h>
72#include <linux/if_vlan.h>
73#include <linux/if_bonding.h>
b63bb739 74#include <linux/jiffies.h>
e843fa50 75#include <linux/preempt.h>
c3ade5ca 76#include <net/route.h>
457c4cbc 77#include <net/net_namespace.h>
ec87fd3b 78#include <net/netns/generic.h>
5ee31c68 79#include <net/pkt_sched.h>
278b2083 80#include <linux/rculist.h>
32819dc1 81#include <net/flow_keys.h>
1da177e4
LT		 82#include "bonding.h"
83#include "bond_3ad.h"
84#include "bond_alb.h"
85
86/*---------------------------- Module parameters ----------------------------*/
87
88/* monitor all links that often (in milliseconds). <=0 disables monitoring */
1da177e4
LT		 89
90static int max_bonds = BOND_DEFAULT_MAX_BONDS;
bb1d9123 91static int tx_queues = BOND_DEFAULT_TX_QUEUES;
ad246c99 92static int num_peer_notif = 1;
b98d9c66 93static int miimon;
3d632c3f
SH		 94static int updelay;
95static int downdelay;
1da177e4 96static int use_carrier = 1;
3d632c3f
SH		 97static char *mode;
98static char *primary;
a549952a 99static char *primary_reselect;
3d632c3f 100static char *lacp_rate;
655f8919 101static int min_links;
3d632c3f
SH		 102static char *ad_select;
103static char *xmit_hash_policy;
7bdb04ed 104static int arp_interval;
3d632c3f
SH		 105static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106static char *arp_validate;
8599b52e 107static char *arp_all_targets;
3d632c3f 108static char *fail_over_mac;
b07ea07b 109static int all_slaves_active;
d2991f75 110static struct bond_params bonding_defaults;
c2952c31 111static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
73958329 112static int packets_per_slave = 1;
3a7129e5 113static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
1da177e4
LT		 114
115module_param(max_bonds, int, 0);
116MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
bb1d9123
AG		 117module_param(tx_queues, int, 0);
118MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
ad246c99 119module_param_named(num_grat_arp, num_peer_notif, int, 0644);
90e62474
AG		 120MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
ad246c99 122module_param_named(num_unsol_na, num_peer_notif, int, 0644);
90e62474
AG		 123MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
1da177e4
LT		 125module_param(miimon, int, 0);
126MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127module_param(updelay, int, 0);
128MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129module_param(downdelay, int, 0);
2ac47660
MW		 130MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
131 "in milliseconds");
1da177e4 132module_param(use_carrier, int, 0);
2ac47660
MW		 133MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
1da177e4 135module_param(mode, charp, 0);
90e62474 136MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
2ac47660
MW		 137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
1da177e4
LT		 140module_param(primary, charp, 0);
141MODULE_PARM_DESC(primary, "Primary network device to use");
a549952a
JP		 142module_param(primary_reselect, charp, 0);
143MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
144 "once it comes up; "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
147 "better, "
148 "2 for only on active slave "
149 "failure");
1da177e4 150module_param(lacp_rate, charp, 0);
90e62474
AG		 151MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
fd989c83 153module_param(ad_select, charp, 0);
90e62474
AG		 154MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
156 "2 for count");
655f8919 157module_param(min_links, int, 0);
158MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
159
169a3e66 160module_param(xmit_hash_policy, charp, 0);
90e62474
AG		 161MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
32819dc1
NA		 163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
1da177e4
LT		 165module_param(arp_interval, int, 0);
166MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167module_param_array(arp_ip_target, charp, NULL, 0);
168MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
f5b2b966 169module_param(arp_validate, charp, 0);
90e62474
AG		 170MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
8599b52e
VF		 173module_param(arp_all_targets, charp, 0);
174MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
3915c1e8 175module_param(fail_over_mac, charp, 0);
90e62474
AG		 176MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
ebd8e497
AG		 179module_param(all_slaves_active, int, 0);
180MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
90e62474 181 "by setting active flag for all slaves; "
ebd8e497 182 "0 for never (default), 1 for always.");
c2952c31 183module_param(resend_igmp, int, 0);
90e62474
AG		 184MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
185 "link failure");
73958329
NA		 186module_param(packets_per_slave, int, 0);
187MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
3a7129e5 190module_param(lp_interval, uint, 0);
191MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
1da177e4
LT		 194
195/*----------------------------- Global variables ----------------------------*/
196
e843fa50 197#ifdef CONFIG_NET_POLL_CONTROLLER
fb4fa76a 198atomic_t netpoll_block_tx = ATOMIC_INIT(0);
e843fa50
NH		 199#endif
200
f99189b1 201int bond_net_id __read_mostly;
1da177e4 202
3d632c3f
SH		 203static __be32 arp_target[BOND_MAX_ARP_TARGETS];
204static int arp_ip_count;
1da177e4 205static int bond_mode = BOND_MODE_ROUNDROBIN;
3d632c3f
SH		 206static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
207static int lacp_fast;
1da177e4 208
1da177e4
LT		 209/*-------------------------- Forward declarations ---------------------------*/
210
181470fc 211static int bond_init(struct net_device *bond_dev);
c67dfb29 212static void bond_uninit(struct net_device *bond_dev);
1da177e4
LT		 213
214/*---------------------------- General routines -----------------------------*/
215
bd33acc3 216const char *bond_mode_name(int mode)
1da177e4 217{
77afc92b
HE		 218 static const char *names[] = {
219 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
3d632c3f 223 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
77afc92b
HE		 224 [BOND_MODE_TLB] = "transmit load balancing",
225 [BOND_MODE_ALB] = "adaptive load balancing",
226 };
227
f5280948 228 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
1da177e4 229 return "unknown";
77afc92b
HE		 230
231 return names[mode];
1da177e4
LT		 232}
233
234/*---------------------------------- VLAN -----------------------------------*/
235
1da177e4
LT		 236/**
237 * bond_dev_queue_xmit - Prepare skb for xmit.
3d632c3f 238 *
1da177e4
LT		 239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
1da177e4 242 */
d316dedd 243void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
3d632c3f 244 struct net_device *slave_dev)
1da177e4 245{
8387451e 246 skb->dev = slave_dev;
374eeb5a 247
5ee31c68 248 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
df4ab5b3
JP		 249 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
250 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
374eeb5a 251
e15c3c22 252 if (unlikely(netpoll_tx_running(bond->dev)))
8a8efa22 253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
080e4130 254 else
f6dc31a8 255 dev_queue_xmit(skb);
1da177e4
LT		 256}
257
258/*
cc0e4070
JP		 259 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
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
ff59c456
JV		 331/*
332 * Set the carrier state for the master according to the state of its
333 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
334 * do special 802.3ad magic.
335 *
336 * Returns zero if carrier state does not change, nonzero if it does.
337 */
338static int bond_set_carrier(struct bonding *bond)
339{
9caff1e7 340 struct list_head *iter;
ff59c456 341 struct slave *slave;
ff59c456 342
0965a1f3 343 if (!bond_has_slaves(bond))
ff59c456
JV		 344 goto down;
345
346 if (bond->params.mode == BOND_MODE_8023AD)
347 return bond_3ad_set_carrier(bond);
348
9caff1e7 349 bond_for_each_slave(bond, slave, iter) {
ff59c456
JV		 350 if (slave->link == BOND_LINK_UP) {
351 if (!netif_carrier_ok(bond->dev)) {
352 netif_carrier_on(bond->dev);
353 return 1;
354 }
355 return 0;
356 }
357 }
358
359down:
360 if (netif_carrier_ok(bond->dev)) {
361 netif_carrier_off(bond->dev);
362 return 1;
363 }
364 return 0;
365}
366
1da177e4
LT		 367/*
368 * Get link speed and duplex from the slave's base driver
369 * using ethtool. If for some reason the call fails or the
98f41f69 370 * values are invalid, set speed and duplex to -1,
13b95fb7 371 * and return.
1da177e4 372 */
13b95fb7 373static void bond_update_speed_duplex(struct slave *slave)
1da177e4
LT		 374{
375 struct net_device *slave_dev = slave->dev;
4bc71cb9 376 struct ethtool_cmd ecmd;
5d30530e 377 u32 slave_speed;
61a44b9c 378 int res;
1da177e4 379
589665f5
DC		 380 slave->speed = SPEED_UNKNOWN;
381 slave->duplex = DUPLEX_UNKNOWN;
1da177e4 382
4bc71cb9 383 res = __ethtool_get_settings(slave_dev, &ecmd);
61a44b9c 384 if (res < 0)
13b95fb7 385 return;
1da177e4 386
4bc71cb9 387 slave_speed = ethtool_cmd_speed(&ecmd);
6f92c66f 388 if (slave_speed == 0 || slave_speed == ((__u32) -1))
13b95fb7 389 return;
1da177e4 390
4bc71cb9 391 switch (ecmd.duplex) {
1da177e4
LT		 392 case DUPLEX_FULL:
393 case DUPLEX_HALF:
394 break;
395 default:
13b95fb7 396 return;
1da177e4
LT		 397 }
398
5d30530e 399 slave->speed = slave_speed;
4bc71cb9 400 slave->duplex = ecmd.duplex;
1da177e4 401
13b95fb7 402 return;
1da177e4
LT		 403}
404
07699f9a 405const char *bond_slave_link_status(s8 link)
406{
407 switch (link) {
408 case BOND_LINK_UP:
409 return "up";
410 case BOND_LINK_FAIL:
411 return "going down";
412 case BOND_LINK_DOWN:
413 return "down";
414 case BOND_LINK_BACK:
415 return "going back";
416 default:
417 return "unknown";
418 }
419}
420
1da177e4
LT		 421/*
422 * if <dev> supports MII link status reporting, check its link status.
423 *
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
3d632c3f 425 * depending upon the setting of the use_carrier parameter.
1da177e4
LT		 426 *
427 * Return either BMSR_LSTATUS, meaning that the link is up (or we
428 * can't tell and just pretend it is), or 0, meaning that the link is
429 * down.
430 *
431 * If reporting is non-zero, instead of faking link up, return -1 if
432 * both ETHTOOL and MII ioctls fail (meaning the device does not
433 * support them). If use_carrier is set, return whatever it says.
434 * It'd be nice if there was a good way to tell if a driver supports
435 * netif_carrier, but there really isn't.
436 */
3d632c3f
SH		 437static int bond_check_dev_link(struct bonding *bond,
438 struct net_device *slave_dev, int reporting)
1da177e4 439{
eb7cc59a 440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
d9d52832 441 int (*ioctl)(struct net_device *, struct ifreq *, int);
1da177e4
LT		 442 struct ifreq ifr;
443 struct mii_ioctl_data *mii;
1da177e4 444
6c988853
PG		 445 if (!reporting && !netif_running(slave_dev))
446 return 0;
447
eb7cc59a 448 if (bond->params.use_carrier)
1da177e4 449 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1da177e4 450
29112f4e 451 /* Try to get link status using Ethtool first. */
c772dde3
BH		 452 if (slave_dev->ethtool_ops->get_link)
453 return slave_dev->ethtool_ops->get_link(slave_dev) ?
454 BMSR_LSTATUS : 0;
29112f4e 455
3d632c3f 456 /* Ethtool can't be used, fallback to MII ioctls. */
eb7cc59a 457 ioctl = slave_ops->ndo_do_ioctl;
1da177e4
LT		 458 if (ioctl) {
459 /* TODO: set pointer to correct ioctl on a per team member */
460 /* bases to make this more efficient. that is, once */
461 /* we determine the correct ioctl, we will always */
462 /* call it and not the others for that team */
463 /* member. */
464
465 /*
466 * We cannot assume that SIOCGMIIPHY will also read a
467 * register; not all network drivers (e.g., e100)
468 * support that.
469 */
470
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
473 mii = if_mii(&ifr);
474 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
475 mii->reg_num = MII_BMSR;
3d632c3f
SH		 476 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
477 return mii->val_out & BMSR_LSTATUS;
1da177e4
LT		 478 }
479 }
480
1da177e4
LT		 481 /*
482 * If reporting, report that either there's no dev->do_ioctl,
61a44b9c 483 * or both SIOCGMIIREG and get_link failed (meaning that we
1da177e4
LT		 484 * cannot report link status). If not reporting, pretend
485 * we're ok.
486 */
3d632c3f 487 return reporting ? -1 : BMSR_LSTATUS;
1da177e4
LT		 488}
489
490/*----------------------------- Multicast list ------------------------------*/
491
1da177e4
LT		 492/*
493 * Push the promiscuity flag down to appropriate slaves
494 */
7e1a1ac1 495static int bond_set_promiscuity(struct bonding *bond, int inc)
1da177e4 496{
9caff1e7 497 struct list_head *iter;
7e1a1ac1 498 int err = 0;
9caff1e7 499
1da177e4
LT		 500 if (USES_PRIMARY(bond->params.mode)) {
501 /* write lock already acquired */
502 if (bond->curr_active_slave) {
7e1a1ac1
WC		 503 err = dev_set_promiscuity(bond->curr_active_slave->dev,
504 inc);
1da177e4
LT		 505 }
506 } else {
507 struct slave *slave;
dec1e90e 508
9caff1e7 509 bond_for_each_slave(bond, slave, iter) {
7e1a1ac1
WC		 510 err = dev_set_promiscuity(slave->dev, inc);
511 if (err)
512 return err;
1da177e4
LT		 513 }
514 }
7e1a1ac1 515 return err;
1da177e4
LT		 516}
517
518/*
519 * Push the allmulti flag down to all slaves
520 */
7e1a1ac1 521static int bond_set_allmulti(struct bonding *bond, int inc)
1da177e4 522{
9caff1e7 523 struct list_head *iter;
7e1a1ac1 524 int err = 0;
9caff1e7 525
1da177e4
LT		 526 if (USES_PRIMARY(bond->params.mode)) {
527 /* write lock already acquired */
528 if (bond->curr_active_slave) {
7e1a1ac1
WC		 529 err = dev_set_allmulti(bond->curr_active_slave->dev,
530 inc);
1da177e4
LT		 531 }
532 } else {
533 struct slave *slave;
dec1e90e 534
9caff1e7 535 bond_for_each_slave(bond, slave, iter) {
7e1a1ac1
WC		 536 err = dev_set_allmulti(slave->dev, inc);
537 if (err)
538 return err;
1da177e4
LT		 539 }
540 }
7e1a1ac1 541 return err;
1da177e4
LT		 542}
543
5a37e8ca
FL		 544/*
545 * Retrieve the list of registered multicast addresses for the bonding
546 * device and retransmit an IGMP JOIN request to the current active
547 * slave.
548 */
f2369109 549static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
5a37e8ca 550{
f2369109 551 struct bonding *bond = container_of(work, struct bonding,
552 mcast_work.work);
553
4aa5dee4 554 if (!rtnl_trylock()) {
4beac029 555 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
4aa5dee4 556 return;
5a37e8ca 557 }
4aa5dee4 558 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
5a37e8ca 559
4f5474e7
NA		 560 if (bond->igmp_retrans > 1) {
561 bond->igmp_retrans--;
c2952c31 562 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
4f5474e7 563 }
f2369109 564 rtnl_unlock();
5a37e8ca
FL		 565}
566
303d1cbf 567/* Flush bond's hardware addresses from slave
1da177e4 568 */
303d1cbf 569static void bond_hw_addr_flush(struct net_device *bond_dev,
3d632c3f 570 struct net_device *slave_dev)
1da177e4 571{
454d7c9b 572 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 573
303d1cbf
JV		 574 dev_uc_unsync(slave_dev, bond_dev);
575 dev_mc_unsync(slave_dev, bond_dev);
1da177e4
LT		 576
577 if (bond->params.mode == BOND_MODE_8023AD) {
578 /* del lacpdu mc addr from mc list */
579 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
580
22bedad3 581 dev_mc_del(slave_dev, lacpdu_multicast);
1da177e4
LT		 582 }
583}
584
585/*--------------------------- Active slave change ---------------------------*/
586
303d1cbf
JV		 587/* Update the hardware address list and promisc/allmulti for the new and
588 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
589 * slaves up date at all times; only the USES_PRIMARY modes need to call
590 * this function to swap these settings during a failover.
1da177e4 591 */
303d1cbf
JV		 592static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
593 struct slave *old_active)
1da177e4 594{
7864a1ad
VF		 595 ASSERT_RTNL();
596
1da177e4 597 if (old_active) {
3d632c3f 598 if (bond->dev->flags & IFF_PROMISC)
1da177e4 599 dev_set_promiscuity(old_active->dev, -1);
1da177e4 600
3d632c3f 601 if (bond->dev->flags & IFF_ALLMULTI)
1da177e4 602 dev_set_allmulti(old_active->dev, -1);
1da177e4 603
303d1cbf 604 bond_hw_addr_flush(bond->dev, old_active->dev);
1da177e4
LT		 605 }
606
607 if (new_active) {
7e1a1ac1 608 /* FIXME: Signal errors upstream. */
3d632c3f 609 if (bond->dev->flags & IFF_PROMISC)
1da177e4 610 dev_set_promiscuity(new_active->dev, 1);
1da177e4 611
3d632c3f 612 if (bond->dev->flags & IFF_ALLMULTI)
1da177e4 613 dev_set_allmulti(new_active->dev, 1);
1da177e4 614
d632ce98 615 netif_addr_lock_bh(bond->dev);
303d1cbf
JV		 616 dev_uc_sync(new_active->dev, bond->dev);
617 dev_mc_sync(new_active->dev, bond->dev);
d632ce98 618 netif_addr_unlock_bh(bond->dev);
1da177e4
LT		 619 }
620}
621
ae0d6750 622/**
623 * bond_set_dev_addr - clone slave's address to bond
624 * @bond_dev: bond net device
625 * @slave_dev: slave net device
626 *
627 * Should be called with RTNL held.
628 */
629static void bond_set_dev_addr(struct net_device *bond_dev,
630 struct net_device *slave_dev)
631{
008aebde
NA		 632 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
633 bond_dev, slave_dev, slave_dev->addr_len);
ae0d6750 634 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
635 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
636 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
637}
638
3915c1e8
JV		 639/*
640 * bond_do_fail_over_mac
641 *
642 * Perform special MAC address swapping for fail_over_mac settings
643 *
b2e7aceb 644 * Called with RTNL, curr_slave_lock for write_bh.
3915c1e8
JV		 645 */
646static void bond_do_fail_over_mac(struct bonding *bond,
647 struct slave *new_active,
648 struct slave *old_active)
1f78d9f9 649 __releases(&bond->curr_slave_lock)
1f78d9f9 650 __acquires(&bond->curr_slave_lock)
3915c1e8
JV		 651{
652 u8 tmp_mac[ETH_ALEN];
653 struct sockaddr saddr;
654 int rv;
655
656 switch (bond->params.fail_over_mac) {
657 case BOND_FOM_ACTIVE:
7d26bb10 658 if (new_active) {
7d26bb10 659 write_unlock_bh(&bond->curr_slave_lock);
ae0d6750 660 bond_set_dev_addr(bond->dev, new_active->dev);
7d26bb10
WP		 661 write_lock_bh(&bond->curr_slave_lock);
662 }
3915c1e8
JV		 663 break;
664 case BOND_FOM_FOLLOW:
665 /*
666 * if new_active && old_active, swap them
667 * if just old_active, do nothing (going to no active slave)
668 * if just new_active, set new_active to bond's MAC
669 */
670 if (!new_active)
671 return;
672
673 write_unlock_bh(&bond->curr_slave_lock);
3915c1e8
JV		 674
675 if (old_active) {
676 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
677 memcpy(saddr.sa_data, old_active->dev->dev_addr,
678 ETH_ALEN);
679 saddr.sa_family = new_active->dev->type;
680 } else {
681 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
682 saddr.sa_family = bond->dev->type;
683 }
684
685 rv = dev_set_mac_address(new_active->dev, &saddr);
686 if (rv) {
a4aee5c8 687 pr_err("%s: Error %d setting MAC of slave %s\n",
3915c1e8
JV		 688 bond->dev->name, -rv, new_active->dev->name);
689 goto out;
690 }
691
692 if (!old_active)
693 goto out;
694
695 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
696 saddr.sa_family = old_active->dev->type;
697
698 rv = dev_set_mac_address(old_active->dev, &saddr);
699 if (rv)
a4aee5c8 700 pr_err("%s: Error %d setting MAC of slave %s\n",
3915c1e8
JV		 701 bond->dev->name, -rv, new_active->dev->name);
702out:
3915c1e8
JV		 703 write_lock_bh(&bond->curr_slave_lock);
704 break;
705 default:
a4aee5c8 706 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
3915c1e8
JV		 707 bond->dev->name, bond->params.fail_over_mac);
708 break;
709 }
710
711}
712
a549952a
JP		 713static bool bond_should_change_active(struct bonding *bond)
714{
715 struct slave *prim = bond->primary_slave;
716 struct slave *curr = bond->curr_active_slave;
717
718 if (!prim || !curr || curr->link != BOND_LINK_UP)
719 return true;
720 if (bond->force_primary) {
721 bond->force_primary = false;
722 return true;
723 }
724 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
725 (prim->speed < curr->speed ||
726 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
727 return false;
728 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
729 return false;
730 return true;
731}
3915c1e8 732
1da177e4
LT		 733/**
734 * find_best_interface - select the best available slave to be the active one
735 * @bond: our bonding struct
1da177e4
LT		 736 */
737static struct slave *bond_find_best_slave(struct bonding *bond)
738{
77140d29
VF		 739 struct slave *slave, *bestslave = NULL;
740 struct list_head *iter;
1da177e4 741 int mintime = bond->params.updelay;
1da177e4 742
77140d29
VF		 743 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
744 bond_should_change_active(bond))
745 return bond->primary_slave;
1da177e4 746
77140d29
VF		 747 bond_for_each_slave(bond, slave, iter) {
748 if (slave->link == BOND_LINK_UP)
749 return slave;
750 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
751 slave->delay < mintime) {
752 mintime = slave->delay;
753 bestslave = slave;
1da177e4
LT		 754 }
755 }
756
757 return bestslave;
758}
759
ad246c99
BH		 760static bool bond_should_notify_peers(struct bonding *bond)
761{
4cb4f97b 762 struct slave *slave;
763
764 rcu_read_lock();
765 slave = rcu_dereference(bond->curr_active_slave);
766 rcu_read_unlock();
ad246c99
BH		 767
768 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
769 bond->dev->name, slave ? slave->dev->name : "NULL");
770
771 if (!slave || !bond->send_peer_notif ||
772 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
773 return false;
774
ad246c99
BH		 775 return true;
776}
777
1da177e4
LT		 778/**
779 * change_active_interface - change the active slave into the specified one
780 * @bond: our bonding struct
781 * @new: the new slave to make the active one
782 *
783 * Set the new slave to the bond's settings and unset them on the old
784 * curr_active_slave.
785 * Setting include flags, mc-list, promiscuity, allmulti, etc.
786 *
787 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
788 * because it is apparently the best available slave we have, even though its
789 * updelay hasn't timed out yet.
790 *
b2e7aceb 791 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
1da177e4 792 */
a77b5325 793void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1da177e4
LT		 794{
795 struct slave *old_active = bond->curr_active_slave;
796
3d632c3f 797 if (old_active == new_active)
1da177e4 798 return;
1da177e4
LT		 799
800 if (new_active) {
b2220cad
JV		 801 new_active->jiffies = jiffies;
802
1da177e4
LT		 803 if (new_active->link == BOND_LINK_BACK) {
804 if (USES_PRIMARY(bond->params.mode)) {
a4aee5c8
JP		 805 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
806 bond->dev->name, new_active->dev->name,
807 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1da177e4
LT		 808 }
809
810 new_active->delay = 0;
811 new_active->link = BOND_LINK_UP;
1da177e4 812
3d632c3f 813 if (bond->params.mode == BOND_MODE_8023AD)
1da177e4 814 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1da177e4 815
58402054 816 if (bond_is_lb(bond))
1da177e4 817 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1da177e4
LT		 818 } else {
819 if (USES_PRIMARY(bond->params.mode)) {
a4aee5c8
JP		 820 pr_info("%s: making interface %s the new active one.\n",
821 bond->dev->name, new_active->dev->name);
1da177e4
LT		 822 }
823 }
824 }
825
3d632c3f 826 if (USES_PRIMARY(bond->params.mode))
303d1cbf 827 bond_hw_addr_swap(bond, new_active, old_active);
1da177e4 828
58402054 829 if (bond_is_lb(bond)) {
1da177e4 830 bond_alb_handle_active_change(bond, new_active);
8f903c70 831 if (old_active)
5e5b0665 832 bond_set_slave_inactive_flags(old_active,
833 BOND_SLAVE_NOTIFY_NOW);
8f903c70 834 if (new_active)
5e5b0665 835 bond_set_slave_active_flags(new_active,
836 BOND_SLAVE_NOTIFY_NOW);
1da177e4 837 } else {
278b2083 838 rcu_assign_pointer(bond->curr_active_slave, new_active);
1da177e4 839 }
c3ade5ca
JV		 840
841 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3d632c3f 842 if (old_active)
5e5b0665 843 bond_set_slave_inactive_flags(old_active,
844 BOND_SLAVE_NOTIFY_NOW);
c3ade5ca
JV		 845
846 if (new_active) {
ad246c99
BH		 847 bool should_notify_peers = false;
848
5e5b0665 849 bond_set_slave_active_flags(new_active,
850 BOND_SLAVE_NOTIFY_NOW);
2ab82852 851
709f8a45
OG		 852 if (bond->params.fail_over_mac)
853 bond_do_fail_over_mac(bond, new_active,
854 old_active);
3915c1e8 855
ad246c99
BH		 856 if (netif_running(bond->dev)) {
857 bond->send_peer_notif =
858 bond->params.num_peer_notif;
859 should_notify_peers =
860 bond_should_notify_peers(bond);
861 }
862
01f3109d 863 write_unlock_bh(&bond->curr_slave_lock);
01f3109d 864
b7bc2a5b 865 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
ad246c99 866 if (should_notify_peers)
b7bc2a5b
AW		 867 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
868 bond->dev);
01f3109d 869
01f3109d 870 write_lock_bh(&bond->curr_slave_lock);
7893b249 871 }
c3ade5ca 872 }
a2fd940f 873
5a37e8ca 874 /* resend IGMP joins since active slave has changed or
94265cf5
FL		 875 * all were sent on curr_active_slave.
876 * resend only if bond is brought up with the affected
877 * bonding modes and the retransmission is enabled */
878 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
879 ((USES_PRIMARY(bond->params.mode) && new_active) ||
880 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
c2952c31 881 bond->igmp_retrans = bond->params.resend_igmp;
4beac029 882 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
a2fd940f 883 }
1da177e4
LT		 884}
885
886/**
887 * bond_select_active_slave - select a new active slave, if needed
888 * @bond: our bonding struct
889 *
3d632c3f 890 * This functions should be called when one of the following occurs:
1da177e4
LT		 891 * - The old curr_active_slave has been released or lost its link.
892 * - The primary_slave has got its link back.
893 * - A slave has got its link back and there's no old curr_active_slave.
894 *
b2e7aceb 895 * Caller must hold curr_slave_lock for write_bh.
1da177e4 896 */
a77b5325 897void bond_select_active_slave(struct bonding *bond)
1da177e4
LT		 898{
899 struct slave *best_slave;
ff59c456 900 int rv;
1da177e4
LT		 901
902 best_slave = bond_find_best_slave(bond);
903 if (best_slave != bond->curr_active_slave) {
904 bond_change_active_slave(bond, best_slave);
ff59c456
JV		 905 rv = bond_set_carrier(bond);
906 if (!rv)
907 return;
908
909 if (netif_carrier_ok(bond->dev)) {
a4aee5c8
JP		 910 pr_info("%s: first active interface up!\n",
911 bond->dev->name);
ff59c456 912 } else {
a4aee5c8
JP		 913 pr_info("%s: now running without any active interface !\n",
914 bond->dev->name);
ff59c456 915 }
1da177e4
LT		 916 }
917}
918
f6dc31a8 919#ifdef CONFIG_NET_POLL_CONTROLLER
8a8efa22 920static inline int slave_enable_netpoll(struct slave *slave)
f6dc31a8 921{
8a8efa22
AW		 922 struct netpoll *np;
923 int err = 0;
f6dc31a8 924
47be03a2 925 np = kzalloc(sizeof(*np), GFP_ATOMIC);
8a8efa22
AW		 926 err = -ENOMEM;
927 if (!np)
928 goto out;
929
47be03a2 930 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
8a8efa22
AW		 931 if (err) {
932 kfree(np);
933 goto out;
f6dc31a8 934 }
8a8efa22
AW		 935 slave->np = np;
936out:
937 return err;
938}
939static inline void slave_disable_netpoll(struct slave *slave)
940{
941 struct netpoll *np = slave->np;
942
943 if (!np)
944 return;
945
946 slave->np = NULL;
2cde6acd 947 __netpoll_free_async(np);
8a8efa22
AW		 948}
949static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
950{
951 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
952 return false;
953 if (!slave_dev->netdev_ops->ndo_poll_controller)
954 return false;
955 return true;
f6dc31a8
WC		 956}
957
958static void bond_poll_controller(struct net_device *bond_dev)
959{
8a8efa22
AW		 960}
961
c4cdef9b 962static void bond_netpoll_cleanup(struct net_device *bond_dev)
8a8efa22 963{
c4cdef9b 964 struct bonding *bond = netdev_priv(bond_dev);
9caff1e7 965 struct list_head *iter;
c2355e1a 966 struct slave *slave;
c2355e1a 967
9caff1e7 968 bond_for_each_slave(bond, slave, iter)
8a8efa22
AW		 969 if (IS_UP(slave->dev))
970 slave_disable_netpoll(slave);
f6dc31a8 971}
8a8efa22 972
47be03a2 973static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
8a8efa22
AW		 974{
975 struct bonding *bond = netdev_priv(dev);
9caff1e7 976 struct list_head *iter;
f6dc31a8 977 struct slave *slave;
dec1e90e 978 int err = 0;
f6dc31a8 979
9caff1e7 980 bond_for_each_slave(bond, slave, iter) {
8a8efa22
AW		 981 err = slave_enable_netpoll(slave);
982 if (err) {
c4cdef9b 983 bond_netpoll_cleanup(dev);
8a8efa22 984 break;
f6dc31a8
WC		 985 }
986 }
8a8efa22 987 return err;
f6dc31a8 988}
8a8efa22
AW		 989#else
990static inline int slave_enable_netpoll(struct slave *slave)
991{
992 return 0;
993}
994static inline void slave_disable_netpoll(struct slave *slave)
995{
996}
f6dc31a8
WC		 997static void bond_netpoll_cleanup(struct net_device *bond_dev)
998{
999}
f6dc31a8
WC		 1000#endif
1001
1da177e4
LT		 1002/*---------------------------------- IOCTL ----------------------------------*/
1003
c8f44aff 1004static netdev_features_t bond_fix_features(struct net_device *dev,
9b7b165a 1005 netdev_features_t features)
8531c5ff 1006{
b2a103e6 1007 struct bonding *bond = netdev_priv(dev);
9caff1e7 1008 struct list_head *iter;
c8f44aff 1009 netdev_features_t mask;
9b7b165a 1010 struct slave *slave;
b63365a2 1011
0965a1f3 1012 if (!bond_has_slaves(bond)) {
b2a103e6
MM		 1013 /* Disable adding VLANs to empty bond. But why? --mq */
1014 features |= NETIF_F_VLAN_CHALLENGED;
9b7b165a 1015 return features;
b2a103e6 1016 }
b63365a2 1017
b2a103e6 1018 mask = features;
b63365a2 1019 features &= ~NETIF_F_ONE_FOR_ALL;
b2a103e6 1020 features |= NETIF_F_ALL_FOR_ALL;
7f353bf2 1021
9caff1e7 1022 bond_for_each_slave(bond, slave, iter) {
b63365a2
HX		 1023 features = netdev_increment_features(features,
1024 slave->dev->features,
b2a103e6
MM		 1025 mask);
1026 }
b0ce3508 1027 features = netdev_add_tso_features(features, mask);
b2a103e6 1028
b2a103e6
MM		 1029 return features;
1030}
1031
62f2a3a4
MM		 1032#define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1033 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1034 NETIF_F_HIGHDMA | NETIF_F_LRO)
b2a103e6
MM		 1035
1036static void bond_compute_features(struct bonding *bond)
1037{
c4826861 1038 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
c8f44aff 1039 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
9caff1e7
VF		 1040 struct net_device *bond_dev = bond->dev;
1041 struct list_head *iter;
1042 struct slave *slave;
b2a103e6 1043 unsigned short max_hard_header_len = ETH_HLEN;
0e376bd0
SB		 1044 unsigned int gso_max_size = GSO_MAX_SIZE;
1045 u16 gso_max_segs = GSO_MAX_SEGS;
b2a103e6 1046
0965a1f3 1047 if (!bond_has_slaves(bond))
b2a103e6
MM		 1048 goto done;
1049
9caff1e7 1050 bond_for_each_slave(bond, slave, iter) {
278339a4 1051 vlan_features = netdev_increment_features(vlan_features,
b2a103e6
MM		 1052 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1053
b6fe83e9 1054 dst_release_flag &= slave->dev->priv_flags;
54ef3137
JV		 1055 if (slave->dev->hard_header_len > max_hard_header_len)
1056 max_hard_header_len = slave->dev->hard_header_len;
0e376bd0
SB		 1057
1058 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1059 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
54ef3137 1060 }
8531c5ff 1061
b63365a2 1062done:
b2a103e6 1063 bond_dev->vlan_features = vlan_features;
54ef3137 1064 bond_dev->hard_header_len = max_hard_header_len;
0e376bd0
SB		 1065 bond_dev->gso_max_segs = gso_max_segs;
1066 netif_set_gso_max_size(bond_dev, gso_max_size);
8531c5ff 1067
b6fe83e9
ED		 1068 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1069 bond_dev->priv_flags = flags | dst_release_flag;
1070
b2a103e6 1071 netdev_change_features(bond_dev);
8531c5ff
AK		 1072}
1073
872254dd
MS		 1074static void bond_setup_by_slave(struct net_device *bond_dev,
1075 struct net_device *slave_dev)
1076{
00829823 1077 bond_dev->header_ops = slave_dev->header_ops;
872254dd
MS		 1078
1079 bond_dev->type = slave_dev->type;
1080 bond_dev->hard_header_len = slave_dev->hard_header_len;
1081 bond_dev->addr_len = slave_dev->addr_len;
1082
1083 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1084 slave_dev->addr_len);
1085}
1086
5b2c4dd2 1087/* On bonding slaves other than the currently active slave, suppress
3aba891d 1088 * duplicates except for alb non-mcast/bcast.
5b2c4dd2
JP		 1089 */
1090static bool bond_should_deliver_exact_match(struct sk_buff *skb,
0bd80dad
JP		 1091 struct slave *slave,
1092 struct bonding *bond)
5b2c4dd2 1093{
2d7011ca 1094 if (bond_is_slave_inactive(slave)) {
0bd80dad 1095 if (bond->params.mode == BOND_MODE_ALB &&
5b2c4dd2
JP		 1096 skb->pkt_type != PACKET_BROADCAST &&
1097 skb->pkt_type != PACKET_MULTICAST)
5b2c4dd2 1098 return false;
5b2c4dd2
JP		 1099 return true;
1100 }
1101 return false;
1102}
1103
8a4eb573 1104static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
5b2c4dd2 1105{
8a4eb573 1106 struct sk_buff *skb = *pskb;
f1c1775a 1107 struct slave *slave;
0bd80dad 1108 struct bonding *bond;
de063b70
ED		 1109 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1110 struct slave *);
13a8e0c8 1111 int ret = RX_HANDLER_ANOTHER;
5b2c4dd2 1112
8a4eb573
JP		 1113 skb = skb_share_check(skb, GFP_ATOMIC);
1114 if (unlikely(!skb))
1115 return RX_HANDLER_CONSUMED;
1116
1117 *pskb = skb;
5b2c4dd2 1118
35d48903
JP		 1119 slave = bond_slave_get_rcu(skb->dev);
1120 bond = slave->bond;
0bd80dad
JP		 1121
1122 if (bond->params.arp_interval)
f1c1775a 1123 slave->dev->last_rx = jiffies;
5b2c4dd2 1124
4d97480b
MH		 1125 recv_probe = ACCESS_ONCE(bond->recv_probe);
1126 if (recv_probe) {
de063b70
ED		 1127 ret = recv_probe(skb, bond, slave);
1128 if (ret == RX_HANDLER_CONSUMED) {
1129 consume_skb(skb);
1130 return ret;
3aba891d
JP		 1131 }
1132 }
1133
0bd80dad 1134 if (bond_should_deliver_exact_match(skb, slave, bond)) {
8a4eb573 1135 return RX_HANDLER_EXACT;
5b2c4dd2
JP		 1136 }
1137
35d48903 1138 skb->dev = bond->dev;
5b2c4dd2 1139
0bd80dad 1140 if (bond->params.mode == BOND_MODE_ALB &&
35d48903 1141 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
5b2c4dd2 1142 skb->pkt_type == PACKET_HOST) {
5b2c4dd2 1143
541ac7c9
CG		 1144 if (unlikely(skb_cow_head(skb,
1145 skb->data - skb_mac_header(skb)))) {
1146 kfree_skb(skb);
8a4eb573 1147 return RX_HANDLER_CONSUMED;
541ac7c9 1148 }
35d48903 1149 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
5b2c4dd2
JP		 1150 }
1151
13a8e0c8 1152 return ret;
5b2c4dd2
JP		 1153}
1154
471cb5a3 1155static int bond_master_upper_dev_link(struct net_device *bond_dev,
1f718f0f
VF		 1156 struct net_device *slave_dev,
1157 struct slave *slave)
471cb5a3
JP		 1158{
1159 int err;
1160
1f718f0f 1161 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
471cb5a3
JP		 1162 if (err)
1163 return err;
1164 slave_dev->flags |= IFF_SLAVE;
7f294054 1165 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
471cb5a3
JP		 1166 return 0;
1167}
1168
1169static void bond_upper_dev_unlink(struct net_device *bond_dev,
1170 struct net_device *slave_dev)
1171{
1172 netdev_upper_dev_unlink(slave_dev, bond_dev);
1173 slave_dev->flags &= ~IFF_SLAVE;
7f294054 1174 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
471cb5a3
JP		 1175}
1176
1da177e4 1177/* enslave device <slave> to bond device <master> */
a77b5325 1178int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1da177e4 1179{
454d7c9b 1180 struct bonding *bond = netdev_priv(bond_dev);
eb7cc59a 1181 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
c8c23903 1182 struct slave *new_slave = NULL, *prev_slave;
1da177e4
LT		 1183 struct sockaddr addr;
1184 int link_reporting;
8599b52e 1185 int res = 0, i;
1da177e4 1186
c772dde3
BH		 1187 if (!bond->params.use_carrier &&
1188 slave_dev->ethtool_ops->get_link == NULL &&
1189 slave_ops->ndo_do_ioctl == NULL) {
a4aee5c8
JP		 1190 pr_warning("%s: Warning: no link monitoring support for %s\n",
1191 bond_dev->name, slave_dev->name);
1da177e4
LT		 1192 }
1193
1da177e4
LT		 1194 /* already enslaved */
1195 if (slave_dev->flags & IFF_SLAVE) {
5a03cdb7 1196 pr_debug("Error, Device was already enslaved\n");
1da177e4
LT		 1197 return -EBUSY;
1198 }
1199
09a89c21
JB		 1200 if (bond_dev == slave_dev) {
1201 pr_err("%s: cannot enslave bond to itself.\n", bond_dev->name);
1202 return -EPERM;
1203 }
1204
1da177e4
LT		 1205 /* vlan challenged mutual exclusion */
1206 /* no need to lock since we're protected by rtnl_lock */
1207 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
5a03cdb7 1208 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
55462cf3 1209 if (vlan_uses_dev(bond_dev)) {
a4aee5c8
JP		 1210 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1211 bond_dev->name, slave_dev->name, bond_dev->name);
1da177e4
LT		 1212 return -EPERM;
1213 } else {
a4aee5c8
JP		 1214 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1215 bond_dev->name, slave_dev->name,
1216 slave_dev->name, bond_dev->name);
1da177e4
LT		 1217 }
1218 } else {
5a03cdb7 1219 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1da177e4
LT		 1220 }
1221
217df670
JV		 1222 /*
1223 * Old ifenslave binaries are no longer supported. These can
3d632c3f 1224 * be identified with moderate accuracy by the state of the slave:
217df670
JV		 1225 * the current ifenslave will set the interface down prior to
1226 * enslaving it; the old ifenslave will not.
1227 */
1228 if ((slave_dev->flags & IFF_UP)) {
a4aee5c8 1229 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
217df670
JV		 1230 slave_dev->name);
1231 res = -EPERM;
1232 goto err_undo_flags;
1233 }
1da177e4 1234
872254dd
MS		 1235 /* set bonding device ether type by slave - bonding netdevices are
1236 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1237 * there is a need to override some of the type dependent attribs/funcs.
1238 *
1239 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1240 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1241 */
0965a1f3 1242 if (!bond_has_slaves(bond)) {
e36b9d16 1243 if (bond_dev->type != slave_dev->type) {
e36b9d16 1244 pr_debug("%s: change device type from %d to %d\n",
a4aee5c8
JP		 1245 bond_dev->name,
1246 bond_dev->type, slave_dev->type);
75c78500 1247
b7bc2a5b
AW		 1248 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1249 bond_dev);
3ca5b404
JP		 1250 res = notifier_to_errno(res);
1251 if (res) {
1252 pr_err("%s: refused to change device type\n",
1253 bond_dev->name);
1254 res = -EBUSY;
1255 goto err_undo_flags;
1256 }
75c78500 1257
32a806c1 1258 /* Flush unicast and multicast addresses */
a748ee24 1259 dev_uc_flush(bond_dev);
22bedad3 1260 dev_mc_flush(bond_dev);
32a806c1 1261
e36b9d16
MS		 1262 if (slave_dev->type != ARPHRD_ETHER)
1263 bond_setup_by_slave(bond_dev, slave_dev);
550fd08c 1264 else {
e36b9d16 1265 ether_setup(bond_dev);
550fd08c
NH		 1266 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1267 }
75c78500 1268
b7bc2a5b
AW		 1269 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1270 bond_dev);
e36b9d16 1271 }
872254dd 1272 } else if (bond_dev->type != slave_dev->type) {
a4aee5c8
JP		 1273 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1274 slave_dev->name,
1275 slave_dev->type, bond_dev->type);
1276 res = -EINVAL;
1277 goto err_undo_flags;
872254dd
MS		 1278 }
1279
eb7cc59a 1280 if (slave_ops->ndo_set_mac_address == NULL) {
0965a1f3 1281 if (!bond_has_slaves(bond)) {
00503b6f 1282 pr_warn("%s: Warning: The first slave device specified does not support setting the MAC address.\n",
1283 bond_dev->name);
1284 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1285 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1286 pr_warn("%s: Setting fail_over_mac to active for active-backup mode.\n",
1287 bond_dev->name);
1288 }
3915c1e8 1289 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
a4aee5c8
JP		 1290 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1291 bond_dev->name);
2ab82852
MS		 1292 res = -EOPNOTSUPP;
1293 goto err_undo_flags;
1294 }
1da177e4
LT		 1295 }
1296
8d8fc29d
AW		 1297 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1298
c20811a7
JP		 1299 /* If this is the first slave, then we need to set the master's hardware
1300 * address to be the same as the slave's. */
0965a1f3 1301 if (!bond_has_slaves(bond) &&
dec1e90e 1302 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
409cc1f8 1303 bond_set_dev_addr(bond->dev, slave_dev);
c20811a7 1304
243cb4e5 1305 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1da177e4
LT		 1306 if (!new_slave) {
1307 res = -ENOMEM;
1308 goto err_undo_flags;
1309 }
bb1d9123
AG		 1310 /*
1311 * Set the new_slave's queue_id to be zero. Queue ID mapping
1312 * is set via sysfs or module option if desired.
1313 */
1314 new_slave->queue_id = 0;
1315
b15ba0fb
JP		 1316 /* Save slave's original mtu and then set it to match the bond */
1317 new_slave->original_mtu = slave_dev->mtu;
1318 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1319 if (res) {
1320 pr_debug("Error %d calling dev_set_mtu\n", res);
1321 goto err_free;
1322 }
1323
217df670
JV		 1324 /*
1325 * Save slave's original ("permanent") mac address for modes
1326 * that need it, and for restoring it upon release, and then
1327 * set it to the master's address
1328 */
1329 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1da177e4 1330
00503b6f 1331 if (!bond->params.fail_over_mac ||
1332 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2ab82852
MS		 1333 /*
1334 * Set slave to master's mac address. The application already
1335 * set the master's mac address to that of the first slave
1336 */
1337 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1338 addr.sa_family = slave_dev->type;
1339 res = dev_set_mac_address(slave_dev, &addr);
1340 if (res) {
5a03cdb7 1341 pr_debug("Error %d calling set_mac_address\n", res);
b15ba0fb 1342 goto err_restore_mtu;
2ab82852 1343 }
217df670 1344 }
1da177e4 1345
217df670
JV		 1346 /* open the slave since the application closed it */
1347 res = dev_open(slave_dev);
1348 if (res) {
3d632c3f 1349 pr_debug("Opening slave %s failed\n", slave_dev->name);
1f718f0f 1350 goto err_restore_mac;
1da177e4
LT		 1351 }
1352
35d48903 1353 new_slave->bond = bond;
1da177e4 1354 new_slave->dev = slave_dev;
0b680e75 1355 slave_dev->priv_flags |= IFF_BONDING;
1da177e4 1356
58402054 1357 if (bond_is_lb(bond)) {
1da177e4
LT		 1358 /* bond_alb_init_slave() must be called before all other stages since
1359 * it might fail and we do not want to have to undo everything
1360 */
1361 res = bond_alb_init_slave(bond, new_slave);
3d632c3f 1362 if (res)
569f0c4d 1363 goto err_close;
1da177e4
LT		 1364 }
1365
303d1cbf
JV		 1366 /* If the mode USES_PRIMARY, then the following is handled by
1367 * bond_change_active_slave().
1da177e4
LT		 1368 */
1369 if (!USES_PRIMARY(bond->params.mode)) {
1370 /* set promiscuity level to new slave */
1371 if (bond_dev->flags & IFF_PROMISC) {
7e1a1ac1
WC		 1372 res = dev_set_promiscuity(slave_dev, 1);
1373 if (res)
1374 goto err_close;
1da177e4
LT		 1375 }
1376
1377 /* set allmulti level to new slave */
1378 if (bond_dev->flags & IFF_ALLMULTI) {
7e1a1ac1
WC		 1379 res = dev_set_allmulti(slave_dev, 1);
1380 if (res)
1381 goto err_close;
1da177e4
LT		 1382 }
1383
b9e40857 1384 netif_addr_lock_bh(bond_dev);
303d1cbf
JV		 1385
1386 dev_mc_sync_multiple(slave_dev, bond_dev);
1387 dev_uc_sync_multiple(slave_dev, bond_dev);
1388
b9e40857 1389 netif_addr_unlock_bh(bond_dev);
1da177e4
LT		 1390 }
1391
1392 if (bond->params.mode == BOND_MODE_8023AD) {
1393 /* add lacpdu mc addr to mc list */
1394 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1395
22bedad3 1396 dev_mc_add(slave_dev, lacpdu_multicast);
1da177e4
LT		 1397 }
1398
b8e2fde4
WY		 1399 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1400 if (res) {
1ff412ad 1401 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1402 bond_dev->name, slave_dev->name);
1403 goto err_close;
1404 }
1da177e4 1405
c8c23903 1406 prev_slave = bond_last_slave(bond);
1da177e4
LT		 1407
1408 new_slave->delay = 0;
1409 new_slave->link_failure_count = 0;
1410
876254ae
VF		 1411 bond_update_speed_duplex(new_slave);
1412
f31c7937
MK		 1413 new_slave->last_arp_rx = jiffies -
1414 (msecs_to_jiffies(bond->params.arp_interval) + 1);
8599b52e
VF		 1415 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1416 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
f5b2b966 1417
1da177e4
LT		 1418 if (bond->params.miimon && !bond->params.use_carrier) {
1419 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1420
1421 if ((link_reporting == -1) && !bond->params.arp_interval) {
1422 /*
1423 * miimon is set but a bonded network driver
1424 * does not support ETHTOOL/MII and
1425 * arp_interval is not set. Note: if
1426 * use_carrier is enabled, we will never go
1427 * here (because netif_carrier is always
1428 * supported); thus, we don't need to change
1429 * the messages for netif_carrier.
1430 */
a4aee5c8 1431 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
4e0952c7 1432 bond_dev->name, slave_dev->name);
1da177e4
LT		 1433 } else if (link_reporting == -1) {
1434 /* unable get link status using mii/ethtool */
a4aee5c8
JP		 1435 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1436 bond_dev->name, slave_dev->name);
1da177e4
LT		 1437 }
1438 }
1439
1440 /* check for initial state */
f31c7937
MK		 1441 if (bond->params.miimon) {
1442 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1443 if (bond->params.updelay) {
1444 new_slave->link = BOND_LINK_BACK;
1445 new_slave->delay = bond->params.updelay;
1446 } else {
1447 new_slave->link = BOND_LINK_UP;
1448 }
1da177e4 1449 } else {
f31c7937 1450 new_slave->link = BOND_LINK_DOWN;
1da177e4 1451 }
f31c7937
MK		 1452 } else if (bond->params.arp_interval) {
1453 new_slave->link = (netif_carrier_ok(slave_dev) ?
1454 BOND_LINK_UP : BOND_LINK_DOWN);
1da177e4 1455 } else {
f31c7937 1456 new_slave->link = BOND_LINK_UP;
1da177e4
LT		 1457 }
1458
f31c7937
MK		 1459 if (new_slave->link != BOND_LINK_DOWN)
1460 new_slave->jiffies = jiffies;
1461 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1462 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1463 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1464
1da177e4
LT		 1465 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1466 /* if there is a primary slave, remember it */
a549952a 1467 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1da177e4 1468 bond->primary_slave = new_slave;
a549952a
JP		 1469 bond->force_primary = true;
1470 }
1da177e4
LT		 1471 }
1472
1473 switch (bond->params.mode) {
1474 case BOND_MODE_ACTIVEBACKUP:
5e5b0665 1475 bond_set_slave_inactive_flags(new_slave,
1476 BOND_SLAVE_NOTIFY_NOW);
1da177e4
LT		 1477 break;
1478 case BOND_MODE_8023AD:
1479 /* in 802.3ad mode, the internal mechanism
1480 * will activate the slaves in the selected
1481 * aggregator
1482 */
5e5b0665 1483 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1da177e4 1484 /* if this is the first slave */
23c147e0 1485 if (!prev_slave) {
1da177e4
LT		 1486 SLAVE_AD_INFO(new_slave).id = 1;
1487 /* Initialize AD with the number of times that the AD timer is called in 1 second
1488 * can be called only after the mac address of the bond is set
1489 */
56d00c67 1490 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1da177e4
LT		 1491 } else {
1492 SLAVE_AD_INFO(new_slave).id =
dec1e90e 1493 SLAVE_AD_INFO(prev_slave).id + 1;
1da177e4
LT		 1494 }
1495
1496 bond_3ad_bind_slave(new_slave);
1497 break;
1498 case BOND_MODE_TLB:
1499 case BOND_MODE_ALB:
e30bc066 1500 bond_set_active_slave(new_slave);
5e5b0665 1501 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1da177e4
LT		 1502 break;
1503 default:
5a03cdb7 1504 pr_debug("This slave is always active in trunk mode\n");
1da177e4
LT		 1505
1506 /* always active in trunk mode */
e30bc066 1507 bond_set_active_slave(new_slave);
1da177e4
LT		 1508
1509 /* In trunking mode there is little meaning to curr_active_slave
1510 * anyway (it holds no special properties of the bond device),
1511 * so we can change it without calling change_active_interface()
1512 */
4e591b93 1513 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
278b2083 1514 rcu_assign_pointer(bond->curr_active_slave, new_slave);
3d632c3f 1515
1da177e4
LT		 1516 break;
1517 } /* switch(bond_mode) */
1518
f6dc31a8 1519#ifdef CONFIG_NET_POLL_CONTROLLER
10eccb46 1520 slave_dev->npinfo = bond->dev->npinfo;
8a8efa22
AW		 1521 if (slave_dev->npinfo) {
1522 if (slave_enable_netpoll(new_slave)) {
8a8efa22
AW		 1523 pr_info("Error, %s: master_dev is using netpoll, "
1524 "but new slave device does not support netpoll.\n",
1525 bond_dev->name);
1526 res = -EBUSY;
f7d9821a 1527 goto err_detach;
8a8efa22 1528 }
f6dc31a8
WC		 1529 }
1530#endif
8a8efa22 1531
35d48903
JP		 1532 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1533 new_slave);
1534 if (res) {
1535 pr_debug("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) {
1541 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1542 goto err_unregister;
1543 }
1544
07699f9a 1545 res = bond_sysfs_slave_add(new_slave);
1546 if (res) {
1547 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1548 goto err_upper_unlink;
1549 }
1550
5378c2e6
VF		 1551 bond->slave_cnt++;
1552 bond_compute_features(bond);
1553 bond_set_carrier(bond);
1554
1555 if (USES_PRIMARY(bond->params.mode)) {
f80889a5 1556 block_netpoll_tx();
5378c2e6
VF		 1557 write_lock_bh(&bond->curr_slave_lock);
1558 bond_select_active_slave(bond);
1559 write_unlock_bh(&bond->curr_slave_lock);
f80889a5 1560 unblock_netpoll_tx();
5378c2e6 1561 }
1f718f0f 1562
a4aee5c8
JP		 1563 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1564 bond_dev->name, slave_dev->name,
e30bc066 1565 bond_is_active_slave(new_slave) ? "n active" : " backup",
a4aee5c8 1566 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1da177e4
LT		 1567
1568 /* enslave is successful */
1569 return 0;
1570
1571/* Undo stages on error */
07699f9a 1572err_upper_unlink:
1573 bond_upper_dev_unlink(bond_dev, slave_dev);
1574
1f718f0f
VF		 1575err_unregister:
1576 netdev_rx_handler_unregister(slave_dev);
1577
f7d9821a 1578err_detach:
303d1cbf
JV		 1579 if (!USES_PRIMARY(bond->params.mode))
1580 bond_hw_addr_flush(bond_dev, slave_dev);
1581
1ff412ad 1582 vlan_vids_del_by_dev(slave_dev, bond_dev);
3c5913b5 1583 if (bond->primary_slave == new_slave)
1584 bond->primary_slave = NULL;
3c5913b5 1585 if (bond->curr_active_slave == new_slave) {
f80889a5 1586 block_netpoll_tx();
3c5913b5 1587 write_lock_bh(&bond->curr_slave_lock);
c8517035 1588 bond_change_active_slave(bond, NULL);
3c5913b5 1589 bond_select_active_slave(bond);
1590 write_unlock_bh(&bond->curr_slave_lock);
f80889a5 1591 unblock_netpoll_tx();
3c5913b5 1592 }
fc7a72ac 1593 slave_disable_netpoll(new_slave);
f7d9821a 1594
1da177e4 1595err_close:
b6a5a7b9 1596 slave_dev->priv_flags &= ~IFF_BONDING;
1da177e4
LT		 1597 dev_close(slave_dev);
1598
1599err_restore_mac:
00503b6f 1600 if (!bond->params.fail_over_mac ||
1601 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
3915c1e8
JV		 1602 /* XXX TODO - fom follow mode needs to change master's
1603 * MAC if this slave's MAC is in use by the bond, or at
1604 * least print a warning.
1605 */
2ab82852
MS		 1606 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1607 addr.sa_family = slave_dev->type;
1608 dev_set_mac_address(slave_dev, &addr);
1609 }
1da177e4 1610
b15ba0fb
JP		 1611err_restore_mtu:
1612 dev_set_mtu(slave_dev, new_slave->original_mtu);
1613
1da177e4
LT		 1614err_free:
1615 kfree(new_slave);
1616
1617err_undo_flags:
b8fad459 1618 /* Enslave of first slave has failed and we need to fix master's mac */
0965a1f3 1619 if (!bond_has_slaves(bond) &&
844223ab 1620 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
b8fad459 1621 eth_hw_addr_random(bond_dev);
3d632c3f 1622
1da177e4
LT		 1623 return res;
1624}
1625
1626/*
1627 * Try to release the slave device <slave> from the bond device <master>
1628 * It is legal to access curr_active_slave without a lock because all the function
0896341a 1629 * is write-locked. If "all" is true it means that the function is being called
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)) {
a4aee5c8 1651 pr_err("%s: Error: cannot release %s.\n",
1da177e4
LT		 1652 bond_dev->name, slave_dev->name);
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 */
a4aee5c8
JP		 1661 pr_info("%s: %s not enslaved\n",
1662 bond_dev->name, slave_dev->name);
e843fa50 1663 unblock_netpoll_tx();
1da177e4
LT		 1664 return -EINVAL;
1665 }
1666
07699f9a 1667 bond_sysfs_slave_del(slave);
1668
1f718f0f 1669 bond_upper_dev_unlink(bond_dev, slave_dev);
35d48903
JP		 1670 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1671 * for this slave anymore.
1672 */
1673 netdev_rx_handler_unregister(slave_dev);
35d48903
JP		 1674 write_lock_bh(&bond->lock);
1675
1da177e4 1676 /* Inform AD package of unbinding of slave. */
be79bd04 1677 if (bond->params.mode == BOND_MODE_8023AD)
1da177e4 1678 bond_3ad_unbind_slave(slave);
be79bd04 1679
c8517035 1680 write_unlock_bh(&bond->lock);
1da177e4 1681
a4aee5c8
JP		 1682 pr_info("%s: releasing %s interface %s\n",
1683 bond_dev->name,
e30bc066 1684 bond_is_active_slave(slave) ? "active" : "backup",
a4aee5c8 1685 slave_dev->name);
1da177e4
LT		 1686
1687 oldcurrent = bond->curr_active_slave;
1688
1689 bond->current_arp_slave = NULL;
1690
00503b6f 1691 if (!all && (!bond->params.fail_over_mac ||
1692 bond->params.mode != BOND_MODE_ACTIVEBACKUP)) {
844223ab 1693 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
0965a1f3 1694 bond_has_slaves(bond))
dec1e90e 1695 pr_warn("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1696 bond_dev->name, slave_dev->name,
1697 slave->perm_hwaddr,
1698 bond_dev->name, slave_dev->name);
1699 }
1700
3d632c3f 1701 if (bond->primary_slave == slave)
1da177e4 1702 bond->primary_slave = NULL;
1da177e4 1703
c8517035 1704 if (oldcurrent == slave) {
1705 write_lock_bh(&bond->curr_slave_lock);
1da177e4 1706 bond_change_active_slave(bond, NULL);
c8517035 1707 write_unlock_bh(&bond->curr_slave_lock);
1708 }
1da177e4 1709
58402054 1710 if (bond_is_lb(bond)) {
1da177e4
LT		 1711 /* Must be called only after the slave has been
1712 * detached from the list and the curr_active_slave
1713 * has been cleared (if our_slave == old_current),
1714 * but before a new active slave is selected.
1715 */
1716 bond_alb_deinit_slave(bond, slave);
1717 }
1718
0896341a 1719 if (all) {
36708b89 1720 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
0896341a 1721 } else if (oldcurrent == slave) {
059fe7a5
JV		 1722 /*
1723 * Note that we hold RTNL over this sequence, so there
1724 * is no concern that another slave add/remove event
1725 * will interfere.
1726 */
059fe7a5
JV		 1727 write_lock_bh(&bond->curr_slave_lock);
1728
1da177e4
LT		 1729 bond_select_active_slave(bond);
1730
059fe7a5 1731 write_unlock_bh(&bond->curr_slave_lock);
059fe7a5
JV		 1732 }
1733
0965a1f3 1734 if (!bond_has_slaves(bond)) {
ff59c456 1735 bond_set_carrier(bond);
409cc1f8 1736 eth_hw_addr_random(bond_dev);
1da177e4 1737
a59d3d21 1738 if (vlan_uses_dev(bond_dev)) {
a4aee5c8
JP		 1739 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1740 bond_dev->name, bond_dev->name);
1741 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1742 bond_dev->name);
1da177e4 1743 }
1da177e4
LT		 1744 }
1745
e843fa50 1746 unblock_netpoll_tx();
278b2083 1747 synchronize_rcu();
ee6154e1 1748 bond->slave_cnt--;
1da177e4 1749
0965a1f3 1750 if (!bond_has_slaves(bond)) {
2af73d4b 1751 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
80028ea1
VF		 1752 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1753 }
2af73d4b 1754
b2a103e6
MM		 1755 bond_compute_features(bond);
1756 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1757 (old_features & NETIF_F_VLAN_CHALLENGED))
1758 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1759 bond_dev->name, slave_dev->name, bond_dev->name);
1760
b76cdba9 1761 /* must do this from outside any spinlocks */
1ff412ad 1762 vlan_vids_del_by_dev(slave_dev, bond_dev);
1da177e4 1763
303d1cbf
JV		 1764 /* If the mode USES_PRIMARY, then this cases was handled above by
1765 * bond_change_active_slave(..., NULL)
1da177e4
LT		 1766 */
1767 if (!USES_PRIMARY(bond->params.mode)) {
5a0068de
NH		 1768 /* unset promiscuity level from slave
1769 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1770 * of the IFF_PROMISC flag in the bond_dev, but we need the
1771 * value of that flag before that change, as that was the value
1772 * when this slave was attached, so we cache at the start of the
1773 * function and use it here. Same goes for ALLMULTI below
1774 */
1775 if (old_flags & IFF_PROMISC)
1da177e4 1776 dev_set_promiscuity(slave_dev, -1);
1da177e4
LT		 1777
1778 /* unset allmulti level from slave */
5a0068de 1779 if (old_flags & IFF_ALLMULTI)
1da177e4 1780 dev_set_allmulti(slave_dev, -1);
1da177e4 1781
303d1cbf 1782 bond_hw_addr_flush(bond_dev, slave_dev);
1da177e4
LT		 1783 }
1784
8a8efa22 1785 slave_disable_netpoll(slave);
f6dc31a8 1786
1da177e4
LT		 1787 /* close slave before restoring its mac address */
1788 dev_close(slave_dev);
1789
00503b6f 1790 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1791 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2ab82852
MS		 1792 /* restore original ("permanent") mac address */
1793 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1794 addr.sa_family = slave_dev->type;
1795 dev_set_mac_address(slave_dev, &addr);
1796 }
1da177e4 1797
b15ba0fb
JP		 1798 dev_set_mtu(slave_dev, slave->original_mtu);
1799
2d7011ca 1800 slave_dev->priv_flags &= ~IFF_BONDING;
1da177e4
LT		 1801
1802 kfree(slave);
1803
1804 return 0; /* deletion OK */
1805}
1806
0896341a 1807/* A wrapper used because of ndo_del_link */
1808int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1809{
1810 return __bond_release_one(bond_dev, slave_dev, false);
1811}
1812
d90a162a 1813/*
dadaa10b 1814* First release a slave and then destroy the bond if no more slaves are left.
d90a162a
MS		 1815* Must be under rtnl_lock when this function is called.
1816*/
26d8ee75 1817static int bond_release_and_destroy(struct net_device *bond_dev,
1818 struct net_device *slave_dev)
d90a162a 1819{
454d7c9b 1820 struct bonding *bond = netdev_priv(bond_dev);
d90a162a
MS		 1821 int ret;
1822
1823 ret = bond_release(bond_dev, slave_dev);
0965a1f3 1824 if (ret == 0 && !bond_has_slaves(bond)) {
8a8efa22 1825 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
a4aee5c8
JP		 1826 pr_info("%s: destroying bond %s.\n",
1827 bond_dev->name, bond_dev->name);
9e71626c 1828 unregister_netdevice(bond_dev);
d90a162a
MS		 1829 }
1830 return ret;
1831}
1832
1da177e4
LT		 1833static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1834{
454d7c9b 1835 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 1836
1837 info->bond_mode = bond->params.mode;
1838 info->miimon = bond->params.miimon;
1839
6603a6f2 1840 read_lock(&bond->lock);
1da177e4 1841 info->num_slaves = bond->slave_cnt;
6603a6f2 1842 read_unlock(&bond->lock);
1da177e4
LT		 1843
1844 return 0;
1845}
1846
1847static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1848{
454d7c9b 1849 struct bonding *bond = netdev_priv(bond_dev);
9caff1e7 1850 struct list_head *iter;
dec1e90e 1851 int i = 0, res = -ENODEV;
1da177e4 1852 struct slave *slave;
1da177e4 1853
6603a6f2 1854 read_lock(&bond->lock);
9caff1e7 1855 bond_for_each_slave(bond, slave, iter) {
dec1e90e 1856 if (i++ == (int)info->slave_id) {
689c96cc
ED		 1857 res = 0;
1858 strcpy(info->slave_name, slave->dev->name);
1859 info->link = slave->link;
e30bc066 1860 info->state = bond_slave_state(slave);
689c96cc 1861 info->link_failure_count = slave->link_failure_count;
1da177e4
LT		 1862 break;
1863 }
1864 }
6603a6f2 1865 read_unlock(&bond->lock);
1da177e4 1866
689c96cc 1867 return res;
1da177e4
LT		 1868}
1869
1870/*-------------------------------- Monitoring -------------------------------*/
1871
1da177e4 1872
f0c76d61
JV		 1873static int bond_miimon_inspect(struct bonding *bond)
1874{
dec1e90e 1875 int link_state, commit = 0;
9caff1e7 1876 struct list_head *iter;
f0c76d61 1877 struct slave *slave;
41f89100
JP		 1878 bool ignore_updelay;
1879
1880 ignore_updelay = !bond->curr_active_slave ? true : false;
1da177e4 1881
4cb4f97b 1882 bond_for_each_slave_rcu(bond, slave, iter) {
f0c76d61 1883 slave->new_link = BOND_LINK_NOCHANGE;
1da177e4 1884
f0c76d61 1885 link_state = bond_check_dev_link(bond, slave->dev, 0);
1da177e4
LT		 1886
1887 switch (slave->link) {
f0c76d61
JV		 1888 case BOND_LINK_UP:
1889 if (link_state)
1890 continue;
1da177e4 1891
f0c76d61
JV		 1892 slave->link = BOND_LINK_FAIL;
1893 slave->delay = bond->params.downdelay;
1894 if (slave->delay) {
a4aee5c8
JP		 1895 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1896 bond->dev->name,
1897 (bond->params.mode ==
1898 BOND_MODE_ACTIVEBACKUP) ?
e30bc066 1899 (bond_is_active_slave(slave) ?
a4aee5c8
JP		 1900 "active " : "backup ") : "",
1901 slave->dev->name,
1902 bond->params.downdelay * bond->params.miimon);
1da177e4 1903 }
f0c76d61
JV		 1904 /*FALLTHRU*/
1905 case BOND_LINK_FAIL:
1906 if (link_state) {
1907 /*
1908 * recovered before downdelay expired
1909 */
1910 slave->link = BOND_LINK_UP;
1da177e4 1911 slave->jiffies = jiffies;
a4aee5c8
JP		 1912 pr_info("%s: link status up again after %d ms for interface %s.\n",
1913 bond->dev->name,
1914 (bond->params.downdelay - slave->delay) *
1915 bond->params.miimon,
1916 slave->dev->name);
f0c76d61 1917 continue;
1da177e4 1918 }
f0c76d61
JV		 1919
1920 if (slave->delay <= 0) {
1921 slave->new_link = BOND_LINK_DOWN;
1922 commit++;
1923 continue;
1da177e4 1924 }
1da177e4 1925
f0c76d61
JV		 1926 slave->delay--;
1927 break;
1928
1929 case BOND_LINK_DOWN:
1930 if (!link_state)
1931 continue;
1932
1933 slave->link = BOND_LINK_BACK;
1934 slave->delay = bond->params.updelay;
1935
1936 if (slave->delay) {
a4aee5c8
JP		 1937 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1938 bond->dev->name, slave->dev->name,
1939 ignore_updelay ? 0 :
1940 bond->params.updelay *
1941 bond->params.miimon);
f0c76d61
JV		 1942 }
1943 /*FALLTHRU*/
1944 case BOND_LINK_BACK:
1945 if (!link_state) {
1946 slave->link = BOND_LINK_DOWN;
a4aee5c8
JP		 1947 pr_info("%s: link status down again after %d ms for interface %s.\n",
1948 bond->dev->name,
1949 (bond->params.updelay - slave->delay) *
1950 bond->params.miimon,
1951 slave->dev->name);
f0c76d61
JV		 1952
1953 continue;
1954 }
1955
41f89100
JP		 1956 if (ignore_updelay)
1957 slave->delay = 0;
1958
f0c76d61
JV		 1959 if (slave->delay <= 0) {
1960 slave->new_link = BOND_LINK_UP;
1961 commit++;
41f89100 1962 ignore_updelay = false;
f0c76d61 1963 continue;
1da177e4 1964 }
f0c76d61
JV		 1965
1966 slave->delay--;
1da177e4 1967 break;
f0c76d61
JV		 1968 }
1969 }
1da177e4 1970
f0c76d61
JV		 1971 return commit;
1972}
1da177e4 1973
f0c76d61
JV		 1974static void bond_miimon_commit(struct bonding *bond)
1975{
9caff1e7 1976 struct list_head *iter;
f0c76d61 1977 struct slave *slave;
f0c76d61 1978
9caff1e7 1979 bond_for_each_slave(bond, slave, iter) {
f0c76d61
JV		 1980 switch (slave->new_link) {
1981 case BOND_LINK_NOCHANGE:
1982 continue;
1da177e4 1983
f0c76d61
JV		 1984 case BOND_LINK_UP:
1985 slave->link = BOND_LINK_UP;
1986 slave->jiffies = jiffies;
1987
1988 if (bond->params.mode == BOND_MODE_8023AD) {
1989 /* prevent it from being the active one */
e30bc066 1990 bond_set_backup_slave(slave);
f0c76d61
JV		 1991 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1992 /* make it immediately active */
e30bc066 1993 bond_set_active_slave(slave);
f0c76d61
JV		 1994 } else if (slave != bond->primary_slave) {
1995 /* prevent it from being the active one */
e30bc066 1996 bond_set_backup_slave(slave);
1da177e4 1997 }
1da177e4 1998
5d30530e 1999 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
700c2a77 2000 bond->dev->name, slave->dev->name,
db4e9b2b
NA		 2001 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2002 slave->duplex ? "full" : "half");
1da177e4 2003
f0c76d61
JV		 2004 /* notify ad that the link status has changed */
2005 if (bond->params.mode == BOND_MODE_8023AD)
2006 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
059fe7a5 2007
58402054 2008 if (bond_is_lb(bond))
f0c76d61
JV		 2009 bond_alb_handle_link_change(bond, slave,
2010 BOND_LINK_UP);
1da177e4 2011
f0c76d61
JV		 2012 if (!bond->curr_active_slave ||
2013 (slave == bond->primary_slave))
2014 goto do_failover;
1da177e4 2015
f0c76d61 2016 continue;
059fe7a5 2017
f0c76d61 2018 case BOND_LINK_DOWN:
fba4acda
JV		 2019 if (slave->link_failure_count < UINT_MAX)
2020 slave->link_failure_count++;
2021
f0c76d61 2022 slave->link = BOND_LINK_DOWN;
1da177e4 2023
f0c76d61
JV		 2024 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2025 bond->params.mode == BOND_MODE_8023AD)
5e5b0665 2026 bond_set_slave_inactive_flags(slave,
2027 BOND_SLAVE_NOTIFY_NOW);
f0c76d61 2028
a4aee5c8
JP		 2029 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2030 bond->dev->name, slave->dev->name);
f0c76d61
JV		 2031
2032 if (bond->params.mode == BOND_MODE_8023AD)
2033 bond_3ad_handle_link_change(slave,
2034 BOND_LINK_DOWN);
2035
ae63e808 2036 if (bond_is_lb(bond))
f0c76d61
JV		 2037 bond_alb_handle_link_change(bond, slave,
2038 BOND_LINK_DOWN);
2039
2040 if (slave == bond->curr_active_slave)
2041 goto do_failover;
2042
2043 continue;
2044
2045 default:
a4aee5c8 2046 pr_err("%s: invalid new link %d on slave %s\n",
f0c76d61
JV		 2047 bond->dev->name, slave->new_link,
2048 slave->dev->name);
2049 slave->new_link = BOND_LINK_NOCHANGE;
2050
2051 continue;
2052 }
2053
2054do_failover:
2055 ASSERT_RTNL();
e843fa50 2056 block_netpoll_tx();
f0c76d61
JV		 2057 write_lock_bh(&bond->curr_slave_lock);
2058 bond_select_active_slave(bond);
2059 write_unlock_bh(&bond->curr_slave_lock);
e843fa50 2060 unblock_netpoll_tx();
f0c76d61
JV		 2061 }
2062
2063 bond_set_carrier(bond);
1da177e4
LT		 2064}
2065
0b0eef66
JV		 2066/*
2067 * bond_mii_monitor
2068 *
2069 * Really a wrapper that splits the mii monitor into two phases: an
f0c76d61
JV		 2070 * inspection, then (if inspection indicates something needs to be done)
2071 * an acquisition of appropriate locks followed by a commit phase to
2072 * implement whatever link state changes are indicated.
0b0eef66 2073 */
6da67d26 2074static void bond_mii_monitor(struct work_struct *work)
0b0eef66
JV		 2075{
2076 struct bonding *bond = container_of(work, struct bonding,
2077 mii_work.work);
ad246c99 2078 bool should_notify_peers = false;
1f2cd845 2079 unsigned long delay;
0b0eef66 2080
1f2cd845
DM		 2081 delay = msecs_to_jiffies(bond->params.miimon);
2082
2083 if (!bond_has_slaves(bond))
f0c76d61 2084 goto re_arm;
b59f9f74 2085
4cb4f97b 2086 rcu_read_lock();
2087
ad246c99
BH		 2088 should_notify_peers = bond_should_notify_peers(bond);
2089
1f2cd845 2090 if (bond_miimon_inspect(bond)) {
4cb4f97b 2091 rcu_read_unlock();
f0c76d61 2092
1f2cd845
DM		 2093 /* Race avoidance with bond_close cancel of workqueue */
2094 if (!rtnl_trylock()) {
1f2cd845
DM		 2095 delay = 1;
2096 should_notify_peers = false;
2097 goto re_arm;
2098 }
6b6c5261 2099
1f2cd845
DM		 2100 bond_miimon_commit(bond);
2101
1f2cd845 2102 rtnl_unlock(); /* might sleep, hold no other locks */
4cb4f97b 2103 } else
2104 rcu_read_unlock();
0b0eef66 2105
f0c76d61 2106re_arm:
e6d265e8 2107 if (bond->params.miimon)
1f2cd845
DM		 2108 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2109
1f2cd845
DM		 2110 if (should_notify_peers) {
2111 if (!rtnl_trylock())
2112 return;
2113 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2114 rtnl_unlock();
2115 }
0b0eef66 2116}
c3ade5ca 2117
50223ce4 2118static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
f5b2b966 2119{
50223ce4
VF		 2120 struct net_device *upper;
2121 struct list_head *iter;
2122 bool ret = false;
f5b2b966 2123
eaddcd76 2124 if (ip == bond_confirm_addr(bond->dev, 0, ip))
50223ce4 2125 return true;
f5b2b966 2126
50223ce4 2127 rcu_read_lock();
2f268f12 2128 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
50223ce4
VF		 2129 if (ip == bond_confirm_addr(upper, 0, ip)) {
2130 ret = true;
2131 break;
2132 }
f5b2b966 2133 }
50223ce4 2134 rcu_read_unlock();
f5b2b966 2135
50223ce4 2136 return ret;
f5b2b966
JV		 2137}
2138
c3ade5ca
JV		 2139/*
2140 * We go to the (large) trouble of VLAN tagging ARP frames because
2141 * switches in VLAN mode (especially if ports are configured as
2142 * "native" to a VLAN) might not pass non-tagged frames.
2143 */
d3bb52b0 2144static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
c3ade5ca
JV		 2145{
2146 struct sk_buff *skb;
2147
acca2674 2148 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2149 slave_dev->name, &dest_ip, &src_ip, vlan_id);
3d632c3f 2150
c3ade5ca
JV		 2151 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2152 NULL, slave_dev->dev_addr, NULL);
2153
2154 if (!skb) {
a4aee5c8 2155 pr_err("ARP packet allocation failed\n");
c3ade5ca
JV		 2156 return;
2157 }
2158 if (vlan_id) {
1fd9b1fc 2159 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
c3ade5ca 2160 if (!skb) {
a4aee5c8 2161 pr_err("failed to insert VLAN tag\n");
c3ade5ca
JV		 2162 return;
2163 }
2164 }
2165 arp_xmit(skb);
2166}
2167
2168
1da177e4
LT		 2169static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2170{
27bc11e6
VF		 2171 struct net_device *upper, *vlan_upper;
2172 struct list_head *iter, *vlan_iter;
c3ade5ca 2173 struct rtable *rt;
27bc11e6
VF		 2174 __be32 *targets = bond->params.arp_targets, addr;
2175 int i, vlan_id;
1da177e4 2176
27bc11e6 2177 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
acca2674 2178 pr_debug("basa: target %pI4\n", &targets[i]);
c3ade5ca 2179
27bc11e6 2180 /* Find out through which dev should the packet go */
78fbfd8a
DM		 2181 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2182 RTO_ONLINK, 0);
b23dd4fe 2183 if (IS_ERR(rt)) {
3e32582f
VF		 2184 pr_debug("%s: no route to arp_ip_target %pI4\n",
2185 bond->dev->name, &targets[i]);
c3ade5ca
JV		 2186 continue;
2187 }
2188
c3ade5ca 2189 vlan_id = 0;
27bc11e6
VF		 2190
2191 /* bond device itself */
2192 if (rt->dst.dev == bond->dev)
2193 goto found;
2194
2195 rcu_read_lock();
2196 /* first we search only for vlan devices. for every vlan
2197 * found we verify its upper dev list, searching for the
2198 * rt->dst.dev. If found we save the tag of the vlan and
2199 * proceed to send the packet.
2200 *
2201 * TODO: QinQ?
2202 */
2f268f12
VF		 2203 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2204 vlan_iter) {
27bc11e6
VF		 2205 if (!is_vlan_dev(vlan_upper))
2206 continue;
2f268f12
VF		 2207 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2208 iter) {
27bc11e6
VF		 2209 if (upper == rt->dst.dev) {
2210 vlan_id = vlan_dev_vlan_id(vlan_upper);
2211 rcu_read_unlock();
2212 goto found;
2213 }
c3ade5ca
JV		 2214 }
2215 }
2216
27bc11e6
VF		 2217 /* if the device we're looking for is not on top of any of
2218 * our upper vlans, then just search for any dev that
2219 * matches, and in case it's a vlan - save the id
2220 */
2f268f12 2221 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
27bc11e6
VF		 2222 if (upper == rt->dst.dev) {
2223 /* if it's a vlan - get its VID */
2224 if (is_vlan_dev(upper))
2225 vlan_id = vlan_dev_vlan_id(upper);
2226
2227 rcu_read_unlock();
2228 goto found;
2229 }
c3ade5ca 2230 }
27bc11e6 2231 rcu_read_unlock();
c3ade5ca 2232
27bc11e6 2233 /* Not our device - skip */
3e32582f
VF		 2234 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2235 bond->dev->name, &targets[i],
2236 rt->dst.dev ? rt->dst.dev->name : "NULL");
2237
ed4b9f80 2238 ip_rt_put(rt);
27bc11e6
VF		 2239 continue;
2240
2241found:
2242 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2243 ip_rt_put(rt);
2244 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2245 addr, vlan_id);
c3ade5ca
JV		 2246 }
2247}
2248
d3bb52b0 2249static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
f5b2b966 2250{
8599b52e
VF		 2251 int i;
2252
87a7b84b
VF		 2253 if (!sip || !bond_has_this_ip(bond, tip)) {
2254 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2255 return;
2256 }
f5b2b966 2257
8599b52e
VF		 2258 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2259 if (i == -1) {
87a7b84b
VF		 2260 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2261 return;
f5b2b966 2262 }
87a7b84b 2263 slave->last_arp_rx = jiffies;
8599b52e 2264 slave->target_last_arp_rx[i] = jiffies;
f5b2b966
JV		 2265}
2266
5bb9e0b5 2267int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2268 struct slave *slave)
f5b2b966 2269{
de063b70 2270 struct arphdr *arp = (struct arphdr *)skb->data;
f5b2b966 2271 unsigned char *arp_ptr;
d3bb52b0 2272 __be32 sip, tip;
de063b70 2273 int alen;
f5b2b966 2274
3aba891d 2275 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
13a8e0c8 2276 return RX_HANDLER_ANOTHER;
f5b2b966 2277
f5b2b966 2278 read_lock(&bond->lock);
2c146102
VF		 2279
2280 if (!slave_do_arp_validate(bond, slave))
2281 goto out_unlock;
2282
de063b70 2283 alen = arp_hdr_len(bond->dev);
f5b2b966 2284
3aba891d
JP		 2285 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2286 bond->dev->name, skb->dev->name);
f5b2b966 2287
de063b70
ED		 2288 if (alen > skb_headlen(skb)) {
2289 arp = kmalloc(alen, GFP_ATOMIC);
2290 if (!arp)
2291 goto out_unlock;
2292 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2293 goto out_unlock;
2294 }
f5b2b966 2295
3aba891d 2296 if (arp->ar_hln != bond->dev->addr_len ||
f5b2b966
JV		 2297 skb->pkt_type == PACKET_OTHERHOST ||
2298 skb->pkt_type == PACKET_LOOPBACK ||
2299 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2300 arp->ar_pro != htons(ETH_P_IP) ||
2301 arp->ar_pln != 4)
2302 goto out_unlock;
2303
2304 arp_ptr = (unsigned char *)(arp + 1);
3aba891d 2305 arp_ptr += bond->dev->addr_len;
f5b2b966 2306 memcpy(&sip, arp_ptr, 4);
3aba891d 2307 arp_ptr += 4 + bond->dev->addr_len;
f5b2b966
JV		 2308 memcpy(&tip, arp_ptr, 4);
2309
5a03cdb7 2310 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
e30bc066 2311 bond->dev->name, slave->dev->name, bond_slave_state(slave),
a4aee5c8
JP		 2312 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2313 &sip, &tip);
f5b2b966
JV		 2314
2315 /*
2316 * Backup slaves won't see the ARP reply, but do come through
2317 * here for each ARP probe (so we swap the sip/tip to validate
2318 * the probe). In a "redundant switch, common router" type of
2319 * configuration, the ARP probe will (hopefully) travel from
2320 * the active, through one switch, the router, then the other
2321 * switch before reaching the backup.
aeea64ac
VF		 2322 *
2323 * We 'trust' the arp requests if there is an active slave and
2324 * it received valid arp reply(s) after it became active. This
2325 * is done to avoid endless looping when we can't reach the
2326 * arp_ip_target and fool ourselves with our own arp requests.
f5b2b966 2327 */
e30bc066 2328 if (bond_is_active_slave(slave))
f5b2b966 2329 bond_validate_arp(bond, slave, sip, tip);
aeea64ac
VF		 2330 else if (bond->curr_active_slave &&
2331 time_after(slave_last_rx(bond, bond->curr_active_slave),
2332 bond->curr_active_slave->jiffies))
f5b2b966
JV		 2333 bond_validate_arp(bond, slave, tip, sip);
2334
2335out_unlock:
2336 read_unlock(&bond->lock);
de063b70
ED		 2337 if (arp != (struct arphdr *)skb->data)
2338 kfree(arp);
13a8e0c8 2339 return RX_HANDLER_ANOTHER;
f5b2b966
JV		 2340}
2341
e7f63f1d
VF		 2342/* function to verify if we're in the arp_interval timeslice, returns true if
2343 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2344 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2345 */
2346static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2347 int mod)
2348{
2349 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2350
2351 return time_in_range(jiffies,
2352 last_act - delta_in_ticks,
2353 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2354}
2355
1da177e4
LT		 2356/*
2357 * this function is called regularly to monitor each slave's link
2358 * ensuring that traffic is being sent and received when arp monitoring
2359 * is used in load-balancing mode. if the adapter has been dormant, then an
2360 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2361 * arp monitoring in active backup mode.
2362 */
6da67d26 2363static void bond_loadbalance_arp_mon(struct work_struct *work)
1da177e4 2364{
1b76b316
JV		 2365 struct bonding *bond = container_of(work, struct bonding,
2366 arp_work.work);
1da177e4 2367 struct slave *slave, *oldcurrent;
9caff1e7 2368 struct list_head *iter;
6fde8f03 2369 int do_failover = 0, slave_state_changed = 0;
1da177e4 2370
1f2cd845 2371 if (!bond_has_slaves(bond))
1da177e4 2372 goto re_arm;
1da177e4 2373
2e52f4fe 2374 rcu_read_lock();
2375
2376 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
1da177e4
LT		 2377 /* see if any of the previous devices are up now (i.e. they have
2378 * xmt and rcv traffic). the curr_active_slave does not come into
2379 * the picture unless it is null. also, slave->jiffies is not needed
2380 * here because we send an arp on each slave and give a slave as
2381 * long as it needs to get the tx/rx within the delta.
2382 * TODO: what about up/down delay in arp mode? it wasn't here before
2383 * so it can wait
2384 */
2e52f4fe 2385 bond_for_each_slave_rcu(bond, slave, iter) {
cb32f2a0
JB		 2386 unsigned long trans_start = dev_trans_start(slave->dev);
2387
1da177e4 2388 if (slave->link != BOND_LINK_UP) {
e7f63f1d
VF		 2389 if (bond_time_in_interval(bond, trans_start, 1) &&
2390 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
1da177e4
LT		 2391
2392 slave->link = BOND_LINK_UP;
6fde8f03 2393 slave_state_changed = 1;
1da177e4
LT		 2394
2395 /* primary_slave has no meaning in round-robin
2396 * mode. the window of a slave being up and
2397 * curr_active_slave being null after enslaving
2398 * is closed.
2399 */
2400 if (!oldcurrent) {
a4aee5c8
JP		 2401 pr_info("%s: link status definitely up for interface %s, ",
2402 bond->dev->name,
2403 slave->dev->name);
1da177e4
LT		 2404 do_failover = 1;
2405 } else {
a4aee5c8
JP		 2406 pr_info("%s: interface %s is now up\n",
2407 bond->dev->name,
2408 slave->dev->name);
1da177e4
LT		 2409 }
2410 }
2411 } else {
2412 /* slave->link == BOND_LINK_UP */
2413
2414 /* not all switches will respond to an arp request
2415 * when the source ip is 0, so don't take the link down
2416 * if we don't know our ip yet
2417 */
e7f63f1d
VF		 2418 if (!bond_time_in_interval(bond, trans_start, 2) ||
2419 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
1da177e4
LT		 2420
2421 slave->link = BOND_LINK_DOWN;
6fde8f03 2422 slave_state_changed = 1;
1da177e4 2423
3d632c3f 2424 if (slave->link_failure_count < UINT_MAX)
1da177e4 2425 slave->link_failure_count++;
1da177e4 2426
a4aee5c8
JP		 2427 pr_info("%s: interface %s is now down.\n",
2428 bond->dev->name,
2429 slave->dev->name);
1da177e4 2430
3d632c3f 2431 if (slave == oldcurrent)
1da177e4 2432 do_failover = 1;
1da177e4
LT		 2433 }
2434 }
2435
2436 /* note: if switch is in round-robin mode, all links
2437 * must tx arp to ensure all links rx an arp - otherwise
2438 * links may oscillate or not come up at all; if switch is
2439 * in something like xor mode, there is nothing we can
2440 * do - all replies will be rx'ed on same link causing slaves
2441 * to be unstable during low/no traffic periods
2442 */
3d632c3f 2443 if (IS_UP(slave->dev))
1da177e4 2444 bond_arp_send_all(bond, slave);
1da177e4
LT		 2445 }
2446
2e52f4fe 2447 rcu_read_unlock();
2448
6fde8f03 2449 if (do_failover || slave_state_changed) {
2e52f4fe 2450 if (!rtnl_trylock())
2451 goto re_arm;
1da177e4 2452
6fde8f03
DT		 2453 if (slave_state_changed) {
2454 bond_slave_state_change(bond);
2455 } else if (do_failover) {
2456 /* the bond_select_active_slave must hold RTNL
2457 * and curr_slave_lock for write.
2458 */
2459 block_netpoll_tx();
2460 write_lock_bh(&bond->curr_slave_lock);
1da177e4 2461
6fde8f03
DT		 2462 bond_select_active_slave(bond);
2463
2464 write_unlock_bh(&bond->curr_slave_lock);
2465 unblock_netpoll_tx();
2466 }
2e52f4fe 2467 rtnl_unlock();
1da177e4
LT		 2468 }
2469
2470re_arm:
e6d265e8 2471 if (bond->params.arp_interval)
e7f63f1d
VF		 2472 queue_delayed_work(bond->wq, &bond->arp_work,
2473 msecs_to_jiffies(bond->params.arp_interval));
1da177e4
LT		 2474}
2475
2476/*
b2220cad
JV		 2477 * Called to inspect slaves for active-backup mode ARP monitor link state
2478 * changes. Sets new_link in slaves to specify what action should take
2479 * place for the slave. Returns 0 if no changes are found, >0 if changes
2480 * to link states must be committed.
2481 *
eb9fa4b0 2482 * Called with rcu_read_lock hold.
1da177e4 2483 */
e7f63f1d 2484static int bond_ab_arp_inspect(struct bonding *bond)
1da177e4 2485{
def4460c 2486 unsigned long trans_start, last_rx;
9caff1e7 2487 struct list_head *iter;
dec1e90e 2488 struct slave *slave;
dec1e90e 2489 int commit = 0;
da210f55 2490
eb9fa4b0 2491 bond_for_each_slave_rcu(bond, slave, iter) {
b2220cad 2492 slave->new_link = BOND_LINK_NOCHANGE;
def4460c 2493 last_rx = slave_last_rx(bond, slave);
1da177e4 2494
b2220cad 2495 if (slave->link != BOND_LINK_UP) {
e7f63f1d 2496 if (bond_time_in_interval(bond, last_rx, 1)) {
b2220cad
JV		 2497 slave->new_link = BOND_LINK_UP;
2498 commit++;
2499 }
b2220cad
JV		 2500 continue;
2501 }
1da177e4 2502
b2220cad
JV		 2503 /*
2504 * Give slaves 2*delta after being enslaved or made
2505 * active. This avoids bouncing, as the last receive
2506 * times need a full ARP monitor cycle to be updated.
2507 */
e7f63f1d 2508 if (bond_time_in_interval(bond, slave->jiffies, 2))
b2220cad
JV		 2509 continue;
2510
2511 /*
2512 * Backup slave is down if:
2513 * - No current_arp_slave AND
2514 * - more than 3*delta since last receive AND
2515 * - the bond has an IP address
2516 *
2517 * Note: a non-null current_arp_slave indicates
2518 * the curr_active_slave went down and we are
2519 * searching for a new one; under this condition
2520 * we only take the curr_active_slave down - this
2521 * gives each slave a chance to tx/rx traffic
2522 * before being taken out
2523 */
e30bc066 2524 if (!bond_is_active_slave(slave) &&
b2220cad 2525 !bond->current_arp_slave &&
e7f63f1d 2526 !bond_time_in_interval(bond, last_rx, 3)) {
b2220cad
JV		 2527 slave->new_link = BOND_LINK_DOWN;
2528 commit++;
2529 }
2530
2531 /*
2532 * Active slave is down if:
2533 * - more than 2*delta since transmitting OR
2534 * - (more than 2*delta since receive AND
2535 * the bond has an IP address)
2536 */
cb32f2a0 2537 trans_start = dev_trans_start(slave->dev);
e30bc066 2538 if (bond_is_active_slave(slave) &&
e7f63f1d
VF		 2539 (!bond_time_in_interval(bond, trans_start, 2) ||
2540 !bond_time_in_interval(bond, last_rx, 2))) {
b2220cad
JV		 2541 slave->new_link = BOND_LINK_DOWN;
2542 commit++;
2543 }
1da177e4
LT		 2544 }
2545
b2220cad
JV		 2546 return commit;
2547}
1da177e4 2548
b2220cad
JV		 2549/*
2550 * Called to commit link state changes noted by inspection step of
2551 * active-backup mode ARP monitor.
2552 *
eb9fa4b0 2553 * Called with RTNL hold.
b2220cad 2554 */
e7f63f1d 2555static void bond_ab_arp_commit(struct bonding *bond)
b2220cad 2556{
cb32f2a0 2557 unsigned long trans_start;
9caff1e7 2558 struct list_head *iter;
dec1e90e 2559 struct slave *slave;
1da177e4 2560
9caff1e7 2561 bond_for_each_slave(bond, slave, iter) {
b2220cad
JV		 2562 switch (slave->new_link) {
2563 case BOND_LINK_NOCHANGE:
2564 continue;
ff59c456 2565
b2220cad 2566 case BOND_LINK_UP:
cb32f2a0 2567 trans_start = dev_trans_start(slave->dev);
e7f63f1d
VF		 2568 if (bond->curr_active_slave != slave ||
2569 (!bond->curr_active_slave &&
2570 bond_time_in_interval(bond, trans_start, 1))) {
b2220cad 2571 slave->link = BOND_LINK_UP;
5a430974
VF		 2572 if (bond->current_arp_slave) {
2573 bond_set_slave_inactive_flags(
5e5b0665 2574 bond->current_arp_slave,
2575 BOND_SLAVE_NOTIFY_NOW);
5a430974
VF		 2576 bond->current_arp_slave = NULL;
2577 }
b2220cad 2578
a4aee5c8 2579 pr_info("%s: link status definitely up for interface %s.\n",
b9f60253 2580 bond->dev->name, slave->dev->name);
b2220cad 2581
b9f60253
JP		 2582 if (!bond->curr_active_slave ||
2583 (slave == bond->primary_slave))
2584 goto do_failover;
1da177e4 2585
b9f60253 2586 }
1da177e4 2587
b9f60253 2588 continue;
1da177e4 2589
b2220cad
JV		 2590 case BOND_LINK_DOWN:
2591 if (slave->link_failure_count < UINT_MAX)
2592 slave->link_failure_count++;
2593
2594 slave->link = BOND_LINK_DOWN;
5e5b0665 2595 bond_set_slave_inactive_flags(slave,
2596 BOND_SLAVE_NOTIFY_NOW);
b2220cad 2597
a4aee5c8 2598 pr_info("%s: link status definitely down for interface %s, disabling it\n",
b9f60253 2599 bond->dev->name, slave->dev->name);
b2220cad 2600
b9f60253 2601 if (slave == bond->curr_active_slave) {
b2220cad 2602 bond->current_arp_slave = NULL;
b9f60253 2603 goto do_failover;
1da177e4 2604 }
b9f60253
JP		 2605
2606 continue;
b2220cad
JV		 2607
2608 default:
a4aee5c8 2609 pr_err("%s: impossible: new_link %d on slave %s\n",
b2220cad
JV		 2610 bond->dev->name, slave->new_link,
2611 slave->dev->name);
b9f60253 2612 continue;
1da177e4 2613 }
1da177e4 2614
b9f60253
JP		 2615do_failover:
2616 ASSERT_RTNL();
e843fa50 2617 block_netpoll_tx();
b2220cad 2618 write_lock_bh(&bond->curr_slave_lock);
b9f60253 2619 bond_select_active_slave(bond);
b2220cad 2620 write_unlock_bh(&bond->curr_slave_lock);
e843fa50 2621 unblock_netpoll_tx();
b2220cad 2622 }
1da177e4 2623
b2220cad
JV		 2624 bond_set_carrier(bond);
2625}
1da177e4 2626
b2220cad
JV		 2627/*
2628 * Send ARP probes for active-backup mode ARP monitor.
b0929915 2629 *
2630 * Called with rcu_read_lock hold.
b2220cad 2631 */
f2ebd477 2632static bool bond_ab_arp_probe(struct bonding *bond)
b2220cad 2633{
eb9fa4b0 2634 struct slave *slave, *before = NULL, *new_slave = NULL,
b0929915 2635 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2636 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
4087df87
VF		 2637 struct list_head *iter;
2638 bool found = false;
b0929915 2639 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
f2ebd477 2640
98b90f26 2641 if (curr_arp_slave && curr_active_slave)
a4aee5c8 2642 pr_info("PROBE: c_arp %s && cas %s BAD\n",
eb9fa4b0 2643 curr_arp_slave->dev->name,
98b90f26 2644 curr_active_slave->dev->name);
1da177e4 2645
98b90f26
VF		 2646 if (curr_active_slave) {
2647 bond_arp_send_all(bond, curr_active_slave);
b0929915 2648 return should_notify_rtnl;
b2220cad 2649 }
1da177e4 2650
b2220cad
JV		 2651 /* if we don't have a curr_active_slave, search for the next available
2652 * backup slave from the current_arp_slave and make it the candidate
2653 * for becoming the curr_active_slave
2654 */
1da177e4 2655
eb9fa4b0 2656 if (!curr_arp_slave) {
b0929915 2657 curr_arp_slave = bond_first_slave_rcu(bond);
2658 if (!curr_arp_slave)
2659 return should_notify_rtnl;
b2220cad 2660 }
1da177e4 2661
b0929915 2662 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
059fe7a5 2663
b0929915 2664 bond_for_each_slave_rcu(bond, slave, iter) {
4087df87
VF		 2665 if (!found && !before && IS_UP(slave->dev))
2666 before = slave;
1da177e4 2667
4087df87
VF		 2668 if (found && !new_slave && IS_UP(slave->dev))
2669 new_slave = slave;
b2220cad
JV		 2670 /* if the link state is up at this point, we
2671 * mark it down - this can happen if we have
2672 * simultaneous link failures and
2673 * reselect_active_interface doesn't make this
2674 * one the current slave so it is still marked
2675 * up when it is actually down
1da177e4 2676 */
4087df87 2677 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
b2220cad
JV		 2678 slave->link = BOND_LINK_DOWN;
2679 if (slave->link_failure_count < UINT_MAX)
2680 slave->link_failure_count++;
1da177e4 2681
5e5b0665 2682 bond_set_slave_inactive_flags(slave,
b0929915 2683 BOND_SLAVE_NOTIFY_LATER);
b2220cad 2684
a4aee5c8
JP		 2685 pr_info("%s: backup interface %s is now down.\n",
2686 bond->dev->name, slave->dev->name);
1da177e4 2687 }
eb9fa4b0 2688 if (slave == curr_arp_slave)
4087df87 2689 found = true;
b2220cad 2690 }
4087df87
VF		 2691
2692 if (!new_slave && before)
2693 new_slave = before;
2694
b0929915 2695 if (!new_slave)
2696 goto check_state;
4087df87
VF		 2697
2698 new_slave->link = BOND_LINK_BACK;
b0929915 2699 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
4087df87
VF		 2700 bond_arp_send_all(bond, new_slave);
2701 new_slave->jiffies = jiffies;
eb9fa4b0 2702 rcu_assign_pointer(bond->current_arp_slave, new_slave);
f2ebd477 2703
b0929915 2704check_state:
2705 bond_for_each_slave_rcu(bond, slave, iter) {
2706 if (slave->should_notify) {
2707 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2708 break;
2709 }
2710 }
2711 return should_notify_rtnl;
b2220cad 2712}
1da177e4 2713
6da67d26 2714static void bond_activebackup_arp_mon(struct work_struct *work)
b2220cad
JV		 2715{
2716 struct bonding *bond = container_of(work, struct bonding,
2717 arp_work.work);
b0929915 2718 bool should_notify_peers = false;
2719 bool should_notify_rtnl = false;
1f2cd845 2720 int delta_in_ticks;
1da177e4 2721
1f2cd845
DM		 2722 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2723
2724 if (!bond_has_slaves(bond))
b2220cad
JV		 2725 goto re_arm;
2726
eb9fa4b0 2727 rcu_read_lock();
b0929915 2728
ad246c99
BH		 2729 should_notify_peers = bond_should_notify_peers(bond);
2730
b0929915 2731 if (bond_ab_arp_inspect(bond)) {
2732 rcu_read_unlock();
2733
1f2cd845
DM		 2734 /* Race avoidance with bond_close flush of workqueue */
2735 if (!rtnl_trylock()) {
1f2cd845
DM		 2736 delta_in_ticks = 1;
2737 should_notify_peers = false;
2738 goto re_arm;
2739 }
b2220cad 2740
1f2cd845 2741 bond_ab_arp_commit(bond);
b0929915 2742
1f2cd845 2743 rtnl_unlock();
b0929915 2744 rcu_read_lock();
1f2cd845
DM		 2745 }
2746
b0929915 2747 should_notify_rtnl = bond_ab_arp_probe(bond);
2748 rcu_read_unlock();
ad246c99 2749
80b9d236 2750re_arm:
2751 if (bond->params.arp_interval)
1f2cd845
DM		 2752 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2753
b0929915 2754 if (should_notify_peers || should_notify_rtnl) {
1f2cd845
DM		 2755 if (!rtnl_trylock())
2756 return;
b0929915 2757
2758 if (should_notify_peers)
2759 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2760 bond->dev);
2761 if (should_notify_rtnl)
2762 bond_slave_state_notify(bond);
2763
1f2cd845
DM		 2764 rtnl_unlock();
2765 }
1da177e4
LT		 2766}
2767
1da177e4
LT		 2768/*-------------------------- netdev event handling --------------------------*/
2769
2770/*
2771 * Change device name
2772 */
2773static int bond_event_changename(struct bonding *bond)
2774{
1da177e4
LT		 2775 bond_remove_proc_entry(bond);
2776 bond_create_proc_entry(bond);
7e083840 2777
f073c7ca
TI		 2778 bond_debug_reregister(bond);
2779
1da177e4
LT		 2780 return NOTIFY_DONE;
2781}
2782
3d632c3f
SH		 2783static int bond_master_netdev_event(unsigned long event,
2784 struct net_device *bond_dev)
1da177e4 2785{
454d7c9b 2786 struct bonding *event_bond = netdev_priv(bond_dev);
1da177e4
LT		 2787
2788 switch (event) {
2789 case NETDEV_CHANGENAME:
2790 return bond_event_changename(event_bond);
a64d49c3
EB		 2791 case NETDEV_UNREGISTER:
2792 bond_remove_proc_entry(event_bond);
2793 break;
2794 case NETDEV_REGISTER:
2795 bond_create_proc_entry(event_bond);
2796 break;
6c8d23f7 2797 case NETDEV_NOTIFY_PEERS:
2798 if (event_bond->send_peer_notif)
2799 event_bond->send_peer_notif--;
2800 break;
1da177e4
LT		 2801 default:
2802 break;
2803 }
2804
2805 return NOTIFY_DONE;
2806}
2807
3d632c3f
SH		 2808static int bond_slave_netdev_event(unsigned long event,
2809 struct net_device *slave_dev)
1da177e4 2810{
471cb5a3 2811 struct slave *slave = bond_slave_get_rtnl(slave_dev);
6101391d 2812 struct bonding *bond;
2813 struct net_device *bond_dev;
471cb5a3
JP		 2814 u32 old_speed;
2815 u8 old_duplex;
1da177e4 2816
6101391d 2817 /* A netdev event can be generated while enslaving a device
2818 * before netdev_rx_handler_register is called in which case
2819 * slave will be NULL
2820 */
2821 if (!slave)
2822 return NOTIFY_DONE;
2823 bond_dev = slave->bond->dev;
2824 bond = slave->bond;
2825
1da177e4
LT		 2826 switch (event) {
2827 case NETDEV_UNREGISTER:
8d2ada77 2828 if (bond_dev->type != ARPHRD_ETHER)
471cb5a3
JP		 2829 bond_release_and_destroy(bond_dev, slave_dev);
2830 else
2831 bond_release(bond_dev, slave_dev);
1da177e4 2832 break;
98f41f69 2833 case NETDEV_UP:
1da177e4 2834 case NETDEV_CHANGE:
471cb5a3
JP		 2835 old_speed = slave->speed;
2836 old_duplex = slave->duplex;
17d04500 2837
471cb5a3 2838 bond_update_speed_duplex(slave);
17d04500 2839
471cb5a3
JP		 2840 if (bond->params.mode == BOND_MODE_8023AD) {
2841 if (old_speed != slave->speed)
2842 bond_3ad_adapter_speed_changed(slave);
2843 if (old_duplex != slave->duplex)
2844 bond_3ad_adapter_duplex_changed(slave);
17d04500 2845 }
1da177e4
LT		 2846 break;
2847 case NETDEV_DOWN:
2848 /*
2849 * ... Or is it this?
2850 */
2851 break;
2852 case NETDEV_CHANGEMTU:
2853 /*
2854 * TODO: Should slaves be allowed to
2855 * independently alter their MTU? For
2856 * an active-backup bond, slaves need
2857 * not be the same type of device, so
2858 * MTUs may vary. For other modes,
2859 * slaves arguably should have the
2860 * same MTUs. To do this, we'd need to
2861 * take over the slave's change_mtu
2862 * function for the duration of their
2863 * servitude.
2864 */
2865 break;
2866 case NETDEV_CHANGENAME:
3ec775b9
VF		 2867 /* we don't care if we don't have primary set */
2868 if (!USES_PRIMARY(bond->params.mode) ||
2869 !bond->params.primary[0])
2870 break;
2871
2872 if (slave == bond->primary_slave) {
2873 /* slave's name changed - he's no longer primary */
2874 bond->primary_slave = NULL;
2875 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2876 /* we have a new primary slave */
2877 bond->primary_slave = slave;
2878 } else { /* we didn't change primary - exit */
2879 break;
2880 }
2881
2882 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2883 bond->dev->name, bond->primary_slave ? slave_dev->name :
2884 "none");
f80889a5 2885
2886 block_netpoll_tx();
3ec775b9
VF		 2887 write_lock_bh(&bond->curr_slave_lock);
2888 bond_select_active_slave(bond);
2889 write_unlock_bh(&bond->curr_slave_lock);
f80889a5 2890 unblock_netpoll_tx();
1da177e4 2891 break;
8531c5ff
AK		 2892 case NETDEV_FEAT_CHANGE:
2893 bond_compute_features(bond);
2894 break;
4aa5dee4
JP		 2895 case NETDEV_RESEND_IGMP:
2896 /* Propagate to master device */
2897 call_netdevice_notifiers(event, slave->bond->dev);
2898 break;
1da177e4
LT		 2899 default:
2900 break;
2901 }
2902
2903 return NOTIFY_DONE;
2904}
2905
2906/*
2907 * bond_netdev_event: handle netdev notifier chain events.
2908 *
2909 * This function receives events for the netdev chain. The caller (an
e041c683 2910 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
1da177e4
LT		 2911 * locks for us to safely manipulate the slave devices (RTNL lock,
2912 * dev_probe_lock).
2913 */
3d632c3f
SH		 2914static int bond_netdev_event(struct notifier_block *this,
2915 unsigned long event, void *ptr)
1da177e4 2916{
351638e7 2917 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1da177e4 2918
5a03cdb7 2919 pr_debug("event_dev: %s, event: %lx\n",
a4aee5c8
JP		 2920 event_dev ? event_dev->name : "None",
2921 event);
1da177e4 2922
0b680e75
JV		 2923 if (!(event_dev->priv_flags & IFF_BONDING))
2924 return NOTIFY_DONE;
2925
1da177e4 2926 if (event_dev->flags & IFF_MASTER) {
5a03cdb7 2927 pr_debug("IFF_MASTER\n");
1da177e4
LT		 2928 return bond_master_netdev_event(event, event_dev);
2929 }
2930
2931 if (event_dev->flags & IFF_SLAVE) {
5a03cdb7 2932 pr_debug("IFF_SLAVE\n");
1da177e4
LT		 2933 return bond_slave_netdev_event(event, event_dev);
2934 }
2935
2936 return NOTIFY_DONE;
2937}
2938
2939static struct notifier_block bond_netdev_notifier = {
2940 .notifier_call = bond_netdev_event,
2941};
2942
169a3e66
JV		 2943/*---------------------------- Hashing Policies -----------------------------*/
2944
32819dc1
NA		 2945/* L2 hash helper */
2946static inline u32 bond_eth_hash(struct sk_buff *skb)
6b923cb7
JE		 2947{
2948 struct ethhdr *data = (struct ethhdr *)skb->data;
2949
2950 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
32819dc1 2951 return data->h_dest[5] ^ data->h_source[5];
6b923cb7
JE		 2952
2953 return 0;
2954}
2955
32819dc1
NA		 2956/* Extract the appropriate headers based on bond's xmit policy */
2957static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2958 struct flow_keys *fk)
6f6652be 2959{
32819dc1 2960 const struct ipv6hdr *iph6;
4394542c 2961 const struct iphdr *iph;
32819dc1 2962 int noff, proto = -1;
6b923cb7 2963
32819dc1
NA		 2964 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2965 return skb_flow_dissect(skb, fk);
2966
2967 fk->ports = 0;
2968 noff = skb_network_offset(skb);
2969 if (skb->protocol == htons(ETH_P_IP)) {
2970 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2971 return false;
6b923cb7 2972 iph = ip_hdr(skb);
32819dc1
NA		 2973 fk->src = iph->saddr;
2974 fk->dst = iph->daddr;
2975 noff += iph->ihl << 2;
2976 if (!ip_is_fragment(iph))
2977 proto = iph->protocol;
2978 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2979 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2980 return false;
2981 iph6 = ipv6_hdr(skb);
2982 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2983 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2984 noff += sizeof(*iph6);
2985 proto = iph6->nexthdr;
2986 } else {
2987 return false;
6f6652be 2988 }
32819dc1
NA		 2989 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2990 fk->ports = skb_flow_get_ports(skb, noff, proto);
6f6652be 2991
32819dc1 2992 return true;
6f6652be
JV		 2993}
2994
32819dc1
NA		 2995/**
2996 * bond_xmit_hash - generate a hash value based on the xmit policy
2997 * @bond: bonding device
2998 * @skb: buffer to use for headers
2999 * @count: modulo value
3000 *
3001 * This function will extract the necessary headers from the skb buffer and use
3002 * them to generate a hash based on the xmit_policy set in the bonding device
3003 * which will be reduced modulo count before returning.
169a3e66 3004 */
32819dc1 3005int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
169a3e66 3006{
32819dc1
NA		 3007 struct flow_keys flow;
3008 u32 hash;
4394542c 3009
32819dc1
NA		 3010 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3011 !bond_flow_dissect(bond, skb, &flow))
3012 return bond_eth_hash(skb) % count;
169a3e66 3013
32819dc1
NA		 3014 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3015 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3016 hash = bond_eth_hash(skb);
3017 else
3018 hash = (__force u32)flow.ports;
3019 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3020 hash ^= (hash >> 16);
3021 hash ^= (hash >> 8);
3022
3023 return hash % count;
169a3e66
JV		 3024}
3025
1da177e4
LT		 3026/*-------------------------- Device entry points ----------------------------*/
3027
fbb0c41b 3028static void bond_work_init_all(struct bonding *bond)
3029{
3030 INIT_DELAYED_WORK(&bond->mcast_work,
3031 bond_resend_igmp_join_requests_delayed);
3032 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3033 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3034 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3035 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3036 else
3037 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3038 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3039}
3040
3041static void bond_work_cancel_all(struct bonding *bond)
3042{
3043 cancel_delayed_work_sync(&bond->mii_work);
3044 cancel_delayed_work_sync(&bond->arp_work);
3045 cancel_delayed_work_sync(&bond->alb_work);
3046 cancel_delayed_work_sync(&bond->ad_work);
3047 cancel_delayed_work_sync(&bond->mcast_work);
3048}
3049
1da177e4
LT		 3050static int bond_open(struct net_device *bond_dev)
3051{
454d7c9b 3052 struct bonding *bond = netdev_priv(bond_dev);
9caff1e7 3053 struct list_head *iter;
ba3211cc 3054 struct slave *slave;
1da177e4 3055
ba3211cc
PP		 3056 /* reset slave->backup and slave->inactive */
3057 read_lock(&bond->lock);
0965a1f3 3058 if (bond_has_slaves(bond)) {
ba3211cc 3059 read_lock(&bond->curr_slave_lock);
9caff1e7 3060 bond_for_each_slave(bond, slave, iter) {
ba3211cc
PP		 3061 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3062 && (slave != bond->curr_active_slave)) {
5e5b0665 3063 bond_set_slave_inactive_flags(slave,
3064 BOND_SLAVE_NOTIFY_NOW);
ba3211cc 3065 } else {
5e5b0665 3066 bond_set_slave_active_flags(slave,
3067 BOND_SLAVE_NOTIFY_NOW);
ba3211cc
PP		 3068 }
3069 }
3070 read_unlock(&bond->curr_slave_lock);
3071 }
3072 read_unlock(&bond->lock);
3073
fbb0c41b 3074 bond_work_init_all(bond);
5a37e8ca 3075
58402054 3076 if (bond_is_lb(bond)) {
1da177e4
LT		 3077 /* bond_alb_initialize must be called before the timer
3078 * is started.
3079 */
fbb0c41b 3080 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
b473946a 3081 return -ENOMEM;
1b76b316 3082 queue_delayed_work(bond->wq, &bond->alb_work, 0);
1da177e4
LT		 3083 }
3084
fbb0c41b 3085 if (bond->params.miimon) /* link check interval, in milliseconds. */
1b76b316 3086 queue_delayed_work(bond->wq, &bond->mii_work, 0);
1da177e4
LT		 3087
3088 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
1b76b316 3089 queue_delayed_work(bond->wq, &bond->arp_work, 0);
f5b2b966 3090 if (bond->params.arp_validate)
3aba891d 3091 bond->recv_probe = bond_arp_rcv;
1da177e4
LT		 3092 }
3093
3094 if (bond->params.mode == BOND_MODE_8023AD) {
1b76b316 3095 queue_delayed_work(bond->wq, &bond->ad_work, 0);
1da177e4 3096 /* register to receive LACPDUs */
3aba891d 3097 bond->recv_probe = bond_3ad_lacpdu_recv;
fd989c83 3098 bond_3ad_initiate_agg_selection(bond, 1);
1da177e4
LT		 3099 }
3100
3101 return 0;
3102}
3103
3104static int bond_close(struct net_device *bond_dev)
3105{
454d7c9b 3106 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3107
fbb0c41b 3108 bond_work_cancel_all(bond);
6c8d23f7 3109 bond->send_peer_notif = 0;
ee8487c0 3110 if (bond_is_lb(bond))
1da177e4 3111 bond_alb_deinitialize(bond);
3aba891d 3112 bond->recv_probe = NULL;
1da177e4
LT		 3113
3114 return 0;
3115}
3116
28172739
ED		 3117static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3118 struct rtnl_link_stats64 *stats)
1da177e4 3119{
454d7c9b 3120 struct bonding *bond = netdev_priv(bond_dev);
28172739 3121 struct rtnl_link_stats64 temp;
9caff1e7 3122 struct list_head *iter;
1da177e4 3123 struct slave *slave;
1da177e4 3124
28172739 3125 memset(stats, 0, sizeof(*stats));
1da177e4
LT		 3126
3127 read_lock_bh(&bond->lock);
9caff1e7 3128 bond_for_each_slave(bond, slave, iter) {
be1f3c2c 3129 const struct rtnl_link_stats64 *sstats =
28172739
ED		 3130 dev_get_stats(slave->dev, &temp);
3131
3132 stats->rx_packets += sstats->rx_packets;
3133 stats->rx_bytes += sstats->rx_bytes;
3134 stats->rx_errors += sstats->rx_errors;
3135 stats->rx_dropped += sstats->rx_dropped;
3136
3137 stats->tx_packets += sstats->tx_packets;
3138 stats->tx_bytes += sstats->tx_bytes;
3139 stats->tx_errors += sstats->tx_errors;
3140 stats->tx_dropped += sstats->tx_dropped;
3141
3142 stats->multicast += sstats->multicast;
3143 stats->collisions += sstats->collisions;
3144
3145 stats->rx_length_errors += sstats->rx_length_errors;
3146 stats->rx_over_errors += sstats->rx_over_errors;
3147 stats->rx_crc_errors += sstats->rx_crc_errors;
3148 stats->rx_frame_errors += sstats->rx_frame_errors;
3149 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3150 stats->rx_missed_errors += sstats->rx_missed_errors;
3151
3152 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3153 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3154 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3155 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3156 stats->tx_window_errors += sstats->tx_window_errors;
2439f9eb 3157 }
1da177e4
LT		 3158 read_unlock_bh(&bond->lock);
3159
3160 return stats;
3161}
3162
3163static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3164{
080a06e1 3165 struct bonding *bond = netdev_priv(bond_dev);
1da177e4
LT		 3166 struct net_device *slave_dev = NULL;
3167 struct ifbond k_binfo;
3168 struct ifbond __user *u_binfo = NULL;
3169 struct ifslave k_sinfo;
3170 struct ifslave __user *u_sinfo = NULL;
3171 struct mii_ioctl_data *mii = NULL;
d1fbd3ed 3172 struct bond_opt_value newval;
387ff911 3173 struct net *net;
1da177e4
LT		 3174 int res = 0;
3175
a4aee5c8 3176 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
1da177e4
LT		 3177
3178 switch (cmd) {
1da177e4
LT		 3179 case SIOCGMIIPHY:
3180 mii = if_mii(ifr);
3d632c3f 3181 if (!mii)
1da177e4 3182 return -EINVAL;
3d632c3f 3183
1da177e4
LT		 3184 mii->phy_id = 0;
3185 /* Fall Through */
3186 case SIOCGMIIREG:
3187 /*
3188 * We do this again just in case we were called by SIOCGMIIREG
3189 * instead of SIOCGMIIPHY.
3190 */
3191 mii = if_mii(ifr);
3d632c3f 3192 if (!mii)
1da177e4 3193 return -EINVAL;
3d632c3f 3194
1da177e4
LT		 3195
3196 if (mii->reg_num == 1) {
1da177e4 3197 mii->val_out = 0;
6603a6f2 3198 read_lock(&bond->lock);
1da177e4 3199 read_lock(&bond->curr_slave_lock);
3d632c3f 3200 if (netif_carrier_ok(bond->dev))
1da177e4 3201 mii->val_out = BMSR_LSTATUS;
3d632c3f 3202
1da177e4 3203 read_unlock(&bond->curr_slave_lock);
6603a6f2 3204 read_unlock(&bond->lock);
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:
3235 /* Go on */
3236 break;
3237 }
3238
387ff911
G		 3239 net = dev_net(bond_dev);
3240
3241 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1da177e4 3242 return -EPERM;
1da177e4 3243
0917b933 3244 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
1da177e4 3245
a4aee5c8 3246 pr_debug("slave_dev=%p:\n", slave_dev);
1da177e4 3247
3d632c3f 3248 if (!slave_dev)
0917b933 3249 return -ENODEV;
1da177e4 3250
0917b933
YX		 3251 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3252 switch (cmd) {
3253 case BOND_ENSLAVE_OLD:
3254 case SIOCBONDENSLAVE:
3255 res = bond_enslave(bond_dev, slave_dev);
3256 break;
3257 case BOND_RELEASE_OLD:
3258 case SIOCBONDRELEASE:
3259 res = bond_release(bond_dev, slave_dev);
3260 break;
3261 case BOND_SETHWADDR_OLD:
3262 case SIOCBONDSETHWADDR:
3263 bond_set_dev_addr(bond_dev, slave_dev);
3264 res = 0;
3265 break;
3266 case BOND_CHANGE_ACTIVE_OLD:
3267 case SIOCBONDCHANGEACTIVE:
d1fbd3ed
NA		 3268 bond_opt_initstr(&newval, slave_dev->name);
3269 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
0917b933
YX		 3270 break;
3271 default:
3272 res = -EOPNOTSUPP;
1da177e4
LT		 3273 }
3274
1da177e4
LT		 3275 return res;
3276}
3277
d03462b9 3278static void bond_change_rx_flags(struct net_device *bond_dev, int change)
1da177e4 3279{
454d7c9b 3280 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3281
d03462b9
JP		 3282 if (change & IFF_PROMISC)
3283 bond_set_promiscuity(bond,
3284 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3d632c3f 3285
d03462b9
JP		 3286 if (change & IFF_ALLMULTI)
3287 bond_set_allmulti(bond,
3288 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3289}
1da177e4 3290
303d1cbf 3291static void bond_set_rx_mode(struct net_device *bond_dev)
d03462b9
JP		 3292{
3293 struct bonding *bond = netdev_priv(bond_dev);
9caff1e7 3294 struct list_head *iter;
303d1cbf 3295 struct slave *slave;
1da177e4 3296
80ee5ad2 3297
b3241870 3298 rcu_read_lock();
303d1cbf 3299 if (USES_PRIMARY(bond->params.mode)) {
b3241870 3300 slave = rcu_dereference(bond->curr_active_slave);
303d1cbf
JV		 3301 if (slave) {
3302 dev_uc_sync(slave->dev, bond_dev);
3303 dev_mc_sync(slave->dev, bond_dev);
3304 }
303d1cbf 3305 } else {
b3241870 3306 bond_for_each_slave_rcu(bond, slave, iter) {
303d1cbf
JV		 3307 dev_uc_sync_multiple(slave->dev, bond_dev);
3308 dev_mc_sync_multiple(slave->dev, bond_dev);
3309 }
1da177e4 3310 }
b3241870 3311 rcu_read_unlock();
1da177e4
LT		 3312}
3313
234bcf8a 3314static int bond_neigh_init(struct neighbour *n)
00829823 3315{
234bcf8a 3316 struct bonding *bond = netdev_priv(n->dev);
234bcf8a
SP		 3317 const struct net_device_ops *slave_ops;
3318 struct neigh_parms parms;
dec1e90e 3319 struct slave *slave;
234bcf8a
SP		 3320 int ret;
3321
dec1e90e 3322 slave = bond_first_slave(bond);
234bcf8a
SP		 3323 if (!slave)
3324 return 0;
234bcf8a 3325 slave_ops = slave->dev->netdev_ops;
234bcf8a
SP		 3326 if (!slave_ops->ndo_neigh_setup)
3327 return 0;
3328
3329 parms.neigh_setup = NULL;
3330 parms.neigh_cleanup = NULL;
3331 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3332 if (ret)
3333 return ret;
3334
3335 /*
3336 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3337 * after the last slave has been detached. Assumes that all slaves
3338 * utilize the same neigh_cleanup (true at this writing as only user
3339 * is ipoib).
3340 */
3341 n->parms->neigh_cleanup = parms.neigh_cleanup;
3342
3343 if (!parms.neigh_setup)
3344 return 0;
3345
3346 return parms.neigh_setup(n);
3347}
3348
3349/*
3350 * The bonding ndo_neigh_setup is called at init time beofre any
3351 * slave exists. So we must declare proxy setup function which will
3352 * be used at run time to resolve the actual slave neigh param setup.
9918d5bf
VF		 3353 *
3354 * It's also called by master devices (such as vlans) to setup their
3355 * underlying devices. In that case - do nothing, we're already set up from
3356 * our init.
234bcf8a
SP		 3357 */
3358static int bond_neigh_setup(struct net_device *dev,
3359 struct neigh_parms *parms)
3360{
9918d5bf
VF		 3361 /* modify only our neigh_parms */
3362 if (parms->dev == dev)
3363 parms->neigh_setup = bond_neigh_init;
00829823 3364
00829823
SH		 3365 return 0;
3366}
3367
1da177e4
LT		 3368/*
3369 * Change the MTU of all of a master's slaves to match the master
3370 */
3371static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3372{
454d7c9b 3373 struct bonding *bond = netdev_priv(bond_dev);
81f23b13 3374 struct slave *slave, *rollback_slave;
9caff1e7 3375 struct list_head *iter;
1da177e4 3376 int res = 0;
1da177e4 3377
5a03cdb7 3378 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
a4aee5c8 3379 (bond_dev ? bond_dev->name : "None"), new_mtu);
1da177e4
LT		 3380
3381 /* Can't hold bond->lock with bh disabled here since
3382 * some base drivers panic. On the other hand we can't
3383 * hold bond->lock without bh disabled because we'll
3384 * deadlock. The only solution is to rely on the fact
3385 * that we're under rtnl_lock here, and the slaves
3386 * list won't change. This doesn't solve the problem
3387 * of setting the slave's MTU while it is
3388 * transmitting, but the assumption is that the base
3389 * driver can handle that.
3390 *
3391 * TODO: figure out a way to safely iterate the slaves
3392 * list, but without holding a lock around the actual
3393 * call to the base driver.
3394 */
3395
9caff1e7 3396 bond_for_each_slave(bond, slave, iter) {
c8c23903 3397 pr_debug("s %p c_m %p\n",
a4aee5c8 3398 slave,
a4aee5c8 3399 slave->dev->netdev_ops->ndo_change_mtu);
e944ef79 3400
1da177e4
LT		 3401 res = dev_set_mtu(slave->dev, new_mtu);
3402
3403 if (res) {
3404 /* If we failed to set the slave's mtu to the new value
3405 * we must abort the operation even in ACTIVE_BACKUP
3406 * mode, because if we allow the backup slaves to have
3407 * different mtu values than the active slave we'll
3408 * need to change their mtu when doing a failover. That
3409 * means changing their mtu from timer context, which
3410 * is probably not a good idea.
3411 */
5a03cdb7 3412 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 3413 goto unwind;
3414 }
3415 }
3416
3417 bond_dev->mtu = new_mtu;
3418
3419 return 0;
3420
3421unwind:
3422 /* unwind from head to the slave that failed */
9caff1e7 3423 bond_for_each_slave(bond, rollback_slave, iter) {
1da177e4
LT		 3424 int tmp_res;
3425
81f23b13
VF		 3426 if (rollback_slave == slave)
3427 break;
3428
3429 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
1da177e4 3430 if (tmp_res) {
a4aee5c8 3431 pr_debug("unwind err %d dev %s\n",
81f23b13 3432 tmp_res, rollback_slave->dev->name);
1da177e4
LT		 3433 }
3434 }
3435
3436 return res;
3437}
3438
3439/*
3440 * Change HW address
3441 *
3442 * Note that many devices must be down to change the HW address, and
3443 * downing the master releases all slaves. We can make bonds full of
3444 * bonding devices to test this, however.
3445 */
3446static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3447{
454d7c9b 3448 struct bonding *bond = netdev_priv(bond_dev);
81f23b13 3449 struct slave *slave, *rollback_slave;
1da177e4 3450 struct sockaddr *sa = addr, tmp_sa;
9caff1e7 3451 struct list_head *iter;
1da177e4 3452 int res = 0;
1da177e4 3453
eb7cc59a
SH		 3454 if (bond->params.mode == BOND_MODE_ALB)
3455 return bond_alb_set_mac_address(bond_dev, addr);
3456
3457
a4aee5c8
JP		 3458 pr_debug("bond=%p, name=%s\n",
3459 bond, bond_dev ? bond_dev->name : "None");
1da177e4 3460
1b5acd29
JV		 3461 /* If fail_over_mac is enabled, do nothing and return success.
3462 * Returning an error causes ifenslave to fail.
dd957c57 3463 */
cc689aaa 3464 if (bond->params.fail_over_mac &&
3465 bond->params.mode == BOND_MODE_ACTIVEBACKUP)
dd957c57 3466 return 0;
2ab82852 3467
3d632c3f 3468 if (!is_valid_ether_addr(sa->sa_data))
1da177e4 3469 return -EADDRNOTAVAIL;
1da177e4
LT		 3470
3471 /* Can't hold bond->lock with bh disabled here since
3472 * some base drivers panic. On the other hand we can't
3473 * hold bond->lock without bh disabled because we'll
3474 * deadlock. The only solution is to rely on the fact
3475 * that we're under rtnl_lock here, and the slaves
3476 * list won't change. This doesn't solve the problem
3477 * of setting the slave's hw address while it is
3478 * transmitting, but the assumption is that the base
3479 * driver can handle that.
3480 *
3481 * TODO: figure out a way to safely iterate the slaves
3482 * list, but without holding a lock around the actual
3483 * call to the base driver.
3484 */
3485
9caff1e7 3486 bond_for_each_slave(bond, slave, iter) {
eb7cc59a 3487 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
5a03cdb7 3488 pr_debug("slave %p %s\n", slave, slave->dev->name);
1da177e4 3489
eb7cc59a 3490 if (slave_ops->ndo_set_mac_address == NULL) {
1da177e4 3491 res = -EOPNOTSUPP;
5a03cdb7 3492 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
1da177e4
LT		 3493 goto unwind;
3494 }
3495
3496 res = dev_set_mac_address(slave->dev, addr);
3497 if (res) {
3498 /* TODO: consider downing the slave
3499 * and retry ?
3500 * User should expect communications
3501 * breakage anyway until ARP finish
3502 * updating, so...
3503 */
5a03cdb7 3504 pr_debug("err %d %s\n", res, slave->dev->name);
1da177e4
LT		 3505 goto unwind;
3506 }
3507 }
3508
3509 /* success */
3510 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3511 return 0;
3512
3513unwind:
3514 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3515 tmp_sa.sa_family = bond_dev->type;
3516
3517 /* unwind from head to the slave that failed */
9caff1e7 3518 bond_for_each_slave(bond, rollback_slave, iter) {
1da177e4
LT		 3519 int tmp_res;
3520
81f23b13
VF		 3521 if (rollback_slave == slave)
3522 break;
3523
3524 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
1da177e4 3525 if (tmp_res) {
a4aee5c8 3526 pr_debug("unwind err %d dev %s\n",
81f23b13 3527 tmp_res, rollback_slave->dev->name);
1da177e4
LT		 3528 }
3529 }
3530
3531 return res;
3532}
3533
15077228
NA		 3534/**
3535 * bond_xmit_slave_id - transmit skb through slave with slave_id
3536 * @bond: bonding device that is transmitting
3537 * @skb: buffer to transmit
3538 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3539 *
3540 * This function tries to transmit through slave with slave_id but in case
3541 * it fails, it tries to find the first available slave for transmission.
3542 * The skb is consumed in all cases, thus the function is void.
3543 */
da131ddb 3544static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
15077228 3545{
9caff1e7 3546 struct list_head *iter;
15077228
NA		 3547 struct slave *slave;
3548 int i = slave_id;
3549
3550 /* Here we start from the slave with slave_id */
9caff1e7 3551 bond_for_each_slave_rcu(bond, slave, iter) {
15077228
NA		 3552 if (--i < 0) {
3553 if (slave_can_tx(slave)) {
3554 bond_dev_queue_xmit(bond, skb, slave->dev);
3555 return;
3556 }
3557 }
3558 }
3559
3560 /* Here we start from the first slave up to slave_id */
3561 i = slave_id;
9caff1e7 3562 bond_for_each_slave_rcu(bond, slave, iter) {
15077228
NA		 3563 if (--i < 0)
3564 break;
3565 if (slave_can_tx(slave)) {
3566 bond_dev_queue_xmit(bond, skb, slave->dev);
3567 return;
3568 }
3569 }
3570 /* no slave that can tx has been found */
3571 kfree_skb(skb);
3572}
3573
73958329
NA		 3574/**
3575 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3576 * @bond: bonding device to use
3577 *
3578 * Based on the value of the bonding device's packets_per_slave parameter
3579 * this function generates a slave id, which is usually used as the next
3580 * slave to transmit through.
3581 */
3582static u32 bond_rr_gen_slave_id(struct bonding *bond)
3583{
73958329 3584 u32 slave_id;
809fa972
HFS		 3585 struct reciprocal_value reciprocal_packets_per_slave;
3586 int packets_per_slave = bond->params.packets_per_slave;
73958329
NA		 3587
3588 switch (packets_per_slave) {
3589 case 0:
3590 slave_id = prandom_u32();
3591 break;
3592 case 1:
3593 slave_id = bond->rr_tx_counter;
3594 break;
3595 default:
809fa972
HFS		 3596 reciprocal_packets_per_slave =
3597 bond->params.reciprocal_packets_per_slave;
73958329 3598 slave_id = reciprocal_divide(bond->rr_tx_counter,
809fa972 3599 reciprocal_packets_per_slave);
73958329
NA		 3600 break;
3601 }
3602 bond->rr_tx_counter++;
3603
3604 return slave_id;
3605}
3606
1da177e4
LT		 3607static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3608{
454d7c9b 3609 struct bonding *bond = netdev_priv(bond_dev);
a2fd940f 3610 struct iphdr *iph = ip_hdr(skb);
15077228 3611 struct slave *slave;
73958329 3612 u32 slave_id;
1da177e4 3613
73958329 3614 /* Start with the curr_active_slave that joined the bond as the
a2fd940f
AG		 3615 * default for sending IGMP traffic. For failover purposes one
3616 * needs to maintain some consistency for the interface that will
3617 * send the join/membership reports. The curr_active_slave found
3618 * will send all of this type of traffic.
cf5f9044 3619 */
15077228 3620 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
278b2083 3621 slave = rcu_dereference(bond->curr_active_slave);
15077228
NA		 3622 if (slave && slave_can_tx(slave))
3623 bond_dev_queue_xmit(bond, skb, slave->dev);
3624 else
3625 bond_xmit_slave_id(bond, skb, 0);
a2fd940f 3626 } else {
73958329
NA		 3627 slave_id = bond_rr_gen_slave_id(bond);
3628 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
1da177e4 3629 }
0693e88e 3630
ec634fe3 3631 return NETDEV_TX_OK;
1da177e4
LT		 3632}
3633
3634/*
3635 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3636 * the bond has a usable interface.
3637 */
3638static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3639{
454d7c9b 3640 struct bonding *bond = netdev_priv(bond_dev);
71bc3b2d 3641 struct slave *slave;
1da177e4 3642
278b2083 3643 slave = rcu_dereference(bond->curr_active_slave);
71bc3b2d 3644 if (slave)
15077228
NA		 3645 bond_dev_queue_xmit(bond, skb, slave->dev);
3646 else
04502430 3647 kfree_skb(skb);
0693e88e 3648
ec634fe3 3649 return NETDEV_TX_OK;
1da177e4
LT		 3650}
3651
32819dc1 3652/* In bond_xmit_xor() , we determine the output device by using a pre-
169a3e66
JV		 3653 * determined xmit_hash_policy(), If the selected device is not enabled,
3654 * find the next active slave.
1da177e4
LT		 3655 */
3656static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3657{
454d7c9b 3658 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3659
32819dc1 3660 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
0693e88e 3661
ec634fe3 3662 return NETDEV_TX_OK;
1da177e4
LT		 3663}
3664
78a646ce 3665/* in broadcast mode, we send everything to all usable interfaces. */
1da177e4
LT		 3666static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3667{
454d7c9b 3668 struct bonding *bond = netdev_priv(bond_dev);
78a646ce 3669 struct slave *slave = NULL;
9caff1e7 3670 struct list_head *iter;
1da177e4 3671
9caff1e7 3672 bond_for_each_slave_rcu(bond, slave, iter) {
78a646ce
NA		 3673 if (bond_is_last_slave(bond, slave))
3674 break;
3675 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3676 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1da177e4 3677
78a646ce
NA		 3678 if (!skb2) {
3679 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3680 bond_dev->name);
3681 continue;
1da177e4 3682 }
78a646ce
NA		 3683 /* bond_dev_queue_xmit always returns 0 */
3684 bond_dev_queue_xmit(bond, skb2, slave->dev);
1da177e4
LT		 3685 }
3686 }
78a646ce
NA		 3687 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3688 bond_dev_queue_xmit(bond, skb, slave->dev);
3689 else
04502430 3690 kfree_skb(skb);
3d632c3f 3691
ec634fe3 3692 return NETDEV_TX_OK;
1da177e4
LT		 3693}
3694
3695/*------------------------- Device initialization ---------------------------*/
3696
bb1d9123
AG		 3697/*
3698 * Lookup the slave that corresponds to a qid
3699 */
3700static inline int bond_slave_override(struct bonding *bond,
3701 struct sk_buff *skb)
3702{
bb1d9123 3703 struct slave *slave = NULL;
9caff1e7 3704 struct list_head *iter;
bb1d9123 3705
0693e88e
MM		 3706 if (!skb->queue_mapping)
3707 return 1;
bb1d9123
AG		 3708
3709 /* Find out if any slaves have the same mapping as this skb. */
3900f290 3710 bond_for_each_slave_rcu(bond, slave, iter) {
3711 if (slave->queue_id == skb->queue_mapping) {
3712 if (slave_can_tx(slave)) {
3713 bond_dev_queue_xmit(bond, skb, slave->dev);
3714 return 0;
3715 }
3716 /* If the slave isn't UP, use default transmit policy. */
bb1d9123
AG		 3717 break;
3718 }
3719 }
3720
3900f290 3721 return 1;
bb1d9123
AG		 3722}
3723
374eeb5a 3724
f663dd9a 3725static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
99932d4f 3726 void *accel_priv, select_queue_fallback_t fallback)
bb1d9123
AG		 3727{
3728 /*
3729 * This helper function exists to help dev_pick_tx get the correct
fd0e435b 3730 * destination queue. Using a helper function skips a call to
bb1d9123
AG		 3731 * skb_tx_hash and will put the skbs in the queue we expect on their
3732 * way down to the bonding driver.
3733 */
fd0e435b
PO		 3734 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3735
374eeb5a
NH		 3736 /*
3737 * Save the original txq to restore before passing to the driver
3738 */
df4ab5b3 3739 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
374eeb5a 3740
fd0e435b 3741 if (unlikely(txq >= dev->real_num_tx_queues)) {
d30ee670 3742 do {
fd0e435b 3743 txq -= dev->real_num_tx_queues;
d30ee670 3744 } while (txq >= dev->real_num_tx_queues);
fd0e435b
PO		 3745 }
3746 return txq;
bb1d9123
AG		 3747}
3748
0693e88e 3749static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
00829823 3750{
bb1d9123
AG		 3751 struct bonding *bond = netdev_priv(dev);
3752
3753 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3754 if (!bond_slave_override(bond, skb))
3755 return NETDEV_TX_OK;
3756 }
00829823
SH		 3757
3758 switch (bond->params.mode) {
3759 case BOND_MODE_ROUNDROBIN:
3760 return bond_xmit_roundrobin(skb, dev);
3761 case BOND_MODE_ACTIVEBACKUP:
3762 return bond_xmit_activebackup(skb, dev);
3763 case BOND_MODE_XOR:
3764 return bond_xmit_xor(skb, dev);
3765 case BOND_MODE_BROADCAST:
3766 return bond_xmit_broadcast(skb, dev);
3767 case BOND_MODE_8023AD:
3768 return bond_3ad_xmit_xor(skb, dev);
3769 case BOND_MODE_ALB:
3770 case BOND_MODE_TLB:
3771 return bond_alb_xmit(skb, dev);
3772 default:
3773 /* Should never happen, mode already checked */
a4aee5c8
JP		 3774 pr_err("%s: Error: Unknown bonding mode %d\n",
3775 dev->name, bond->params.mode);
00829823 3776 WARN_ON_ONCE(1);
04502430 3777 kfree_skb(skb);
00829823
SH		 3778 return NETDEV_TX_OK;
3779 }
3780}
3781
0693e88e
MM		 3782static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3783{
3784 struct bonding *bond = netdev_priv(dev);
3785 netdev_tx_t ret = NETDEV_TX_OK;
3786
3787 /*
3788 * If we risk deadlock from transmitting this in the
3789 * netpoll path, tell netpoll to queue the frame for later tx
3790 */
3791 if (is_netpoll_tx_blocked(dev))
3792 return NETDEV_TX_BUSY;
3793
278b2083 3794 rcu_read_lock();
0965a1f3 3795 if (bond_has_slaves(bond))
0693e88e
MM		 3796 ret = __bond_start_xmit(skb, dev);
3797 else
04502430 3798 kfree_skb(skb);
278b2083 3799 rcu_read_unlock();
0693e88e
MM		 3800
3801 return ret;
3802}
00829823 3803
bb5b052f
AG		 3804static int bond_ethtool_get_settings(struct net_device *bond_dev,
3805 struct ethtool_cmd *ecmd)
3806{
3807 struct bonding *bond = netdev_priv(bond_dev);
bb5b052f 3808 unsigned long speed = 0;
9caff1e7 3809 struct list_head *iter;
dec1e90e 3810 struct slave *slave;
bb5b052f
AG		 3811
3812 ecmd->duplex = DUPLEX_UNKNOWN;
3813 ecmd->port = PORT_OTHER;
3814
3815 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3816 * do not need to check mode. Though link speed might not represent
3817 * the true receive or transmit bandwidth (not all modes are symmetric)
3818 * this is an accurate maximum.
3819 */
3820 read_lock(&bond->lock);
9caff1e7 3821 bond_for_each_slave(bond, slave, iter) {
bb5b052f
AG		 3822 if (SLAVE_IS_OK(slave)) {
3823 if (slave->speed != SPEED_UNKNOWN)
3824 speed += slave->speed;
3825 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3826 slave->duplex != DUPLEX_UNKNOWN)
3827 ecmd->duplex = slave->duplex;
3828 }
3829 }
3830 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3831 read_unlock(&bond->lock);
dec1e90e 3832
bb5b052f
AG		 3833 return 0;
3834}
3835
217df670 3836static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
7826d43f 3837 struct ethtool_drvinfo *drvinfo)
217df670 3838{
7826d43f
JP		 3839 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3840 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3841 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3842 BOND_ABI_VERSION);
217df670
JV		 3843}
3844
7282d491 3845static const struct ethtool_ops bond_ethtool_ops = {
217df670 3846 .get_drvinfo = bond_ethtool_get_drvinfo,
bb5b052f 3847 .get_settings = bond_ethtool_get_settings,
fa53ebac 3848 .get_link = ethtool_op_get_link,
8531c5ff
AK		 3849};
3850
eb7cc59a 3851static const struct net_device_ops bond_netdev_ops = {
181470fc 3852 .ndo_init = bond_init,
9e71626c 3853 .ndo_uninit = bond_uninit,
eb7cc59a
SH		 3854 .ndo_open = bond_open,
3855 .ndo_stop = bond_close,
00829823 3856 .ndo_start_xmit = bond_start_xmit,
bb1d9123 3857 .ndo_select_queue = bond_select_queue,
be1f3c2c 3858 .ndo_get_stats64 = bond_get_stats,
eb7cc59a 3859 .ndo_do_ioctl = bond_do_ioctl,
d03462b9 3860 .ndo_change_rx_flags = bond_change_rx_flags,
303d1cbf 3861 .ndo_set_rx_mode = bond_set_rx_mode,
eb7cc59a 3862 .ndo_change_mtu = bond_change_mtu,
cc0e4070 3863 .ndo_set_mac_address = bond_set_mac_address,
00829823 3864 .ndo_neigh_setup = bond_neigh_setup,
cc0e4070 3865 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
eb7cc59a 3866 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
f6dc31a8 3867#ifdef CONFIG_NET_POLL_CONTROLLER
8a8efa22 3868 .ndo_netpoll_setup = bond_netpoll_setup,
f6dc31a8
WC		 3869 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3870 .ndo_poll_controller = bond_poll_controller,
3871#endif
9232ecca
JP		 3872 .ndo_add_slave = bond_enslave,
3873 .ndo_del_slave = bond_release,
b2a103e6 3874 .ndo_fix_features = bond_fix_features,
eb7cc59a
SH		 3875};
3876
b3f92b63
DG		 3877static const struct device_type bond_type = {
3878 .name = "bond",
3879};
3880
9e2e61fb
AW		 3881static void bond_destructor(struct net_device *bond_dev)
3882{
3883 struct bonding *bond = netdev_priv(bond_dev);
3884 if (bond->wq)
3885 destroy_workqueue(bond->wq);
3886 free_netdev(bond_dev);
3887}
3888
0a2a78c4 3889void bond_setup(struct net_device *bond_dev)
1da177e4 3890{
454d7c9b 3891 struct bonding *bond = netdev_priv(bond_dev);
1da177e4 3892
1da177e4
LT		 3893 /* initialize rwlocks */
3894 rwlock_init(&bond->lock);
3895 rwlock_init(&bond->curr_slave_lock);
d2991f75 3896 bond->params = bonding_defaults;
1da177e4
LT		 3897
3898 /* Initialize pointers */
1da177e4 3899 bond->dev = bond_dev;
1da177e4
LT		 3900
3901 /* Initialize the device entry points */
181470fc 3902 ether_setup(bond_dev);
eb7cc59a 3903 bond_dev->netdev_ops = &bond_netdev_ops;
8531c5ff 3904 bond_dev->ethtool_ops = &bond_ethtool_ops;
1da177e4 3905
9e2e61fb 3906 bond_dev->destructor = bond_destructor;
1da177e4 3907
b3f92b63
DG		 3908 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3909
1da177e4
LT		 3910 /* Initialize the device options */
3911 bond_dev->tx_queue_len = 0;
3912 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
0b680e75 3913 bond_dev->priv_flags |= IFF_BONDING;
550fd08c 3914 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
181470fc 3915
1da177e4
LT		 3916 /* At first, we block adding VLANs. That's the only way to
3917 * prevent problems that occur when adding VLANs over an
3918 * empty bond. The block will be removed once non-challenged
3919 * slaves are enslaved.
3920 */
3921 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3922
932ff279 3923 /* don't acquire bond device's netif_tx_lock when
1da177e4
LT		 3924 * transmitting */
3925 bond_dev->features |= NETIF_F_LLTX;
3926
3927 /* By default, we declare the bond to be fully
3928 * VLAN hardware accelerated capable. Special
3929 * care is taken in the various xmit functions
3930 * when there are slaves that are not hw accel
3931 * capable
3932 */
1da177e4 3933
f9399814
WC		 3934 /* Don't allow bond devices to change network namespaces. */
3935 bond_dev->features |= NETIF_F_NETNS_LOCAL;
3936
b2a103e6 3937 bond_dev->hw_features = BOND_VLAN_FEATURES |
f646968f
PM		 3938 NETIF_F_HW_VLAN_CTAG_TX |
3939 NETIF_F_HW_VLAN_CTAG_RX |
3940 NETIF_F_HW_VLAN_CTAG_FILTER;
b2a103e6 3941
34324dc2 3942 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
b2a103e6 3943 bond_dev->features |= bond_dev->hw_features;
1da177e4
LT		 3944}
3945
c67dfb29
EB		 3946/*
3947* Destroy a bonding device.
3948* Must be under rtnl_lock when this function is called.
3949*/
3950static void bond_uninit(struct net_device *bond_dev)
a434e43f 3951{
454d7c9b 3952 struct bonding *bond = netdev_priv(bond_dev);
544a028e
VF		 3953 struct list_head *iter;
3954 struct slave *slave;
a434e43f 3955
f6dc31a8
WC		 3956 bond_netpoll_cleanup(bond_dev);
3957
c67dfb29 3958 /* Release the bonded slaves */
544a028e 3959 bond_for_each_slave(bond, slave, iter)
dec1e90e 3960 __bond_release_one(bond_dev, slave->dev, true);
0896341a 3961 pr_info("%s: released all slaves\n", bond_dev->name);
c67dfb29 3962
a434e43f
JV		 3963 list_del(&bond->bond_list);
3964
f073c7ca 3965 bond_debug_unregister(bond);
a434e43f
JV		 3966}
3967
1da177e4
LT		 3968/*------------------------- Module initialization ---------------------------*/
3969
ec029fac 3970int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3971{
3972 int i;
3973
3974 for (i = 0; tbl[i].modename; i++)
3975 if (mode == tbl[i].mode)
3976 return tbl[i].mode;
3977
3978 return -1;
3979}
3980
3981static int bond_parm_tbl_lookup_name(const char *modename,
3982 const struct bond_parm_tbl *tbl)
3983{
3984 int i;
3985
3986 for (i = 0; tbl[i].modename; i++)
3987 if (strcmp(modename, tbl[i].modename) == 0)
3988 return tbl[i].mode;
3989
3990 return -1;
3991}
3992
1da177e4
LT		 3993/*
3994 * Convert string input module parms. Accept either the
ece95f7f
JV		 3995 * number of the mode or its string name. A bit complicated because
3996 * some mode names are substrings of other names, and calls from sysfs
3997 * may have whitespace in the name (trailing newlines, for example).
1da177e4 3998 */
325dcf7a 3999int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
1da177e4 4000{
ec029fac 4001 int modeint;
4002 char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
ece95f7f 4003
a42e534f
JV		 4004 for (p = (char *)buf; *p; p++)
4005 if (!(isdigit(*p) || isspace(*p)))
4006 break;
4007
ec029fac 4008 if (*p && sscanf(buf, "%20s", modestr) != 0)
4009 return bond_parm_tbl_lookup_name(modestr, tbl);
4010 else if (sscanf(buf, "%d", &modeint) != 0)
4011 return bond_parm_tbl_lookup(modeint, tbl);
1da177e4
LT		 4012
4013 return -1;
4014}
4015
4016static int bond_check_params(struct bond_params *params)
4017{
5a5c5fd4 4018 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
2b3798d5 4019 struct bond_opt_value newval, *valptr;
8599b52e 4020 int arp_all_targets_value;
f5b2b966 4021
1da177e4
LT		 4022 /*
4023 * Convert string parameters.
4024 */
4025 if (mode) {
2b3798d5
NA		 4026 bond_opt_initstr(&newval, mode);
4027 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4028 if (!valptr) {
4029 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
1da177e4
LT		 4030 return -EINVAL;
4031 }
2b3798d5 4032 bond_mode = valptr->value;
1da177e4
LT		 4033 }
4034
169a3e66
JV		 4035 if (xmit_hash_policy) {
4036 if ((bond_mode != BOND_MODE_XOR) &&
4037 (bond_mode != BOND_MODE_8023AD)) {
a4aee5c8 4038 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
169a3e66
JV		 4039 bond_mode_name(bond_mode));
4040 } else {
a4b32ce7
NA		 4041 bond_opt_initstr(&newval, xmit_hash_policy);
4042 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4043 &newval);
4044 if (!valptr) {
a4aee5c8 4045 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
169a3e66
JV		 4046 xmit_hash_policy);
4047 return -EINVAL;
4048 }
a4b32ce7 4049 xmit_hashtype = valptr->value;
169a3e66
JV		 4050 }
4051 }
4052
1da177e4
LT		 4053 if (lacp_rate) {
4054 if (bond_mode != BOND_MODE_8023AD) {
a4aee5c8
JP		 4055 pr_info("lacp_rate param is irrelevant in mode %s\n",
4056 bond_mode_name(bond_mode));
1da177e4 4057 } else {
d3131de7
NA		 4058 bond_opt_initstr(&newval, lacp_rate);
4059 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4060 &newval);
4061 if (!valptr) {
a4aee5c8 4062 pr_err("Error: Invalid lacp rate \"%s\"\n",
d3131de7 4063 lacp_rate);
1da177e4
LT		 4064 return -EINVAL;
4065 }
d3131de7 4066 lacp_fast = valptr->value;
1da177e4
LT		 4067 }
4068 }
4069
fd989c83 4070 if (ad_select) {
9e5f5eeb
NA		 4071 bond_opt_initstr(&newval, lacp_rate);
4072 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4073 &newval);
4074 if (!valptr) {
4075 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
fd989c83
JV		 4076 return -EINVAL;
4077 }
9e5f5eeb
NA		 4078 params->ad_select = valptr->value;
4079 if (bond_mode != BOND_MODE_8023AD)
a4aee5c8 4080 pr_warning("ad_select param only affects 802.3ad mode\n");
fd989c83
JV		 4081 } else {
4082 params->ad_select = BOND_AD_STABLE;
4083 }
4084
f5841306 4085 if (max_bonds < 0) {
a4aee5c8
JP		 4086 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4087 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
1da177e4
LT		 4088 max_bonds = BOND_DEFAULT_MAX_BONDS;
4089 }
4090
4091 if (miimon < 0) {
b98d9c66
NA		 4092 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4093 miimon, INT_MAX);
4094 miimon = 0;
1da177e4
LT		 4095 }
4096
4097 if (updelay < 0) {
a4aee5c8
JP		 4098 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4099 updelay, INT_MAX);
1da177e4
LT		 4100 updelay = 0;
4101 }
4102
4103 if (downdelay < 0) {
a4aee5c8
JP		 4104 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4105 downdelay, INT_MAX);
1da177e4
LT		 4106 downdelay = 0;
4107 }
4108
4109 if ((use_carrier != 0) && (use_carrier != 1)) {
a4aee5c8
JP		 4110 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4111 use_carrier);
1da177e4
LT		 4112 use_carrier = 1;
4113 }
4114
ad246c99
BH		 4115 if (num_peer_notif < 0 || num_peer_notif > 255) {
4116 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4117 num_peer_notif);
4118 num_peer_notif = 1;
4119 }
4120
834db4bc 4121 /* reset values for 802.3ad/TLB/ALB */
4122 if (BOND_NO_USES_ARP(bond_mode)) {
1da177e4 4123 if (!miimon) {
a4aee5c8 4124 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
3d632c3f 4125 pr_warning("Forcing miimon to 100msec\n");
fe9d04af 4126 miimon = BOND_DEFAULT_MIIMON;
1da177e4
LT		 4127 }
4128 }
4129
bb1d9123
AG		 4130 if (tx_queues < 1 || tx_queues > 255) {
4131 pr_warning("Warning: tx_queues (%d) should be between "
4132 "1 and 255, resetting to %d\n",
4133 tx_queues, BOND_DEFAULT_TX_QUEUES);
4134 tx_queues = BOND_DEFAULT_TX_QUEUES;
4135 }
4136
ebd8e497
AG		 4137 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4138 pr_warning("Warning: all_slaves_active module parameter (%d), "
4139 "not of valid value (0/1), so it was set to "
4140 "0\n", all_slaves_active);
4141 all_slaves_active = 0;
4142 }
4143
c2952c31
FL		 4144 if (resend_igmp < 0 || resend_igmp > 255) {
4145 pr_warning("Warning: resend_igmp (%d) should be between "
4146 "0 and 255, resetting to %d\n",
4147 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4148 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4149 }
4150
aa59d851
NA		 4151 bond_opt_initval(&newval, packets_per_slave);
4152 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
73958329
NA		 4153 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4154 packets_per_slave, USHRT_MAX);
4155 packets_per_slave = 1;
4156 }
4157
1da177e4 4158 if (bond_mode == BOND_MODE_ALB) {
a4aee5c8
JP		 4159 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",
4160 updelay);
1da177e4
LT		 4161 }
4162
4163 if (!miimon) {
4164 if (updelay || downdelay) {
4165 /* just warn the user the up/down delay will have
4166 * no effect since miimon is zero...
4167 */
a4aee5c8
JP		 4168 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4169 updelay, downdelay);
1da177e4
LT		 4170 }
4171 } else {
4172 /* don't allow arp monitoring */
4173 if (arp_interval) {
a4aee5c8
JP		 4174 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4175 miimon, arp_interval);
1da177e4
LT		 4176 arp_interval = 0;
4177 }
4178
4179 if ((updelay % miimon) != 0) {
a4aee5c8
JP		 4180 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4181 updelay, miimon,
4182 (updelay / miimon) * miimon);
1da177e4
LT		 4183 }
4184
4185 updelay /= miimon;
4186
4187 if ((downdelay % miimon) != 0) {
a4aee5c8
JP		 4188 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4189 downdelay, miimon,
4190 (downdelay / miimon) * miimon);
1da177e4
LT		 4191 }
4192
4193 downdelay /= miimon;
4194 }
4195
4196 if (arp_interval < 0) {
7bdb04ed
NA		 4197 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to 0\n",
4198 arp_interval, INT_MAX);
4199 arp_interval = 0;
1da177e4
LT		 4200 }
4201
5a5c5fd4 4202 for (arp_ip_count = 0, i = 0;
4203 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
1da177e4
LT		 4204 /* not complete check, but should be good enough to
4205 catch mistakes */
89015c18 4206 __be32 ip;
4207 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4208 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
a4aee5c8 4209 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5a5c5fd4 4210 arp_ip_target[i]);
1da177e4
LT		 4211 arp_interval = 0;
4212 } else {
0afee4e8
VF		 4213 if (bond_get_targets_ip(arp_target, ip) == -1)
4214 arp_target[arp_ip_count++] = ip;
4215 else
4216 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4217 &ip);
1da177e4
LT		 4218 }
4219 }
4220
4221 if (arp_interval && !arp_ip_count) {
4222 /* don't allow arping if no arp_ip_target given... */
a4aee5c8
JP		 4223 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4224 arp_interval);
1da177e4
LT		 4225 arp_interval = 0;
4226 }
4227
f5b2b966
JV		 4228 if (arp_validate) {
4229 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
a4aee5c8 4230 pr_err("arp_validate only supported in active-backup mode\n");
f5b2b966
JV		 4231 return -EINVAL;
4232 }
4233 if (!arp_interval) {
a4aee5c8 4234 pr_err("arp_validate requires arp_interval\n");
f5b2b966
JV		 4235 return -EINVAL;
4236 }
4237
16228881
NA		 4238 bond_opt_initstr(&newval, arp_validate);
4239 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4240 &newval);
4241 if (!valptr) {
a4aee5c8 4242 pr_err("Error: invalid arp_validate \"%s\"\n",
16228881 4243 arp_validate);
f5b2b966
JV		 4244 return -EINVAL;
4245 }
16228881
NA		 4246 arp_validate_value = valptr->value;
4247 } else {
f5b2b966 4248 arp_validate_value = 0;
16228881 4249 }
f5b2b966 4250
8599b52e
VF		 4251 arp_all_targets_value = 0;
4252 if (arp_all_targets) {
edf36b24
NA		 4253 bond_opt_initstr(&newval, arp_all_targets);
4254 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4255 &newval);
4256 if (!valptr) {
8599b52e
VF		 4257 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4258 arp_all_targets);
4259 arp_all_targets_value = 0;
edf36b24
NA		 4260 } else {
4261 arp_all_targets_value = valptr->value;
8599b52e
VF		 4262 }
4263 }
4264
1da177e4 4265 if (miimon) {
a4aee5c8 4266 pr_info("MII link monitoring set to %d ms\n", miimon);
1da177e4 4267 } else if (arp_interval) {
16228881
NA		 4268 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4269 arp_validate_value);
a4aee5c8 4270 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
16228881 4271 arp_interval, valptr->string, arp_ip_count);
1da177e4
LT		 4272
4273 for (i = 0; i < arp_ip_count; i++)
e5e2a8fd 4274 pr_info(" %s", arp_ip_target[i]);
1da177e4 4275
e5e2a8fd 4276 pr_info("\n");
1da177e4 4277
b8a9787e 4278 } else if (max_bonds) {
1da177e4
LT		 4279 /* miimon and arp_interval not set, we need one so things
4280 * work as expected, see bonding.txt for details
4281 */
b2730f4f 4282 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		 4283 }
4284
4285 if (primary && !USES_PRIMARY(bond_mode)) {
4286 /* currently, using a primary only makes sense
4287 * in active backup, TLB or ALB modes
4288 */
a4aee5c8
JP		 4289 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4290 primary, bond_mode_name(bond_mode));
1da177e4
LT		 4291 primary = NULL;
4292 }
4293
a549952a 4294 if (primary && primary_reselect) {
388d3a6d
NA		 4295 bond_opt_initstr(&newval, primary_reselect);
4296 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4297 &newval);
4298 if (!valptr) {
a4aee5c8 4299 pr_err("Error: Invalid primary_reselect \"%s\"\n",
388d3a6d 4300 primary_reselect);
a549952a
JP		 4301 return -EINVAL;
4302 }
388d3a6d 4303 primary_reselect_value = valptr->value;
a549952a
JP		 4304 } else {
4305 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4306 }
4307
3915c1e8 4308 if (fail_over_mac) {
1df6b6aa
NA		 4309 bond_opt_initstr(&newval, fail_over_mac);
4310 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4311 &newval);
4312 if (!valptr) {
a4aee5c8 4313 pr_err("Error: invalid fail_over_mac \"%s\"\n",
1df6b6aa 4314 fail_over_mac);
3915c1e8
JV		 4315 return -EINVAL;
4316 }
1df6b6aa 4317 fail_over_mac_value = valptr->value;
3915c1e8 4318 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
a4aee5c8 4319 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
3915c1e8
JV		 4320 } else {
4321 fail_over_mac_value = BOND_FOM_NONE;
4322 }
dd957c57 4323
3a7129e5 4324 if (lp_interval == 0) {
4325 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4326 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4327 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4328 }
4329
1da177e4
LT		 4330 /* fill params struct with the proper values */
4331 params->mode = bond_mode;
169a3e66 4332 params->xmit_policy = xmit_hashtype;
1da177e4 4333 params->miimon = miimon;
ad246c99 4334 params->num_peer_notif = num_peer_notif;
1da177e4 4335 params->arp_interval = arp_interval;
f5b2b966 4336 params->arp_validate = arp_validate_value;
8599b52e 4337 params->arp_all_targets = arp_all_targets_value;
1da177e4
LT		 4338 params->updelay = updelay;
4339 params->downdelay = downdelay;
4340 params->use_carrier = use_carrier;
4341 params->lacp_fast = lacp_fast;
4342 params->primary[0] = 0;
a549952a 4343 params->primary_reselect = primary_reselect_value;
3915c1e8 4344 params->fail_over_mac = fail_over_mac_value;
bb1d9123 4345 params->tx_queues = tx_queues;
ebd8e497 4346 params->all_slaves_active = all_slaves_active;
c2952c31 4347 params->resend_igmp = resend_igmp;
655f8919 4348 params->min_links = min_links;
3a7129e5 4349 params->lp_interval = lp_interval;
809fa972
HFS		 4350 params->packets_per_slave = packets_per_slave;
4351 if (packets_per_slave > 0) {
4352 params->reciprocal_packets_per_slave =
4353 reciprocal_value(packets_per_slave);
4354 } else {
4355 /* reciprocal_packets_per_slave is unused if
4356 * packets_per_slave is 0 or 1, just initialize it
4357 */
4358 params->reciprocal_packets_per_slave =
4359 (struct reciprocal_value) { 0 };
4360 }
4361
1da177e4
LT		 4362 if (primary) {
4363 strncpy(params->primary, primary, IFNAMSIZ);
4364 params->primary[IFNAMSIZ - 1] = 0;
4365 }
4366
4367 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4368
4369 return 0;
4370}
4371
0daa2303 4372static struct lock_class_key bonding_netdev_xmit_lock_key;
cf508b12 4373static struct lock_class_key bonding_netdev_addr_lock_key;
49ee4920 4374static struct lock_class_key bonding_tx_busylock_key;
0daa2303 4375
e8a0464c
DM		 4376static void bond_set_lockdep_class_one(struct net_device *dev,
4377 struct netdev_queue *txq,
4378 void *_unused)
c773e847
DM		 4379{
4380 lockdep_set_class(&txq->_xmit_lock,
4381 &bonding_netdev_xmit_lock_key);
4382}
4383
4384static void bond_set_lockdep_class(struct net_device *dev)
4385{
cf508b12
DM		 4386 lockdep_set_class(&dev->addr_list_lock,
4387 &bonding_netdev_addr_lock_key);
e8a0464c 4388 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
49ee4920 4389 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
c773e847
DM		 4390}
4391
181470fc
SH		 4392/*
4393 * Called from registration process
4394 */
4395static int bond_init(struct net_device *bond_dev)
4396{
4397 struct bonding *bond = netdev_priv(bond_dev);
ec87fd3b 4398 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
9fe0617d 4399 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
181470fc
SH		 4400
4401 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4402
9fe0617d
NH		 4403 /*
4404 * Initialize locks that may be required during
4405 * en/deslave operations. All of the bond_open work
4406 * (of which this is part) should really be moved to
4407 * a phase prior to dev_open
4408 */
4409 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4410 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4411
181470fc
SH		 4412 bond->wq = create_singlethread_workqueue(bond_dev->name);
4413 if (!bond->wq)
4414 return -ENOMEM;
4415
4416 bond_set_lockdep_class(bond_dev);
4417
ec87fd3b 4418 list_add_tail(&bond->bond_list, &bn->dev_list);
181470fc 4419
6151b3d4 4420 bond_prepare_sysfs_group(bond);
22bedad3 4421
f073c7ca
TI		 4422 bond_debug_register(bond);
4423
409cc1f8
JP		 4424 /* Ensure valid dev_addr */
4425 if (is_zero_ether_addr(bond_dev->dev_addr) &&
97a1e639 4426 bond_dev->addr_assign_type == NET_ADDR_PERM)
409cc1f8 4427 eth_hw_addr_random(bond_dev);
409cc1f8 4428
181470fc
SH		 4429 return 0;
4430}
4431
0a2a78c4 4432unsigned int bond_get_num_tx_queues(void)
d5da4510 4433{
efacb309 4434 return tx_queues;
d5da4510
JP		 4435}
4436
dfe60397 4437/* Create a new bond based on the specified name and bonding parameters.
e4b91c48 4438 * If name is NULL, obtain a suitable "bond%d" name for us.
dfe60397
MW		 4439 * Caller must NOT hold rtnl_lock; we need to release it here before we
4440 * set up our sysfs entries.
4441 */
ec87fd3b 4442int bond_create(struct net *net, const char *name)
dfe60397
MW		 4443{
4444 struct net_device *bond_dev;
4445 int res;
4446
4447 rtnl_lock();
027ea041 4448
1c5cae81
JP		 4449 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4450 name ? name : "bond%d",
4451 bond_setup, tx_queues);
dfe60397 4452 if (!bond_dev) {
a4aee5c8 4453 pr_err("%s: eek! can't alloc netdev!\n", name);
9e2e61fb
AW		 4454 rtnl_unlock();
4455 return -ENOMEM;
dfe60397
MW		 4456 }
4457
ec87fd3b 4458 dev_net_set(bond_dev, net);
88ead977
EB		 4459 bond_dev->rtnl_link_ops = &bond_link_ops;
4460
dfe60397 4461 res = register_netdevice(bond_dev);
0daa2303 4462
e826eafa
PO		 4463 netif_carrier_off(bond_dev);
4464
7e083840 4465 rtnl_unlock();
9e2e61fb
AW		 4466 if (res < 0)
4467 bond_destructor(bond_dev);
30c15ba9 4468 return res;
dfe60397
MW		 4469}
4470
2c8c1e72 4471static int __net_init bond_net_init(struct net *net)
ec87fd3b 4472{
15449745 4473 struct bond_net *bn = net_generic(net, bond_net_id);
ec87fd3b
EB		 4474
4475 bn->net = net;
4476 INIT_LIST_HEAD(&bn->dev_list);
4477
ec87fd3b 4478 bond_create_proc_dir(bn);
4c22400a 4479 bond_create_sysfs(bn);
87a7b84b 4480
15449745 4481 return 0;
ec87fd3b
EB		 4482}
4483
2c8c1e72 4484static void __net_exit bond_net_exit(struct net *net)
ec87fd3b 4485{
15449745 4486 struct bond_net *bn = net_generic(net, bond_net_id);
69b0216a 4487 struct bonding *bond, *tmp_bond;
4488 LIST_HEAD(list);
ec87fd3b 4489
4c22400a 4490 bond_destroy_sysfs(bn);
ec87fd3b 4491 bond_destroy_proc_dir(bn);
69b0216a 4492
4493 /* Kill off any bonds created after unregistering bond rtnl ops */
4494 rtnl_lock();
4495 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4496 unregister_netdevice_queue(bond->dev, &list);
4497 unregister_netdevice_many(&list);
4498 rtnl_unlock();
ec87fd3b
EB		 4499}
4500
4501static struct pernet_operations bond_net_ops = {
4502 .init = bond_net_init,
4503 .exit = bond_net_exit,
15449745
EB		 4504 .id = &bond_net_id,
4505 .size = sizeof(struct bond_net),
ec87fd3b
EB		 4506};
4507
1da177e4
LT		 4508static int __init bonding_init(void)
4509{
1da177e4
LT		 4510 int i;
4511 int res;
4512
bd33acc3 4513 pr_info("%s", bond_version);
1da177e4 4514
dfe60397 4515 res = bond_check_params(&bonding_defaults);
3d632c3f 4516 if (res)
dfe60397 4517 goto out;
1da177e4 4518
15449745 4519 res = register_pernet_subsys(&bond_net_ops);
ec87fd3b
EB		 4520 if (res)
4521 goto out;
027ea041 4522
0a2a78c4 4523 res = bond_netlink_init();
88ead977 4524 if (res)
6639104b 4525 goto err_link;
88ead977 4526
f073c7ca
TI		 4527 bond_create_debugfs();
4528
1da177e4 4529 for (i = 0; i < max_bonds; i++) {
ec87fd3b 4530 res = bond_create(&init_net, NULL);
dfe60397
MW		 4531 if (res)
4532 goto err;
1da177e4
LT		 4533 }
4534
1da177e4 4535 register_netdevice_notifier(&bond_netdev_notifier);
dfe60397 4536out:
1da177e4 4537 return res;
88ead977 4538err:
0a2a78c4 4539 bond_netlink_fini();
6639104b 4540err_link:
15449745 4541 unregister_pernet_subsys(&bond_net_ops);
88ead977 4542 goto out;
dfe60397 4543
1da177e4
LT		 4544}
4545
4546static void __exit bonding_exit(void)
4547{
4548 unregister_netdevice_notifier(&bond_netdev_notifier);
4549
f073c7ca 4550 bond_destroy_debugfs();
ae68c398 4551
0a2a78c4 4552 bond_netlink_fini();
ffcdedb6 4553 unregister_pernet_subsys(&bond_net_ops);
e843fa50
NH		 4554
4555#ifdef CONFIG_NET_POLL_CONTROLLER
fb4fa76a
NH		 4556 /*
4557 * Make sure we don't have an imbalance on our netpoll blocking
4558 */
4559 WARN_ON(atomic_read(&netpoll_block_tx));
e843fa50 4560#endif
1da177e4
LT		 4561}
4562
4563module_init(bonding_init);
4564module_exit(bonding_exit);
4565MODULE_LICENSE("GPL");
4566MODULE_VERSION(DRV_VERSION);
4567MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4568MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
Ubuntu Jammy Linux kernel mirror