08879d552ae03bd4539c1d51860a3cd329056451
[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 <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
80 #include "bonding.h"
81 #include "bond_3ad.h"
82 #include "bond_alb.h"
83
84 /*---------------------------- Module parameters ----------------------------*/
85
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV    0
88 #define BOND_LINK_ARP_INTERV    0
89
90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
91 static int miimon       = BOND_LINK_MON_INTERV;
92 static int updelay      = 0;
93 static int downdelay    = 0;
94 static int use_carrier  = 1;
95 static char *mode       = NULL;
96 static char *primary    = NULL;
97 static char *lacp_rate  = NULL;
98 static char *xmit_hash_policy = NULL;
99 static int arp_interval = BOND_LINK_ARP_INTERV;
100 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
101 static char *arp_validate = NULL;
102 static int fail_over_mac = 0;
103 struct bond_params bonding_defaults;
104
105 module_param(max_bonds, int, 0);
106 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
107 module_param(miimon, int, 0);
108 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
109 module_param(updelay, int, 0);
110 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
111 module_param(downdelay, int, 0);
112 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
113                             "in milliseconds");
114 module_param(use_carrier, int, 0);
115 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
116                               "0 for off, 1 for on (default)");
117 module_param(mode, charp, 0);
118 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
119                        "1 for active-backup, 2 for balance-xor, "
120                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
121                        "6 for balance-alb");
122 module_param(primary, charp, 0);
123 MODULE_PARM_DESC(primary, "Primary network device to use");
124 module_param(lacp_rate, charp, 0);
125 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
126                             "(slow/fast)");
127 module_param(xmit_hash_policy, charp, 0);
128 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
129                                    ", 1 for layer 3+4");
130 module_param(arp_interval, int, 0);
131 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
132 module_param_array(arp_ip_target, charp, NULL, 0);
133 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
134 module_param(arp_validate, charp, 0);
135 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
136 module_param(fail_over_mac, int, 0);
137 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.");
138
139 /*----------------------------- Global variables ----------------------------*/
140
141 static const char * const version =
142         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
143
144 LIST_HEAD(bond_dev_list);
145
146 #ifdef CONFIG_PROC_FS
147 static struct proc_dir_entry *bond_proc_dir = NULL;
148 #endif
149
150 extern struct rw_semaphore bonding_rwsem;
151 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
152 static int arp_ip_count = 0;
153 static int bond_mode    = BOND_MODE_ROUNDROBIN;
154 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
155 static int lacp_fast    = 0;
156
157
158 struct bond_parm_tbl bond_lacp_tbl[] = {
159 {       "slow",         AD_LACP_SLOW},
160 {       "fast",         AD_LACP_FAST},
161 {       NULL,           -1},
162 };
163
164 struct bond_parm_tbl bond_mode_tbl[] = {
165 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
166 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
167 {       "balance-xor",          BOND_MODE_XOR},
168 {       "broadcast",            BOND_MODE_BROADCAST},
169 {       "802.3ad",              BOND_MODE_8023AD},
170 {       "balance-tlb",          BOND_MODE_TLB},
171 {       "balance-alb",          BOND_MODE_ALB},
172 {       NULL,                   -1},
173 };
174
175 struct bond_parm_tbl xmit_hashtype_tbl[] = {
176 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
177 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
178 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
179 {       NULL,                   -1},
180 };
181
182 struct bond_parm_tbl arp_validate_tbl[] = {
183 {       "none",                 BOND_ARP_VALIDATE_NONE},
184 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
185 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
186 {       "all",                  BOND_ARP_VALIDATE_ALL},
187 {       NULL,                   -1},
188 };
189
190 /*-------------------------- Forward declarations ---------------------------*/
191
192 static void bond_send_gratuitous_arp(struct bonding *bond);
193 static void bond_deinit(struct net_device *bond_dev);
194
195 /*---------------------------- General routines -----------------------------*/
196
197 static const char *bond_mode_name(int mode)
198 {
199         switch (mode) {
200         case BOND_MODE_ROUNDROBIN :
201                 return "load balancing (round-robin)";
202         case BOND_MODE_ACTIVEBACKUP :
203                 return "fault-tolerance (active-backup)";
204         case BOND_MODE_XOR :
205                 return "load balancing (xor)";
206         case BOND_MODE_BROADCAST :
207                 return "fault-tolerance (broadcast)";
208         case BOND_MODE_8023AD:
209                 return "IEEE 802.3ad Dynamic link aggregation";
210         case BOND_MODE_TLB:
211                 return "transmit load balancing";
212         case BOND_MODE_ALB:
213                 return "adaptive load balancing";
214         default:
215                 return "unknown";
216         }
217 }
218
219 /*---------------------------------- VLAN -----------------------------------*/
220
221 /**
222  * bond_add_vlan - add a new vlan id on bond
223  * @bond: bond that got the notification
224  * @vlan_id: the vlan id to add
225  *
226  * Returns -ENOMEM if allocation failed.
227  */
228 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
229 {
230         struct vlan_entry *vlan;
231
232         dprintk("bond: %s, vlan id %d\n",
233                 (bond ? bond->dev->name: "None"), vlan_id);
234
235         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
236         if (!vlan) {
237                 return -ENOMEM;
238         }
239
240         INIT_LIST_HEAD(&vlan->vlan_list);
241         vlan->vlan_id = vlan_id;
242         vlan->vlan_ip = 0;
243
244         write_lock_bh(&bond->lock);
245
246         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
247
248         write_unlock_bh(&bond->lock);
249
250         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
251
252         return 0;
253 }
254
255 /**
256  * bond_del_vlan - delete a vlan id from bond
257  * @bond: bond that got the notification
258  * @vlan_id: the vlan id to delete
259  *
260  * returns -ENODEV if @vlan_id was not found in @bond.
261  */
262 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
263 {
264         struct vlan_entry *vlan, *next;
265         int res = -ENODEV;
266
267         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
268
269         write_lock_bh(&bond->lock);
270
271         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
272                 if (vlan->vlan_id == vlan_id) {
273                         list_del(&vlan->vlan_list);
274
275                         if ((bond->params.mode == BOND_MODE_TLB) ||
276                             (bond->params.mode == BOND_MODE_ALB)) {
277                                 bond_alb_clear_vlan(bond, vlan_id);
278                         }
279
280                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
281                                 bond->dev->name);
282
283                         kfree(vlan);
284
285                         if (list_empty(&bond->vlan_list) &&
286                             (bond->slave_cnt == 0)) {
287                                 /* Last VLAN removed and no slaves, so
288                                  * restore block on adding VLANs. This will
289                                  * be removed once new slaves that are not
290                                  * VLAN challenged will be added.
291                                  */
292                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
293                         }
294
295                         res = 0;
296                         goto out;
297                 }
298         }
299
300         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
301                 bond->dev->name);
302
303 out:
304         write_unlock_bh(&bond->lock);
305         return res;
306 }
307
308 /**
309  * bond_has_challenged_slaves
310  * @bond: the bond we're working on
311  *
312  * Searches the slave list. Returns 1 if a vlan challenged slave
313  * was found, 0 otherwise.
314  *
315  * Assumes bond->lock is held.
316  */
317 static int bond_has_challenged_slaves(struct bonding *bond)
318 {
319         struct slave *slave;
320         int i;
321
322         bond_for_each_slave(bond, slave, i) {
323                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
324                         dprintk("found VLAN challenged slave - %s\n",
325                                 slave->dev->name);
326                         return 1;
327                 }
328         }
329
330         dprintk("no VLAN challenged slaves found\n");
331         return 0;
332 }
333
334 /**
335  * bond_next_vlan - safely skip to the next item in the vlans list.
336  * @bond: the bond we're working on
337  * @curr: item we're advancing from
338  *
339  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
340  * or @curr->next otherwise (even if it is @curr itself again).
341  * 
342  * Caller must hold bond->lock
343  */
344 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
345 {
346         struct vlan_entry *next, *last;
347
348         if (list_empty(&bond->vlan_list)) {
349                 return NULL;
350         }
351
352         if (!curr) {
353                 next = list_entry(bond->vlan_list.next,
354                                   struct vlan_entry, vlan_list);
355         } else {
356                 last = list_entry(bond->vlan_list.prev,
357                                   struct vlan_entry, vlan_list);
358                 if (last == curr) {
359                         next = list_entry(bond->vlan_list.next,
360                                           struct vlan_entry, vlan_list);
361                 } else {
362                         next = list_entry(curr->vlan_list.next,
363                                           struct vlan_entry, vlan_list);
364                 }
365         }
366
367         return next;
368 }
369
370 /**
371  * bond_dev_queue_xmit - Prepare skb for xmit.
372  * 
373  * @bond: bond device that got this skb for tx.
374  * @skb: hw accel VLAN tagged skb to transmit
375  * @slave_dev: slave that is supposed to xmit this skbuff
376  * 
377  * When the bond gets an skb to transmit that is
378  * already hardware accelerated VLAN tagged, and it
379  * needs to relay this skb to a slave that is not
380  * hw accel capable, the skb needs to be "unaccelerated",
381  * i.e. strip the hwaccel tag and re-insert it as part
382  * of the payload.
383  */
384 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
385 {
386         unsigned short vlan_id;
387
388         if (!list_empty(&bond->vlan_list) &&
389             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
390             vlan_get_tag(skb, &vlan_id) == 0) {
391                 skb->dev = slave_dev;
392                 skb = vlan_put_tag(skb, vlan_id);
393                 if (!skb) {
394                         /* vlan_put_tag() frees the skb in case of error,
395                          * so return success here so the calling functions
396                          * won't attempt to free is again.
397                          */
398                         return 0;
399                 }
400         } else {
401                 skb->dev = slave_dev;
402         }
403
404         skb->priority = 1;
405         dev_queue_xmit(skb);
406
407         return 0;
408 }
409
410 /*
411  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
412  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
413  * lock because:
414  * a. This operation is performed in IOCTL context,
415  * b. The operation is protected by the RTNL semaphore in the 8021q code,
416  * c. Holding a lock with BH disabled while directly calling a base driver
417  *    entry point is generally a BAD idea.
418  * 
419  * The design of synchronization/protection for this operation in the 8021q
420  * module is good for one or more VLAN devices over a single physical device
421  * and cannot be extended for a teaming solution like bonding, so there is a
422  * potential race condition here where a net device from the vlan group might
423  * be referenced (either by a base driver or the 8021q code) while it is being
424  * removed from the system. However, it turns out we're not making matters
425  * worse, and if it works for regular VLAN usage it will work here too.
426 */
427
428 /**
429  * bond_vlan_rx_register - Propagates registration to slaves
430  * @bond_dev: bonding net device that got called
431  * @grp: vlan group being registered
432  */
433 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
434 {
435         struct bonding *bond = bond_dev->priv;
436         struct slave *slave;
437         int i;
438
439         bond->vlgrp = grp;
440
441         bond_for_each_slave(bond, slave, i) {
442                 struct net_device *slave_dev = slave->dev;
443
444                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
445                     slave_dev->vlan_rx_register) {
446                         slave_dev->vlan_rx_register(slave_dev, grp);
447                 }
448         }
449 }
450
451 /**
452  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
453  * @bond_dev: bonding net device that got called
454  * @vid: vlan id being added
455  */
456 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
457 {
458         struct bonding *bond = bond_dev->priv;
459         struct slave *slave;
460         int i, res;
461
462         bond_for_each_slave(bond, slave, i) {
463                 struct net_device *slave_dev = slave->dev;
464
465                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
466                     slave_dev->vlan_rx_add_vid) {
467                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
468                 }
469         }
470
471         res = bond_add_vlan(bond, vid);
472         if (res) {
473                 printk(KERN_ERR DRV_NAME
474                        ": %s: Error: Failed to add vlan id %d\n",
475                        bond_dev->name, vid);
476         }
477 }
478
479 /**
480  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
481  * @bond_dev: bonding net device that got called
482  * @vid: vlan id being removed
483  */
484 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
485 {
486         struct bonding *bond = bond_dev->priv;
487         struct slave *slave;
488         struct net_device *vlan_dev;
489         int i, res;
490
491         bond_for_each_slave(bond, slave, i) {
492                 struct net_device *slave_dev = slave->dev;
493
494                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
495                     slave_dev->vlan_rx_kill_vid) {
496                         /* Save and then restore vlan_dev in the grp array,
497                          * since the slave's driver might clear it.
498                          */
499                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
500                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
501                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
502                 }
503         }
504
505         res = bond_del_vlan(bond, vid);
506         if (res) {
507                 printk(KERN_ERR DRV_NAME
508                        ": %s: Error: Failed to remove vlan id %d\n",
509                        bond_dev->name, vid);
510         }
511 }
512
513 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
514 {
515         struct vlan_entry *vlan;
516
517         write_lock_bh(&bond->lock);
518
519         if (list_empty(&bond->vlan_list)) {
520                 goto out;
521         }
522
523         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
524             slave_dev->vlan_rx_register) {
525                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
526         }
527
528         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
529             !(slave_dev->vlan_rx_add_vid)) {
530                 goto out;
531         }
532
533         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
534                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
535         }
536
537 out:
538         write_unlock_bh(&bond->lock);
539 }
540
541 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
542 {
543         struct vlan_entry *vlan;
544         struct net_device *vlan_dev;
545
546         write_lock_bh(&bond->lock);
547
548         if (list_empty(&bond->vlan_list)) {
549                 goto out;
550         }
551
552         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
553             !(slave_dev->vlan_rx_kill_vid)) {
554                 goto unreg;
555         }
556
557         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
558                 /* Save and then restore vlan_dev in the grp array,
559                  * since the slave's driver might clear it.
560                  */
561                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
562                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
563                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
564         }
565
566 unreg:
567         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
568             slave_dev->vlan_rx_register) {
569                 slave_dev->vlan_rx_register(slave_dev, NULL);
570         }
571
572 out:
573         write_unlock_bh(&bond->lock);
574 }
575
576 /*------------------------------- Link status -------------------------------*/
577
578 /*
579  * Set the carrier state for the master according to the state of its
580  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
581  * do special 802.3ad magic.
582  *
583  * Returns zero if carrier state does not change, nonzero if it does.
584  */
585 static int bond_set_carrier(struct bonding *bond)
586 {
587         struct slave *slave;
588         int i;
589
590         if (bond->slave_cnt == 0)
591                 goto down;
592
593         if (bond->params.mode == BOND_MODE_8023AD)
594                 return bond_3ad_set_carrier(bond);
595
596         bond_for_each_slave(bond, slave, i) {
597                 if (slave->link == BOND_LINK_UP) {
598                         if (!netif_carrier_ok(bond->dev)) {
599                                 netif_carrier_on(bond->dev);
600                                 return 1;
601                         }
602                         return 0;
603                 }
604         }
605
606 down:
607         if (netif_carrier_ok(bond->dev)) {
608                 netif_carrier_off(bond->dev);
609                 return 1;
610         }
611         return 0;
612 }
613
614 /*
615  * Get link speed and duplex from the slave's base driver
616  * using ethtool. If for some reason the call fails or the
617  * values are invalid, fake speed and duplex to 100/Full
618  * and return error.
619  */
620 static int bond_update_speed_duplex(struct slave *slave)
621 {
622         struct net_device *slave_dev = slave->dev;
623         struct ethtool_cmd etool;
624         int res;
625
626         /* Fake speed and duplex */
627         slave->speed = SPEED_100;
628         slave->duplex = DUPLEX_FULL;
629
630         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
631                 return -1;
632
633         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
634         if (res < 0)
635                 return -1;
636
637         switch (etool.speed) {
638         case SPEED_10:
639         case SPEED_100:
640         case SPEED_1000:
641         case SPEED_10000:
642                 break;
643         default:
644                 return -1;
645         }
646
647         switch (etool.duplex) {
648         case DUPLEX_FULL:
649         case DUPLEX_HALF:
650                 break;
651         default:
652                 return -1;
653         }
654
655         slave->speed = etool.speed;
656         slave->duplex = etool.duplex;
657
658         return 0;
659 }
660
661 /*
662  * if <dev> supports MII link status reporting, check its link status.
663  *
664  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
665  * depening upon the setting of the use_carrier parameter.
666  *
667  * Return either BMSR_LSTATUS, meaning that the link is up (or we
668  * can't tell and just pretend it is), or 0, meaning that the link is
669  * down.
670  *
671  * If reporting is non-zero, instead of faking link up, return -1 if
672  * both ETHTOOL and MII ioctls fail (meaning the device does not
673  * support them).  If use_carrier is set, return whatever it says.
674  * It'd be nice if there was a good way to tell if a driver supports
675  * netif_carrier, but there really isn't.
676  */
677 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
678 {
679         static int (* ioctl)(struct net_device *, struct ifreq *, int);
680         struct ifreq ifr;
681         struct mii_ioctl_data *mii;
682
683         if (bond->params.use_carrier) {
684                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
685         }
686
687         ioctl = slave_dev->do_ioctl;
688         if (ioctl) {
689                 /* TODO: set pointer to correct ioctl on a per team member */
690                 /*       bases to make this more efficient. that is, once  */
691                 /*       we determine the correct ioctl, we will always    */
692                 /*       call it and not the others for that team          */
693                 /*       member.                                           */
694
695                 /*
696                  * We cannot assume that SIOCGMIIPHY will also read a
697                  * register; not all network drivers (e.g., e100)
698                  * support that.
699                  */
700
701                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
702                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
703                 mii = if_mii(&ifr);
704                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
705                         mii->reg_num = MII_BMSR;
706                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
707                                 return (mii->val_out & BMSR_LSTATUS);
708                         }
709                 }
710         }
711
712         /*
713          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
714          * attempt to get link status from it if the above MII ioctls fail.
715          */
716         if (slave_dev->ethtool_ops) {
717                 if (slave_dev->ethtool_ops->get_link) {
718                         u32 link;
719
720                         link = slave_dev->ethtool_ops->get_link(slave_dev);
721
722                         return link ? BMSR_LSTATUS : 0;
723                 }
724         }
725
726         /*
727          * If reporting, report that either there's no dev->do_ioctl,
728          * or both SIOCGMIIREG and get_link failed (meaning that we
729          * cannot report link status).  If not reporting, pretend
730          * we're ok.
731          */
732         return (reporting ? -1 : BMSR_LSTATUS);
733 }
734
735 /*----------------------------- Multicast list ------------------------------*/
736
737 /*
738  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
739  */
740 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
741 {
742         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
743                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
744 }
745
746 /*
747  * returns dmi entry if found, NULL otherwise
748  */
749 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
750 {
751         struct dev_mc_list *idmi;
752
753         for (idmi = mc_list; idmi; idmi = idmi->next) {
754                 if (bond_is_dmi_same(dmi, idmi)) {
755                         return idmi;
756                 }
757         }
758
759         return NULL;
760 }
761
762 /*
763  * Push the promiscuity flag down to appropriate slaves
764  */
765 static void bond_set_promiscuity(struct bonding *bond, int inc)
766 {
767         if (USES_PRIMARY(bond->params.mode)) {
768                 /* write lock already acquired */
769                 if (bond->curr_active_slave) {
770                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
771                 }
772         } else {
773                 struct slave *slave;
774                 int i;
775                 bond_for_each_slave(bond, slave, i) {
776                         dev_set_promiscuity(slave->dev, inc);
777                 }
778         }
779 }
780
781 /*
782  * Push the allmulti flag down to all slaves
783  */
784 static void bond_set_allmulti(struct bonding *bond, int inc)
785 {
786         if (USES_PRIMARY(bond->params.mode)) {
787                 /* write lock already acquired */
788                 if (bond->curr_active_slave) {
789                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
790                 }
791         } else {
792                 struct slave *slave;
793                 int i;
794                 bond_for_each_slave(bond, slave, i) {
795                         dev_set_allmulti(slave->dev, inc);
796                 }
797         }
798 }
799
800 /*
801  * Add a Multicast address to slaves
802  * according to mode
803  */
804 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
805 {
806         if (USES_PRIMARY(bond->params.mode)) {
807                 /* write lock already acquired */
808                 if (bond->curr_active_slave) {
809                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
810                 }
811         } else {
812                 struct slave *slave;
813                 int i;
814                 bond_for_each_slave(bond, slave, i) {
815                         dev_mc_add(slave->dev, addr, alen, 0);
816                 }
817         }
818 }
819
820 /*
821  * Remove a multicast address from slave
822  * according to mode
823  */
824 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
825 {
826         if (USES_PRIMARY(bond->params.mode)) {
827                 /* write lock already acquired */
828                 if (bond->curr_active_slave) {
829                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
830                 }
831         } else {
832                 struct slave *slave;
833                 int i;
834                 bond_for_each_slave(bond, slave, i) {
835                         dev_mc_delete(slave->dev, addr, alen, 0);
836                 }
837         }
838 }
839
840
841 /*
842  * Retrieve the list of registered multicast addresses for the bonding
843  * device and retransmit an IGMP JOIN request to the current active
844  * slave.
845  */
846 static void bond_resend_igmp_join_requests(struct bonding *bond)
847 {
848         struct in_device *in_dev;
849         struct ip_mc_list *im;
850
851         rcu_read_lock();
852         in_dev = __in_dev_get_rcu(bond->dev);
853         if (in_dev) {
854                 for (im = in_dev->mc_list; im; im = im->next) {
855                         ip_mc_rejoin_group(im);
856                 }
857         }
858
859         rcu_read_unlock();
860 }
861
862 /*
863  * Totally destroys the mc_list in bond
864  */
865 static void bond_mc_list_destroy(struct bonding *bond)
866 {
867         struct dev_mc_list *dmi;
868
869         dmi = bond->mc_list;
870         while (dmi) {
871                 bond->mc_list = dmi->next;
872                 kfree(dmi);
873                 dmi = bond->mc_list;
874         }
875         bond->mc_list = NULL;
876 }
877
878 /*
879  * Copy all the Multicast addresses from src to the bonding device dst
880  */
881 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
882                              gfp_t gfp_flag)
883 {
884         struct dev_mc_list *dmi, *new_dmi;
885
886         for (dmi = mc_list; dmi; dmi = dmi->next) {
887                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
888
889                 if (!new_dmi) {
890                         /* FIXME: Potential memory leak !!! */
891                         return -ENOMEM;
892                 }
893
894                 new_dmi->next = bond->mc_list;
895                 bond->mc_list = new_dmi;
896                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
897                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
898                 new_dmi->dmi_users = dmi->dmi_users;
899                 new_dmi->dmi_gusers = dmi->dmi_gusers;
900         }
901
902         return 0;
903 }
904
905 /*
906  * flush all members of flush->mc_list from device dev->mc_list
907  */
908 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
909 {
910         struct bonding *bond = bond_dev->priv;
911         struct dev_mc_list *dmi;
912
913         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
914                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
915         }
916
917         if (bond->params.mode == BOND_MODE_8023AD) {
918                 /* del lacpdu mc addr from mc list */
919                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
920
921                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
922         }
923 }
924
925 /*--------------------------- Active slave change ---------------------------*/
926
927 /*
928  * Update the mc list and multicast-related flags for the new and
929  * old active slaves (if any) according to the multicast mode, and
930  * promiscuous flags unconditionally.
931  */
932 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
933 {
934         struct dev_mc_list *dmi;
935
936         if (!USES_PRIMARY(bond->params.mode)) {
937                 /* nothing to do -  mc list is already up-to-date on
938                  * all slaves
939                  */
940                 return;
941         }
942
943         if (old_active) {
944                 if (bond->dev->flags & IFF_PROMISC) {
945                         dev_set_promiscuity(old_active->dev, -1);
946                 }
947
948                 if (bond->dev->flags & IFF_ALLMULTI) {
949                         dev_set_allmulti(old_active->dev, -1);
950                 }
951
952                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
953                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
954                 }
955         }
956
957         if (new_active) {
958                 if (bond->dev->flags & IFF_PROMISC) {
959                         dev_set_promiscuity(new_active->dev, 1);
960                 }
961
962                 if (bond->dev->flags & IFF_ALLMULTI) {
963                         dev_set_allmulti(new_active->dev, 1);
964                 }
965
966                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
967                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
968                 }
969                 bond_resend_igmp_join_requests(bond);
970         }
971 }
972
973 /**
974  * find_best_interface - select the best available slave to be the active one
975  * @bond: our bonding struct
976  *
977  * Warning: Caller must hold curr_slave_lock for writing.
978  */
979 static struct slave *bond_find_best_slave(struct bonding *bond)
980 {
981         struct slave *new_active, *old_active;
982         struct slave *bestslave = NULL;
983         int mintime = bond->params.updelay;
984         int i;
985
986         new_active = old_active = bond->curr_active_slave;
987
988         if (!new_active) { /* there were no active slaves left */
989                 if (bond->slave_cnt > 0) {  /* found one slave */
990                         new_active = bond->first_slave;
991                 } else {
992                         return NULL; /* still no slave, return NULL */
993                 }
994         }
995
996         /* first try the primary link; if arping, a link must tx/rx traffic
997          * before it can be considered the curr_active_slave - also, we would skip
998          * slaves between the curr_active_slave and primary_slave that may be up
999          * and able to arp
1000          */
1001         if ((bond->primary_slave) &&
1002             (!bond->params.arp_interval) &&
1003             (IS_UP(bond->primary_slave->dev))) {
1004                 new_active = bond->primary_slave;
1005         }
1006
1007         /* remember where to stop iterating over the slaves */
1008         old_active = new_active;
1009
1010         bond_for_each_slave_from(bond, new_active, i, old_active) {
1011                 if (IS_UP(new_active->dev)) {
1012                         if (new_active->link == BOND_LINK_UP) {
1013                                 return new_active;
1014                         } else if (new_active->link == BOND_LINK_BACK) {
1015                                 /* link up, but waiting for stabilization */
1016                                 if (new_active->delay < mintime) {
1017                                         mintime = new_active->delay;
1018                                         bestslave = new_active;
1019                                 }
1020                         }
1021                 }
1022         }
1023
1024         return bestslave;
1025 }
1026
1027 /**
1028  * change_active_interface - change the active slave into the specified one
1029  * @bond: our bonding struct
1030  * @new: the new slave to make the active one
1031  *
1032  * Set the new slave to the bond's settings and unset them on the old
1033  * curr_active_slave.
1034  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1035  *
1036  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1037  * because it is apparently the best available slave we have, even though its
1038  * updelay hasn't timed out yet.
1039  *
1040  * Warning: Caller must hold curr_slave_lock for writing.
1041  */
1042 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1043 {
1044         struct slave *old_active = bond->curr_active_slave;
1045
1046         if (old_active == new_active) {
1047                 return;
1048         }
1049
1050         if (new_active) {
1051                 if (new_active->link == BOND_LINK_BACK) {
1052                         if (USES_PRIMARY(bond->params.mode)) {
1053                                 printk(KERN_INFO DRV_NAME
1054                                        ": %s: making interface %s the new "
1055                                        "active one %d ms earlier.\n",
1056                                        bond->dev->name, new_active->dev->name,
1057                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1058                         }
1059
1060                         new_active->delay = 0;
1061                         new_active->link = BOND_LINK_UP;
1062                         new_active->jiffies = jiffies;
1063
1064                         if (bond->params.mode == BOND_MODE_8023AD) {
1065                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1066                         }
1067
1068                         if ((bond->params.mode == BOND_MODE_TLB) ||
1069                             (bond->params.mode == BOND_MODE_ALB)) {
1070                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1071                         }
1072                 } else {
1073                         if (USES_PRIMARY(bond->params.mode)) {
1074                                 printk(KERN_INFO DRV_NAME
1075                                        ": %s: making interface %s the new "
1076                                        "active one.\n",
1077                                        bond->dev->name, new_active->dev->name);
1078                         }
1079                 }
1080         }
1081
1082         if (USES_PRIMARY(bond->params.mode)) {
1083                 bond_mc_swap(bond, new_active, old_active);
1084         }
1085
1086         if ((bond->params.mode == BOND_MODE_TLB) ||
1087             (bond->params.mode == BOND_MODE_ALB)) {
1088                 bond_alb_handle_active_change(bond, new_active);
1089                 if (old_active)
1090                         bond_set_slave_inactive_flags(old_active);
1091                 if (new_active)
1092                         bond_set_slave_active_flags(new_active);
1093         } else {
1094                 bond->curr_active_slave = new_active;
1095         }
1096
1097         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1098                 if (old_active) {
1099                         bond_set_slave_inactive_flags(old_active);
1100                 }
1101
1102                 if (new_active) {
1103                         bond_set_slave_active_flags(new_active);
1104                 }
1105
1106                 /* when bonding does not set the slave MAC address, the bond MAC
1107                  * address is the one of the active slave.
1108                  */
1109                 if (new_active && bond->params.fail_over_mac)
1110                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1111                                 new_active->dev->addr_len);
1112                 if (bond->curr_active_slave &&
1113                         test_bit(__LINK_STATE_LINKWATCH_PENDING,
1114                                         &bond->curr_active_slave->dev->state)) {
1115                         dprintk("delaying gratuitous arp on %s\n",
1116                                 bond->curr_active_slave->dev->name);
1117                         bond->send_grat_arp = 1;
1118                 } else
1119                         bond_send_gratuitous_arp(bond);
1120         }
1121 }
1122
1123 /**
1124  * bond_select_active_slave - select a new active slave, if needed
1125  * @bond: our bonding struct
1126  *
1127  * This functions shoud be called when one of the following occurs:
1128  * - The old curr_active_slave has been released or lost its link.
1129  * - The primary_slave has got its link back.
1130  * - A slave has got its link back and there's no old curr_active_slave.
1131  *
1132  * Warning: Caller must hold curr_slave_lock for writing.
1133  */
1134 void bond_select_active_slave(struct bonding *bond)
1135 {
1136         struct slave *best_slave;
1137         int rv;
1138
1139         best_slave = bond_find_best_slave(bond);
1140         if (best_slave != bond->curr_active_slave) {
1141                 bond_change_active_slave(bond, best_slave);
1142                 rv = bond_set_carrier(bond);
1143                 if (!rv)
1144                         return;
1145
1146                 if (netif_carrier_ok(bond->dev)) {
1147                         printk(KERN_INFO DRV_NAME
1148                                ": %s: first active interface up!\n",
1149                                bond->dev->name);
1150                 } else {
1151                         printk(KERN_INFO DRV_NAME ": %s: "
1152                                "now running without any active interface !\n",
1153                                bond->dev->name);
1154                 }
1155         }
1156 }
1157
1158 /*--------------------------- slave list handling ---------------------------*/
1159
1160 /*
1161  * This function attaches the slave to the end of list.
1162  *
1163  * bond->lock held for writing by caller.
1164  */
1165 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1166 {
1167         if (bond->first_slave == NULL) { /* attaching the first slave */
1168                 new_slave->next = new_slave;
1169                 new_slave->prev = new_slave;
1170                 bond->first_slave = new_slave;
1171         } else {
1172                 new_slave->next = bond->first_slave;
1173                 new_slave->prev = bond->first_slave->prev;
1174                 new_slave->next->prev = new_slave;
1175                 new_slave->prev->next = new_slave;
1176         }
1177
1178         bond->slave_cnt++;
1179 }
1180
1181 /*
1182  * This function detaches the slave from the list.
1183  * WARNING: no check is made to verify if the slave effectively
1184  * belongs to <bond>.
1185  * Nothing is freed on return, structures are just unchained.
1186  * If any slave pointer in bond was pointing to <slave>,
1187  * it should be changed by the calling function.
1188  *
1189  * bond->lock held for writing by caller.
1190  */
1191 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1192 {
1193         if (slave->next) {
1194                 slave->next->prev = slave->prev;
1195         }
1196
1197         if (slave->prev) {
1198                 slave->prev->next = slave->next;
1199         }
1200
1201         if (bond->first_slave == slave) { /* slave is the first slave */
1202                 if (bond->slave_cnt > 1) { /* there are more slave */
1203                         bond->first_slave = slave->next;
1204                 } else {
1205                         bond->first_slave = NULL; /* slave was the last one */
1206                 }
1207         }
1208
1209         slave->next = NULL;
1210         slave->prev = NULL;
1211         bond->slave_cnt--;
1212 }
1213
1214 /*---------------------------------- IOCTL ----------------------------------*/
1215
1216 static int bond_sethwaddr(struct net_device *bond_dev,
1217                           struct net_device *slave_dev)
1218 {
1219         dprintk("bond_dev=%p\n", bond_dev);
1220         dprintk("slave_dev=%p\n", slave_dev);
1221         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1222         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1223         return 0;
1224 }
1225
1226 #define BOND_VLAN_FEATURES \
1227         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1228          NETIF_F_HW_VLAN_FILTER)
1229
1230 /* 
1231  * Compute the common dev->feature set available to all slaves.  Some
1232  * feature bits are managed elsewhere, so preserve those feature bits
1233  * on the master device.
1234  */
1235 static int bond_compute_features(struct bonding *bond)
1236 {
1237         struct slave *slave;
1238         struct net_device *bond_dev = bond->dev;
1239         unsigned long features = bond_dev->features;
1240         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1241                                                 bond_dev->hard_header_len);
1242         int i;
1243
1244         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1245         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1246                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1247
1248         bond_for_each_slave(bond, slave, i) {
1249                 features = netdev_compute_features(features,
1250                                                    slave->dev->features);
1251                 if (slave->dev->hard_header_len > max_hard_header_len)
1252                         max_hard_header_len = slave->dev->hard_header_len;
1253         }
1254
1255         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1256         bond_dev->features = features;
1257         bond_dev->hard_header_len = max_hard_header_len;
1258
1259         return 0;
1260 }
1261
1262
1263 static void bond_setup_by_slave(struct net_device *bond_dev,
1264                                 struct net_device *slave_dev)
1265 {
1266         struct bonding *bond = bond_dev->priv;
1267
1268         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1269         bond_dev->header_ops            = slave_dev->header_ops;
1270
1271         bond_dev->type              = slave_dev->type;
1272         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1273         bond_dev->addr_len          = slave_dev->addr_len;
1274
1275         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1276                 slave_dev->addr_len);
1277         bond->setup_by_slave = 1;
1278 }
1279
1280 /* enslave device <slave> to bond device <master> */
1281 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1282 {
1283         struct bonding *bond = bond_dev->priv;
1284         struct slave *new_slave = NULL;
1285         struct dev_mc_list *dmi;
1286         struct sockaddr addr;
1287         int link_reporting;
1288         int old_features = bond_dev->features;
1289         int res = 0;
1290
1291         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1292                 slave_dev->do_ioctl == NULL) {
1293                 printk(KERN_WARNING DRV_NAME
1294                        ": %s: Warning: no link monitoring support for %s\n",
1295                        bond_dev->name, slave_dev->name);
1296         }
1297
1298         /* bond must be initialized by bond_open() before enslaving */
1299         if (!(bond_dev->flags & IFF_UP)) {
1300                 printk(KERN_WARNING DRV_NAME
1301                         " %s: master_dev is not up in bond_enslave\n",
1302                         bond_dev->name);
1303         }
1304
1305         /* already enslaved */
1306         if (slave_dev->flags & IFF_SLAVE) {
1307                 dprintk("Error, Device was already enslaved\n");
1308                 return -EBUSY;
1309         }
1310
1311         /* vlan challenged mutual exclusion */
1312         /* no need to lock since we're protected by rtnl_lock */
1313         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1314                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1315                 if (!list_empty(&bond->vlan_list)) {
1316                         printk(KERN_ERR DRV_NAME
1317                                ": %s: Error: cannot enslave VLAN "
1318                                "challenged slave %s on VLAN enabled "
1319                                "bond %s\n", bond_dev->name, slave_dev->name,
1320                                bond_dev->name);
1321                         return -EPERM;
1322                 } else {
1323                         printk(KERN_WARNING DRV_NAME
1324                                ": %s: Warning: enslaved VLAN challenged "
1325                                "slave %s. Adding VLANs will be blocked as "
1326                                "long as %s is part of bond %s\n",
1327                                bond_dev->name, slave_dev->name, slave_dev->name,
1328                                bond_dev->name);
1329                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1330                 }
1331         } else {
1332                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1333                 if (bond->slave_cnt == 0) {
1334                         /* First slave, and it is not VLAN challenged,
1335                          * so remove the block of adding VLANs over the bond.
1336                          */
1337                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1338                 }
1339         }
1340
1341         /*
1342          * Old ifenslave binaries are no longer supported.  These can
1343          * be identified with moderate accurary by the state of the slave:
1344          * the current ifenslave will set the interface down prior to
1345          * enslaving it; the old ifenslave will not.
1346          */
1347         if ((slave_dev->flags & IFF_UP)) {
1348                 printk(KERN_ERR DRV_NAME ": %s is up. "
1349                        "This may be due to an out of date ifenslave.\n",
1350                        slave_dev->name);
1351                 res = -EPERM;
1352                 goto err_undo_flags;
1353         }
1354
1355         /* set bonding device ether type by slave - bonding netdevices are
1356          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1357          * there is a need to override some of the type dependent attribs/funcs.
1358          *
1359          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1360          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1361          */
1362         if (bond->slave_cnt == 0) {
1363                 if (slave_dev->type != ARPHRD_ETHER)
1364                         bond_setup_by_slave(bond_dev, slave_dev);
1365         } else if (bond_dev->type != slave_dev->type) {
1366                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1367                         "from other slaves (%d), can not enslave it.\n",
1368                         slave_dev->name,
1369                         slave_dev->type, bond_dev->type);
1370                         res = -EINVAL;
1371                         goto err_undo_flags;
1372         }
1373
1374         if (slave_dev->set_mac_address == NULL) {
1375                 if (bond->slave_cnt == 0) {
1376                         printk(KERN_WARNING DRV_NAME
1377                                ": %s: Warning: The first slave device "
1378                                "specified does not support setting the MAC "
1379                                "address. Enabling the fail_over_mac option.",
1380                                bond_dev->name);
1381                         bond->params.fail_over_mac = 1;
1382                 } else if (!bond->params.fail_over_mac) {
1383                         printk(KERN_ERR DRV_NAME
1384                                 ": %s: Error: The slave device specified "
1385                                 "does not support setting the MAC address, "
1386                                 "but fail_over_mac is not enabled.\n"
1387                                 , bond_dev->name);
1388                         res = -EOPNOTSUPP;
1389                         goto err_undo_flags;
1390                 }
1391         }
1392
1393         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1394         if (!new_slave) {
1395                 res = -ENOMEM;
1396                 goto err_undo_flags;
1397         }
1398
1399         /* save slave's original flags before calling
1400          * netdev_set_master and dev_open
1401          */
1402         new_slave->original_flags = slave_dev->flags;
1403
1404         /*
1405          * Save slave's original ("permanent") mac address for modes
1406          * that need it, and for restoring it upon release, and then
1407          * set it to the master's address
1408          */
1409         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1410
1411         if (!bond->params.fail_over_mac) {
1412                 /*
1413                  * Set slave to master's mac address.  The application already
1414                  * set the master's mac address to that of the first slave
1415                  */
1416                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1417                 addr.sa_family = slave_dev->type;
1418                 res = dev_set_mac_address(slave_dev, &addr);
1419                 if (res) {
1420                         dprintk("Error %d calling set_mac_address\n", res);
1421                         goto err_free;
1422                 }
1423         }
1424
1425         res = netdev_set_master(slave_dev, bond_dev);
1426         if (res) {
1427                 dprintk("Error %d calling netdev_set_master\n", res);
1428                 goto err_close;
1429         }
1430         /* open the slave since the application closed it */
1431         res = dev_open(slave_dev);
1432         if (res) {
1433                 dprintk("Openning slave %s failed\n", slave_dev->name);
1434                 goto err_restore_mac;
1435         }
1436
1437         new_slave->dev = slave_dev;
1438         slave_dev->priv_flags |= IFF_BONDING;
1439
1440         if ((bond->params.mode == BOND_MODE_TLB) ||
1441             (bond->params.mode == BOND_MODE_ALB)) {
1442                 /* bond_alb_init_slave() must be called before all other stages since
1443                  * it might fail and we do not want to have to undo everything
1444                  */
1445                 res = bond_alb_init_slave(bond, new_slave);
1446                 if (res) {
1447                         goto err_unset_master;
1448                 }
1449         }
1450
1451         /* If the mode USES_PRIMARY, then the new slave gets the
1452          * master's promisc (and mc) settings only if it becomes the
1453          * curr_active_slave, and that is taken care of later when calling
1454          * bond_change_active()
1455          */
1456         if (!USES_PRIMARY(bond->params.mode)) {
1457                 /* set promiscuity level to new slave */
1458                 if (bond_dev->flags & IFF_PROMISC) {
1459                         dev_set_promiscuity(slave_dev, 1);
1460                 }
1461
1462                 /* set allmulti level to new slave */
1463                 if (bond_dev->flags & IFF_ALLMULTI) {
1464                         dev_set_allmulti(slave_dev, 1);
1465                 }
1466
1467                 /* upload master's mc_list to new slave */
1468                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1469                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1470                 }
1471         }
1472
1473         if (bond->params.mode == BOND_MODE_8023AD) {
1474                 /* add lacpdu mc addr to mc list */
1475                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1476
1477                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1478         }
1479
1480         bond_add_vlans_on_slave(bond, slave_dev);
1481
1482         write_lock_bh(&bond->lock);
1483
1484         bond_attach_slave(bond, new_slave);
1485
1486         new_slave->delay = 0;
1487         new_slave->link_failure_count = 0;
1488
1489         bond_compute_features(bond);
1490
1491         new_slave->last_arp_rx = jiffies;
1492
1493         if (bond->params.miimon && !bond->params.use_carrier) {
1494                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1495
1496                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1497                         /*
1498                          * miimon is set but a bonded network driver
1499                          * does not support ETHTOOL/MII and
1500                          * arp_interval is not set.  Note: if
1501                          * use_carrier is enabled, we will never go
1502                          * here (because netif_carrier is always
1503                          * supported); thus, we don't need to change
1504                          * the messages for netif_carrier.
1505                          */
1506                         printk(KERN_WARNING DRV_NAME
1507                                ": %s: Warning: MII and ETHTOOL support not "
1508                                "available for interface %s, and "
1509                                "arp_interval/arp_ip_target module parameters "
1510                                "not specified, thus bonding will not detect "
1511                                "link failures! see bonding.txt for details.\n",
1512                                bond_dev->name, slave_dev->name);
1513                 } else if (link_reporting == -1) {
1514                         /* unable get link status using mii/ethtool */
1515                         printk(KERN_WARNING DRV_NAME
1516                                ": %s: Warning: can't get link status from "
1517                                "interface %s; the network driver associated "
1518                                "with this interface does not support MII or "
1519                                "ETHTOOL link status reporting, thus miimon "
1520                                "has no effect on this interface.\n",
1521                                bond_dev->name, slave_dev->name);
1522                 }
1523         }
1524
1525         /* check for initial state */
1526         if (!bond->params.miimon ||
1527             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1528                 if (bond->params.updelay) {
1529                         dprintk("Initial state of slave_dev is "
1530                                 "BOND_LINK_BACK\n");
1531                         new_slave->link  = BOND_LINK_BACK;
1532                         new_slave->delay = bond->params.updelay;
1533                 } else {
1534                         dprintk("Initial state of slave_dev is "
1535                                 "BOND_LINK_UP\n");
1536                         new_slave->link  = BOND_LINK_UP;
1537                 }
1538                 new_slave->jiffies = jiffies;
1539         } else {
1540                 dprintk("Initial state of slave_dev is "
1541                         "BOND_LINK_DOWN\n");
1542                 new_slave->link  = BOND_LINK_DOWN;
1543         }
1544
1545         if (bond_update_speed_duplex(new_slave) &&
1546             (new_slave->link != BOND_LINK_DOWN)) {
1547                 printk(KERN_WARNING DRV_NAME
1548                        ": %s: Warning: failed to get speed and duplex from %s, "
1549                        "assumed to be 100Mb/sec and Full.\n",
1550                        bond_dev->name, new_slave->dev->name);
1551
1552                 if (bond->params.mode == BOND_MODE_8023AD) {
1553                         printk(KERN_WARNING DRV_NAME
1554                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1555                                "support in base driver for proper aggregator "
1556                                "selection.\n", bond_dev->name);
1557                 }
1558         }
1559
1560         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1561                 /* if there is a primary slave, remember it */
1562                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1563                         bond->primary_slave = new_slave;
1564                 }
1565         }
1566
1567         switch (bond->params.mode) {
1568         case BOND_MODE_ACTIVEBACKUP:
1569                 bond_set_slave_inactive_flags(new_slave);
1570                 bond_select_active_slave(bond);
1571                 break;
1572         case BOND_MODE_8023AD:
1573                 /* in 802.3ad mode, the internal mechanism
1574                  * will activate the slaves in the selected
1575                  * aggregator
1576                  */
1577                 bond_set_slave_inactive_flags(new_slave);
1578                 /* if this is the first slave */
1579                 if (bond->slave_cnt == 1) {
1580                         SLAVE_AD_INFO(new_slave).id = 1;
1581                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1582                          * can be called only after the mac address of the bond is set
1583                          */
1584                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1585                                             bond->params.lacp_fast);
1586                 } else {
1587                         SLAVE_AD_INFO(new_slave).id =
1588                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1589                 }
1590
1591                 bond_3ad_bind_slave(new_slave);
1592                 break;
1593         case BOND_MODE_TLB:
1594         case BOND_MODE_ALB:
1595                 new_slave->state = BOND_STATE_ACTIVE;
1596                 bond_set_slave_inactive_flags(new_slave);
1597                 break;
1598         default:
1599                 dprintk("This slave is always active in trunk mode\n");
1600
1601                 /* always active in trunk mode */
1602                 new_slave->state = BOND_STATE_ACTIVE;
1603
1604                 /* In trunking mode there is little meaning to curr_active_slave
1605                  * anyway (it holds no special properties of the bond device),
1606                  * so we can change it without calling change_active_interface()
1607                  */
1608                 if (!bond->curr_active_slave) {
1609                         bond->curr_active_slave = new_slave;
1610                 }
1611                 break;
1612         } /* switch(bond_mode) */
1613
1614         bond_set_carrier(bond);
1615
1616         write_unlock_bh(&bond->lock);
1617
1618         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1619         if (res)
1620                 goto err_unset_master;
1621
1622         printk(KERN_INFO DRV_NAME
1623                ": %s: enslaving %s as a%s interface with a%s link.\n",
1624                bond_dev->name, slave_dev->name,
1625                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1626                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1627
1628         /* enslave is successful */
1629         return 0;
1630
1631 /* Undo stages on error */
1632 err_unset_master:
1633         netdev_set_master(slave_dev, NULL);
1634
1635 err_close:
1636         dev_close(slave_dev);
1637
1638 err_restore_mac:
1639         if (!bond->params.fail_over_mac) {
1640                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1641                 addr.sa_family = slave_dev->type;
1642                 dev_set_mac_address(slave_dev, &addr);
1643         }
1644
1645 err_free:
1646         kfree(new_slave);
1647
1648 err_undo_flags:
1649         bond_dev->features = old_features;
1650  
1651         return res;
1652 }
1653
1654 /*
1655  * Try to release the slave device <slave> from the bond device <master>
1656  * It is legal to access curr_active_slave without a lock because all the function
1657  * is write-locked.
1658  *
1659  * The rules for slave state should be:
1660  *   for Active/Backup:
1661  *     Active stays on all backups go down
1662  *   for Bonded connections:
1663  *     The first up interface should be left on and all others downed.
1664  */
1665 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1666 {
1667         struct bonding *bond = bond_dev->priv;
1668         struct slave *slave, *oldcurrent;
1669         struct sockaddr addr;
1670         int mac_addr_differ;
1671         DECLARE_MAC_BUF(mac);
1672
1673         /* slave is not a slave or master is not master of this slave */
1674         if (!(slave_dev->flags & IFF_SLAVE) ||
1675             (slave_dev->master != bond_dev)) {
1676                 printk(KERN_ERR DRV_NAME
1677                        ": %s: Error: cannot release %s.\n",
1678                        bond_dev->name, slave_dev->name);
1679                 return -EINVAL;
1680         }
1681
1682         write_lock_bh(&bond->lock);
1683
1684         slave = bond_get_slave_by_dev(bond, slave_dev);
1685         if (!slave) {
1686                 /* not a slave of this bond */
1687                 printk(KERN_INFO DRV_NAME
1688                        ": %s: %s not enslaved\n",
1689                        bond_dev->name, slave_dev->name);
1690                 write_unlock_bh(&bond->lock);
1691                 return -EINVAL;
1692         }
1693
1694         mac_addr_differ = memcmp(bond_dev->dev_addr,
1695                                  slave->perm_hwaddr,
1696                                  ETH_ALEN);
1697         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1698                 printk(KERN_WARNING DRV_NAME
1699                        ": %s: Warning: the permanent HWaddr of %s - "
1700                        "%s - is still in use by %s. "
1701                        "Set the HWaddr of %s to a different address "
1702                        "to avoid conflicts.\n",
1703                        bond_dev->name,
1704                        slave_dev->name,
1705                        print_mac(mac, slave->perm_hwaddr),
1706                        bond_dev->name,
1707                        slave_dev->name);
1708         }
1709
1710         /* Inform AD package of unbinding of slave. */
1711         if (bond->params.mode == BOND_MODE_8023AD) {
1712                 /* must be called before the slave is
1713                  * detached from the list
1714                  */
1715                 bond_3ad_unbind_slave(slave);
1716         }
1717
1718         printk(KERN_INFO DRV_NAME
1719                ": %s: releasing %s interface %s\n",
1720                bond_dev->name,
1721                (slave->state == BOND_STATE_ACTIVE)
1722                ? "active" : "backup",
1723                slave_dev->name);
1724
1725         oldcurrent = bond->curr_active_slave;
1726
1727         bond->current_arp_slave = NULL;
1728
1729         /* release the slave from its bond */
1730         bond_detach_slave(bond, slave);
1731
1732         bond_compute_features(bond);
1733
1734         if (bond->primary_slave == slave) {
1735                 bond->primary_slave = NULL;
1736         }
1737
1738         if (oldcurrent == slave) {
1739                 bond_change_active_slave(bond, NULL);
1740         }
1741
1742         if ((bond->params.mode == BOND_MODE_TLB) ||
1743             (bond->params.mode == BOND_MODE_ALB)) {
1744                 /* Must be called only after the slave has been
1745                  * detached from the list and the curr_active_slave
1746                  * has been cleared (if our_slave == old_current),
1747                  * but before a new active slave is selected.
1748                  */
1749                 bond_alb_deinit_slave(bond, slave);
1750         }
1751
1752         if (oldcurrent == slave) {
1753                 /*
1754                  * Note that we hold RTNL over this sequence, so there
1755                  * is no concern that another slave add/remove event
1756                  * will interfere.
1757                  */
1758                 write_unlock_bh(&bond->lock);
1759                 read_lock(&bond->lock);
1760                 write_lock_bh(&bond->curr_slave_lock);
1761
1762                 bond_select_active_slave(bond);
1763
1764                 write_unlock_bh(&bond->curr_slave_lock);
1765                 read_unlock(&bond->lock);
1766                 write_lock_bh(&bond->lock);
1767         }
1768
1769         if (bond->slave_cnt == 0) {
1770                 bond_set_carrier(bond);
1771
1772                 /* if the last slave was removed, zero the mac address
1773                  * of the master so it will be set by the application
1774                  * to the mac address of the first slave
1775                  */
1776                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1777
1778                 if (list_empty(&bond->vlan_list)) {
1779                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1780                 } else {
1781                         printk(KERN_WARNING DRV_NAME
1782                                ": %s: Warning: clearing HW address of %s while it "
1783                                "still has VLANs.\n",
1784                                bond_dev->name, bond_dev->name);
1785                         printk(KERN_WARNING DRV_NAME
1786                                ": %s: When re-adding slaves, make sure the bond's "
1787                                "HW address matches its VLANs'.\n",
1788                                bond_dev->name);
1789                 }
1790         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1791                    !bond_has_challenged_slaves(bond)) {
1792                 printk(KERN_INFO DRV_NAME
1793                        ": %s: last VLAN challenged slave %s "
1794                        "left bond %s. VLAN blocking is removed\n",
1795                        bond_dev->name, slave_dev->name, bond_dev->name);
1796                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1797         }
1798
1799         write_unlock_bh(&bond->lock);
1800
1801         /* must do this from outside any spinlocks */
1802         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1803
1804         bond_del_vlans_from_slave(bond, slave_dev);
1805
1806         /* If the mode USES_PRIMARY, then we should only remove its
1807          * promisc and mc settings if it was the curr_active_slave, but that was
1808          * already taken care of above when we detached the slave
1809          */
1810         if (!USES_PRIMARY(bond->params.mode)) {
1811                 /* unset promiscuity level from slave */
1812                 if (bond_dev->flags & IFF_PROMISC) {
1813                         dev_set_promiscuity(slave_dev, -1);
1814                 }
1815
1816                 /* unset allmulti level from slave */
1817                 if (bond_dev->flags & IFF_ALLMULTI) {
1818                         dev_set_allmulti(slave_dev, -1);
1819                 }
1820
1821                 /* flush master's mc_list from slave */
1822                 bond_mc_list_flush(bond_dev, slave_dev);
1823         }
1824
1825         netdev_set_master(slave_dev, NULL);
1826
1827         /* close slave before restoring its mac address */
1828         dev_close(slave_dev);
1829
1830         if (!bond->params.fail_over_mac) {
1831                 /* restore original ("permanent") mac address */
1832                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1833                 addr.sa_family = slave_dev->type;
1834                 dev_set_mac_address(slave_dev, &addr);
1835         }
1836
1837         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1838                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1839                                    IFF_SLAVE_NEEDARP);
1840
1841         kfree(slave);
1842
1843         return 0;  /* deletion OK */
1844 }
1845
1846 /*
1847 * Destroy a bonding device.
1848 * Must be under rtnl_lock when this function is called.
1849 */
1850 void bond_destroy(struct bonding *bond)
1851 {
1852         bond_deinit(bond->dev);
1853         bond_destroy_sysfs_entry(bond);
1854         unregister_netdevice(bond->dev);
1855 }
1856
1857 /*
1858 * First release a slave and than destroy the bond if no more slaves iare left.
1859 * Must be under rtnl_lock when this function is called.
1860 */
1861 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1862 {
1863         struct bonding *bond = bond_dev->priv;
1864         int ret;
1865
1866         ret = bond_release(bond_dev, slave_dev);
1867         if ((ret == 0) && (bond->slave_cnt == 0)) {
1868                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1869                        bond_dev->name, bond_dev->name);
1870                 bond_destroy(bond);
1871         }
1872         return ret;
1873 }
1874
1875 /*
1876  * This function releases all slaves.
1877  */
1878 static int bond_release_all(struct net_device *bond_dev)
1879 {
1880         struct bonding *bond = bond_dev->priv;
1881         struct slave *slave;
1882         struct net_device *slave_dev;
1883         struct sockaddr addr;
1884
1885         write_lock_bh(&bond->lock);
1886
1887         netif_carrier_off(bond_dev);
1888
1889         if (bond->slave_cnt == 0) {
1890                 goto out;
1891         }
1892
1893         bond->current_arp_slave = NULL;
1894         bond->primary_slave = NULL;
1895         bond_change_active_slave(bond, NULL);
1896
1897         while ((slave = bond->first_slave) != NULL) {
1898                 /* Inform AD package of unbinding of slave
1899                  * before slave is detached from the list.
1900                  */
1901                 if (bond->params.mode == BOND_MODE_8023AD) {
1902                         bond_3ad_unbind_slave(slave);
1903                 }
1904
1905                 slave_dev = slave->dev;
1906                 bond_detach_slave(bond, slave);
1907
1908                 if ((bond->params.mode == BOND_MODE_TLB) ||
1909                     (bond->params.mode == BOND_MODE_ALB)) {
1910                         /* must be called only after the slave
1911                          * has been detached from the list
1912                          */
1913                         bond_alb_deinit_slave(bond, slave);
1914                 }
1915
1916                 bond_compute_features(bond);
1917
1918                 /* now that the slave is detached, unlock and perform
1919                  * all the undo steps that should not be called from
1920                  * within a lock.
1921                  */
1922                 write_unlock_bh(&bond->lock);
1923
1924                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1925                 bond_del_vlans_from_slave(bond, slave_dev);
1926
1927                 /* If the mode USES_PRIMARY, then we should only remove its
1928                  * promisc and mc settings if it was the curr_active_slave, but that was
1929                  * already taken care of above when we detached the slave
1930                  */
1931                 if (!USES_PRIMARY(bond->params.mode)) {
1932                         /* unset promiscuity level from slave */
1933                         if (bond_dev->flags & IFF_PROMISC) {
1934                                 dev_set_promiscuity(slave_dev, -1);
1935                         }
1936
1937                         /* unset allmulti level from slave */
1938                         if (bond_dev->flags & IFF_ALLMULTI) {
1939                                 dev_set_allmulti(slave_dev, -1);
1940                         }
1941
1942                         /* flush master's mc_list from slave */
1943                         bond_mc_list_flush(bond_dev, slave_dev);
1944                 }
1945
1946                 netdev_set_master(slave_dev, NULL);
1947
1948                 /* close slave before restoring its mac address */
1949                 dev_close(slave_dev);
1950
1951                 if (!bond->params.fail_over_mac) {
1952                         /* restore original ("permanent") mac address*/
1953                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1954                         addr.sa_family = slave_dev->type;
1955                         dev_set_mac_address(slave_dev, &addr);
1956                 }
1957
1958                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1959                                            IFF_SLAVE_INACTIVE);
1960
1961                 kfree(slave);
1962
1963                 /* re-acquire the lock before getting the next slave */
1964                 write_lock_bh(&bond->lock);
1965         }
1966
1967         /* zero the mac address of the master so it will be
1968          * set by the application to the mac address of the
1969          * first slave
1970          */
1971         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1972
1973         if (list_empty(&bond->vlan_list)) {
1974                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1975         } else {
1976                 printk(KERN_WARNING DRV_NAME
1977                        ": %s: Warning: clearing HW address of %s while it "
1978                        "still has VLANs.\n",
1979                        bond_dev->name, bond_dev->name);
1980                 printk(KERN_WARNING DRV_NAME
1981                        ": %s: When re-adding slaves, make sure the bond's "
1982                        "HW address matches its VLANs'.\n",
1983                        bond_dev->name);
1984         }
1985
1986         printk(KERN_INFO DRV_NAME
1987                ": %s: released all slaves\n",
1988                bond_dev->name);
1989
1990 out:
1991         write_unlock_bh(&bond->lock);
1992
1993         return 0;
1994 }
1995
1996 /*
1997  * This function changes the active slave to slave <slave_dev>.
1998  * It returns -EINVAL in the following cases.
1999  *  - <slave_dev> is not found in the list.
2000  *  - There is not active slave now.
2001  *  - <slave_dev> is already active.
2002  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2003  *  - <slave_dev> is not running.
2004  * In these cases, this fuction does nothing.
2005  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2006  */
2007 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2008 {
2009         struct bonding *bond = bond_dev->priv;
2010         struct slave *old_active = NULL;
2011         struct slave *new_active = NULL;
2012         int res = 0;
2013
2014         if (!USES_PRIMARY(bond->params.mode)) {
2015                 return -EINVAL;
2016         }
2017
2018         /* Verify that master_dev is indeed the master of slave_dev */
2019         if (!(slave_dev->flags & IFF_SLAVE) ||
2020             (slave_dev->master != bond_dev)) {
2021                 return -EINVAL;
2022         }
2023
2024         read_lock(&bond->lock);
2025
2026         read_lock(&bond->curr_slave_lock);
2027         old_active = bond->curr_active_slave;
2028         read_unlock(&bond->curr_slave_lock);
2029
2030         new_active = bond_get_slave_by_dev(bond, slave_dev);
2031
2032         /*
2033          * Changing to the current active: do nothing; return success.
2034          */
2035         if (new_active && (new_active == old_active)) {
2036                 read_unlock(&bond->lock);
2037                 return 0;
2038         }
2039
2040         if ((new_active) &&
2041             (old_active) &&
2042             (new_active->link == BOND_LINK_UP) &&
2043             IS_UP(new_active->dev)) {
2044                 write_lock_bh(&bond->curr_slave_lock);
2045                 bond_change_active_slave(bond, new_active);
2046                 write_unlock_bh(&bond->curr_slave_lock);
2047         } else {
2048                 res = -EINVAL;
2049         }
2050
2051         read_unlock(&bond->lock);
2052
2053         return res;
2054 }
2055
2056 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2057 {
2058         struct bonding *bond = bond_dev->priv;
2059
2060         info->bond_mode = bond->params.mode;
2061         info->miimon = bond->params.miimon;
2062
2063         read_lock(&bond->lock);
2064         info->num_slaves = bond->slave_cnt;
2065         read_unlock(&bond->lock);
2066
2067         return 0;
2068 }
2069
2070 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2071 {
2072         struct bonding *bond = bond_dev->priv;
2073         struct slave *slave;
2074         int i, found = 0;
2075
2076         if (info->slave_id < 0) {
2077                 return -ENODEV;
2078         }
2079
2080         read_lock(&bond->lock);
2081
2082         bond_for_each_slave(bond, slave, i) {
2083                 if (i == (int)info->slave_id) {
2084                         found = 1;
2085                         break;
2086                 }
2087         }
2088
2089         read_unlock(&bond->lock);
2090
2091         if (found) {
2092                 strcpy(info->slave_name, slave->dev->name);
2093                 info->link = slave->link;
2094                 info->state = slave->state;
2095                 info->link_failure_count = slave->link_failure_count;
2096         } else {
2097                 return -ENODEV;
2098         }
2099
2100         return 0;
2101 }
2102
2103 /*-------------------------------- Monitoring -------------------------------*/
2104
2105 /*
2106  * if !have_locks, return nonzero if a failover is necessary.  if
2107  * have_locks, do whatever failover activities are needed.
2108  *
2109  * This is to separate the inspection and failover steps for locking
2110  * purposes; failover requires rtnl, but acquiring it for every
2111  * inspection is undesirable, so a wrapper first does inspection, and
2112  * the acquires the necessary locks and calls again to perform
2113  * failover if needed.  Since all locks are dropped, a complete
2114  * restart is needed between calls.
2115  */
2116 static int __bond_mii_monitor(struct bonding *bond, int have_locks)
2117 {
2118         struct slave *slave, *oldcurrent;
2119         int do_failover = 0;
2120         int i;
2121
2122         if (bond->slave_cnt == 0)
2123                 goto out;
2124
2125         /* we will try to read the link status of each of our slaves, and
2126          * set their IFF_RUNNING flag appropriately. For each slave not
2127          * supporting MII status, we won't do anything so that a user-space
2128          * program could monitor the link itself if needed.
2129          */
2130
2131         if (bond->send_grat_arp) {
2132                 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2133                                 &bond->curr_active_slave->dev->state))
2134                         dprintk("Needs to send gratuitous arp but not yet\n");
2135                 else {
2136                         dprintk("sending delayed gratuitous arp on on %s\n",
2137                                 bond->curr_active_slave->dev->name);
2138                         bond_send_gratuitous_arp(bond);
2139                         bond->send_grat_arp = 0;
2140                 }
2141         }
2142         read_lock(&bond->curr_slave_lock);
2143         oldcurrent = bond->curr_active_slave;
2144         read_unlock(&bond->curr_slave_lock);
2145
2146         bond_for_each_slave(bond, slave, i) {
2147                 struct net_device *slave_dev = slave->dev;
2148                 int link_state;
2149                 u16 old_speed = slave->speed;
2150                 u8 old_duplex = slave->duplex;
2151
2152                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2153
2154                 switch (slave->link) {
2155                 case BOND_LINK_UP:      /* the link was up */
2156                         if (link_state == BMSR_LSTATUS) {
2157                                 if (!oldcurrent) {
2158                                         if (!have_locks)
2159                                                 return 1;
2160                                         do_failover = 1;
2161                                 }
2162                                 break;
2163                         } else { /* link going down */
2164                                 slave->link  = BOND_LINK_FAIL;
2165                                 slave->delay = bond->params.downdelay;
2166
2167                                 if (slave->link_failure_count < UINT_MAX) {
2168                                         slave->link_failure_count++;
2169                                 }
2170
2171                                 if (bond->params.downdelay) {
2172                                         printk(KERN_INFO DRV_NAME
2173                                                ": %s: link status down for %s "
2174                                                "interface %s, disabling it in "
2175                                                "%d ms.\n",
2176                                                bond->dev->name,
2177                                                IS_UP(slave_dev)
2178                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2179                                                   ? ((slave == oldcurrent)
2180                                                      ? "active " : "backup ")
2181                                                   : "")
2182                                                : "idle ",
2183                                                slave_dev->name,
2184                                                bond->params.downdelay * bond->params.miimon);
2185                                 }
2186                         }
2187                         /* no break ! fall through the BOND_LINK_FAIL test to
2188                            ensure proper action to be taken
2189                         */
2190                 case BOND_LINK_FAIL:    /* the link has just gone down */
2191                         if (link_state != BMSR_LSTATUS) {
2192                                 /* link stays down */
2193                                 if (slave->delay <= 0) {
2194                                         if (!have_locks)
2195                                                 return 1;
2196
2197                                         /* link down for too long time */
2198                                         slave->link = BOND_LINK_DOWN;
2199
2200                                         /* in active/backup mode, we must
2201                                          * completely disable this interface
2202                                          */
2203                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2204                                             (bond->params.mode == BOND_MODE_8023AD)) {
2205                                                 bond_set_slave_inactive_flags(slave);
2206                                         }
2207
2208                                         printk(KERN_INFO DRV_NAME
2209                                                ": %s: link status definitely "
2210                                                "down for interface %s, "
2211                                                "disabling it\n",
2212                                                bond->dev->name,
2213                                                slave_dev->name);
2214
2215                                         /* notify ad that the link status has changed */
2216                                         if (bond->params.mode == BOND_MODE_8023AD) {
2217                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2218                                         }
2219
2220                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2221                                             (bond->params.mode == BOND_MODE_ALB)) {
2222                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2223                                         }
2224
2225                                         if (slave == oldcurrent) {
2226                                                 do_failover = 1;
2227                                         }
2228                                 } else {
2229                                         slave->delay--;
2230                                 }
2231                         } else {
2232                                 /* link up again */
2233                                 slave->link  = BOND_LINK_UP;
2234                                 slave->jiffies = jiffies;
2235                                 printk(KERN_INFO DRV_NAME
2236                                        ": %s: link status up again after %d "
2237                                        "ms for interface %s.\n",
2238                                        bond->dev->name,
2239                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2240                                        slave_dev->name);
2241                         }
2242                         break;
2243                 case BOND_LINK_DOWN:    /* the link was down */
2244                         if (link_state != BMSR_LSTATUS) {
2245                                 /* the link stays down, nothing more to do */
2246                                 break;
2247                         } else {        /* link going up */
2248                                 slave->link  = BOND_LINK_BACK;
2249                                 slave->delay = bond->params.updelay;
2250
2251                                 if (bond->params.updelay) {
2252                                         /* if updelay == 0, no need to
2253                                            advertise about a 0 ms delay */
2254                                         printk(KERN_INFO DRV_NAME
2255                                                ": %s: link status up for "
2256                                                "interface %s, enabling it "
2257                                                "in %d ms.\n",
2258                                                bond->dev->name,
2259                                                slave_dev->name,
2260                                                bond->params.updelay * bond->params.miimon);
2261                                 }
2262                         }
2263                         /* no break ! fall through the BOND_LINK_BACK state in
2264                            case there's something to do.
2265                         */
2266                 case BOND_LINK_BACK:    /* the link has just come back */
2267                         if (link_state != BMSR_LSTATUS) {
2268                                 /* link down again */
2269                                 slave->link  = BOND_LINK_DOWN;
2270
2271                                 printk(KERN_INFO DRV_NAME
2272                                        ": %s: link status down again after %d "
2273                                        "ms for interface %s.\n",
2274                                        bond->dev->name,
2275                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2276                                        slave_dev->name);
2277                         } else {
2278                                 /* link stays up */
2279                                 if (slave->delay == 0) {
2280                                         if (!have_locks)
2281                                                 return 1;
2282
2283                                         /* now the link has been up for long time enough */
2284                                         slave->link = BOND_LINK_UP;
2285                                         slave->jiffies = jiffies;
2286
2287                                         if (bond->params.mode == BOND_MODE_8023AD) {
2288                                                 /* prevent it from being the active one */
2289                                                 slave->state = BOND_STATE_BACKUP;
2290                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2291                                                 /* make it immediately active */
2292                                                 slave->state = BOND_STATE_ACTIVE;
2293                                         } else if (slave != bond->primary_slave) {
2294                                                 /* prevent it from being the active one */
2295                                                 slave->state = BOND_STATE_BACKUP;
2296                                         }
2297
2298                                         printk(KERN_INFO DRV_NAME
2299                                                ": %s: link status definitely "
2300                                                "up for interface %s.\n",
2301                                                bond->dev->name,
2302                                                slave_dev->name);
2303
2304                                         /* notify ad that the link status has changed */
2305                                         if (bond->params.mode == BOND_MODE_8023AD) {
2306                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2307                                         }
2308
2309                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2310                                             (bond->params.mode == BOND_MODE_ALB)) {
2311                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2312                                         }
2313
2314                                         if ((!oldcurrent) ||
2315                                             (slave == bond->primary_slave)) {
2316                                                 do_failover = 1;
2317                                         }
2318                                 } else {
2319                                         slave->delay--;
2320                                 }
2321                         }
2322                         break;
2323                 default:
2324                         /* Should not happen */
2325                         printk(KERN_ERR DRV_NAME
2326                                ": %s: Error: %s Illegal value (link=%d)\n",
2327                                bond->dev->name,
2328                                slave->dev->name,
2329                                slave->link);
2330                         goto out;
2331                 } /* end of switch (slave->link) */
2332
2333                 bond_update_speed_duplex(slave);
2334
2335                 if (bond->params.mode == BOND_MODE_8023AD) {
2336                         if (old_speed != slave->speed) {
2337                                 bond_3ad_adapter_speed_changed(slave);
2338                         }
2339
2340                         if (old_duplex != slave->duplex) {
2341                                 bond_3ad_adapter_duplex_changed(slave);
2342                         }
2343                 }
2344
2345         } /* end of for */
2346
2347         if (do_failover) {
2348                 ASSERT_RTNL();
2349
2350                 write_lock_bh(&bond->curr_slave_lock);
2351
2352                 bond_select_active_slave(bond);
2353
2354                 write_unlock_bh(&bond->curr_slave_lock);
2355
2356         } else
2357                 bond_set_carrier(bond);
2358
2359 out:
2360         return 0;
2361 }
2362
2363 /*
2364  * bond_mii_monitor
2365  *
2366  * Really a wrapper that splits the mii monitor into two phases: an
2367  * inspection, then (if inspection indicates something needs to be
2368  * done) an acquisition of appropriate locks followed by another pass
2369  * to implement whatever link state changes are indicated.
2370  */
2371 void bond_mii_monitor(struct work_struct *work)
2372 {
2373         struct bonding *bond = container_of(work, struct bonding,
2374                                             mii_work.work);
2375         unsigned long delay;
2376
2377         read_lock(&bond->lock);
2378         if (bond->kill_timers) {
2379                 read_unlock(&bond->lock);
2380                 return;
2381         }
2382         if (__bond_mii_monitor(bond, 0)) {
2383                 read_unlock(&bond->lock);
2384                 rtnl_lock();
2385                 read_lock(&bond->lock);
2386                 __bond_mii_monitor(bond, 1);
2387                 rtnl_unlock();
2388         }
2389
2390         delay = ((bond->params.miimon * HZ) / 1000) ? : 1;
2391         read_unlock(&bond->lock);
2392         queue_delayed_work(bond->wq, &bond->mii_work, delay);
2393 }
2394
2395 static __be32 bond_glean_dev_ip(struct net_device *dev)
2396 {
2397         struct in_device *idev;
2398         struct in_ifaddr *ifa;
2399         __be32 addr = 0;
2400
2401         if (!dev)
2402                 return 0;
2403
2404         rcu_read_lock();
2405         idev = __in_dev_get_rcu(dev);
2406         if (!idev)
2407                 goto out;
2408
2409         ifa = idev->ifa_list;
2410         if (!ifa)
2411                 goto out;
2412
2413         addr = ifa->ifa_local;
2414 out:
2415         rcu_read_unlock();
2416         return addr;
2417 }
2418
2419 static int bond_has_ip(struct bonding *bond)
2420 {
2421         struct vlan_entry *vlan, *vlan_next;
2422
2423         if (bond->master_ip)
2424                 return 1;
2425
2426         if (list_empty(&bond->vlan_list))
2427                 return 0;
2428
2429         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2430                                  vlan_list) {
2431                 if (vlan->vlan_ip)
2432                         return 1;
2433         }
2434
2435         return 0;
2436 }
2437
2438 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2439 {
2440         struct vlan_entry *vlan, *vlan_next;
2441
2442         if (ip == bond->master_ip)
2443                 return 1;
2444
2445         if (list_empty(&bond->vlan_list))
2446                 return 0;
2447
2448         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2449                                  vlan_list) {
2450                 if (ip == vlan->vlan_ip)
2451                         return 1;
2452         }
2453
2454         return 0;
2455 }
2456
2457 /*
2458  * We go to the (large) trouble of VLAN tagging ARP frames because
2459  * switches in VLAN mode (especially if ports are configured as
2460  * "native" to a VLAN) might not pass non-tagged frames.
2461  */
2462 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2463 {
2464         struct sk_buff *skb;
2465
2466         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2467                slave_dev->name, dest_ip, src_ip, vlan_id);
2468                
2469         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2470                          NULL, slave_dev->dev_addr, NULL);
2471
2472         if (!skb) {
2473                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2474                 return;
2475         }
2476         if (vlan_id) {
2477                 skb = vlan_put_tag(skb, vlan_id);
2478                 if (!skb) {
2479                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2480                         return;
2481                 }
2482         }
2483         arp_xmit(skb);
2484 }
2485
2486
2487 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2488 {
2489         int i, vlan_id, rv;
2490         __be32 *targets = bond->params.arp_targets;
2491         struct vlan_entry *vlan, *vlan_next;
2492         struct net_device *vlan_dev;
2493         struct flowi fl;
2494         struct rtable *rt;
2495
2496         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2497                 if (!targets[i])
2498                         continue;
2499                 dprintk("basa: target %x\n", targets[i]);
2500                 if (list_empty(&bond->vlan_list)) {
2501                         dprintk("basa: empty vlan: arp_send\n");
2502                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2503                                       bond->master_ip, 0);
2504                         continue;
2505                 }
2506
2507                 /*
2508                  * If VLANs are configured, we do a route lookup to
2509                  * determine which VLAN interface would be used, so we
2510                  * can tag the ARP with the proper VLAN tag.
2511                  */
2512                 memset(&fl, 0, sizeof(fl));
2513                 fl.fl4_dst = targets[i];
2514                 fl.fl4_tos = RTO_ONLINK;
2515
2516                 rv = ip_route_output_key(&rt, &fl);
2517                 if (rv) {
2518                         if (net_ratelimit()) {
2519                                 printk(KERN_WARNING DRV_NAME
2520                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2521                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2522                         }
2523                         continue;
2524                 }
2525
2526                 /*
2527                  * This target is not on a VLAN
2528                  */
2529                 if (rt->u.dst.dev == bond->dev) {
2530                         ip_rt_put(rt);
2531                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2532                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2533                                       bond->master_ip, 0);
2534                         continue;
2535                 }
2536
2537                 vlan_id = 0;
2538                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2539                                          vlan_list) {
2540                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2541                         if (vlan_dev == rt->u.dst.dev) {
2542                                 vlan_id = vlan->vlan_id;
2543                                 dprintk("basa: vlan match on %s %d\n",
2544                                        vlan_dev->name, vlan_id);
2545                                 break;
2546                         }
2547                 }
2548
2549                 if (vlan_id) {
2550                         ip_rt_put(rt);
2551                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2552                                       vlan->vlan_ip, vlan_id);
2553                         continue;
2554                 }
2555
2556                 if (net_ratelimit()) {
2557                         printk(KERN_WARNING DRV_NAME
2558                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2559                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2560                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2561                 }
2562                 ip_rt_put(rt);
2563         }
2564 }
2565
2566 /*
2567  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2568  * for each VLAN above us.
2569  */
2570 static void bond_send_gratuitous_arp(struct bonding *bond)
2571 {
2572         struct slave *slave = bond->curr_active_slave;
2573         struct vlan_entry *vlan;
2574         struct net_device *vlan_dev;
2575
2576         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2577                                 slave ? slave->dev->name : "NULL");
2578         if (!slave)
2579                 return;
2580
2581         if (bond->master_ip) {
2582                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2583                                 bond->master_ip, 0);
2584         }
2585
2586         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2587                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2588                 if (vlan->vlan_ip) {
2589                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2590                                       vlan->vlan_ip, vlan->vlan_id);
2591                 }
2592         }
2593 }
2594
2595 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2596 {
2597         int i;
2598         __be32 *targets = bond->params.arp_targets;
2599
2600         targets = bond->params.arp_targets;
2601         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2602                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2603                         "%u.%u.%u.%u bhti(tip) %d\n",
2604                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2605                        bond_has_this_ip(bond, tip));
2606                 if (sip == targets[i]) {
2607                         if (bond_has_this_ip(bond, tip))
2608                                 slave->last_arp_rx = jiffies;
2609                         return;
2610                 }
2611         }
2612 }
2613
2614 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2615 {
2616         struct arphdr *arp;
2617         struct slave *slave;
2618         struct bonding *bond;
2619         unsigned char *arp_ptr;
2620         __be32 sip, tip;
2621
2622         if (dev->nd_net != &init_net)
2623                 goto out;
2624
2625         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2626                 goto out;
2627
2628         bond = dev->priv;
2629         read_lock(&bond->lock);
2630
2631         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2632                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2633                 orig_dev ? orig_dev->name : "NULL");
2634
2635         slave = bond_get_slave_by_dev(bond, orig_dev);
2636         if (!slave || !slave_do_arp_validate(bond, slave))
2637                 goto out_unlock;
2638
2639         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
2640         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2641                                  (2 * dev->addr_len) +
2642                                  (2 * sizeof(u32)))))
2643                 goto out_unlock;
2644
2645         arp = arp_hdr(skb);
2646         if (arp->ar_hln != dev->addr_len ||
2647             skb->pkt_type == PACKET_OTHERHOST ||
2648             skb->pkt_type == PACKET_LOOPBACK ||
2649             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2650             arp->ar_pro != htons(ETH_P_IP) ||
2651             arp->ar_pln != 4)
2652                 goto out_unlock;
2653
2654         arp_ptr = (unsigned char *)(arp + 1);
2655         arp_ptr += dev->addr_len;
2656         memcpy(&sip, arp_ptr, 4);
2657         arp_ptr += 4 + dev->addr_len;
2658         memcpy(&tip, arp_ptr, 4);
2659
2660         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2661                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2662                 slave->state, bond->params.arp_validate,
2663                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2664
2665         /*
2666          * Backup slaves won't see the ARP reply, but do come through
2667          * here for each ARP probe (so we swap the sip/tip to validate
2668          * the probe).  In a "redundant switch, common router" type of
2669          * configuration, the ARP probe will (hopefully) travel from
2670          * the active, through one switch, the router, then the other
2671          * switch before reaching the backup.
2672          */
2673         if (slave->state == BOND_STATE_ACTIVE)
2674                 bond_validate_arp(bond, slave, sip, tip);
2675         else
2676                 bond_validate_arp(bond, slave, tip, sip);
2677
2678 out_unlock:
2679         read_unlock(&bond->lock);
2680 out:
2681         dev_kfree_skb(skb);
2682         return NET_RX_SUCCESS;
2683 }
2684
2685 /*
2686  * this function is called regularly to monitor each slave's link
2687  * ensuring that traffic is being sent and received when arp monitoring
2688  * is used in load-balancing mode. if the adapter has been dormant, then an
2689  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2690  * arp monitoring in active backup mode.
2691  */
2692 void bond_loadbalance_arp_mon(struct work_struct *work)
2693 {
2694         struct bonding *bond = container_of(work, struct bonding,
2695                                             arp_work.work);
2696         struct slave *slave, *oldcurrent;
2697         int do_failover = 0;
2698         int delta_in_ticks;
2699         int i;
2700
2701         read_lock(&bond->lock);
2702
2703         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2704
2705         if (bond->kill_timers) {
2706                 goto out;
2707         }
2708
2709         if (bond->slave_cnt == 0) {
2710                 goto re_arm;
2711         }
2712
2713         read_lock(&bond->curr_slave_lock);
2714         oldcurrent = bond->curr_active_slave;
2715         read_unlock(&bond->curr_slave_lock);
2716
2717         /* see if any of the previous devices are up now (i.e. they have
2718          * xmt and rcv traffic). the curr_active_slave does not come into
2719          * the picture unless it is null. also, slave->jiffies is not needed
2720          * here because we send an arp on each slave and give a slave as
2721          * long as it needs to get the tx/rx within the delta.
2722          * TODO: what about up/down delay in arp mode? it wasn't here before
2723          *       so it can wait
2724          */
2725         bond_for_each_slave(bond, slave, i) {
2726                 if (slave->link != BOND_LINK_UP) {
2727                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2728                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2729
2730                                 slave->link  = BOND_LINK_UP;
2731                                 slave->state = BOND_STATE_ACTIVE;
2732
2733                                 /* primary_slave has no meaning in round-robin
2734                                  * mode. the window of a slave being up and
2735                                  * curr_active_slave being null after enslaving
2736                                  * is closed.
2737                                  */
2738                                 if (!oldcurrent) {
2739                                         printk(KERN_INFO DRV_NAME
2740                                                ": %s: link status definitely "
2741                                                "up for interface %s, ",
2742                                                bond->dev->name,
2743                                                slave->dev->name);
2744                                         do_failover = 1;
2745                                 } else {
2746                                         printk(KERN_INFO DRV_NAME
2747                                                ": %s: interface %s is now up\n",
2748                                                bond->dev->name,
2749                                                slave->dev->name);
2750                                 }
2751                         }
2752                 } else {
2753                         /* slave->link == BOND_LINK_UP */
2754
2755                         /* not all switches will respond to an arp request
2756                          * when the source ip is 0, so don't take the link down
2757                          * if we don't know our ip yet
2758                          */
2759                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2760                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks) &&
2761                              bond_has_ip(bond))) {
2762
2763                                 slave->link  = BOND_LINK_DOWN;
2764                                 slave->state = BOND_STATE_BACKUP;
2765
2766                                 if (slave->link_failure_count < UINT_MAX) {
2767                                         slave->link_failure_count++;
2768                                 }
2769
2770                                 printk(KERN_INFO DRV_NAME
2771                                        ": %s: interface %s is now down.\n",
2772                                        bond->dev->name,
2773                                        slave->dev->name);
2774
2775                                 if (slave == oldcurrent) {
2776                                         do_failover = 1;
2777                                 }
2778                         }
2779                 }
2780
2781                 /* note: if switch is in round-robin mode, all links
2782                  * must tx arp to ensure all links rx an arp - otherwise
2783                  * links may oscillate or not come up at all; if switch is
2784                  * in something like xor mode, there is nothing we can
2785                  * do - all replies will be rx'ed on same link causing slaves
2786                  * to be unstable during low/no traffic periods
2787                  */
2788                 if (IS_UP(slave->dev)) {
2789                         bond_arp_send_all(bond, slave);
2790                 }
2791         }
2792
2793         if (do_failover) {
2794                 rtnl_lock();
2795                 write_lock_bh(&bond->curr_slave_lock);
2796
2797                 bond_select_active_slave(bond);
2798
2799                 write_unlock_bh(&bond->curr_slave_lock);
2800                 rtnl_unlock();
2801
2802         }
2803
2804 re_arm:
2805         if (bond->params.arp_interval)
2806                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2807 out:
2808         read_unlock(&bond->lock);
2809 }
2810
2811 /*
2812  * When using arp monitoring in active-backup mode, this function is
2813  * called to determine if any backup slaves have went down or a new
2814  * current slave needs to be found.
2815  * The backup slaves never generate traffic, they are considered up by merely
2816  * receiving traffic. If the current slave goes down, each backup slave will
2817  * be given the opportunity to tx/rx an arp before being taken down - this
2818  * prevents all slaves from being taken down due to the current slave not
2819  * sending any traffic for the backups to receive. The arps are not necessarily
2820  * necessary, any tx and rx traffic will keep the current slave up. While any
2821  * rx traffic will keep the backup slaves up, the current slave is responsible
2822  * for generating traffic to keep them up regardless of any other traffic they
2823  * may have received.
2824  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2825  */
2826 void bond_activebackup_arp_mon(struct work_struct *work)
2827 {
2828         struct bonding *bond = container_of(work, struct bonding,
2829                                             arp_work.work);
2830         struct slave *slave;
2831         int delta_in_ticks;
2832         int i;
2833
2834         read_lock(&bond->lock);
2835
2836         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2837
2838         if (bond->kill_timers) {
2839                 goto out;
2840         }
2841
2842         if (bond->slave_cnt == 0) {
2843                 goto re_arm;
2844         }
2845
2846         /* determine if any slave has come up or any backup slave has
2847          * gone down
2848          * TODO: what about up/down delay in arp mode? it wasn't here before
2849          *       so it can wait
2850          */
2851         bond_for_each_slave(bond, slave, i) {
2852                 if (slave->link != BOND_LINK_UP) {
2853                         if (time_before_eq(jiffies,
2854                             slave_last_rx(bond, slave) + delta_in_ticks)) {
2855
2856                                 slave->link = BOND_LINK_UP;
2857
2858                                 rtnl_lock();
2859
2860                                 write_lock_bh(&bond->curr_slave_lock);
2861
2862                                 if ((!bond->curr_active_slave) &&
2863                                     time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks)) {
2864                                         bond_change_active_slave(bond, slave);
2865                                         bond->current_arp_slave = NULL;
2866                                 } else if (bond->curr_active_slave != slave) {
2867                                         /* this slave has just come up but we
2868                                          * already have a current slave; this
2869                                          * can also happen if bond_enslave adds
2870                                          * a new slave that is up while we are
2871                                          * searching for a new slave
2872                                          */
2873                                         bond_set_slave_inactive_flags(slave);
2874                                         bond->current_arp_slave = NULL;
2875                                 }
2876
2877                                 bond_set_carrier(bond);
2878
2879                                 if (slave == bond->curr_active_slave) {
2880                                         printk(KERN_INFO DRV_NAME
2881                                                ": %s: %s is up and now the "
2882                                                "active interface\n",
2883                                                bond->dev->name,
2884                                                slave->dev->name);
2885                                         netif_carrier_on(bond->dev);
2886                                 } else {
2887                                         printk(KERN_INFO DRV_NAME
2888                                                ": %s: backup interface %s is "
2889                                                "now up\n",
2890                                                bond->dev->name,
2891                                                slave->dev->name);
2892                                 }
2893
2894                                 write_unlock_bh(&bond->curr_slave_lock);
2895                                 rtnl_unlock();
2896                         }
2897                 } else {
2898                         read_lock(&bond->curr_slave_lock);
2899
2900                         if ((slave != bond->curr_active_slave) &&
2901                             (!bond->current_arp_slave) &&
2902                             (time_after_eq(jiffies, slave_last_rx(bond, slave) + 3*delta_in_ticks) &&
2903                              bond_has_ip(bond))) {
2904                                 /* a backup slave has gone down; three times
2905                                  * the delta allows the current slave to be
2906                                  * taken out before the backup slave.
2907                                  * note: a non-null current_arp_slave indicates
2908                                  * the curr_active_slave went down and we are
2909                                  * searching for a new one; under this
2910                                  * condition we only take the curr_active_slave
2911                                  * down - this gives each slave a chance to
2912                                  * tx/rx traffic before being taken out
2913                                  */
2914
2915                                 read_unlock(&bond->curr_slave_lock);
2916
2917                                 slave->link  = BOND_LINK_DOWN;
2918
2919                                 if (slave->link_failure_count < UINT_MAX) {
2920                                         slave->link_failure_count++;
2921                                 }
2922
2923                                 bond_set_slave_inactive_flags(slave);
2924
2925                                 printk(KERN_INFO DRV_NAME
2926                                        ": %s: backup interface %s is now down\n",
2927                                        bond->dev->name,
2928                                        slave->dev->name);
2929                         } else {
2930                                 read_unlock(&bond->curr_slave_lock);
2931                         }
2932                 }
2933         }
2934
2935         read_lock(&bond->curr_slave_lock);
2936         slave = bond->curr_active_slave;
2937         read_unlock(&bond->curr_slave_lock);
2938
2939         if (slave) {
2940                 /* if we have sent traffic in the past 2*arp_intervals but
2941                  * haven't xmit and rx traffic in that time interval, select
2942                  * a different slave. slave->jiffies is only updated when
2943                  * a slave first becomes the curr_active_slave - not necessarily
2944                  * after every arp; this ensures the slave has a full 2*delta
2945                  * before being taken out. if a primary is being used, check
2946                  * if it is up and needs to take over as the curr_active_slave
2947                  */
2948                 if ((time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2949                         (time_after_eq(jiffies, slave_last_rx(bond, slave) + 2*delta_in_ticks) &&
2950                          bond_has_ip(bond))) &&
2951                         time_after_eq(jiffies, slave->jiffies + 2*delta_in_ticks)) {
2952
2953                         slave->link  = BOND_LINK_DOWN;
2954
2955                         if (slave->link_failure_count < UINT_MAX) {
2956                                 slave->link_failure_count++;
2957                         }
2958
2959                         printk(KERN_INFO DRV_NAME
2960                                ": %s: link status down for active interface "
2961                                "%s, disabling it\n",
2962                                bond->dev->name,
2963                                slave->dev->name);
2964
2965                         rtnl_lock();
2966                         write_lock_bh(&bond->curr_slave_lock);
2967
2968                         bond_select_active_slave(bond);
2969                         slave = bond->curr_active_slave;
2970
2971                         write_unlock_bh(&bond->curr_slave_lock);
2972
2973                         rtnl_unlock();
2974
2975                         bond->current_arp_slave = slave;
2976
2977                         if (slave) {
2978                                 slave->jiffies = jiffies;
2979                         }
2980                 } else if ((bond->primary_slave) &&
2981                            (bond->primary_slave != slave) &&
2982                            (bond->primary_slave->link == BOND_LINK_UP)) {
2983                         /* at this point, slave is the curr_active_slave */
2984                         printk(KERN_INFO DRV_NAME
2985                                ": %s: changing from interface %s to primary "
2986                                "interface %s\n",
2987                                bond->dev->name,
2988                                slave->dev->name,
2989                                bond->primary_slave->dev->name);
2990
2991                         /* primary is up so switch to it */
2992                         rtnl_lock();
2993                         write_lock_bh(&bond->curr_slave_lock);
2994                         bond_change_active_slave(bond, bond->primary_slave);
2995                         write_unlock_bh(&bond->curr_slave_lock);
2996
2997                         rtnl_unlock();
2998
2999                         slave = bond->primary_slave;
3000                         slave->jiffies = jiffies;
3001                 } else {
3002                         bond->current_arp_slave = NULL;
3003                 }
3004
3005                 /* the current slave must tx an arp to ensure backup slaves
3006                  * rx traffic
3007                  */
3008                 if (slave && bond_has_ip(bond)) {
3009                         bond_arp_send_all(bond, slave);
3010                 }
3011         }
3012
3013         /* if we don't have a curr_active_slave, search for the next available
3014          * backup slave from the current_arp_slave and make it the candidate
3015          * for becoming the curr_active_slave
3016          */
3017         if (!slave) {
3018                 if (!bond->current_arp_slave) {
3019                         bond->current_arp_slave = bond->first_slave;
3020                 }
3021
3022                 if (bond->current_arp_slave) {
3023                         bond_set_slave_inactive_flags(bond->current_arp_slave);
3024
3025                         /* search for next candidate */
3026                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3027                                 if (IS_UP(slave->dev)) {
3028                                         slave->link = BOND_LINK_BACK;
3029                                         bond_set_slave_active_flags(slave);
3030                                         bond_arp_send_all(bond, slave);
3031                                         slave->jiffies = jiffies;
3032                                         bond->current_arp_slave = slave;
3033                                         break;
3034                                 }
3035
3036                                 /* if the link state is up at this point, we
3037                                  * mark it down - this can happen if we have
3038                                  * simultaneous link failures and
3039                                  * reselect_active_interface doesn't make this
3040                                  * one the current slave so it is still marked
3041                                  * up when it is actually down
3042                                  */
3043                                 if (slave->link == BOND_LINK_UP) {
3044                                         slave->link  = BOND_LINK_DOWN;
3045                                         if (slave->link_failure_count < UINT_MAX) {
3046                                                 slave->link_failure_count++;
3047                                         }
3048
3049                                         bond_set_slave_inactive_flags(slave);
3050
3051                                         printk(KERN_INFO DRV_NAME
3052                                                ": %s: backup interface %s is "
3053                                                "now down.\n",
3054                                                bond->dev->name,
3055                                                slave->dev->name);
3056                                 }
3057                         }
3058                 }
3059         }
3060
3061 re_arm:
3062         if (bond->params.arp_interval) {
3063                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3064         }
3065 out:
3066         read_unlock(&bond->lock);
3067 }
3068
3069 /*------------------------------ proc/seq_file-------------------------------*/
3070
3071 #ifdef CONFIG_PROC_FS
3072
3073 #define SEQ_START_TOKEN ((void *)1)
3074
3075 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3076 {
3077         struct bonding *bond = seq->private;
3078         loff_t off = 0;
3079         struct slave *slave;
3080         int i;
3081
3082         /* make sure the bond won't be taken away */
3083         read_lock(&dev_base_lock);
3084         read_lock(&bond->lock);
3085
3086         if (*pos == 0) {
3087                 return SEQ_START_TOKEN;
3088         }
3089
3090         bond_for_each_slave(bond, slave, i) {
3091                 if (++off == *pos) {
3092                         return slave;
3093                 }
3094         }
3095
3096         return NULL;
3097 }
3098
3099 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3100 {
3101         struct bonding *bond = seq->private;
3102         struct slave *slave = v;
3103
3104         ++*pos;
3105         if (v == SEQ_START_TOKEN) {
3106                 return bond->first_slave;
3107         }
3108
3109         slave = slave->next;
3110
3111         return (slave == bond->first_slave) ? NULL : slave;
3112 }
3113
3114 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3115 {
3116         struct bonding *bond = seq->private;
3117
3118         read_unlock(&bond->lock);
3119         read_unlock(&dev_base_lock);
3120 }
3121
3122 static void bond_info_show_master(struct seq_file *seq)
3123 {
3124         struct bonding *bond = seq->private;
3125         struct slave *curr;
3126         int i;
3127         u32 target;
3128
3129         read_lock(&bond->curr_slave_lock);
3130         curr = bond->curr_active_slave;
3131         read_unlock(&bond->curr_slave_lock);
3132
3133         seq_printf(seq, "Bonding Mode: %s",
3134                    bond_mode_name(bond->params.mode));
3135
3136         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3137             bond->params.fail_over_mac)
3138                 seq_printf(seq, " (fail_over_mac)");
3139
3140         seq_printf(seq, "\n");
3141
3142         if (bond->params.mode == BOND_MODE_XOR ||
3143                 bond->params.mode == BOND_MODE_8023AD) {
3144                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3145                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3146                         bond->params.xmit_policy);
3147         }
3148
3149         if (USES_PRIMARY(bond->params.mode)) {
3150                 seq_printf(seq, "Primary Slave: %s\n",
3151                            (bond->primary_slave) ?
3152                            bond->primary_slave->dev->name : "None");
3153
3154                 seq_printf(seq, "Currently Active Slave: %s\n",
3155                            (curr) ? curr->dev->name : "None");
3156         }
3157
3158         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3159                    "up" : "down");
3160         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3161         seq_printf(seq, "Up Delay (ms): %d\n",
3162                    bond->params.updelay * bond->params.miimon);
3163         seq_printf(seq, "Down Delay (ms): %d\n",
3164                    bond->params.downdelay * bond->params.miimon);
3165
3166
3167         /* ARP information */
3168         if(bond->params.arp_interval > 0) {
3169                 int printed=0;
3170                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3171                                 bond->params.arp_interval);
3172
3173                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3174
3175                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3176                         if (!bond->params.arp_targets[i])
3177                                 continue;
3178                         if (printed)
3179                                 seq_printf(seq, ",");
3180                         target = ntohl(bond->params.arp_targets[i]);
3181                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3182                         printed = 1;
3183                 }
3184                 seq_printf(seq, "\n");
3185         }
3186
3187         if (bond->params.mode == BOND_MODE_8023AD) {
3188                 struct ad_info ad_info;
3189                 DECLARE_MAC_BUF(mac);
3190
3191                 seq_puts(seq, "\n802.3ad info\n");
3192                 seq_printf(seq, "LACP rate: %s\n",
3193                            (bond->params.lacp_fast) ? "fast" : "slow");
3194
3195                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3196                         seq_printf(seq, "bond %s has no active aggregator\n",
3197                                    bond->dev->name);
3198                 } else {
3199                         seq_printf(seq, "Active Aggregator Info:\n");
3200
3201                         seq_printf(seq, "\tAggregator ID: %d\n",
3202                                    ad_info.aggregator_id);
3203                         seq_printf(seq, "\tNumber of ports: %d\n",
3204                                    ad_info.ports);
3205                         seq_printf(seq, "\tActor Key: %d\n",
3206                                    ad_info.actor_key);
3207                         seq_printf(seq, "\tPartner Key: %d\n",
3208                                    ad_info.partner_key);
3209                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3210                                    print_mac(mac, ad_info.partner_system));
3211                 }
3212         }
3213 }
3214
3215 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3216 {
3217         struct bonding *bond = seq->private;
3218         DECLARE_MAC_BUF(mac);
3219
3220         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3221         seq_printf(seq, "MII Status: %s\n",
3222                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3223         seq_printf(seq, "Link Failure Count: %u\n",
3224                    slave->link_failure_count);
3225
3226         seq_printf(seq,
3227                    "Permanent HW addr: %s\n",
3228                    print_mac(mac, slave->perm_hwaddr));
3229
3230         if (bond->params.mode == BOND_MODE_8023AD) {
3231                 const struct aggregator *agg
3232                         = SLAVE_AD_INFO(slave).port.aggregator;
3233
3234                 if (agg) {
3235                         seq_printf(seq, "Aggregator ID: %d\n",
3236                                    agg->aggregator_identifier);
3237                 } else {
3238                         seq_puts(seq, "Aggregator ID: N/A\n");
3239                 }
3240         }
3241 }
3242
3243 static int bond_info_seq_show(struct seq_file *seq, void *v)
3244 {
3245         if (v == SEQ_START_TOKEN) {
3246                 seq_printf(seq, "%s\n", version);
3247                 bond_info_show_master(seq);
3248         } else {
3249                 bond_info_show_slave(seq, v);
3250         }
3251
3252         return 0;
3253 }
3254
3255 static struct seq_operations bond_info_seq_ops = {
3256         .start = bond_info_seq_start,
3257         .next  = bond_info_seq_next,
3258         .stop  = bond_info_seq_stop,
3259         .show  = bond_info_seq_show,
3260 };
3261
3262 static int bond_info_open(struct inode *inode, struct file *file)
3263 {
3264         struct seq_file *seq;
3265         struct proc_dir_entry *proc;
3266         int res;
3267
3268         res = seq_open(file, &bond_info_seq_ops);
3269         if (!res) {
3270                 /* recover the pointer buried in proc_dir_entry data */
3271                 seq = file->private_data;
3272                 proc = PDE(inode);
3273                 seq->private = proc->data;
3274         }
3275
3276         return res;
3277 }
3278
3279 static const struct file_operations bond_info_fops = {
3280         .owner   = THIS_MODULE,
3281         .open    = bond_info_open,
3282         .read    = seq_read,
3283         .llseek  = seq_lseek,
3284         .release = seq_release,
3285 };
3286
3287 static int bond_create_proc_entry(struct bonding *bond)
3288 {
3289         struct net_device *bond_dev = bond->dev;
3290
3291         if (bond_proc_dir) {
3292                 bond->proc_entry = create_proc_entry(bond_dev->name,
3293                                                      S_IRUGO,
3294                                                      bond_proc_dir);
3295                 if (bond->proc_entry == NULL) {
3296                         printk(KERN_WARNING DRV_NAME
3297                                ": Warning: Cannot create /proc/net/%s/%s\n",
3298                                DRV_NAME, bond_dev->name);
3299                 } else {
3300                         bond->proc_entry->data = bond;
3301                         bond->proc_entry->proc_fops = &bond_info_fops;
3302                         bond->proc_entry->owner = THIS_MODULE;
3303                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3304                 }
3305         }
3306
3307         return 0;
3308 }
3309
3310 static void bond_remove_proc_entry(struct bonding *bond)
3311 {
3312         if (bond_proc_dir && bond->proc_entry) {
3313                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3314                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3315                 bond->proc_entry = NULL;
3316         }
3317 }
3318
3319 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3320  * Caller must hold rtnl_lock.
3321  */
3322 static void bond_create_proc_dir(void)
3323 {
3324         int len = strlen(DRV_NAME);
3325
3326         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3327              bond_proc_dir = bond_proc_dir->next) {
3328                 if ((bond_proc_dir->namelen == len) &&
3329                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3330                         break;
3331                 }
3332         }
3333
3334         if (!bond_proc_dir) {
3335                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3336                 if (bond_proc_dir) {
3337                         bond_proc_dir->owner = THIS_MODULE;
3338                 } else {
3339                         printk(KERN_WARNING DRV_NAME
3340                                 ": Warning: cannot create /proc/net/%s\n",
3341                                 DRV_NAME);
3342                 }
3343         }
3344 }
3345
3346 /* Destroy the bonding directory under /proc/net, if empty.
3347  * Caller must hold rtnl_lock.
3348  */
3349 static void bond_destroy_proc_dir(void)
3350 {
3351         struct proc_dir_entry *de;
3352
3353         if (!bond_proc_dir) {
3354                 return;
3355         }
3356
3357         /* verify that the /proc dir is empty */
3358         for (de = bond_proc_dir->subdir; de; de = de->next) {
3359                 /* ignore . and .. */
3360                 if (*(de->name) != '.') {
3361                         break;
3362                 }
3363         }
3364
3365         if (de) {
3366                 if (bond_proc_dir->owner == THIS_MODULE) {
3367                         bond_proc_dir->owner = NULL;
3368                 }
3369         } else {
3370                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3371                 bond_proc_dir = NULL;
3372         }
3373 }
3374 #endif /* CONFIG_PROC_FS */
3375
3376 /*-------------------------- netdev event handling --------------------------*/
3377
3378 /*
3379  * Change device name
3380  */
3381 static int bond_event_changename(struct bonding *bond)
3382 {
3383 #ifdef CONFIG_PROC_FS
3384         bond_remove_proc_entry(bond);
3385         bond_create_proc_entry(bond);
3386 #endif
3387         down_write(&(bonding_rwsem));
3388         bond_destroy_sysfs_entry(bond);
3389         bond_create_sysfs_entry(bond);
3390         up_write(&(bonding_rwsem));
3391         return NOTIFY_DONE;
3392 }
3393
3394 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3395 {
3396         struct bonding *event_bond = bond_dev->priv;
3397
3398         switch (event) {
3399         case NETDEV_CHANGENAME:
3400                 return bond_event_changename(event_bond);
3401         case NETDEV_UNREGISTER:
3402                 /*
3403                  * TODO: remove a bond from the list?
3404                  */
3405                 break;
3406         default:
3407                 break;
3408         }
3409
3410         return NOTIFY_DONE;
3411 }
3412
3413 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3414 {
3415         struct net_device *bond_dev = slave_dev->master;
3416         struct bonding *bond = bond_dev->priv;
3417
3418         switch (event) {
3419         case NETDEV_UNREGISTER:
3420                 if (bond_dev) {
3421                         if (bond->setup_by_slave)
3422                                 bond_release_and_destroy(bond_dev, slave_dev);
3423                         else
3424                                 bond_release(bond_dev, slave_dev);
3425                 }
3426                 break;
3427         case NETDEV_CHANGE:
3428                 /*
3429                  * TODO: is this what we get if somebody
3430                  * sets up a hierarchical bond, then rmmod's
3431                  * one of the slave bonding devices?
3432                  */
3433                 break;
3434         case NETDEV_DOWN:
3435                 /*
3436                  * ... Or is it this?
3437                  */
3438                 break;
3439         case NETDEV_CHANGEMTU:
3440                 /*
3441                  * TODO: Should slaves be allowed to
3442                  * independently alter their MTU?  For
3443                  * an active-backup bond, slaves need
3444                  * not be the same type of device, so
3445                  * MTUs may vary.  For other modes,
3446                  * slaves arguably should have the
3447                  * same MTUs. To do this, we'd need to
3448                  * take over the slave's change_mtu
3449                  * function for the duration of their
3450                  * servitude.
3451                  */
3452                 break;
3453         case NETDEV_CHANGENAME:
3454                 /*
3455                  * TODO: handle changing the primary's name
3456                  */
3457                 break;
3458         case NETDEV_FEAT_CHANGE:
3459                 bond_compute_features(bond);
3460                 break;
3461         default:
3462                 break;
3463         }
3464
3465         return NOTIFY_DONE;
3466 }
3467
3468 /*
3469  * bond_netdev_event: handle netdev notifier chain events.
3470  *
3471  * This function receives events for the netdev chain.  The caller (an
3472  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3473  * locks for us to safely manipulate the slave devices (RTNL lock,
3474  * dev_probe_lock).
3475  */
3476 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3477 {
3478         struct net_device *event_dev = (struct net_device *)ptr;
3479
3480         if (event_dev->nd_net != &init_net)
3481                 return NOTIFY_DONE;
3482
3483         dprintk("event_dev: %s, event: %lx\n",
3484                 (event_dev ? event_dev->name : "None"),
3485                 event);
3486
3487         if (!(event_dev->priv_flags & IFF_BONDING))
3488                 return NOTIFY_DONE;
3489
3490         if (event_dev->flags & IFF_MASTER) {
3491                 dprintk("IFF_MASTER\n");
3492                 return bond_master_netdev_event(event, event_dev);
3493         }
3494
3495         if (event_dev->flags & IFF_SLAVE) {
3496                 dprintk("IFF_SLAVE\n");
3497                 return bond_slave_netdev_event(event, event_dev);
3498         }
3499
3500         return NOTIFY_DONE;
3501 }
3502
3503 /*
3504  * bond_inetaddr_event: handle inetaddr notifier chain events.
3505  *
3506  * We keep track of device IPs primarily to use as source addresses in
3507  * ARP monitor probes (rather than spewing out broadcasts all the time).
3508  *
3509  * We track one IP for the main device (if it has one), plus one per VLAN.
3510  */
3511 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3512 {
3513         struct in_ifaddr *ifa = ptr;
3514         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3515         struct bonding *bond, *bond_next;
3516         struct vlan_entry *vlan, *vlan_next;
3517
3518         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3519                 if (bond->dev == event_dev) {
3520                         switch (event) {
3521                         case NETDEV_UP:
3522                                 bond->master_ip = ifa->ifa_local;
3523                                 return NOTIFY_OK;
3524                         case NETDEV_DOWN:
3525                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3526                                 return NOTIFY_OK;
3527                         default:
3528                                 return NOTIFY_DONE;
3529                         }
3530                 }
3531
3532                 if (list_empty(&bond->vlan_list))
3533                         continue;
3534
3535                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3536                                          vlan_list) {
3537                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3538                         if (vlan_dev == event_dev) {
3539                                 switch (event) {
3540                                 case NETDEV_UP:
3541                                         vlan->vlan_ip = ifa->ifa_local;
3542                                         return NOTIFY_OK;
3543                                 case NETDEV_DOWN:
3544                                         vlan->vlan_ip =
3545                                                 bond_glean_dev_ip(vlan_dev);
3546                                         return NOTIFY_OK;
3547                                 default:
3548                                         return NOTIFY_DONE;
3549                                 }
3550                         }
3551                 }
3552         }
3553         return NOTIFY_DONE;
3554 }
3555
3556 static struct notifier_block bond_netdev_notifier = {
3557         .notifier_call = bond_netdev_event,
3558 };
3559
3560 static struct notifier_block bond_inetaddr_notifier = {
3561         .notifier_call = bond_inetaddr_event,
3562 };
3563
3564 /*-------------------------- Packet type handling ---------------------------*/
3565
3566 /* register to receive lacpdus on a bond */
3567 static void bond_register_lacpdu(struct bonding *bond)
3568 {
3569         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3570
3571         /* initialize packet type */
3572         pk_type->type = PKT_TYPE_LACPDU;
3573         pk_type->dev = bond->dev;
3574         pk_type->func = bond_3ad_lacpdu_recv;
3575
3576         dev_add_pack(pk_type);
3577 }
3578
3579 /* unregister to receive lacpdus on a bond */
3580 static void bond_unregister_lacpdu(struct bonding *bond)
3581 {
3582         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3583 }
3584
3585 void bond_register_arp(struct bonding *bond)
3586 {
3587         struct packet_type *pt = &bond->arp_mon_pt;
3588
3589         if (pt->type)
3590                 return;
3591
3592         pt->type = htons(ETH_P_ARP);
3593         pt->dev = bond->dev;
3594         pt->func = bond_arp_rcv;
3595         dev_add_pack(pt);
3596 }
3597
3598 void bond_unregister_arp(struct bonding *bond)
3599 {
3600         struct packet_type *pt = &bond->arp_mon_pt;
3601
3602         dev_remove_pack(pt);
3603         pt->type = 0;
3604 }
3605
3606 /*---------------------------- Hashing Policies -----------------------------*/
3607
3608 /*
3609  * Hash for the output device based upon layer 2 and layer 3 data. If
3610  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3611  */
3612 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3613                                      struct net_device *bond_dev, int count)
3614 {
3615         struct ethhdr *data = (struct ethhdr *)skb->data;
3616         struct iphdr *iph = ip_hdr(skb);
3617
3618         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3619                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3620                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3621         }
3622
3623         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3624 }
3625
3626 /*
3627  * Hash for the output device based upon layer 3 and layer 4 data. If
3628  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3629  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3630  */
3631 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3632                                     struct net_device *bond_dev, int count)
3633 {
3634         struct ethhdr *data = (struct ethhdr *)skb->data;
3635         struct iphdr *iph = ip_hdr(skb);
3636         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3637         int layer4_xor = 0;
3638
3639         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3640                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3641                     (iph->protocol == IPPROTO_TCP ||
3642                      iph->protocol == IPPROTO_UDP)) {
3643                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3644                 }
3645                 return (layer4_xor ^
3646                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3647
3648         }
3649
3650         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3651 }
3652
3653 /*
3654  * Hash for the output device based upon layer 2 data
3655  */
3656 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3657                                    struct net_device *bond_dev, int count)
3658 {
3659         struct ethhdr *data = (struct ethhdr *)skb->data;
3660
3661         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3662 }
3663
3664 /*-------------------------- Device entry points ----------------------------*/
3665
3666 static int bond_open(struct net_device *bond_dev)
3667 {
3668         struct bonding *bond = bond_dev->priv;
3669
3670         bond->kill_timers = 0;
3671
3672         if ((bond->params.mode == BOND_MODE_TLB) ||
3673             (bond->params.mode == BOND_MODE_ALB)) {
3674                 /* bond_alb_initialize must be called before the timer
3675                  * is started.
3676                  */
3677                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3678                         /* something went wrong - fail the open operation */
3679                         return -1;
3680                 }
3681
3682                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3683                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3684         }
3685
3686         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3687                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3688                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3689         }
3690
3691         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3692                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3693                         INIT_DELAYED_WORK(&bond->arp_work,
3694                                           bond_activebackup_arp_mon);
3695                 else
3696                         INIT_DELAYED_WORK(&bond->arp_work,
3697                                           bond_loadbalance_arp_mon);
3698
3699                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3700                 if (bond->params.arp_validate)
3701                         bond_register_arp(bond);
3702         }
3703
3704         if (bond->params.mode == BOND_MODE_8023AD) {
3705                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3706                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3707                 /* register to receive LACPDUs */
3708                 bond_register_lacpdu(bond);
3709         }
3710
3711         return 0;
3712 }
3713
3714 static int bond_close(struct net_device *bond_dev)
3715 {
3716         struct bonding *bond = bond_dev->priv;
3717
3718         if (bond->params.mode == BOND_MODE_8023AD) {
3719                 /* Unregister the receive of LACPDUs */
3720                 bond_unregister_lacpdu(bond);
3721         }
3722
3723         if (bond->params.arp_validate)
3724                 bond_unregister_arp(bond);
3725
3726         write_lock_bh(&bond->lock);
3727
3728
3729         /* signal timers not to re-arm */
3730         bond->kill_timers = 1;
3731
3732         write_unlock_bh(&bond->lock);
3733
3734         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3735                 cancel_delayed_work(&bond->mii_work);
3736         }
3737
3738         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3739                 cancel_delayed_work(&bond->arp_work);
3740         }
3741
3742         switch (bond->params.mode) {
3743         case BOND_MODE_8023AD:
3744                 cancel_delayed_work(&bond->ad_work);
3745                 break;
3746         case BOND_MODE_TLB:
3747         case BOND_MODE_ALB:
3748                 cancel_delayed_work(&bond->alb_work);
3749                 break;
3750         default:
3751                 break;
3752         }
3753
3754
3755         if ((bond->params.mode == BOND_MODE_TLB) ||
3756             (bond->params.mode == BOND_MODE_ALB)) {
3757                 /* Must be called only after all
3758                  * slaves have been released
3759                  */
3760                 bond_alb_deinitialize(bond);
3761         }
3762
3763         return 0;
3764 }
3765
3766 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3767 {
3768         struct bonding *bond = bond_dev->priv;
3769         struct net_device_stats *stats = &(bond->stats), *sstats;
3770         struct slave *slave;
3771         int i;
3772
3773         memset(stats, 0, sizeof(struct net_device_stats));
3774
3775         read_lock_bh(&bond->lock);
3776
3777         bond_for_each_slave(bond, slave, i) {
3778                 sstats = slave->dev->get_stats(slave->dev);
3779                 stats->rx_packets += sstats->rx_packets;
3780                 stats->rx_bytes += sstats->rx_bytes;
3781                 stats->rx_errors += sstats->rx_errors;
3782                 stats->rx_dropped += sstats->rx_dropped;
3783
3784                 stats->tx_packets += sstats->tx_packets;
3785                 stats->tx_bytes += sstats->tx_bytes;
3786                 stats->tx_errors += sstats->tx_errors;
3787                 stats->tx_dropped += sstats->tx_dropped;
3788
3789                 stats->multicast += sstats->multicast;
3790                 stats->collisions += sstats->collisions;
3791
3792                 stats->rx_length_errors += sstats->rx_length_errors;
3793                 stats->rx_over_errors += sstats->rx_over_errors;
3794                 stats->rx_crc_errors += sstats->rx_crc_errors;
3795                 stats->rx_frame_errors += sstats->rx_frame_errors;
3796                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3797                 stats->rx_missed_errors += sstats->rx_missed_errors;
3798
3799                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3800                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3801                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3802                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3803                 stats->tx_window_errors += sstats->tx_window_errors;
3804         }
3805
3806         read_unlock_bh(&bond->lock);
3807
3808         return stats;
3809 }
3810
3811 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3812 {
3813         struct net_device *slave_dev = NULL;
3814         struct ifbond k_binfo;
3815         struct ifbond __user *u_binfo = NULL;
3816         struct ifslave k_sinfo;
3817         struct ifslave __user *u_sinfo = NULL;
3818         struct mii_ioctl_data *mii = NULL;
3819         int res = 0;
3820
3821         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3822                 bond_dev->name, cmd);
3823
3824         switch (cmd) {
3825         case SIOCGMIIPHY:
3826                 mii = if_mii(ifr);
3827                 if (!mii) {
3828                         return -EINVAL;
3829                 }
3830                 mii->phy_id = 0;
3831                 /* Fall Through */
3832         case SIOCGMIIREG:
3833                 /*
3834                  * We do this again just in case we were called by SIOCGMIIREG
3835                  * instead of SIOCGMIIPHY.
3836                  */
3837                 mii = if_mii(ifr);
3838                 if (!mii) {
3839                         return -EINVAL;
3840                 }
3841
3842                 if (mii->reg_num == 1) {
3843                         struct bonding *bond = bond_dev->priv;
3844                         mii->val_out = 0;
3845                         read_lock(&bond->lock);
3846                         read_lock(&bond->curr_slave_lock);
3847                         if (netif_carrier_ok(bond->dev)) {
3848                                 mii->val_out = BMSR_LSTATUS;
3849                         }
3850                         read_unlock(&bond->curr_slave_lock);
3851                         read_unlock(&bond->lock);
3852                 }
3853
3854                 return 0;
3855         case BOND_INFO_QUERY_OLD:
3856         case SIOCBONDINFOQUERY:
3857                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3858
3859                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3860                         return -EFAULT;
3861                 }
3862
3863                 res = bond_info_query(bond_dev, &k_binfo);
3864                 if (res == 0) {
3865                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3866                                 return -EFAULT;
3867                         }
3868                 }
3869
3870                 return res;
3871         case BOND_SLAVE_INFO_QUERY_OLD:
3872         case SIOCBONDSLAVEINFOQUERY:
3873                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3874
3875                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3876                         return -EFAULT;
3877                 }
3878
3879                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3880                 if (res == 0) {
3881                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3882                                 return -EFAULT;
3883                         }
3884                 }
3885
3886                 return res;
3887         default:
3888                 /* Go on */
3889                 break;
3890         }
3891
3892         if (!capable(CAP_NET_ADMIN)) {
3893                 return -EPERM;
3894         }
3895
3896         down_write(&(bonding_rwsem));
3897         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3898
3899         dprintk("slave_dev=%p: \n", slave_dev);
3900
3901         if (!slave_dev) {
3902                 res = -ENODEV;
3903         } else {
3904                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3905                 switch (cmd) {
3906                 case BOND_ENSLAVE_OLD:
3907                 case SIOCBONDENSLAVE:
3908                         res = bond_enslave(bond_dev, slave_dev);
3909                         break;
3910                 case BOND_RELEASE_OLD:
3911                 case SIOCBONDRELEASE:
3912                         res = bond_release(bond_dev, slave_dev);
3913                         break;
3914                 case BOND_SETHWADDR_OLD:
3915                 case SIOCBONDSETHWADDR:
3916                         res = bond_sethwaddr(bond_dev, slave_dev);
3917                         break;
3918                 case BOND_CHANGE_ACTIVE_OLD:
3919                 case SIOCBONDCHANGEACTIVE:
3920                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3921                         break;
3922                 default:
3923                         res = -EOPNOTSUPP;
3924                 }
3925
3926                 dev_put(slave_dev);
3927         }
3928
3929         up_write(&(bonding_rwsem));
3930         return res;
3931 }
3932
3933 static void bond_set_multicast_list(struct net_device *bond_dev)
3934 {
3935         struct bonding *bond = bond_dev->priv;
3936         struct dev_mc_list *dmi;
3937
3938         write_lock_bh(&bond->lock);
3939
3940         /*
3941          * Do promisc before checking multicast_mode
3942          */
3943         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3944                 bond_set_promiscuity(bond, 1);
3945         }
3946
3947         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3948                 bond_set_promiscuity(bond, -1);
3949         }
3950
3951         /* set allmulti flag to slaves */
3952         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3953                 bond_set_allmulti(bond, 1);
3954         }
3955
3956         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3957                 bond_set_allmulti(bond, -1);
3958         }
3959
3960         bond->flags = bond_dev->flags;
3961
3962         /* looking for addresses to add to slaves' mc list */
3963         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3964                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3965                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3966                 }
3967         }
3968
3969         /* looking for addresses to delete from slaves' list */
3970         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3971                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3972                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3973                 }
3974         }
3975
3976         /* save master's multicast list */
3977         bond_mc_list_destroy(bond);
3978         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3979
3980         write_unlock_bh(&bond->lock);
3981 }
3982
3983 /*
3984  * Change the MTU of all of a master's slaves to match the master
3985  */
3986 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3987 {
3988         struct bonding *bond = bond_dev->priv;
3989         struct slave *slave, *stop_at;
3990         int res = 0;
3991         int i;
3992
3993         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3994                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3995
3996         /* Can't hold bond->lock with bh disabled here since
3997          * some base drivers panic. On the other hand we can't
3998          * hold bond->lock without bh disabled because we'll
3999          * deadlock. The only solution is to rely on the fact
4000          * that we're under rtnl_lock here, and the slaves
4001          * list won't change. This doesn't solve the problem
4002          * of setting the slave's MTU while it is
4003          * transmitting, but the assumption is that the base
4004          * driver can handle that.
4005          *
4006          * TODO: figure out a way to safely iterate the slaves
4007          * list, but without holding a lock around the actual
4008          * call to the base driver.
4009          */
4010
4011         bond_for_each_slave(bond, slave, i) {
4012                 dprintk("s %p s->p %p c_m %p\n", slave,
4013                         slave->prev, slave->dev->change_mtu);
4014
4015                 res = dev_set_mtu(slave->dev, new_mtu);
4016
4017                 if (res) {
4018                         /* If we failed to set the slave's mtu to the new value
4019                          * we must abort the operation even in ACTIVE_BACKUP
4020                          * mode, because if we allow the backup slaves to have
4021                          * different mtu values than the active slave we'll
4022                          * need to change their mtu when doing a failover. That
4023                          * means changing their mtu from timer context, which
4024                          * is probably not a good idea.
4025                          */
4026                         dprintk("err %d %s\n", res, slave->dev->name);
4027                         goto unwind;
4028                 }
4029         }
4030
4031         bond_dev->mtu = new_mtu;
4032
4033         return 0;
4034
4035 unwind:
4036         /* unwind from head to the slave that failed */
4037         stop_at = slave;
4038         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4039                 int tmp_res;
4040
4041                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4042                 if (tmp_res) {
4043                         dprintk("unwind err %d dev %s\n", tmp_res,
4044                                 slave->dev->name);
4045                 }
4046         }
4047
4048         return res;
4049 }
4050
4051 /*
4052  * Change HW address
4053  *
4054  * Note that many devices must be down to change the HW address, and
4055  * downing the master releases all slaves.  We can make bonds full of
4056  * bonding devices to test this, however.
4057  */
4058 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4059 {
4060         struct bonding *bond = bond_dev->priv;
4061         struct sockaddr *sa = addr, tmp_sa;
4062         struct slave *slave, *stop_at;
4063         int res = 0;
4064         int i;
4065
4066         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4067
4068         /*
4069          * If fail_over_mac is enabled, do nothing and return success.
4070          * Returning an error causes ifenslave to fail.
4071          */
4072         if (bond->params.fail_over_mac)
4073                 return 0;
4074
4075         if (!is_valid_ether_addr(sa->sa_data)) {
4076                 return -EADDRNOTAVAIL;
4077         }
4078
4079         /* Can't hold bond->lock with bh disabled here since
4080          * some base drivers panic. On the other hand we can't
4081          * hold bond->lock without bh disabled because we'll
4082          * deadlock. The only solution is to rely on the fact
4083          * that we're under rtnl_lock here, and the slaves
4084          * list won't change. This doesn't solve the problem
4085          * of setting the slave's hw address while it is
4086          * transmitting, but the assumption is that the base
4087          * driver can handle that.
4088          *
4089          * TODO: figure out a way to safely iterate the slaves
4090          * list, but without holding a lock around the actual
4091          * call to the base driver.
4092          */
4093
4094         bond_for_each_slave(bond, slave, i) {
4095                 dprintk("slave %p %s\n", slave, slave->dev->name);
4096
4097                 if (slave->dev->set_mac_address == NULL) {
4098                         res = -EOPNOTSUPP;
4099                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4100                         goto unwind;
4101                 }
4102
4103                 res = dev_set_mac_address(slave->dev, addr);
4104                 if (res) {
4105                         /* TODO: consider downing the slave
4106                          * and retry ?
4107                          * User should expect communications
4108                          * breakage anyway until ARP finish
4109                          * updating, so...
4110                          */
4111                         dprintk("err %d %s\n", res, slave->dev->name);
4112                         goto unwind;
4113                 }
4114         }
4115
4116         /* success */
4117         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4118         return 0;
4119
4120 unwind:
4121         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4122         tmp_sa.sa_family = bond_dev->type;
4123
4124         /* unwind from head to the slave that failed */
4125         stop_at = slave;
4126         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4127                 int tmp_res;
4128
4129                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4130                 if (tmp_res) {
4131                         dprintk("unwind err %d dev %s\n", tmp_res,
4132                                 slave->dev->name);
4133                 }
4134         }
4135
4136         return res;
4137 }
4138
4139 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4140 {
4141         struct bonding *bond = bond_dev->priv;
4142         struct slave *slave, *start_at;
4143         int i, slave_no, res = 1;
4144
4145         read_lock(&bond->lock);
4146
4147         if (!BOND_IS_OK(bond)) {
4148                 goto out;
4149         }
4150
4151         /*
4152          * Concurrent TX may collide on rr_tx_counter; we accept that
4153          * as being rare enough not to justify using an atomic op here
4154          */
4155         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4156
4157         bond_for_each_slave(bond, slave, i) {
4158                 slave_no--;
4159                 if (slave_no < 0) {
4160                         break;
4161                 }
4162         }
4163
4164         start_at = slave;
4165         bond_for_each_slave_from(bond, slave, i, start_at) {
4166                 if (IS_UP(slave->dev) &&
4167                     (slave->link == BOND_LINK_UP) &&
4168                     (slave->state == BOND_STATE_ACTIVE)) {
4169                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4170                         break;
4171                 }
4172         }
4173
4174 out:
4175         if (res) {
4176                 /* no suitable interface, frame not sent */
4177                 dev_kfree_skb(skb);
4178         }
4179         read_unlock(&bond->lock);
4180         return 0;
4181 }
4182
4183
4184 /*
4185  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4186  * the bond has a usable interface.
4187  */
4188 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4189 {
4190         struct bonding *bond = bond_dev->priv;
4191         int res = 1;
4192
4193         read_lock(&bond->lock);
4194         read_lock(&bond->curr_slave_lock);
4195
4196         if (!BOND_IS_OK(bond)) {
4197                 goto out;
4198         }
4199
4200         if (!bond->curr_active_slave)
4201                 goto out;
4202
4203         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4204
4205 out:
4206         if (res) {
4207                 /* no suitable interface, frame not sent */
4208                 dev_kfree_skb(skb);
4209         }
4210         read_unlock(&bond->curr_slave_lock);
4211         read_unlock(&bond->lock);
4212         return 0;
4213 }
4214
4215 /*
4216  * In bond_xmit_xor() , we determine the output device by using a pre-
4217  * determined xmit_hash_policy(), If the selected device is not enabled,
4218  * find the next active slave.
4219  */
4220 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4221 {
4222         struct bonding *bond = bond_dev->priv;
4223         struct slave *slave, *start_at;
4224         int slave_no;
4225         int i;
4226         int res = 1;
4227
4228         read_lock(&bond->lock);
4229
4230         if (!BOND_IS_OK(bond)) {
4231                 goto out;
4232         }
4233
4234         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4235
4236         bond_for_each_slave(bond, slave, i) {
4237                 slave_no--;
4238                 if (slave_no < 0) {
4239                         break;
4240                 }
4241         }
4242
4243         start_at = slave;
4244
4245         bond_for_each_slave_from(bond, slave, i, start_at) {
4246                 if (IS_UP(slave->dev) &&
4247                     (slave->link == BOND_LINK_UP) &&
4248                     (slave->state == BOND_STATE_ACTIVE)) {
4249                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4250                         break;
4251                 }
4252         }
4253
4254 out:
4255         if (res) {
4256                 /* no suitable interface, frame not sent */
4257                 dev_kfree_skb(skb);
4258         }
4259         read_unlock(&bond->lock);
4260         return 0;
4261 }
4262
4263 /*
4264  * in broadcast mode, we send everything to all usable interfaces.
4265  */
4266 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4267 {
4268         struct bonding *bond = bond_dev->priv;
4269         struct slave *slave, *start_at;
4270         struct net_device *tx_dev = NULL;
4271         int i;
4272         int res = 1;
4273
4274         read_lock(&bond->lock);
4275
4276         if (!BOND_IS_OK(bond)) {
4277                 goto out;
4278         }
4279
4280         read_lock(&bond->curr_slave_lock);
4281         start_at = bond->curr_active_slave;
4282         read_unlock(&bond->curr_slave_lock);
4283
4284         if (!start_at) {
4285                 goto out;
4286         }
4287
4288         bond_for_each_slave_from(bond, slave, i, start_at) {
4289                 if (IS_UP(slave->dev) &&
4290                     (slave->link == BOND_LINK_UP) &&
4291                     (slave->state == BOND_STATE_ACTIVE)) {
4292                         if (tx_dev) {
4293                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4294                                 if (!skb2) {
4295                                         printk(KERN_ERR DRV_NAME
4296                                                ": %s: Error: bond_xmit_broadcast(): "
4297                                                "skb_clone() failed\n",
4298                                                bond_dev->name);
4299                                         continue;
4300                                 }
4301
4302                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4303                                 if (res) {
4304                                         dev_kfree_skb(skb2);
4305                                         continue;
4306                                 }
4307                         }
4308                         tx_dev = slave->dev;
4309                 }
4310         }
4311
4312         if (tx_dev) {
4313                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4314         }
4315
4316 out:
4317         if (res) {
4318                 /* no suitable interface, frame not sent */
4319                 dev_kfree_skb(skb);
4320         }
4321         /* frame sent to all suitable interfaces */
4322         read_unlock(&bond->lock);
4323         return 0;
4324 }
4325
4326 /*------------------------- Device initialization ---------------------------*/
4327
4328 static void bond_set_xmit_hash_policy(struct bonding *bond)
4329 {
4330         switch (bond->params.xmit_policy) {
4331         case BOND_XMIT_POLICY_LAYER23:
4332                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4333                 break;
4334         case BOND_XMIT_POLICY_LAYER34:
4335                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4336                 break;
4337         case BOND_XMIT_POLICY_LAYER2:
4338         default:
4339                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4340                 break;
4341         }
4342 }
4343
4344 /*
4345  * set bond mode specific net device operations
4346  */
4347 void bond_set_mode_ops(struct bonding *bond, int mode)
4348 {
4349         struct net_device *bond_dev = bond->dev;
4350
4351         switch (mode) {
4352         case BOND_MODE_ROUNDROBIN:
4353                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4354                 break;
4355         case BOND_MODE_ACTIVEBACKUP:
4356                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4357                 break;
4358         case BOND_MODE_XOR:
4359                 bond_dev->hard_start_xmit = bond_xmit_xor;
4360                 bond_set_xmit_hash_policy(bond);
4361                 break;
4362         case BOND_MODE_BROADCAST:
4363                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4364                 break;
4365         case BOND_MODE_8023AD:
4366                 bond_set_master_3ad_flags(bond);
4367                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4368                 bond_set_xmit_hash_policy(bond);
4369                 break;
4370         case BOND_MODE_ALB:
4371                 bond_set_master_alb_flags(bond);
4372                 /* FALLTHRU */
4373         case BOND_MODE_TLB:
4374                 bond_dev->hard_start_xmit = bond_alb_xmit;
4375                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4376                 break;
4377         default:
4378                 /* Should never happen, mode already checked */
4379                 printk(KERN_ERR DRV_NAME
4380                        ": %s: Error: Unknown bonding mode %d\n",
4381                        bond_dev->name,
4382                        mode);
4383                 break;
4384         }
4385 }
4386
4387 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4388                                     struct ethtool_drvinfo *drvinfo)
4389 {
4390         strncpy(drvinfo->driver, DRV_NAME, 32);
4391         strncpy(drvinfo->version, DRV_VERSION, 32);
4392         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4393 }
4394
4395 static const struct ethtool_ops bond_ethtool_ops = {
4396         .get_drvinfo            = bond_ethtool_get_drvinfo,
4397 };
4398
4399 /*
4400  * Does not allocate but creates a /proc entry.
4401  * Allowed to fail.
4402  */
4403 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4404 {
4405         struct bonding *bond = bond_dev->priv;
4406
4407         dprintk("Begin bond_init for %s\n", bond_dev->name);
4408
4409         /* initialize rwlocks */
4410         rwlock_init(&bond->lock);
4411         rwlock_init(&bond->curr_slave_lock);
4412
4413         bond->params = *params; /* copy params struct */
4414
4415         bond->wq = create_singlethread_workqueue(bond_dev->name);
4416         if (!bond->wq)
4417                 return -ENOMEM;
4418
4419         /* Initialize pointers */
4420         bond->first_slave = NULL;
4421         bond->curr_active_slave = NULL;
4422         bond->current_arp_slave = NULL;
4423         bond->primary_slave = NULL;
4424         bond->dev = bond_dev;
4425         bond->send_grat_arp = 0;
4426         bond->setup_by_slave = 0;
4427         INIT_LIST_HEAD(&bond->vlan_list);
4428
4429         /* Initialize the device entry points */
4430         bond_dev->open = bond_open;
4431         bond_dev->stop = bond_close;
4432         bond_dev->get_stats = bond_get_stats;
4433         bond_dev->do_ioctl = bond_do_ioctl;
4434         bond_dev->ethtool_ops = &bond_ethtool_ops;
4435         bond_dev->set_multicast_list = bond_set_multicast_list;
4436         bond_dev->change_mtu = bond_change_mtu;
4437         bond_dev->set_mac_address = bond_set_mac_address;
4438         bond_dev->validate_addr = NULL;
4439
4440         bond_set_mode_ops(bond, bond->params.mode);
4441
4442         bond_dev->destructor = free_netdev;
4443
4444         /* Initialize the device options */
4445         bond_dev->tx_queue_len = 0;
4446         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4447         bond_dev->priv_flags |= IFF_BONDING;
4448
4449         /* At first, we block adding VLANs. That's the only way to
4450          * prevent problems that occur when adding VLANs over an
4451          * empty bond. The block will be removed once non-challenged
4452          * slaves are enslaved.
4453          */
4454         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4455
4456         /* don't acquire bond device's netif_tx_lock when
4457          * transmitting */
4458         bond_dev->features |= NETIF_F_LLTX;
4459
4460         /* By default, we declare the bond to be fully
4461          * VLAN hardware accelerated capable. Special
4462          * care is taken in the various xmit functions
4463          * when there are slaves that are not hw accel
4464          * capable
4465          */
4466         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4467         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4468         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4469         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4470                                NETIF_F_HW_VLAN_RX |
4471                                NETIF_F_HW_VLAN_FILTER);
4472
4473 #ifdef CONFIG_PROC_FS
4474         bond_create_proc_entry(bond);
4475 #endif
4476         list_add_tail(&bond->bond_list, &bond_dev_list);
4477
4478         return 0;
4479 }
4480
4481 /* De-initialize device specific data.
4482  * Caller must hold rtnl_lock.
4483  */
4484 static void bond_deinit(struct net_device *bond_dev)
4485 {
4486         struct bonding *bond = bond_dev->priv;
4487
4488         list_del(&bond->bond_list);
4489
4490 #ifdef CONFIG_PROC_FS
4491         bond_remove_proc_entry(bond);
4492 #endif
4493 }
4494
4495 /* Unregister and free all bond devices.
4496  * Caller must hold rtnl_lock.
4497  */
4498 static void bond_free_all(void)
4499 {
4500         struct bonding *bond, *nxt;
4501
4502         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4503                 struct net_device *bond_dev = bond->dev;
4504
4505                 bond_mc_list_destroy(bond);
4506                 /* Release the bonded slaves */
4507                 bond_release_all(bond_dev);
4508                 bond_deinit(bond_dev);
4509                 unregister_netdevice(bond_dev);
4510         }
4511
4512 #ifdef CONFIG_PROC_FS
4513         bond_destroy_proc_dir();
4514 #endif
4515 }
4516
4517 /*------------------------- Module initialization ---------------------------*/
4518
4519 /*
4520  * Convert string input module parms.  Accept either the
4521  * number of the mode or its string name.
4522  */
4523 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4524 {
4525         int i;
4526
4527         for (i = 0; tbl[i].modename; i++) {
4528                 if ((isdigit(*mode_arg) &&
4529                      tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4530                     (strcmp(mode_arg, tbl[i].modename) == 0)) {
4531                         return tbl[i].mode;
4532                 }
4533         }
4534
4535         return -1;
4536 }
4537
4538 static int bond_check_params(struct bond_params *params)
4539 {
4540         int arp_validate_value;
4541
4542         /*
4543          * Convert string parameters.
4544          */
4545         if (mode) {
4546                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4547                 if (bond_mode == -1) {
4548                         printk(KERN_ERR DRV_NAME
4549                                ": Error: Invalid bonding mode \"%s\"\n",
4550                                mode == NULL ? "NULL" : mode);
4551                         return -EINVAL;
4552                 }
4553         }
4554
4555         if (xmit_hash_policy) {
4556                 if ((bond_mode != BOND_MODE_XOR) &&
4557                     (bond_mode != BOND_MODE_8023AD)) {
4558                         printk(KERN_INFO DRV_NAME
4559                                ": xor_mode param is irrelevant in mode %s\n",
4560                                bond_mode_name(bond_mode));
4561                 } else {
4562                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4563                                                         xmit_hashtype_tbl);
4564                         if (xmit_hashtype == -1) {
4565                                 printk(KERN_ERR DRV_NAME
4566                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4567                                 xmit_hash_policy == NULL ? "NULL" :
4568                                        xmit_hash_policy);
4569                                 return -EINVAL;
4570                         }
4571                 }
4572         }
4573
4574         if (lacp_rate) {
4575                 if (bond_mode != BOND_MODE_8023AD) {
4576                         printk(KERN_INFO DRV_NAME
4577                                ": lacp_rate param is irrelevant in mode %s\n",
4578                                bond_mode_name(bond_mode));
4579                 } else {
4580                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4581                         if (lacp_fast == -1) {
4582                                 printk(KERN_ERR DRV_NAME
4583                                        ": Error: Invalid lacp rate \"%s\"\n",
4584                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4585                                 return -EINVAL;
4586                         }
4587                 }
4588         }
4589
4590         if (max_bonds < 1 || max_bonds > INT_MAX) {
4591                 printk(KERN_WARNING DRV_NAME
4592                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4593                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4594                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4595                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4596         }
4597
4598         if (miimon < 0) {
4599                 printk(KERN_WARNING DRV_NAME
4600                        ": Warning: miimon module parameter (%d), "
4601                        "not in range 0-%d, so it was reset to %d\n",
4602                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4603                 miimon = BOND_LINK_MON_INTERV;
4604         }
4605
4606         if (updelay < 0) {
4607                 printk(KERN_WARNING DRV_NAME
4608                        ": Warning: updelay module parameter (%d), "
4609                        "not in range 0-%d, so it was reset to 0\n",
4610                        updelay, INT_MAX);
4611                 updelay = 0;
4612         }
4613
4614         if (downdelay < 0) {
4615                 printk(KERN_WARNING DRV_NAME
4616                        ": Warning: downdelay module parameter (%d), "
4617                        "not in range 0-%d, so it was reset to 0\n",
4618                        downdelay, INT_MAX);
4619                 downdelay = 0;
4620         }
4621
4622         if ((use_carrier != 0) && (use_carrier != 1)) {
4623                 printk(KERN_WARNING DRV_NAME
4624                        ": Warning: use_carrier module parameter (%d), "
4625                        "not of valid value (0/1), so it was set to 1\n",
4626                        use_carrier);
4627                 use_carrier = 1;
4628         }
4629
4630         /* reset values for 802.3ad */
4631         if (bond_mode == BOND_MODE_8023AD) {
4632                 if (!miimon) {
4633                         printk(KERN_WARNING DRV_NAME
4634                                ": Warning: miimon must be specified, "
4635                                "otherwise bonding will not detect link "
4636                                "failure, speed and duplex which are "
4637                                "essential for 802.3ad operation\n");
4638                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4639                         miimon = 100;
4640                 }
4641         }
4642
4643         /* reset values for TLB/ALB */
4644         if ((bond_mode == BOND_MODE_TLB) ||
4645             (bond_mode == BOND_MODE_ALB)) {
4646                 if (!miimon) {
4647                         printk(KERN_WARNING DRV_NAME
4648                                ": Warning: miimon must be specified, "
4649                                "otherwise bonding will not detect link "
4650                                "failure and link speed which are essential "
4651                                "for TLB/ALB load balancing\n");
4652                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4653                         miimon = 100;
4654                 }
4655         }
4656
4657         if (bond_mode == BOND_MODE_ALB) {
4658                 printk(KERN_NOTICE DRV_NAME
4659                        ": In ALB mode you might experience client "
4660                        "disconnections upon reconnection of a link if the "
4661                        "bonding module updelay parameter (%d msec) is "
4662                        "incompatible with the forwarding delay time of the "
4663                        "switch\n",
4664                        updelay);
4665         }
4666
4667         if (!miimon) {
4668                 if (updelay || downdelay) {
4669                         /* just warn the user the up/down delay will have
4670                          * no effect since miimon is zero...
4671                          */
4672                         printk(KERN_WARNING DRV_NAME
4673                                ": Warning: miimon module parameter not set "
4674                                "and updelay (%d) or downdelay (%d) module "
4675                                "parameter is set; updelay and downdelay have "
4676                                "no effect unless miimon is set\n",
4677                                updelay, downdelay);
4678                 }
4679         } else {
4680                 /* don't allow arp monitoring */
4681                 if (arp_interval) {
4682                         printk(KERN_WARNING DRV_NAME
4683                                ": Warning: miimon (%d) and arp_interval (%d) "
4684                                "can't be used simultaneously, disabling ARP "
4685                                "monitoring\n",
4686                                miimon, arp_interval);
4687                         arp_interval = 0;
4688                 }
4689
4690                 if ((updelay % miimon) != 0) {
4691                         printk(KERN_WARNING DRV_NAME
4692                                ": Warning: updelay (%d) is not a multiple "
4693                                "of miimon (%d), updelay rounded to %d ms\n",
4694                                updelay, miimon, (updelay / miimon) * miimon);
4695                 }
4696
4697                 updelay /= miimon;
4698
4699                 if ((downdelay % miimon) != 0) {
4700                         printk(KERN_WARNING DRV_NAME
4701                                ": Warning: downdelay (%d) is not a multiple "
4702                                "of miimon (%d), downdelay rounded to %d ms\n",
4703                                downdelay, miimon,
4704                                (downdelay / miimon) * miimon);
4705                 }
4706
4707                 downdelay /= miimon;
4708         }
4709
4710         if (arp_interval < 0) {
4711                 printk(KERN_WARNING DRV_NAME
4712                        ": Warning: arp_interval module parameter (%d) "
4713                        ", not in range 0-%d, so it was reset to %d\n",
4714                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4715                 arp_interval = BOND_LINK_ARP_INTERV;
4716         }
4717
4718         for (arp_ip_count = 0;
4719              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4720              arp_ip_count++) {
4721                 /* not complete check, but should be good enough to
4722                    catch mistakes */
4723                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4724                         printk(KERN_WARNING DRV_NAME
4725                                ": Warning: bad arp_ip_target module parameter "
4726                                "(%s), ARP monitoring will not be performed\n",
4727                                arp_ip_target[arp_ip_count]);
4728                         arp_interval = 0;
4729                 } else {
4730                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4731                         arp_target[arp_ip_count] = ip;
4732                 }
4733         }
4734
4735         if (arp_interval && !arp_ip_count) {
4736                 /* don't allow arping if no arp_ip_target given... */
4737                 printk(KERN_WARNING DRV_NAME
4738                        ": Warning: arp_interval module parameter (%d) "
4739                        "specified without providing an arp_ip_target "
4740                        "parameter, arp_interval was reset to 0\n",
4741                        arp_interval);
4742                 arp_interval = 0;
4743         }
4744
4745         if (arp_validate) {
4746                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4747                         printk(KERN_ERR DRV_NAME
4748                ": arp_validate only supported in active-backup mode\n");
4749                         return -EINVAL;
4750                 }
4751                 if (!arp_interval) {
4752                         printk(KERN_ERR DRV_NAME
4753                                ": arp_validate requires arp_interval\n");
4754                         return -EINVAL;
4755                 }
4756
4757                 arp_validate_value = bond_parse_parm(arp_validate,
4758                                                      arp_validate_tbl);
4759                 if (arp_validate_value == -1) {
4760                         printk(KERN_ERR DRV_NAME
4761                                ": Error: invalid arp_validate \"%s\"\n",
4762                                arp_validate == NULL ? "NULL" : arp_validate);
4763                         return -EINVAL;
4764                 }
4765         } else
4766                 arp_validate_value = 0;
4767
4768         if (miimon) {
4769                 printk(KERN_INFO DRV_NAME
4770                        ": MII link monitoring set to %d ms\n",
4771                        miimon);
4772         } else if (arp_interval) {
4773                 int i;
4774
4775                 printk(KERN_INFO DRV_NAME
4776                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4777                        arp_interval,
4778                        arp_validate_tbl[arp_validate_value].modename,
4779                        arp_ip_count);
4780
4781                 for (i = 0; i < arp_ip_count; i++)
4782                         printk (" %s", arp_ip_target[i]);
4783
4784                 printk("\n");
4785
4786         } else {
4787                 /* miimon and arp_interval not set, we need one so things
4788                  * work as expected, see bonding.txt for details
4789                  */
4790                 printk(KERN_WARNING DRV_NAME
4791                        ": Warning: either miimon or arp_interval and "
4792                        "arp_ip_target module parameters must be specified, "
4793                        "otherwise bonding will not detect link failures! see "
4794                        "bonding.txt for details.\n");
4795         }
4796
4797         if (primary && !USES_PRIMARY(bond_mode)) {
4798                 /* currently, using a primary only makes sense
4799                  * in active backup, TLB or ALB modes
4800                  */
4801                 printk(KERN_WARNING DRV_NAME
4802                        ": Warning: %s primary device specified but has no "
4803                        "effect in %s mode\n",
4804                        primary, bond_mode_name(bond_mode));
4805                 primary = NULL;
4806         }
4807
4808         if (fail_over_mac && (bond_mode != BOND_MODE_ACTIVEBACKUP))
4809                 printk(KERN_WARNING DRV_NAME
4810                        ": Warning: fail_over_mac only affects "
4811                        "active-backup mode.\n");
4812
4813         /* fill params struct with the proper values */
4814         params->mode = bond_mode;
4815         params->xmit_policy = xmit_hashtype;
4816         params->miimon = miimon;
4817         params->arp_interval = arp_interval;
4818         params->arp_validate = arp_validate_value;
4819         params->updelay = updelay;
4820         params->downdelay = downdelay;
4821         params->use_carrier = use_carrier;
4822         params->lacp_fast = lacp_fast;
4823         params->primary[0] = 0;
4824         params->fail_over_mac = fail_over_mac;
4825
4826         if (primary) {
4827                 strncpy(params->primary, primary, IFNAMSIZ);
4828                 params->primary[IFNAMSIZ - 1] = 0;
4829         }
4830
4831         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4832
4833         return 0;
4834 }
4835
4836 static struct lock_class_key bonding_netdev_xmit_lock_key;
4837
4838 /* Create a new bond based on the specified name and bonding parameters.
4839  * If name is NULL, obtain a suitable "bond%d" name for us.
4840  * Caller must NOT hold rtnl_lock; we need to release it here before we
4841  * set up our sysfs entries.
4842  */
4843 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4844 {
4845         struct net_device *bond_dev;
4846         int res;
4847
4848         rtnl_lock();
4849         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4850                                 ether_setup);
4851         if (!bond_dev) {
4852                 printk(KERN_ERR DRV_NAME
4853                        ": %s: eek! can't alloc netdev!\n",
4854                        name);
4855                 res = -ENOMEM;
4856                 goto out_rtnl;
4857         }
4858
4859         if (!name) {
4860                 res = dev_alloc_name(bond_dev, "bond%d");
4861                 if (res < 0)
4862                         goto out_netdev;
4863         }
4864
4865         /* bond_init() must be called after dev_alloc_name() (for the
4866          * /proc files), but before register_netdevice(), because we
4867          * need to set function pointers.
4868          */
4869
4870         res = bond_init(bond_dev, params);
4871         if (res < 0) {
4872                 goto out_netdev;
4873         }
4874
4875         res = register_netdevice(bond_dev);
4876         if (res < 0) {
4877                 goto out_bond;
4878         }
4879
4880         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4881
4882         if (newbond)
4883                 *newbond = bond_dev->priv;
4884
4885         netif_carrier_off(bond_dev);
4886
4887         rtnl_unlock(); /* allows sysfs registration of net device */
4888         res = bond_create_sysfs_entry(bond_dev->priv);
4889         if (res < 0) {
4890                 rtnl_lock();
4891                 goto out_bond;
4892         }
4893
4894         return 0;
4895
4896 out_bond:
4897         bond_deinit(bond_dev);
4898 out_netdev:
4899         free_netdev(bond_dev);
4900 out_rtnl:
4901         rtnl_unlock();
4902         return res;
4903 }
4904
4905 static void bond_work_cancel_all(struct bonding *bond)
4906 {
4907         write_lock_bh(&bond->lock);
4908         bond->kill_timers = 1;
4909         write_unlock_bh(&bond->lock);
4910
4911         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4912                 cancel_delayed_work(&bond->mii_work);
4913
4914         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4915                 cancel_delayed_work(&bond->arp_work);
4916
4917         if (bond->params.mode == BOND_MODE_ALB &&
4918             delayed_work_pending(&bond->alb_work))
4919                 cancel_delayed_work(&bond->alb_work);
4920
4921         if (bond->params.mode == BOND_MODE_8023AD &&
4922             delayed_work_pending(&bond->ad_work))
4923                 cancel_delayed_work(&bond->ad_work);
4924 }
4925
4926 static int __init bonding_init(void)
4927 {
4928         int i;
4929         int res;
4930         struct bonding *bond, *nxt;
4931
4932         printk(KERN_INFO "%s", version);
4933
4934         res = bond_check_params(&bonding_defaults);
4935         if (res) {
4936                 goto out;
4937         }
4938
4939 #ifdef CONFIG_PROC_FS
4940         bond_create_proc_dir();
4941 #endif
4942         for (i = 0; i < max_bonds; i++) {
4943                 res = bond_create(NULL, &bonding_defaults, NULL);
4944                 if (res)
4945                         goto err;
4946         }
4947
4948         res = bond_create_sysfs();
4949         if (res)
4950                 goto err;
4951
4952         register_netdevice_notifier(&bond_netdev_notifier);
4953         register_inetaddr_notifier(&bond_inetaddr_notifier);
4954
4955         goto out;
4956 err:
4957         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4958                 bond_work_cancel_all(bond);
4959                 destroy_workqueue(bond->wq);
4960         }
4961
4962         rtnl_lock();
4963         bond_free_all();
4964         bond_destroy_sysfs();
4965         rtnl_unlock();
4966 out:
4967         return res;
4968
4969 }
4970
4971 static void __exit bonding_exit(void)
4972 {
4973         unregister_netdevice_notifier(&bond_netdev_notifier);
4974         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4975
4976         rtnl_lock();
4977         bond_free_all();
4978         bond_destroy_sysfs();
4979         rtnl_unlock();
4980 }
4981
4982 module_init(bonding_init);
4983 module_exit(bonding_exit);
4984 MODULE_LICENSE("GPL");
4985 MODULE_VERSION(DRV_VERSION);
4986 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4987 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4988 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4989
4990 /*
4991  * Local variables:
4992  *  c-indent-level: 8
4993  *  c-basic-offset: 8
4994  *  tab-width: 8
4995  * End:
4996  */
4997