net/bonding: Optionally allow ethernet slaves to keep own MAC
[linux-2.6.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/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <asm/system.h>
57 #include <asm/io.h>
58 #include <asm/dma.h>
59 #include <asm/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
66 #include <net/sock.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
72 #include <net/arp.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <net/route.h>
78 #include <net/net_namespace.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 static char *arp_validate = NULL;
101 static int fail_over_mac = 0;
102 struct bond_params bonding_defaults;
103
104 module_param(max_bonds, int, 0);
105 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
106 module_param(miimon, int, 0);
107 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
108 module_param(updelay, int, 0);
109 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
110 module_param(downdelay, int, 0);
111 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
112                             "in milliseconds");
113 module_param(use_carrier, int, 0);
114 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
115                               "0 for off, 1 for on (default)");
116 module_param(mode, charp, 0);
117 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
118                        "1 for active-backup, 2 for balance-xor, "
119                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
120                        "6 for balance-alb");
121 module_param(primary, charp, 0);
122 MODULE_PARM_DESC(primary, "Primary network device to use");
123 module_param(lacp_rate, charp, 0);
124 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
125                             "(slow/fast)");
126 module_param(xmit_hash_policy, charp, 0);
127 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
128                                    ", 1 for layer 3+4");
129 module_param(arp_interval, int, 0);
130 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
131 module_param_array(arp_ip_target, charp, NULL, 0);
132 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
133 module_param(arp_validate, charp, 0);
134 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
135 module_param(fail_over_mac, int, 0);
136 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  0 of off (default), 1 for on.");
137
138 /*----------------------------- Global variables ----------------------------*/
139
140 static const char * const version =
141         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
142
143 LIST_HEAD(bond_dev_list);
144
145 #ifdef CONFIG_PROC_FS
146 static struct proc_dir_entry *bond_proc_dir = NULL;
147 #endif
148
149 extern struct rw_semaphore bonding_rwsem;
150 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
151 static int arp_ip_count = 0;
152 static int bond_mode    = BOND_MODE_ROUNDROBIN;
153 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
154 static int lacp_fast    = 0;
155
156
157 struct bond_parm_tbl bond_lacp_tbl[] = {
158 {       "slow",         AD_LACP_SLOW},
159 {       "fast",         AD_LACP_FAST},
160 {       NULL,           -1},
161 };
162
163 struct bond_parm_tbl bond_mode_tbl[] = {
164 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
165 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
166 {       "balance-xor",          BOND_MODE_XOR},
167 {       "broadcast",            BOND_MODE_BROADCAST},
168 {       "802.3ad",              BOND_MODE_8023AD},
169 {       "balance-tlb",          BOND_MODE_TLB},
170 {       "balance-alb",          BOND_MODE_ALB},
171 {       NULL,                   -1},
172 };
173
174 struct bond_parm_tbl xmit_hashtype_tbl[] = {
175 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
176 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
177 {       NULL,                   -1},
178 };
179
180 struct bond_parm_tbl arp_validate_tbl[] = {
181 {       "none",                 BOND_ARP_VALIDATE_NONE},
182 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
183 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
184 {       "all",                  BOND_ARP_VALIDATE_ALL},
185 {       NULL,                   -1},
186 };
187
188 /*-------------------------- Forward declarations ---------------------------*/
189
190 static void bond_send_gratuitous_arp(struct bonding *bond);
191
192 /*---------------------------- General routines -----------------------------*/
193
194 static const char *bond_mode_name(int mode)
195 {
196         switch (mode) {
197         case BOND_MODE_ROUNDROBIN :
198                 return "load balancing (round-robin)";
199         case BOND_MODE_ACTIVEBACKUP :
200                 return "fault-tolerance (active-backup)";
201         case BOND_MODE_XOR :
202                 return "load balancing (xor)";
203         case BOND_MODE_BROADCAST :
204                 return "fault-tolerance (broadcast)";
205         case BOND_MODE_8023AD:
206                 return "IEEE 802.3ad Dynamic link aggregation";
207         case BOND_MODE_TLB:
208                 return "transmit load balancing";
209         case BOND_MODE_ALB:
210                 return "adaptive load balancing";
211         default:
212                 return "unknown";
213         }
214 }
215
216 /*---------------------------------- VLAN -----------------------------------*/
217
218 /**
219  * bond_add_vlan - add a new vlan id on bond
220  * @bond: bond that got the notification
221  * @vlan_id: the vlan id to add
222  *
223  * Returns -ENOMEM if allocation failed.
224  */
225 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
226 {
227         struct vlan_entry *vlan;
228
229         dprintk("bond: %s, vlan id %d\n",
230                 (bond ? bond->dev->name: "None"), vlan_id);
231
232         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
233         if (!vlan) {
234                 return -ENOMEM;
235         }
236
237         INIT_LIST_HEAD(&vlan->vlan_list);
238         vlan->vlan_id = vlan_id;
239         vlan->vlan_ip = 0;
240
241         write_lock_bh(&bond->lock);
242
243         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
244
245         write_unlock_bh(&bond->lock);
246
247         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
248
249         return 0;
250 }
251
252 /**
253  * bond_del_vlan - delete a vlan id from bond
254  * @bond: bond that got the notification
255  * @vlan_id: the vlan id to delete
256  *
257  * returns -ENODEV if @vlan_id was not found in @bond.
258  */
259 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
260 {
261         struct vlan_entry *vlan, *next;
262         int res = -ENODEV;
263
264         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
265
266         write_lock_bh(&bond->lock);
267
268         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
269                 if (vlan->vlan_id == vlan_id) {
270                         list_del(&vlan->vlan_list);
271
272                         if ((bond->params.mode == BOND_MODE_TLB) ||
273                             (bond->params.mode == BOND_MODE_ALB)) {
274                                 bond_alb_clear_vlan(bond, vlan_id);
275                         }
276
277                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
278                                 bond->dev->name);
279
280                         kfree(vlan);
281
282                         if (list_empty(&bond->vlan_list) &&
283                             (bond->slave_cnt == 0)) {
284                                 /* Last VLAN removed and no slaves, so
285                                  * restore block on adding VLANs. This will
286                                  * be removed once new slaves that are not
287                                  * VLAN challenged will be added.
288                                  */
289                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
290                         }
291
292                         res = 0;
293                         goto out;
294                 }
295         }
296
297         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
298                 bond->dev->name);
299
300 out:
301         write_unlock_bh(&bond->lock);
302         return res;
303 }
304
305 /**
306  * bond_has_challenged_slaves
307  * @bond: the bond we're working on
308  *
309  * Searches the slave list. Returns 1 if a vlan challenged slave
310  * was found, 0 otherwise.
311  *
312  * Assumes bond->lock is held.
313  */
314 static int bond_has_challenged_slaves(struct bonding *bond)
315 {
316         struct slave *slave;
317         int i;
318
319         bond_for_each_slave(bond, slave, i) {
320                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
321                         dprintk("found VLAN challenged slave - %s\n",
322                                 slave->dev->name);
323                         return 1;
324                 }
325         }
326
327         dprintk("no VLAN challenged slaves found\n");
328         return 0;
329 }
330
331 /**
332  * bond_next_vlan - safely skip to the next item in the vlans list.
333  * @bond: the bond we're working on
334  * @curr: item we're advancing from
335  *
336  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
337  * or @curr->next otherwise (even if it is @curr itself again).
338  * 
339  * Caller must hold bond->lock
340  */
341 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
342 {
343         struct vlan_entry *next, *last;
344
345         if (list_empty(&bond->vlan_list)) {
346                 return NULL;
347         }
348
349         if (!curr) {
350                 next = list_entry(bond->vlan_list.next,
351                                   struct vlan_entry, vlan_list);
352         } else {
353                 last = list_entry(bond->vlan_list.prev,
354                                   struct vlan_entry, vlan_list);
355                 if (last == curr) {
356                         next = list_entry(bond->vlan_list.next,
357                                           struct vlan_entry, vlan_list);
358                 } else {
359                         next = list_entry(curr->vlan_list.next,
360                                           struct vlan_entry, vlan_list);
361                 }
362         }
363
364         return next;
365 }
366
367 /**
368  * bond_dev_queue_xmit - Prepare skb for xmit.
369  * 
370  * @bond: bond device that got this skb for tx.
371  * @skb: hw accel VLAN tagged skb to transmit
372  * @slave_dev: slave that is supposed to xmit this skbuff
373  * 
374  * When the bond gets an skb to transmit that is
375  * already hardware accelerated VLAN tagged, and it
376  * needs to relay this skb to a slave that is not
377  * hw accel capable, the skb needs to be "unaccelerated",
378  * i.e. strip the hwaccel tag and re-insert it as part
379  * of the payload.
380  */
381 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
382 {
383         unsigned short vlan_id;
384
385         if (!list_empty(&bond->vlan_list) &&
386             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
387             vlan_get_tag(skb, &vlan_id) == 0) {
388                 skb->dev = slave_dev;
389                 skb = vlan_put_tag(skb, vlan_id);
390                 if (!skb) {
391                         /* vlan_put_tag() frees the skb in case of error,
392                          * so return success here so the calling functions
393                          * won't attempt to free is again.
394                          */
395                         return 0;
396                 }
397         } else {
398                 skb->dev = slave_dev;
399         }
400
401         skb->priority = 1;
402         dev_queue_xmit(skb);
403
404         return 0;
405 }
406
407 /*
408  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
409  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
410  * lock because:
411  * a. This operation is performed in IOCTL context,
412  * b. The operation is protected by the RTNL semaphore in the 8021q code,
413  * c. Holding a lock with BH disabled while directly calling a base driver
414  *    entry point is generally a BAD idea.
415  * 
416  * The design of synchronization/protection for this operation in the 8021q
417  * module is good for one or more VLAN devices over a single physical device
418  * and cannot be extended for a teaming solution like bonding, so there is a
419  * potential race condition here where a net device from the vlan group might
420  * be referenced (either by a base driver or the 8021q code) while it is being
421  * removed from the system. However, it turns out we're not making matters
422  * worse, and if it works for regular VLAN usage it will work here too.
423 */
424
425 /**
426  * bond_vlan_rx_register - Propagates registration to slaves
427  * @bond_dev: bonding net device that got called
428  * @grp: vlan group being registered
429  */
430 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
431 {
432         struct bonding *bond = bond_dev->priv;
433         struct slave *slave;
434         int i;
435
436         bond->vlgrp = grp;
437
438         bond_for_each_slave(bond, slave, i) {
439                 struct net_device *slave_dev = slave->dev;
440
441                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
442                     slave_dev->vlan_rx_register) {
443                         slave_dev->vlan_rx_register(slave_dev, grp);
444                 }
445         }
446 }
447
448 /**
449  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
450  * @bond_dev: bonding net device that got called
451  * @vid: vlan id being added
452  */
453 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
454 {
455         struct bonding *bond = bond_dev->priv;
456         struct slave *slave;
457         int i, res;
458
459         bond_for_each_slave(bond, slave, i) {
460                 struct net_device *slave_dev = slave->dev;
461
462                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
463                     slave_dev->vlan_rx_add_vid) {
464                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
465                 }
466         }
467
468         res = bond_add_vlan(bond, vid);
469         if (res) {
470                 printk(KERN_ERR DRV_NAME
471                        ": %s: Error: Failed to add vlan id %d\n",
472                        bond_dev->name, vid);
473         }
474 }
475
476 /**
477  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
478  * @bond_dev: bonding net device that got called
479  * @vid: vlan id being removed
480  */
481 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
482 {
483         struct bonding *bond = bond_dev->priv;
484         struct slave *slave;
485         struct net_device *vlan_dev;
486         int i, res;
487
488         bond_for_each_slave(bond, slave, i) {
489                 struct net_device *slave_dev = slave->dev;
490
491                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
492                     slave_dev->vlan_rx_kill_vid) {
493                         /* Save and then restore vlan_dev in the grp array,
494                          * since the slave's driver might clear it.
495                          */
496                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
497                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
498                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
499                 }
500         }
501
502         res = bond_del_vlan(bond, vid);
503         if (res) {
504                 printk(KERN_ERR DRV_NAME
505                        ": %s: Error: Failed to remove vlan id %d\n",
506                        bond_dev->name, vid);
507         }
508 }
509
510 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
511 {
512         struct vlan_entry *vlan;
513
514         write_lock_bh(&bond->lock);
515
516         if (list_empty(&bond->vlan_list)) {
517                 goto out;
518         }
519
520         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
521             slave_dev->vlan_rx_register) {
522                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
523         }
524
525         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
526             !(slave_dev->vlan_rx_add_vid)) {
527                 goto out;
528         }
529
530         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
531                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
532         }
533
534 out:
535         write_unlock_bh(&bond->lock);
536 }
537
538 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
539 {
540         struct vlan_entry *vlan;
541         struct net_device *vlan_dev;
542
543         write_lock_bh(&bond->lock);
544
545         if (list_empty(&bond->vlan_list)) {
546                 goto out;
547         }
548
549         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
550             !(slave_dev->vlan_rx_kill_vid)) {
551                 goto unreg;
552         }
553
554         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
555                 /* Save and then restore vlan_dev in the grp array,
556                  * since the slave's driver might clear it.
557                  */
558                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
559                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
560                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
561         }
562
563 unreg:
564         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
565             slave_dev->vlan_rx_register) {
566                 slave_dev->vlan_rx_register(slave_dev, NULL);
567         }
568
569 out:
570         write_unlock_bh(&bond->lock);
571 }
572
573 /*------------------------------- Link status -------------------------------*/
574
575 /*
576  * Set the carrier state for the master according to the state of its
577  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
578  * do special 802.3ad magic.
579  *
580  * Returns zero if carrier state does not change, nonzero if it does.
581  */
582 static int bond_set_carrier(struct bonding *bond)
583 {
584         struct slave *slave;
585         int i;
586
587         if (bond->slave_cnt == 0)
588                 goto down;
589
590         if (bond->params.mode == BOND_MODE_8023AD)
591                 return bond_3ad_set_carrier(bond);
592
593         bond_for_each_slave(bond, slave, i) {
594                 if (slave->link == BOND_LINK_UP) {
595                         if (!netif_carrier_ok(bond->dev)) {
596                                 netif_carrier_on(bond->dev);
597                                 return 1;
598                         }
599                         return 0;
600                 }
601         }
602
603 down:
604         if (netif_carrier_ok(bond->dev)) {
605                 netif_carrier_off(bond->dev);
606                 return 1;
607         }
608         return 0;
609 }
610
611 /*
612  * Get link speed and duplex from the slave's base driver
613  * using ethtool. If for some reason the call fails or the
614  * values are invalid, fake speed and duplex to 100/Full
615  * and return error.
616  */
617 static int bond_update_speed_duplex(struct slave *slave)
618 {
619         struct net_device *slave_dev = slave->dev;
620         struct ethtool_cmd etool;
621         int res;
622
623         /* Fake speed and duplex */
624         slave->speed = SPEED_100;
625         slave->duplex = DUPLEX_FULL;
626
627         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
628                 return -1;
629
630         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
631         if (res < 0)
632                 return -1;
633
634         switch (etool.speed) {
635         case SPEED_10:
636         case SPEED_100:
637         case SPEED_1000:
638         case SPEED_10000:
639                 break;
640         default:
641                 return -1;
642         }
643
644         switch (etool.duplex) {
645         case DUPLEX_FULL:
646         case DUPLEX_HALF:
647                 break;
648         default:
649                 return -1;
650         }
651
652         slave->speed = etool.speed;
653         slave->duplex = etool.duplex;
654
655         return 0;
656 }
657
658 /*
659  * if <dev> supports MII link status reporting, check its link status.
660  *
661  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
662  * depening upon the setting of the use_carrier parameter.
663  *
664  * Return either BMSR_LSTATUS, meaning that the link is up (or we
665  * can't tell and just pretend it is), or 0, meaning that the link is
666  * down.
667  *
668  * If reporting is non-zero, instead of faking link up, return -1 if
669  * both ETHTOOL and MII ioctls fail (meaning the device does not
670  * support them).  If use_carrier is set, return whatever it says.
671  * It'd be nice if there was a good way to tell if a driver supports
672  * netif_carrier, but there really isn't.
673  */
674 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
675 {
676         static int (* ioctl)(struct net_device *, struct ifreq *, int);
677         struct ifreq ifr;
678         struct mii_ioctl_data *mii;
679
680         if (bond->params.use_carrier) {
681                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
682         }
683
684         ioctl = slave_dev->do_ioctl;
685         if (ioctl) {
686                 /* TODO: set pointer to correct ioctl on a per team member */
687                 /*       bases to make this more efficient. that is, once  */
688                 /*       we determine the correct ioctl, we will always    */
689                 /*       call it and not the others for that team          */
690                 /*       member.                                           */
691
692                 /*
693                  * We cannot assume that SIOCGMIIPHY will also read a
694                  * register; not all network drivers (e.g., e100)
695                  * support that.
696                  */
697
698                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
699                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
700                 mii = if_mii(&ifr);
701                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
702                         mii->reg_num = MII_BMSR;
703                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
704                                 return (mii->val_out & BMSR_LSTATUS);
705                         }
706                 }
707         }
708
709         /*
710          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
711          * attempt to get link status from it if the above MII ioctls fail.
712          */
713         if (slave_dev->ethtool_ops) {
714                 if (slave_dev->ethtool_ops->get_link) {
715                         u32 link;
716
717                         link = slave_dev->ethtool_ops->get_link(slave_dev);
718
719                         return link ? BMSR_LSTATUS : 0;
720                 }
721         }
722
723         /*
724          * If reporting, report that either there's no dev->do_ioctl,
725          * or both SIOCGMIIREG and get_link failed (meaning that we
726          * cannot report link status).  If not reporting, pretend
727          * we're ok.
728          */
729         return (reporting ? -1 : BMSR_LSTATUS);
730 }
731
732 /*----------------------------- Multicast list ------------------------------*/
733
734 /*
735  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
736  */
737 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
738 {
739         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
740                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
741 }
742
743 /*
744  * returns dmi entry if found, NULL otherwise
745  */
746 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
747 {
748         struct dev_mc_list *idmi;
749
750         for (idmi = mc_list; idmi; idmi = idmi->next) {
751                 if (bond_is_dmi_same(dmi, idmi)) {
752                         return idmi;
753                 }
754         }
755
756         return NULL;
757 }
758
759 /*
760  * Push the promiscuity flag down to appropriate slaves
761  */
762 static void bond_set_promiscuity(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_promiscuity(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_promiscuity(slave->dev, inc);
774                 }
775         }
776 }
777
778 /*
779  * Push the allmulti flag down to all slaves
780  */
781 static void bond_set_allmulti(struct bonding *bond, int inc)
782 {
783         if (USES_PRIMARY(bond->params.mode)) {
784                 /* write lock already acquired */
785                 if (bond->curr_active_slave) {
786                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
787                 }
788         } else {
789                 struct slave *slave;
790                 int i;
791                 bond_for_each_slave(bond, slave, i) {
792                         dev_set_allmulti(slave->dev, inc);
793                 }
794         }
795 }
796
797 /*
798  * Add a Multicast address to slaves
799  * according to mode
800  */
801 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
802 {
803         if (USES_PRIMARY(bond->params.mode)) {
804                 /* write lock already acquired */
805                 if (bond->curr_active_slave) {
806                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
807                 }
808         } else {
809                 struct slave *slave;
810                 int i;
811                 bond_for_each_slave(bond, slave, i) {
812                         dev_mc_add(slave->dev, addr, alen, 0);
813                 }
814         }
815 }
816
817 /*
818  * Remove a multicast address from slave
819  * according to mode
820  */
821 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
822 {
823         if (USES_PRIMARY(bond->params.mode)) {
824                 /* write lock already acquired */
825                 if (bond->curr_active_slave) {
826                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
827                 }
828         } else {
829                 struct slave *slave;
830                 int i;
831                 bond_for_each_slave(bond, slave, i) {
832                         dev_mc_delete(slave->dev, addr, alen, 0);
833                 }
834         }
835 }
836
837
838 /*
839  * Retrieve the list of registered multicast addresses for the bonding
840  * device and retransmit an IGMP JOIN request to the current active
841  * slave.
842  */
843 static void bond_resend_igmp_join_requests(struct bonding *bond)
844 {
845         struct in_device *in_dev;
846         struct ip_mc_list *im;
847
848         rcu_read_lock();
849         in_dev = __in_dev_get_rcu(bond->dev);
850         if (in_dev) {
851                 for (im = in_dev->mc_list; im; im = im->next) {
852                         ip_mc_rejoin_group(im);
853                 }
854         }
855
856         rcu_read_unlock();
857 }
858
859 /*
860  * Totally destroys the mc_list in bond
861  */
862 static void bond_mc_list_destroy(struct bonding *bond)
863 {
864         struct dev_mc_list *dmi;
865
866         dmi = bond->mc_list;
867         while (dmi) {
868                 bond->mc_list = dmi->next;
869                 kfree(dmi);
870                 dmi = bond->mc_list;
871         }
872         bond->mc_list = NULL;
873 }
874
875 /*
876  * Copy all the Multicast addresses from src to the bonding device dst
877  */
878 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
879                              gfp_t gfp_flag)
880 {
881         struct dev_mc_list *dmi, *new_dmi;
882
883         for (dmi = mc_list; dmi; dmi = dmi->next) {
884                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
885
886                 if (!new_dmi) {
887                         /* FIXME: Potential memory leak !!! */
888                         return -ENOMEM;
889                 }
890
891                 new_dmi->next = bond->mc_list;
892                 bond->mc_list = new_dmi;
893                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
894                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
895                 new_dmi->dmi_users = dmi->dmi_users;
896                 new_dmi->dmi_gusers = dmi->dmi_gusers;
897         }
898
899         return 0;
900 }
901
902 /*
903  * flush all members of flush->mc_list from device dev->mc_list
904  */
905 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
906 {
907         struct bonding *bond = bond_dev->priv;
908         struct dev_mc_list *dmi;
909
910         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
911                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
912         }
913
914         if (bond->params.mode == BOND_MODE_8023AD) {
915                 /* del lacpdu mc addr from mc list */
916                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
917
918                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
919         }
920 }
921
922 /*--------------------------- Active slave change ---------------------------*/
923
924 /*
925  * Update the mc list and multicast-related flags for the new and
926  * old active slaves (if any) according to the multicast mode, and
927  * promiscuous flags unconditionally.
928  */
929 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
930 {
931         struct dev_mc_list *dmi;
932
933         if (!USES_PRIMARY(bond->params.mode)) {
934                 /* nothing to do -  mc list is already up-to-date on
935                  * all slaves
936                  */
937                 return;
938         }
939
940         if (old_active) {
941                 if (bond->dev->flags & IFF_PROMISC) {
942                         dev_set_promiscuity(old_active->dev, -1);
943                 }
944
945                 if (bond->dev->flags & IFF_ALLMULTI) {
946                         dev_set_allmulti(old_active->dev, -1);
947                 }
948
949                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
950                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
951                 }
952         }
953
954         if (new_active) {
955                 if (bond->dev->flags & IFF_PROMISC) {
956                         dev_set_promiscuity(new_active->dev, 1);
957                 }
958
959                 if (bond->dev->flags & IFF_ALLMULTI) {
960                         dev_set_allmulti(new_active->dev, 1);
961                 }
962
963                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
964                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
965                 }
966                 bond_resend_igmp_join_requests(bond);
967         }
968 }
969
970 /**
971  * find_best_interface - select the best available slave to be the active one
972  * @bond: our bonding struct
973  *
974  * Warning: Caller must hold curr_slave_lock for writing.
975  */
976 static struct slave *bond_find_best_slave(struct bonding *bond)
977 {
978         struct slave *new_active, *old_active;
979         struct slave *bestslave = NULL;
980         int mintime = bond->params.updelay;
981         int i;
982
983         new_active = old_active = bond->curr_active_slave;
984
985         if (!new_active) { /* there were no active slaves left */
986                 if (bond->slave_cnt > 0) {  /* found one slave */
987                         new_active = bond->first_slave;
988                 } else {
989                         return NULL; /* still no slave, return NULL */
990                 }
991         }
992
993         /* first try the primary link; if arping, a link must tx/rx traffic
994          * before it can be considered the curr_active_slave - also, we would skip
995          * slaves between the curr_active_slave and primary_slave that may be up
996          * and able to arp
997          */
998         if ((bond->primary_slave) &&
999             (!bond->params.arp_interval) &&
1000             (IS_UP(bond->primary_slave->dev))) {
1001                 new_active = bond->primary_slave;
1002         }
1003
1004         /* remember where to stop iterating over the slaves */
1005         old_active = new_active;
1006
1007         bond_for_each_slave_from(bond, new_active, i, old_active) {
1008                 if (IS_UP(new_active->dev)) {
1009                         if (new_active->link == BOND_LINK_UP) {
1010                                 return new_active;
1011                         } else if (new_active->link == BOND_LINK_BACK) {
1012                                 /* link up, but waiting for stabilization */
1013                                 if (new_active->delay < mintime) {
1014                                         mintime = new_active->delay;
1015                                         bestslave = new_active;
1016                                 }
1017                         }
1018                 }
1019         }
1020
1021         return bestslave;
1022 }
1023
1024 /**
1025  * change_active_interface - change the active slave into the specified one
1026  * @bond: our bonding struct
1027  * @new: the new slave to make the active one
1028  *
1029  * Set the new slave to the bond's settings and unset them on the old
1030  * curr_active_slave.
1031  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1032  *
1033  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1034  * because it is apparently the best available slave we have, even though its
1035  * updelay hasn't timed out yet.
1036  *
1037  * Warning: Caller must hold curr_slave_lock for writing.
1038  */
1039 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1040 {
1041         struct slave *old_active = bond->curr_active_slave;
1042
1043         if (old_active == new_active) {
1044                 return;
1045         }
1046
1047         if (new_active) {
1048                 if (new_active->link == BOND_LINK_BACK) {
1049                         if (USES_PRIMARY(bond->params.mode)) {
1050                                 printk(KERN_INFO DRV_NAME
1051                                        ": %s: making interface %s the new "
1052                                        "active one %d ms earlier.\n",
1053                                        bond->dev->name, new_active->dev->name,
1054                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1055                         }
1056
1057                         new_active->delay = 0;
1058                         new_active->link = BOND_LINK_UP;
1059                         new_active->jiffies = jiffies;
1060
1061                         if (bond->params.mode == BOND_MODE_8023AD) {
1062                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1063                         }
1064
1065                         if ((bond->params.mode == BOND_MODE_TLB) ||
1066                             (bond->params.mode == BOND_MODE_ALB)) {
1067                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1068                         }
1069                 } else {
1070                         if (USES_PRIMARY(bond->params.mode)) {
1071                                 printk(KERN_INFO DRV_NAME
1072                                        ": %s: making interface %s the new "
1073                                        "active one.\n",
1074                                        bond->dev->name, new_active->dev->name);
1075                         }
1076                 }
1077         }
1078
1079         if (USES_PRIMARY(bond->params.mode)) {
1080                 bond_mc_swap(bond, new_active, old_active);
1081         }
1082
1083         if ((bond->params.mode == BOND_MODE_TLB) ||
1084             (bond->params.mode == BOND_MODE_ALB)) {
1085                 bond_alb_handle_active_change(bond, new_active);
1086                 if (old_active)
1087                         bond_set_slave_inactive_flags(old_active);
1088                 if (new_active)
1089                         bond_set_slave_active_flags(new_active);
1090         } else {
1091                 bond->curr_active_slave = new_active;
1092         }
1093
1094         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1095                 if (old_active) {
1096                         bond_set_slave_inactive_flags(old_active);
1097                 }
1098
1099                 if (new_active) {
1100                         bond_set_slave_active_flags(new_active);
1101                 }
1102
1103                 /* when bonding does not set the slave MAC address, the bond MAC
1104                  * address is the one of the active slave.
1105                  */
1106                 if (new_active && bond->params.fail_over_mac)
1107                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1108                                 new_active->dev->addr_len);
1109                 if (bond->curr_active_slave &&
1110                         test_bit(__LINK_STATE_LINKWATCH_PENDING,
1111                                         &bond->curr_active_slave->dev->state)) {
1112                         dprintk("delaying gratuitous arp on %s\n",
1113                                 bond->curr_active_slave->dev->name);
1114                         bond->send_grat_arp = 1;
1115                 } else
1116                         bond_send_gratuitous_arp(bond);
1117         }
1118 }
1119
1120 /**
1121  * bond_select_active_slave - select a new active slave, if needed
1122  * @bond: our bonding struct
1123  *
1124  * This functions shoud be called when one of the following occurs:
1125  * - The old curr_active_slave has been released or lost its link.
1126  * - The primary_slave has got its link back.
1127  * - A slave has got its link back and there's no old curr_active_slave.
1128  *
1129  * Warning: Caller must hold curr_slave_lock for writing.
1130  */
1131 void bond_select_active_slave(struct bonding *bond)
1132 {
1133         struct slave *best_slave;
1134         int rv;
1135
1136         best_slave = bond_find_best_slave(bond);
1137         if (best_slave != bond->curr_active_slave) {
1138                 bond_change_active_slave(bond, best_slave);
1139                 rv = bond_set_carrier(bond);
1140                 if (!rv)
1141                         return;
1142
1143                 if (netif_carrier_ok(bond->dev)) {
1144                         printk(KERN_INFO DRV_NAME
1145                                ": %s: first active interface up!\n",
1146                                bond->dev->name);
1147                 } else {
1148                         printk(KERN_INFO DRV_NAME ": %s: "
1149                                "now running without any active interface !\n",
1150                                bond->dev->name);
1151                 }
1152         }
1153 }
1154
1155 /*--------------------------- slave list handling ---------------------------*/
1156
1157 /*
1158  * This function attaches the slave to the end of list.
1159  *
1160  * bond->lock held for writing by caller.
1161  */
1162 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1163 {
1164         if (bond->first_slave == NULL) { /* attaching the first slave */
1165                 new_slave->next = new_slave;
1166                 new_slave->prev = new_slave;
1167                 bond->first_slave = new_slave;
1168         } else {
1169                 new_slave->next = bond->first_slave;
1170                 new_slave->prev = bond->first_slave->prev;
1171                 new_slave->next->prev = new_slave;
1172                 new_slave->prev->next = new_slave;
1173         }
1174
1175         bond->slave_cnt++;
1176 }
1177
1178 /*
1179  * This function detaches the slave from the list.
1180  * WARNING: no check is made to verify if the slave effectively
1181  * belongs to <bond>.
1182  * Nothing is freed on return, structures are just unchained.
1183  * If any slave pointer in bond was pointing to <slave>,
1184  * it should be changed by the calling function.
1185  *
1186  * bond->lock held for writing by caller.
1187  */
1188 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1189 {
1190         if (slave->next) {
1191                 slave->next->prev = slave->prev;
1192         }
1193
1194         if (slave->prev) {
1195                 slave->prev->next = slave->next;
1196         }
1197
1198         if (bond->first_slave == slave) { /* slave is the first slave */
1199                 if (bond->slave_cnt > 1) { /* there are more slave */
1200                         bond->first_slave = slave->next;
1201                 } else {
1202                         bond->first_slave = NULL; /* slave was the last one */
1203                 }
1204         }
1205
1206         slave->next = NULL;
1207         slave->prev = NULL;
1208         bond->slave_cnt--;
1209 }
1210
1211 /*---------------------------------- IOCTL ----------------------------------*/
1212
1213 static int bond_sethwaddr(struct net_device *bond_dev,
1214                           struct net_device *slave_dev)
1215 {
1216         dprintk("bond_dev=%p\n", bond_dev);
1217         dprintk("slave_dev=%p\n", slave_dev);
1218         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1219         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1220         return 0;
1221 }
1222
1223 #define BOND_VLAN_FEATURES \
1224         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1225          NETIF_F_HW_VLAN_FILTER)
1226
1227 /* 
1228  * Compute the common dev->feature set available to all slaves.  Some
1229  * feature bits are managed elsewhere, so preserve those feature bits
1230  * on the master device.
1231  */
1232 static int bond_compute_features(struct bonding *bond)
1233 {
1234         struct slave *slave;
1235         struct net_device *bond_dev = bond->dev;
1236         unsigned long features = bond_dev->features;
1237         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1238                                                 bond_dev->hard_header_len);
1239         int i;
1240
1241         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1242         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1243                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1244
1245         bond_for_each_slave(bond, slave, i) {
1246                 features = netdev_compute_features(features,
1247                                                    slave->dev->features);
1248                 if (slave->dev->hard_header_len > max_hard_header_len)
1249                         max_hard_header_len = slave->dev->hard_header_len;
1250         }
1251
1252         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1253         bond_dev->features = features;
1254         bond_dev->hard_header_len = max_hard_header_len;
1255
1256         return 0;
1257 }
1258
1259
1260 static void bond_setup_by_slave(struct net_device *bond_dev,
1261                                 struct net_device *slave_dev)
1262 {
1263         struct bonding *bond = bond_dev->priv;
1264
1265         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1266
1267         bond_dev->type              = slave_dev->type;
1268         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1269         bond_dev->addr_len          = slave_dev->addr_len;
1270
1271         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1272                 slave_dev->addr_len);
1273         bond->setup_by_slave = 1;
1274 }
1275
1276 /* enslave device <slave> to bond device <master> */
1277 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1278 {
1279         struct bonding *bond = bond_dev->priv;
1280         struct slave *new_slave = NULL;
1281         struct dev_mc_list *dmi;
1282         struct sockaddr addr;
1283         int link_reporting;
1284         int old_features = bond_dev->features;
1285         int res = 0;
1286
1287         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1288                 slave_dev->do_ioctl == NULL) {
1289                 printk(KERN_WARNING DRV_NAME
1290                        ": %s: Warning: no link monitoring support for %s\n",
1291                        bond_dev->name, slave_dev->name);
1292         }
1293
1294         /* bond must be initialized by bond_open() before enslaving */
1295         if (!(bond_dev->flags & IFF_UP)) {
1296                 printk(KERN_WARNING DRV_NAME
1297                         " %s: master_dev is not up in bond_enslave\n",
1298                         bond_dev->name);
1299         }
1300
1301         /* already enslaved */
1302         if (slave_dev->flags & IFF_SLAVE) {
1303                 dprintk("Error, Device was already enslaved\n");
1304                 return -EBUSY;
1305         }
1306
1307         /* vlan challenged mutual exclusion */
1308         /* no need to lock since we're protected by rtnl_lock */
1309         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1310                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1311                 if (!list_empty(&bond->vlan_list)) {
1312                         printk(KERN_ERR DRV_NAME
1313                                ": %s: Error: cannot enslave VLAN "
1314                                "challenged slave %s on VLAN enabled "
1315                                "bond %s\n", bond_dev->name, slave_dev->name,
1316                                bond_dev->name);
1317                         return -EPERM;
1318                 } else {
1319                         printk(KERN_WARNING DRV_NAME
1320                                ": %s: Warning: enslaved VLAN challenged "
1321                                "slave %s. Adding VLANs will be blocked as "
1322                                "long as %s is part of bond %s\n",
1323                                bond_dev->name, slave_dev->name, slave_dev->name,
1324                                bond_dev->name);
1325                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1326                 }
1327         } else {
1328                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1329                 if (bond->slave_cnt == 0) {
1330                         /* First slave, and it is not VLAN challenged,
1331                          * so remove the block of adding VLANs over the bond.
1332                          */
1333                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1334                 }
1335         }
1336
1337         /*
1338          * Old ifenslave binaries are no longer supported.  These can
1339          * be identified with moderate accurary by the state of the slave:
1340          * the current ifenslave will set the interface down prior to
1341          * enslaving it; the old ifenslave will not.
1342          */
1343         if ((slave_dev->flags & IFF_UP)) {
1344                 printk(KERN_ERR DRV_NAME ": %s is up. "
1345                        "This may be due to an out of date ifenslave.\n",
1346                        slave_dev->name);
1347                 res = -EPERM;
1348                 goto err_undo_flags;
1349         }
1350
1351         /* set bonding device ether type by slave - bonding netdevices are
1352          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1353          * there is a need to override some of the type dependent attribs/funcs.
1354          *
1355          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1356          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1357          */
1358         if (bond->slave_cnt == 0) {
1359                 if (slave_dev->type != ARPHRD_ETHER)
1360                         bond_setup_by_slave(bond_dev, slave_dev);
1361         } else if (bond_dev->type != slave_dev->type) {
1362                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1363                         "from other slaves (%d), can not enslave it.\n",
1364                         slave_dev->name,
1365                         slave_dev->type, bond_dev->type);
1366                         res = -EINVAL;
1367                         goto err_undo_flags;
1368         }
1369
1370         if (slave_dev->set_mac_address == NULL) {
1371                 if (bond->slave_cnt == 0) {
1372                         printk(KERN_WARNING DRV_NAME
1373                                ": %s: Warning: The first slave device "
1374                                "specified does not support setting the MAC "
1375                                "address. Enabling the fail_over_mac option.",
1376                                bond_dev->name);
1377                         bond->params.fail_over_mac = 1;
1378                 } else if (!bond->params.fail_over_mac) {
1379                         printk(KERN_ERR DRV_NAME
1380                                 ": %s: Error: The slave device specified "
1381                                 "does not support setting the MAC address, "
1382                                 "but fail_over_mac is not enabled.\n"
1383                                 , bond_dev->name);
1384                         res = -EOPNOTSUPP;
1385                         goto err_undo_flags;
1386                 }
1387         }
1388
1389         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1390         if (!new_slave) {
1391                 res = -ENOMEM;
1392                 goto err_undo_flags;
1393         }
1394
1395         /* save slave's original flags before calling
1396          * netdev_set_master and dev_open
1397          */
1398         new_slave->original_flags = slave_dev->flags;
1399
1400         /*
1401          * Save slave's original ("permanent") mac address for modes
1402          * that need it, and for restoring it upon release, and then
1403          * set it to the master's address
1404          */
1405         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1406
1407         if (!bond->params.fail_over_mac) {
1408                 /*
1409                  * Set slave to master's mac address.  The application already
1410                  * set the master's mac address to that of the first slave
1411                  */
1412                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1413                 addr.sa_family = slave_dev->type;
1414                 res = dev_set_mac_address(slave_dev, &addr);
1415                 if (res) {
1416                         dprintk("Error %d calling set_mac_address\n", res);
1417                         goto err_free;
1418                 }
1419         }
1420
1421         res = netdev_set_master(slave_dev, bond_dev);
1422         if (res) {
1423                 dprintk("Error %d calling netdev_set_master\n", res);
1424                 goto err_close;
1425         }
1426         /* open the slave since the application closed it */
1427         res = dev_open(slave_dev);
1428         if (res) {
1429                 dprintk("Openning slave %s failed\n", slave_dev->name);
1430                 goto err_restore_mac;
1431         }
1432
1433         new_slave->dev = slave_dev;
1434         slave_dev->priv_flags |= IFF_BONDING;
1435
1436         if ((bond->params.mode == BOND_MODE_TLB) ||
1437             (bond->params.mode == BOND_MODE_ALB)) {
1438                 /* bond_alb_init_slave() must be called before all other stages since
1439                  * it might fail and we do not want to have to undo everything
1440                  */
1441                 res = bond_alb_init_slave(bond, new_slave);
1442                 if (res) {
1443                         goto err_unset_master;
1444                 }
1445         }
1446
1447         /* If the mode USES_PRIMARY, then the new slave gets the
1448          * master's promisc (and mc) settings only if it becomes the
1449          * curr_active_slave, and that is taken care of later when calling
1450          * bond_change_active()
1451          */
1452         if (!USES_PRIMARY(bond->params.mode)) {
1453                 /* set promiscuity level to new slave */
1454                 if (bond_dev->flags & IFF_PROMISC) {
1455                         dev_set_promiscuity(slave_dev, 1);
1456                 }
1457
1458                 /* set allmulti level to new slave */
1459                 if (bond_dev->flags & IFF_ALLMULTI) {
1460                         dev_set_allmulti(slave_dev, 1);
1461                 }
1462
1463                 /* upload master's mc_list to new slave */
1464                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1465                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1466                 }
1467         }
1468
1469         if (bond->params.mode == BOND_MODE_8023AD) {
1470                 /* add lacpdu mc addr to mc list */
1471                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1472
1473                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1474         }
1475
1476         bond_add_vlans_on_slave(bond, slave_dev);
1477
1478         write_lock_bh(&bond->lock);
1479
1480         bond_attach_slave(bond, new_slave);
1481
1482         new_slave->delay = 0;
1483         new_slave->link_failure_count = 0;
1484
1485         bond_compute_features(bond);
1486
1487         new_slave->last_arp_rx = jiffies;
1488
1489         if (bond->params.miimon && !bond->params.use_carrier) {
1490                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1491
1492                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1493                         /*
1494                          * miimon is set but a bonded network driver
1495                          * does not support ETHTOOL/MII and
1496                          * arp_interval is not set.  Note: if
1497                          * use_carrier is enabled, we will never go
1498                          * here (because netif_carrier is always
1499                          * supported); thus, we don't need to change
1500                          * the messages for netif_carrier.
1501                          */
1502                         printk(KERN_WARNING DRV_NAME
1503                                ": %s: Warning: MII and ETHTOOL support not "
1504                                "available for interface %s, and "
1505                                "arp_interval/arp_ip_target module parameters "
1506                                "not specified, thus bonding will not detect "
1507                                "link failures! see bonding.txt for details.\n",
1508                                bond_dev->name, slave_dev->name);
1509                 } else if (link_reporting == -1) {
1510                         /* unable get link status using mii/ethtool */
1511                         printk(KERN_WARNING DRV_NAME
1512                                ": %s: Warning: can't get link status from "
1513                                "interface %s; the network driver associated "
1514                                "with this interface does not support MII or "
1515                                "ETHTOOL link status reporting, thus miimon "
1516                                "has no effect on this interface.\n",
1517                                bond_dev->name, slave_dev->name);
1518                 }
1519         }
1520
1521         /* check for initial state */
1522         if (!bond->params.miimon ||
1523             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1524                 if (bond->params.updelay) {
1525                         dprintk("Initial state of slave_dev is "
1526                                 "BOND_LINK_BACK\n");
1527                         new_slave->link  = BOND_LINK_BACK;
1528                         new_slave->delay = bond->params.updelay;
1529                 } else {
1530                         dprintk("Initial state of slave_dev is "
1531                                 "BOND_LINK_UP\n");
1532                         new_slave->link  = BOND_LINK_UP;
1533                 }
1534                 new_slave->jiffies = jiffies;
1535         } else {
1536                 dprintk("Initial state of slave_dev is "
1537                         "BOND_LINK_DOWN\n");
1538                 new_slave->link  = BOND_LINK_DOWN;
1539         }
1540
1541         if (bond_update_speed_duplex(new_slave) &&
1542             (new_slave->link != BOND_LINK_DOWN)) {
1543                 printk(KERN_WARNING DRV_NAME
1544                        ": %s: Warning: failed to get speed and duplex from %s, "
1545                        "assumed to be 100Mb/sec and Full.\n",
1546                        bond_dev->name, new_slave->dev->name);
1547
1548                 if (bond->params.mode == BOND_MODE_8023AD) {
1549                         printk(KERN_WARNING DRV_NAME
1550                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1551                                "support in base driver for proper aggregator "
1552                                "selection.\n", bond_dev->name);
1553                 }
1554         }
1555
1556         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1557                 /* if there is a primary slave, remember it */
1558                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1559                         bond->primary_slave = new_slave;
1560                 }
1561         }
1562
1563         switch (bond->params.mode) {
1564         case BOND_MODE_ACTIVEBACKUP:
1565                 bond_set_slave_inactive_flags(new_slave);
1566                 bond_select_active_slave(bond);
1567                 break;
1568         case BOND_MODE_8023AD:
1569                 /* in 802.3ad mode, the internal mechanism
1570                  * will activate the slaves in the selected
1571                  * aggregator
1572                  */
1573                 bond_set_slave_inactive_flags(new_slave);
1574                 /* if this is the first slave */
1575                 if (bond->slave_cnt == 1) {
1576                         SLAVE_AD_INFO(new_slave).id = 1;
1577                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1578                          * can be called only after the mac address of the bond is set
1579                          */
1580                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1581                                             bond->params.lacp_fast);
1582                 } else {
1583                         SLAVE_AD_INFO(new_slave).id =
1584                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1585                 }
1586
1587                 bond_3ad_bind_slave(new_slave);
1588                 break;
1589         case BOND_MODE_TLB:
1590         case BOND_MODE_ALB:
1591                 new_slave->state = BOND_STATE_ACTIVE;
1592                 if ((!bond->curr_active_slave) &&
1593                     (new_slave->link != BOND_LINK_DOWN)) {
1594                         /* first slave or no active slave yet, and this link
1595                          * is OK, so make this interface the active one
1596                          */
1597                         bond_change_active_slave(bond, new_slave);
1598                 } else {
1599                         bond_set_slave_inactive_flags(new_slave);
1600                 }
1601                 break;
1602         default:
1603                 dprintk("This slave is always active in trunk mode\n");
1604
1605                 /* always active in trunk mode */
1606                 new_slave->state = BOND_STATE_ACTIVE;
1607
1608                 /* In trunking mode there is little meaning to curr_active_slave
1609                  * anyway (it holds no special properties of the bond device),
1610                  * so we can change it without calling change_active_interface()
1611                  */
1612                 if (!bond->curr_active_slave) {
1613                         bond->curr_active_slave = new_slave;
1614                 }
1615                 break;
1616         } /* switch(bond_mode) */
1617
1618         bond_set_carrier(bond);
1619
1620         write_unlock_bh(&bond->lock);
1621
1622         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1623         if (res)
1624                 goto err_unset_master;
1625
1626         printk(KERN_INFO DRV_NAME
1627                ": %s: enslaving %s as a%s interface with a%s link.\n",
1628                bond_dev->name, slave_dev->name,
1629                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1630                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1631
1632         /* enslave is successful */
1633         return 0;
1634
1635 /* Undo stages on error */
1636 err_unset_master:
1637         netdev_set_master(slave_dev, NULL);
1638
1639 err_close:
1640         dev_close(slave_dev);
1641
1642 err_restore_mac:
1643         if (!bond->params.fail_over_mac) {
1644                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1645                 addr.sa_family = slave_dev->type;
1646                 dev_set_mac_address(slave_dev, &addr);
1647         }
1648
1649 err_free:
1650         kfree(new_slave);
1651
1652 err_undo_flags:
1653         bond_dev->features = old_features;
1654  
1655         return res;
1656 }
1657
1658 /*
1659  * Try to release the slave device <slave> from the bond device <master>
1660  * It is legal to access curr_active_slave without a lock because all the function
1661  * is write-locked.
1662  *
1663  * The rules for slave state should be:
1664  *   for Active/Backup:
1665  *     Active stays on all backups go down
1666  *   for Bonded connections:
1667  *     The first up interface should be left on and all others downed.
1668  */
1669 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1670 {
1671         struct bonding *bond = bond_dev->priv;
1672         struct slave *slave, *oldcurrent;
1673         struct sockaddr addr;
1674         int mac_addr_differ;
1675         DECLARE_MAC_BUF(mac);
1676
1677         /* slave is not a slave or master is not master of this slave */
1678         if (!(slave_dev->flags & IFF_SLAVE) ||
1679             (slave_dev->master != bond_dev)) {
1680                 printk(KERN_ERR DRV_NAME
1681                        ": %s: Error: cannot release %s.\n",
1682                        bond_dev->name, slave_dev->name);
1683                 return -EINVAL;
1684         }
1685
1686         write_lock_bh(&bond->lock);
1687
1688         slave = bond_get_slave_by_dev(bond, slave_dev);
1689         if (!slave) {
1690                 /* not a slave of this bond */
1691                 printk(KERN_INFO DRV_NAME
1692                        ": %s: %s not enslaved\n",
1693                        bond_dev->name, slave_dev->name);
1694                 write_unlock_bh(&bond->lock);
1695                 return -EINVAL;
1696         }
1697
1698         mac_addr_differ = memcmp(bond_dev->dev_addr,
1699                                  slave->perm_hwaddr,
1700                                  ETH_ALEN);
1701         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1702                 printk(KERN_WARNING DRV_NAME
1703                        ": %s: Warning: the permanent HWaddr of %s - "
1704                        "%s - is still in use by %s. "
1705                        "Set the HWaddr of %s to a different address "
1706                        "to avoid conflicts.\n",
1707                        bond_dev->name,
1708                        slave_dev->name,
1709                        print_mac(mac, slave->perm_hwaddr),
1710                        bond_dev->name,
1711                        slave_dev->name);
1712         }
1713
1714         /* Inform AD package of unbinding of slave. */
1715         if (bond->params.mode == BOND_MODE_8023AD) {
1716                 /* must be called before the slave is
1717                  * detached from the list
1718                  */
1719                 bond_3ad_unbind_slave(slave);
1720         }
1721
1722         printk(KERN_INFO DRV_NAME
1723                ": %s: releasing %s interface %s\n",
1724                bond_dev->name,
1725                (slave->state == BOND_STATE_ACTIVE)
1726                ? "active" : "backup",
1727                slave_dev->name);
1728
1729         oldcurrent = bond->curr_active_slave;
1730
1731         bond->current_arp_slave = NULL;
1732
1733         /* release the slave from its bond */
1734         bond_detach_slave(bond, slave);
1735
1736         bond_compute_features(bond);
1737
1738         if (bond->primary_slave == slave) {
1739                 bond->primary_slave = NULL;
1740         }
1741
1742         if (oldcurrent == slave) {
1743                 bond_change_active_slave(bond, NULL);
1744         }
1745
1746         if ((bond->params.mode == BOND_MODE_TLB) ||
1747             (bond->params.mode == BOND_MODE_ALB)) {
1748                 /* Must be called only after the slave has been
1749                  * detached from the list and the curr_active_slave
1750                  * has been cleared (if our_slave == old_current),
1751                  * but before a new active slave is selected.
1752                  */
1753                 bond_alb_deinit_slave(bond, slave);
1754         }
1755
1756         if (oldcurrent == slave)
1757                 bond_select_active_slave(bond);
1758
1759         if (bond->slave_cnt == 0) {
1760                 bond_set_carrier(bond);
1761
1762                 /* if the last slave was removed, zero the mac address
1763                  * of the master so it will be set by the application
1764                  * to the mac address of the first slave
1765                  */
1766                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1767
1768                 if (list_empty(&bond->vlan_list)) {
1769                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1770                 } else {
1771                         printk(KERN_WARNING DRV_NAME
1772                                ": %s: Warning: clearing HW address of %s while it "
1773                                "still has VLANs.\n",
1774                                bond_dev->name, bond_dev->name);
1775                         printk(KERN_WARNING DRV_NAME
1776                                ": %s: When re-adding slaves, make sure the bond's "
1777                                "HW address matches its VLANs'.\n",
1778                                bond_dev->name);
1779                 }
1780         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1781                    !bond_has_challenged_slaves(bond)) {
1782                 printk(KERN_INFO DRV_NAME
1783                        ": %s: last VLAN challenged slave %s "
1784                        "left bond %s. VLAN blocking is removed\n",
1785                        bond_dev->name, slave_dev->name, bond_dev->name);
1786                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1787         }
1788
1789         write_unlock_bh(&bond->lock);
1790
1791         /* must do this from outside any spinlocks */
1792         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1793
1794         bond_del_vlans_from_slave(bond, slave_dev);
1795
1796         /* If the mode USES_PRIMARY, then we should only remove its
1797          * promisc and mc settings if it was the curr_active_slave, but that was
1798          * already taken care of above when we detached the slave
1799          */
1800         if (!USES_PRIMARY(bond->params.mode)) {
1801                 /* unset promiscuity level from slave */
1802                 if (bond_dev->flags & IFF_PROMISC) {
1803                         dev_set_promiscuity(slave_dev, -1);
1804                 }
1805
1806                 /* unset allmulti level from slave */
1807                 if (bond_dev->flags & IFF_ALLMULTI) {
1808                         dev_set_allmulti(slave_dev, -1);
1809                 }
1810
1811                 /* flush master's mc_list from slave */
1812                 bond_mc_list_flush(bond_dev, slave_dev);
1813         }
1814
1815         netdev_set_master(slave_dev, NULL);
1816
1817         /* close slave before restoring its mac address */
1818         dev_close(slave_dev);
1819
1820         if (!bond->params.fail_over_mac) {
1821                 /* restore original ("permanent") mac address */
1822                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1823                 addr.sa_family = slave_dev->type;
1824                 dev_set_mac_address(slave_dev, &addr);
1825         }
1826
1827         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1828                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1829                                    IFF_SLAVE_NEEDARP);
1830
1831         kfree(slave);
1832
1833         return 0;  /* deletion OK */
1834 }
1835
1836 /*
1837 * Destroy a bonding device.
1838 * Must be under rtnl_lock when this function is called.
1839 */
1840 void bond_destroy(struct bonding *bond)
1841 {
1842         bond_deinit(bond->dev);
1843         bond_destroy_sysfs_entry(bond);
1844         unregister_netdevice(bond->dev);
1845 }
1846
1847 /*
1848 * First release a slave and than destroy the bond if no more slaves iare left.
1849 * Must be under rtnl_lock when this function is called.
1850 */
1851 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1852 {
1853         struct bonding *bond = bond_dev->priv;
1854         int ret;
1855
1856         ret = bond_release(bond_dev, slave_dev);
1857         if ((ret == 0) && (bond->slave_cnt == 0)) {
1858                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1859                        bond_dev->name, bond_dev->name);
1860                 bond_destroy(bond);
1861         }
1862         return ret;
1863 }
1864
1865 /*
1866  * This function releases all slaves.
1867  */
1868 static int bond_release_all(struct net_device *bond_dev)
1869 {
1870         struct bonding *bond = bond_dev->priv;
1871         struct slave *slave;
1872         struct net_device *slave_dev;
1873         struct sockaddr addr;
1874
1875         write_lock_bh(&bond->lock);
1876
1877         netif_carrier_off(bond_dev);
1878
1879         if (bond->slave_cnt == 0) {
1880                 goto out;
1881         }
1882
1883         bond->current_arp_slave = NULL;
1884         bond->primary_slave = NULL;
1885         bond_change_active_slave(bond, NULL);
1886
1887         while ((slave = bond->first_slave) != NULL) {
1888                 /* Inform AD package of unbinding of slave
1889                  * before slave is detached from the list.
1890                  */
1891                 if (bond->params.mode == BOND_MODE_8023AD) {
1892                         bond_3ad_unbind_slave(slave);
1893                 }
1894
1895                 slave_dev = slave->dev;
1896                 bond_detach_slave(bond, slave);
1897
1898                 if ((bond->params.mode == BOND_MODE_TLB) ||
1899                     (bond->params.mode == BOND_MODE_ALB)) {
1900                         /* must be called only after the slave
1901                          * has been detached from the list
1902                          */
1903                         bond_alb_deinit_slave(bond, slave);
1904                 }
1905
1906                 bond_compute_features(bond);
1907
1908                 /* now that the slave is detached, unlock and perform
1909                  * all the undo steps that should not be called from
1910                  * within a lock.
1911                  */
1912                 write_unlock_bh(&bond->lock);
1913
1914                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1915                 bond_del_vlans_from_slave(bond, slave_dev);
1916
1917                 /* If the mode USES_PRIMARY, then we should only remove its
1918                  * promisc and mc settings if it was the curr_active_slave, but that was
1919                  * already taken care of above when we detached the slave
1920                  */
1921                 if (!USES_PRIMARY(bond->params.mode)) {
1922                         /* unset promiscuity level from slave */
1923                         if (bond_dev->flags & IFF_PROMISC) {
1924                                 dev_set_promiscuity(slave_dev, -1);
1925                         }
1926
1927                         /* unset allmulti level from slave */
1928                         if (bond_dev->flags & IFF_ALLMULTI) {
1929                                 dev_set_allmulti(slave_dev, -1);
1930                         }
1931
1932                         /* flush master's mc_list from slave */
1933                         bond_mc_list_flush(bond_dev, slave_dev);
1934                 }
1935
1936                 netdev_set_master(slave_dev, NULL);
1937
1938                 /* close slave before restoring its mac address */
1939                 dev_close(slave_dev);
1940
1941                 if (!bond->params.fail_over_mac) {
1942                         /* restore original ("permanent") mac address*/
1943                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1944                         addr.sa_family = slave_dev->type;
1945                         dev_set_mac_address(slave_dev, &addr);
1946                 }
1947
1948                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1949                                            IFF_SLAVE_INACTIVE);
1950
1951                 kfree(slave);
1952
1953                 /* re-acquire the lock before getting the next slave */
1954                 write_lock_bh(&bond->lock);
1955         }
1956
1957         /* zero the mac address of the master so it will be
1958          * set by the application to the mac address of the
1959          * first slave
1960          */
1961         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1962
1963         if (list_empty(&bond->vlan_list)) {
1964                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1965         } else {
1966                 printk(KERN_WARNING DRV_NAME
1967                        ": %s: Warning: clearing HW address of %s while it "
1968                        "still has VLANs.\n",
1969                        bond_dev->name, bond_dev->name);
1970                 printk(KERN_WARNING DRV_NAME
1971                        ": %s: When re-adding slaves, make sure the bond's "
1972                        "HW address matches its VLANs'.\n",
1973                        bond_dev->name);
1974         }
1975
1976         printk(KERN_INFO DRV_NAME
1977                ": %s: released all slaves\n",
1978                bond_dev->name);
1979
1980 out:
1981         write_unlock_bh(&bond->lock);
1982
1983         return 0;
1984 }
1985
1986 /*
1987  * This function changes the active slave to slave <slave_dev>.
1988  * It returns -EINVAL in the following cases.
1989  *  - <slave_dev> is not found in the list.
1990  *  - There is not active slave now.
1991  *  - <slave_dev> is already active.
1992  *  - The link state of <slave_dev> is not BOND_LINK_UP.
1993  *  - <slave_dev> is not running.
1994  * In these cases, this fuction does nothing.
1995  * In the other cases, currnt_slave pointer is changed and 0 is returned.
1996  */
1997 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1998 {
1999         struct bonding *bond = bond_dev->priv;
2000         struct slave *old_active = NULL;
2001         struct slave *new_active = NULL;
2002         int res = 0;
2003
2004         if (!USES_PRIMARY(bond->params.mode)) {
2005                 return -EINVAL;
2006         }
2007
2008         /* Verify that master_dev is indeed the master of slave_dev */
2009         if (!(slave_dev->flags & IFF_SLAVE) ||
2010             (slave_dev->master != bond_dev)) {
2011                 return -EINVAL;
2012         }
2013
2014         write_lock_bh(&bond->lock);
2015
2016         old_active = bond->curr_active_slave;
2017         new_active = bond_get_slave_by_dev(bond, slave_dev);
2018
2019         /*
2020          * Changing to the current active: do nothing; return success.
2021          */
2022         if (new_active && (new_active == old_active)) {
2023                 write_unlock_bh(&bond->lock);
2024                 return 0;
2025         }
2026
2027         if ((new_active) &&
2028             (old_active) &&
2029             (new_active->link == BOND_LINK_UP) &&
2030             IS_UP(new_active->dev)) {
2031                 bond_change_active_slave(bond, new_active);
2032         } else {
2033                 res = -EINVAL;
2034         }
2035
2036         write_unlock_bh(&bond->lock);
2037
2038         return res;
2039 }
2040
2041 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2042 {
2043         struct bonding *bond = bond_dev->priv;
2044
2045         info->bond_mode = bond->params.mode;
2046         info->miimon = bond->params.miimon;
2047
2048         read_lock_bh(&bond->lock);
2049         info->num_slaves = bond->slave_cnt;
2050         read_unlock_bh(&bond->lock);
2051
2052         return 0;
2053 }
2054
2055 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2056 {
2057         struct bonding *bond = bond_dev->priv;
2058         struct slave *slave;
2059         int i, found = 0;
2060
2061         if (info->slave_id < 0) {
2062                 return -ENODEV;
2063         }
2064
2065         read_lock_bh(&bond->lock);
2066
2067         bond_for_each_slave(bond, slave, i) {
2068                 if (i == (int)info->slave_id) {
2069                         found = 1;
2070                         break;
2071                 }
2072         }
2073
2074         read_unlock_bh(&bond->lock);
2075
2076         if (found) {
2077                 strcpy(info->slave_name, slave->dev->name);
2078                 info->link = slave->link;
2079                 info->state = slave->state;
2080                 info->link_failure_count = slave->link_failure_count;
2081         } else {
2082                 return -ENODEV;
2083         }
2084
2085         return 0;
2086 }
2087
2088 /*-------------------------------- Monitoring -------------------------------*/
2089
2090 /* this function is called regularly to monitor each slave's link. */
2091 void bond_mii_monitor(struct net_device *bond_dev)
2092 {
2093         struct bonding *bond = bond_dev->priv;
2094         struct slave *slave, *oldcurrent;
2095         int do_failover = 0;
2096         int delta_in_ticks;
2097         int i;
2098
2099         read_lock(&bond->lock);
2100
2101         delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2102
2103         if (bond->kill_timers) {
2104                 goto out;
2105         }
2106
2107         if (bond->slave_cnt == 0) {
2108                 goto re_arm;
2109         }
2110
2111         /* we will try to read the link status of each of our slaves, and
2112          * set their IFF_RUNNING flag appropriately. For each slave not
2113          * supporting MII status, we won't do anything so that a user-space
2114          * program could monitor the link itself if needed.
2115          */
2116
2117         if (bond->send_grat_arp) {
2118                 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2119                                 &bond->curr_active_slave->dev->state))
2120                         dprintk("Needs to send gratuitous arp but not yet\n");
2121                 else {
2122                         dprintk("sending delayed gratuitous arp on on %s\n",
2123                                 bond->curr_active_slave->dev->name);
2124                         bond_send_gratuitous_arp(bond);
2125                         bond->send_grat_arp = 0;
2126                 }
2127         }
2128         read_lock(&bond->curr_slave_lock);
2129         oldcurrent = bond->curr_active_slave;
2130         read_unlock(&bond->curr_slave_lock);
2131
2132         bond_for_each_slave(bond, slave, i) {
2133                 struct net_device *slave_dev = slave->dev;
2134                 int link_state;
2135                 u16 old_speed = slave->speed;
2136                 u8 old_duplex = slave->duplex;
2137
2138                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2139
2140                 switch (slave->link) {
2141                 case BOND_LINK_UP:      /* the link was up */
2142                         if (link_state == BMSR_LSTATUS) {
2143                                 /* link stays up, nothing more to do */
2144                                 break;
2145                         } else { /* link going down */
2146                                 slave->link  = BOND_LINK_FAIL;
2147                                 slave->delay = bond->params.downdelay;
2148
2149                                 if (slave->link_failure_count < UINT_MAX) {
2150                                         slave->link_failure_count++;
2151                                 }
2152
2153                                 if (bond->params.downdelay) {
2154                                         printk(KERN_INFO DRV_NAME
2155                                                ": %s: link status down for %s "
2156                                                "interface %s, disabling it in "
2157                                                "%d ms.\n",
2158                                                bond_dev->name,
2159                                                IS_UP(slave_dev)
2160                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2161                                                   ? ((slave == oldcurrent)
2162                                                      ? "active " : "backup ")
2163                                                   : "")
2164                                                : "idle ",
2165                                                slave_dev->name,
2166                                                bond->params.downdelay * bond->params.miimon);
2167                                 }
2168                         }
2169                         /* no break ! fall through the BOND_LINK_FAIL test to
2170                            ensure proper action to be taken
2171                         */
2172                 case BOND_LINK_FAIL:    /* the link has just gone down */
2173                         if (link_state != BMSR_LSTATUS) {
2174                                 /* link stays down */
2175                                 if (slave->delay <= 0) {
2176                                         /* link down for too long time */
2177                                         slave->link = BOND_LINK_DOWN;
2178
2179                                         /* in active/backup mode, we must
2180                                          * completely disable this interface
2181                                          */
2182                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2183                                             (bond->params.mode == BOND_MODE_8023AD)) {
2184                                                 bond_set_slave_inactive_flags(slave);
2185                                         }
2186
2187                                         printk(KERN_INFO DRV_NAME
2188                                                ": %s: link status definitely "
2189                                                "down for interface %s, "
2190                                                "disabling it\n",
2191                                                bond_dev->name,
2192                                                slave_dev->name);
2193
2194                                         /* notify ad that the link status has changed */
2195                                         if (bond->params.mode == BOND_MODE_8023AD) {
2196                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2197                                         }
2198
2199                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2200                                             (bond->params.mode == BOND_MODE_ALB)) {
2201                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2202                                         }
2203
2204                                         if (slave == oldcurrent) {
2205                                                 do_failover = 1;
2206                                         }
2207                                 } else {
2208                                         slave->delay--;
2209                                 }
2210                         } else {
2211                                 /* link up again */
2212                                 slave->link  = BOND_LINK_UP;
2213                                 slave->jiffies = jiffies;
2214                                 printk(KERN_INFO DRV_NAME
2215                                        ": %s: link status up again after %d "
2216                                        "ms for interface %s.\n",
2217                                        bond_dev->name,
2218                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2219                                        slave_dev->name);
2220                         }
2221                         break;
2222                 case BOND_LINK_DOWN:    /* the link was down */
2223                         if (link_state != BMSR_LSTATUS) {
2224                                 /* the link stays down, nothing more to do */
2225                                 break;
2226                         } else {        /* link going up */
2227                                 slave->link  = BOND_LINK_BACK;
2228                                 slave->delay = bond->params.updelay;
2229
2230                                 if (bond->params.updelay) {
2231                                         /* if updelay == 0, no need to
2232                                            advertise about a 0 ms delay */
2233                                         printk(KERN_INFO DRV_NAME
2234                                                ": %s: link status up for "
2235                                                "interface %s, enabling it "
2236                                                "in %d ms.\n",
2237                                                bond_dev->name,
2238                                                slave_dev->name,
2239                                                bond->params.updelay * bond->params.miimon);
2240                                 }
2241                         }
2242                         /* no break ! fall through the BOND_LINK_BACK state in
2243                            case there's something to do.
2244                         */
2245                 case BOND_LINK_BACK:    /* the link has just come back */
2246                         if (link_state != BMSR_LSTATUS) {
2247                                 /* link down again */
2248                                 slave->link  = BOND_LINK_DOWN;
2249
2250                                 printk(KERN_INFO DRV_NAME
2251                                        ": %s: link status down again after %d "
2252                                        "ms for interface %s.\n",
2253                                        bond_dev->name,
2254                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2255                                        slave_dev->name);
2256                         } else {
2257                                 /* link stays up */
2258                                 if (slave->delay == 0) {
2259                                         /* now the link has been up for long time enough */
2260                                         slave->link = BOND_LINK_UP;
2261                                         slave->jiffies = jiffies;
2262
2263                                         if (bond->params.mode == BOND_MODE_8023AD) {
2264                                                 /* prevent it from being the active one */
2265                                                 slave->state = BOND_STATE_BACKUP;
2266                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2267                                                 /* make it immediately active */
2268                                                 slave->state = BOND_STATE_ACTIVE;
2269                                         } else if (slave != bond->primary_slave) {
2270                                                 /* prevent it from being the active one */
2271                                                 slave->state = BOND_STATE_BACKUP;
2272                                         }
2273
2274                                         printk(KERN_INFO DRV_NAME
2275                                                ": %s: link status definitely "
2276                                                "up for interface %s.\n",
2277                                                bond_dev->name,
2278                                                slave_dev->name);
2279
2280                                         /* notify ad that the link status has changed */
2281                                         if (bond->params.mode == BOND_MODE_8023AD) {
2282                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2283                                         }
2284
2285                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2286                                             (bond->params.mode == BOND_MODE_ALB)) {
2287                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2288                                         }
2289
2290                                         if ((!oldcurrent) ||
2291                                             (slave == bond->primary_slave)) {
2292                                                 do_failover = 1;
2293                                         }
2294                                 } else {
2295                                         slave->delay--;
2296                                 }
2297                         }
2298                         break;
2299                 default:
2300                         /* Should not happen */
2301                         printk(KERN_ERR DRV_NAME
2302                                ": %s: Error: %s Illegal value (link=%d)\n",
2303                                bond_dev->name,
2304                                slave->dev->name,
2305                                slave->link);
2306                         goto out;
2307                 } /* end of switch (slave->link) */
2308
2309                 bond_update_speed_duplex(slave);
2310
2311                 if (bond->params.mode == BOND_MODE_8023AD) {
2312                         if (old_speed != slave->speed) {
2313                                 bond_3ad_adapter_speed_changed(slave);
2314                         }
2315
2316                         if (old_duplex != slave->duplex) {
2317                                 bond_3ad_adapter_duplex_changed(slave);
2318                         }
2319                 }
2320
2321         } /* end of for */
2322
2323         if (do_failover) {
2324                 write_lock(&bond->curr_slave_lock);
2325
2326                 bond_select_active_slave(bond);
2327
2328                 write_unlock(&bond->curr_slave_lock);
2329         } else
2330                 bond_set_carrier(bond);
2331
2332 re_arm:
2333         if (bond->params.miimon) {
2334                 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2335         }
2336 out:
2337         read_unlock(&bond->lock);
2338 }
2339
2340
2341 static __be32 bond_glean_dev_ip(struct net_device *dev)
2342 {
2343         struct in_device *idev;
2344         struct in_ifaddr *ifa;
2345         __be32 addr = 0;
2346
2347         if (!dev)
2348                 return 0;
2349
2350         rcu_read_lock();
2351         idev = __in_dev_get_rcu(dev);
2352         if (!idev)
2353                 goto out;
2354
2355         ifa = idev->ifa_list;
2356         if (!ifa)
2357                 goto out;
2358
2359         addr = ifa->ifa_local;
2360 out:
2361         rcu_read_unlock();
2362         return addr;
2363 }
2364
2365 static int bond_has_ip(struct bonding *bond)
2366 {
2367         struct vlan_entry *vlan, *vlan_next;
2368
2369         if (bond->master_ip)
2370                 return 1;
2371
2372         if (list_empty(&bond->vlan_list))
2373                 return 0;
2374
2375         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2376                                  vlan_list) {
2377                 if (vlan->vlan_ip)
2378                         return 1;
2379         }
2380
2381         return 0;
2382 }
2383
2384 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2385 {
2386         struct vlan_entry *vlan, *vlan_next;
2387
2388         if (ip == bond->master_ip)
2389                 return 1;
2390
2391         if (list_empty(&bond->vlan_list))
2392                 return 0;
2393
2394         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2395                                  vlan_list) {
2396                 if (ip == vlan->vlan_ip)
2397                         return 1;
2398         }
2399
2400         return 0;
2401 }
2402
2403 /*
2404  * We go to the (large) trouble of VLAN tagging ARP frames because
2405  * switches in VLAN mode (especially if ports are configured as
2406  * "native" to a VLAN) might not pass non-tagged frames.
2407  */
2408 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2409 {
2410         struct sk_buff *skb;
2411
2412         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2413                slave_dev->name, dest_ip, src_ip, vlan_id);
2414                
2415         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2416                          NULL, slave_dev->dev_addr, NULL);
2417
2418         if (!skb) {
2419                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2420                 return;
2421         }
2422         if (vlan_id) {
2423                 skb = vlan_put_tag(skb, vlan_id);
2424                 if (!skb) {
2425                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2426                         return;
2427                 }
2428         }
2429         arp_xmit(skb);
2430 }
2431
2432
2433 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2434 {
2435         int i, vlan_id, rv;
2436         __be32 *targets = bond->params.arp_targets;
2437         struct vlan_entry *vlan, *vlan_next;
2438         struct net_device *vlan_dev;
2439         struct flowi fl;
2440         struct rtable *rt;
2441
2442         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2443                 if (!targets[i])
2444                         continue;
2445                 dprintk("basa: target %x\n", targets[i]);
2446                 if (list_empty(&bond->vlan_list)) {
2447                         dprintk("basa: empty vlan: arp_send\n");
2448                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2449                                       bond->master_ip, 0);
2450                         continue;
2451                 }
2452
2453                 /*
2454                  * If VLANs are configured, we do a route lookup to
2455                  * determine which VLAN interface would be used, so we
2456                  * can tag the ARP with the proper VLAN tag.
2457                  */
2458                 memset(&fl, 0, sizeof(fl));
2459                 fl.fl4_dst = targets[i];
2460                 fl.fl4_tos = RTO_ONLINK;
2461
2462                 rv = ip_route_output_key(&rt, &fl);
2463                 if (rv) {
2464                         if (net_ratelimit()) {
2465                                 printk(KERN_WARNING DRV_NAME
2466                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2467                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2468                         }
2469                         continue;
2470                 }
2471
2472                 /*
2473                  * This target is not on a VLAN
2474                  */
2475                 if (rt->u.dst.dev == bond->dev) {
2476                         ip_rt_put(rt);
2477                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2478                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2479                                       bond->master_ip, 0);
2480                         continue;
2481                 }
2482
2483                 vlan_id = 0;
2484                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2485                                          vlan_list) {
2486                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2487                         if (vlan_dev == rt->u.dst.dev) {
2488                                 vlan_id = vlan->vlan_id;
2489                                 dprintk("basa: vlan match on %s %d\n",
2490                                        vlan_dev->name, vlan_id);
2491                                 break;
2492                         }
2493                 }
2494
2495                 if (vlan_id) {
2496                         ip_rt_put(rt);
2497                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2498                                       vlan->vlan_ip, vlan_id);
2499                         continue;
2500                 }
2501
2502                 if (net_ratelimit()) {
2503                         printk(KERN_WARNING DRV_NAME
2504                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2505                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2506                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2507                 }
2508                 ip_rt_put(rt);
2509         }
2510 }
2511
2512 /*
2513  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2514  * for each VLAN above us.
2515  */
2516 static void bond_send_gratuitous_arp(struct bonding *bond)
2517 {
2518         struct slave *slave = bond->curr_active_slave;
2519         struct vlan_entry *vlan;
2520         struct net_device *vlan_dev;
2521
2522         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2523                                 slave ? slave->dev->name : "NULL");
2524         if (!slave)
2525                 return;
2526
2527         if (bond->master_ip) {
2528                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2529                                 bond->master_ip, 0);
2530         }
2531
2532         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2533                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2534                 if (vlan->vlan_ip) {
2535                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2536                                       vlan->vlan_ip, vlan->vlan_id);
2537                 }
2538         }
2539 }
2540
2541 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2542 {
2543         int i;
2544         __be32 *targets = bond->params.arp_targets;
2545
2546         targets = bond->params.arp_targets;
2547         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2548                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2549                         "%u.%u.%u.%u bhti(tip) %d\n",
2550                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2551                        bond_has_this_ip(bond, tip));
2552                 if (sip == targets[i]) {
2553                         if (bond_has_this_ip(bond, tip))
2554                                 slave->last_arp_rx = jiffies;
2555                         return;
2556                 }
2557         }
2558 }
2559
2560 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2561 {
2562         struct arphdr *arp;
2563         struct slave *slave;
2564         struct bonding *bond;
2565         unsigned char *arp_ptr;
2566         __be32 sip, tip;
2567
2568         if (dev->nd_net != &init_net)
2569                 goto out;
2570
2571         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2572                 goto out;
2573
2574         bond = dev->priv;
2575         read_lock(&bond->lock);
2576
2577         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2578                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2579                 orig_dev ? orig_dev->name : "NULL");
2580
2581         slave = bond_get_slave_by_dev(bond, orig_dev);
2582         if (!slave || !slave_do_arp_validate(bond, slave))
2583                 goto out_unlock;
2584
2585         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
2586         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2587                                  (2 * dev->addr_len) +
2588                                  (2 * sizeof(u32)))))
2589                 goto out_unlock;
2590
2591         arp = arp_hdr(skb);
2592         if (arp->ar_hln != dev->addr_len ||
2593             skb->pkt_type == PACKET_OTHERHOST ||
2594             skb->pkt_type == PACKET_LOOPBACK ||
2595             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2596             arp->ar_pro != htons(ETH_P_IP) ||
2597             arp->ar_pln != 4)
2598                 goto out_unlock;
2599
2600         arp_ptr = (unsigned char *)(arp + 1);
2601         arp_ptr += dev->addr_len;
2602         memcpy(&sip, arp_ptr, 4);
2603         arp_ptr += 4 + dev->addr_len;
2604         memcpy(&tip, arp_ptr, 4);
2605
2606         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2607                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2608                 slave->state, bond->params.arp_validate,
2609                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2610
2611         /*
2612          * Backup slaves won't see the ARP reply, but do come through
2613          * here for each ARP probe (so we swap the sip/tip to validate
2614          * the probe).  In a "redundant switch, common router" type of
2615          * configuration, the ARP probe will (hopefully) travel from
2616          * the active, through one switch, the router, then the other
2617          * switch before reaching the backup.
2618          */
2619         if (slave->state == BOND_STATE_ACTIVE)
2620                 bond_validate_arp(bond, slave, sip, tip);
2621         else
2622                 bond_validate_arp(bond, slave, tip, sip);
2623
2624 out_unlock:
2625         read_unlock(&bond->lock);
2626 out:
2627         dev_kfree_skb(skb);
2628         return NET_RX_SUCCESS;
2629 }
2630
2631 /*
2632  * this function is called regularly to monitor each slave's link
2633  * ensuring that traffic is being sent and received when arp monitoring
2634  * is used in load-balancing mode. if the adapter has been dormant, then an
2635  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2636  * arp monitoring in active backup mode.
2637  */
2638 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2639 {
2640         struct bonding *bond = bond_dev->priv;
2641         struct slave *slave, *oldcurrent;
2642         int do_failover = 0;
2643         int delta_in_ticks;
2644         int i;
2645
2646         read_lock(&bond->lock);
2647
2648         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2649
2650         if (bond->kill_timers) {
2651                 goto out;
2652         }
2653
2654         if (bond->slave_cnt == 0) {
2655                 goto re_arm;
2656         }
2657
2658         read_lock(&bond->curr_slave_lock);
2659         oldcurrent = bond->curr_active_slave;
2660         read_unlock(&bond->curr_slave_lock);
2661
2662         /* see if any of the previous devices are up now (i.e. they have
2663          * xmt and rcv traffic). the curr_active_slave does not come into
2664          * the picture unless it is null. also, slave->jiffies is not needed
2665          * here because we send an arp on each slave and give a slave as
2666          * long as it needs to get the tx/rx within the delta.
2667          * TODO: what about up/down delay in arp mode? it wasn't here before
2668          *       so it can wait
2669          */
2670         bond_for_each_slave(bond, slave, i) {
2671                 if (slave->link != BOND_LINK_UP) {
2672                         if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2673                             ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2674
2675                                 slave->link  = BOND_LINK_UP;
2676                                 slave->state = BOND_STATE_ACTIVE;
2677
2678                                 /* primary_slave has no meaning in round-robin
2679                                  * mode. the window of a slave being up and
2680                                  * curr_active_slave being null after enslaving
2681                                  * is closed.
2682                                  */
2683                                 if (!oldcurrent) {
2684                                         printk(KERN_INFO DRV_NAME
2685                                                ": %s: link status definitely "
2686                                                "up for interface %s, ",
2687                                                bond_dev->name,
2688                                                slave->dev->name);
2689                                         do_failover = 1;
2690                                 } else {
2691                                         printk(KERN_INFO DRV_NAME
2692                                                ": %s: interface %s is now up\n",
2693                                                bond_dev->name,
2694                                                slave->dev->name);
2695                                 }
2696                         }
2697                 } else {
2698                         /* slave->link == BOND_LINK_UP */
2699
2700                         /* not all switches will respond to an arp request
2701                          * when the source ip is 0, so don't take the link down
2702                          * if we don't know our ip yet
2703                          */
2704                         if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2705                             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2706                              bond_has_ip(bond))) {
2707
2708                                 slave->link  = BOND_LINK_DOWN;
2709                                 slave->state = BOND_STATE_BACKUP;
2710
2711                                 if (slave->link_failure_count < UINT_MAX) {
2712                                         slave->link_failure_count++;
2713                                 }
2714
2715                                 printk(KERN_INFO DRV_NAME
2716                                        ": %s: interface %s is now down.\n",
2717                                        bond_dev->name,
2718                                        slave->dev->name);
2719
2720                                 if (slave == oldcurrent) {
2721                                         do_failover = 1;
2722                                 }
2723                         }
2724                 }
2725
2726                 /* note: if switch is in round-robin mode, all links
2727                  * must tx arp to ensure all links rx an arp - otherwise
2728                  * links may oscillate or not come up at all; if switch is
2729                  * in something like xor mode, there is nothing we can
2730                  * do - all replies will be rx'ed on same link causing slaves
2731                  * to be unstable during low/no traffic periods
2732                  */
2733                 if (IS_UP(slave->dev)) {
2734                         bond_arp_send_all(bond, slave);
2735                 }
2736         }
2737
2738         if (do_failover) {
2739                 write_lock(&bond->curr_slave_lock);
2740
2741                 bond_select_active_slave(bond);
2742
2743                 write_unlock(&bond->curr_slave_lock);
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 /*
2755  * When using arp monitoring in active-backup mode, this function is
2756  * called to determine if any backup slaves have went down or a new
2757  * current slave needs to be found.
2758  * The backup slaves never generate traffic, they are considered up by merely
2759  * receiving traffic. If the current slave goes down, each backup slave will
2760  * be given the opportunity to tx/rx an arp before being taken down - this
2761  * prevents all slaves from being taken down due to the current slave not
2762  * sending any traffic for the backups to receive. The arps are not necessarily
2763  * necessary, any tx and rx traffic will keep the current slave up. While any
2764  * rx traffic will keep the backup slaves up, the current slave is responsible
2765  * for generating traffic to keep them up regardless of any other traffic they
2766  * may have received.
2767  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2768  */
2769 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2770 {
2771         struct bonding *bond = bond_dev->priv;
2772         struct slave *slave;
2773         int delta_in_ticks;
2774         int i;
2775
2776         read_lock(&bond->lock);
2777
2778         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2779
2780         if (bond->kill_timers) {
2781                 goto out;
2782         }
2783
2784         if (bond->slave_cnt == 0) {
2785                 goto re_arm;
2786         }
2787
2788         /* determine if any slave has come up or any backup slave has
2789          * gone down
2790          * TODO: what about up/down delay in arp mode? it wasn't here before
2791          *       so it can wait
2792          */
2793         bond_for_each_slave(bond, slave, i) {
2794                 if (slave->link != BOND_LINK_UP) {
2795                         if ((jiffies - slave_last_rx(bond, slave)) <=
2796                              delta_in_ticks) {
2797
2798                                 slave->link = BOND_LINK_UP;
2799
2800                                 write_lock(&bond->curr_slave_lock);
2801
2802                                 if ((!bond->curr_active_slave) &&
2803                                     ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2804                                         bond_change_active_slave(bond, slave);
2805                                         bond->current_arp_slave = NULL;
2806                                 } else if (bond->curr_active_slave != slave) {
2807                                         /* this slave has just come up but we
2808                                          * already have a current slave; this
2809                                          * can also happen if bond_enslave adds
2810                                          * a new slave that is up while we are
2811                                          * searching for a new slave
2812                                          */
2813                                         bond_set_slave_inactive_flags(slave);
2814                                         bond->current_arp_slave = NULL;
2815                                 }
2816
2817                                 bond_set_carrier(bond);
2818
2819                                 if (slave == bond->curr_active_slave) {
2820                                         printk(KERN_INFO DRV_NAME
2821                                                ": %s: %s is up and now the "
2822                                                "active interface\n",
2823                                                bond_dev->name,
2824                                                slave->dev->name);
2825                                         netif_carrier_on(bond->dev);
2826                                 } else {
2827                                         printk(KERN_INFO DRV_NAME
2828                                                ": %s: backup interface %s is "
2829                                                "now up\n",
2830                                                bond_dev->name,
2831                                                slave->dev->name);
2832                                 }
2833
2834                                 write_unlock(&bond->curr_slave_lock);
2835                         }
2836                 } else {
2837                         read_lock(&bond->curr_slave_lock);
2838
2839                         if ((slave != bond->curr_active_slave) &&
2840                             (!bond->current_arp_slave) &&
2841                             (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2842                              bond_has_ip(bond))) {
2843                                 /* a backup slave has gone down; three times
2844                                  * the delta allows the current slave to be
2845                                  * taken out before the backup slave.
2846                                  * note: a non-null current_arp_slave indicates
2847                                  * the curr_active_slave went down and we are
2848                                  * searching for a new one; under this
2849                                  * condition we only take the curr_active_slave
2850                                  * down - this gives each slave a chance to
2851                                  * tx/rx traffic before being taken out
2852                                  */
2853
2854                                 read_unlock(&bond->curr_slave_lock);
2855
2856                                 slave->link  = BOND_LINK_DOWN;
2857
2858                                 if (slave->link_failure_count < UINT_MAX) {
2859                                         slave->link_failure_count++;
2860                                 }
2861
2862                                 bond_set_slave_inactive_flags(slave);
2863
2864                                 printk(KERN_INFO DRV_NAME
2865                                        ": %s: backup interface %s is now down\n",
2866                                        bond_dev->name,
2867                                        slave->dev->name);
2868                         } else {
2869                                 read_unlock(&bond->curr_slave_lock);
2870                         }
2871                 }
2872         }
2873
2874         read_lock(&bond->curr_slave_lock);
2875         slave = bond->curr_active_slave;
2876         read_unlock(&bond->curr_slave_lock);
2877
2878         if (slave) {
2879                 /* if we have sent traffic in the past 2*arp_intervals but
2880                  * haven't xmit and rx traffic in that time interval, select
2881                  * a different slave. slave->jiffies is only updated when
2882                  * a slave first becomes the curr_active_slave - not necessarily
2883                  * after every arp; this ensures the slave has a full 2*delta
2884                  * before being taken out. if a primary is being used, check
2885                  * if it is up and needs to take over as the curr_active_slave
2886                  */
2887                 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2888             (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2889              bond_has_ip(bond))) &&
2890                     ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2891
2892                         slave->link  = BOND_LINK_DOWN;
2893
2894                         if (slave->link_failure_count < UINT_MAX) {
2895                                 slave->link_failure_count++;
2896                         }
2897
2898                         printk(KERN_INFO DRV_NAME
2899                                ": %s: link status down for active interface "
2900                                "%s, disabling it\n",
2901                                bond_dev->name,
2902                                slave->dev->name);
2903
2904                         write_lock(&bond->curr_slave_lock);
2905
2906                         bond_select_active_slave(bond);
2907                         slave = bond->curr_active_slave;
2908
2909                         write_unlock(&bond->curr_slave_lock);
2910
2911                         bond->current_arp_slave = slave;
2912
2913                         if (slave) {
2914                                 slave->jiffies = jiffies;
2915                         }
2916                 } else if ((bond->primary_slave) &&
2917                            (bond->primary_slave != slave) &&
2918                            (bond->primary_slave->link == BOND_LINK_UP)) {
2919                         /* at this point, slave is the curr_active_slave */
2920                         printk(KERN_INFO DRV_NAME
2921                                ": %s: changing from interface %s to primary "
2922                                "interface %s\n",
2923                                bond_dev->name,
2924                                slave->dev->name,
2925                                bond->primary_slave->dev->name);
2926
2927                         /* primary is up so switch to it */
2928                         write_lock(&bond->curr_slave_lock);
2929                         bond_change_active_slave(bond, bond->primary_slave);
2930                         write_unlock(&bond->curr_slave_lock);
2931
2932                         slave = bond->primary_slave;
2933                         slave->jiffies = jiffies;
2934                 } else {
2935                         bond->current_arp_slave = NULL;
2936                 }
2937
2938                 /* the current slave must tx an arp to ensure backup slaves
2939                  * rx traffic
2940                  */
2941                 if (slave && bond_has_ip(bond)) {
2942                         bond_arp_send_all(bond, slave);
2943                 }
2944         }
2945
2946         /* if we don't have a curr_active_slave, search for the next available
2947          * backup slave from the current_arp_slave and make it the candidate
2948          * for becoming the curr_active_slave
2949          */
2950         if (!slave) {
2951                 if (!bond->current_arp_slave) {
2952                         bond->current_arp_slave = bond->first_slave;
2953                 }
2954
2955                 if (bond->current_arp_slave) {
2956                         bond_set_slave_inactive_flags(bond->current_arp_slave);
2957
2958                         /* search for next candidate */
2959                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2960                                 if (IS_UP(slave->dev)) {
2961                                         slave->link = BOND_LINK_BACK;
2962                                         bond_set_slave_active_flags(slave);
2963                                         bond_arp_send_all(bond, slave);
2964                                         slave->jiffies = jiffies;
2965                                         bond->current_arp_slave = slave;
2966                                         break;
2967                                 }
2968
2969                                 /* if the link state is up at this point, we
2970                                  * mark it down - this can happen if we have
2971                                  * simultaneous link failures and
2972                                  * reselect_active_interface doesn't make this
2973                                  * one the current slave so it is still marked
2974                                  * up when it is actually down
2975                                  */
2976                                 if (slave->link == BOND_LINK_UP) {
2977                                         slave->link  = BOND_LINK_DOWN;
2978                                         if (slave->link_failure_count < UINT_MAX) {
2979                                                 slave->link_failure_count++;
2980                                         }
2981
2982                                         bond_set_slave_inactive_flags(slave);
2983
2984                                         printk(KERN_INFO DRV_NAME
2985                                                ": %s: backup interface %s is "
2986                                                "now down.\n",
2987                                                bond_dev->name,
2988                                                slave->dev->name);
2989                                 }
2990                         }
2991                 }
2992         }
2993
2994 re_arm:
2995         if (bond->params.arp_interval) {
2996                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2997         }
2998 out:
2999         read_unlock(&bond->lock);
3000 }
3001
3002 /*------------------------------ proc/seq_file-------------------------------*/
3003
3004 #ifdef CONFIG_PROC_FS
3005
3006 #define SEQ_START_TOKEN ((void *)1)
3007
3008 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3009 {
3010         struct bonding *bond = seq->private;
3011         loff_t off = 0;
3012         struct slave *slave;
3013         int i;
3014
3015         /* make sure the bond won't be taken away */
3016         read_lock(&dev_base_lock);
3017         read_lock_bh(&bond->lock);
3018
3019         if (*pos == 0) {
3020                 return SEQ_START_TOKEN;
3021         }
3022
3023         bond_for_each_slave(bond, slave, i) {
3024                 if (++off == *pos) {
3025                         return slave;
3026                 }
3027         }
3028
3029         return NULL;
3030 }
3031
3032 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3033 {
3034         struct bonding *bond = seq->private;
3035         struct slave *slave = v;
3036
3037         ++*pos;
3038         if (v == SEQ_START_TOKEN) {
3039                 return bond->first_slave;
3040         }
3041
3042         slave = slave->next;
3043
3044         return (slave == bond->first_slave) ? NULL : slave;
3045 }
3046
3047 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3048 {
3049         struct bonding *bond = seq->private;
3050
3051         read_unlock_bh(&bond->lock);
3052         read_unlock(&dev_base_lock);
3053 }
3054
3055 static void bond_info_show_master(struct seq_file *seq)
3056 {
3057         struct bonding *bond = seq->private;
3058         struct slave *curr;
3059         int i;
3060         u32 target;
3061
3062         read_lock(&bond->curr_slave_lock);
3063         curr = bond->curr_active_slave;
3064         read_unlock(&bond->curr_slave_lock);
3065
3066         seq_printf(seq, "Bonding Mode: %s",
3067                    bond_mode_name(bond->params.mode));
3068
3069         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3070             bond->params.fail_over_mac)
3071                 seq_printf(seq, " (fail_over_mac)");
3072
3073         seq_printf(seq, "\n");
3074
3075         if (bond->params.mode == BOND_MODE_XOR ||
3076                 bond->params.mode == BOND_MODE_8023AD) {
3077                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3078                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3079                         bond->params.xmit_policy);
3080         }
3081
3082         if (USES_PRIMARY(bond->params.mode)) {
3083                 seq_printf(seq, "Primary Slave: %s\n",
3084                            (bond->primary_slave) ?
3085                            bond->primary_slave->dev->name : "None");
3086
3087                 seq_printf(seq, "Currently Active Slave: %s\n",
3088                            (curr) ? curr->dev->name : "None");
3089         }
3090
3091         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3092                    "up" : "down");
3093         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3094         seq_printf(seq, "Up Delay (ms): %d\n",
3095                    bond->params.updelay * bond->params.miimon);
3096         seq_printf(seq, "Down Delay (ms): %d\n",
3097                    bond->params.downdelay * bond->params.miimon);
3098
3099
3100         /* ARP information */
3101         if(bond->params.arp_interval > 0) {
3102                 int printed=0;
3103                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3104                                 bond->params.arp_interval);
3105
3106                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3107
3108                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3109                         if (!bond->params.arp_targets[i])
3110                                 continue;
3111                         if (printed)
3112                                 seq_printf(seq, ",");
3113                         target = ntohl(bond->params.arp_targets[i]);
3114                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3115                         printed = 1;
3116                 }
3117                 seq_printf(seq, "\n");
3118         }
3119
3120         if (bond->params.mode == BOND_MODE_8023AD) {
3121                 struct ad_info ad_info;
3122                 DECLARE_MAC_BUF(mac);
3123
3124                 seq_puts(seq, "\n802.3ad info\n");
3125                 seq_printf(seq, "LACP rate: %s\n",
3126                            (bond->params.lacp_fast) ? "fast" : "slow");
3127
3128                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3129                         seq_printf(seq, "bond %s has no active aggregator\n",
3130                                    bond->dev->name);
3131                 } else {
3132                         seq_printf(seq, "Active Aggregator Info:\n");
3133
3134                         seq_printf(seq, "\tAggregator ID: %d\n",
3135                                    ad_info.aggregator_id);
3136                         seq_printf(seq, "\tNumber of ports: %d\n",
3137                                    ad_info.ports);
3138                         seq_printf(seq, "\tActor Key: %d\n",
3139                                    ad_info.actor_key);
3140                         seq_printf(seq, "\tPartner Key: %d\n",
3141                                    ad_info.partner_key);
3142                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3143                                    print_mac(mac, ad_info.partner_system));
3144                 }
3145         }
3146 }
3147
3148 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3149 {
3150         struct bonding *bond = seq->private;
3151         DECLARE_MAC_BUF(mac);
3152
3153         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3154         seq_printf(seq, "MII Status: %s\n",
3155                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3156         seq_printf(seq, "Link Failure Count: %u\n",
3157                    slave->link_failure_count);
3158
3159         seq_printf(seq,
3160                    "Permanent HW addr: %s\n",
3161                    print_mac(mac, slave->perm_hwaddr));
3162
3163         if (bond->params.mode == BOND_MODE_8023AD) {
3164                 const struct aggregator *agg
3165                         = SLAVE_AD_INFO(slave).port.aggregator;
3166
3167                 if (agg) {
3168                         seq_printf(seq, "Aggregator ID: %d\n",
3169                                    agg->aggregator_identifier);
3170                 } else {
3171                         seq_puts(seq, "Aggregator ID: N/A\n");
3172                 }
3173         }
3174 }
3175
3176 static int bond_info_seq_show(struct seq_file *seq, void *v)
3177 {
3178         if (v == SEQ_START_TOKEN) {
3179                 seq_printf(seq, "%s\n", version);
3180                 bond_info_show_master(seq);
3181         } else {
3182                 bond_info_show_slave(seq, v);
3183         }
3184
3185         return 0;
3186 }
3187
3188 static struct seq_operations bond_info_seq_ops = {
3189         .start = bond_info_seq_start,
3190         .next  = bond_info_seq_next,
3191         .stop  = bond_info_seq_stop,
3192         .show  = bond_info_seq_show,
3193 };
3194
3195 static int bond_info_open(struct inode *inode, struct file *file)
3196 {
3197         struct seq_file *seq;
3198         struct proc_dir_entry *proc;
3199         int res;
3200
3201         res = seq_open(file, &bond_info_seq_ops);
3202         if (!res) {
3203                 /* recover the pointer buried in proc_dir_entry data */
3204                 seq = file->private_data;
3205                 proc = PDE(inode);
3206                 seq->private = proc->data;
3207         }
3208
3209         return res;
3210 }
3211
3212 static const struct file_operations bond_info_fops = {
3213         .owner   = THIS_MODULE,
3214         .open    = bond_info_open,
3215         .read    = seq_read,
3216         .llseek  = seq_lseek,
3217         .release = seq_release,
3218 };
3219
3220 static int bond_create_proc_entry(struct bonding *bond)
3221 {
3222         struct net_device *bond_dev = bond->dev;
3223
3224         if (bond_proc_dir) {
3225                 bond->proc_entry = create_proc_entry(bond_dev->name,
3226                                                      S_IRUGO,
3227                                                      bond_proc_dir);
3228                 if (bond->proc_entry == NULL) {
3229                         printk(KERN_WARNING DRV_NAME
3230                                ": Warning: Cannot create /proc/net/%s/%s\n",
3231                                DRV_NAME, bond_dev->name);
3232                 } else {
3233                         bond->proc_entry->data = bond;
3234                         bond->proc_entry->proc_fops = &bond_info_fops;
3235                         bond->proc_entry->owner = THIS_MODULE;
3236                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3237                 }
3238         }
3239
3240         return 0;
3241 }
3242
3243 static void bond_remove_proc_entry(struct bonding *bond)
3244 {
3245         if (bond_proc_dir && bond->proc_entry) {
3246                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3247                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3248                 bond->proc_entry = NULL;
3249         }
3250 }
3251
3252 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3253  * Caller must hold rtnl_lock.
3254  */
3255 static void bond_create_proc_dir(void)
3256 {
3257         int len = strlen(DRV_NAME);
3258
3259         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3260              bond_proc_dir = bond_proc_dir->next) {
3261                 if ((bond_proc_dir->namelen == len) &&
3262                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3263                         break;
3264                 }
3265         }
3266
3267         if (!bond_proc_dir) {
3268                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3269                 if (bond_proc_dir) {
3270                         bond_proc_dir->owner = THIS_MODULE;
3271                 } else {
3272                         printk(KERN_WARNING DRV_NAME
3273                                 ": Warning: cannot create /proc/net/%s\n",
3274                                 DRV_NAME);
3275                 }
3276         }
3277 }
3278
3279 /* Destroy the bonding directory under /proc/net, if empty.
3280  * Caller must hold rtnl_lock.
3281  */
3282 static void bond_destroy_proc_dir(void)
3283 {
3284         struct proc_dir_entry *de;
3285
3286         if (!bond_proc_dir) {
3287                 return;
3288         }
3289
3290         /* verify that the /proc dir is empty */
3291         for (de = bond_proc_dir->subdir; de; de = de->next) {
3292                 /* ignore . and .. */
3293                 if (*(de->name) != '.') {
3294                         break;
3295                 }
3296         }
3297
3298         if (de) {
3299                 if (bond_proc_dir->owner == THIS_MODULE) {
3300                         bond_proc_dir->owner = NULL;
3301                 }
3302         } else {
3303                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3304                 bond_proc_dir = NULL;
3305         }
3306 }
3307 #endif /* CONFIG_PROC_FS */
3308
3309 /*-------------------------- netdev event handling --------------------------*/
3310
3311 /*
3312  * Change device name
3313  */
3314 static int bond_event_changename(struct bonding *bond)
3315 {
3316 #ifdef CONFIG_PROC_FS
3317         bond_remove_proc_entry(bond);
3318         bond_create_proc_entry(bond);
3319 #endif
3320         down_write(&(bonding_rwsem));
3321         bond_destroy_sysfs_entry(bond);
3322         bond_create_sysfs_entry(bond);
3323         up_write(&(bonding_rwsem));
3324         return NOTIFY_DONE;
3325 }
3326
3327 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3328 {
3329         struct bonding *event_bond = bond_dev->priv;
3330
3331         switch (event) {
3332         case NETDEV_CHANGENAME:
3333                 return bond_event_changename(event_bond);
3334         case NETDEV_UNREGISTER:
3335                 /*
3336                  * TODO: remove a bond from the list?
3337                  */
3338                 break;
3339         default:
3340                 break;
3341         }
3342
3343         return NOTIFY_DONE;
3344 }
3345
3346 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3347 {
3348         struct net_device *bond_dev = slave_dev->master;
3349         struct bonding *bond = bond_dev->priv;
3350
3351         switch (event) {
3352         case NETDEV_UNREGISTER:
3353                 if (bond_dev) {
3354                         bond_release(bond_dev, slave_dev);
3355                 }
3356                 break;
3357         case NETDEV_CHANGE:
3358                 /*
3359                  * TODO: is this what we get if somebody
3360                  * sets up a hierarchical bond, then rmmod's
3361                  * one of the slave bonding devices?
3362                  */
3363                 break;
3364         case NETDEV_DOWN:
3365                 /*
3366                  * ... Or is it this?
3367                  */
3368                 break;
3369         case NETDEV_GOING_DOWN:
3370                 dprintk("slave %s is going down\n", slave_dev->name);
3371                 if (bond->setup_by_slave)
3372                         bond_release_and_destroy(bond_dev, slave_dev);
3373                 break;
3374         case NETDEV_CHANGEMTU:
3375                 /*
3376                  * TODO: Should slaves be allowed to
3377                  * independently alter their MTU?  For
3378                  * an active-backup bond, slaves need
3379                  * not be the same type of device, so
3380                  * MTUs may vary.  For other modes,
3381                  * slaves arguably should have the
3382                  * same MTUs. To do this, we'd need to
3383                  * take over the slave's change_mtu
3384                  * function for the duration of their
3385                  * servitude.
3386                  */
3387                 break;
3388         case NETDEV_CHANGENAME:
3389                 /*
3390                  * TODO: handle changing the primary's name
3391                  */
3392                 break;
3393         case NETDEV_FEAT_CHANGE:
3394                 bond_compute_features(bond);
3395                 break;
3396         default:
3397                 break;
3398         }
3399
3400         return NOTIFY_DONE;
3401 }
3402
3403 /*
3404  * bond_netdev_event: handle netdev notifier chain events.
3405  *
3406  * This function receives events for the netdev chain.  The caller (an
3407  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3408  * locks for us to safely manipulate the slave devices (RTNL lock,
3409  * dev_probe_lock).
3410  */
3411 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3412 {
3413         struct net_device *event_dev = (struct net_device *)ptr;
3414
3415         if (event_dev->nd_net != &init_net)
3416                 return NOTIFY_DONE;
3417
3418         dprintk("event_dev: %s, event: %lx\n",
3419                 (event_dev ? event_dev->name : "None"),
3420                 event);
3421
3422         if (!(event_dev->priv_flags & IFF_BONDING))
3423                 return NOTIFY_DONE;
3424
3425         if (event_dev->flags & IFF_MASTER) {
3426                 dprintk("IFF_MASTER\n");
3427                 return bond_master_netdev_event(event, event_dev);
3428         }
3429
3430         if (event_dev->flags & IFF_SLAVE) {
3431                 dprintk("IFF_SLAVE\n");
3432                 return bond_slave_netdev_event(event, event_dev);
3433         }
3434
3435         return NOTIFY_DONE;
3436 }
3437
3438 /*
3439  * bond_inetaddr_event: handle inetaddr notifier chain events.
3440  *
3441  * We keep track of device IPs primarily to use as source addresses in
3442  * ARP monitor probes (rather than spewing out broadcasts all the time).
3443  *
3444  * We track one IP for the main device (if it has one), plus one per VLAN.
3445  */
3446 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3447 {
3448         struct in_ifaddr *ifa = ptr;
3449         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3450         struct bonding *bond, *bond_next;
3451         struct vlan_entry *vlan, *vlan_next;
3452
3453         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3454                 if (bond->dev == event_dev) {
3455                         switch (event) {
3456                         case NETDEV_UP:
3457                                 bond->master_ip = ifa->ifa_local;
3458                                 return NOTIFY_OK;
3459                         case NETDEV_DOWN:
3460                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3461                                 return NOTIFY_OK;
3462                         default:
3463                                 return NOTIFY_DONE;
3464                         }
3465                 }
3466
3467                 if (list_empty(&bond->vlan_list))
3468                         continue;
3469
3470                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3471                                          vlan_list) {
3472                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3473                         if (vlan_dev == event_dev) {
3474                                 switch (event) {
3475                                 case NETDEV_UP:
3476                                         vlan->vlan_ip = ifa->ifa_local;
3477                                         return NOTIFY_OK;
3478                                 case NETDEV_DOWN:
3479                                         vlan->vlan_ip =
3480                                                 bond_glean_dev_ip(vlan_dev);
3481                                         return NOTIFY_OK;
3482                                 default:
3483                                         return NOTIFY_DONE;
3484                                 }
3485                         }
3486                 }
3487         }
3488         return NOTIFY_DONE;
3489 }
3490
3491 static struct notifier_block bond_netdev_notifier = {
3492         .notifier_call = bond_netdev_event,
3493 };
3494
3495 static struct notifier_block bond_inetaddr_notifier = {
3496         .notifier_call = bond_inetaddr_event,
3497 };
3498
3499 /*-------------------------- Packet type handling ---------------------------*/
3500
3501 /* register to receive lacpdus on a bond */
3502 static void bond_register_lacpdu(struct bonding *bond)
3503 {
3504         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3505
3506         /* initialize packet type */
3507         pk_type->type = PKT_TYPE_LACPDU;
3508         pk_type->dev = bond->dev;
3509         pk_type->func = bond_3ad_lacpdu_recv;
3510
3511         dev_add_pack(pk_type);
3512 }
3513
3514 /* unregister to receive lacpdus on a bond */
3515 static void bond_unregister_lacpdu(struct bonding *bond)
3516 {
3517         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3518 }
3519
3520 void bond_register_arp(struct bonding *bond)
3521 {
3522         struct packet_type *pt = &bond->arp_mon_pt;
3523
3524         if (pt->type)
3525                 return;
3526
3527         pt->type = htons(ETH_P_ARP);
3528         pt->dev = bond->dev;
3529         pt->func = bond_arp_rcv;
3530         dev_add_pack(pt);
3531 }
3532
3533 void bond_unregister_arp(struct bonding *bond)
3534 {
3535         struct packet_type *pt = &bond->arp_mon_pt;
3536
3537         dev_remove_pack(pt);
3538         pt->type = 0;
3539 }
3540
3541 /*---------------------------- Hashing Policies -----------------------------*/
3542
3543 /*
3544  * Hash for the output device based upon layer 3 and layer 4 data. If
3545  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3546  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3547  */
3548 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3549                                     struct net_device *bond_dev, int count)
3550 {
3551         struct ethhdr *data = (struct ethhdr *)skb->data;
3552         struct iphdr *iph = ip_hdr(skb);
3553         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3554         int layer4_xor = 0;
3555
3556         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3557                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3558                     (iph->protocol == IPPROTO_TCP ||
3559                      iph->protocol == IPPROTO_UDP)) {
3560                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3561                 }
3562                 return (layer4_xor ^
3563                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3564
3565         }
3566
3567         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3568 }
3569
3570 /*
3571  * Hash for the output device based upon layer 2 data
3572  */
3573 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3574                                    struct net_device *bond_dev, int count)
3575 {
3576         struct ethhdr *data = (struct ethhdr *)skb->data;
3577
3578         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3579 }
3580
3581 /*-------------------------- Device entry points ----------------------------*/
3582
3583 static int bond_open(struct net_device *bond_dev)
3584 {
3585         struct bonding *bond = bond_dev->priv;
3586         struct timer_list *mii_timer = &bond->mii_timer;
3587         struct timer_list *arp_timer = &bond->arp_timer;
3588
3589         bond->kill_timers = 0;
3590
3591         if ((bond->params.mode == BOND_MODE_TLB) ||
3592             (bond->params.mode == BOND_MODE_ALB)) {
3593                 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3594
3595                 /* bond_alb_initialize must be called before the timer
3596                  * is started.
3597                  */
3598                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3599                         /* something went wrong - fail the open operation */
3600                         return -1;
3601                 }
3602
3603                 init_timer(alb_timer);
3604                 alb_timer->expires  = jiffies + 1;
3605                 alb_timer->data     = (unsigned long)bond;
3606                 alb_timer->function = (void *)&bond_alb_monitor;
3607                 add_timer(alb_timer);
3608         }
3609
3610         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3611                 init_timer(mii_timer);
3612                 mii_timer->expires  = jiffies + 1;
3613                 mii_timer->data     = (unsigned long)bond_dev;
3614                 mii_timer->function = (void *)&bond_mii_monitor;
3615                 add_timer(mii_timer);
3616         }
3617
3618         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3619                 init_timer(arp_timer);
3620                 arp_timer->expires  = jiffies + 1;
3621                 arp_timer->data     = (unsigned long)bond_dev;
3622                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3623                         arp_timer->function = (void *)&bond_activebackup_arp_mon;
3624                 } else {
3625                         arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3626                 }
3627                 if (bond->params.arp_validate)
3628                         bond_register_arp(bond);
3629
3630                 add_timer(arp_timer);
3631         }
3632
3633         if (bond->params.mode == BOND_MODE_8023AD) {
3634                 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3635                 init_timer(ad_timer);
3636                 ad_timer->expires  = jiffies + 1;
3637                 ad_timer->data     = (unsigned long)bond;
3638                 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3639                 add_timer(ad_timer);
3640
3641                 /* register to receive LACPDUs */
3642                 bond_register_lacpdu(bond);
3643         }
3644
3645         return 0;
3646 }
3647
3648 static int bond_close(struct net_device *bond_dev)
3649 {
3650         struct bonding *bond = bond_dev->priv;
3651
3652         if (bond->params.mode == BOND_MODE_8023AD) {
3653                 /* Unregister the receive of LACPDUs */
3654                 bond_unregister_lacpdu(bond);
3655         }
3656
3657         if (bond->params.arp_validate)
3658                 bond_unregister_arp(bond);
3659
3660         write_lock_bh(&bond->lock);
3661
3662
3663         /* signal timers not to re-arm */
3664         bond->kill_timers = 1;
3665
3666         write_unlock_bh(&bond->lock);
3667
3668         /* del_timer_sync must run without holding the bond->lock
3669          * because a running timer might be trying to hold it too
3670          */
3671
3672         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3673                 del_timer_sync(&bond->mii_timer);
3674         }
3675
3676         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3677                 del_timer_sync(&bond->arp_timer);
3678         }
3679
3680         switch (bond->params.mode) {
3681         case BOND_MODE_8023AD:
3682                 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3683                 break;
3684         case BOND_MODE_TLB:
3685         case BOND_MODE_ALB:
3686                 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3687                 break;
3688         default:
3689                 break;
3690         }
3691
3692
3693         if ((bond->params.mode == BOND_MODE_TLB) ||
3694             (bond->params.mode == BOND_MODE_ALB)) {
3695                 /* Must be called only after all
3696                  * slaves have been released
3697                  */
3698                 bond_alb_deinitialize(bond);
3699         }
3700
3701         return 0;
3702 }
3703
3704 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3705 {
3706         struct bonding *bond = bond_dev->priv;
3707         struct net_device_stats *stats = &(bond->stats), *sstats;
3708         struct slave *slave;
3709         int i;
3710
3711         memset(stats, 0, sizeof(struct net_device_stats));
3712
3713         read_lock_bh(&bond->lock);
3714
3715         bond_for_each_slave(bond, slave, i) {
3716                 sstats = slave->dev->get_stats(slave->dev);
3717                 stats->rx_packets += sstats->rx_packets;
3718                 stats->rx_bytes += sstats->rx_bytes;
3719                 stats->rx_errors += sstats->rx_errors;
3720                 stats->rx_dropped += sstats->rx_dropped;
3721
3722                 stats->tx_packets += sstats->tx_packets;
3723                 stats->tx_bytes += sstats->tx_bytes;
3724                 stats->tx_errors += sstats->tx_errors;
3725                 stats->tx_dropped += sstats->tx_dropped;
3726
3727                 stats->multicast += sstats->multicast;
3728                 stats->collisions += sstats->collisions;
3729
3730                 stats->rx_length_errors += sstats->rx_length_errors;
3731                 stats->rx_over_errors += sstats->rx_over_errors;
3732                 stats->rx_crc_errors += sstats->rx_crc_errors;
3733                 stats->rx_frame_errors += sstats->rx_frame_errors;
3734                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3735                 stats->rx_missed_errors += sstats->rx_missed_errors;
3736
3737                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3738                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3739                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3740                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3741                 stats->tx_window_errors += sstats->tx_window_errors;
3742         }
3743
3744         read_unlock_bh(&bond->lock);
3745
3746         return stats;
3747 }
3748
3749 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3750 {
3751         struct net_device *slave_dev = NULL;
3752         struct ifbond k_binfo;
3753         struct ifbond __user *u_binfo = NULL;
3754         struct ifslave k_sinfo;
3755         struct ifslave __user *u_sinfo = NULL;
3756         struct mii_ioctl_data *mii = NULL;
3757         int res = 0;
3758
3759         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3760                 bond_dev->name, cmd);
3761
3762         switch (cmd) {
3763         case SIOCGMIIPHY:
3764                 mii = if_mii(ifr);
3765                 if (!mii) {
3766                         return -EINVAL;
3767                 }
3768                 mii->phy_id = 0;
3769                 /* Fall Through */
3770         case SIOCGMIIREG:
3771                 /*
3772                  * We do this again just in case we were called by SIOCGMIIREG
3773                  * instead of SIOCGMIIPHY.
3774                  */
3775                 mii = if_mii(ifr);
3776                 if (!mii) {
3777                         return -EINVAL;
3778                 }
3779
3780                 if (mii->reg_num == 1) {
3781                         struct bonding *bond = bond_dev->priv;
3782                         mii->val_out = 0;
3783                         read_lock_bh(&bond->lock);
3784                         read_lock(&bond->curr_slave_lock);
3785                         if (netif_carrier_ok(bond->dev)) {
3786                                 mii->val_out = BMSR_LSTATUS;
3787                         }
3788                         read_unlock(&bond->curr_slave_lock);
3789                         read_unlock_bh(&bond->lock);
3790                 }
3791
3792                 return 0;
3793         case BOND_INFO_QUERY_OLD:
3794         case SIOCBONDINFOQUERY:
3795                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3796
3797                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3798                         return -EFAULT;
3799                 }
3800
3801                 res = bond_info_query(bond_dev, &k_binfo);
3802                 if (res == 0) {
3803                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3804                                 return -EFAULT;
3805                         }
3806                 }
3807
3808                 return res;
3809         case BOND_SLAVE_INFO_QUERY_OLD:
3810         case SIOCBONDSLAVEINFOQUERY:
3811                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3812
3813                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3814                         return -EFAULT;
3815                 }
3816
3817                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3818                 if (res == 0) {
3819                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3820                                 return -EFAULT;
3821                         }
3822                 }
3823
3824                 return res;
3825         default:
3826                 /* Go on */
3827                 break;
3828         }
3829
3830         if (!capable(CAP_NET_ADMIN)) {
3831                 return -EPERM;
3832         }
3833
3834         down_write(&(bonding_rwsem));
3835         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3836
3837         dprintk("slave_dev=%p: \n", slave_dev);
3838
3839         if (!slave_dev) {
3840                 res = -ENODEV;
3841         } else {
3842                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3843                 switch (cmd) {
3844                 case BOND_ENSLAVE_OLD:
3845                 case SIOCBONDENSLAVE:
3846                         res = bond_enslave(bond_dev, slave_dev);
3847                         break;
3848                 case BOND_RELEASE_OLD:
3849                 case SIOCBONDRELEASE:
3850                         res = bond_release(bond_dev, slave_dev);
3851                         break;
3852                 case BOND_SETHWADDR_OLD:
3853                 case SIOCBONDSETHWADDR:
3854                         res = bond_sethwaddr(bond_dev, slave_dev);
3855                         break;
3856                 case BOND_CHANGE_ACTIVE_OLD:
3857                 case SIOCBONDCHANGEACTIVE:
3858                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3859                         break;
3860                 default:
3861                         res = -EOPNOTSUPP;
3862                 }
3863
3864                 dev_put(slave_dev);
3865         }
3866
3867         up_write(&(bonding_rwsem));
3868         return res;
3869 }
3870
3871 static void bond_set_multicast_list(struct net_device *bond_dev)
3872 {
3873         struct bonding *bond = bond_dev->priv;
3874         struct dev_mc_list *dmi;
3875
3876         write_lock_bh(&bond->lock);
3877
3878         /*
3879          * Do promisc before checking multicast_mode
3880          */
3881         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3882                 bond_set_promiscuity(bond, 1);
3883         }
3884
3885         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3886                 bond_set_promiscuity(bond, -1);
3887         }
3888
3889         /* set allmulti flag to slaves */
3890         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3891                 bond_set_allmulti(bond, 1);
3892         }
3893
3894         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3895                 bond_set_allmulti(bond, -1);
3896         }
3897
3898         bond->flags = bond_dev->flags;
3899
3900         /* looking for addresses to add to slaves' mc list */
3901         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3902                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3903                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3904                 }
3905         }
3906
3907         /* looking for addresses to delete from slaves' list */
3908         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3909                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3910                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3911                 }
3912         }
3913
3914         /* save master's multicast list */
3915         bond_mc_list_destroy(bond);
3916         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3917
3918         write_unlock_bh(&bond->lock);
3919 }
3920
3921 /*
3922  * Change the MTU of all of a master's slaves to match the master
3923  */
3924 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3925 {
3926         struct bonding *bond = bond_dev->priv;
3927         struct slave *slave, *stop_at;
3928         int res = 0;
3929         int i;
3930
3931         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3932                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3933
3934         /* Can't hold bond->lock with bh disabled here since
3935          * some base drivers panic. On the other hand we can't
3936          * hold bond->lock without bh disabled because we'll
3937          * deadlock. The only solution is to rely on the fact
3938          * that we're under rtnl_lock here, and the slaves
3939          * list won't change. This doesn't solve the problem
3940          * of setting the slave's MTU while it is
3941          * transmitting, but the assumption is that the base
3942          * driver can handle that.
3943          *
3944          * TODO: figure out a way to safely iterate the slaves
3945          * list, but without holding a lock around the actual
3946          * call to the base driver.
3947          */
3948
3949         bond_for_each_slave(bond, slave, i) {
3950                 dprintk("s %p s->p %p c_m %p\n", slave,
3951                         slave->prev, slave->dev->change_mtu);
3952
3953                 res = dev_set_mtu(slave->dev, new_mtu);
3954
3955                 if (res) {
3956                         /* If we failed to set the slave's mtu to the new value
3957                          * we must abort the operation even in ACTIVE_BACKUP
3958                          * mode, because if we allow the backup slaves to have
3959                          * different mtu values than the active slave we'll
3960                          * need to change their mtu when doing a failover. That
3961                          * means changing their mtu from timer context, which
3962                          * is probably not a good idea.
3963                          */
3964                         dprintk("err %d %s\n", res, slave->dev->name);
3965                         goto unwind;
3966                 }
3967         }
3968
3969         bond_dev->mtu = new_mtu;
3970
3971         return 0;
3972
3973 unwind:
3974         /* unwind from head to the slave that failed */
3975         stop_at = slave;
3976         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3977                 int tmp_res;
3978
3979                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3980                 if (tmp_res) {
3981                         dprintk("unwind err %d dev %s\n", tmp_res,
3982                                 slave->dev->name);
3983                 }
3984         }
3985
3986         return res;
3987 }
3988
3989 /*
3990  * Change HW address
3991  *
3992  * Note that many devices must be down to change the HW address, and
3993  * downing the master releases all slaves.  We can make bonds full of
3994  * bonding devices to test this, however.
3995  */
3996 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3997 {
3998         struct bonding *bond = bond_dev->priv;
3999         struct sockaddr *sa = addr, tmp_sa;
4000         struct slave *slave, *stop_at;
4001         int res = 0;
4002         int i;
4003
4004         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4005
4006         /*
4007          * If fail_over_mac is enabled, do nothing and return success.
4008          * Returning an error causes ifenslave to fail.
4009          */
4010         if (bond->params.fail_over_mac)
4011                 return 0;
4012
4013         if (!is_valid_ether_addr(sa->sa_data)) {
4014                 return -EADDRNOTAVAIL;
4015         }
4016
4017         /* Can't hold bond->lock with bh disabled here since
4018          * some base drivers panic. On the other hand we can't
4019          * hold bond->lock without bh disabled because we'll
4020          * deadlock. The only solution is to rely on the fact
4021          * that we're under rtnl_lock here, and the slaves
4022          * list won't change. This doesn't solve the problem
4023          * of setting the slave's hw address while it is
4024          * transmitting, but the assumption is that the base
4025          * driver can handle that.
4026          *
4027          * TODO: figure out a way to safely iterate the slaves
4028          * list, but without holding a lock around the actual
4029          * call to the base driver.
4030          */
4031
4032         bond_for_each_slave(bond, slave, i) {
4033                 dprintk("slave %p %s\n", slave, slave->dev->name);
4034
4035                 if (slave->dev->set_mac_address == NULL) {
4036                         res = -EOPNOTSUPP;
4037                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4038                         goto unwind;
4039                 }
4040
4041                 res = dev_set_mac_address(slave->dev, addr);
4042                 if (res) {
4043                         /* TODO: consider downing the slave
4044                          * and retry ?
4045                          * User should expect communications
4046                          * breakage anyway until ARP finish
4047                          * updating, so...
4048                          */
4049                         dprintk("err %d %s\n", res, slave->dev->name);
4050                         goto unwind;
4051                 }
4052         }
4053
4054         /* success */
4055         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4056         return 0;
4057
4058 unwind:
4059         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4060         tmp_sa.sa_family = bond_dev->type;
4061
4062         /* unwind from head to the slave that failed */
4063         stop_at = slave;
4064         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4065                 int tmp_res;
4066
4067                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4068                 if (tmp_res) {
4069                         dprintk("unwind err %d dev %s\n", tmp_res,
4070                                 slave->dev->name);
4071                 }
4072         }
4073
4074         return res;
4075 }
4076
4077 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4078 {
4079         struct bonding *bond = bond_dev->priv;
4080         struct slave *slave, *start_at;
4081         int i;
4082         int res = 1;
4083
4084         read_lock(&bond->lock);
4085
4086         if (!BOND_IS_OK(bond)) {
4087                 goto out;
4088         }
4089
4090         read_lock(&bond->curr_slave_lock);
4091         slave = start_at = bond->curr_active_slave;
4092         read_unlock(&bond->curr_slave_lock);
4093
4094         if (!slave) {
4095                 goto out;
4096         }
4097
4098         bond_for_each_slave_from(bond, slave, i, start_at) {
4099                 if (IS_UP(slave->dev) &&
4100                     (slave->link == BOND_LINK_UP) &&
4101                     (slave->state == BOND_STATE_ACTIVE)) {
4102                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4103
4104                         write_lock(&bond->curr_slave_lock);
4105                         bond->curr_active_slave = slave->next;
4106                         write_unlock(&bond->curr_slave_lock);
4107
4108                         break;
4109                 }
4110         }
4111
4112
4113 out:
4114         if (res) {
4115                 /* no suitable interface, frame not sent */
4116                 dev_kfree_skb(skb);
4117         }
4118         read_unlock(&bond->lock);
4119         return 0;
4120 }
4121
4122
4123 /*
4124  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4125  * the bond has a usable interface.
4126  */
4127 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4128 {
4129         struct bonding *bond = bond_dev->priv;
4130         int res = 1;
4131
4132         read_lock(&bond->lock);
4133         read_lock(&bond->curr_slave_lock);
4134
4135         if (!BOND_IS_OK(bond)) {
4136                 goto out;
4137         }
4138
4139         if (!bond->curr_active_slave)
4140                 goto out;
4141
4142         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4143
4144 out:
4145         if (res) {
4146                 /* no suitable interface, frame not sent */
4147                 dev_kfree_skb(skb);
4148         }
4149         read_unlock(&bond->curr_slave_lock);
4150         read_unlock(&bond->lock);
4151         return 0;
4152 }
4153
4154 /*
4155  * In bond_xmit_xor() , we determine the output device by using a pre-
4156  * determined xmit_hash_policy(), If the selected device is not enabled,
4157  * find the next active slave.
4158  */
4159 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4160 {
4161         struct bonding *bond = bond_dev->priv;
4162         struct slave *slave, *start_at;
4163         int slave_no;
4164         int i;
4165         int res = 1;
4166
4167         read_lock(&bond->lock);
4168
4169         if (!BOND_IS_OK(bond)) {
4170                 goto out;
4171         }
4172
4173         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4174
4175         bond_for_each_slave(bond, slave, i) {
4176                 slave_no--;
4177                 if (slave_no < 0) {
4178                         break;
4179                 }
4180         }
4181
4182         start_at = slave;
4183
4184         bond_for_each_slave_from(bond, slave, i, start_at) {
4185                 if (IS_UP(slave->dev) &&
4186                     (slave->link == BOND_LINK_UP) &&
4187                     (slave->state == BOND_STATE_ACTIVE)) {
4188                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4189                         break;
4190                 }
4191         }
4192
4193 out:
4194         if (res) {
4195                 /* no suitable interface, frame not sent */
4196                 dev_kfree_skb(skb);
4197         }
4198         read_unlock(&bond->lock);
4199         return 0;
4200 }
4201
4202 /*
4203  * in broadcast mode, we send everything to all usable interfaces.
4204  */
4205 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4206 {
4207         struct bonding *bond = bond_dev->priv;
4208         struct slave *slave, *start_at;
4209         struct net_device *tx_dev = NULL;
4210         int i;
4211         int res = 1;
4212
4213         read_lock(&bond->lock);
4214
4215         if (!BOND_IS_OK(bond)) {
4216                 goto out;
4217         }
4218
4219         read_lock(&bond->curr_slave_lock);
4220         start_at = bond->curr_active_slave;
4221         read_unlock(&bond->curr_slave_lock);
4222
4223         if (!start_at) {
4224                 goto out;
4225         }
4226
4227         bond_for_each_slave_from(bond, slave, i, start_at) {
4228                 if (IS_UP(slave->dev) &&
4229                     (slave->link == BOND_LINK_UP) &&
4230                     (slave->state == BOND_STATE_ACTIVE)) {
4231                         if (tx_dev) {
4232                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4233                                 if (!skb2) {
4234                                         printk(KERN_ERR DRV_NAME
4235                                                ": %s: Error: bond_xmit_broadcast(): "
4236                                                "skb_clone() failed\n",
4237                                                bond_dev->name);
4238                                         continue;
4239                                 }
4240
4241                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4242                                 if (res) {
4243                                         dev_kfree_skb(skb2);
4244                                         continue;
4245                                 }
4246                         }
4247                         tx_dev = slave->dev;
4248                 }
4249         }
4250
4251         if (tx_dev) {
4252                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4253         }
4254
4255 out:
4256         if (res) {
4257                 /* no suitable interface, frame not sent */
4258                 dev_kfree_skb(skb);
4259         }
4260         /* frame sent to all suitable interfaces */
4261         read_unlock(&bond->lock);
4262         return 0;
4263 }
4264
4265 /*------------------------- Device initialization ---------------------------*/
4266
4267 /*
4268  * set bond mode specific net device operations
4269  */
4270 void bond_set_mode_ops(struct bonding *bond, int mode)
4271 {
4272         struct net_device *bond_dev = bond->dev;
4273
4274         switch (mode) {
4275         case BOND_MODE_ROUNDROBIN:
4276                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4277                 break;
4278         case BOND_MODE_ACTIVEBACKUP:
4279                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4280                 break;
4281         case BOND_MODE_XOR:
4282                 bond_dev->hard_start_xmit = bond_xmit_xor;
4283                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4284                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4285                 else
4286                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4287                 break;
4288         case BOND_MODE_BROADCAST:
4289                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4290                 break;
4291         case BOND_MODE_8023AD:
4292                 bond_set_master_3ad_flags(bond);
4293                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4294                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4295                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4296                 else
4297                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4298                 break;
4299         case BOND_MODE_ALB:
4300                 bond_set_master_alb_flags(bond);
4301                 /* FALLTHRU */
4302         case BOND_MODE_TLB:
4303                 bond_dev->hard_start_xmit = bond_alb_xmit;
4304                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4305                 break;
4306         default:
4307                 /* Should never happen, mode already checked */
4308                 printk(KERN_ERR DRV_NAME
4309                        ": %s: Error: Unknown bonding mode %d\n",
4310                        bond_dev->name,
4311                        mode);
4312                 break;
4313         }
4314 }
4315
4316 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4317                                     struct ethtool_drvinfo *drvinfo)
4318 {
4319         strncpy(drvinfo->driver, DRV_NAME, 32);
4320         strncpy(drvinfo->version, DRV_VERSION, 32);
4321         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4322 }
4323
4324 static const struct ethtool_ops bond_ethtool_ops = {
4325         .get_drvinfo            = bond_ethtool_get_drvinfo,
4326 };
4327
4328 /*
4329  * Does not allocate but creates a /proc entry.
4330  * Allowed to fail.
4331  */
4332 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4333 {
4334         struct bonding *bond = bond_dev->priv;
4335
4336         dprintk("Begin bond_init for %s\n", bond_dev->name);
4337
4338         /* initialize rwlocks */
4339         rwlock_init(&bond->lock);
4340         rwlock_init(&bond->curr_slave_lock);
4341
4342         bond->params = *params; /* copy params struct */
4343
4344         /* Initialize pointers */
4345         bond->first_slave = NULL;
4346         bond->curr_active_slave = NULL;
4347         bond->current_arp_slave = NULL;
4348         bond->primary_slave = NULL;
4349         bond->dev = bond_dev;
4350         bond->send_grat_arp = 0;
4351         bond->setup_by_slave = 0;
4352         INIT_LIST_HEAD(&bond->vlan_list);
4353
4354         /* Initialize the device entry points */
4355         bond_dev->open = bond_open;
4356         bond_dev->stop = bond_close;
4357         bond_dev->get_stats = bond_get_stats;
4358         bond_dev->do_ioctl = bond_do_ioctl;
4359         bond_dev->ethtool_ops = &bond_ethtool_ops;
4360         bond_dev->set_multicast_list = bond_set_multicast_list;
4361         bond_dev->change_mtu = bond_change_mtu;
4362         bond_dev->set_mac_address = bond_set_mac_address;
4363
4364         bond_set_mode_ops(bond, bond->params.mode);
4365
4366         bond_dev->destructor = free_netdev;
4367
4368         /* Initialize the device options */
4369         bond_dev->tx_queue_len = 0;
4370         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4371         bond_dev->priv_flags |= IFF_BONDING;
4372
4373         /* At first, we block adding VLANs. That's the only way to
4374          * prevent problems that occur when adding VLANs over an
4375          * empty bond. The block will be removed once non-challenged
4376          * slaves are enslaved.
4377          */
4378         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4379
4380         /* don't acquire bond device's netif_tx_lock when
4381          * transmitting */
4382         bond_dev->features |= NETIF_F_LLTX;
4383
4384         /* By default, we declare the bond to be fully
4385          * VLAN hardware accelerated capable. Special
4386          * care is taken in the various xmit functions
4387          * when there are slaves that are not hw accel
4388          * capable
4389          */
4390         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4391         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4392         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4393         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4394                                NETIF_F_HW_VLAN_RX |
4395                                NETIF_F_HW_VLAN_FILTER);
4396
4397 #ifdef CONFIG_PROC_FS
4398         bond_create_proc_entry(bond);
4399 #endif
4400         list_add_tail(&bond->bond_list, &bond_dev_list);
4401
4402         return 0;
4403 }
4404
4405 /* De-initialize device specific data.
4406  * Caller must hold rtnl_lock.
4407  */
4408 void bond_deinit(struct net_device *bond_dev)
4409 {
4410         struct bonding *bond = bond_dev->priv;
4411
4412         list_del(&bond->bond_list);
4413
4414 #ifdef CONFIG_PROC_FS
4415         bond_remove_proc_entry(bond);
4416 #endif
4417 }
4418
4419 /* Unregister and free all bond devices.
4420  * Caller must hold rtnl_lock.
4421  */
4422 static void bond_free_all(void)
4423 {
4424         struct bonding *bond, *nxt;
4425
4426         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4427                 struct net_device *bond_dev = bond->dev;
4428
4429                 bond_mc_list_destroy(bond);
4430                 /* Release the bonded slaves */
4431                 bond_release_all(bond_dev);
4432                 bond_deinit(bond_dev);
4433                 unregister_netdevice(bond_dev);
4434         }
4435
4436 #ifdef CONFIG_PROC_FS
4437         bond_destroy_proc_dir();
4438 #endif
4439 }
4440
4441 /*------------------------- Module initialization ---------------------------*/
4442
4443 /*
4444  * Convert string input module parms.  Accept either the
4445  * number of the mode or its string name.
4446  */
4447 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4448 {
4449         int i;
4450
4451         for (i = 0; tbl[i].modename; i++) {
4452                 if ((isdigit(*mode_arg) &&
4453                      tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4454                     (strncmp(mode_arg, tbl[i].modename,
4455                              strlen(tbl[i].modename)) == 0)) {
4456                         return tbl[i].mode;
4457                 }
4458         }
4459
4460         return -1;
4461 }
4462
4463 static int bond_check_params(struct bond_params *params)
4464 {
4465         int arp_validate_value;
4466
4467         /*
4468          * Convert string parameters.
4469          */
4470         if (mode) {
4471                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4472                 if (bond_mode == -1) {
4473                         printk(KERN_ERR DRV_NAME
4474                                ": Error: Invalid bonding mode \"%s\"\n",
4475                                mode == NULL ? "NULL" : mode);
4476                         return -EINVAL;
4477                 }
4478         }
4479
4480         if (xmit_hash_policy) {
4481                 if ((bond_mode != BOND_MODE_XOR) &&
4482                     (bond_mode != BOND_MODE_8023AD)) {
4483                         printk(KERN_INFO DRV_NAME
4484                                ": xor_mode param is irrelevant in mode %s\n",
4485                                bond_mode_name(bond_mode));
4486                 } else {
4487                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4488                                                         xmit_hashtype_tbl);
4489                         if (xmit_hashtype == -1) {
4490                                 printk(KERN_ERR DRV_NAME
4491                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4492                                 xmit_hash_policy == NULL ? "NULL" :
4493                                        xmit_hash_policy);
4494                                 return -EINVAL;
4495                         }
4496                 }
4497         }
4498
4499         if (lacp_rate) {
4500                 if (bond_mode != BOND_MODE_8023AD) {
4501                         printk(KERN_INFO DRV_NAME
4502                                ": lacp_rate param is irrelevant in mode %s\n",
4503                                bond_mode_name(bond_mode));
4504                 } else {
4505                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4506                         if (lacp_fast == -1) {
4507                                 printk(KERN_ERR DRV_NAME
4508                                        ": Error: Invalid lacp rate \"%s\"\n",
4509                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4510                                 return -EINVAL;
4511                         }
4512                 }
4513         }
4514
4515         if (max_bonds < 1 || max_bonds > INT_MAX) {
4516                 printk(KERN_WARNING DRV_NAME
4517                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4518                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4519                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4520                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4521         }
4522
4523         if (miimon < 0) {
4524                 printk(KERN_WARNING DRV_NAME
4525                        ": Warning: miimon module parameter (%d), "
4526                        "not in range 0-%d, so it was reset to %d\n",
4527                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4528                 miimon = BOND_LINK_MON_INTERV;
4529         }
4530
4531         if (updelay < 0) {
4532                 printk(KERN_WARNING DRV_NAME
4533                        ": Warning: updelay module parameter (%d), "
4534                        "not in range 0-%d, so it was reset to 0\n",
4535                        updelay, INT_MAX);
4536                 updelay = 0;
4537         }
4538
4539         if (downdelay < 0) {
4540                 printk(KERN_WARNING DRV_NAME
4541                        ": Warning: downdelay module parameter (%d), "
4542                        "not in range 0-%d, so it was reset to 0\n",
4543                        downdelay, INT_MAX);
4544                 downdelay = 0;
4545         }
4546
4547         if ((use_carrier != 0) && (use_carrier != 1)) {
4548                 printk(KERN_WARNING DRV_NAME
4549                        ": Warning: use_carrier module parameter (%d), "
4550                        "not of valid value (0/1), so it was set to 1\n",
4551                        use_carrier);
4552                 use_carrier = 1;
4553         }
4554
4555         /* reset values for 802.3ad */
4556         if (bond_mode == BOND_MODE_8023AD) {
4557                 if (!miimon) {
4558                         printk(KERN_WARNING DRV_NAME
4559                                ": Warning: miimon must be specified, "
4560                                "otherwise bonding will not detect link "
4561                                "failure, speed and duplex which are "
4562                                "essential for 802.3ad operation\n");
4563                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4564                         miimon = 100;
4565                 }
4566         }
4567
4568         /* reset values for TLB/ALB */
4569         if ((bond_mode == BOND_MODE_TLB) ||
4570             (bond_mode == BOND_MODE_ALB)) {
4571                 if (!miimon) {
4572                         printk(KERN_WARNING DRV_NAME
4573                                ": Warning: miimon must be specified, "
4574                                "otherwise bonding will not detect link "
4575                                "failure and link speed which are essential "
4576                                "for TLB/ALB load balancing\n");
4577                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4578                         miimon = 100;
4579                 }
4580         }
4581
4582         if (bond_mode == BOND_MODE_ALB) {
4583                 printk(KERN_NOTICE DRV_NAME
4584                        ": In ALB mode you might experience client "
4585                        "disconnections upon reconnection of a link if the "
4586                        "bonding module updelay parameter (%d msec) is "
4587                        "incompatible with the forwarding delay time of the "
4588                        "switch\n",
4589                        updelay);
4590         }
4591
4592         if (!miimon) {
4593                 if (updelay || downdelay) {
4594                         /* just warn the user the up/down delay will have
4595                          * no effect since miimon is zero...
4596                          */
4597                         printk(KERN_WARNING DRV_NAME
4598                                ": Warning: miimon module parameter not set "
4599                                "and updelay (%d) or downdelay (%d) module "
4600                                "parameter is set; updelay and downdelay have "
4601                                "no effect unless miimon is set\n",
4602                                updelay, downdelay);
4603                 }
4604         } else {
4605                 /* don't allow arp monitoring */
4606                 if (arp_interval) {
4607                         printk(KERN_WARNING DRV_NAME
4608                                ": Warning: miimon (%d) and arp_interval (%d) "
4609                                "can't be used simultaneously, disabling ARP "
4610                                "monitoring\n",
4611                                miimon, arp_interval);
4612                         arp_interval = 0;
4613                 }
4614
4615                 if ((updelay % miimon) != 0) {
4616                         printk(KERN_WARNING DRV_NAME
4617                                ": Warning: updelay (%d) is not a multiple "
4618                                "of miimon (%d), updelay rounded to %d ms\n",
4619                                updelay, miimon, (updelay / miimon) * miimon);
4620                 }
4621
4622                 updelay /= miimon;
4623
4624                 if ((downdelay % miimon) != 0) {
4625                         printk(KERN_WARNING DRV_NAME
4626                                ": Warning: downdelay (%d) is not a multiple "
4627                                "of miimon (%d), downdelay rounded to %d ms\n",
4628                                downdelay, miimon,
4629                                (downdelay / miimon) * miimon);
4630                 }
4631
4632                 downdelay /= miimon;
4633         }
4634
4635         if (arp_interval < 0) {
4636                 printk(KERN_WARNING DRV_NAME
4637                        ": Warning: arp_interval module parameter (%d) "
4638                        ", not in range 0-%d, so it was reset to %d\n",
4639                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4640                 arp_interval = BOND_LINK_ARP_INTERV;
4641         }
4642
4643         for (arp_ip_count = 0;
4644              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4645              arp_ip_count++) {
4646                 /* not complete check, but should be good enough to
4647                    catch mistakes */
4648                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4649                         printk(KERN_WARNING DRV_NAME
4650                                ": Warning: bad arp_ip_target module parameter "
4651                                "(%s), ARP monitoring will not be performed\n",
4652                                arp_ip_target[arp_ip_count]);
4653                         arp_interval = 0;
4654                 } else {
4655                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4656                         arp_target[arp_ip_count] = ip;
4657                 }
4658         }
4659
4660         if (arp_interval && !arp_ip_count) {
4661                 /* don't allow arping if no arp_ip_target given... */
4662                 printk(KERN_WARNING DRV_NAME
4663                        ": Warning: arp_interval module parameter (%d) "
4664                        "specified without providing an arp_ip_target "
4665                        "parameter, arp_interval was reset to 0\n",
4666                        arp_interval);
4667                 arp_interval = 0;
4668         }
4669
4670         if (arp_validate) {
4671                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4672                         printk(KERN_ERR DRV_NAME
4673                ": arp_validate only supported in active-backup mode\n");
4674                         return -EINVAL;
4675                 }
4676                 if (!arp_interval) {
4677                         printk(KERN_ERR DRV_NAME
4678                                ": arp_validate requires arp_interval\n");
4679                         return -EINVAL;
4680                 }
4681
4682                 arp_validate_value = bond_parse_parm(arp_validate,
4683                                                      arp_validate_tbl);
4684                 if (arp_validate_value == -1) {
4685                         printk(KERN_ERR DRV_NAME
4686                                ": Error: invalid arp_validate \"%s\"\n",
4687                                arp_validate == NULL ? "NULL" : arp_validate);
4688                         return -EINVAL;
4689                 }
4690         } else
4691                 arp_validate_value = 0;
4692
4693         if (miimon) {
4694                 printk(KERN_INFO DRV_NAME
4695                        ": MII link monitoring set to %d ms\n",
4696                        miimon);
4697         } else if (arp_interval) {
4698                 int i;
4699
4700                 printk(KERN_INFO DRV_NAME
4701                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4702                        arp_interval,
4703                        arp_validate_tbl[arp_validate_value].modename,
4704                        arp_ip_count);
4705
4706                 for (i = 0; i < arp_ip_count; i++)
4707                         printk (" %s", arp_ip_target[i]);
4708
4709                 printk("\n");
4710
4711         } else {
4712                 /* miimon and arp_interval not set, we need one so things
4713                  * work as expected, see bonding.txt for details
4714                  */
4715                 printk(KERN_WARNING DRV_NAME
4716                        ": Warning: either miimon or arp_interval and "
4717                        "arp_ip_target module parameters must be specified, "
4718                        "otherwise bonding will not detect link failures! see "
4719                        "bonding.txt for details.\n");
4720         }
4721
4722         if (primary && !USES_PRIMARY(bond_mode)) {
4723                 /* currently, using a primary only makes sense
4724                  * in active backup, TLB or ALB modes
4725                  */
4726                 printk(KERN_WARNING DRV_NAME
4727                        ": Warning: %s primary device specified but has no "
4728                        "effect in %s mode\n",
4729                        primary, bond_mode_name(bond_mode));
4730                 primary = NULL;
4731         }
4732
4733         if (fail_over_mac && (bond_mode != BOND_MODE_ACTIVEBACKUP))
4734                 printk(KERN_WARNING DRV_NAME
4735                        ": Warning: fail_over_mac only affects "
4736                        "active-backup mode.\n");
4737
4738         /* fill params struct with the proper values */
4739         params->mode = bond_mode;
4740         params->xmit_policy = xmit_hashtype;
4741         params->miimon = miimon;
4742         params->arp_interval = arp_interval;
4743         params->arp_validate = arp_validate_value;
4744         params->updelay = updelay;
4745         params->downdelay = downdelay;
4746         params->use_carrier = use_carrier;
4747         params->lacp_fast = lacp_fast;
4748         params->primary[0] = 0;
4749         params->fail_over_mac = fail_over_mac;
4750
4751         if (primary) {
4752                 strncpy(params->primary, primary, IFNAMSIZ);
4753                 params->primary[IFNAMSIZ - 1] = 0;
4754         }
4755
4756         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4757
4758         return 0;
4759 }
4760
4761 static struct lock_class_key bonding_netdev_xmit_lock_key;
4762
4763 /* Create a new bond based on the specified name and bonding parameters.
4764  * If name is NULL, obtain a suitable "bond%d" name for us.
4765  * Caller must NOT hold rtnl_lock; we need to release it here before we
4766  * set up our sysfs entries.
4767  */
4768 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4769 {
4770         struct net_device *bond_dev;
4771         int res;
4772
4773         rtnl_lock();
4774         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4775                                 ether_setup);
4776         if (!bond_dev) {
4777                 printk(KERN_ERR DRV_NAME
4778                        ": %s: eek! can't alloc netdev!\n",
4779                        name);
4780                 res = -ENOMEM;
4781                 goto out_rtnl;
4782         }
4783
4784         if (!name) {
4785                 res = dev_alloc_name(bond_dev, "bond%d");
4786                 if (res < 0)
4787                         goto out_netdev;
4788         }
4789
4790         /* bond_init() must be called after dev_alloc_name() (for the
4791          * /proc files), but before register_netdevice(), because we
4792          * need to set function pointers.
4793          */
4794
4795         res = bond_init(bond_dev, params);
4796         if (res < 0) {
4797                 goto out_netdev;
4798         }
4799
4800         res = register_netdevice(bond_dev);
4801         if (res < 0) {
4802                 goto out_bond;
4803         }
4804
4805         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4806
4807         if (newbond)
4808                 *newbond = bond_dev->priv;
4809
4810         netif_carrier_off(bond_dev);
4811
4812         rtnl_unlock(); /* allows sysfs registration of net device */
4813         res = bond_create_sysfs_entry(bond_dev->priv);
4814         if (res < 0) {
4815                 rtnl_lock();
4816                 goto out_bond;
4817         }
4818
4819         return 0;
4820
4821 out_bond:
4822         bond_deinit(bond_dev);
4823 out_netdev:
4824         free_netdev(bond_dev);
4825 out_rtnl:
4826         rtnl_unlock();
4827         return res;
4828 }
4829
4830 static int __init bonding_init(void)
4831 {
4832         int i;
4833         int res;
4834
4835         printk(KERN_INFO "%s", version);
4836
4837         res = bond_check_params(&bonding_defaults);
4838         if (res) {
4839                 goto out;
4840         }
4841
4842 #ifdef CONFIG_PROC_FS
4843         bond_create_proc_dir();
4844 #endif
4845         for (i = 0; i < max_bonds; i++) {
4846                 res = bond_create(NULL, &bonding_defaults, NULL);
4847                 if (res)
4848                         goto err;
4849         }
4850
4851         res = bond_create_sysfs();
4852         if (res)
4853                 goto err;
4854
4855         register_netdevice_notifier(&bond_netdev_notifier);
4856         register_inetaddr_notifier(&bond_inetaddr_notifier);
4857
4858         goto out;
4859 err:
4860         rtnl_lock();
4861         bond_free_all();
4862         bond_destroy_sysfs();
4863         rtnl_unlock();
4864 out:
4865         return res;
4866
4867 }
4868
4869 static void __exit bonding_exit(void)
4870 {
4871         unregister_netdevice_notifier(&bond_netdev_notifier);
4872         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4873
4874         rtnl_lock();
4875         bond_free_all();
4876         bond_destroy_sysfs();
4877         rtnl_unlock();
4878 }
4879
4880 module_init(bonding_init);
4881 module_exit(bonding_exit);
4882 MODULE_LICENSE("GPL");
4883 MODULE_VERSION(DRV_VERSION);
4884 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4885 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4886 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4887
4888 /*
4889  * Local variables:
4890  *  c-indent-level: 8
4891  *  c-basic-offset: 8
4892  *  tab-width: 8
4893  * End:
4894  */
4895