]> pilppa.org Git - linux-2.6-omap-h63xx.git/blob - drivers/net/bonding/bond_main.c
Merge branch 'upstream-fixes'
[linux-2.6-omap-h63xx.git] / drivers / net / bonding / bond_main.c
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  *
32  */
33
34 //#define BONDING_DEBUG 1
35
36 #include <linux/config.h>
37 #include <linux/kernel.h>
38 #include <linux/module.h>
39 #include <linux/sched.h>
40 #include <linux/types.h>
41 #include <linux/fcntl.h>
42 #include <linux/interrupt.h>
43 #include <linux/ptrace.h>
44 #include <linux/ioport.h>
45 #include <linux/in.h>
46 #include <net/ip.h>
47 #include <linux/ip.h>
48 #include <linux/tcp.h>
49 #include <linux/udp.h>
50 #include <linux/slab.h>
51 #include <linux/string.h>
52 #include <linux/init.h>
53 #include <linux/timer.h>
54 #include <linux/socket.h>
55 #include <linux/ctype.h>
56 #include <linux/inet.h>
57 #include <linux/bitops.h>
58 #include <asm/system.h>
59 #include <asm/io.h>
60 #include <asm/dma.h>
61 #include <asm/uaccess.h>
62 #include <linux/errno.h>
63 #include <linux/netdevice.h>
64 #include <linux/inetdevice.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
73 #include <net/arp.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <net/route.h>
79 #include "bonding.h"
80 #include "bond_3ad.h"
81 #include "bond_alb.h"
82
83 /*---------------------------- Module parameters ----------------------------*/
84
85 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
86 #define BOND_LINK_MON_INTERV    0
87 #define BOND_LINK_ARP_INTERV    0
88
89 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
90 static int miimon       = BOND_LINK_MON_INTERV;
91 static int updelay      = 0;
92 static int downdelay    = 0;
93 static int use_carrier  = 1;
94 static char *mode       = NULL;
95 static char *primary    = NULL;
96 static char *lacp_rate  = NULL;
97 static char *xmit_hash_policy = NULL;
98 static int arp_interval = BOND_LINK_ARP_INTERV;
99 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
100 struct bond_params bonding_defaults;
101
102 module_param(max_bonds, int, 0);
103 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
104 module_param(miimon, int, 0);
105 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
106 module_param(updelay, int, 0);
107 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
108 module_param(downdelay, int, 0);
109 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
110                             "in milliseconds");
111 module_param(use_carrier, int, 0);
112 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
113                               "0 for off, 1 for on (default)");
114 module_param(mode, charp, 0);
115 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
116                        "1 for active-backup, 2 for balance-xor, "
117                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
118                        "6 for balance-alb");
119 module_param(primary, charp, 0);
120 MODULE_PARM_DESC(primary, "Primary network device to use");
121 module_param(lacp_rate, charp, 0);
122 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
123                             "(slow/fast)");
124 module_param(xmit_hash_policy, charp, 0);
125 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
126                                    ", 1 for layer 3+4");
127 module_param(arp_interval, int, 0);
128 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
129 module_param_array(arp_ip_target, charp, NULL, 0);
130 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
131
132 /*----------------------------- Global variables ----------------------------*/
133
134 static const char *version =
135         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
136
137 LIST_HEAD(bond_dev_list);
138
139 #ifdef CONFIG_PROC_FS
140 static struct proc_dir_entry *bond_proc_dir = NULL;
141 #endif
142
143 extern struct rw_semaphore bonding_rwsem;
144 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
145 static int arp_ip_count = 0;
146 static int bond_mode    = BOND_MODE_ROUNDROBIN;
147 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
148 static int lacp_fast    = 0;
149
150
151 struct bond_parm_tbl bond_lacp_tbl[] = {
152 {       "slow",         AD_LACP_SLOW},
153 {       "fast",         AD_LACP_FAST},
154 {       NULL,           -1},
155 };
156
157 struct bond_parm_tbl bond_mode_tbl[] = {
158 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
159 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
160 {       "balance-xor",          BOND_MODE_XOR},
161 {       "broadcast",            BOND_MODE_BROADCAST},
162 {       "802.3ad",              BOND_MODE_8023AD},
163 {       "balance-tlb",          BOND_MODE_TLB},
164 {       "balance-alb",          BOND_MODE_ALB},
165 {       NULL,                   -1},
166 };
167
168 struct bond_parm_tbl xmit_hashtype_tbl[] = {
169 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
170 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
171 {       NULL,                   -1},
172 };
173
174 /*-------------------------- Forward declarations ---------------------------*/
175
176 static void bond_send_gratuitous_arp(struct bonding *bond);
177
178 /*---------------------------- General routines -----------------------------*/
179
180 const char *bond_mode_name(int mode)
181 {
182         switch (mode) {
183         case BOND_MODE_ROUNDROBIN :
184                 return "load balancing (round-robin)";
185         case BOND_MODE_ACTIVEBACKUP :
186                 return "fault-tolerance (active-backup)";
187         case BOND_MODE_XOR :
188                 return "load balancing (xor)";
189         case BOND_MODE_BROADCAST :
190                 return "fault-tolerance (broadcast)";
191         case BOND_MODE_8023AD:
192                 return "IEEE 802.3ad Dynamic link aggregation";
193         case BOND_MODE_TLB:
194                 return "transmit load balancing";
195         case BOND_MODE_ALB:
196                 return "adaptive load balancing";
197         default:
198                 return "unknown";
199         }
200 }
201
202 /*---------------------------------- VLAN -----------------------------------*/
203
204 /**
205  * bond_add_vlan - add a new vlan id on bond
206  * @bond: bond that got the notification
207  * @vlan_id: the vlan id to add
208  *
209  * Returns -ENOMEM if allocation failed.
210  */
211 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
212 {
213         struct vlan_entry *vlan;
214
215         dprintk("bond: %s, vlan id %d\n",
216                 (bond ? bond->dev->name: "None"), vlan_id);
217
218         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
219         if (!vlan) {
220                 return -ENOMEM;
221         }
222
223         INIT_LIST_HEAD(&vlan->vlan_list);
224         vlan->vlan_id = vlan_id;
225         vlan->vlan_ip = 0;
226
227         write_lock_bh(&bond->lock);
228
229         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
230
231         write_unlock_bh(&bond->lock);
232
233         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
234
235         return 0;
236 }
237
238 /**
239  * bond_del_vlan - delete a vlan id from bond
240  * @bond: bond that got the notification
241  * @vlan_id: the vlan id to delete
242  *
243  * returns -ENODEV if @vlan_id was not found in @bond.
244  */
245 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
246 {
247         struct vlan_entry *vlan, *next;
248         int res = -ENODEV;
249
250         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
251
252         write_lock_bh(&bond->lock);
253
254         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
255                 if (vlan->vlan_id == vlan_id) {
256                         list_del(&vlan->vlan_list);
257
258                         if ((bond->params.mode == BOND_MODE_TLB) ||
259                             (bond->params.mode == BOND_MODE_ALB)) {
260                                 bond_alb_clear_vlan(bond, vlan_id);
261                         }
262
263                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
264                                 bond->dev->name);
265
266                         kfree(vlan);
267
268                         if (list_empty(&bond->vlan_list) &&
269                             (bond->slave_cnt == 0)) {
270                                 /* Last VLAN removed and no slaves, so
271                                  * restore block on adding VLANs. This will
272                                  * be removed once new slaves that are not
273                                  * VLAN challenged will be added.
274                                  */
275                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
276                         }
277
278                         res = 0;
279                         goto out;
280                 }
281         }
282
283         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
284                 bond->dev->name);
285
286 out:
287         write_unlock_bh(&bond->lock);
288         return res;
289 }
290
291 /**
292  * bond_has_challenged_slaves
293  * @bond: the bond we're working on
294  *
295  * Searches the slave list. Returns 1 if a vlan challenged slave
296  * was found, 0 otherwise.
297  *
298  * Assumes bond->lock is held.
299  */
300 static int bond_has_challenged_slaves(struct bonding *bond)
301 {
302         struct slave *slave;
303         int i;
304
305         bond_for_each_slave(bond, slave, i) {
306                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
307                         dprintk("found VLAN challenged slave - %s\n",
308                                 slave->dev->name);
309                         return 1;
310                 }
311         }
312
313         dprintk("no VLAN challenged slaves found\n");
314         return 0;
315 }
316
317 /**
318  * bond_next_vlan - safely skip to the next item in the vlans list.
319  * @bond: the bond we're working on
320  * @curr: item we're advancing from
321  *
322  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
323  * or @curr->next otherwise (even if it is @curr itself again).
324  * 
325  * Caller must hold bond->lock
326  */
327 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
328 {
329         struct vlan_entry *next, *last;
330
331         if (list_empty(&bond->vlan_list)) {
332                 return NULL;
333         }
334
335         if (!curr) {
336                 next = list_entry(bond->vlan_list.next,
337                                   struct vlan_entry, vlan_list);
338         } else {
339                 last = list_entry(bond->vlan_list.prev,
340                                   struct vlan_entry, vlan_list);
341                 if (last == curr) {
342                         next = list_entry(bond->vlan_list.next,
343                                           struct vlan_entry, vlan_list);
344                 } else {
345                         next = list_entry(curr->vlan_list.next,
346                                           struct vlan_entry, vlan_list);
347                 }
348         }
349
350         return next;
351 }
352
353 /**
354  * bond_dev_queue_xmit - Prepare skb for xmit.
355  * 
356  * @bond: bond device that got this skb for tx.
357  * @skb: hw accel VLAN tagged skb to transmit
358  * @slave_dev: slave that is supposed to xmit this skbuff
359  * 
360  * When the bond gets an skb to transmit that is
361  * already hardware accelerated VLAN tagged, and it
362  * needs to relay this skb to a slave that is not
363  * hw accel capable, the skb needs to be "unaccelerated",
364  * i.e. strip the hwaccel tag and re-insert it as part
365  * of the payload.
366  */
367 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
368 {
369         unsigned short vlan_id;
370
371         if (!list_empty(&bond->vlan_list) &&
372             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
373             vlan_get_tag(skb, &vlan_id) == 0) {
374                 skb->dev = slave_dev;
375                 skb = vlan_put_tag(skb, vlan_id);
376                 if (!skb) {
377                         /* vlan_put_tag() frees the skb in case of error,
378                          * so return success here so the calling functions
379                          * won't attempt to free is again.
380                          */
381                         return 0;
382                 }
383         } else {
384                 skb->dev = slave_dev;
385         }
386
387         skb->priority = 1;
388         dev_queue_xmit(skb);
389
390         return 0;
391 }
392
393 /*
394  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
395  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
396  * lock because:
397  * a. This operation is performed in IOCTL context,
398  * b. The operation is protected by the RTNL semaphore in the 8021q code,
399  * c. Holding a lock with BH disabled while directly calling a base driver
400  *    entry point is generally a BAD idea.
401  * 
402  * The design of synchronization/protection for this operation in the 8021q
403  * module is good for one or more VLAN devices over a single physical device
404  * and cannot be extended for a teaming solution like bonding, so there is a
405  * potential race condition here where a net device from the vlan group might
406  * be referenced (either by a base driver or the 8021q code) while it is being
407  * removed from the system. However, it turns out we're not making matters
408  * worse, and if it works for regular VLAN usage it will work here too.
409 */
410
411 /**
412  * bond_vlan_rx_register - Propagates registration to slaves
413  * @bond_dev: bonding net device that got called
414  * @grp: vlan group being registered
415  */
416 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
417 {
418         struct bonding *bond = bond_dev->priv;
419         struct slave *slave;
420         int i;
421
422         bond->vlgrp = grp;
423
424         bond_for_each_slave(bond, slave, i) {
425                 struct net_device *slave_dev = slave->dev;
426
427                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
428                     slave_dev->vlan_rx_register) {
429                         slave_dev->vlan_rx_register(slave_dev, grp);
430                 }
431         }
432 }
433
434 /**
435  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
436  * @bond_dev: bonding net device that got called
437  * @vid: vlan id being added
438  */
439 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
440 {
441         struct bonding *bond = bond_dev->priv;
442         struct slave *slave;
443         int i, res;
444
445         bond_for_each_slave(bond, slave, i) {
446                 struct net_device *slave_dev = slave->dev;
447
448                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
449                     slave_dev->vlan_rx_add_vid) {
450                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
451                 }
452         }
453
454         res = bond_add_vlan(bond, vid);
455         if (res) {
456                 printk(KERN_ERR DRV_NAME
457                        ": %s: Error: Failed to add vlan id %d\n",
458                        bond_dev->name, vid);
459         }
460 }
461
462 /**
463  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
464  * @bond_dev: bonding net device that got called
465  * @vid: vlan id being removed
466  */
467 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
468 {
469         struct bonding *bond = bond_dev->priv;
470         struct slave *slave;
471         struct net_device *vlan_dev;
472         int i, res;
473
474         bond_for_each_slave(bond, slave, i) {
475                 struct net_device *slave_dev = slave->dev;
476
477                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
478                     slave_dev->vlan_rx_kill_vid) {
479                         /* Save and then restore vlan_dev in the grp array,
480                          * since the slave's driver might clear it.
481                          */
482                         vlan_dev = bond->vlgrp->vlan_devices[vid];
483                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
484                         bond->vlgrp->vlan_devices[vid] = vlan_dev;
485                 }
486         }
487
488         res = bond_del_vlan(bond, vid);
489         if (res) {
490                 printk(KERN_ERR DRV_NAME
491                        ": %s: Error: Failed to remove vlan id %d\n",
492                        bond_dev->name, vid);
493         }
494 }
495
496 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
497 {
498         struct vlan_entry *vlan;
499
500         write_lock_bh(&bond->lock);
501
502         if (list_empty(&bond->vlan_list)) {
503                 goto out;
504         }
505
506         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
507             slave_dev->vlan_rx_register) {
508                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
509         }
510
511         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
512             !(slave_dev->vlan_rx_add_vid)) {
513                 goto out;
514         }
515
516         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
517                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
518         }
519
520 out:
521         write_unlock_bh(&bond->lock);
522 }
523
524 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
525 {
526         struct vlan_entry *vlan;
527         struct net_device *vlan_dev;
528
529         write_lock_bh(&bond->lock);
530
531         if (list_empty(&bond->vlan_list)) {
532                 goto out;
533         }
534
535         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
536             !(slave_dev->vlan_rx_kill_vid)) {
537                 goto unreg;
538         }
539
540         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
541                 /* Save and then restore vlan_dev in the grp array,
542                  * since the slave's driver might clear it.
543                  */
544                 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
545                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
546                 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
547         }
548
549 unreg:
550         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
551             slave_dev->vlan_rx_register) {
552                 slave_dev->vlan_rx_register(slave_dev, NULL);
553         }
554
555 out:
556         write_unlock_bh(&bond->lock);
557 }
558
559 /*------------------------------- Link status -------------------------------*/
560
561 /*
562  * Get link speed and duplex from the slave's base driver
563  * using ethtool. If for some reason the call fails or the
564  * values are invalid, fake speed and duplex to 100/Full
565  * and return error.
566  */
567 static int bond_update_speed_duplex(struct slave *slave)
568 {
569         struct net_device *slave_dev = slave->dev;
570         static int (* ioctl)(struct net_device *, struct ifreq *, int);
571         struct ifreq ifr;
572         struct ethtool_cmd etool;
573
574         /* Fake speed and duplex */
575         slave->speed = SPEED_100;
576         slave->duplex = DUPLEX_FULL;
577
578         if (slave_dev->ethtool_ops) {
579                 u32 res;
580
581                 if (!slave_dev->ethtool_ops->get_settings) {
582                         return -1;
583                 }
584
585                 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
586                 if (res < 0) {
587                         return -1;
588                 }
589
590                 goto verify;
591         }
592
593         ioctl = slave_dev->do_ioctl;
594         strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
595         etool.cmd = ETHTOOL_GSET;
596         ifr.ifr_data = (char*)&etool;
597         if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
598                 return -1;
599         }
600
601 verify:
602         switch (etool.speed) {
603         case SPEED_10:
604         case SPEED_100:
605         case SPEED_1000:
606                 break;
607         default:
608                 return -1;
609         }
610
611         switch (etool.duplex) {
612         case DUPLEX_FULL:
613         case DUPLEX_HALF:
614                 break;
615         default:
616                 return -1;
617         }
618
619         slave->speed = etool.speed;
620         slave->duplex = etool.duplex;
621
622         return 0;
623 }
624
625 /*
626  * if <dev> supports MII link status reporting, check its link status.
627  *
628  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
629  * depening upon the setting of the use_carrier parameter.
630  *
631  * Return either BMSR_LSTATUS, meaning that the link is up (or we
632  * can't tell and just pretend it is), or 0, meaning that the link is
633  * down.
634  *
635  * If reporting is non-zero, instead of faking link up, return -1 if
636  * both ETHTOOL and MII ioctls fail (meaning the device does not
637  * support them).  If use_carrier is set, return whatever it says.
638  * It'd be nice if there was a good way to tell if a driver supports
639  * netif_carrier, but there really isn't.
640  */
641 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
642 {
643         static int (* ioctl)(struct net_device *, struct ifreq *, int);
644         struct ifreq ifr;
645         struct mii_ioctl_data *mii;
646         struct ethtool_value etool;
647
648         if (bond->params.use_carrier) {
649                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
650         }
651
652         ioctl = slave_dev->do_ioctl;
653         if (ioctl) {
654                 /* TODO: set pointer to correct ioctl on a per team member */
655                 /*       bases to make this more efficient. that is, once  */
656                 /*       we determine the correct ioctl, we will always    */
657                 /*       call it and not the others for that team          */
658                 /*       member.                                           */
659
660                 /*
661                  * We cannot assume that SIOCGMIIPHY will also read a
662                  * register; not all network drivers (e.g., e100)
663                  * support that.
664                  */
665
666                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
667                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
668                 mii = if_mii(&ifr);
669                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
670                         mii->reg_num = MII_BMSR;
671                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
672                                 return (mii->val_out & BMSR_LSTATUS);
673                         }
674                 }
675         }
676
677         /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
678         /* for a period of time so we attempt to get link status   */
679         /* from it last if the above MII ioctls fail...            */
680         if (slave_dev->ethtool_ops) {
681                 if (slave_dev->ethtool_ops->get_link) {
682                         u32 link;
683
684                         link = slave_dev->ethtool_ops->get_link(slave_dev);
685
686                         return link ? BMSR_LSTATUS : 0;
687                 }
688         }
689
690         if (ioctl) {
691                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
692                 etool.cmd = ETHTOOL_GLINK;
693                 ifr.ifr_data = (char*)&etool;
694                 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
695                         if (etool.data == 1) {
696                                 return BMSR_LSTATUS;
697                         } else {
698                                 dprintk("SIOCETHTOOL shows link down\n");
699                                 return 0;
700                         }
701                 }
702         }
703
704         /*
705          * If reporting, report that either there's no dev->do_ioctl,
706          * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
707          * cannot report link status).  If not reporting, pretend
708          * we're ok.
709          */
710         return (reporting ? -1 : BMSR_LSTATUS);
711 }
712
713 /*----------------------------- Multicast list ------------------------------*/
714
715 /*
716  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
717  */
718 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
719 {
720         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
721                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
722 }
723
724 /*
725  * returns dmi entry if found, NULL otherwise
726  */
727 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
728 {
729         struct dev_mc_list *idmi;
730
731         for (idmi = mc_list; idmi; idmi = idmi->next) {
732                 if (bond_is_dmi_same(dmi, idmi)) {
733                         return idmi;
734                 }
735         }
736
737         return NULL;
738 }
739
740 /*
741  * Push the promiscuity flag down to appropriate slaves
742  */
743 static void bond_set_promiscuity(struct bonding *bond, int inc)
744 {
745         if (USES_PRIMARY(bond->params.mode)) {
746                 /* write lock already acquired */
747                 if (bond->curr_active_slave) {
748                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
749                 }
750         } else {
751                 struct slave *slave;
752                 int i;
753                 bond_for_each_slave(bond, slave, i) {
754                         dev_set_promiscuity(slave->dev, inc);
755                 }
756         }
757 }
758
759 /*
760  * Push the allmulti flag down to all slaves
761  */
762 static void bond_set_allmulti(struct bonding *bond, int inc)
763 {
764         if (USES_PRIMARY(bond->params.mode)) {
765                 /* write lock already acquired */
766                 if (bond->curr_active_slave) {
767                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
768                 }
769         } else {
770                 struct slave *slave;
771                 int i;
772                 bond_for_each_slave(bond, slave, i) {
773                         dev_set_allmulti(slave->dev, inc);
774                 }
775         }
776 }
777
778 /*
779  * Add a Multicast address to slaves
780  * according to mode
781  */
782 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
783 {
784         if (USES_PRIMARY(bond->params.mode)) {
785                 /* write lock already acquired */
786                 if (bond->curr_active_slave) {
787                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
788                 }
789         } else {
790                 struct slave *slave;
791                 int i;
792                 bond_for_each_slave(bond, slave, i) {
793                         dev_mc_add(slave->dev, addr, alen, 0);
794                 }
795         }
796 }
797
798 /*
799  * Remove a multicast address from slave
800  * according to mode
801  */
802 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
803 {
804         if (USES_PRIMARY(bond->params.mode)) {
805                 /* write lock already acquired */
806                 if (bond->curr_active_slave) {
807                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
808                 }
809         } else {
810                 struct slave *slave;
811                 int i;
812                 bond_for_each_slave(bond, slave, i) {
813                         dev_mc_delete(slave->dev, addr, alen, 0);
814                 }
815         }
816 }
817
818 /*
819  * Totally destroys the mc_list in bond
820  */
821 static void bond_mc_list_destroy(struct bonding *bond)
822 {
823         struct dev_mc_list *dmi;
824
825         dmi = bond->mc_list;
826         while (dmi) {
827                 bond->mc_list = dmi->next;
828                 kfree(dmi);
829                 dmi = bond->mc_list;
830         }
831 }
832
833 /*
834  * Copy all the Multicast addresses from src to the bonding device dst
835  */
836 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
837                              gfp_t gfp_flag)
838 {
839         struct dev_mc_list *dmi, *new_dmi;
840
841         for (dmi = mc_list; dmi; dmi = dmi->next) {
842                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
843
844                 if (!new_dmi) {
845                         /* FIXME: Potential memory leak !!! */
846                         return -ENOMEM;
847                 }
848
849                 new_dmi->next = bond->mc_list;
850                 bond->mc_list = new_dmi;
851                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
852                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
853                 new_dmi->dmi_users = dmi->dmi_users;
854                 new_dmi->dmi_gusers = dmi->dmi_gusers;
855         }
856
857         return 0;
858 }
859
860 /*
861  * flush all members of flush->mc_list from device dev->mc_list
862  */
863 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
864 {
865         struct bonding *bond = bond_dev->priv;
866         struct dev_mc_list *dmi;
867
868         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
869                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
870         }
871
872         if (bond->params.mode == BOND_MODE_8023AD) {
873                 /* del lacpdu mc addr from mc list */
874                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
875
876                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
877         }
878 }
879
880 /*--------------------------- Active slave change ---------------------------*/
881
882 /*
883  * Update the mc list and multicast-related flags for the new and
884  * old active slaves (if any) according to the multicast mode, and
885  * promiscuous flags unconditionally.
886  */
887 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
888 {
889         struct dev_mc_list *dmi;
890
891         if (!USES_PRIMARY(bond->params.mode)) {
892                 /* nothing to do -  mc list is already up-to-date on
893                  * all slaves
894                  */
895                 return;
896         }
897
898         if (old_active) {
899                 if (bond->dev->flags & IFF_PROMISC) {
900                         dev_set_promiscuity(old_active->dev, -1);
901                 }
902
903                 if (bond->dev->flags & IFF_ALLMULTI) {
904                         dev_set_allmulti(old_active->dev, -1);
905                 }
906
907                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
908                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
909                 }
910         }
911
912         if (new_active) {
913                 if (bond->dev->flags & IFF_PROMISC) {
914                         dev_set_promiscuity(new_active->dev, 1);
915                 }
916
917                 if (bond->dev->flags & IFF_ALLMULTI) {
918                         dev_set_allmulti(new_active->dev, 1);
919                 }
920
921                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
922                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
923                 }
924         }
925 }
926
927 /**
928  * find_best_interface - select the best available slave to be the active one
929  * @bond: our bonding struct
930  *
931  * Warning: Caller must hold curr_slave_lock for writing.
932  */
933 static struct slave *bond_find_best_slave(struct bonding *bond)
934 {
935         struct slave *new_active, *old_active;
936         struct slave *bestslave = NULL;
937         int mintime = bond->params.updelay;
938         int i;
939
940         new_active = old_active = bond->curr_active_slave;
941
942         if (!new_active) { /* there were no active slaves left */
943                 if (bond->slave_cnt > 0) {  /* found one slave */
944                         new_active = bond->first_slave;
945                 } else {
946                         return NULL; /* still no slave, return NULL */
947                 }
948         }
949
950         /* first try the primary link; if arping, a link must tx/rx traffic
951          * before it can be considered the curr_active_slave - also, we would skip
952          * slaves between the curr_active_slave and primary_slave that may be up
953          * and able to arp
954          */
955         if ((bond->primary_slave) &&
956             (!bond->params.arp_interval) &&
957             (IS_UP(bond->primary_slave->dev))) {
958                 new_active = bond->primary_slave;
959         }
960
961         /* remember where to stop iterating over the slaves */
962         old_active = new_active;
963
964         bond_for_each_slave_from(bond, new_active, i, old_active) {
965                 if (IS_UP(new_active->dev)) {
966                         if (new_active->link == BOND_LINK_UP) {
967                                 return new_active;
968                         } else if (new_active->link == BOND_LINK_BACK) {
969                                 /* link up, but waiting for stabilization */
970                                 if (new_active->delay < mintime) {
971                                         mintime = new_active->delay;
972                                         bestslave = new_active;
973                                 }
974                         }
975                 }
976         }
977
978         return bestslave;
979 }
980
981 /**
982  * change_active_interface - change the active slave into the specified one
983  * @bond: our bonding struct
984  * @new: the new slave to make the active one
985  *
986  * Set the new slave to the bond's settings and unset them on the old
987  * curr_active_slave.
988  * Setting include flags, mc-list, promiscuity, allmulti, etc.
989  *
990  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
991  * because it is apparently the best available slave we have, even though its
992  * updelay hasn't timed out yet.
993  *
994  * Warning: Caller must hold curr_slave_lock for writing.
995  */
996 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
997 {
998         struct slave *old_active = bond->curr_active_slave;
999
1000         if (old_active == new_active) {
1001                 return;
1002         }
1003
1004         if (new_active) {
1005                 if (new_active->link == BOND_LINK_BACK) {
1006                         if (USES_PRIMARY(bond->params.mode)) {
1007                                 printk(KERN_INFO DRV_NAME
1008                                        ": %s: making interface %s the new "
1009                                        "active one %d ms earlier.\n",
1010                                        bond->dev->name, new_active->dev->name,
1011                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1012                         }
1013
1014                         new_active->delay = 0;
1015                         new_active->link = BOND_LINK_UP;
1016                         new_active->jiffies = jiffies;
1017
1018                         if (bond->params.mode == BOND_MODE_8023AD) {
1019                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1020                         }
1021
1022                         if ((bond->params.mode == BOND_MODE_TLB) ||
1023                             (bond->params.mode == BOND_MODE_ALB)) {
1024                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1025                         }
1026                 } else {
1027                         if (USES_PRIMARY(bond->params.mode)) {
1028                                 printk(KERN_INFO DRV_NAME
1029                                        ": %s: making interface %s the new "
1030                                        "active one.\n",
1031                                        bond->dev->name, new_active->dev->name);
1032                         }
1033                 }
1034         }
1035
1036         if (USES_PRIMARY(bond->params.mode)) {
1037                 bond_mc_swap(bond, new_active, old_active);
1038         }
1039
1040         if ((bond->params.mode == BOND_MODE_TLB) ||
1041             (bond->params.mode == BOND_MODE_ALB)) {
1042                 bond_alb_handle_active_change(bond, new_active);
1043         } else {
1044                 bond->curr_active_slave = new_active;
1045         }
1046
1047         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1048                 if (old_active) {
1049                         bond_set_slave_inactive_flags(old_active);
1050                 }
1051
1052                 if (new_active) {
1053                         bond_set_slave_active_flags(new_active);
1054                 }
1055                 bond_send_gratuitous_arp(bond);
1056         }
1057 }
1058
1059 /**
1060  * bond_select_active_slave - select a new active slave, if needed
1061  * @bond: our bonding struct
1062  *
1063  * This functions shoud be called when one of the following occurs:
1064  * - The old curr_active_slave has been released or lost its link.
1065  * - The primary_slave has got its link back.
1066  * - A slave has got its link back and there's no old curr_active_slave.
1067  *
1068  * Warning: Caller must hold curr_slave_lock for writing.
1069  */
1070 void bond_select_active_slave(struct bonding *bond)
1071 {
1072         struct slave *best_slave;
1073
1074         best_slave = bond_find_best_slave(bond);
1075         if (best_slave != bond->curr_active_slave) {
1076                 bond_change_active_slave(bond, best_slave);
1077         }
1078 }
1079
1080 /*--------------------------- slave list handling ---------------------------*/
1081
1082 /*
1083  * This function attaches the slave to the end of list.
1084  *
1085  * bond->lock held for writing by caller.
1086  */
1087 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1088 {
1089         if (bond->first_slave == NULL) { /* attaching the first slave */
1090                 new_slave->next = new_slave;
1091                 new_slave->prev = new_slave;
1092                 bond->first_slave = new_slave;
1093         } else {
1094                 new_slave->next = bond->first_slave;
1095                 new_slave->prev = bond->first_slave->prev;
1096                 new_slave->next->prev = new_slave;
1097                 new_slave->prev->next = new_slave;
1098         }
1099
1100         bond->slave_cnt++;
1101 }
1102
1103 /*
1104  * This function detaches the slave from the list.
1105  * WARNING: no check is made to verify if the slave effectively
1106  * belongs to <bond>.
1107  * Nothing is freed on return, structures are just unchained.
1108  * If any slave pointer in bond was pointing to <slave>,
1109  * it should be changed by the calling function.
1110  *
1111  * bond->lock held for writing by caller.
1112  */
1113 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1114 {
1115         if (slave->next) {
1116                 slave->next->prev = slave->prev;
1117         }
1118
1119         if (slave->prev) {
1120                 slave->prev->next = slave->next;
1121         }
1122
1123         if (bond->first_slave == slave) { /* slave is the first slave */
1124                 if (bond->slave_cnt > 1) { /* there are more slave */
1125                         bond->first_slave = slave->next;
1126                 } else {
1127                         bond->first_slave = NULL; /* slave was the last one */
1128                 }
1129         }
1130
1131         slave->next = NULL;
1132         slave->prev = NULL;
1133         bond->slave_cnt--;
1134 }
1135
1136 /*---------------------------------- IOCTL ----------------------------------*/
1137
1138 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1139 {
1140         dprintk("bond_dev=%p\n", bond_dev);
1141         dprintk("slave_dev=%p\n", slave_dev);
1142         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1143         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1144         return 0;
1145 }
1146
1147 #define BOND_INTERSECT_FEATURES \
1148         (NETIF_F_SG|NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM)
1149
1150 /* 
1151  * Compute the common dev->feature set available to all slaves.  Some
1152  * feature bits are managed elsewhere, so preserve feature bits set on
1153  * master device that are not part of the examined set.
1154  */
1155 static int bond_compute_features(struct bonding *bond)
1156 {
1157         unsigned long features = BOND_INTERSECT_FEATURES;
1158         struct slave *slave;
1159         struct net_device *bond_dev = bond->dev;
1160         int i;
1161
1162         bond_for_each_slave(bond, slave, i)
1163                 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1164
1165         if ((features & NETIF_F_SG) && 
1166             !(features & (NETIF_F_IP_CSUM |
1167                           NETIF_F_NO_CSUM |
1168                           NETIF_F_HW_CSUM)))
1169                 features &= ~NETIF_F_SG;
1170
1171         features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1172         bond_dev->features = features;
1173
1174         return 0;
1175 }
1176
1177 /* enslave device <slave> to bond device <master> */
1178 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1179 {
1180         struct bonding *bond = bond_dev->priv;
1181         struct slave *new_slave = NULL;
1182         struct dev_mc_list *dmi;
1183         struct sockaddr addr;
1184         int link_reporting;
1185         int old_features = bond_dev->features;
1186         int res = 0;
1187
1188         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1189                 slave_dev->do_ioctl == NULL) {
1190                 printk(KERN_WARNING DRV_NAME
1191                        ": %s: Warning: no link monitoring support for %s\n",
1192                        bond_dev->name, slave_dev->name);
1193         }
1194
1195         /* bond must be initialized by bond_open() before enslaving */
1196         if (!(bond_dev->flags & IFF_UP)) {
1197                 dprintk("Error, master_dev is not up\n");
1198                 return -EPERM;
1199         }
1200
1201         /* already enslaved */
1202         if (slave_dev->flags & IFF_SLAVE) {
1203                 dprintk("Error, Device was already enslaved\n");
1204                 return -EBUSY;
1205         }
1206
1207         /* vlan challenged mutual exclusion */
1208         /* no need to lock since we're protected by rtnl_lock */
1209         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1210                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1211                 if (!list_empty(&bond->vlan_list)) {
1212                         printk(KERN_ERR DRV_NAME
1213                                ": %s: Error: cannot enslave VLAN "
1214                                "challenged slave %s on VLAN enabled "
1215                                "bond %s\n", bond_dev->name, slave_dev->name,
1216                                bond_dev->name);
1217                         return -EPERM;
1218                 } else {
1219                         printk(KERN_WARNING DRV_NAME
1220                                ": %s: Warning: enslaved VLAN challenged "
1221                                "slave %s. Adding VLANs will be blocked as "
1222                                "long as %s is part of bond %s\n",
1223                                bond_dev->name, slave_dev->name, slave_dev->name,
1224                                bond_dev->name);
1225                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1226                 }
1227         } else {
1228                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1229                 if (bond->slave_cnt == 0) {
1230                         /* First slave, and it is not VLAN challenged,
1231                          * so remove the block of adding VLANs over the bond.
1232                          */
1233                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1234                 }
1235         }
1236
1237         /*
1238          * Old ifenslave binaries are no longer supported.  These can
1239          * be identified with moderate accurary by the state of the slave:
1240          * the current ifenslave will set the interface down prior to
1241          * enslaving it; the old ifenslave will not.
1242          */
1243         if ((slave_dev->flags & IFF_UP)) {
1244                 printk(KERN_ERR DRV_NAME ": %s is up. "
1245                        "This may be due to an out of date ifenslave.\n",
1246                        slave_dev->name);
1247                 res = -EPERM;
1248                 goto err_undo_flags;
1249         }
1250
1251         if (slave_dev->set_mac_address == NULL) {
1252                 printk(KERN_ERR DRV_NAME
1253                         ": %s: Error: The slave device you specified does "
1254                         "not support setting the MAC address. "
1255                         "Your kernel likely does not support slave "
1256                         "devices.\n", bond_dev->name);
1257                 res = -EOPNOTSUPP;
1258                 goto err_undo_flags;
1259         }
1260
1261         new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1262         if (!new_slave) {
1263                 res = -ENOMEM;
1264                 goto err_undo_flags;
1265         }
1266
1267         memset(new_slave, 0, sizeof(struct slave));
1268
1269         /* save slave's original flags before calling
1270          * netdev_set_master and dev_open
1271          */
1272         new_slave->original_flags = slave_dev->flags;
1273
1274         /*
1275          * Save slave's original ("permanent") mac address for modes
1276          * that need it, and for restoring it upon release, and then
1277          * set it to the master's address
1278          */
1279         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1280
1281         /*
1282          * Set slave to master's mac address.  The application already
1283          * set the master's mac address to that of the first slave
1284          */
1285         memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1286         addr.sa_family = slave_dev->type;
1287         res = dev_set_mac_address(slave_dev, &addr);
1288         if (res) {
1289                 dprintk("Error %d calling set_mac_address\n", res);
1290                 goto err_free;
1291         }
1292
1293         /* open the slave since the application closed it */
1294         res = dev_open(slave_dev);
1295         if (res) {
1296                 dprintk("Openning slave %s failed\n", slave_dev->name);
1297                 goto err_restore_mac;
1298         }
1299
1300         res = netdev_set_master(slave_dev, bond_dev);
1301         if (res) {
1302                 dprintk("Error %d calling netdev_set_master\n", res);
1303                 goto err_close;
1304         }
1305
1306         new_slave->dev = slave_dev;
1307
1308         if ((bond->params.mode == BOND_MODE_TLB) ||
1309             (bond->params.mode == BOND_MODE_ALB)) {
1310                 /* bond_alb_init_slave() must be called before all other stages since
1311                  * it might fail and we do not want to have to undo everything
1312                  */
1313                 res = bond_alb_init_slave(bond, new_slave);
1314                 if (res) {
1315                         goto err_unset_master;
1316                 }
1317         }
1318
1319         /* If the mode USES_PRIMARY, then the new slave gets the
1320          * master's promisc (and mc) settings only if it becomes the
1321          * curr_active_slave, and that is taken care of later when calling
1322          * bond_change_active()
1323          */
1324         if (!USES_PRIMARY(bond->params.mode)) {
1325                 /* set promiscuity level to new slave */
1326                 if (bond_dev->flags & IFF_PROMISC) {
1327                         dev_set_promiscuity(slave_dev, 1);
1328                 }
1329
1330                 /* set allmulti level to new slave */
1331                 if (bond_dev->flags & IFF_ALLMULTI) {
1332                         dev_set_allmulti(slave_dev, 1);
1333                 }
1334
1335                 /* upload master's mc_list to new slave */
1336                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1337                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1338                 }
1339         }
1340
1341         if (bond->params.mode == BOND_MODE_8023AD) {
1342                 /* add lacpdu mc addr to mc list */
1343                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1344
1345                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1346         }
1347
1348         bond_add_vlans_on_slave(bond, slave_dev);
1349
1350         write_lock_bh(&bond->lock);
1351
1352         bond_attach_slave(bond, new_slave);
1353
1354         new_slave->delay = 0;
1355         new_slave->link_failure_count = 0;
1356
1357         bond_compute_features(bond);
1358
1359         if (bond->params.miimon && !bond->params.use_carrier) {
1360                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1361
1362                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1363                         /*
1364                          * miimon is set but a bonded network driver
1365                          * does not support ETHTOOL/MII and
1366                          * arp_interval is not set.  Note: if
1367                          * use_carrier is enabled, we will never go
1368                          * here (because netif_carrier is always
1369                          * supported); thus, we don't need to change
1370                          * the messages for netif_carrier.
1371                          */
1372                         printk(KERN_WARNING DRV_NAME
1373                                ": %s: Warning: MII and ETHTOOL support not "
1374                                "available for interface %s, and "
1375                                "arp_interval/arp_ip_target module parameters "
1376                                "not specified, thus bonding will not detect "
1377                                "link failures! see bonding.txt for details.\n",
1378                                bond_dev->name, slave_dev->name);
1379                 } else if (link_reporting == -1) {
1380                         /* unable get link status using mii/ethtool */
1381                         printk(KERN_WARNING DRV_NAME
1382                                ": %s: Warning: can't get link status from "
1383                                "interface %s; the network driver associated "
1384                                "with this interface does not support MII or "
1385                                "ETHTOOL link status reporting, thus miimon "
1386                                "has no effect on this interface.\n",
1387                                bond_dev->name, slave_dev->name);
1388                 }
1389         }
1390
1391         /* check for initial state */
1392         if (!bond->params.miimon ||
1393             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1394                 if (bond->params.updelay) {
1395                         dprintk("Initial state of slave_dev is "
1396                                 "BOND_LINK_BACK\n");
1397                         new_slave->link  = BOND_LINK_BACK;
1398                         new_slave->delay = bond->params.updelay;
1399                 } else {
1400                         dprintk("Initial state of slave_dev is "
1401                                 "BOND_LINK_UP\n");
1402                         new_slave->link  = BOND_LINK_UP;
1403                 }
1404                 new_slave->jiffies = jiffies;
1405         } else {
1406                 dprintk("Initial state of slave_dev is "
1407                         "BOND_LINK_DOWN\n");
1408                 new_slave->link  = BOND_LINK_DOWN;
1409         }
1410
1411         if (bond_update_speed_duplex(new_slave) &&
1412             (new_slave->link != BOND_LINK_DOWN)) {
1413                 printk(KERN_WARNING DRV_NAME
1414                        ": %s: Warning: failed to get speed and duplex from %s, "
1415                        "assumed to be 100Mb/sec and Full.\n",
1416                        bond_dev->name, new_slave->dev->name);
1417
1418                 if (bond->params.mode == BOND_MODE_8023AD) {
1419                         printk(KERN_WARNING DRV_NAME
1420                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1421                                "support in base driver for proper aggregator "
1422                                "selection.\n", bond_dev->name);
1423                 }
1424         }
1425
1426         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1427                 /* if there is a primary slave, remember it */
1428                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1429                         bond->primary_slave = new_slave;
1430                 }
1431         }
1432
1433         switch (bond->params.mode) {
1434         case BOND_MODE_ACTIVEBACKUP:
1435                 /* if we're in active-backup mode, we need one and only one active
1436                  * interface. The backup interfaces will have their NOARP flag set
1437                  * because we need them to be completely deaf and not to respond to
1438                  * any ARP request on the network to avoid fooling a switch. Thus,
1439                  * since we guarantee that curr_active_slave always point to the last
1440                  * usable interface, we just have to verify this interface's flag.
1441                  */
1442                 if (((!bond->curr_active_slave) ||
1443                      (bond->curr_active_slave->dev->flags & IFF_NOARP)) &&
1444                     (new_slave->link != BOND_LINK_DOWN)) {
1445                         dprintk("This is the first active slave\n");
1446                         /* first slave or no active slave yet, and this link
1447                            is OK, so make this interface the active one */
1448                         bond_change_active_slave(bond, new_slave);
1449                 } else {
1450                         dprintk("This is just a backup slave\n");
1451                         bond_set_slave_inactive_flags(new_slave);
1452                 }
1453                 break;
1454         case BOND_MODE_8023AD:
1455                 /* in 802.3ad mode, the internal mechanism
1456                  * will activate the slaves in the selected
1457                  * aggregator
1458                  */
1459                 bond_set_slave_inactive_flags(new_slave);
1460                 /* if this is the first slave */
1461                 if (bond->slave_cnt == 1) {
1462                         SLAVE_AD_INFO(new_slave).id = 1;
1463                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1464                          * can be called only after the mac address of the bond is set
1465                          */
1466                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1467                                             bond->params.lacp_fast);
1468                 } else {
1469                         SLAVE_AD_INFO(new_slave).id =
1470                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1471                 }
1472
1473                 bond_3ad_bind_slave(new_slave);
1474                 break;
1475         case BOND_MODE_TLB:
1476         case BOND_MODE_ALB:
1477                 new_slave->state = BOND_STATE_ACTIVE;
1478                 if ((!bond->curr_active_slave) &&
1479                     (new_slave->link != BOND_LINK_DOWN)) {
1480                         /* first slave or no active slave yet, and this link
1481                          * is OK, so make this interface the active one
1482                          */
1483                         bond_change_active_slave(bond, new_slave);
1484                 }
1485                 break;
1486         default:
1487                 dprintk("This slave is always active in trunk mode\n");
1488
1489                 /* always active in trunk mode */
1490                 new_slave->state = BOND_STATE_ACTIVE;
1491
1492                 /* In trunking mode there is little meaning to curr_active_slave
1493                  * anyway (it holds no special properties of the bond device),
1494                  * so we can change it without calling change_active_interface()
1495                  */
1496                 if (!bond->curr_active_slave) {
1497                         bond->curr_active_slave = new_slave;
1498                 }
1499                 break;
1500         } /* switch(bond_mode) */
1501
1502         write_unlock_bh(&bond->lock);
1503
1504         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1505         if (res)
1506                 goto err_unset_master;
1507
1508         printk(KERN_INFO DRV_NAME
1509                ": %s: enslaving %s as a%s interface with a%s link.\n",
1510                bond_dev->name, slave_dev->name,
1511                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1512                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1513
1514         /* enslave is successful */
1515         return 0;
1516
1517 /* Undo stages on error */
1518 err_unset_master:
1519         netdev_set_master(slave_dev, NULL);
1520
1521 err_close:
1522         dev_close(slave_dev);
1523
1524 err_restore_mac:
1525         memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1526         addr.sa_family = slave_dev->type;
1527         dev_set_mac_address(slave_dev, &addr);
1528
1529 err_free:
1530         kfree(new_slave);
1531
1532 err_undo_flags:
1533         bond_dev->features = old_features;
1534  
1535         return res;
1536 }
1537
1538 /*
1539  * Try to release the slave device <slave> from the bond device <master>
1540  * It is legal to access curr_active_slave without a lock because all the function
1541  * is write-locked.
1542  *
1543  * The rules for slave state should be:
1544  *   for Active/Backup:
1545  *     Active stays on all backups go down
1546  *   for Bonded connections:
1547  *     The first up interface should be left on and all others downed.
1548  */
1549 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1550 {
1551         struct bonding *bond = bond_dev->priv;
1552         struct slave *slave, *oldcurrent;
1553         struct sockaddr addr;
1554         int mac_addr_differ;
1555
1556         /* slave is not a slave or master is not master of this slave */
1557         if (!(slave_dev->flags & IFF_SLAVE) ||
1558             (slave_dev->master != bond_dev)) {
1559                 printk(KERN_ERR DRV_NAME
1560                        ": %s: Error: cannot release %s.\n",
1561                        bond_dev->name, slave_dev->name);
1562                 return -EINVAL;
1563         }
1564
1565         write_lock_bh(&bond->lock);
1566
1567         slave = bond_get_slave_by_dev(bond, slave_dev);
1568         if (!slave) {
1569                 /* not a slave of this bond */
1570                 printk(KERN_INFO DRV_NAME
1571                        ": %s: %s not enslaved\n",
1572                        bond_dev->name, slave_dev->name);
1573                 return -EINVAL;
1574         }
1575
1576         mac_addr_differ = memcmp(bond_dev->dev_addr,
1577                                  slave->perm_hwaddr,
1578                                  ETH_ALEN);
1579         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1580                 printk(KERN_WARNING DRV_NAME
1581                        ": %s: Warning: the permanent HWaddr of %s "
1582                        "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1583                        "still in use by %s. Set the HWaddr of "
1584                        "%s to a different address to avoid "
1585                        "conflicts.\n",
1586                        bond_dev->name,
1587                        slave_dev->name,
1588                        slave->perm_hwaddr[0],
1589                        slave->perm_hwaddr[1],
1590                        slave->perm_hwaddr[2],
1591                        slave->perm_hwaddr[3],
1592                        slave->perm_hwaddr[4],
1593                        slave->perm_hwaddr[5],
1594                        bond_dev->name,
1595                        slave_dev->name);
1596         }
1597
1598         /* Inform AD package of unbinding of slave. */
1599         if (bond->params.mode == BOND_MODE_8023AD) {
1600                 /* must be called before the slave is
1601                  * detached from the list
1602                  */
1603                 bond_3ad_unbind_slave(slave);
1604         }
1605
1606         printk(KERN_INFO DRV_NAME
1607                ": %s: releasing %s interface %s\n",
1608                bond_dev->name,
1609                (slave->state == BOND_STATE_ACTIVE)
1610                ? "active" : "backup",
1611                slave_dev->name);
1612
1613         oldcurrent = bond->curr_active_slave;
1614
1615         bond->current_arp_slave = NULL;
1616
1617         /* release the slave from its bond */
1618         bond_detach_slave(bond, slave);
1619
1620         bond_compute_features(bond);
1621
1622         if (bond->primary_slave == slave) {
1623                 bond->primary_slave = NULL;
1624         }
1625
1626         if (oldcurrent == slave) {
1627                 bond_change_active_slave(bond, NULL);
1628         }
1629
1630         if ((bond->params.mode == BOND_MODE_TLB) ||
1631             (bond->params.mode == BOND_MODE_ALB)) {
1632                 /* Must be called only after the slave has been
1633                  * detached from the list and the curr_active_slave
1634                  * has been cleared (if our_slave == old_current),
1635                  * but before a new active slave is selected.
1636                  */
1637                 bond_alb_deinit_slave(bond, slave);
1638         }
1639
1640         if (oldcurrent == slave) {
1641                 bond_select_active_slave(bond);
1642
1643                 if (!bond->curr_active_slave) {
1644                         printk(KERN_INFO DRV_NAME
1645                                ": %s: now running without any active "
1646                                "interface !\n",
1647                                bond_dev->name);
1648                 }
1649         }
1650
1651         if (bond->slave_cnt == 0) {
1652                 /* if the last slave was removed, zero the mac address
1653                  * of the master so it will be set by the application
1654                  * to the mac address of the first slave
1655                  */
1656                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1657
1658                 if (list_empty(&bond->vlan_list)) {
1659                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1660                 } else {
1661                         printk(KERN_WARNING DRV_NAME
1662                                ": %s: Warning: clearing HW address of %s while it "
1663                                "still has VLANs.\n",
1664                                bond_dev->name, bond_dev->name);
1665                         printk(KERN_WARNING DRV_NAME
1666                                ": %s: When re-adding slaves, make sure the bond's "
1667                                "HW address matches its VLANs'.\n",
1668                                bond_dev->name);
1669                 }
1670         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1671                    !bond_has_challenged_slaves(bond)) {
1672                 printk(KERN_INFO DRV_NAME
1673                        ": %s: last VLAN challenged slave %s "
1674                        "left bond %s. VLAN blocking is removed\n",
1675                        bond_dev->name, slave_dev->name, bond_dev->name);
1676                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1677         }
1678
1679         write_unlock_bh(&bond->lock);
1680
1681         /* must do this from outside any spinlocks */
1682         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1683
1684         bond_del_vlans_from_slave(bond, slave_dev);
1685
1686         /* If the mode USES_PRIMARY, then we should only remove its
1687          * promisc and mc settings if it was the curr_active_slave, but that was
1688          * already taken care of above when we detached the slave
1689          */
1690         if (!USES_PRIMARY(bond->params.mode)) {
1691                 /* unset promiscuity level from slave */
1692                 if (bond_dev->flags & IFF_PROMISC) {
1693                         dev_set_promiscuity(slave_dev, -1);
1694                 }
1695
1696                 /* unset allmulti level from slave */
1697                 if (bond_dev->flags & IFF_ALLMULTI) {
1698                         dev_set_allmulti(slave_dev, -1);
1699                 }
1700
1701                 /* flush master's mc_list from slave */
1702                 bond_mc_list_flush(bond_dev, slave_dev);
1703         }
1704
1705         netdev_set_master(slave_dev, NULL);
1706
1707         /* close slave before restoring its mac address */
1708         dev_close(slave_dev);
1709
1710         /* restore original ("permanent") mac address */
1711         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1712         addr.sa_family = slave_dev->type;
1713         dev_set_mac_address(slave_dev, &addr);
1714
1715         /* restore the original state of the
1716          * IFF_NOARP flag that might have been
1717          * set by bond_set_slave_inactive_flags()
1718          */
1719         if ((slave->original_flags & IFF_NOARP) == 0) {
1720                 slave_dev->flags &= ~IFF_NOARP;
1721         }
1722
1723         kfree(slave);
1724
1725         return 0;  /* deletion OK */
1726 }
1727
1728 /*
1729  * This function releases all slaves.
1730  */
1731 static int bond_release_all(struct net_device *bond_dev)
1732 {
1733         struct bonding *bond = bond_dev->priv;
1734         struct slave *slave;
1735         struct net_device *slave_dev;
1736         struct sockaddr addr;
1737
1738         write_lock_bh(&bond->lock);
1739
1740         if (bond->slave_cnt == 0) {
1741                 goto out;
1742         }
1743
1744         bond->current_arp_slave = NULL;
1745         bond->primary_slave = NULL;
1746         bond_change_active_slave(bond, NULL);
1747
1748         while ((slave = bond->first_slave) != NULL) {
1749                 /* Inform AD package of unbinding of slave
1750                  * before slave is detached from the list.
1751                  */
1752                 if (bond->params.mode == BOND_MODE_8023AD) {
1753                         bond_3ad_unbind_slave(slave);
1754                 }
1755
1756                 slave_dev = slave->dev;
1757                 bond_detach_slave(bond, slave);
1758
1759                 if ((bond->params.mode == BOND_MODE_TLB) ||
1760                     (bond->params.mode == BOND_MODE_ALB)) {
1761                         /* must be called only after the slave
1762                          * has been detached from the list
1763                          */
1764                         bond_alb_deinit_slave(bond, slave);
1765                 }
1766
1767                 bond_compute_features(bond);
1768
1769                 /* now that the slave is detached, unlock and perform
1770                  * all the undo steps that should not be called from
1771                  * within a lock.
1772                  */
1773                 write_unlock_bh(&bond->lock);
1774
1775                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1776                 bond_del_vlans_from_slave(bond, slave_dev);
1777
1778                 /* If the mode USES_PRIMARY, then we should only remove its
1779                  * promisc and mc settings if it was the curr_active_slave, but that was
1780                  * already taken care of above when we detached the slave
1781                  */
1782                 if (!USES_PRIMARY(bond->params.mode)) {
1783                         /* unset promiscuity level from slave */
1784                         if (bond_dev->flags & IFF_PROMISC) {
1785                                 dev_set_promiscuity(slave_dev, -1);
1786                         }
1787
1788                         /* unset allmulti level from slave */
1789                         if (bond_dev->flags & IFF_ALLMULTI) {
1790                                 dev_set_allmulti(slave_dev, -1);
1791                         }
1792
1793                         /* flush master's mc_list from slave */
1794                         bond_mc_list_flush(bond_dev, slave_dev);
1795                 }
1796
1797                 netdev_set_master(slave_dev, NULL);
1798
1799                 /* close slave before restoring its mac address */
1800                 dev_close(slave_dev);
1801
1802                 /* restore original ("permanent") mac address*/
1803                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1804                 addr.sa_family = slave_dev->type;
1805                 dev_set_mac_address(slave_dev, &addr);
1806
1807                 /* restore the original state of the IFF_NOARP flag that might have
1808                  * been set by bond_set_slave_inactive_flags()
1809                  */
1810                 if ((slave->original_flags & IFF_NOARP) == 0) {
1811                         slave_dev->flags &= ~IFF_NOARP;
1812                 }
1813
1814                 kfree(slave);
1815
1816                 /* re-acquire the lock before getting the next slave */
1817                 write_lock_bh(&bond->lock);
1818         }
1819
1820         /* zero the mac address of the master so it will be
1821          * set by the application to the mac address of the
1822          * first slave
1823          */
1824         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1825
1826         if (list_empty(&bond->vlan_list)) {
1827                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1828         } else {
1829                 printk(KERN_WARNING DRV_NAME
1830                        ": %s: Warning: clearing HW address of %s while it "
1831                        "still has VLANs.\n",
1832                        bond_dev->name, bond_dev->name);
1833                 printk(KERN_WARNING DRV_NAME
1834                        ": %s: When re-adding slaves, make sure the bond's "
1835                        "HW address matches its VLANs'.\n",
1836                        bond_dev->name);
1837         }
1838
1839         printk(KERN_INFO DRV_NAME
1840                ": %s: released all slaves\n",
1841                bond_dev->name);
1842
1843 out:
1844         write_unlock_bh(&bond->lock);
1845
1846         return 0;
1847 }
1848
1849 /*
1850  * This function changes the active slave to slave <slave_dev>.
1851  * It returns -EINVAL in the following cases.
1852  *  - <slave_dev> is not found in the list.
1853  *  - There is not active slave now.
1854  *  - <slave_dev> is already active.
1855  *  - The link state of <slave_dev> is not BOND_LINK_UP.
1856  *  - <slave_dev> is not running.
1857  * In these cases, this fuction does nothing.
1858  * In the other cases, currnt_slave pointer is changed and 0 is returned.
1859  */
1860 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1861 {
1862         struct bonding *bond = bond_dev->priv;
1863         struct slave *old_active = NULL;
1864         struct slave *new_active = NULL;
1865         int res = 0;
1866
1867         if (!USES_PRIMARY(bond->params.mode)) {
1868                 return -EINVAL;
1869         }
1870
1871         /* Verify that master_dev is indeed the master of slave_dev */
1872         if (!(slave_dev->flags & IFF_SLAVE) ||
1873             (slave_dev->master != bond_dev)) {
1874                 return -EINVAL;
1875         }
1876
1877         write_lock_bh(&bond->lock);
1878
1879         old_active = bond->curr_active_slave;
1880         new_active = bond_get_slave_by_dev(bond, slave_dev);
1881
1882         /*
1883          * Changing to the current active: do nothing; return success.
1884          */
1885         if (new_active && (new_active == old_active)) {
1886                 write_unlock_bh(&bond->lock);
1887                 return 0;
1888         }
1889
1890         if ((new_active) &&
1891             (old_active) &&
1892             (new_active->link == BOND_LINK_UP) &&
1893             IS_UP(new_active->dev)) {
1894                 bond_change_active_slave(bond, new_active);
1895         } else {
1896                 res = -EINVAL;
1897         }
1898
1899         write_unlock_bh(&bond->lock);
1900
1901         return res;
1902 }
1903
1904 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1905 {
1906         struct bonding *bond = bond_dev->priv;
1907
1908         info->bond_mode = bond->params.mode;
1909         info->miimon = bond->params.miimon;
1910
1911         read_lock_bh(&bond->lock);
1912         info->num_slaves = bond->slave_cnt;
1913         read_unlock_bh(&bond->lock);
1914
1915         return 0;
1916 }
1917
1918 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1919 {
1920         struct bonding *bond = bond_dev->priv;
1921         struct slave *slave;
1922         int i, found = 0;
1923
1924         if (info->slave_id < 0) {
1925                 return -ENODEV;
1926         }
1927
1928         read_lock_bh(&bond->lock);
1929
1930         bond_for_each_slave(bond, slave, i) {
1931                 if (i == (int)info->slave_id) {
1932                         found = 1;
1933                         break;
1934                 }
1935         }
1936
1937         read_unlock_bh(&bond->lock);
1938
1939         if (found) {
1940                 strcpy(info->slave_name, slave->dev->name);
1941                 info->link = slave->link;
1942                 info->state = slave->state;
1943                 info->link_failure_count = slave->link_failure_count;
1944         } else {
1945                 return -ENODEV;
1946         }
1947
1948         return 0;
1949 }
1950
1951 /*-------------------------------- Monitoring -------------------------------*/
1952
1953 /* this function is called regularly to monitor each slave's link. */
1954 void bond_mii_monitor(struct net_device *bond_dev)
1955 {
1956         struct bonding *bond = bond_dev->priv;
1957         struct slave *slave, *oldcurrent;
1958         int do_failover = 0;
1959         int delta_in_ticks;
1960         int i;
1961
1962         read_lock(&bond->lock);
1963
1964         delta_in_ticks = (bond->params.miimon * HZ) / 1000;
1965
1966         if (bond->kill_timers) {
1967                 goto out;
1968         }
1969
1970         if (bond->slave_cnt == 0) {
1971                 goto re_arm;
1972         }
1973
1974         /* we will try to read the link status of each of our slaves, and
1975          * set their IFF_RUNNING flag appropriately. For each slave not
1976          * supporting MII status, we won't do anything so that a user-space
1977          * program could monitor the link itself if needed.
1978          */
1979
1980         read_lock(&bond->curr_slave_lock);
1981         oldcurrent = bond->curr_active_slave;
1982         read_unlock(&bond->curr_slave_lock);
1983
1984         bond_for_each_slave(bond, slave, i) {
1985                 struct net_device *slave_dev = slave->dev;
1986                 int link_state;
1987                 u16 old_speed = slave->speed;
1988                 u8 old_duplex = slave->duplex;
1989
1990                 link_state = bond_check_dev_link(bond, slave_dev, 0);
1991
1992                 switch (slave->link) {
1993                 case BOND_LINK_UP:      /* the link was up */
1994                         if (link_state == BMSR_LSTATUS) {
1995                                 /* link stays up, nothing more to do */
1996                                 break;
1997                         } else { /* link going down */
1998                                 slave->link  = BOND_LINK_FAIL;
1999                                 slave->delay = bond->params.downdelay;
2000
2001                                 if (slave->link_failure_count < UINT_MAX) {
2002                                         slave->link_failure_count++;
2003                                 }
2004
2005                                 if (bond->params.downdelay) {
2006                                         printk(KERN_INFO DRV_NAME
2007                                                ": %s: link status down for %s "
2008                                                "interface %s, disabling it in "
2009                                                "%d ms.\n",
2010                                                bond_dev->name,
2011                                                IS_UP(slave_dev)
2012                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2013                                                   ? ((slave == oldcurrent)
2014                                                      ? "active " : "backup ")
2015                                                   : "")
2016                                                : "idle ",
2017                                                slave_dev->name,
2018                                                bond->params.downdelay * bond->params.miimon);
2019                                 }
2020                         }
2021                         /* no break ! fall through the BOND_LINK_FAIL test to
2022                            ensure proper action to be taken
2023                         */
2024                 case BOND_LINK_FAIL:    /* the link has just gone down */
2025                         if (link_state != BMSR_LSTATUS) {
2026                                 /* link stays down */
2027                                 if (slave->delay <= 0) {
2028                                         /* link down for too long time */
2029                                         slave->link = BOND_LINK_DOWN;
2030
2031                                         /* in active/backup mode, we must
2032                                          * completely disable this interface
2033                                          */
2034                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2035                                             (bond->params.mode == BOND_MODE_8023AD)) {
2036                                                 bond_set_slave_inactive_flags(slave);
2037                                         }
2038
2039                                         printk(KERN_INFO DRV_NAME
2040                                                ": %s: link status definitely "
2041                                                "down for interface %s, "
2042                                                "disabling it\n",
2043                                                bond_dev->name,
2044                                                slave_dev->name);
2045
2046                                         /* notify ad that the link status has changed */
2047                                         if (bond->params.mode == BOND_MODE_8023AD) {
2048                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2049                                         }
2050
2051                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2052                                             (bond->params.mode == BOND_MODE_ALB)) {
2053                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2054                                         }
2055
2056                                         if (slave == oldcurrent) {
2057                                                 do_failover = 1;
2058                                         }
2059                                 } else {
2060                                         slave->delay--;
2061                                 }
2062                         } else {
2063                                 /* link up again */
2064                                 slave->link  = BOND_LINK_UP;
2065                                 slave->jiffies = jiffies;
2066                                 printk(KERN_INFO DRV_NAME
2067                                        ": %s: link status up again after %d "
2068                                        "ms for interface %s.\n",
2069                                        bond_dev->name,
2070                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2071                                        slave_dev->name);
2072                         }
2073                         break;
2074                 case BOND_LINK_DOWN:    /* the link was down */
2075                         if (link_state != BMSR_LSTATUS) {
2076                                 /* the link stays down, nothing more to do */
2077                                 break;
2078                         } else {        /* link going up */
2079                                 slave->link  = BOND_LINK_BACK;
2080                                 slave->delay = bond->params.updelay;
2081
2082                                 if (bond->params.updelay) {
2083                                         /* if updelay == 0, no need to
2084                                            advertise about a 0 ms delay */
2085                                         printk(KERN_INFO DRV_NAME
2086                                                ": %s: link status up for "
2087                                                "interface %s, enabling it "
2088                                                "in %d ms.\n",
2089                                                bond_dev->name,
2090                                                slave_dev->name,
2091                                                bond->params.updelay * bond->params.miimon);
2092                                 }
2093                         }
2094                         /* no break ! fall through the BOND_LINK_BACK state in
2095                            case there's something to do.
2096                         */
2097                 case BOND_LINK_BACK:    /* the link has just come back */
2098                         if (link_state != BMSR_LSTATUS) {
2099                                 /* link down again */
2100                                 slave->link  = BOND_LINK_DOWN;
2101
2102                                 printk(KERN_INFO DRV_NAME
2103                                        ": %s: link status down again after %d "
2104                                        "ms for interface %s.\n",
2105                                        bond_dev->name,
2106                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2107                                        slave_dev->name);
2108                         } else {
2109                                 /* link stays up */
2110                                 if (slave->delay == 0) {
2111                                         /* now the link has been up for long time enough */
2112                                         slave->link = BOND_LINK_UP;
2113                                         slave->jiffies = jiffies;
2114
2115                                         if (bond->params.mode == BOND_MODE_8023AD) {
2116                                                 /* prevent it from being the active one */
2117                                                 slave->state = BOND_STATE_BACKUP;
2118                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2119                                                 /* make it immediately active */
2120                                                 slave->state = BOND_STATE_ACTIVE;
2121                                         } else if (slave != bond->primary_slave) {
2122                                                 /* prevent it from being the active one */
2123                                                 slave->state = BOND_STATE_BACKUP;
2124                                         }
2125
2126                                         printk(KERN_INFO DRV_NAME
2127                                                ": %s: link status definitely "
2128                                                "up for interface %s.\n",
2129                                                bond_dev->name,
2130                                                slave_dev->name);
2131
2132                                         /* notify ad that the link status has changed */
2133                                         if (bond->params.mode == BOND_MODE_8023AD) {
2134                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2135                                         }
2136
2137                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2138                                             (bond->params.mode == BOND_MODE_ALB)) {
2139                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2140                                         }
2141
2142                                         if ((!oldcurrent) ||
2143                                             (slave == bond->primary_slave)) {
2144                                                 do_failover = 1;
2145                                         }
2146                                 } else {
2147                                         slave->delay--;
2148                                 }
2149                         }
2150                         break;
2151                 default:
2152                         /* Should not happen */
2153                         printk(KERN_ERR DRV_NAME
2154                                ": %s: Error: %s Illegal value (link=%d)\n",
2155                                bond_dev->name,
2156                                slave->dev->name,
2157                                slave->link);
2158                         goto out;
2159                 } /* end of switch (slave->link) */
2160
2161                 bond_update_speed_duplex(slave);
2162
2163                 if (bond->params.mode == BOND_MODE_8023AD) {
2164                         if (old_speed != slave->speed) {
2165                                 bond_3ad_adapter_speed_changed(slave);
2166                         }
2167
2168                         if (old_duplex != slave->duplex) {
2169                                 bond_3ad_adapter_duplex_changed(slave);
2170                         }
2171                 }
2172
2173         } /* end of for */
2174
2175         if (do_failover) {
2176                 write_lock(&bond->curr_slave_lock);
2177
2178                 bond_select_active_slave(bond);
2179
2180                 if (oldcurrent && !bond->curr_active_slave) {
2181                         printk(KERN_INFO DRV_NAME
2182                                ": %s: now running without any active "
2183                                "interface !\n",
2184                                bond_dev->name);
2185                 }
2186
2187                 write_unlock(&bond->curr_slave_lock);
2188         }
2189
2190 re_arm:
2191         if (bond->params.miimon) {
2192                 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2193         }
2194 out:
2195         read_unlock(&bond->lock);
2196 }
2197
2198
2199 static u32 bond_glean_dev_ip(struct net_device *dev)
2200 {
2201         struct in_device *idev;
2202         struct in_ifaddr *ifa;
2203         u32 addr = 0;
2204
2205         if (!dev)
2206                 return 0;
2207
2208         rcu_read_lock();
2209         idev = __in_dev_get_rcu(dev);
2210         if (!idev)
2211                 goto out;
2212
2213         ifa = idev->ifa_list;
2214         if (!ifa)
2215                 goto out;
2216
2217         addr = ifa->ifa_local;
2218 out:
2219         rcu_read_unlock();
2220         return addr;
2221 }
2222
2223 static int bond_has_ip(struct bonding *bond)
2224 {
2225         struct vlan_entry *vlan, *vlan_next;
2226
2227         if (bond->master_ip)
2228                 return 1;
2229
2230         if (list_empty(&bond->vlan_list))
2231                 return 0;
2232
2233         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2234                                  vlan_list) {
2235                 if (vlan->vlan_ip)
2236                         return 1;
2237         }
2238
2239         return 0;
2240 }
2241
2242 /*
2243  * We go to the (large) trouble of VLAN tagging ARP frames because
2244  * switches in VLAN mode (especially if ports are configured as
2245  * "native" to a VLAN) might not pass non-tagged frames.
2246  */
2247 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2248 {
2249         struct sk_buff *skb;
2250
2251         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2252                slave_dev->name, dest_ip, src_ip, vlan_id);
2253                
2254         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2255                          NULL, slave_dev->dev_addr, NULL);
2256
2257         if (!skb) {
2258                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2259                 return;
2260         }
2261         if (vlan_id) {
2262                 skb = vlan_put_tag(skb, vlan_id);
2263                 if (!skb) {
2264                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2265                         return;
2266                 }
2267         }
2268         arp_xmit(skb);
2269 }
2270
2271
2272 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2273 {
2274         int i, vlan_id, rv;
2275         u32 *targets = bond->params.arp_targets;
2276         struct vlan_entry *vlan, *vlan_next;
2277         struct net_device *vlan_dev;
2278         struct flowi fl;
2279         struct rtable *rt;
2280
2281         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2282                 if (!targets[i])
2283                         continue;
2284                 dprintk("basa: target %x\n", targets[i]);
2285                 if (list_empty(&bond->vlan_list)) {
2286                         dprintk("basa: empty vlan: arp_send\n");
2287                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2288                                       bond->master_ip, 0);
2289                         continue;
2290                 }
2291
2292                 /*
2293                  * If VLANs are configured, we do a route lookup to
2294                  * determine which VLAN interface would be used, so we
2295                  * can tag the ARP with the proper VLAN tag.
2296                  */
2297                 memset(&fl, 0, sizeof(fl));
2298                 fl.fl4_dst = targets[i];
2299                 fl.fl4_tos = RTO_ONLINK;
2300
2301                 rv = ip_route_output_key(&rt, &fl);
2302                 if (rv) {
2303                         if (net_ratelimit()) {
2304                                 printk(KERN_WARNING DRV_NAME
2305                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2306                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2307                         }
2308                         continue;
2309                 }
2310
2311                 /*
2312                  * This target is not on a VLAN
2313                  */
2314                 if (rt->u.dst.dev == bond->dev) {
2315                         ip_rt_put(rt);
2316                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2317                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2318                                       bond->master_ip, 0);
2319                         continue;
2320                 }
2321
2322                 vlan_id = 0;
2323                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2324                                          vlan_list) {
2325                         vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2326                         if (vlan_dev == rt->u.dst.dev) {
2327                                 vlan_id = vlan->vlan_id;
2328                                 dprintk("basa: vlan match on %s %d\n",
2329                                        vlan_dev->name, vlan_id);
2330                                 break;
2331                         }
2332                 }
2333
2334                 if (vlan_id) {
2335                         ip_rt_put(rt);
2336                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2337                                       vlan->vlan_ip, vlan_id);
2338                         continue;
2339                 }
2340
2341                 if (net_ratelimit()) {
2342                         printk(KERN_WARNING DRV_NAME
2343                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2344                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2345                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2346                 }
2347                 ip_rt_put(rt);
2348         }
2349 }
2350
2351 /*
2352  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2353  * for each VLAN above us.
2354  */
2355 static void bond_send_gratuitous_arp(struct bonding *bond)
2356 {
2357         struct slave *slave = bond->curr_active_slave;
2358         struct vlan_entry *vlan;
2359         struct net_device *vlan_dev;
2360
2361         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2362                                 slave ? slave->dev->name : "NULL");
2363         if (!slave)
2364                 return;
2365
2366         if (bond->master_ip) {
2367                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2368                                   bond->master_ip, 0);
2369         }
2370
2371         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2372                 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2373                 if (vlan->vlan_ip) {
2374                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2375                                       vlan->vlan_ip, vlan->vlan_id);
2376                 }
2377         }
2378 }
2379
2380 /*
2381  * this function is called regularly to monitor each slave's link
2382  * ensuring that traffic is being sent and received when arp monitoring
2383  * is used in load-balancing mode. if the adapter has been dormant, then an
2384  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2385  * arp monitoring in active backup mode.
2386  */
2387 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2388 {
2389         struct bonding *bond = bond_dev->priv;
2390         struct slave *slave, *oldcurrent;
2391         int do_failover = 0;
2392         int delta_in_ticks;
2393         int i;
2394
2395         read_lock(&bond->lock);
2396
2397         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2398
2399         if (bond->kill_timers) {
2400                 goto out;
2401         }
2402
2403         if (bond->slave_cnt == 0) {
2404                 goto re_arm;
2405         }
2406
2407         read_lock(&bond->curr_slave_lock);
2408         oldcurrent = bond->curr_active_slave;
2409         read_unlock(&bond->curr_slave_lock);
2410
2411         /* see if any of the previous devices are up now (i.e. they have
2412          * xmt and rcv traffic). the curr_active_slave does not come into
2413          * the picture unless it is null. also, slave->jiffies is not needed
2414          * here because we send an arp on each slave and give a slave as
2415          * long as it needs to get the tx/rx within the delta.
2416          * TODO: what about up/down delay in arp mode? it wasn't here before
2417          *       so it can wait
2418          */
2419         bond_for_each_slave(bond, slave, i) {
2420                 if (slave->link != BOND_LINK_UP) {
2421                         if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2422                             ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2423
2424                                 slave->link  = BOND_LINK_UP;
2425                                 slave->state = BOND_STATE_ACTIVE;
2426
2427                                 /* primary_slave has no meaning in round-robin
2428                                  * mode. the window of a slave being up and
2429                                  * curr_active_slave being null after enslaving
2430                                  * is closed.
2431                                  */
2432                                 if (!oldcurrent) {
2433                                         printk(KERN_INFO DRV_NAME
2434                                                ": %s: link status definitely "
2435                                                "up for interface %s, ",
2436                                                bond_dev->name,
2437                                                slave->dev->name);
2438                                         do_failover = 1;
2439                                 } else {
2440                                         printk(KERN_INFO DRV_NAME
2441                                                ": %s: interface %s is now up\n",
2442                                                bond_dev->name,
2443                                                slave->dev->name);
2444                                 }
2445                         }
2446                 } else {
2447                         /* slave->link == BOND_LINK_UP */
2448
2449                         /* not all switches will respond to an arp request
2450                          * when the source ip is 0, so don't take the link down
2451                          * if we don't know our ip yet
2452                          */
2453                         if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2454                             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2455                              bond_has_ip(bond))) {
2456
2457                                 slave->link  = BOND_LINK_DOWN;
2458                                 slave->state = BOND_STATE_BACKUP;
2459
2460                                 if (slave->link_failure_count < UINT_MAX) {
2461                                         slave->link_failure_count++;
2462                                 }
2463
2464                                 printk(KERN_INFO DRV_NAME
2465                                        ": %s: interface %s is now down.\n",
2466                                        bond_dev->name,
2467                                        slave->dev->name);
2468
2469                                 if (slave == oldcurrent) {
2470                                         do_failover = 1;
2471                                 }
2472                         }
2473                 }
2474
2475                 /* note: if switch is in round-robin mode, all links
2476                  * must tx arp to ensure all links rx an arp - otherwise
2477                  * links may oscillate or not come up at all; if switch is
2478                  * in something like xor mode, there is nothing we can
2479                  * do - all replies will be rx'ed on same link causing slaves
2480                  * to be unstable during low/no traffic periods
2481                  */
2482                 if (IS_UP(slave->dev)) {
2483                         bond_arp_send_all(bond, slave);
2484                 }
2485         }
2486
2487         if (do_failover) {
2488                 write_lock(&bond->curr_slave_lock);
2489
2490                 bond_select_active_slave(bond);
2491
2492                 if (oldcurrent && !bond->curr_active_slave) {
2493                         printk(KERN_INFO DRV_NAME
2494                                ": %s: now running without any active "
2495                                "interface !\n",
2496                                bond_dev->name);
2497                 }
2498
2499                 write_unlock(&bond->curr_slave_lock);
2500         }
2501
2502 re_arm:
2503         if (bond->params.arp_interval) {
2504                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2505         }
2506 out:
2507         read_unlock(&bond->lock);
2508 }
2509
2510 /*
2511  * When using arp monitoring in active-backup mode, this function is
2512  * called to determine if any backup slaves have went down or a new
2513  * current slave needs to be found.
2514  * The backup slaves never generate traffic, they are considered up by merely
2515  * receiving traffic. If the current slave goes down, each backup slave will
2516  * be given the opportunity to tx/rx an arp before being taken down - this
2517  * prevents all slaves from being taken down due to the current slave not
2518  * sending any traffic for the backups to receive. The arps are not necessarily
2519  * necessary, any tx and rx traffic will keep the current slave up. While any
2520  * rx traffic will keep the backup slaves up, the current slave is responsible
2521  * for generating traffic to keep them up regardless of any other traffic they
2522  * may have received.
2523  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2524  */
2525 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2526 {
2527         struct bonding *bond = bond_dev->priv;
2528         struct slave *slave;
2529         int delta_in_ticks;
2530         int i;
2531
2532         read_lock(&bond->lock);
2533
2534         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2535
2536         if (bond->kill_timers) {
2537                 goto out;
2538         }
2539
2540         if (bond->slave_cnt == 0) {
2541                 goto re_arm;
2542         }
2543
2544         /* determine if any slave has come up or any backup slave has
2545          * gone down
2546          * TODO: what about up/down delay in arp mode? it wasn't here before
2547          *       so it can wait
2548          */
2549         bond_for_each_slave(bond, slave, i) {
2550                 if (slave->link != BOND_LINK_UP) {
2551                         if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
2552
2553                                 slave->link = BOND_LINK_UP;
2554
2555                                 write_lock(&bond->curr_slave_lock);
2556
2557                                 if ((!bond->curr_active_slave) &&
2558                                     ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2559                                         bond_change_active_slave(bond, slave);
2560                                         bond->current_arp_slave = NULL;
2561                                 } else if (bond->curr_active_slave != slave) {
2562                                         /* this slave has just come up but we
2563                                          * already have a current slave; this
2564                                          * can also happen if bond_enslave adds
2565                                          * a new slave that is up while we are
2566                                          * searching for a new slave
2567                                          */
2568                                         bond_set_slave_inactive_flags(slave);
2569                                         bond->current_arp_slave = NULL;
2570                                 }
2571
2572                                 if (slave == bond->curr_active_slave) {
2573                                         printk(KERN_INFO DRV_NAME
2574                                                ": %s: %s is up and now the "
2575                                                "active interface\n",
2576                                                bond_dev->name,
2577                                                slave->dev->name);
2578                                 } else {
2579                                         printk(KERN_INFO DRV_NAME
2580                                                ": %s: backup interface %s is "
2581                                                "now up\n",
2582                                                bond_dev->name,
2583                                                slave->dev->name);
2584                                 }
2585
2586                                 write_unlock(&bond->curr_slave_lock);
2587                         }
2588                 } else {
2589                         read_lock(&bond->curr_slave_lock);
2590
2591                         if ((slave != bond->curr_active_slave) &&
2592                             (!bond->current_arp_slave) &&
2593                             (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
2594                              bond_has_ip(bond))) {
2595                                 /* a backup slave has gone down; three times
2596                                  * the delta allows the current slave to be
2597                                  * taken out before the backup slave.
2598                                  * note: a non-null current_arp_slave indicates
2599                                  * the curr_active_slave went down and we are
2600                                  * searching for a new one; under this
2601                                  * condition we only take the curr_active_slave
2602                                  * down - this gives each slave a chance to
2603                                  * tx/rx traffic before being taken out
2604                                  */
2605
2606                                 read_unlock(&bond->curr_slave_lock);
2607
2608                                 slave->link  = BOND_LINK_DOWN;
2609
2610                                 if (slave->link_failure_count < UINT_MAX) {
2611                                         slave->link_failure_count++;
2612                                 }
2613
2614                                 bond_set_slave_inactive_flags(slave);
2615
2616                                 printk(KERN_INFO DRV_NAME
2617                                        ": %s: backup interface %s is now down\n",
2618                                        bond_dev->name,
2619                                        slave->dev->name);
2620                         } else {
2621                                 read_unlock(&bond->curr_slave_lock);
2622                         }
2623                 }
2624         }
2625
2626         read_lock(&bond->curr_slave_lock);
2627         slave = bond->curr_active_slave;
2628         read_unlock(&bond->curr_slave_lock);
2629
2630         if (slave) {
2631                 /* if we have sent traffic in the past 2*arp_intervals but
2632                  * haven't xmit and rx traffic in that time interval, select
2633                  * a different slave. slave->jiffies is only updated when
2634                  * a slave first becomes the curr_active_slave - not necessarily
2635                  * after every arp; this ensures the slave has a full 2*delta
2636                  * before being taken out. if a primary is being used, check
2637                  * if it is up and needs to take over as the curr_active_slave
2638                  */
2639                 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2640             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2641              bond_has_ip(bond))) &&
2642                     ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2643
2644                         slave->link  = BOND_LINK_DOWN;
2645
2646                         if (slave->link_failure_count < UINT_MAX) {
2647                                 slave->link_failure_count++;
2648                         }
2649
2650                         printk(KERN_INFO DRV_NAME
2651                                ": %s: link status down for active interface "
2652                                "%s, disabling it\n",
2653                                bond_dev->name,
2654                                slave->dev->name);
2655
2656                         write_lock(&bond->curr_slave_lock);
2657
2658                         bond_select_active_slave(bond);
2659                         slave = bond->curr_active_slave;
2660
2661                         write_unlock(&bond->curr_slave_lock);
2662
2663                         bond->current_arp_slave = slave;
2664
2665                         if (slave) {
2666                                 slave->jiffies = jiffies;
2667                         }
2668                 } else if ((bond->primary_slave) &&
2669                            (bond->primary_slave != slave) &&
2670                            (bond->primary_slave->link == BOND_LINK_UP)) {
2671                         /* at this point, slave is the curr_active_slave */
2672                         printk(KERN_INFO DRV_NAME
2673                                ": %s: changing from interface %s to primary "
2674                                "interface %s\n",
2675                                bond_dev->name,
2676                                slave->dev->name,
2677                                bond->primary_slave->dev->name);
2678
2679                         /* primary is up so switch to it */
2680                         write_lock(&bond->curr_slave_lock);
2681                         bond_change_active_slave(bond, bond->primary_slave);
2682                         write_unlock(&bond->curr_slave_lock);
2683
2684                         slave = bond->primary_slave;
2685                         slave->jiffies = jiffies;
2686                 } else {
2687                         bond->current_arp_slave = NULL;
2688                 }
2689
2690                 /* the current slave must tx an arp to ensure backup slaves
2691                  * rx traffic
2692                  */
2693                 if (slave && bond_has_ip(bond)) {
2694                         bond_arp_send_all(bond, slave);
2695                 }
2696         }
2697
2698         /* if we don't have a curr_active_slave, search for the next available
2699          * backup slave from the current_arp_slave and make it the candidate
2700          * for becoming the curr_active_slave
2701          */
2702         if (!slave) {
2703                 if (!bond->current_arp_slave) {
2704                         bond->current_arp_slave = bond->first_slave;
2705                 }
2706
2707                 if (bond->current_arp_slave) {
2708                         bond_set_slave_inactive_flags(bond->current_arp_slave);
2709
2710                         /* search for next candidate */
2711                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2712                                 if (IS_UP(slave->dev)) {
2713                                         slave->link = BOND_LINK_BACK;
2714                                         bond_set_slave_active_flags(slave);
2715                                         bond_arp_send_all(bond, slave);
2716                                         slave->jiffies = jiffies;
2717                                         bond->current_arp_slave = slave;
2718                                         break;
2719                                 }
2720
2721                                 /* if the link state is up at this point, we
2722                                  * mark it down - this can happen if we have
2723                                  * simultaneous link failures and
2724                                  * reselect_active_interface doesn't make this
2725                                  * one the current slave so it is still marked
2726                                  * up when it is actually down
2727                                  */
2728                                 if (slave->link == BOND_LINK_UP) {
2729                                         slave->link  = BOND_LINK_DOWN;
2730                                         if (slave->link_failure_count < UINT_MAX) {
2731                                                 slave->link_failure_count++;
2732                                         }
2733
2734                                         bond_set_slave_inactive_flags(slave);
2735
2736                                         printk(KERN_INFO DRV_NAME
2737                                                ": %s: backup interface %s is "
2738                                                "now down.\n",
2739                                                bond_dev->name,
2740                                                slave->dev->name);
2741                                 }
2742                         }
2743                 }
2744         }
2745
2746 re_arm:
2747         if (bond->params.arp_interval) {
2748                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2749         }
2750 out:
2751         read_unlock(&bond->lock);
2752 }
2753
2754 /*------------------------------ proc/seq_file-------------------------------*/
2755
2756 #ifdef CONFIG_PROC_FS
2757
2758 #define SEQ_START_TOKEN ((void *)1)
2759
2760 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2761 {
2762         struct bonding *bond = seq->private;
2763         loff_t off = 0;
2764         struct slave *slave;
2765         int i;
2766
2767         /* make sure the bond won't be taken away */
2768         read_lock(&dev_base_lock);
2769         read_lock_bh(&bond->lock);
2770
2771         if (*pos == 0) {
2772                 return SEQ_START_TOKEN;
2773         }
2774
2775         bond_for_each_slave(bond, slave, i) {
2776                 if (++off == *pos) {
2777                         return slave;
2778                 }
2779         }
2780
2781         return NULL;
2782 }
2783
2784 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2785 {
2786         struct bonding *bond = seq->private;
2787         struct slave *slave = v;
2788
2789         ++*pos;
2790         if (v == SEQ_START_TOKEN) {
2791                 return bond->first_slave;
2792         }
2793
2794         slave = slave->next;
2795
2796         return (slave == bond->first_slave) ? NULL : slave;
2797 }
2798
2799 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2800 {
2801         struct bonding *bond = seq->private;
2802
2803         read_unlock_bh(&bond->lock);
2804         read_unlock(&dev_base_lock);
2805 }
2806
2807 static void bond_info_show_master(struct seq_file *seq)
2808 {
2809         struct bonding *bond = seq->private;
2810         struct slave *curr;
2811         int i;
2812         u32 target;
2813
2814         read_lock(&bond->curr_slave_lock);
2815         curr = bond->curr_active_slave;
2816         read_unlock(&bond->curr_slave_lock);
2817
2818         seq_printf(seq, "Bonding Mode: %s\n",
2819                    bond_mode_name(bond->params.mode));
2820
2821         if (bond->params.mode == BOND_MODE_XOR ||
2822                 bond->params.mode == BOND_MODE_8023AD) {
2823                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2824                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2825                         bond->params.xmit_policy);
2826         }
2827
2828         if (USES_PRIMARY(bond->params.mode)) {
2829                 seq_printf(seq, "Primary Slave: %s\n",
2830                            (bond->primary_slave) ?
2831                            bond->primary_slave->dev->name : "None");
2832
2833                 seq_printf(seq, "Currently Active Slave: %s\n",
2834                            (curr) ? curr->dev->name : "None");
2835         }
2836
2837         seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down");
2838         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
2839         seq_printf(seq, "Up Delay (ms): %d\n",
2840                    bond->params.updelay * bond->params.miimon);
2841         seq_printf(seq, "Down Delay (ms): %d\n",
2842                    bond->params.downdelay * bond->params.miimon);
2843
2844
2845         /* ARP information */
2846         if(bond->params.arp_interval > 0) {
2847                 int printed=0;
2848                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
2849                                 bond->params.arp_interval);
2850
2851                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
2852
2853                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
2854                         if (!bond->params.arp_targets[i])
2855                                 continue;
2856                         if (printed)
2857                                 seq_printf(seq, ",");
2858                         target = ntohl(bond->params.arp_targets[i]);
2859                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
2860                         printed = 1;
2861                 }
2862                 seq_printf(seq, "\n");
2863         }
2864
2865         if (bond->params.mode == BOND_MODE_8023AD) {
2866                 struct ad_info ad_info;
2867
2868                 seq_puts(seq, "\n802.3ad info\n");
2869                 seq_printf(seq, "LACP rate: %s\n",
2870                            (bond->params.lacp_fast) ? "fast" : "slow");
2871
2872                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
2873                         seq_printf(seq, "bond %s has no active aggregator\n",
2874                                    bond->dev->name);
2875                 } else {
2876                         seq_printf(seq, "Active Aggregator Info:\n");
2877
2878                         seq_printf(seq, "\tAggregator ID: %d\n",
2879                                    ad_info.aggregator_id);
2880                         seq_printf(seq, "\tNumber of ports: %d\n",
2881                                    ad_info.ports);
2882                         seq_printf(seq, "\tActor Key: %d\n",
2883                                    ad_info.actor_key);
2884                         seq_printf(seq, "\tPartner Key: %d\n",
2885                                    ad_info.partner_key);
2886                         seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2887                                    ad_info.partner_system[0],
2888                                    ad_info.partner_system[1],
2889                                    ad_info.partner_system[2],
2890                                    ad_info.partner_system[3],
2891                                    ad_info.partner_system[4],
2892                                    ad_info.partner_system[5]);
2893                 }
2894         }
2895 }
2896
2897 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
2898 {
2899         struct bonding *bond = seq->private;
2900
2901         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
2902         seq_printf(seq, "MII Status: %s\n",
2903                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
2904         seq_printf(seq, "Link Failure Count: %d\n",
2905                    slave->link_failure_count);
2906
2907         seq_printf(seq,
2908                    "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
2909                    slave->perm_hwaddr[0], slave->perm_hwaddr[1],
2910                    slave->perm_hwaddr[2], slave->perm_hwaddr[3],
2911                    slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
2912
2913         if (bond->params.mode == BOND_MODE_8023AD) {
2914                 const struct aggregator *agg
2915                         = SLAVE_AD_INFO(slave).port.aggregator;
2916
2917                 if (agg) {
2918                         seq_printf(seq, "Aggregator ID: %d\n",
2919                                    agg->aggregator_identifier);
2920                 } else {
2921                         seq_puts(seq, "Aggregator ID: N/A\n");
2922                 }
2923         }
2924 }
2925
2926 static int bond_info_seq_show(struct seq_file *seq, void *v)
2927 {
2928         if (v == SEQ_START_TOKEN) {
2929                 seq_printf(seq, "%s\n", version);
2930                 bond_info_show_master(seq);
2931         } else {
2932                 bond_info_show_slave(seq, v);
2933         }
2934
2935         return 0;
2936 }
2937
2938 static struct seq_operations bond_info_seq_ops = {
2939         .start = bond_info_seq_start,
2940         .next  = bond_info_seq_next,
2941         .stop  = bond_info_seq_stop,
2942         .show  = bond_info_seq_show,
2943 };
2944
2945 static int bond_info_open(struct inode *inode, struct file *file)
2946 {
2947         struct seq_file *seq;
2948         struct proc_dir_entry *proc;
2949         int res;
2950
2951         res = seq_open(file, &bond_info_seq_ops);
2952         if (!res) {
2953                 /* recover the pointer buried in proc_dir_entry data */
2954                 seq = file->private_data;
2955                 proc = PDE(inode);
2956                 seq->private = proc->data;
2957         }
2958
2959         return res;
2960 }
2961
2962 static struct file_operations bond_info_fops = {
2963         .owner   = THIS_MODULE,
2964         .open    = bond_info_open,
2965         .read    = seq_read,
2966         .llseek  = seq_lseek,
2967         .release = seq_release,
2968 };
2969
2970 static int bond_create_proc_entry(struct bonding *bond)
2971 {
2972         struct net_device *bond_dev = bond->dev;
2973
2974         if (bond_proc_dir) {
2975                 bond->proc_entry = create_proc_entry(bond_dev->name,
2976                                                      S_IRUGO,
2977                                                      bond_proc_dir);
2978                 if (bond->proc_entry == NULL) {
2979                         printk(KERN_WARNING DRV_NAME
2980                                ": Warning: Cannot create /proc/net/%s/%s\n",
2981                                DRV_NAME, bond_dev->name);
2982                 } else {
2983                         bond->proc_entry->data = bond;
2984                         bond->proc_entry->proc_fops = &bond_info_fops;
2985                         bond->proc_entry->owner = THIS_MODULE;
2986                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
2987                 }
2988         }
2989
2990         return 0;
2991 }
2992
2993 static void bond_remove_proc_entry(struct bonding *bond)
2994 {
2995         if (bond_proc_dir && bond->proc_entry) {
2996                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
2997                 memset(bond->proc_file_name, 0, IFNAMSIZ);
2998                 bond->proc_entry = NULL;
2999         }
3000 }
3001
3002 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3003  * Caller must hold rtnl_lock.
3004  */
3005 static void bond_create_proc_dir(void)
3006 {
3007         int len = strlen(DRV_NAME);
3008
3009         for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3010              bond_proc_dir = bond_proc_dir->next) {
3011                 if ((bond_proc_dir->namelen == len) &&
3012                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3013                         break;
3014                 }
3015         }
3016
3017         if (!bond_proc_dir) {
3018                 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3019                 if (bond_proc_dir) {
3020                         bond_proc_dir->owner = THIS_MODULE;
3021                 } else {
3022                         printk(KERN_WARNING DRV_NAME
3023                                 ": Warning: cannot create /proc/net/%s\n",
3024                                 DRV_NAME);
3025                 }
3026         }
3027 }
3028
3029 /* Destroy the bonding directory under /proc/net, if empty.
3030  * Caller must hold rtnl_lock.
3031  */
3032 static void bond_destroy_proc_dir(void)
3033 {
3034         struct proc_dir_entry *de;
3035
3036         if (!bond_proc_dir) {
3037                 return;
3038         }
3039
3040         /* verify that the /proc dir is empty */
3041         for (de = bond_proc_dir->subdir; de; de = de->next) {
3042                 /* ignore . and .. */
3043                 if (*(de->name) != '.') {
3044                         break;
3045                 }
3046         }
3047
3048         if (de) {
3049                 if (bond_proc_dir->owner == THIS_MODULE) {
3050                         bond_proc_dir->owner = NULL;
3051                 }
3052         } else {
3053                 remove_proc_entry(DRV_NAME, proc_net);
3054                 bond_proc_dir = NULL;
3055         }
3056 }
3057 #endif /* CONFIG_PROC_FS */
3058
3059 /*-------------------------- netdev event handling --------------------------*/
3060
3061 /*
3062  * Change device name
3063  */
3064 static int bond_event_changename(struct bonding *bond)
3065 {
3066 #ifdef CONFIG_PROC_FS
3067         bond_remove_proc_entry(bond);
3068         bond_create_proc_entry(bond);
3069 #endif
3070         down_write(&(bonding_rwsem));
3071         bond_destroy_sysfs_entry(bond);
3072         bond_create_sysfs_entry(bond);
3073         up_write(&(bonding_rwsem));
3074         return NOTIFY_DONE;
3075 }
3076
3077 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3078 {
3079         struct bonding *event_bond = bond_dev->priv;
3080
3081         switch (event) {
3082         case NETDEV_CHANGENAME:
3083                 return bond_event_changename(event_bond);
3084         case NETDEV_UNREGISTER:
3085                 /*
3086                  * TODO: remove a bond from the list?
3087                  */
3088                 break;
3089         default:
3090                 break;
3091         }
3092
3093         return NOTIFY_DONE;
3094 }
3095
3096 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3097 {
3098         struct net_device *bond_dev = slave_dev->master;
3099         struct bonding *bond = bond_dev->priv;
3100
3101         switch (event) {
3102         case NETDEV_UNREGISTER:
3103                 if (bond_dev) {
3104                         bond_release(bond_dev, slave_dev);
3105                 }
3106                 break;
3107         case NETDEV_CHANGE:
3108                 /*
3109                  * TODO: is this what we get if somebody
3110                  * sets up a hierarchical bond, then rmmod's
3111                  * one of the slave bonding devices?
3112                  */
3113                 break;
3114         case NETDEV_DOWN:
3115                 /*
3116                  * ... Or is it this?
3117                  */
3118                 break;
3119         case NETDEV_CHANGEMTU:
3120                 /*
3121                  * TODO: Should slaves be allowed to
3122                  * independently alter their MTU?  For
3123                  * an active-backup bond, slaves need
3124                  * not be the same type of device, so
3125                  * MTUs may vary.  For other modes,
3126                  * slaves arguably should have the
3127                  * same MTUs. To do this, we'd need to
3128                  * take over the slave's change_mtu
3129                  * function for the duration of their
3130                  * servitude.
3131                  */
3132                 break;
3133         case NETDEV_CHANGENAME:
3134                 /*
3135                  * TODO: handle changing the primary's name
3136                  */
3137                 break;
3138         case NETDEV_FEAT_CHANGE:
3139                 bond_compute_features(bond);
3140                 break;
3141         default:
3142                 break;
3143         }
3144
3145         return NOTIFY_DONE;
3146 }
3147
3148 /*
3149  * bond_netdev_event: handle netdev notifier chain events.
3150  *
3151  * This function receives events for the netdev chain.  The caller (an
3152  * ioctl handler calling notifier_call_chain) holds the necessary
3153  * locks for us to safely manipulate the slave devices (RTNL lock,
3154  * dev_probe_lock).
3155  */
3156 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3157 {
3158         struct net_device *event_dev = (struct net_device *)ptr;
3159
3160         dprintk("event_dev: %s, event: %lx\n",
3161                 (event_dev ? event_dev->name : "None"),
3162                 event);
3163
3164         if (event_dev->flags & IFF_MASTER) {
3165                 dprintk("IFF_MASTER\n");
3166                 return bond_master_netdev_event(event, event_dev);
3167         }
3168
3169         if (event_dev->flags & IFF_SLAVE) {
3170                 dprintk("IFF_SLAVE\n");
3171                 return bond_slave_netdev_event(event, event_dev);
3172         }
3173
3174         return NOTIFY_DONE;
3175 }
3176
3177 /*
3178  * bond_inetaddr_event: handle inetaddr notifier chain events.
3179  *
3180  * We keep track of device IPs primarily to use as source addresses in
3181  * ARP monitor probes (rather than spewing out broadcasts all the time).
3182  *
3183  * We track one IP for the main device (if it has one), plus one per VLAN.
3184  */
3185 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3186 {
3187         struct in_ifaddr *ifa = ptr;
3188         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3189         struct bonding *bond, *bond_next;
3190         struct vlan_entry *vlan, *vlan_next;
3191
3192         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3193                 if (bond->dev == event_dev) {
3194                         switch (event) {
3195                         case NETDEV_UP:
3196                                 bond->master_ip = ifa->ifa_local;
3197                                 return NOTIFY_OK;
3198                         case NETDEV_DOWN:
3199                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3200                                 return NOTIFY_OK;
3201                         default:
3202                                 return NOTIFY_DONE;
3203                         }
3204                 }
3205
3206                 if (list_empty(&bond->vlan_list))
3207                         continue;
3208
3209                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3210                                          vlan_list) {
3211                         vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3212                         if (vlan_dev == event_dev) {
3213                                 switch (event) {
3214                                 case NETDEV_UP:
3215                                         vlan->vlan_ip = ifa->ifa_local;
3216                                         return NOTIFY_OK;
3217                                 case NETDEV_DOWN:
3218                                         vlan->vlan_ip =
3219                                                 bond_glean_dev_ip(vlan_dev);
3220                                         return NOTIFY_OK;
3221                                 default:
3222                                         return NOTIFY_DONE;
3223                                 }
3224                         }
3225                 }
3226         }
3227         return NOTIFY_DONE;
3228 }
3229
3230 static struct notifier_block bond_netdev_notifier = {
3231         .notifier_call = bond_netdev_event,
3232 };
3233
3234 static struct notifier_block bond_inetaddr_notifier = {
3235         .notifier_call = bond_inetaddr_event,
3236 };
3237
3238 /*-------------------------- Packet type handling ---------------------------*/
3239
3240 /* register to receive lacpdus on a bond */
3241 static void bond_register_lacpdu(struct bonding *bond)
3242 {
3243         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3244
3245         /* initialize packet type */
3246         pk_type->type = PKT_TYPE_LACPDU;
3247         pk_type->dev = bond->dev;
3248         pk_type->func = bond_3ad_lacpdu_recv;
3249
3250         dev_add_pack(pk_type);
3251 }
3252
3253 /* unregister to receive lacpdus on a bond */
3254 static void bond_unregister_lacpdu(struct bonding *bond)
3255 {
3256         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3257 }
3258
3259 /*---------------------------- Hashing Policies -----------------------------*/
3260
3261 /*
3262  * Hash for the the output device based upon layer 3 and layer 4 data. If
3263  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3264  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3265  */
3266 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3267                                     struct net_device *bond_dev, int count)
3268 {
3269         struct ethhdr *data = (struct ethhdr *)skb->data;
3270         struct iphdr *iph = skb->nh.iph;
3271         u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3272         int layer4_xor = 0;
3273
3274         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3275                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3276                     (iph->protocol == IPPROTO_TCP ||
3277                      iph->protocol == IPPROTO_UDP)) {
3278                         layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3279                 }
3280                 return (layer4_xor ^
3281                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3282
3283         }
3284
3285         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3286 }
3287
3288 /*
3289  * Hash for the output device based upon layer 2 data
3290  */
3291 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3292                                    struct net_device *bond_dev, int count)
3293 {
3294         struct ethhdr *data = (struct ethhdr *)skb->data;
3295
3296         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3297 }
3298
3299 /*-------------------------- Device entry points ----------------------------*/
3300
3301 static int bond_open(struct net_device *bond_dev)
3302 {
3303         struct bonding *bond = bond_dev->priv;
3304         struct timer_list *mii_timer = &bond->mii_timer;
3305         struct timer_list *arp_timer = &bond->arp_timer;
3306
3307         bond->kill_timers = 0;
3308
3309         if ((bond->params.mode == BOND_MODE_TLB) ||
3310             (bond->params.mode == BOND_MODE_ALB)) {
3311                 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3312
3313                 /* bond_alb_initialize must be called before the timer
3314                  * is started.
3315                  */
3316                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3317                         /* something went wrong - fail the open operation */
3318                         return -1;
3319                 }
3320
3321                 init_timer(alb_timer);
3322                 alb_timer->expires  = jiffies + 1;
3323                 alb_timer->data     = (unsigned long)bond;
3324                 alb_timer->function = (void *)&bond_alb_monitor;
3325                 add_timer(alb_timer);
3326         }
3327
3328         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3329                 init_timer(mii_timer);
3330                 mii_timer->expires  = jiffies + 1;
3331                 mii_timer->data     = (unsigned long)bond_dev;
3332                 mii_timer->function = (void *)&bond_mii_monitor;
3333                 add_timer(mii_timer);
3334         }
3335
3336         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3337                 init_timer(arp_timer);
3338                 arp_timer->expires  = jiffies + 1;
3339                 arp_timer->data     = (unsigned long)bond_dev;
3340                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3341                         arp_timer->function = (void *)&bond_activebackup_arp_mon;
3342                 } else {
3343                         arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3344                 }
3345                 add_timer(arp_timer);
3346         }
3347
3348         if (bond->params.mode == BOND_MODE_8023AD) {
3349                 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3350                 init_timer(ad_timer);
3351                 ad_timer->expires  = jiffies + 1;
3352                 ad_timer->data     = (unsigned long)bond;
3353                 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3354                 add_timer(ad_timer);
3355
3356                 /* register to receive LACPDUs */
3357                 bond_register_lacpdu(bond);
3358         }
3359
3360         return 0;
3361 }
3362
3363 static int bond_close(struct net_device *bond_dev)
3364 {
3365         struct bonding *bond = bond_dev->priv;
3366
3367         if (bond->params.mode == BOND_MODE_8023AD) {
3368                 /* Unregister the receive of LACPDUs */
3369                 bond_unregister_lacpdu(bond);
3370         }
3371
3372         write_lock_bh(&bond->lock);
3373
3374         bond_mc_list_destroy(bond);
3375
3376         /* signal timers not to re-arm */
3377         bond->kill_timers = 1;
3378
3379         write_unlock_bh(&bond->lock);
3380
3381         /* del_timer_sync must run without holding the bond->lock
3382          * because a running timer might be trying to hold it too
3383          */
3384
3385         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3386                 del_timer_sync(&bond->mii_timer);
3387         }
3388
3389         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3390                 del_timer_sync(&bond->arp_timer);
3391         }
3392
3393         switch (bond->params.mode) {
3394         case BOND_MODE_8023AD:
3395                 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3396                 break;
3397         case BOND_MODE_TLB:
3398         case BOND_MODE_ALB:
3399                 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3400                 break;
3401         default:
3402                 break;
3403         }
3404
3405         /* Release the bonded slaves */
3406         bond_release_all(bond_dev);
3407
3408         if ((bond->params.mode == BOND_MODE_TLB) ||
3409             (bond->params.mode == BOND_MODE_ALB)) {
3410                 /* Must be called only after all
3411                  * slaves have been released
3412                  */
3413                 bond_alb_deinitialize(bond);
3414         }
3415
3416         return 0;
3417 }
3418
3419 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3420 {
3421         struct bonding *bond = bond_dev->priv;
3422         struct net_device_stats *stats = &(bond->stats), *sstats;
3423         struct slave *slave;
3424         int i;
3425
3426         memset(stats, 0, sizeof(struct net_device_stats));
3427
3428         read_lock_bh(&bond->lock);
3429
3430         bond_for_each_slave(bond, slave, i) {
3431                 sstats = slave->dev->get_stats(slave->dev);
3432
3433                 stats->rx_packets += sstats->rx_packets;
3434                 stats->rx_bytes += sstats->rx_bytes;
3435                 stats->rx_errors += sstats->rx_errors;
3436                 stats->rx_dropped += sstats->rx_dropped;
3437
3438                 stats->tx_packets += sstats->tx_packets;
3439                 stats->tx_bytes += sstats->tx_bytes;
3440                 stats->tx_errors += sstats->tx_errors;
3441                 stats->tx_dropped += sstats->tx_dropped;
3442
3443                 stats->multicast += sstats->multicast;
3444                 stats->collisions += sstats->collisions;
3445
3446                 stats->rx_length_errors += sstats->rx_length_errors;
3447                 stats->rx_over_errors += sstats->rx_over_errors;
3448                 stats->rx_crc_errors += sstats->rx_crc_errors;
3449                 stats->rx_frame_errors += sstats->rx_frame_errors;
3450                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3451                 stats->rx_missed_errors += sstats->rx_missed_errors;
3452
3453                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3454                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3455                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3456                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3457                 stats->tx_window_errors += sstats->tx_window_errors;
3458         }
3459
3460         read_unlock_bh(&bond->lock);
3461
3462         return stats;
3463 }
3464
3465 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3466 {
3467         struct net_device *slave_dev = NULL;
3468         struct ifbond k_binfo;
3469         struct ifbond __user *u_binfo = NULL;
3470         struct ifslave k_sinfo;
3471         struct ifslave __user *u_sinfo = NULL;
3472         struct mii_ioctl_data *mii = NULL;
3473         int res = 0;
3474
3475         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3476                 bond_dev->name, cmd);
3477
3478         switch (cmd) {
3479         case SIOCGMIIPHY:
3480                 mii = if_mii(ifr);
3481                 if (!mii) {
3482                         return -EINVAL;
3483                 }
3484                 mii->phy_id = 0;
3485                 /* Fall Through */
3486         case SIOCGMIIREG:
3487                 /*
3488                  * We do this again just in case we were called by SIOCGMIIREG
3489                  * instead of SIOCGMIIPHY.
3490                  */
3491                 mii = if_mii(ifr);
3492                 if (!mii) {
3493                         return -EINVAL;
3494                 }
3495
3496                 if (mii->reg_num == 1) {
3497                         struct bonding *bond = bond_dev->priv;
3498                         mii->val_out = 0;
3499                         read_lock_bh(&bond->lock);
3500                         read_lock(&bond->curr_slave_lock);
3501                         if (bond->curr_active_slave) {
3502                                 mii->val_out = BMSR_LSTATUS;
3503                         }
3504                         read_unlock(&bond->curr_slave_lock);
3505                         read_unlock_bh(&bond->lock);
3506                 }
3507
3508                 return 0;
3509         case BOND_INFO_QUERY_OLD:
3510         case SIOCBONDINFOQUERY:
3511                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3512
3513                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3514                         return -EFAULT;
3515                 }
3516
3517                 res = bond_info_query(bond_dev, &k_binfo);
3518                 if (res == 0) {
3519                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3520                                 return -EFAULT;
3521                         }
3522                 }
3523
3524                 return res;
3525         case BOND_SLAVE_INFO_QUERY_OLD:
3526         case SIOCBONDSLAVEINFOQUERY:
3527                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3528
3529                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3530                         return -EFAULT;
3531                 }
3532
3533                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3534                 if (res == 0) {
3535                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3536                                 return -EFAULT;
3537                         }
3538                 }
3539
3540                 return res;
3541         default:
3542                 /* Go on */
3543                 break;
3544         }
3545
3546         if (!capable(CAP_NET_ADMIN)) {
3547                 return -EPERM;
3548         }
3549
3550         down_write(&(bonding_rwsem));
3551         slave_dev = dev_get_by_name(ifr->ifr_slave);
3552
3553         dprintk("slave_dev=%p: \n", slave_dev);
3554
3555         if (!slave_dev) {
3556                 res = -ENODEV;
3557         } else {
3558                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3559                 switch (cmd) {
3560                 case BOND_ENSLAVE_OLD:
3561                 case SIOCBONDENSLAVE:
3562                         res = bond_enslave(bond_dev, slave_dev);
3563                         break;
3564                 case BOND_RELEASE_OLD:
3565                 case SIOCBONDRELEASE:
3566                         res = bond_release(bond_dev, slave_dev);
3567                         break;
3568                 case BOND_SETHWADDR_OLD:
3569                 case SIOCBONDSETHWADDR:
3570                         res = bond_sethwaddr(bond_dev, slave_dev);
3571                         break;
3572                 case BOND_CHANGE_ACTIVE_OLD:
3573                 case SIOCBONDCHANGEACTIVE:
3574                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3575                         break;
3576                 default:
3577                         res = -EOPNOTSUPP;
3578                 }
3579
3580                 dev_put(slave_dev);
3581         }
3582
3583         up_write(&(bonding_rwsem));
3584         return res;
3585 }
3586
3587 static void bond_set_multicast_list(struct net_device *bond_dev)
3588 {
3589         struct bonding *bond = bond_dev->priv;
3590         struct dev_mc_list *dmi;
3591
3592         write_lock_bh(&bond->lock);
3593
3594         /*
3595          * Do promisc before checking multicast_mode
3596          */
3597         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3598                 bond_set_promiscuity(bond, 1);
3599         }
3600
3601         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3602                 bond_set_promiscuity(bond, -1);
3603         }
3604
3605         /* set allmulti flag to slaves */
3606         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3607                 bond_set_allmulti(bond, 1);
3608         }
3609
3610         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3611                 bond_set_allmulti(bond, -1);
3612         }
3613
3614         bond->flags = bond_dev->flags;
3615
3616         /* looking for addresses to add to slaves' mc list */
3617         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3618                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3619                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3620                 }
3621         }
3622
3623         /* looking for addresses to delete from slaves' list */
3624         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3625                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3626                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3627                 }
3628         }
3629
3630         /* save master's multicast list */
3631         bond_mc_list_destroy(bond);
3632         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3633
3634         write_unlock_bh(&bond->lock);
3635 }
3636
3637 /*
3638  * Change the MTU of all of a master's slaves to match the master
3639  */
3640 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3641 {
3642         struct bonding *bond = bond_dev->priv;
3643         struct slave *slave, *stop_at;
3644         int res = 0;
3645         int i;
3646
3647         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3648                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3649
3650         /* Can't hold bond->lock with bh disabled here since
3651          * some base drivers panic. On the other hand we can't
3652          * hold bond->lock without bh disabled because we'll
3653          * deadlock. The only solution is to rely on the fact
3654          * that we're under rtnl_lock here, and the slaves
3655          * list won't change. This doesn't solve the problem
3656          * of setting the slave's MTU while it is
3657          * transmitting, but the assumption is that the base
3658          * driver can handle that.
3659          *
3660          * TODO: figure out a way to safely iterate the slaves
3661          * list, but without holding a lock around the actual
3662          * call to the base driver.
3663          */
3664
3665         bond_for_each_slave(bond, slave, i) {
3666                 dprintk("s %p s->p %p c_m %p\n", slave,
3667                         slave->prev, slave->dev->change_mtu);
3668
3669                 res = dev_set_mtu(slave->dev, new_mtu);
3670
3671                 if (res) {
3672                         /* If we failed to set the slave's mtu to the new value
3673                          * we must abort the operation even in ACTIVE_BACKUP
3674                          * mode, because if we allow the backup slaves to have
3675                          * different mtu values than the active slave we'll
3676                          * need to change their mtu when doing a failover. That
3677                          * means changing their mtu from timer context, which
3678                          * is probably not a good idea.
3679                          */
3680                         dprintk("err %d %s\n", res, slave->dev->name);
3681                         goto unwind;
3682                 }
3683         }
3684
3685         bond_dev->mtu = new_mtu;
3686
3687         return 0;
3688
3689 unwind:
3690         /* unwind from head to the slave that failed */
3691         stop_at = slave;
3692         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3693                 int tmp_res;
3694
3695                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3696                 if (tmp_res) {
3697                         dprintk("unwind err %d dev %s\n", tmp_res,
3698                                 slave->dev->name);
3699                 }
3700         }
3701
3702         return res;
3703 }
3704
3705 /*
3706  * Change HW address
3707  *
3708  * Note that many devices must be down to change the HW address, and
3709  * downing the master releases all slaves.  We can make bonds full of
3710  * bonding devices to test this, however.
3711  */
3712 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3713 {
3714         struct bonding *bond = bond_dev->priv;
3715         struct sockaddr *sa = addr, tmp_sa;
3716         struct slave *slave, *stop_at;
3717         int res = 0;
3718         int i;
3719
3720         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3721
3722         if (!is_valid_ether_addr(sa->sa_data)) {
3723                 return -EADDRNOTAVAIL;
3724         }
3725
3726         /* Can't hold bond->lock with bh disabled here since
3727          * some base drivers panic. On the other hand we can't
3728          * hold bond->lock without bh disabled because we'll
3729          * deadlock. The only solution is to rely on the fact
3730          * that we're under rtnl_lock here, and the slaves
3731          * list won't change. This doesn't solve the problem
3732          * of setting the slave's hw address while it is
3733          * transmitting, but the assumption is that the base
3734          * driver can handle that.
3735          *
3736          * TODO: figure out a way to safely iterate the slaves
3737          * list, but without holding a lock around the actual
3738          * call to the base driver.
3739          */
3740
3741         bond_for_each_slave(bond, slave, i) {
3742                 dprintk("slave %p %s\n", slave, slave->dev->name);
3743
3744                 if (slave->dev->set_mac_address == NULL) {
3745                         res = -EOPNOTSUPP;
3746                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3747                         goto unwind;
3748                 }
3749
3750                 res = dev_set_mac_address(slave->dev, addr);
3751                 if (res) {
3752                         /* TODO: consider downing the slave
3753                          * and retry ?
3754                          * User should expect communications
3755                          * breakage anyway until ARP finish
3756                          * updating, so...
3757                          */
3758                         dprintk("err %d %s\n", res, slave->dev->name);
3759                         goto unwind;
3760                 }
3761         }
3762
3763         /* success */
3764         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3765         return 0;
3766
3767 unwind:
3768         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3769         tmp_sa.sa_family = bond_dev->type;
3770
3771         /* unwind from head to the slave that failed */
3772         stop_at = slave;
3773         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3774                 int tmp_res;
3775
3776                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3777                 if (tmp_res) {
3778                         dprintk("unwind err %d dev %s\n", tmp_res,
3779                                 slave->dev->name);
3780                 }
3781         }
3782
3783         return res;
3784 }
3785
3786 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3787 {
3788         struct bonding *bond = bond_dev->priv;
3789         struct slave *slave, *start_at;
3790         int i;
3791         int res = 1;
3792
3793         read_lock(&bond->lock);
3794
3795         if (!BOND_IS_OK(bond)) {
3796                 goto out;
3797         }
3798
3799         read_lock(&bond->curr_slave_lock);
3800         slave = start_at = bond->curr_active_slave;
3801         read_unlock(&bond->curr_slave_lock);
3802
3803         if (!slave) {
3804                 goto out;
3805         }
3806
3807         bond_for_each_slave_from(bond, slave, i, start_at) {
3808                 if (IS_UP(slave->dev) &&
3809                     (slave->link == BOND_LINK_UP) &&
3810                     (slave->state == BOND_STATE_ACTIVE)) {
3811                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3812
3813                         write_lock(&bond->curr_slave_lock);
3814                         bond->curr_active_slave = slave->next;
3815                         write_unlock(&bond->curr_slave_lock);
3816
3817                         break;
3818                 }
3819         }
3820
3821
3822 out:
3823         if (res) {
3824                 /* no suitable interface, frame not sent */
3825                 dev_kfree_skb(skb);
3826         }
3827         read_unlock(&bond->lock);
3828         return 0;
3829 }
3830
3831 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
3832                                         struct bonding *bond,
3833                                         struct slave *slave)
3834 {
3835         struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
3836         struct ethhdr *eth_data;
3837         u8 *hwaddr;
3838         int res;
3839
3840         if (!skb2) {
3841                 printk(KERN_ERR DRV_NAME ": Error: "
3842                        "bond_activebackup_xmit_copy(): skb_copy() failed\n");
3843                 return;
3844         }
3845
3846         skb2->mac.raw = (unsigned char *)skb2->data;
3847         eth_data = eth_hdr(skb2);
3848
3849         /* Pick an appropriate source MAC address
3850          *      -- use slave's perm MAC addr, unless used by bond
3851          *      -- otherwise, borrow active slave's perm MAC addr
3852          *         since that will not be used
3853          */
3854         hwaddr = slave->perm_hwaddr;
3855         if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
3856                 hwaddr = bond->curr_active_slave->perm_hwaddr;
3857
3858         /* Set source MAC address appropriately */
3859         memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
3860
3861         res = bond_dev_queue_xmit(bond, skb2, slave->dev);
3862         if (res)
3863                 dev_kfree_skb(skb2);
3864
3865         return;
3866 }
3867
3868 /*
3869  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3870  * the bond has a usable interface.
3871  */
3872 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3873 {
3874         struct bonding *bond = bond_dev->priv;
3875         int res = 1;
3876
3877         read_lock(&bond->lock);
3878         read_lock(&bond->curr_slave_lock);
3879
3880         if (!BOND_IS_OK(bond)) {
3881                 goto out;
3882         }
3883
3884         if (!bond->curr_active_slave)
3885                 goto out;
3886
3887         /* Xmit IGMP frames on all slaves to ensure rapid fail-over
3888            for multicast traffic on snooping switches */
3889         if (skb->protocol == __constant_htons(ETH_P_IP) &&
3890             skb->nh.iph->protocol == IPPROTO_IGMP) {
3891                 struct slave *slave, *active_slave;
3892                 int i;
3893
3894                 active_slave = bond->curr_active_slave;
3895                 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
3896                                             active_slave->prev)
3897                         if (IS_UP(slave->dev) &&
3898                             (slave->link == BOND_LINK_UP))
3899                                 bond_activebackup_xmit_copy(skb, bond, slave);
3900         }
3901
3902         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
3903
3904 out:
3905         if (res) {
3906                 /* no suitable interface, frame not sent */
3907                 dev_kfree_skb(skb);
3908         }
3909         read_unlock(&bond->curr_slave_lock);
3910         read_unlock(&bond->lock);
3911         return 0;
3912 }
3913
3914 /*
3915  * In bond_xmit_xor() , we determine the output device by using a pre-
3916  * determined xmit_hash_policy(), If the selected device is not enabled,
3917  * find the next active slave.
3918  */
3919 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3920 {
3921         struct bonding *bond = bond_dev->priv;
3922         struct slave *slave, *start_at;
3923         int slave_no;
3924         int i;
3925         int res = 1;
3926
3927         read_lock(&bond->lock);
3928
3929         if (!BOND_IS_OK(bond)) {
3930                 goto out;
3931         }
3932
3933         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
3934
3935         bond_for_each_slave(bond, slave, i) {
3936                 slave_no--;
3937                 if (slave_no < 0) {
3938                         break;
3939                 }
3940         }
3941
3942         start_at = slave;
3943
3944         bond_for_each_slave_from(bond, slave, i, start_at) {
3945                 if (IS_UP(slave->dev) &&
3946                     (slave->link == BOND_LINK_UP) &&
3947                     (slave->state == BOND_STATE_ACTIVE)) {
3948                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3949                         break;
3950                 }
3951         }
3952
3953 out:
3954         if (res) {
3955                 /* no suitable interface, frame not sent */
3956                 dev_kfree_skb(skb);
3957         }
3958         read_unlock(&bond->lock);
3959         return 0;
3960 }
3961
3962 /*
3963  * in broadcast mode, we send everything to all usable interfaces.
3964  */
3965 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3966 {
3967         struct bonding *bond = bond_dev->priv;
3968         struct slave *slave, *start_at;
3969         struct net_device *tx_dev = NULL;
3970         int i;
3971         int res = 1;
3972
3973         read_lock(&bond->lock);
3974
3975         if (!BOND_IS_OK(bond)) {
3976                 goto out;
3977         }
3978
3979         read_lock(&bond->curr_slave_lock);
3980         start_at = bond->curr_active_slave;
3981         read_unlock(&bond->curr_slave_lock);
3982
3983         if (!start_at) {
3984                 goto out;
3985         }
3986
3987         bond_for_each_slave_from(bond, slave, i, start_at) {
3988                 if (IS_UP(slave->dev) &&
3989                     (slave->link == BOND_LINK_UP) &&
3990                     (slave->state == BOND_STATE_ACTIVE)) {
3991                         if (tx_dev) {
3992                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3993                                 if (!skb2) {
3994                                         printk(KERN_ERR DRV_NAME
3995                                                ": %s: Error: bond_xmit_broadcast(): "
3996                                                "skb_clone() failed\n",
3997                                                bond_dev->name);
3998                                         continue;
3999                                 }
4000
4001                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4002                                 if (res) {
4003                                         dev_kfree_skb(skb2);
4004                                         continue;
4005                                 }
4006                         }
4007                         tx_dev = slave->dev;
4008                 }
4009         }
4010
4011         if (tx_dev) {
4012                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4013         }
4014
4015 out:
4016         if (res) {
4017                 /* no suitable interface, frame not sent */
4018                 dev_kfree_skb(skb);
4019         }
4020         /* frame sent to all suitable interfaces */
4021         read_unlock(&bond->lock);
4022         return 0;
4023 }
4024
4025 /*------------------------- Device initialization ---------------------------*/
4026
4027 /*
4028  * set bond mode specific net device operations
4029  */
4030 void bond_set_mode_ops(struct bonding *bond, int mode)
4031 {
4032         struct net_device *bond_dev = bond->dev;
4033
4034         switch (mode) {
4035         case BOND_MODE_ROUNDROBIN:
4036                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4037                 break;
4038         case BOND_MODE_ACTIVEBACKUP:
4039                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4040                 break;
4041         case BOND_MODE_XOR:
4042                 bond_dev->hard_start_xmit = bond_xmit_xor;
4043                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4044                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4045                 else
4046                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4047                 break;
4048         case BOND_MODE_BROADCAST:
4049                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4050                 break;
4051         case BOND_MODE_8023AD:
4052                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4053                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4054                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4055                 else
4056                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4057                 break;
4058         case BOND_MODE_TLB:
4059         case BOND_MODE_ALB:
4060                 bond_dev->hard_start_xmit = bond_alb_xmit;
4061                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4062                 break;
4063         default:
4064                 /* Should never happen, mode already checked */
4065                 printk(KERN_ERR DRV_NAME
4066                        ": %s: Error: Unknown bonding mode %d\n",
4067                        bond_dev->name,
4068                        mode);
4069                 break;
4070         }
4071 }
4072
4073 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4074                                     struct ethtool_drvinfo *drvinfo)
4075 {
4076         strncpy(drvinfo->driver, DRV_NAME, 32);
4077         strncpy(drvinfo->version, DRV_VERSION, 32);
4078         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4079 }
4080
4081 static struct ethtool_ops bond_ethtool_ops = {
4082         .get_tx_csum            = ethtool_op_get_tx_csum,
4083         .get_sg                 = ethtool_op_get_sg,
4084         .get_drvinfo            = bond_ethtool_get_drvinfo,
4085 };
4086
4087 /*
4088  * Does not allocate but creates a /proc entry.
4089  * Allowed to fail.
4090  */
4091 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4092 {
4093         struct bonding *bond = bond_dev->priv;
4094
4095         dprintk("Begin bond_init for %s\n", bond_dev->name);
4096
4097         /* initialize rwlocks */
4098         rwlock_init(&bond->lock);
4099         rwlock_init(&bond->curr_slave_lock);
4100
4101         bond->params = *params; /* copy params struct */
4102
4103         /* Initialize pointers */
4104         bond->first_slave = NULL;
4105         bond->curr_active_slave = NULL;
4106         bond->current_arp_slave = NULL;
4107         bond->primary_slave = NULL;
4108         bond->dev = bond_dev;
4109         INIT_LIST_HEAD(&bond->vlan_list);
4110
4111         /* Initialize the device entry points */
4112         bond_dev->open = bond_open;
4113         bond_dev->stop = bond_close;
4114         bond_dev->get_stats = bond_get_stats;
4115         bond_dev->do_ioctl = bond_do_ioctl;
4116         bond_dev->ethtool_ops = &bond_ethtool_ops;
4117         bond_dev->set_multicast_list = bond_set_multicast_list;
4118         bond_dev->change_mtu = bond_change_mtu;
4119         bond_dev->set_mac_address = bond_set_mac_address;
4120
4121         bond_set_mode_ops(bond, bond->params.mode);
4122
4123         bond_dev->destructor = free_netdev;
4124
4125         /* Initialize the device options */
4126         bond_dev->tx_queue_len = 0;
4127         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4128
4129         /* At first, we block adding VLANs. That's the only way to
4130          * prevent problems that occur when adding VLANs over an
4131          * empty bond. The block will be removed once non-challenged
4132          * slaves are enslaved.
4133          */
4134         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4135
4136         /* don't acquire bond device's xmit_lock when 
4137          * transmitting */
4138         bond_dev->features |= NETIF_F_LLTX;
4139
4140         /* By default, we declare the bond to be fully
4141          * VLAN hardware accelerated capable. Special
4142          * care is taken in the various xmit functions
4143          * when there are slaves that are not hw accel
4144          * capable
4145          */
4146         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4147         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4148         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4149         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4150                                NETIF_F_HW_VLAN_RX |
4151                                NETIF_F_HW_VLAN_FILTER);
4152
4153 #ifdef CONFIG_PROC_FS
4154         bond_create_proc_entry(bond);
4155 #endif
4156
4157         list_add_tail(&bond->bond_list, &bond_dev_list);
4158
4159         return 0;
4160 }
4161
4162 /* De-initialize device specific data.
4163  * Caller must hold rtnl_lock.
4164  */
4165 void bond_deinit(struct net_device *bond_dev)
4166 {
4167         struct bonding *bond = bond_dev->priv;
4168
4169         list_del(&bond->bond_list);
4170
4171 #ifdef CONFIG_PROC_FS
4172         bond_remove_proc_entry(bond);
4173 #endif
4174 }
4175
4176 /* Unregister and free all bond devices.
4177  * Caller must hold rtnl_lock.
4178  */
4179 static void bond_free_all(void)
4180 {
4181         struct bonding *bond, *nxt;
4182
4183         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4184                 struct net_device *bond_dev = bond->dev;
4185
4186                 unregister_netdevice(bond_dev);
4187                 bond_deinit(bond_dev);
4188         }
4189
4190 #ifdef CONFIG_PROC_FS
4191         bond_destroy_proc_dir();
4192 #endif
4193 }
4194
4195 /*------------------------- Module initialization ---------------------------*/
4196
4197 /*
4198  * Convert string input module parms.  Accept either the
4199  * number of the mode or its string name.
4200  */
4201 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4202 {
4203         int i;
4204
4205         for (i = 0; tbl[i].modename; i++) {
4206                 if ((isdigit(*mode_arg) &&
4207                      tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4208                     (strncmp(mode_arg, tbl[i].modename,
4209                              strlen(tbl[i].modename)) == 0)) {
4210                         return tbl[i].mode;
4211                 }
4212         }
4213
4214         return -1;
4215 }
4216
4217 static int bond_check_params(struct bond_params *params)
4218 {
4219         /*
4220          * Convert string parameters.
4221          */
4222         if (mode) {
4223                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4224                 if (bond_mode == -1) {
4225                         printk(KERN_ERR DRV_NAME
4226                                ": Error: Invalid bonding mode \"%s\"\n",
4227                                mode == NULL ? "NULL" : mode);
4228                         return -EINVAL;
4229                 }
4230         }
4231
4232         if (xmit_hash_policy) {
4233                 if ((bond_mode != BOND_MODE_XOR) &&
4234                     (bond_mode != BOND_MODE_8023AD)) {
4235                         printk(KERN_INFO DRV_NAME
4236                                ": xor_mode param is irrelevant in mode %s\n",
4237                                bond_mode_name(bond_mode));
4238                 } else {
4239                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4240                                                         xmit_hashtype_tbl);
4241                         if (xmit_hashtype == -1) {
4242                                 printk(KERN_ERR DRV_NAME
4243                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4244                                 xmit_hash_policy == NULL ? "NULL" :
4245                                        xmit_hash_policy);
4246                                 return -EINVAL;
4247                         }
4248                 }
4249         }
4250
4251         if (lacp_rate) {
4252                 if (bond_mode != BOND_MODE_8023AD) {
4253                         printk(KERN_INFO DRV_NAME
4254                                ": lacp_rate param is irrelevant in mode %s\n",
4255                                bond_mode_name(bond_mode));
4256                 } else {
4257                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4258                         if (lacp_fast == -1) {
4259                                 printk(KERN_ERR DRV_NAME
4260                                        ": Error: Invalid lacp rate \"%s\"\n",
4261                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4262                                 return -EINVAL;
4263                         }
4264                 }
4265         }
4266
4267         if (max_bonds < 1 || max_bonds > INT_MAX) {
4268                 printk(KERN_WARNING DRV_NAME
4269                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4270                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4271                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4272                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4273         }
4274
4275         if (miimon < 0) {
4276                 printk(KERN_WARNING DRV_NAME
4277                        ": Warning: miimon module parameter (%d), "
4278                        "not in range 0-%d, so it was reset to %d\n",
4279                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4280                 miimon = BOND_LINK_MON_INTERV;
4281         }
4282
4283         if (updelay < 0) {
4284                 printk(KERN_WARNING DRV_NAME
4285                        ": Warning: updelay module parameter (%d), "
4286                        "not in range 0-%d, so it was reset to 0\n",
4287                        updelay, INT_MAX);
4288                 updelay = 0;
4289         }
4290
4291         if (downdelay < 0) {
4292                 printk(KERN_WARNING DRV_NAME
4293                        ": Warning: downdelay module parameter (%d), "
4294                        "not in range 0-%d, so it was reset to 0\n",
4295                        downdelay, INT_MAX);
4296                 downdelay = 0;
4297         }
4298
4299         if ((use_carrier != 0) && (use_carrier != 1)) {
4300                 printk(KERN_WARNING DRV_NAME
4301                        ": Warning: use_carrier module parameter (%d), "
4302                        "not of valid value (0/1), so it was set to 1\n",
4303                        use_carrier);
4304                 use_carrier = 1;
4305         }
4306
4307         /* reset values for 802.3ad */
4308         if (bond_mode == BOND_MODE_8023AD) {
4309                 if (!miimon) {
4310                         printk(KERN_WARNING DRV_NAME
4311                                ": Warning: miimon must be specified, "
4312                                "otherwise bonding will not detect link "
4313                                "failure, speed and duplex which are "
4314                                "essential for 802.3ad operation\n");
4315                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4316                         miimon = 100;
4317                 }
4318         }
4319
4320         /* reset values for TLB/ALB */
4321         if ((bond_mode == BOND_MODE_TLB) ||
4322             (bond_mode == BOND_MODE_ALB)) {
4323                 if (!miimon) {
4324                         printk(KERN_WARNING DRV_NAME
4325                                ": Warning: miimon must be specified, "
4326                                "otherwise bonding will not detect link "
4327                                "failure and link speed which are essential "
4328                                "for TLB/ALB load balancing\n");
4329                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4330                         miimon = 100;
4331                 }
4332         }
4333
4334         if (bond_mode == BOND_MODE_ALB) {
4335                 printk(KERN_NOTICE DRV_NAME
4336                        ": In ALB mode you might experience client "
4337                        "disconnections upon reconnection of a link if the "
4338                        "bonding module updelay parameter (%d msec) is "
4339                        "incompatible with the forwarding delay time of the "
4340                        "switch\n",
4341                        updelay);
4342         }
4343
4344         if (!miimon) {
4345                 if (updelay || downdelay) {
4346                         /* just warn the user the up/down delay will have
4347                          * no effect since miimon is zero...
4348                          */
4349                         printk(KERN_WARNING DRV_NAME
4350                                ": Warning: miimon module parameter not set "
4351                                "and updelay (%d) or downdelay (%d) module "
4352                                "parameter is set; updelay and downdelay have "
4353                                "no effect unless miimon is set\n",
4354                                updelay, downdelay);
4355                 }
4356         } else {
4357                 /* don't allow arp monitoring */
4358                 if (arp_interval) {
4359                         printk(KERN_WARNING DRV_NAME
4360                                ": Warning: miimon (%d) and arp_interval (%d) "
4361                                "can't be used simultaneously, disabling ARP "
4362                                "monitoring\n",
4363                                miimon, arp_interval);
4364                         arp_interval = 0;
4365                 }
4366
4367                 if ((updelay % miimon) != 0) {
4368                         printk(KERN_WARNING DRV_NAME
4369                                ": Warning: updelay (%d) is not a multiple "
4370                                "of miimon (%d), updelay rounded to %d ms\n",
4371                                updelay, miimon, (updelay / miimon) * miimon);
4372                 }
4373
4374                 updelay /= miimon;
4375
4376                 if ((downdelay % miimon) != 0) {
4377                         printk(KERN_WARNING DRV_NAME
4378                                ": Warning: downdelay (%d) is not a multiple "
4379                                "of miimon (%d), downdelay rounded to %d ms\n",
4380                                downdelay, miimon,
4381                                (downdelay / miimon) * miimon);
4382                 }
4383
4384                 downdelay /= miimon;
4385         }
4386
4387         if (arp_interval < 0) {
4388                 printk(KERN_WARNING DRV_NAME
4389                        ": Warning: arp_interval module parameter (%d) "
4390                        ", not in range 0-%d, so it was reset to %d\n",
4391                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4392                 arp_interval = BOND_LINK_ARP_INTERV;
4393         }
4394
4395         for (arp_ip_count = 0;
4396              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4397              arp_ip_count++) {
4398                 /* not complete check, but should be good enough to
4399                    catch mistakes */
4400                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4401                         printk(KERN_WARNING DRV_NAME
4402                                ": Warning: bad arp_ip_target module parameter "
4403                                "(%s), ARP monitoring will not be performed\n",
4404                                arp_ip_target[arp_ip_count]);
4405                         arp_interval = 0;
4406                 } else {
4407                         u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4408                         arp_target[arp_ip_count] = ip;
4409                 }
4410         }
4411
4412         if (arp_interval && !arp_ip_count) {
4413                 /* don't allow arping if no arp_ip_target given... */
4414                 printk(KERN_WARNING DRV_NAME
4415                        ": Warning: arp_interval module parameter (%d) "
4416                        "specified without providing an arp_ip_target "
4417                        "parameter, arp_interval was reset to 0\n",
4418                        arp_interval);
4419                 arp_interval = 0;
4420         }
4421
4422         if (miimon) {
4423                 printk(KERN_INFO DRV_NAME
4424                        ": MII link monitoring set to %d ms\n",
4425                        miimon);
4426         } else if (arp_interval) {
4427                 int i;
4428
4429                 printk(KERN_INFO DRV_NAME
4430                        ": ARP monitoring set to %d ms with %d target(s):",
4431                        arp_interval, arp_ip_count);
4432
4433                 for (i = 0; i < arp_ip_count; i++)
4434                         printk (" %s", arp_ip_target[i]);
4435
4436                 printk("\n");
4437
4438         } else {
4439                 /* miimon and arp_interval not set, we need one so things
4440                  * work as expected, see bonding.txt for details
4441                  */
4442                 printk(KERN_WARNING DRV_NAME
4443                        ": Warning: either miimon or arp_interval and "
4444                        "arp_ip_target module parameters must be specified, "
4445                        "otherwise bonding will not detect link failures! see "
4446                        "bonding.txt for details.\n");
4447         }
4448
4449         if (primary && !USES_PRIMARY(bond_mode)) {
4450                 /* currently, using a primary only makes sense
4451                  * in active backup, TLB or ALB modes
4452                  */
4453                 printk(KERN_WARNING DRV_NAME
4454                        ": Warning: %s primary device specified but has no "
4455                        "effect in %s mode\n",
4456                        primary, bond_mode_name(bond_mode));
4457                 primary = NULL;
4458         }
4459
4460         /* fill params struct with the proper values */
4461         params->mode = bond_mode;
4462         params->xmit_policy = xmit_hashtype;
4463         params->miimon = miimon;
4464         params->arp_interval = arp_interval;
4465         params->updelay = updelay;
4466         params->downdelay = downdelay;
4467         params->use_carrier = use_carrier;
4468         params->lacp_fast = lacp_fast;
4469         params->primary[0] = 0;
4470
4471         if (primary) {
4472                 strncpy(params->primary, primary, IFNAMSIZ);
4473                 params->primary[IFNAMSIZ - 1] = 0;
4474         }
4475
4476         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4477
4478         return 0;
4479 }
4480
4481 /* Create a new bond based on the specified name and bonding parameters.
4482  * Caller must NOT hold rtnl_lock; we need to release it here before we
4483  * set up our sysfs entries.
4484  */
4485 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4486 {
4487         struct net_device *bond_dev;
4488         int res;
4489
4490         rtnl_lock();
4491         bond_dev = alloc_netdev(sizeof(struct bonding), name, ether_setup);
4492         if (!bond_dev) {
4493                 printk(KERN_ERR DRV_NAME
4494                        ": %s: eek! can't alloc netdev!\n",
4495                        name);
4496                 res = -ENOMEM;
4497                 goto out_rtnl;
4498         }
4499
4500         /* bond_init() must be called after dev_alloc_name() (for the
4501          * /proc files), but before register_netdevice(), because we
4502          * need to set function pointers.
4503          */
4504
4505         res = bond_init(bond_dev, params);
4506         if (res < 0) {
4507                 goto out_netdev;
4508         }
4509
4510         SET_MODULE_OWNER(bond_dev);
4511
4512         res = register_netdevice(bond_dev);
4513         if (res < 0) {
4514                 goto out_bond;
4515         }
4516         if (newbond)
4517                 *newbond = bond_dev->priv;
4518
4519         rtnl_unlock(); /* allows sysfs registration of net device */
4520         res = bond_create_sysfs_entry(bond_dev->priv);
4521         goto done;
4522 out_bond:
4523         bond_deinit(bond_dev);
4524 out_netdev:
4525         free_netdev(bond_dev);
4526 out_rtnl:
4527         rtnl_unlock();
4528 done:
4529         return res;
4530 }
4531
4532 static int __init bonding_init(void)
4533 {
4534         int i;
4535         int res;
4536         char new_bond_name[8];  /* Enough room for 999 bonds at init. */
4537
4538         printk(KERN_INFO "%s", version);
4539
4540         res = bond_check_params(&bonding_defaults);
4541         if (res) {
4542                 goto out;
4543         }
4544
4545 #ifdef CONFIG_PROC_FS
4546         bond_create_proc_dir();
4547 #endif
4548         for (i = 0; i < max_bonds; i++) {
4549                 sprintf(new_bond_name, "bond%d",i);
4550                 res = bond_create(new_bond_name,&bonding_defaults, NULL);
4551                 if (res)
4552                         goto err;
4553         }
4554
4555         res = bond_create_sysfs();
4556         if (res)
4557                 goto err;
4558
4559         register_netdevice_notifier(&bond_netdev_notifier);
4560         register_inetaddr_notifier(&bond_inetaddr_notifier);
4561
4562         goto out;
4563 err:
4564         rtnl_lock();
4565         bond_free_all();
4566         bond_destroy_sysfs();
4567         rtnl_unlock();
4568 out:
4569         return res;
4570
4571 }
4572
4573 static void __exit bonding_exit(void)
4574 {
4575         unregister_netdevice_notifier(&bond_netdev_notifier);
4576         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4577
4578         rtnl_lock();
4579         bond_free_all();
4580         bond_destroy_sysfs();
4581         rtnl_unlock();
4582 }
4583
4584 module_init(bonding_init);
4585 module_exit(bonding_exit);
4586 MODULE_LICENSE("GPL");
4587 MODULE_VERSION(DRV_VERSION);
4588 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4589 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4590 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4591
4592 /*
4593  * Local variables:
4594  *  c-indent-level: 8
4595  *  c-basic-offset: 8
4596  *  tab-width: 8
4597  * End:
4598  */
4599