Merge branch 'upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/linville...
[linux-2.6] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 //#define BONDING_DEBUG 1
35
36 #include <linux/config.h>
37 #include <linux/kernel.h>
38 #include <linux/module.h>
39 #include <linux/sched.h>
40 #include <linux/types.h>
41 #include <linux/fcntl.h>
42 #include <linux/interrupt.h>
43 #include <linux/ptrace.h>
44 #include <linux/ioport.h>
45 #include <linux/in.h>
46 #include <net/ip.h>
47 #include <linux/ip.h>
48 #include <linux/tcp.h>
49 #include <linux/udp.h>
50 #include <linux/slab.h>
51 #include <linux/string.h>
52 #include <linux/init.h>
53 #include <linux/timer.h>
54 #include <linux/socket.h>
55 #include <linux/ctype.h>
56 #include <linux/inet.h>
57 #include <linux/bitops.h>
58 #include <asm/system.h>
59 #include <asm/io.h>
60 #include <asm/dma.h>
61 #include <asm/uaccess.h>
62 #include <linux/errno.h>
63 #include <linux/netdevice.h>
64 #include <linux/inetdevice.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
73 #include <net/arp.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <net/route.h>
79 #include "bonding.h"
80 #include "bond_3ad.h"
81 #include "bond_alb.h"
82
83 /*---------------------------- Module parameters ----------------------------*/
84
85 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
86 #define BOND_LINK_MON_INTERV    0
87 #define BOND_LINK_ARP_INTERV    0
88
89 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
90 static int miimon       = BOND_LINK_MON_INTERV;
91 static int updelay      = 0;
92 static int downdelay    = 0;
93 static int use_carrier  = 1;
94 static char *mode       = NULL;
95 static char *primary    = NULL;
96 static char *lacp_rate  = NULL;
97 static char *xmit_hash_policy = NULL;
98 static int arp_interval = BOND_LINK_ARP_INTERV;
99 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
100 struct bond_params bonding_defaults;
101
102 module_param(max_bonds, int, 0);
103 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
104 module_param(miimon, int, 0);
105 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
106 module_param(updelay, int, 0);
107 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
108 module_param(downdelay, int, 0);
109 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
110                             "in milliseconds");
111 module_param(use_carrier, int, 0);
112 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
113                               "0 for off, 1 for on (default)");
114 module_param(mode, charp, 0);
115 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
116                        "1 for active-backup, 2 for balance-xor, "
117                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
118                        "6 for balance-alb");
119 module_param(primary, charp, 0);
120 MODULE_PARM_DESC(primary, "Primary network device to use");
121 module_param(lacp_rate, charp, 0);
122 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
123                             "(slow/fast)");
124 module_param(xmit_hash_policy, charp, 0);
125 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
126                                    ", 1 for layer 3+4");
127 module_param(arp_interval, int, 0);
128 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
129 module_param_array(arp_ip_target, charp, NULL, 0);
130 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
131
132 /*----------------------------- Global variables ----------------------------*/
133
134 static const char * const version =
135         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
136
137 LIST_HEAD(bond_dev_list);
138
139 #ifdef CONFIG_PROC_FS
140 static struct proc_dir_entry *bond_proc_dir = NULL;
141 #endif
142
143 extern struct rw_semaphore bonding_rwsem;
144 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
145 static int arp_ip_count = 0;
146 static int bond_mode    = BOND_MODE_ROUNDROBIN;
147 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
148 static int lacp_fast    = 0;
149
150
151 struct bond_parm_tbl bond_lacp_tbl[] = {
152 {       "slow",         AD_LACP_SLOW},
153 {       "fast",         AD_LACP_FAST},
154 {       NULL,           -1},
155 };
156
157 struct bond_parm_tbl bond_mode_tbl[] = {
158 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
159 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
160 {       "balance-xor",          BOND_MODE_XOR},
161 {       "broadcast",            BOND_MODE_BROADCAST},
162 {       "802.3ad",              BOND_MODE_8023AD},
163 {       "balance-tlb",          BOND_MODE_TLB},
164 {       "balance-alb",          BOND_MODE_ALB},
165 {       NULL,                   -1},
166 };
167
168 struct bond_parm_tbl xmit_hashtype_tbl[] = {
169 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
170 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
171 {       NULL,                   -1},
172 };
173
174 /*-------------------------- Forward declarations ---------------------------*/
175
176 static void bond_send_gratuitous_arp(struct bonding *bond);
177
178 /*---------------------------- General routines -----------------------------*/
179
180 const char *bond_mode_name(int mode)
181 {
182         switch (mode) {
183         case BOND_MODE_ROUNDROBIN :
184                 return "load balancing (round-robin)";
185         case BOND_MODE_ACTIVEBACKUP :
186                 return "fault-tolerance (active-backup)";
187         case BOND_MODE_XOR :
188                 return "load balancing (xor)";
189         case BOND_MODE_BROADCAST :
190                 return "fault-tolerance (broadcast)";
191         case BOND_MODE_8023AD:
192                 return "IEEE 802.3ad Dynamic link aggregation";
193         case BOND_MODE_TLB:
194                 return "transmit load balancing";
195         case BOND_MODE_ALB:
196                 return "adaptive load balancing";
197         default:
198                 return "unknown";
199         }
200 }
201
202 /*---------------------------------- VLAN -----------------------------------*/
203
204 /**
205  * bond_add_vlan - add a new vlan id on bond
206  * @bond: bond that got the notification
207  * @vlan_id: the vlan id to add
208  *
209  * Returns -ENOMEM if allocation failed.
210  */
211 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
212 {
213         struct vlan_entry *vlan;
214
215         dprintk("bond: %s, vlan id %d\n",
216                 (bond ? bond->dev->name: "None"), vlan_id);
217
218         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
219         if (!vlan) {
220                 return -ENOMEM;
221         }
222
223         INIT_LIST_HEAD(&vlan->vlan_list);
224         vlan->vlan_id = vlan_id;
225         vlan->vlan_ip = 0;
226
227         write_lock_bh(&bond->lock);
228
229         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
230
231         write_unlock_bh(&bond->lock);
232
233         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
234
235         return 0;
236 }
237
238 /**
239  * bond_del_vlan - delete a vlan id from bond
240  * @bond: bond that got the notification
241  * @vlan_id: the vlan id to delete
242  *
243  * returns -ENODEV if @vlan_id was not found in @bond.
244  */
245 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
246 {
247         struct vlan_entry *vlan, *next;
248         int res = -ENODEV;
249
250         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
251
252         write_lock_bh(&bond->lock);
253
254         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
255                 if (vlan->vlan_id == vlan_id) {
256                         list_del(&vlan->vlan_list);
257
258                         if ((bond->params.mode == BOND_MODE_TLB) ||
259                             (bond->params.mode == BOND_MODE_ALB)) {
260                                 bond_alb_clear_vlan(bond, vlan_id);
261                         }
262
263                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
264                                 bond->dev->name);
265
266                         kfree(vlan);
267
268                         if (list_empty(&bond->vlan_list) &&
269                             (bond->slave_cnt == 0)) {
270                                 /* Last VLAN removed and no slaves, so
271                                  * restore block on adding VLANs. This will
272                                  * be removed once new slaves that are not
273                                  * VLAN challenged will be added.
274                                  */
275                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
276                         }
277
278                         res = 0;
279                         goto out;
280                 }
281         }
282
283         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
284                 bond->dev->name);
285
286 out:
287         write_unlock_bh(&bond->lock);
288         return res;
289 }
290
291 /**
292  * bond_has_challenged_slaves
293  * @bond: the bond we're working on
294  *
295  * Searches the slave list. Returns 1 if a vlan challenged slave
296  * was found, 0 otherwise.
297  *
298  * Assumes bond->lock is held.
299  */
300 static int bond_has_challenged_slaves(struct bonding *bond)
301 {
302         struct slave *slave;
303         int i;
304
305         bond_for_each_slave(bond, slave, i) {
306                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
307                         dprintk("found VLAN challenged slave - %s\n",
308                                 slave->dev->name);
309                         return 1;
310                 }
311         }
312
313         dprintk("no VLAN challenged slaves found\n");
314         return 0;
315 }
316
317 /**
318  * bond_next_vlan - safely skip to the next item in the vlans list.
319  * @bond: the bond we're working on
320  * @curr: item we're advancing from
321  *
322  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
323  * or @curr->next otherwise (even if it is @curr itself again).
324  * 
325  * Caller must hold bond->lock
326  */
327 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
328 {
329         struct vlan_entry *next, *last;
330
331         if (list_empty(&bond->vlan_list)) {
332                 return NULL;
333         }
334
335         if (!curr) {
336                 next = list_entry(bond->vlan_list.next,
337                                   struct vlan_entry, vlan_list);
338         } else {
339                 last = list_entry(bond->vlan_list.prev,
340                                   struct vlan_entry, vlan_list);
341                 if (last == curr) {
342                         next = list_entry(bond->vlan_list.next,
343                                           struct vlan_entry, vlan_list);
344                 } else {
345                         next = list_entry(curr->vlan_list.next,
346                                           struct vlan_entry, vlan_list);
347                 }
348         }
349
350         return next;
351 }
352
353 /**
354  * bond_dev_queue_xmit - Prepare skb for xmit.
355  * 
356  * @bond: bond device that got this skb for tx.
357  * @skb: hw accel VLAN tagged skb to transmit
358  * @slave_dev: slave that is supposed to xmit this skbuff
359  * 
360  * When the bond gets an skb to transmit that is
361  * already hardware accelerated VLAN tagged, and it
362  * needs to relay this skb to a slave that is not
363  * hw accel capable, the skb needs to be "unaccelerated",
364  * i.e. strip the hwaccel tag and re-insert it as part
365  * of the payload.
366  */
367 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
368 {
369         unsigned short vlan_id;
370
371         if (!list_empty(&bond->vlan_list) &&
372             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
373             vlan_get_tag(skb, &vlan_id) == 0) {
374                 skb->dev = slave_dev;
375                 skb = vlan_put_tag(skb, vlan_id);
376                 if (!skb) {
377                         /* vlan_put_tag() frees the skb in case of error,
378                          * so return success here so the calling functions
379                          * won't attempt to free is again.
380                          */
381                         return 0;
382                 }
383         } else {
384                 skb->dev = slave_dev;
385         }
386
387         skb->priority = 1;
388         dev_queue_xmit(skb);
389
390         return 0;
391 }
392
393 /*
394  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
395  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
396  * lock because:
397  * a. This operation is performed in IOCTL context,
398  * b. The operation is protected by the RTNL semaphore in the 8021q code,
399  * c. Holding a lock with BH disabled while directly calling a base driver
400  *    entry point is generally a BAD idea.
401  * 
402  * The design of synchronization/protection for this operation in the 8021q
403  * module is good for one or more VLAN devices over a single physical device
404  * and cannot be extended for a teaming solution like bonding, so there is a
405  * potential race condition here where a net device from the vlan group might
406  * be referenced (either by a base driver or the 8021q code) while it is being
407  * removed from the system. However, it turns out we're not making matters
408  * worse, and if it works for regular VLAN usage it will work here too.
409 */
410
411 /**
412  * bond_vlan_rx_register - Propagates registration to slaves
413  * @bond_dev: bonding net device that got called
414  * @grp: vlan group being registered
415  */
416 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
417 {
418         struct bonding *bond = bond_dev->priv;
419         struct slave *slave;
420         int i;
421
422         bond->vlgrp = grp;
423
424         bond_for_each_slave(bond, slave, i) {
425                 struct net_device *slave_dev = slave->dev;
426
427                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
428                     slave_dev->vlan_rx_register) {
429                         slave_dev->vlan_rx_register(slave_dev, grp);
430                 }
431         }
432 }
433
434 /**
435  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
436  * @bond_dev: bonding net device that got called
437  * @vid: vlan id being added
438  */
439 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
440 {
441         struct bonding *bond = bond_dev->priv;
442         struct slave *slave;
443         int i, res;
444
445         bond_for_each_slave(bond, slave, i) {
446                 struct net_device *slave_dev = slave->dev;
447
448                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
449                     slave_dev->vlan_rx_add_vid) {
450                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
451                 }
452         }
453
454         res = bond_add_vlan(bond, vid);
455         if (res) {
456                 printk(KERN_ERR DRV_NAME
457                        ": %s: Error: Failed to add vlan id %d\n",
458                        bond_dev->name, vid);
459         }
460 }
461
462 /**
463  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
464  * @bond_dev: bonding net device that got called
465  * @vid: vlan id being removed
466  */
467 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
468 {
469         struct bonding *bond = bond_dev->priv;
470         struct slave *slave;
471         struct net_device *vlan_dev;
472         int i, res;
473
474         bond_for_each_slave(bond, slave, i) {
475                 struct net_device *slave_dev = slave->dev;
476
477                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
478                     slave_dev->vlan_rx_kill_vid) {
479                         /* Save and then restore vlan_dev in the grp array,
480                          * since the slave's driver might clear it.
481                          */
482                         vlan_dev = bond->vlgrp->vlan_devices[vid];
483                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
484                         bond->vlgrp->vlan_devices[vid] = vlan_dev;
485                 }
486         }
487
488         res = bond_del_vlan(bond, vid);
489         if (res) {
490                 printk(KERN_ERR DRV_NAME
491                        ": %s: Error: Failed to remove vlan id %d\n",
492                        bond_dev->name, vid);
493         }
494 }
495
496 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
497 {
498         struct vlan_entry *vlan;
499
500         write_lock_bh(&bond->lock);
501
502         if (list_empty(&bond->vlan_list)) {
503                 goto out;
504         }
505
506         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
507             slave_dev->vlan_rx_register) {
508                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
509         }
510
511         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
512             !(slave_dev->vlan_rx_add_vid)) {
513                 goto out;
514         }
515
516         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
517                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
518         }
519
520 out:
521         write_unlock_bh(&bond->lock);
522 }
523
524 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
525 {
526         struct vlan_entry *vlan;
527         struct net_device *vlan_dev;
528
529         write_lock_bh(&bond->lock);
530
531         if (list_empty(&bond->vlan_list)) {
532                 goto out;
533         }
534
535         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
536             !(slave_dev->vlan_rx_kill_vid)) {
537                 goto unreg;
538         }
539
540         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
541                 /* Save and then restore vlan_dev in the grp array,
542                  * since the slave's driver might clear it.
543                  */
544                 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
545                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
546                 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
547         }
548
549 unreg:
550         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
551             slave_dev->vlan_rx_register) {
552                 slave_dev->vlan_rx_register(slave_dev, NULL);
553         }
554
555 out:
556         write_unlock_bh(&bond->lock);
557 }
558
559 /*------------------------------- Link status -------------------------------*/
560
561 /*
562  * Get link speed and duplex from the slave's base driver
563  * using ethtool. If for some reason the call fails or the
564  * values are invalid, fake speed and duplex to 100/Full
565  * and return error.
566  */
567 static int bond_update_speed_duplex(struct slave *slave)
568 {
569         struct net_device *slave_dev = slave->dev;
570         static int (* ioctl)(struct net_device *, struct ifreq *, int);
571         struct ifreq ifr;
572         struct ethtool_cmd etool;
573
574         /* Fake speed and duplex */
575         slave->speed = SPEED_100;
576         slave->duplex = DUPLEX_FULL;
577
578         if (slave_dev->ethtool_ops) {
579                 int res;
580
581                 if (!slave_dev->ethtool_ops->get_settings) {
582                         return -1;
583                 }
584
585                 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
586                 if (res < 0) {
587                         return -1;
588                 }
589
590                 goto verify;
591         }
592
593         ioctl = slave_dev->do_ioctl;
594         strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
595         etool.cmd = ETHTOOL_GSET;
596         ifr.ifr_data = (char*)&etool;
597         if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
598                 return -1;
599         }
600
601 verify:
602         switch (etool.speed) {
603         case SPEED_10:
604         case SPEED_100:
605         case SPEED_1000:
606                 break;
607         default:
608                 return -1;
609         }
610
611         switch (etool.duplex) {
612         case DUPLEX_FULL:
613         case DUPLEX_HALF:
614                 break;
615         default:
616                 return -1;
617         }
618
619         slave->speed = etool.speed;
620         slave->duplex = etool.duplex;
621
622         return 0;
623 }
624
625 /*
626  * if <dev> supports MII link status reporting, check its link status.
627  *
628  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
629  * depening upon the setting of the use_carrier parameter.
630  *
631  * Return either BMSR_LSTATUS, meaning that the link is up (or we
632  * can't tell and just pretend it is), or 0, meaning that the link is
633  * down.
634  *
635  * If reporting is non-zero, instead of faking link up, return -1 if
636  * both ETHTOOL and MII ioctls fail (meaning the device does not
637  * support them).  If use_carrier is set, return whatever it says.
638  * It'd be nice if there was a good way to tell if a driver supports
639  * netif_carrier, but there really isn't.
640  */
641 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
642 {
643         static int (* ioctl)(struct net_device *, struct ifreq *, int);
644         struct ifreq ifr;
645         struct mii_ioctl_data *mii;
646         struct ethtool_value etool;
647
648         if (bond->params.use_carrier) {
649                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
650         }
651
652         ioctl = slave_dev->do_ioctl;
653         if (ioctl) {
654                 /* TODO: set pointer to correct ioctl on a per team member */
655                 /*       bases to make this more efficient. that is, once  */
656                 /*       we determine the correct ioctl, we will always    */
657                 /*       call it and not the others for that team          */
658                 /*       member.                                           */
659
660                 /*
661                  * We cannot assume that SIOCGMIIPHY will also read a
662                  * register; not all network drivers (e.g., e100)
663                  * support that.
664                  */
665
666                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
667                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
668                 mii = if_mii(&ifr);
669                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
670                         mii->reg_num = MII_BMSR;
671                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
672                                 return (mii->val_out & BMSR_LSTATUS);
673                         }
674                 }
675         }
676
677         /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
678         /* for a period of time so we attempt to get link status   */
679         /* from it last if the above MII ioctls fail...            */
680         if (slave_dev->ethtool_ops) {
681                 if (slave_dev->ethtool_ops->get_link) {
682                         u32 link;
683
684                         link = slave_dev->ethtool_ops->get_link(slave_dev);
685
686                         return link ? BMSR_LSTATUS : 0;
687                 }
688         }
689
690         if (ioctl) {
691                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
692                 etool.cmd = ETHTOOL_GLINK;
693                 ifr.ifr_data = (char*)&etool;
694                 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
695                         if (etool.data == 1) {
696                                 return BMSR_LSTATUS;
697                         } else {
698                                 dprintk("SIOCETHTOOL shows link down\n");
699                                 return 0;
700                         }
701                 }
702         }
703
704         /*
705          * If reporting, report that either there's no dev->do_ioctl,
706          * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
707          * cannot report link status).  If not reporting, pretend
708          * we're ok.
709          */
710         return (reporting ? -1 : BMSR_LSTATUS);
711 }
712
713 /*----------------------------- Multicast list ------------------------------*/
714
715 /*
716  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
717  */
718 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
719 {
720         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
721                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
722 }
723
724 /*
725  * returns dmi entry if found, NULL otherwise
726  */
727 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
728 {
729         struct dev_mc_list *idmi;
730
731         for (idmi = mc_list; idmi; idmi = idmi->next) {
732                 if (bond_is_dmi_same(dmi, idmi)) {
733                         return idmi;
734                 }
735         }
736
737         return NULL;
738 }
739
740 /*
741  * Push the promiscuity flag down to appropriate slaves
742  */
743 static void bond_set_promiscuity(struct bonding *bond, int inc)
744 {
745         if (USES_PRIMARY(bond->params.mode)) {
746                 /* write lock already acquired */
747                 if (bond->curr_active_slave) {
748                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
749                 }
750         } else {
751                 struct slave *slave;
752                 int i;
753                 bond_for_each_slave(bond, slave, i) {
754                         dev_set_promiscuity(slave->dev, inc);
755                 }
756         }
757 }
758
759 /*
760  * Push the allmulti flag down to all slaves
761  */
762 static void bond_set_allmulti(struct bonding *bond, int inc)
763 {
764         if (USES_PRIMARY(bond->params.mode)) {
765                 /* write lock already acquired */
766                 if (bond->curr_active_slave) {
767                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
768                 }
769         } else {
770                 struct slave *slave;
771                 int i;
772                 bond_for_each_slave(bond, slave, i) {
773                         dev_set_allmulti(slave->dev, inc);
774                 }
775         }
776 }
777
778 /*
779  * Add a Multicast address to slaves
780  * according to mode
781  */
782 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
783 {
784         if (USES_PRIMARY(bond->params.mode)) {
785                 /* write lock already acquired */
786                 if (bond->curr_active_slave) {
787                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
788                 }
789         } else {
790                 struct slave *slave;
791                 int i;
792                 bond_for_each_slave(bond, slave, i) {
793                         dev_mc_add(slave->dev, addr, alen, 0);
794                 }
795         }
796 }
797
798 /*
799  * Remove a multicast address from slave
800  * according to mode
801  */
802 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
803 {
804         if (USES_PRIMARY(bond->params.mode)) {
805                 /* write lock already acquired */
806                 if (bond->curr_active_slave) {
807                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
808                 }
809         } else {
810                 struct slave *slave;
811                 int i;
812                 bond_for_each_slave(bond, slave, i) {
813                         dev_mc_delete(slave->dev, addr, alen, 0);
814                 }
815         }
816 }
817
818 /*
819  * Totally destroys the mc_list in bond
820  */
821 static void bond_mc_list_destroy(struct bonding *bond)
822 {
823         struct dev_mc_list *dmi;
824
825         dmi = bond->mc_list;
826         while (dmi) {
827                 bond->mc_list = dmi->next;
828                 kfree(dmi);
829                 dmi = bond->mc_list;
830         }
831 }
832
833 /*
834  * Copy all the Multicast addresses from src to the bonding device dst
835  */
836 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
837                              gfp_t gfp_flag)
838 {
839         struct dev_mc_list *dmi, *new_dmi;
840
841         for (dmi = mc_list; dmi; dmi = dmi->next) {
842                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
843
844                 if (!new_dmi) {
845                         /* FIXME: Potential memory leak !!! */
846                         return -ENOMEM;
847                 }
848
849                 new_dmi->next = bond->mc_list;
850                 bond->mc_list = new_dmi;
851                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
852                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
853                 new_dmi->dmi_users = dmi->dmi_users;
854                 new_dmi->dmi_gusers = dmi->dmi_gusers;
855         }
856
857         return 0;
858 }
859
860 /*
861  * flush all members of flush->mc_list from device dev->mc_list
862  */
863 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
864 {
865         struct bonding *bond = bond_dev->priv;
866         struct dev_mc_list *dmi;
867
868         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
869                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
870         }
871
872         if (bond->params.mode == BOND_MODE_8023AD) {
873                 /* del lacpdu mc addr from mc list */
874                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
875
876                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
877         }
878 }
879
880 /*--------------------------- Active slave change ---------------------------*/
881
882 /*
883  * Update the mc list and multicast-related flags for the new and
884  * old active slaves (if any) according to the multicast mode, and
885  * promiscuous flags unconditionally.
886  */
887 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
888 {
889         struct dev_mc_list *dmi;
890
891         if (!USES_PRIMARY(bond->params.mode)) {
892                 /* nothing to do -  mc list is already up-to-date on
893                  * all slaves
894                  */
895                 return;
896         }
897
898         if (old_active) {
899                 if (bond->dev->flags & IFF_PROMISC) {
900                         dev_set_promiscuity(old_active->dev, -1);
901                 }
902
903                 if (bond->dev->flags & IFF_ALLMULTI) {
904                         dev_set_allmulti(old_active->dev, -1);
905                 }
906
907                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
908                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
909                 }
910         }
911
912         if (new_active) {
913                 if (bond->dev->flags & IFF_PROMISC) {
914                         dev_set_promiscuity(new_active->dev, 1);
915                 }
916
917                 if (bond->dev->flags & IFF_ALLMULTI) {
918                         dev_set_allmulti(new_active->dev, 1);
919                 }
920
921                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
922                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
923                 }
924         }
925 }
926
927 /**
928  * find_best_interface - select the best available slave to be the active one
929  * @bond: our bonding struct
930  *
931  * Warning: Caller must hold curr_slave_lock for writing.
932  */
933 static struct slave *bond_find_best_slave(struct bonding *bond)
934 {
935         struct slave *new_active, *old_active;
936         struct slave *bestslave = NULL;
937         int mintime = bond->params.updelay;
938         int i;
939
940         new_active = old_active = bond->curr_active_slave;
941
942         if (!new_active) { /* there were no active slaves left */
943                 if (bond->slave_cnt > 0) {  /* found one slave */
944                         new_active = bond->first_slave;
945                 } else {
946                         return NULL; /* still no slave, return NULL */
947                 }
948         }
949
950         /* first try the primary link; if arping, a link must tx/rx traffic
951          * before it can be considered the curr_active_slave - also, we would skip
952          * slaves between the curr_active_slave and primary_slave that may be up
953          * and able to arp
954          */
955         if ((bond->primary_slave) &&
956             (!bond->params.arp_interval) &&
957             (IS_UP(bond->primary_slave->dev))) {
958                 new_active = bond->primary_slave;
959         }
960
961         /* remember where to stop iterating over the slaves */
962         old_active = new_active;
963
964         bond_for_each_slave_from(bond, new_active, i, old_active) {
965                 if (IS_UP(new_active->dev)) {
966                         if (new_active->link == BOND_LINK_UP) {
967                                 return new_active;
968                         } else if (new_active->link == BOND_LINK_BACK) {
969                                 /* link up, but waiting for stabilization */
970                                 if (new_active->delay < mintime) {
971                                         mintime = new_active->delay;
972                                         bestslave = new_active;
973                                 }
974                         }
975                 }
976         }
977
978         return bestslave;
979 }
980
981 /**
982  * change_active_interface - change the active slave into the specified one
983  * @bond: our bonding struct
984  * @new: the new slave to make the active one
985  *
986  * Set the new slave to the bond's settings and unset them on the old
987  * curr_active_slave.
988  * Setting include flags, mc-list, promiscuity, allmulti, etc.
989  *
990  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
991  * because it is apparently the best available slave we have, even though its
992  * updelay hasn't timed out yet.
993  *
994  * Warning: Caller must hold curr_slave_lock for writing.
995  */
996 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
997 {
998         struct slave *old_active = bond->curr_active_slave;
999
1000         if (old_active == new_active) {
1001                 return;
1002         }
1003
1004         if (new_active) {
1005                 if (new_active->link == BOND_LINK_BACK) {
1006                         if (USES_PRIMARY(bond->params.mode)) {
1007                                 printk(KERN_INFO DRV_NAME
1008                                        ": %s: making interface %s the new "
1009                                        "active one %d ms earlier.\n",
1010                                        bond->dev->name, new_active->dev->name,
1011                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1012                         }
1013
1014                         new_active->delay = 0;
1015                         new_active->link = BOND_LINK_UP;
1016                         new_active->jiffies = jiffies;
1017
1018                         if (bond->params.mode == BOND_MODE_8023AD) {
1019                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1020                         }
1021
1022                         if ((bond->params.mode == BOND_MODE_TLB) ||
1023                             (bond->params.mode == BOND_MODE_ALB)) {
1024                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1025                         }
1026                 } else {
1027                         if (USES_PRIMARY(bond->params.mode)) {
1028                                 printk(KERN_INFO DRV_NAME
1029                                        ": %s: making interface %s the new "
1030                                        "active one.\n",
1031                                        bond->dev->name, new_active->dev->name);
1032                         }
1033                 }
1034         }
1035
1036         if (USES_PRIMARY(bond->params.mode)) {
1037                 bond_mc_swap(bond, new_active, old_active);
1038         }
1039
1040         if ((bond->params.mode == BOND_MODE_TLB) ||
1041             (bond->params.mode == BOND_MODE_ALB)) {
1042                 bond_alb_handle_active_change(bond, new_active);
1043                 if (old_active)
1044                         bond_set_slave_inactive_flags(old_active);
1045                 if (new_active)
1046                         bond_set_slave_active_flags(new_active);
1047         } else {
1048                 bond->curr_active_slave = new_active;
1049         }
1050
1051         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1052                 if (old_active) {
1053                         bond_set_slave_inactive_flags(old_active);
1054                 }
1055
1056                 if (new_active) {
1057                         bond_set_slave_active_flags(new_active);
1058                 }
1059                 bond_send_gratuitous_arp(bond);
1060         }
1061 }
1062
1063 /**
1064  * bond_select_active_slave - select a new active slave, if needed
1065  * @bond: our bonding struct
1066  *
1067  * This functions shoud be called when one of the following occurs:
1068  * - The old curr_active_slave has been released or lost its link.
1069  * - The primary_slave has got its link back.
1070  * - A slave has got its link back and there's no old curr_active_slave.
1071  *
1072  * Warning: Caller must hold curr_slave_lock for writing.
1073  */
1074 void bond_select_active_slave(struct bonding *bond)
1075 {
1076         struct slave *best_slave;
1077
1078         best_slave = bond_find_best_slave(bond);
1079         if (best_slave != bond->curr_active_slave) {
1080                 bond_change_active_slave(bond, best_slave);
1081         }
1082 }
1083
1084 /*--------------------------- slave list handling ---------------------------*/
1085
1086 /*
1087  * This function attaches the slave to the end of list.
1088  *
1089  * bond->lock held for writing by caller.
1090  */
1091 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1092 {
1093         if (bond->first_slave == NULL) { /* attaching the first slave */
1094                 new_slave->next = new_slave;
1095                 new_slave->prev = new_slave;
1096                 bond->first_slave = new_slave;
1097         } else {
1098                 new_slave->next = bond->first_slave;
1099                 new_slave->prev = bond->first_slave->prev;
1100                 new_slave->next->prev = new_slave;
1101                 new_slave->prev->next = new_slave;
1102         }
1103
1104         bond->slave_cnt++;
1105 }
1106
1107 /*
1108  * This function detaches the slave from the list.
1109  * WARNING: no check is made to verify if the slave effectively
1110  * belongs to <bond>.
1111  * Nothing is freed on return, structures are just unchained.
1112  * If any slave pointer in bond was pointing to <slave>,
1113  * it should be changed by the calling function.
1114  *
1115  * bond->lock held for writing by caller.
1116  */
1117 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1118 {
1119         if (slave->next) {
1120                 slave->next->prev = slave->prev;
1121         }
1122
1123         if (slave->prev) {
1124                 slave->prev->next = slave->next;
1125         }
1126
1127         if (bond->first_slave == slave) { /* slave is the first slave */
1128                 if (bond->slave_cnt > 1) { /* there are more slave */
1129                         bond->first_slave = slave->next;
1130                 } else {
1131                         bond->first_slave = NULL; /* slave was the last one */
1132                 }
1133         }
1134
1135         slave->next = NULL;
1136         slave->prev = NULL;
1137         bond->slave_cnt--;
1138 }
1139
1140 /*---------------------------------- IOCTL ----------------------------------*/
1141
1142 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1143 {
1144         dprintk("bond_dev=%p\n", bond_dev);
1145         dprintk("slave_dev=%p\n", slave_dev);
1146         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1147         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1148         return 0;
1149 }
1150
1151 #define BOND_INTERSECT_FEATURES \
1152         (NETIF_F_SG|NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM|\
1153         NETIF_F_TSO|NETIF_F_UFO)
1154
1155 /* 
1156  * Compute the common dev->feature set available to all slaves.  Some
1157  * feature bits are managed elsewhere, so preserve feature bits set on
1158  * master device that are not part of the examined set.
1159  */
1160 static int bond_compute_features(struct bonding *bond)
1161 {
1162         unsigned long features = BOND_INTERSECT_FEATURES;
1163         struct slave *slave;
1164         struct net_device *bond_dev = bond->dev;
1165         int i;
1166
1167         bond_for_each_slave(bond, slave, i)
1168                 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1169
1170         if ((features & NETIF_F_SG) && 
1171             !(features & (NETIF_F_IP_CSUM |
1172                           NETIF_F_NO_CSUM |
1173                           NETIF_F_HW_CSUM)))
1174                 features &= ~NETIF_F_SG;
1175
1176         /* 
1177          * features will include NETIF_F_TSO (NETIF_F_UFO) iff all 
1178          * slave devices support NETIF_F_TSO (NETIF_F_UFO), which 
1179          * implies that all slaves also support scatter-gather 
1180          * (NETIF_F_SG), which implies that features also includes 
1181          * NETIF_F_SG. So no need to check whether we have an  
1182          * illegal combination of NETIF_F_{TSO,UFO} and 
1183          * !NETIF_F_SG 
1184          */
1185
1186         features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1187         bond_dev->features = features;
1188
1189         return 0;
1190 }
1191
1192 /* enslave device <slave> to bond device <master> */
1193 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1194 {
1195         struct bonding *bond = bond_dev->priv;
1196         struct slave *new_slave = NULL;
1197         struct dev_mc_list *dmi;
1198         struct sockaddr addr;
1199         int link_reporting;
1200         int old_features = bond_dev->features;
1201         int res = 0;
1202
1203         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1204                 slave_dev->do_ioctl == NULL) {
1205                 printk(KERN_WARNING DRV_NAME
1206                        ": %s: Warning: no link monitoring support for %s\n",
1207                        bond_dev->name, slave_dev->name);
1208         }
1209
1210         /* bond must be initialized by bond_open() before enslaving */
1211         if (!(bond_dev->flags & IFF_UP)) {
1212                 dprintk("Error, master_dev is not up\n");
1213                 return -EPERM;
1214         }
1215
1216         /* already enslaved */
1217         if (slave_dev->flags & IFF_SLAVE) {
1218                 dprintk("Error, Device was already enslaved\n");
1219                 return -EBUSY;
1220         }
1221
1222         /* vlan challenged mutual exclusion */
1223         /* no need to lock since we're protected by rtnl_lock */
1224         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1225                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1226                 if (!list_empty(&bond->vlan_list)) {
1227                         printk(KERN_ERR DRV_NAME
1228                                ": %s: Error: cannot enslave VLAN "
1229                                "challenged slave %s on VLAN enabled "
1230                                "bond %s\n", bond_dev->name, slave_dev->name,
1231                                bond_dev->name);
1232                         return -EPERM;
1233                 } else {
1234                         printk(KERN_WARNING DRV_NAME
1235                                ": %s: Warning: enslaved VLAN challenged "
1236                                "slave %s. Adding VLANs will be blocked as "
1237                                "long as %s is part of bond %s\n",
1238                                bond_dev->name, slave_dev->name, slave_dev->name,
1239                                bond_dev->name);
1240                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1241                 }
1242         } else {
1243                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1244                 if (bond->slave_cnt == 0) {
1245                         /* First slave, and it is not VLAN challenged,
1246                          * so remove the block of adding VLANs over the bond.
1247                          */
1248                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1249                 }
1250         }
1251
1252         /*
1253          * Old ifenslave binaries are no longer supported.  These can
1254          * be identified with moderate accurary by the state of the slave:
1255          * the current ifenslave will set the interface down prior to
1256          * enslaving it; the old ifenslave will not.
1257          */
1258         if ((slave_dev->flags & IFF_UP)) {
1259                 printk(KERN_ERR DRV_NAME ": %s is up. "
1260                        "This may be due to an out of date ifenslave.\n",
1261                        slave_dev->name);
1262                 res = -EPERM;
1263                 goto err_undo_flags;
1264         }
1265
1266         if (slave_dev->set_mac_address == NULL) {
1267                 printk(KERN_ERR DRV_NAME
1268                         ": %s: Error: The slave device you specified does "
1269                         "not support setting the MAC address. "
1270                         "Your kernel likely does not support slave "
1271                         "devices.\n", bond_dev->name);
1272                 res = -EOPNOTSUPP;
1273                 goto err_undo_flags;
1274         }
1275
1276         new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1277         if (!new_slave) {
1278                 res = -ENOMEM;
1279                 goto err_undo_flags;
1280         }
1281
1282         memset(new_slave, 0, sizeof(struct slave));
1283
1284         /* save slave's original flags before calling
1285          * netdev_set_master and dev_open
1286          */
1287         new_slave->original_flags = slave_dev->flags;
1288
1289         /*
1290          * Save slave's original ("permanent") mac address for modes
1291          * that need it, and for restoring it upon release, and then
1292          * set it to the master's address
1293          */
1294         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1295
1296         /*
1297          * Set slave to master's mac address.  The application already
1298          * set the master's mac address to that of the first slave
1299          */
1300         memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1301         addr.sa_family = slave_dev->type;
1302         res = dev_set_mac_address(slave_dev, &addr);
1303         if (res) {
1304                 dprintk("Error %d calling set_mac_address\n", res);
1305                 goto err_free;
1306         }
1307
1308         /* open the slave since the application closed it */
1309         res = dev_open(slave_dev);
1310         if (res) {
1311                 dprintk("Openning slave %s failed\n", slave_dev->name);
1312                 goto err_restore_mac;
1313         }
1314
1315         res = netdev_set_master(slave_dev, bond_dev);
1316         if (res) {
1317                 dprintk("Error %d calling netdev_set_master\n", res);
1318                 goto err_close;
1319         }
1320
1321         new_slave->dev = slave_dev;
1322
1323         if ((bond->params.mode == BOND_MODE_TLB) ||
1324             (bond->params.mode == BOND_MODE_ALB)) {
1325                 /* bond_alb_init_slave() must be called before all other stages since
1326                  * it might fail and we do not want to have to undo everything
1327                  */
1328                 res = bond_alb_init_slave(bond, new_slave);
1329                 if (res) {
1330                         goto err_unset_master;
1331                 }
1332         }
1333
1334         /* If the mode USES_PRIMARY, then the new slave gets the
1335          * master's promisc (and mc) settings only if it becomes the
1336          * curr_active_slave, and that is taken care of later when calling
1337          * bond_change_active()
1338          */
1339         if (!USES_PRIMARY(bond->params.mode)) {
1340                 /* set promiscuity level to new slave */
1341                 if (bond_dev->flags & IFF_PROMISC) {
1342                         dev_set_promiscuity(slave_dev, 1);
1343                 }
1344
1345                 /* set allmulti level to new slave */
1346                 if (bond_dev->flags & IFF_ALLMULTI) {
1347                         dev_set_allmulti(slave_dev, 1);
1348                 }
1349
1350                 /* upload master's mc_list to new slave */
1351                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1352                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1353                 }
1354         }
1355
1356         if (bond->params.mode == BOND_MODE_8023AD) {
1357                 /* add lacpdu mc addr to mc list */
1358                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1359
1360                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1361         }
1362
1363         bond_add_vlans_on_slave(bond, slave_dev);
1364
1365         write_lock_bh(&bond->lock);
1366
1367         bond_attach_slave(bond, new_slave);
1368
1369         new_slave->delay = 0;
1370         new_slave->link_failure_count = 0;
1371
1372         bond_compute_features(bond);
1373
1374         if (bond->params.miimon && !bond->params.use_carrier) {
1375                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1376
1377                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1378                         /*
1379                          * miimon is set but a bonded network driver
1380                          * does not support ETHTOOL/MII and
1381                          * arp_interval is not set.  Note: if
1382                          * use_carrier is enabled, we will never go
1383                          * here (because netif_carrier is always
1384                          * supported); thus, we don't need to change
1385                          * the messages for netif_carrier.
1386                          */
1387                         printk(KERN_WARNING DRV_NAME
1388                                ": %s: Warning: MII and ETHTOOL support not "
1389                                "available for interface %s, and "
1390                                "arp_interval/arp_ip_target module parameters "
1391                                "not specified, thus bonding will not detect "
1392                                "link failures! see bonding.txt for details.\n",
1393                                bond_dev->name, slave_dev->name);
1394                 } else if (link_reporting == -1) {
1395                         /* unable get link status using mii/ethtool */
1396                         printk(KERN_WARNING DRV_NAME
1397                                ": %s: Warning: can't get link status from "
1398                                "interface %s; the network driver associated "
1399                                "with this interface does not support MII or "
1400                                "ETHTOOL link status reporting, thus miimon "
1401                                "has no effect on this interface.\n",
1402                                bond_dev->name, slave_dev->name);
1403                 }
1404         }
1405
1406         /* check for initial state */
1407         if (!bond->params.miimon ||
1408             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1409                 if (bond->params.updelay) {
1410                         dprintk("Initial state of slave_dev is "
1411                                 "BOND_LINK_BACK\n");
1412                         new_slave->link  = BOND_LINK_BACK;
1413                         new_slave->delay = bond->params.updelay;
1414                 } else {
1415                         dprintk("Initial state of slave_dev is "
1416                                 "BOND_LINK_UP\n");
1417                         new_slave->link  = BOND_LINK_UP;
1418                 }
1419                 new_slave->jiffies = jiffies;
1420         } else {
1421                 dprintk("Initial state of slave_dev is "
1422                         "BOND_LINK_DOWN\n");
1423                 new_slave->link  = BOND_LINK_DOWN;
1424         }
1425
1426         if (bond_update_speed_duplex(new_slave) &&
1427             (new_slave->link != BOND_LINK_DOWN)) {
1428                 printk(KERN_WARNING DRV_NAME
1429                        ": %s: Warning: failed to get speed and duplex from %s, "
1430                        "assumed to be 100Mb/sec and Full.\n",
1431                        bond_dev->name, new_slave->dev->name);
1432
1433                 if (bond->params.mode == BOND_MODE_8023AD) {
1434                         printk(KERN_WARNING DRV_NAME
1435                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1436                                "support in base driver for proper aggregator "
1437                                "selection.\n", bond_dev->name);
1438                 }
1439         }
1440
1441         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1442                 /* if there is a primary slave, remember it */
1443                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1444                         bond->primary_slave = new_slave;
1445                 }
1446         }
1447
1448         switch (bond->params.mode) {
1449         case BOND_MODE_ACTIVEBACKUP:
1450                 /* if we're in active-backup mode, we need one and
1451                  * only one active interface. The backup interfaces
1452                  * will have their SLAVE_INACTIVE flag set because we
1453                  * need them to be drop all packets. Thus, since we
1454                  * guarantee that curr_active_slave always point to
1455                  * the last usable interface, we just have to verify
1456                  * this interface's flag.
1457                  */
1458                 if (((!bond->curr_active_slave) ||
1459                      (bond->curr_active_slave->dev->priv_flags & IFF_SLAVE_INACTIVE)) &&
1460                     (new_slave->link != BOND_LINK_DOWN)) {
1461                         dprintk("This is the first active slave\n");
1462                         /* first slave or no active slave yet, and this link
1463                            is OK, so make this interface the active one */
1464                         bond_change_active_slave(bond, new_slave);
1465                 } else {
1466                         dprintk("This is just a backup slave\n");
1467                         bond_set_slave_inactive_flags(new_slave);
1468                 }
1469                 break;
1470         case BOND_MODE_8023AD:
1471                 /* in 802.3ad mode, the internal mechanism
1472                  * will activate the slaves in the selected
1473                  * aggregator
1474                  */
1475                 bond_set_slave_inactive_flags(new_slave);
1476                 /* if this is the first slave */
1477                 if (bond->slave_cnt == 1) {
1478                         SLAVE_AD_INFO(new_slave).id = 1;
1479                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1480                          * can be called only after the mac address of the bond is set
1481                          */
1482                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1483                                             bond->params.lacp_fast);
1484                 } else {
1485                         SLAVE_AD_INFO(new_slave).id =
1486                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1487                 }
1488
1489                 bond_3ad_bind_slave(new_slave);
1490                 break;
1491         case BOND_MODE_TLB:
1492         case BOND_MODE_ALB:
1493                 new_slave->state = BOND_STATE_ACTIVE;
1494                 if ((!bond->curr_active_slave) &&
1495                     (new_slave->link != BOND_LINK_DOWN)) {
1496                         /* first slave or no active slave yet, and this link
1497                          * is OK, so make this interface the active one
1498                          */
1499                         bond_change_active_slave(bond, new_slave);
1500                 } else {
1501                         bond_set_slave_inactive_flags(new_slave);
1502                 }
1503                 break;
1504         default:
1505                 dprintk("This slave is always active in trunk mode\n");
1506
1507                 /* always active in trunk mode */
1508                 new_slave->state = BOND_STATE_ACTIVE;
1509
1510                 /* In trunking mode there is little meaning to curr_active_slave
1511                  * anyway (it holds no special properties of the bond device),
1512                  * so we can change it without calling change_active_interface()
1513                  */
1514                 if (!bond->curr_active_slave) {
1515                         bond->curr_active_slave = new_slave;
1516                 }
1517                 break;
1518         } /* switch(bond_mode) */
1519
1520         write_unlock_bh(&bond->lock);
1521
1522         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1523         if (res)
1524                 goto err_unset_master;
1525
1526         printk(KERN_INFO DRV_NAME
1527                ": %s: enslaving %s as a%s interface with a%s link.\n",
1528                bond_dev->name, slave_dev->name,
1529                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1530                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1531
1532         /* enslave is successful */
1533         return 0;
1534
1535 /* Undo stages on error */
1536 err_unset_master:
1537         netdev_set_master(slave_dev, NULL);
1538
1539 err_close:
1540         dev_close(slave_dev);
1541
1542 err_restore_mac:
1543         memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1544         addr.sa_family = slave_dev->type;
1545         dev_set_mac_address(slave_dev, &addr);
1546
1547 err_free:
1548         kfree(new_slave);
1549
1550 err_undo_flags:
1551         bond_dev->features = old_features;
1552  
1553         return res;
1554 }
1555
1556 /*
1557  * Try to release the slave device <slave> from the bond device <master>
1558  * It is legal to access curr_active_slave without a lock because all the function
1559  * is write-locked.
1560  *
1561  * The rules for slave state should be:
1562  *   for Active/Backup:
1563  *     Active stays on all backups go down
1564  *   for Bonded connections:
1565  *     The first up interface should be left on and all others downed.
1566  */
1567 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1568 {
1569         struct bonding *bond = bond_dev->priv;
1570         struct slave *slave, *oldcurrent;
1571         struct sockaddr addr;
1572         int mac_addr_differ;
1573
1574         /* slave is not a slave or master is not master of this slave */
1575         if (!(slave_dev->flags & IFF_SLAVE) ||
1576             (slave_dev->master != bond_dev)) {
1577                 printk(KERN_ERR DRV_NAME
1578                        ": %s: Error: cannot release %s.\n",
1579                        bond_dev->name, slave_dev->name);
1580                 return -EINVAL;
1581         }
1582
1583         write_lock_bh(&bond->lock);
1584
1585         slave = bond_get_slave_by_dev(bond, slave_dev);
1586         if (!slave) {
1587                 /* not a slave of this bond */
1588                 printk(KERN_INFO DRV_NAME
1589                        ": %s: %s not enslaved\n",
1590                        bond_dev->name, slave_dev->name);
1591                 write_unlock_bh(&bond->lock);
1592                 return -EINVAL;
1593         }
1594
1595         mac_addr_differ = memcmp(bond_dev->dev_addr,
1596                                  slave->perm_hwaddr,
1597                                  ETH_ALEN);
1598         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1599                 printk(KERN_WARNING DRV_NAME
1600                        ": %s: Warning: the permanent HWaddr of %s "
1601                        "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1602                        "still in use by %s. Set the HWaddr of "
1603                        "%s to a different address to avoid "
1604                        "conflicts.\n",
1605                        bond_dev->name,
1606                        slave_dev->name,
1607                        slave->perm_hwaddr[0],
1608                        slave->perm_hwaddr[1],
1609                        slave->perm_hwaddr[2],
1610                        slave->perm_hwaddr[3],
1611                        slave->perm_hwaddr[4],
1612                        slave->perm_hwaddr[5],
1613                        bond_dev->name,
1614                        slave_dev->name);
1615         }
1616
1617         /* Inform AD package of unbinding of slave. */
1618         if (bond->params.mode == BOND_MODE_8023AD) {
1619                 /* must be called before the slave is
1620                  * detached from the list
1621                  */
1622                 bond_3ad_unbind_slave(slave);
1623         }
1624
1625         printk(KERN_INFO DRV_NAME
1626                ": %s: releasing %s interface %s\n",
1627                bond_dev->name,
1628                (slave->state == BOND_STATE_ACTIVE)
1629                ? "active" : "backup",
1630                slave_dev->name);
1631
1632         oldcurrent = bond->curr_active_slave;
1633
1634         bond->current_arp_slave = NULL;
1635
1636         /* release the slave from its bond */
1637         bond_detach_slave(bond, slave);
1638
1639         bond_compute_features(bond);
1640
1641         if (bond->primary_slave == slave) {
1642                 bond->primary_slave = NULL;
1643         }
1644
1645         if (oldcurrent == slave) {
1646                 bond_change_active_slave(bond, NULL);
1647         }
1648
1649         if ((bond->params.mode == BOND_MODE_TLB) ||
1650             (bond->params.mode == BOND_MODE_ALB)) {
1651                 /* Must be called only after the slave has been
1652                  * detached from the list and the curr_active_slave
1653                  * has been cleared (if our_slave == old_current),
1654                  * but before a new active slave is selected.
1655                  */
1656                 bond_alb_deinit_slave(bond, slave);
1657         }
1658
1659         if (oldcurrent == slave) {
1660                 bond_select_active_slave(bond);
1661
1662                 if (!bond->curr_active_slave) {
1663                         printk(KERN_INFO DRV_NAME
1664                                ": %s: now running without any active "
1665                                "interface !\n",
1666                                bond_dev->name);
1667                 }
1668         }
1669
1670         if (bond->slave_cnt == 0) {
1671                 /* if the last slave was removed, zero the mac address
1672                  * of the master so it will be set by the application
1673                  * to the mac address of the first slave
1674                  */
1675                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1676
1677                 if (list_empty(&bond->vlan_list)) {
1678                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1679                 } else {
1680                         printk(KERN_WARNING DRV_NAME
1681                                ": %s: Warning: clearing HW address of %s while it "
1682                                "still has VLANs.\n",
1683                                bond_dev->name, bond_dev->name);
1684                         printk(KERN_WARNING DRV_NAME
1685                                ": %s: When re-adding slaves, make sure the bond's "
1686                                "HW address matches its VLANs'.\n",
1687                                bond_dev->name);
1688                 }
1689         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1690                    !bond_has_challenged_slaves(bond)) {
1691                 printk(KERN_INFO DRV_NAME
1692                        ": %s: last VLAN challenged slave %s "
1693                        "left bond %s. VLAN blocking is removed\n",
1694                        bond_dev->name, slave_dev->name, bond_dev->name);
1695                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1696         }
1697
1698         write_unlock_bh(&bond->lock);
1699
1700         /* must do this from outside any spinlocks */
1701         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1702
1703         bond_del_vlans_from_slave(bond, slave_dev);
1704
1705         /* If the mode USES_PRIMARY, then we should only remove its
1706          * promisc and mc settings if it was the curr_active_slave, but that was
1707          * already taken care of above when we detached the slave
1708          */
1709         if (!USES_PRIMARY(bond->params.mode)) {
1710                 /* unset promiscuity level from slave */
1711                 if (bond_dev->flags & IFF_PROMISC) {
1712                         dev_set_promiscuity(slave_dev, -1);
1713                 }
1714
1715                 /* unset allmulti level from slave */
1716                 if (bond_dev->flags & IFF_ALLMULTI) {
1717                         dev_set_allmulti(slave_dev, -1);
1718                 }
1719
1720                 /* flush master's mc_list from slave */
1721                 bond_mc_list_flush(bond_dev, slave_dev);
1722         }
1723
1724         netdev_set_master(slave_dev, NULL);
1725
1726         /* close slave before restoring its mac address */
1727         dev_close(slave_dev);
1728
1729         /* restore original ("permanent") mac address */
1730         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1731         addr.sa_family = slave_dev->type;
1732         dev_set_mac_address(slave_dev, &addr);
1733
1734         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1735                                    IFF_SLAVE_INACTIVE);
1736
1737         kfree(slave);
1738
1739         return 0;  /* deletion OK */
1740 }
1741
1742 /*
1743  * This function releases all slaves.
1744  */
1745 static int bond_release_all(struct net_device *bond_dev)
1746 {
1747         struct bonding *bond = bond_dev->priv;
1748         struct slave *slave;
1749         struct net_device *slave_dev;
1750         struct sockaddr addr;
1751
1752         write_lock_bh(&bond->lock);
1753
1754         if (bond->slave_cnt == 0) {
1755                 goto out;
1756         }
1757
1758         bond->current_arp_slave = NULL;
1759         bond->primary_slave = NULL;
1760         bond_change_active_slave(bond, NULL);
1761
1762         while ((slave = bond->first_slave) != NULL) {
1763                 /* Inform AD package of unbinding of slave
1764                  * before slave is detached from the list.
1765                  */
1766                 if (bond->params.mode == BOND_MODE_8023AD) {
1767                         bond_3ad_unbind_slave(slave);
1768                 }
1769
1770                 slave_dev = slave->dev;
1771                 bond_detach_slave(bond, slave);
1772
1773                 if ((bond->params.mode == BOND_MODE_TLB) ||
1774                     (bond->params.mode == BOND_MODE_ALB)) {
1775                         /* must be called only after the slave
1776                          * has been detached from the list
1777                          */
1778                         bond_alb_deinit_slave(bond, slave);
1779                 }
1780
1781                 bond_compute_features(bond);
1782
1783                 /* now that the slave is detached, unlock and perform
1784                  * all the undo steps that should not be called from
1785                  * within a lock.
1786                  */
1787                 write_unlock_bh(&bond->lock);
1788
1789                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1790                 bond_del_vlans_from_slave(bond, slave_dev);
1791
1792                 /* If the mode USES_PRIMARY, then we should only remove its
1793                  * promisc and mc settings if it was the curr_active_slave, but that was
1794                  * already taken care of above when we detached the slave
1795                  */
1796                 if (!USES_PRIMARY(bond->params.mode)) {
1797                         /* unset promiscuity level from slave */
1798                         if (bond_dev->flags & IFF_PROMISC) {
1799                                 dev_set_promiscuity(slave_dev, -1);
1800                         }
1801
1802                         /* unset allmulti level from slave */
1803                         if (bond_dev->flags & IFF_ALLMULTI) {
1804                                 dev_set_allmulti(slave_dev, -1);
1805                         }
1806
1807                         /* flush master's mc_list from slave */
1808                         bond_mc_list_flush(bond_dev, slave_dev);
1809                 }
1810
1811                 netdev_set_master(slave_dev, NULL);
1812
1813                 /* close slave before restoring its mac address */
1814                 dev_close(slave_dev);
1815
1816                 /* restore original ("permanent") mac address*/
1817                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1818                 addr.sa_family = slave_dev->type;
1819                 dev_set_mac_address(slave_dev, &addr);
1820
1821                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1822                                            IFF_SLAVE_INACTIVE);
1823
1824                 kfree(slave);
1825
1826                 /* re-acquire the lock before getting the next slave */
1827                 write_lock_bh(&bond->lock);
1828         }
1829
1830         /* zero the mac address of the master so it will be
1831          * set by the application to the mac address of the
1832          * first slave
1833          */
1834         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1835
1836         if (list_empty(&bond->vlan_list)) {
1837                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1838         } else {
1839                 printk(KERN_WARNING DRV_NAME
1840                        ": %s: Warning: clearing HW address of %s while it "
1841                        "still has VLANs.\n",
1842                        bond_dev->name, bond_dev->name);
1843                 printk(KERN_WARNING DRV_NAME
1844                        ": %s: When re-adding slaves, make sure the bond's "
1845                        "HW address matches its VLANs'.\n",
1846                        bond_dev->name);
1847         }
1848
1849         printk(KERN_INFO DRV_NAME
1850                ": %s: released all slaves\n",
1851                bond_dev->name);
1852
1853 out:
1854         write_unlock_bh(&bond->lock);
1855
1856         return 0;
1857 }
1858
1859 /*
1860  * This function changes the active slave to slave <slave_dev>.
1861  * It returns -EINVAL in the following cases.
1862  *  - <slave_dev> is not found in the list.
1863  *  - There is not active slave now.
1864  *  - <slave_dev> is already active.
1865  *  - The link state of <slave_dev> is not BOND_LINK_UP.
1866  *  - <slave_dev> is not running.
1867  * In these cases, this fuction does nothing.
1868  * In the other cases, currnt_slave pointer is changed and 0 is returned.
1869  */
1870 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1871 {
1872         struct bonding *bond = bond_dev->priv;
1873         struct slave *old_active = NULL;
1874         struct slave *new_active = NULL;
1875         int res = 0;
1876
1877         if (!USES_PRIMARY(bond->params.mode)) {
1878                 return -EINVAL;
1879         }
1880
1881         /* Verify that master_dev is indeed the master of slave_dev */
1882         if (!(slave_dev->flags & IFF_SLAVE) ||
1883             (slave_dev->master != bond_dev)) {
1884                 return -EINVAL;
1885         }
1886
1887         write_lock_bh(&bond->lock);
1888
1889         old_active = bond->curr_active_slave;
1890         new_active = bond_get_slave_by_dev(bond, slave_dev);
1891
1892         /*
1893          * Changing to the current active: do nothing; return success.
1894          */
1895         if (new_active && (new_active == old_active)) {
1896                 write_unlock_bh(&bond->lock);
1897                 return 0;
1898         }
1899
1900         if ((new_active) &&
1901             (old_active) &&
1902             (new_active->link == BOND_LINK_UP) &&
1903             IS_UP(new_active->dev)) {
1904                 bond_change_active_slave(bond, new_active);
1905         } else {
1906                 res = -EINVAL;
1907         }
1908
1909         write_unlock_bh(&bond->lock);
1910
1911         return res;
1912 }
1913
1914 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1915 {
1916         struct bonding *bond = bond_dev->priv;
1917
1918         info->bond_mode = bond->params.mode;
1919         info->miimon = bond->params.miimon;
1920
1921         read_lock_bh(&bond->lock);
1922         info->num_slaves = bond->slave_cnt;
1923         read_unlock_bh(&bond->lock);
1924
1925         return 0;
1926 }
1927
1928 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1929 {
1930         struct bonding *bond = bond_dev->priv;
1931         struct slave *slave;
1932         int i, found = 0;
1933
1934         if (info->slave_id < 0) {
1935                 return -ENODEV;
1936         }
1937
1938         read_lock_bh(&bond->lock);
1939
1940         bond_for_each_slave(bond, slave, i) {
1941                 if (i == (int)info->slave_id) {
1942                         found = 1;
1943                         break;
1944                 }
1945         }
1946
1947         read_unlock_bh(&bond->lock);
1948
1949         if (found) {
1950                 strcpy(info->slave_name, slave->dev->name);
1951                 info->link = slave->link;
1952                 info->state = slave->state;
1953                 info->link_failure_count = slave->link_failure_count;
1954         } else {
1955                 return -ENODEV;
1956         }
1957
1958         return 0;
1959 }
1960
1961 /*-------------------------------- Monitoring -------------------------------*/
1962
1963 /* this function is called regularly to monitor each slave's link. */
1964 void bond_mii_monitor(struct net_device *bond_dev)
1965 {
1966         struct bonding *bond = bond_dev->priv;
1967         struct slave *slave, *oldcurrent;
1968         int do_failover = 0;
1969         int delta_in_ticks;
1970         int i;
1971
1972         read_lock(&bond->lock);
1973
1974         delta_in_ticks = (bond->params.miimon * HZ) / 1000;
1975
1976         if (bond->kill_timers) {
1977                 goto out;
1978         }
1979
1980         if (bond->slave_cnt == 0) {
1981                 goto re_arm;
1982         }
1983
1984         /* we will try to read the link status of each of our slaves, and
1985          * set their IFF_RUNNING flag appropriately. For each slave not
1986          * supporting MII status, we won't do anything so that a user-space
1987          * program could monitor the link itself if needed.
1988          */
1989
1990         read_lock(&bond->curr_slave_lock);
1991         oldcurrent = bond->curr_active_slave;
1992         read_unlock(&bond->curr_slave_lock);
1993
1994         bond_for_each_slave(bond, slave, i) {
1995                 struct net_device *slave_dev = slave->dev;
1996                 int link_state;
1997                 u16 old_speed = slave->speed;
1998                 u8 old_duplex = slave->duplex;
1999
2000                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2001
2002                 switch (slave->link) {
2003                 case BOND_LINK_UP:      /* the link was up */
2004                         if (link_state == BMSR_LSTATUS) {
2005                                 /* link stays up, nothing more to do */
2006                                 break;
2007                         } else { /* link going down */
2008                                 slave->link  = BOND_LINK_FAIL;
2009                                 slave->delay = bond->params.downdelay;
2010
2011                                 if (slave->link_failure_count < UINT_MAX) {
2012                                         slave->link_failure_count++;
2013                                 }
2014
2015                                 if (bond->params.downdelay) {
2016                                         printk(KERN_INFO DRV_NAME
2017                                                ": %s: link status down for %s "
2018                                                "interface %s, disabling it in "
2019                                                "%d ms.\n",
2020                                                bond_dev->name,
2021                                                IS_UP(slave_dev)
2022                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2023                                                   ? ((slave == oldcurrent)
2024                                                      ? "active " : "backup ")
2025                                                   : "")
2026                                                : "idle ",
2027                                                slave_dev->name,
2028                                                bond->params.downdelay * bond->params.miimon);
2029                                 }
2030                         }
2031                         /* no break ! fall through the BOND_LINK_FAIL test to
2032                            ensure proper action to be taken
2033                         */
2034                 case BOND_LINK_FAIL:    /* the link has just gone down */
2035                         if (link_state != BMSR_LSTATUS) {
2036                                 /* link stays down */
2037                                 if (slave->delay <= 0) {
2038                                         /* link down for too long time */
2039                                         slave->link = BOND_LINK_DOWN;
2040
2041                                         /* in active/backup mode, we must
2042                                          * completely disable this interface
2043                                          */
2044                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2045                                             (bond->params.mode == BOND_MODE_8023AD)) {
2046                                                 bond_set_slave_inactive_flags(slave);
2047                                         }
2048
2049                                         printk(KERN_INFO DRV_NAME
2050                                                ": %s: link status definitely "
2051                                                "down for interface %s, "
2052                                                "disabling it\n",
2053                                                bond_dev->name,
2054                                                slave_dev->name);
2055
2056                                         /* notify ad that the link status has changed */
2057                                         if (bond->params.mode == BOND_MODE_8023AD) {
2058                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2059                                         }
2060
2061                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2062                                             (bond->params.mode == BOND_MODE_ALB)) {
2063                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2064                                         }
2065
2066                                         if (slave == oldcurrent) {
2067                                                 do_failover = 1;
2068                                         }
2069                                 } else {
2070                                         slave->delay--;
2071                                 }
2072                         } else {
2073                                 /* link up again */
2074                                 slave->link  = BOND_LINK_UP;
2075                                 slave->jiffies = jiffies;
2076                                 printk(KERN_INFO DRV_NAME
2077                                        ": %s: link status up again after %d "
2078                                        "ms for interface %s.\n",
2079                                        bond_dev->name,
2080                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2081                                        slave_dev->name);
2082                         }
2083                         break;
2084                 case BOND_LINK_DOWN:    /* the link was down */
2085                         if (link_state != BMSR_LSTATUS) {
2086                                 /* the link stays down, nothing more to do */
2087                                 break;
2088                         } else {        /* link going up */
2089                                 slave->link  = BOND_LINK_BACK;
2090                                 slave->delay = bond->params.updelay;
2091
2092                                 if (bond->params.updelay) {
2093                                         /* if updelay == 0, no need to
2094                                            advertise about a 0 ms delay */
2095                                         printk(KERN_INFO DRV_NAME
2096                                                ": %s: link status up for "
2097                                                "interface %s, enabling it "
2098                                                "in %d ms.\n",
2099                                                bond_dev->name,
2100                                                slave_dev->name,
2101                                                bond->params.updelay * bond->params.miimon);
2102                                 }
2103                         }
2104                         /* no break ! fall through the BOND_LINK_BACK state in
2105                            case there's something to do.
2106                         */
2107                 case BOND_LINK_BACK:    /* the link has just come back */
2108                         if (link_state != BMSR_LSTATUS) {
2109                                 /* link down again */
2110                                 slave->link  = BOND_LINK_DOWN;
2111
2112                                 printk(KERN_INFO DRV_NAME
2113                                        ": %s: link status down again after %d "
2114                                        "ms for interface %s.\n",
2115                                        bond_dev->name,
2116                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2117                                        slave_dev->name);
2118                         } else {
2119                                 /* link stays up */
2120                                 if (slave->delay == 0) {
2121                                         /* now the link has been up for long time enough */
2122                                         slave->link = BOND_LINK_UP;
2123                                         slave->jiffies = jiffies;
2124
2125                                         if (bond->params.mode == BOND_MODE_8023AD) {
2126                                                 /* prevent it from being the active one */
2127                                                 slave->state = BOND_STATE_BACKUP;
2128                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2129                                                 /* make it immediately active */
2130                                                 slave->state = BOND_STATE_ACTIVE;
2131                                         } else if (slave != bond->primary_slave) {
2132                                                 /* prevent it from being the active one */
2133                                                 slave->state = BOND_STATE_BACKUP;
2134                                         }
2135
2136                                         printk(KERN_INFO DRV_NAME
2137                                                ": %s: link status definitely "
2138                                                "up for interface %s.\n",
2139                                                bond_dev->name,
2140                                                slave_dev->name);
2141
2142                                         /* notify ad that the link status has changed */
2143                                         if (bond->params.mode == BOND_MODE_8023AD) {
2144                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2145                                         }
2146
2147                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2148                                             (bond->params.mode == BOND_MODE_ALB)) {
2149                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2150                                         }
2151
2152                                         if ((!oldcurrent) ||
2153                                             (slave == bond->primary_slave)) {
2154                                                 do_failover = 1;
2155                                         }
2156                                 } else {
2157                                         slave->delay--;
2158                                 }
2159                         }
2160                         break;
2161                 default:
2162                         /* Should not happen */
2163                         printk(KERN_ERR DRV_NAME
2164                                ": %s: Error: %s Illegal value (link=%d)\n",
2165                                bond_dev->name,
2166                                slave->dev->name,
2167                                slave->link);
2168                         goto out;
2169                 } /* end of switch (slave->link) */
2170
2171                 bond_update_speed_duplex(slave);
2172
2173                 if (bond->params.mode == BOND_MODE_8023AD) {
2174                         if (old_speed != slave->speed) {
2175                                 bond_3ad_adapter_speed_changed(slave);
2176                         }
2177
2178                         if (old_duplex != slave->duplex) {
2179                                 bond_3ad_adapter_duplex_changed(slave);
2180                         }
2181                 }
2182
2183         } /* end of for */
2184
2185         if (do_failover) {
2186                 write_lock(&bond->curr_slave_lock);
2187
2188                 bond_select_active_slave(bond);
2189
2190                 if (oldcurrent && !bond->curr_active_slave) {
2191                         printk(KERN_INFO DRV_NAME
2192                                ": %s: now running without any active "
2193                                "interface !\n",
2194                                bond_dev->name);
2195                 }
2196
2197                 write_unlock(&bond->curr_slave_lock);
2198         }
2199
2200 re_arm:
2201         if (bond->params.miimon) {
2202                 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2203         }
2204 out:
2205         read_unlock(&bond->lock);
2206 }
2207
2208
2209 static u32 bond_glean_dev_ip(struct net_device *dev)
2210 {
2211         struct in_device *idev;
2212         struct in_ifaddr *ifa;
2213         u32 addr = 0;
2214
2215         if (!dev)
2216                 return 0;
2217
2218         rcu_read_lock();
2219         idev = __in_dev_get_rcu(dev);
2220         if (!idev)
2221                 goto out;
2222
2223         ifa = idev->ifa_list;
2224         if (!ifa)
2225                 goto out;
2226
2227         addr = ifa->ifa_local;
2228 out:
2229         rcu_read_unlock();
2230         return addr;
2231 }
2232
2233 static int bond_has_ip(struct bonding *bond)
2234 {
2235         struct vlan_entry *vlan, *vlan_next;
2236
2237         if (bond->master_ip)
2238                 return 1;
2239
2240         if (list_empty(&bond->vlan_list))
2241                 return 0;
2242
2243         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2244                                  vlan_list) {
2245                 if (vlan->vlan_ip)
2246                         return 1;
2247         }
2248
2249         return 0;
2250 }
2251
2252 /*
2253  * We go to the (large) trouble of VLAN tagging ARP frames because
2254  * switches in VLAN mode (especially if ports are configured as
2255  * "native" to a VLAN) might not pass non-tagged frames.
2256  */
2257 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2258 {
2259         struct sk_buff *skb;
2260
2261         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2262                slave_dev->name, dest_ip, src_ip, vlan_id);
2263                
2264         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2265                          NULL, slave_dev->dev_addr, NULL);
2266
2267         if (!skb) {
2268                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2269                 return;
2270         }
2271         if (vlan_id) {
2272                 skb = vlan_put_tag(skb, vlan_id);
2273                 if (!skb) {
2274                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2275                         return;
2276                 }
2277         }
2278         arp_xmit(skb);
2279 }
2280
2281
2282 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2283 {
2284         int i, vlan_id, rv;
2285         u32 *targets = bond->params.arp_targets;
2286         struct vlan_entry *vlan, *vlan_next;
2287         struct net_device *vlan_dev;
2288         struct flowi fl;
2289         struct rtable *rt;
2290
2291         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2292                 if (!targets[i])
2293                         continue;
2294                 dprintk("basa: target %x\n", targets[i]);
2295                 if (list_empty(&bond->vlan_list)) {
2296                         dprintk("basa: empty vlan: arp_send\n");
2297                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2298                                       bond->master_ip, 0);
2299                         continue;
2300                 }
2301
2302                 /*
2303                  * If VLANs are configured, we do a route lookup to
2304                  * determine which VLAN interface would be used, so we
2305                  * can tag the ARP with the proper VLAN tag.
2306                  */
2307                 memset(&fl, 0, sizeof(fl));
2308                 fl.fl4_dst = targets[i];
2309                 fl.fl4_tos = RTO_ONLINK;
2310
2311                 rv = ip_route_output_key(&rt, &fl);
2312                 if (rv) {
2313                         if (net_ratelimit()) {
2314                                 printk(KERN_WARNING DRV_NAME
2315                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2316                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2317                         }
2318                         continue;
2319                 }
2320
2321                 /*
2322                  * This target is not on a VLAN
2323                  */
2324                 if (rt->u.dst.dev == bond->dev) {
2325                         ip_rt_put(rt);
2326                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2327                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2328                                       bond->master_ip, 0);
2329                         continue;
2330                 }
2331
2332                 vlan_id = 0;
2333                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2334                                          vlan_list) {
2335                         vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2336                         if (vlan_dev == rt->u.dst.dev) {
2337                                 vlan_id = vlan->vlan_id;
2338                                 dprintk("basa: vlan match on %s %d\n",
2339                                        vlan_dev->name, vlan_id);
2340                                 break;
2341                         }
2342                 }
2343
2344                 if (vlan_id) {
2345                         ip_rt_put(rt);
2346                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2347                                       vlan->vlan_ip, vlan_id);
2348                         continue;
2349                 }
2350
2351                 if (net_ratelimit()) {
2352                         printk(KERN_WARNING DRV_NAME
2353                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2354                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2355                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2356                 }
2357                 ip_rt_put(rt);
2358         }
2359 }
2360
2361 /*
2362  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2363  * for each VLAN above us.
2364  */
2365 static void bond_send_gratuitous_arp(struct bonding *bond)
2366 {
2367         struct slave *slave = bond->curr_active_slave;
2368         struct vlan_entry *vlan;
2369         struct net_device *vlan_dev;
2370
2371         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2372                                 slave ? slave->dev->name : "NULL");
2373         if (!slave)
2374                 return;
2375
2376         if (bond->master_ip) {
2377                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2378                                   bond->master_ip, 0);
2379         }
2380
2381         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2382                 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2383                 if (vlan->vlan_ip) {
2384                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2385                                       vlan->vlan_ip, vlan->vlan_id);
2386                 }
2387         }
2388 }
2389
2390 /*
2391  * this function is called regularly to monitor each slave's link
2392  * ensuring that traffic is being sent and received when arp monitoring
2393  * is used in load-balancing mode. if the adapter has been dormant, then an
2394  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2395  * arp monitoring in active backup mode.
2396  */
2397 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2398 {
2399         struct bonding *bond = bond_dev->priv;
2400         struct slave *slave, *oldcurrent;
2401         int do_failover = 0;
2402         int delta_in_ticks;
2403         int i;
2404
2405         read_lock(&bond->lock);
2406
2407         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2408
2409         if (bond->kill_timers) {
2410                 goto out;
2411         }
2412
2413         if (bond->slave_cnt == 0) {
2414                 goto re_arm;
2415         }
2416
2417         read_lock(&bond->curr_slave_lock);
2418         oldcurrent = bond->curr_active_slave;
2419         read_unlock(&bond->curr_slave_lock);
2420
2421         /* see if any of the previous devices are up now (i.e. they have
2422          * xmt and rcv traffic). the curr_active_slave does not come into
2423          * the picture unless it is null. also, slave->jiffies is not needed
2424          * here because we send an arp on each slave and give a slave as
2425          * long as it needs to get the tx/rx within the delta.
2426          * TODO: what about up/down delay in arp mode? it wasn't here before
2427          *       so it can wait
2428          */
2429         bond_for_each_slave(bond, slave, i) {
2430                 if (slave->link != BOND_LINK_UP) {
2431                         if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2432                             ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2433
2434                                 slave->link  = BOND_LINK_UP;
2435                                 slave->state = BOND_STATE_ACTIVE;
2436
2437                                 /* primary_slave has no meaning in round-robin
2438                                  * mode. the window of a slave being up and
2439                                  * curr_active_slave being null after enslaving
2440                                  * is closed.
2441                                  */
2442                                 if (!oldcurrent) {
2443                                         printk(KERN_INFO DRV_NAME
2444                                                ": %s: link status definitely "
2445                                                "up for interface %s, ",
2446                                                bond_dev->name,
2447                                                slave->dev->name);
2448                                         do_failover = 1;
2449                                 } else {
2450                                         printk(KERN_INFO DRV_NAME
2451                                                ": %s: interface %s is now up\n",
2452                                                bond_dev->name,
2453                                                slave->dev->name);
2454                                 }
2455                         }
2456                 } else {
2457                         /* slave->link == BOND_LINK_UP */
2458
2459                         /* not all switches will respond to an arp request
2460                          * when the source ip is 0, so don't take the link down
2461                          * if we don't know our ip yet
2462                          */
2463                         if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2464                             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2465                              bond_has_ip(bond))) {
2466
2467                                 slave->link  = BOND_LINK_DOWN;
2468                                 slave->state = BOND_STATE_BACKUP;
2469
2470                                 if (slave->link_failure_count < UINT_MAX) {
2471                                         slave->link_failure_count++;
2472                                 }
2473
2474                                 printk(KERN_INFO DRV_NAME
2475                                        ": %s: interface %s is now down.\n",
2476                                        bond_dev->name,
2477                                        slave->dev->name);
2478
2479                                 if (slave == oldcurrent) {
2480                                         do_failover = 1;
2481                                 }
2482                         }
2483                 }
2484
2485                 /* note: if switch is in round-robin mode, all links
2486                  * must tx arp to ensure all links rx an arp - otherwise
2487                  * links may oscillate or not come up at all; if switch is
2488                  * in something like xor mode, there is nothing we can
2489                  * do - all replies will be rx'ed on same link causing slaves
2490                  * to be unstable during low/no traffic periods
2491                  */
2492                 if (IS_UP(slave->dev)) {
2493                         bond_arp_send_all(bond, slave);
2494                 }
2495         }
2496
2497         if (do_failover) {
2498                 write_lock(&bond->curr_slave_lock);
2499
2500                 bond_select_active_slave(bond);
2501
2502                 if (oldcurrent && !bond->curr_active_slave) {
2503                         printk(KERN_INFO DRV_NAME
2504                                ": %s: now running without any active "
2505                                "interface !\n",
2506                                bond_dev->name);
2507                 }
2508
2509                 write_unlock(&bond->curr_slave_lock);
2510         }
2511
2512 re_arm:
2513         if (bond->params.arp_interval) {
2514                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2515         }
2516 out:
2517         read_unlock(&bond->lock);
2518 }
2519
2520 /*
2521  * When using arp monitoring in active-backup mode, this function is
2522  * called to determine if any backup slaves have went down or a new
2523  * current slave needs to be found.
2524  * The backup slaves never generate traffic, they are considered up by merely
2525  * receiving traffic. If the current slave goes down, each backup slave will
2526  * be given the opportunity to tx/rx an arp before being taken down - this
2527  * prevents all slaves from being taken down due to the current slave not
2528  * sending any traffic for the backups to receive. The arps are not necessarily
2529  * necessary, any tx and rx traffic will keep the current slave up. While any
2530  * rx traffic will keep the backup slaves up, the current slave is responsible
2531  * for generating traffic to keep them up regardless of any other traffic they
2532  * may have received.
2533  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2534  */
2535 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2536 {
2537         struct bonding *bond = bond_dev->priv;
2538         struct slave *slave;
2539         int delta_in_ticks;
2540         int i;
2541
2542         read_lock(&bond->lock);
2543
2544         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2545
2546         if (bond->kill_timers) {
2547                 goto out;
2548         }
2549
2550         if (bond->slave_cnt == 0) {
2551                 goto re_arm;
2552         }
2553
2554         /* determine if any slave has come up or any backup slave has
2555          * gone down
2556          * TODO: what about up/down delay in arp mode? it wasn't here before
2557          *       so it can wait
2558          */
2559         bond_for_each_slave(bond, slave, i) {
2560                 if (slave->link != BOND_LINK_UP) {
2561                         if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
2562
2563                                 slave->link = BOND_LINK_UP;
2564
2565                                 write_lock(&bond->curr_slave_lock);
2566
2567                                 if ((!bond->curr_active_slave) &&
2568                                     ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2569                                         bond_change_active_slave(bond, slave);
2570                                         bond->current_arp_slave = NULL;
2571                                 } else if (bond->curr_active_slave != slave) {
2572                                         /* this slave has just come up but we
2573                                          * already have a current slave; this
2574                                          * can also happen if bond_enslave adds
2575                                          * a new slave that is up while we are
2576                                          * searching for a new slave
2577                                          */
2578                                         bond_set_slave_inactive_flags(slave);
2579                                         bond->current_arp_slave = NULL;
2580                                 }
2581
2582                                 if (slave == bond->curr_active_slave) {
2583                                         printk(KERN_INFO DRV_NAME
2584                                                ": %s: %s is up and now the "
2585                                                "active interface\n",
2586                                                bond_dev->name,
2587                                                slave->dev->name);
2588                                 } else {
2589                                         printk(KERN_INFO DRV_NAME
2590                                                ": %s: backup interface %s is "
2591                                                "now up\n",
2592                                                bond_dev->name,
2593                                                slave->dev->name);
2594                                 }
2595
2596                                 write_unlock(&bond->curr_slave_lock);
2597                         }
2598                 } else {
2599                         read_lock(&bond->curr_slave_lock);
2600
2601                         if ((slave != bond->curr_active_slave) &&
2602                             (!bond->current_arp_slave) &&
2603                             (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
2604                              bond_has_ip(bond))) {
2605                                 /* a backup slave has gone down; three times
2606                                  * the delta allows the current slave to be
2607                                  * taken out before the backup slave.
2608                                  * note: a non-null current_arp_slave indicates
2609                                  * the curr_active_slave went down and we are
2610                                  * searching for a new one; under this
2611                                  * condition we only take the curr_active_slave
2612                                  * down - this gives each slave a chance to
2613                                  * tx/rx traffic before being taken out
2614                                  */
2615
2616                                 read_unlock(&bond->curr_slave_lock);
2617
2618                                 slave->link  = BOND_LINK_DOWN;
2619
2620                                 if (slave->link_failure_count < UINT_MAX) {
2621                                         slave->link_failure_count++;
2622                                 }
2623
2624                                 bond_set_slave_inactive_flags(slave);
2625
2626                                 printk(KERN_INFO DRV_NAME
2627                                        ": %s: backup interface %s is now down\n",
2628                                        bond_dev->name,
2629                                        slave->dev->name);
2630                         } else {
2631                                 read_unlock(&bond->curr_slave_lock);
2632                         }
2633                 }
2634         }
2635
2636         read_lock(&bond->curr_slave_lock);
2637         slave = bond->curr_active_slave;
2638         read_unlock(&bond->curr_slave_lock);
2639
2640         if (slave) {
2641                 /* if we have sent traffic in the past 2*arp_intervals but
2642                  * haven't xmit and rx traffic in that time interval, select
2643                  * a different slave. slave->jiffies is only updated when
2644                  * a slave first becomes the curr_active_slave - not necessarily
2645                  * after every arp; this ensures the slave has a full 2*delta
2646                  * before being taken out. if a primary is being used, check
2647                  * if it is up and needs to take over as the curr_active_slave
2648                  */
2649                 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2650             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2651              bond_has_ip(bond))) &&
2652                     ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2653
2654                         slave->link  = BOND_LINK_DOWN;
2655
2656                         if (slave->link_failure_count < UINT_MAX) {
2657                                 slave->link_failure_count++;
2658                         }
2659
2660                         printk(KERN_INFO DRV_NAME
2661                                ": %s: link status down for active interface "
2662                                "%s, disabling it\n",
2663                                bond_dev->name,
2664                                slave->dev->name);
2665
2666                         write_lock(&bond->curr_slave_lock);
2667
2668                         bond_select_active_slave(bond);
2669                         slave = bond->curr_active_slave;
2670
2671                         write_unlock(&bond->curr_slave_lock);
2672
2673                         bond->current_arp_slave = slave;
2674
2675                         if (slave) {
2676                                 slave->jiffies = jiffies;
2677                         }
2678                 } else if ((bond->primary_slave) &&
2679                            (bond->primary_slave != slave) &&
2680                            (bond->primary_slave->link == BOND_LINK_UP)) {
2681                         /* at this point, slave is the curr_active_slave */
2682                         printk(KERN_INFO DRV_NAME
2683                                ": %s: changing from interface %s to primary "
2684                                "interface %s\n",
2685                                bond_dev->name,
2686                                slave->dev->name,
2687                                bond->primary_slave->dev->name);
2688
2689                         /* primary is up so switch to it */
2690                         write_lock(&bond->curr_slave_lock);
2691                         bond_change_active_slave(bond, bond->primary_slave);
2692                         write_unlock(&bond->curr_slave_lock);
2693
2694                         slave = bond->primary_slave;
2695                         slave->jiffies = jiffies;
2696                 } else {
2697                         bond->current_arp_slave = NULL;
2698                 }
2699
2700                 /* the current slave must tx an arp to ensure backup slaves
2701                  * rx traffic
2702                  */
2703                 if (slave && bond_has_ip(bond)) {
2704                         bond_arp_send_all(bond, slave);
2705                 }
2706         }
2707
2708         /* if we don't have a curr_active_slave, search for the next available
2709          * backup slave from the current_arp_slave and make it the candidate
2710          * for becoming the curr_active_slave
2711          */
2712         if (!slave) {
2713                 if (!bond->current_arp_slave) {
2714                         bond->current_arp_slave = bond->first_slave;
2715                 }
2716
2717                 if (bond->current_arp_slave) {
2718                         bond_set_slave_inactive_flags(bond->current_arp_slave);
2719
2720                         /* search for next candidate */
2721                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2722                                 if (IS_UP(slave->dev)) {
2723                                         slave->link = BOND_LINK_BACK;
2724                                         bond_set_slave_active_flags(slave);
2725                                         bond_arp_send_all(bond, slave);
2726                                         slave->jiffies = jiffies;
2727                                         bond->current_arp_slave = slave;
2728                                         break;
2729                                 }
2730
2731                                 /* if the link state is up at this point, we
2732                                  * mark it down - this can happen if we have
2733                                  * simultaneous link failures and
2734                                  * reselect_active_interface doesn't make this
2735                                  * one the current slave so it is still marked
2736                                  * up when it is actually down
2737                                  */
2738                                 if (slave->link == BOND_LINK_UP) {
2739                                         slave->link  = BOND_LINK_DOWN;
2740                                         if (slave->link_failure_count < UINT_MAX) {
2741                                                 slave->link_failure_count++;
2742                                         }
2743
2744                                         bond_set_slave_inactive_flags(slave);
2745
2746                                         printk(KERN_INFO DRV_NAME
2747                                                ": %s: backup interface %s is "
2748                                                "now down.\n",
2749                                                bond_dev->name,
2750                                                slave->dev->name);
2751                                 }
2752                         }
2753                 }
2754         }
2755
2756 re_arm:
2757         if (bond->params.arp_interval) {
2758                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2759         }
2760 out:
2761         read_unlock(&bond->lock);
2762 }
2763
2764 /*------------------------------ proc/seq_file-------------------------------*/
2765
2766 #ifdef CONFIG_PROC_FS
2767
2768 #define SEQ_START_TOKEN ((void *)1)
2769
2770 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2771 {
2772         struct bonding *bond = seq->private;
2773         loff_t off = 0;
2774         struct slave *slave;
2775         int i;
2776
2777         /* make sure the bond won't be taken away */
2778         read_lock(&dev_base_lock);
2779         read_lock_bh(&bond->lock);
2780
2781         if (*pos == 0) {
2782                 return SEQ_START_TOKEN;
2783         }
2784
2785         bond_for_each_slave(bond, slave, i) {
2786                 if (++off == *pos) {
2787                         return slave;
2788                 }
2789         }
2790
2791         return NULL;
2792 }
2793
2794 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2795 {
2796         struct bonding *bond = seq->private;
2797         struct slave *slave = v;
2798
2799         ++*pos;
2800         if (v == SEQ_START_TOKEN) {
2801                 return bond->first_slave;
2802         }
2803
2804         slave = slave->next;
2805
2806         return (slave == bond->first_slave) ? NULL : slave;
2807 }
2808
2809 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2810 {
2811         struct bonding *bond = seq->private;
2812
2813         read_unlock_bh(&bond->lock);
2814         read_unlock(&dev_base_lock);
2815 }
2816
2817 static void bond_info_show_master(struct seq_file *seq)
2818 {
2819         struct bonding *bond = seq->private;
2820         struct slave *curr;
2821         int i;
2822         u32 target;
2823
2824         read_lock(&bond->curr_slave_lock);
2825         curr = bond->curr_active_slave;
2826         read_unlock(&bond->curr_slave_lock);
2827
2828         seq_printf(seq, "Bonding Mode: %s\n",
2829                    bond_mode_name(bond->params.mode));
2830
2831         if (bond->params.mode == BOND_MODE_XOR ||
2832                 bond->params.mode == BOND_MODE_8023AD) {
2833                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2834                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2835                         bond->params.xmit_policy);
2836         }
2837
2838         if (USES_PRIMARY(bond->params.mode)) {
2839                 seq_printf(seq, "Primary Slave: %s\n",
2840                            (bond->primary_slave) ?
2841                            bond->primary_slave->dev->name : "None");
2842
2843                 seq_printf(seq, "Currently Active Slave: %s\n",
2844                            (curr) ? curr->dev->name : "None");
2845         }
2846
2847         seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down");
2848         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
2849         seq_printf(seq, "Up Delay (ms): %d\n",
2850                    bond->params.updelay * bond->params.miimon);
2851         seq_printf(seq, "Down Delay (ms): %d\n",
2852                    bond->params.downdelay * bond->params.miimon);
2853
2854
2855         /* ARP information */
2856         if(bond->params.arp_interval > 0) {
2857                 int printed=0;
2858                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
2859                                 bond->params.arp_interval);
2860
2861                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
2862
2863                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
2864                         if (!bond->params.arp_targets[i])
2865                                 continue;
2866                         if (printed)
2867                                 seq_printf(seq, ",");
2868                         target = ntohl(bond->params.arp_targets[i]);
2869                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
2870                         printed = 1;
2871                 }
2872                 seq_printf(seq, "\n");
2873         }
2874
2875         if (bond->params.mode == BOND_MODE_8023AD) {
2876                 struct ad_info ad_info;
2877
2878                 seq_puts(seq, "\n802.3ad info\n");
2879                 seq_printf(seq, "LACP rate: %s\n",
2880                            (bond->params.lacp_fast) ? "fast" : "slow");
2881
2882                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
2883                         seq_printf(seq, "bond %s has no active aggregator\n",
2884                                    bond->dev->name);
2885                 } else {
2886                         seq_printf(seq, "Active Aggregator Info:\n");
2887
2888                         seq_printf(seq, "\tAggregator ID: %d\n",
2889                                    ad_info.aggregator_id);
2890                         seq_printf(seq, "\tNumber of ports: %d\n",
2891                                    ad_info.ports);
2892                         seq_printf(seq, "\tActor Key: %d\n",
2893                                    ad_info.actor_key);
2894                         seq_printf(seq, "\tPartner Key: %d\n",
2895                                    ad_info.partner_key);
2896                         seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2897                                    ad_info.partner_system[0],
2898                                    ad_info.partner_system[1],
2899                                    ad_info.partner_system[2],
2900                                    ad_info.partner_system[3],
2901                                    ad_info.partner_system[4],
2902                                    ad_info.partner_system[5]);
2903                 }
2904         }
2905 }
2906
2907 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
2908 {
2909         struct bonding *bond = seq->private;
2910
2911         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
2912         seq_printf(seq, "MII Status: %s\n",
2913                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
2914         seq_printf(seq, "Link Failure Count: %d\n",
2915                    slave->link_failure_count);
2916
2917         seq_printf(seq,
2918                    "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
2919                    slave->perm_hwaddr[0], slave->perm_hwaddr[1],
2920                    slave->perm_hwaddr[2], slave->perm_hwaddr[3],
2921                    slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
2922
2923         if (bond->params.mode == BOND_MODE_8023AD) {
2924                 const struct aggregator *agg
2925                         = SLAVE_AD_INFO(slave).port.aggregator;
2926
2927                 if (agg) {
2928                         seq_printf(seq, "Aggregator ID: %d\n",
2929                                    agg->aggregator_identifier);
2930                 } else {
2931                         seq_puts(seq, "Aggregator ID: N/A\n");
2932                 }
2933         }
2934 }
2935
2936 static int bond_info_seq_show(struct seq_file *seq, void *v)
2937 {
2938         if (v == SEQ_START_TOKEN) {
2939                 seq_printf(seq, "%s\n", version);
2940                 bond_info_show_master(seq);
2941         } else {
2942                 bond_info_show_slave(seq, v);
2943         }
2944
2945         return 0;
2946 }
2947
2948 static struct seq_operations bond_info_seq_ops = {
2949         .start = bond_info_seq_start,
2950         .next  = bond_info_seq_next,
2951         .stop  = bond_info_seq_stop,
2952         .show  = bond_info_seq_show,
2953 };
2954
2955 static int bond_info_open(struct inode *inode, struct file *file)
2956 {
2957         struct seq_file *seq;
2958         struct proc_dir_entry *proc;
2959         int res;
2960
2961         res = seq_open(file, &bond_info_seq_ops);
2962         if (!res) {
2963                 /* recover the pointer buried in proc_dir_entry data */
2964                 seq = file->private_data;
2965                 proc = PDE(inode);
2966                 seq->private = proc->data;
2967         }
2968
2969         return res;
2970 }
2971
2972 static struct file_operations bond_info_fops = {
2973         .owner   = THIS_MODULE,
2974         .open    = bond_info_open,
2975         .read    = seq_read,
2976         .llseek  = seq_lseek,
2977         .release = seq_release,
2978 };
2979
2980 static int bond_create_proc_entry(struct bonding *bond)
2981 {
2982         struct net_device *bond_dev = bond->dev;
2983
2984         if (bond_proc_dir) {
2985                 bond->proc_entry = create_proc_entry(bond_dev->name,
2986                                                      S_IRUGO,
2987                                                      bond_proc_dir);
2988                 if (bond->proc_entry == NULL) {
2989                         printk(KERN_WARNING DRV_NAME
2990                                ": Warning: Cannot create /proc/net/%s/%s\n",
2991                                DRV_NAME, bond_dev->name);
2992                 } else {
2993                         bond->proc_entry->data = bond;
2994                         bond->proc_entry->proc_fops = &bond_info_fops;
2995                         bond->proc_entry->owner = THIS_MODULE;
2996                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
2997                 }
2998         }
2999
3000         return 0;
3001 }
3002
3003 static void bond_remove_proc_entry(struct bonding *bond)
3004 {
3005         if (bond_proc_dir && bond->proc_entry) {
3006                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3007                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3008                 bond->proc_entry = NULL;
3009         }
3010 }
3011
3012 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3013  * Caller must hold rtnl_lock.
3014  */
3015 static void bond_create_proc_dir(void)
3016 {
3017         int len = strlen(DRV_NAME);
3018
3019         for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3020              bond_proc_dir = bond_proc_dir->next) {
3021                 if ((bond_proc_dir->namelen == len) &&
3022                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3023                         break;
3024                 }
3025         }
3026
3027         if (!bond_proc_dir) {
3028                 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3029                 if (bond_proc_dir) {
3030                         bond_proc_dir->owner = THIS_MODULE;
3031                 } else {
3032                         printk(KERN_WARNING DRV_NAME
3033                                 ": Warning: cannot create /proc/net/%s\n",
3034                                 DRV_NAME);
3035                 }
3036         }
3037 }
3038
3039 /* Destroy the bonding directory under /proc/net, if empty.
3040  * Caller must hold rtnl_lock.
3041  */
3042 static void bond_destroy_proc_dir(void)
3043 {
3044         struct proc_dir_entry *de;
3045
3046         if (!bond_proc_dir) {
3047                 return;
3048         }
3049
3050         /* verify that the /proc dir is empty */
3051         for (de = bond_proc_dir->subdir; de; de = de->next) {
3052                 /* ignore . and .. */
3053                 if (*(de->name) != '.') {
3054                         break;
3055                 }
3056         }
3057
3058         if (de) {
3059                 if (bond_proc_dir->owner == THIS_MODULE) {
3060                         bond_proc_dir->owner = NULL;
3061                 }
3062         } else {
3063                 remove_proc_entry(DRV_NAME, proc_net);
3064                 bond_proc_dir = NULL;
3065         }
3066 }
3067 #endif /* CONFIG_PROC_FS */
3068
3069 /*-------------------------- netdev event handling --------------------------*/
3070
3071 /*
3072  * Change device name
3073  */
3074 static int bond_event_changename(struct bonding *bond)
3075 {
3076 #ifdef CONFIG_PROC_FS
3077         bond_remove_proc_entry(bond);
3078         bond_create_proc_entry(bond);
3079 #endif
3080         down_write(&(bonding_rwsem));
3081         bond_destroy_sysfs_entry(bond);
3082         bond_create_sysfs_entry(bond);
3083         up_write(&(bonding_rwsem));
3084         return NOTIFY_DONE;
3085 }
3086
3087 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3088 {
3089         struct bonding *event_bond = bond_dev->priv;
3090
3091         switch (event) {
3092         case NETDEV_CHANGENAME:
3093                 return bond_event_changename(event_bond);
3094         case NETDEV_UNREGISTER:
3095                 /*
3096                  * TODO: remove a bond from the list?
3097                  */
3098                 break;
3099         default:
3100                 break;
3101         }
3102
3103         return NOTIFY_DONE;
3104 }
3105
3106 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3107 {
3108         struct net_device *bond_dev = slave_dev->master;
3109         struct bonding *bond = bond_dev->priv;
3110
3111         switch (event) {
3112         case NETDEV_UNREGISTER:
3113                 if (bond_dev) {
3114                         bond_release(bond_dev, slave_dev);
3115                 }
3116                 break;
3117         case NETDEV_CHANGE:
3118                 /*
3119                  * TODO: is this what we get if somebody
3120                  * sets up a hierarchical bond, then rmmod's
3121                  * one of the slave bonding devices?
3122                  */
3123                 break;
3124         case NETDEV_DOWN:
3125                 /*
3126                  * ... Or is it this?
3127                  */
3128                 break;
3129         case NETDEV_CHANGEMTU:
3130                 /*
3131                  * TODO: Should slaves be allowed to
3132                  * independently alter their MTU?  For
3133                  * an active-backup bond, slaves need
3134                  * not be the same type of device, so
3135                  * MTUs may vary.  For other modes,
3136                  * slaves arguably should have the
3137                  * same MTUs. To do this, we'd need to
3138                  * take over the slave's change_mtu
3139                  * function for the duration of their
3140                  * servitude.
3141                  */
3142                 break;
3143         case NETDEV_CHANGENAME:
3144                 /*
3145                  * TODO: handle changing the primary's name
3146                  */
3147                 break;
3148         case NETDEV_FEAT_CHANGE:
3149                 bond_compute_features(bond);
3150                 break;
3151         default:
3152                 break;
3153         }
3154
3155         return NOTIFY_DONE;
3156 }
3157
3158 /*
3159  * bond_netdev_event: handle netdev notifier chain events.
3160  *
3161  * This function receives events for the netdev chain.  The caller (an
3162  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3163  * locks for us to safely manipulate the slave devices (RTNL lock,
3164  * dev_probe_lock).
3165  */
3166 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3167 {
3168         struct net_device *event_dev = (struct net_device *)ptr;
3169
3170         dprintk("event_dev: %s, event: %lx\n",
3171                 (event_dev ? event_dev->name : "None"),
3172                 event);
3173
3174         if (event_dev->flags & IFF_MASTER) {
3175                 dprintk("IFF_MASTER\n");
3176                 return bond_master_netdev_event(event, event_dev);
3177         }
3178
3179         if (event_dev->flags & IFF_SLAVE) {
3180                 dprintk("IFF_SLAVE\n");
3181                 return bond_slave_netdev_event(event, event_dev);
3182         }
3183
3184         return NOTIFY_DONE;
3185 }
3186
3187 /*
3188  * bond_inetaddr_event: handle inetaddr notifier chain events.
3189  *
3190  * We keep track of device IPs primarily to use as source addresses in
3191  * ARP monitor probes (rather than spewing out broadcasts all the time).
3192  *
3193  * We track one IP for the main device (if it has one), plus one per VLAN.
3194  */
3195 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3196 {
3197         struct in_ifaddr *ifa = ptr;
3198         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3199         struct bonding *bond, *bond_next;
3200         struct vlan_entry *vlan, *vlan_next;
3201
3202         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3203                 if (bond->dev == event_dev) {
3204                         switch (event) {
3205                         case NETDEV_UP:
3206                                 bond->master_ip = ifa->ifa_local;
3207                                 return NOTIFY_OK;
3208                         case NETDEV_DOWN:
3209                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3210                                 return NOTIFY_OK;
3211                         default:
3212                                 return NOTIFY_DONE;
3213                         }
3214                 }
3215
3216                 if (list_empty(&bond->vlan_list))
3217                         continue;
3218
3219                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3220                                          vlan_list) {
3221                         vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3222                         if (vlan_dev == event_dev) {
3223                                 switch (event) {
3224                                 case NETDEV_UP:
3225                                         vlan->vlan_ip = ifa->ifa_local;
3226                                         return NOTIFY_OK;
3227                                 case NETDEV_DOWN:
3228                                         vlan->vlan_ip =
3229                                                 bond_glean_dev_ip(vlan_dev);
3230                                         return NOTIFY_OK;
3231                                 default:
3232                                         return NOTIFY_DONE;
3233                                 }
3234                         }
3235                 }
3236         }
3237         return NOTIFY_DONE;
3238 }
3239
3240 static struct notifier_block bond_netdev_notifier = {
3241         .notifier_call = bond_netdev_event,
3242 };
3243
3244 static struct notifier_block bond_inetaddr_notifier = {
3245         .notifier_call = bond_inetaddr_event,
3246 };
3247
3248 /*-------------------------- Packet type handling ---------------------------*/
3249
3250 /* register to receive lacpdus on a bond */
3251 static void bond_register_lacpdu(struct bonding *bond)
3252 {
3253         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3254
3255         /* initialize packet type */
3256         pk_type->type = PKT_TYPE_LACPDU;
3257         pk_type->dev = bond->dev;
3258         pk_type->func = bond_3ad_lacpdu_recv;
3259
3260         dev_add_pack(pk_type);
3261 }
3262
3263 /* unregister to receive lacpdus on a bond */
3264 static void bond_unregister_lacpdu(struct bonding *bond)
3265 {
3266         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3267 }
3268
3269 /*---------------------------- Hashing Policies -----------------------------*/
3270
3271 /*
3272  * Hash for the the output device based upon layer 3 and layer 4 data. If
3273  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3274  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3275  */
3276 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3277                                     struct net_device *bond_dev, int count)
3278 {
3279         struct ethhdr *data = (struct ethhdr *)skb->data;
3280         struct iphdr *iph = skb->nh.iph;
3281         u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3282         int layer4_xor = 0;
3283
3284         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3285                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3286                     (iph->protocol == IPPROTO_TCP ||
3287                      iph->protocol == IPPROTO_UDP)) {
3288                         layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3289                 }
3290                 return (layer4_xor ^
3291                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3292
3293         }
3294
3295         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3296 }
3297
3298 /*
3299  * Hash for the output device based upon layer 2 data
3300  */
3301 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3302                                    struct net_device *bond_dev, int count)
3303 {
3304         struct ethhdr *data = (struct ethhdr *)skb->data;
3305
3306         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3307 }
3308
3309 /*-------------------------- Device entry points ----------------------------*/
3310
3311 static int bond_open(struct net_device *bond_dev)
3312 {
3313         struct bonding *bond = bond_dev->priv;
3314         struct timer_list *mii_timer = &bond->mii_timer;
3315         struct timer_list *arp_timer = &bond->arp_timer;
3316
3317         bond->kill_timers = 0;
3318
3319         if ((bond->params.mode == BOND_MODE_TLB) ||
3320             (bond->params.mode == BOND_MODE_ALB)) {
3321                 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3322
3323                 /* bond_alb_initialize must be called before the timer
3324                  * is started.
3325                  */
3326                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3327                         /* something went wrong - fail the open operation */
3328                         return -1;
3329                 }
3330
3331                 init_timer(alb_timer);
3332                 alb_timer->expires  = jiffies + 1;
3333                 alb_timer->data     = (unsigned long)bond;
3334                 alb_timer->function = (void *)&bond_alb_monitor;
3335                 add_timer(alb_timer);
3336         }
3337
3338         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3339                 init_timer(mii_timer);
3340                 mii_timer->expires  = jiffies + 1;
3341                 mii_timer->data     = (unsigned long)bond_dev;
3342                 mii_timer->function = (void *)&bond_mii_monitor;
3343                 add_timer(mii_timer);
3344         }
3345
3346         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3347                 init_timer(arp_timer);
3348                 arp_timer->expires  = jiffies + 1;
3349                 arp_timer->data     = (unsigned long)bond_dev;
3350                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3351                         arp_timer->function = (void *)&bond_activebackup_arp_mon;
3352                 } else {
3353                         arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3354                 }
3355                 add_timer(arp_timer);
3356         }
3357
3358         if (bond->params.mode == BOND_MODE_8023AD) {
3359                 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3360                 init_timer(ad_timer);
3361                 ad_timer->expires  = jiffies + 1;
3362                 ad_timer->data     = (unsigned long)bond;
3363                 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3364                 add_timer(ad_timer);
3365
3366                 /* register to receive LACPDUs */
3367                 bond_register_lacpdu(bond);
3368         }
3369
3370         return 0;
3371 }
3372
3373 static int bond_close(struct net_device *bond_dev)
3374 {
3375         struct bonding *bond = bond_dev->priv;
3376
3377         if (bond->params.mode == BOND_MODE_8023AD) {
3378                 /* Unregister the receive of LACPDUs */
3379                 bond_unregister_lacpdu(bond);
3380         }
3381
3382         write_lock_bh(&bond->lock);
3383
3384         bond_mc_list_destroy(bond);
3385
3386         /* signal timers not to re-arm */
3387         bond->kill_timers = 1;
3388
3389         write_unlock_bh(&bond->lock);
3390
3391         /* del_timer_sync must run without holding the bond->lock
3392          * because a running timer might be trying to hold it too
3393          */
3394
3395         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3396                 del_timer_sync(&bond->mii_timer);
3397         }
3398
3399         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3400                 del_timer_sync(&bond->arp_timer);
3401         }
3402
3403         switch (bond->params.mode) {
3404         case BOND_MODE_8023AD:
3405                 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3406                 break;
3407         case BOND_MODE_TLB:
3408         case BOND_MODE_ALB:
3409                 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3410                 break;
3411         default:
3412                 break;
3413         }
3414
3415         /* Release the bonded slaves */
3416         bond_release_all(bond_dev);
3417
3418         if ((bond->params.mode == BOND_MODE_TLB) ||
3419             (bond->params.mode == BOND_MODE_ALB)) {
3420                 /* Must be called only after all
3421                  * slaves have been released
3422                  */
3423                 bond_alb_deinitialize(bond);
3424         }
3425
3426         return 0;
3427 }
3428
3429 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3430 {
3431         struct bonding *bond = bond_dev->priv;
3432         struct net_device_stats *stats = &(bond->stats), *sstats;
3433         struct slave *slave;
3434         int i;
3435
3436         memset(stats, 0, sizeof(struct net_device_stats));
3437
3438         read_lock_bh(&bond->lock);
3439
3440         bond_for_each_slave(bond, slave, i) {
3441                 sstats = slave->dev->get_stats(slave->dev);
3442
3443                 stats->rx_packets += sstats->rx_packets;
3444                 stats->rx_bytes += sstats->rx_bytes;
3445                 stats->rx_errors += sstats->rx_errors;
3446                 stats->rx_dropped += sstats->rx_dropped;
3447
3448                 stats->tx_packets += sstats->tx_packets;
3449                 stats->tx_bytes += sstats->tx_bytes;
3450                 stats->tx_errors += sstats->tx_errors;
3451                 stats->tx_dropped += sstats->tx_dropped;
3452
3453                 stats->multicast += sstats->multicast;
3454                 stats->collisions += sstats->collisions;
3455
3456                 stats->rx_length_errors += sstats->rx_length_errors;
3457                 stats->rx_over_errors += sstats->rx_over_errors;
3458                 stats->rx_crc_errors += sstats->rx_crc_errors;
3459                 stats->rx_frame_errors += sstats->rx_frame_errors;
3460                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3461                 stats->rx_missed_errors += sstats->rx_missed_errors;
3462
3463                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3464                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3465                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3466                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3467                 stats->tx_window_errors += sstats->tx_window_errors;
3468         }
3469
3470         read_unlock_bh(&bond->lock);
3471
3472         return stats;
3473 }
3474
3475 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3476 {
3477         struct net_device *slave_dev = NULL;
3478         struct ifbond k_binfo;
3479         struct ifbond __user *u_binfo = NULL;
3480         struct ifslave k_sinfo;
3481         struct ifslave __user *u_sinfo = NULL;
3482         struct mii_ioctl_data *mii = NULL;
3483         int res = 0;
3484
3485         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3486                 bond_dev->name, cmd);
3487
3488         switch (cmd) {
3489         case SIOCGMIIPHY:
3490                 mii = if_mii(ifr);
3491                 if (!mii) {
3492                         return -EINVAL;
3493                 }
3494                 mii->phy_id = 0;
3495                 /* Fall Through */
3496         case SIOCGMIIREG:
3497                 /*
3498                  * We do this again just in case we were called by SIOCGMIIREG
3499                  * instead of SIOCGMIIPHY.
3500                  */
3501                 mii = if_mii(ifr);
3502                 if (!mii) {
3503                         return -EINVAL;
3504                 }
3505
3506                 if (mii->reg_num == 1) {
3507                         struct bonding *bond = bond_dev->priv;
3508                         mii->val_out = 0;
3509                         read_lock_bh(&bond->lock);
3510                         read_lock(&bond->curr_slave_lock);
3511                         if (bond->curr_active_slave) {
3512                                 mii->val_out = BMSR_LSTATUS;
3513                         }
3514                         read_unlock(&bond->curr_slave_lock);
3515                         read_unlock_bh(&bond->lock);
3516                 }
3517
3518                 return 0;
3519         case BOND_INFO_QUERY_OLD:
3520         case SIOCBONDINFOQUERY:
3521                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3522
3523                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3524                         return -EFAULT;
3525                 }
3526
3527                 res = bond_info_query(bond_dev, &k_binfo);
3528                 if (res == 0) {
3529                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3530                                 return -EFAULT;
3531                         }
3532                 }
3533
3534                 return res;
3535         case BOND_SLAVE_INFO_QUERY_OLD:
3536         case SIOCBONDSLAVEINFOQUERY:
3537                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3538
3539                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3540                         return -EFAULT;
3541                 }
3542
3543                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3544                 if (res == 0) {
3545                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3546                                 return -EFAULT;
3547                         }
3548                 }
3549
3550                 return res;
3551         default:
3552                 /* Go on */
3553                 break;
3554         }
3555
3556         if (!capable(CAP_NET_ADMIN)) {
3557                 return -EPERM;
3558         }
3559
3560         down_write(&(bonding_rwsem));
3561         slave_dev = dev_get_by_name(ifr->ifr_slave);
3562
3563         dprintk("slave_dev=%p: \n", slave_dev);
3564
3565         if (!slave_dev) {
3566                 res = -ENODEV;
3567         } else {
3568                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3569                 switch (cmd) {
3570                 case BOND_ENSLAVE_OLD:
3571                 case SIOCBONDENSLAVE:
3572                         res = bond_enslave(bond_dev, slave_dev);
3573                         break;
3574                 case BOND_RELEASE_OLD:
3575                 case SIOCBONDRELEASE:
3576                         res = bond_release(bond_dev, slave_dev);
3577                         break;
3578                 case BOND_SETHWADDR_OLD:
3579                 case SIOCBONDSETHWADDR:
3580                         res = bond_sethwaddr(bond_dev, slave_dev);
3581                         break;
3582                 case BOND_CHANGE_ACTIVE_OLD:
3583                 case SIOCBONDCHANGEACTIVE:
3584                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3585                         break;
3586                 default:
3587                         res = -EOPNOTSUPP;
3588                 }
3589
3590                 dev_put(slave_dev);
3591         }
3592
3593         up_write(&(bonding_rwsem));
3594         return res;
3595 }
3596
3597 static void bond_set_multicast_list(struct net_device *bond_dev)
3598 {
3599         struct bonding *bond = bond_dev->priv;
3600         struct dev_mc_list *dmi;
3601
3602         write_lock_bh(&bond->lock);
3603
3604         /*
3605          * Do promisc before checking multicast_mode
3606          */
3607         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3608                 bond_set_promiscuity(bond, 1);
3609         }
3610
3611         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3612                 bond_set_promiscuity(bond, -1);
3613         }
3614
3615         /* set allmulti flag to slaves */
3616         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3617                 bond_set_allmulti(bond, 1);
3618         }
3619
3620         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3621                 bond_set_allmulti(bond, -1);
3622         }
3623
3624         bond->flags = bond_dev->flags;
3625
3626         /* looking for addresses to add to slaves' mc list */
3627         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3628                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3629                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3630                 }
3631         }
3632
3633         /* looking for addresses to delete from slaves' list */
3634         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3635                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3636                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3637                 }
3638         }
3639
3640         /* save master's multicast list */
3641         bond_mc_list_destroy(bond);
3642         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3643
3644         write_unlock_bh(&bond->lock);
3645 }
3646
3647 /*
3648  * Change the MTU of all of a master's slaves to match the master
3649  */
3650 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3651 {
3652         struct bonding *bond = bond_dev->priv;
3653         struct slave *slave, *stop_at;
3654         int res = 0;
3655         int i;
3656
3657         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3658                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3659
3660         /* Can't hold bond->lock with bh disabled here since
3661          * some base drivers panic. On the other hand we can't
3662          * hold bond->lock without bh disabled because we'll
3663          * deadlock. The only solution is to rely on the fact
3664          * that we're under rtnl_lock here, and the slaves
3665          * list won't change. This doesn't solve the problem
3666          * of setting the slave's MTU while it is
3667          * transmitting, but the assumption is that the base
3668          * driver can handle that.
3669          *
3670          * TODO: figure out a way to safely iterate the slaves
3671          * list, but without holding a lock around the actual
3672          * call to the base driver.
3673          */
3674
3675         bond_for_each_slave(bond, slave, i) {
3676                 dprintk("s %p s->p %p c_m %p\n", slave,
3677                         slave->prev, slave->dev->change_mtu);
3678
3679                 res = dev_set_mtu(slave->dev, new_mtu);
3680
3681                 if (res) {
3682                         /* If we failed to set the slave's mtu to the new value
3683                          * we must abort the operation even in ACTIVE_BACKUP
3684                          * mode, because if we allow the backup slaves to have
3685                          * different mtu values than the active slave we'll
3686                          * need to change their mtu when doing a failover. That
3687                          * means changing their mtu from timer context, which
3688                          * is probably not a good idea.
3689                          */
3690                         dprintk("err %d %s\n", res, slave->dev->name);
3691                         goto unwind;
3692                 }
3693         }
3694
3695         bond_dev->mtu = new_mtu;
3696
3697         return 0;
3698
3699 unwind:
3700         /* unwind from head to the slave that failed */
3701         stop_at = slave;
3702         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3703                 int tmp_res;
3704
3705                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3706                 if (tmp_res) {
3707                         dprintk("unwind err %d dev %s\n", tmp_res,
3708                                 slave->dev->name);
3709                 }
3710         }
3711
3712         return res;
3713 }
3714
3715 /*
3716  * Change HW address
3717  *
3718  * Note that many devices must be down to change the HW address, and
3719  * downing the master releases all slaves.  We can make bonds full of
3720  * bonding devices to test this, however.
3721  */
3722 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3723 {
3724         struct bonding *bond = bond_dev->priv;
3725         struct sockaddr *sa = addr, tmp_sa;
3726         struct slave *slave, *stop_at;
3727         int res = 0;
3728         int i;
3729
3730         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3731
3732         if (!is_valid_ether_addr(sa->sa_data)) {
3733                 return -EADDRNOTAVAIL;
3734         }
3735
3736         /* Can't hold bond->lock with bh disabled here since
3737          * some base drivers panic. On the other hand we can't
3738          * hold bond->lock without bh disabled because we'll
3739          * deadlock. The only solution is to rely on the fact
3740          * that we're under rtnl_lock here, and the slaves
3741          * list won't change. This doesn't solve the problem
3742          * of setting the slave's hw address while it is
3743          * transmitting, but the assumption is that the base
3744          * driver can handle that.
3745          *
3746          * TODO: figure out a way to safely iterate the slaves
3747          * list, but without holding a lock around the actual
3748          * call to the base driver.
3749          */
3750
3751         bond_for_each_slave(bond, slave, i) {
3752                 dprintk("slave %p %s\n", slave, slave->dev->name);
3753
3754                 if (slave->dev->set_mac_address == NULL) {
3755                         res = -EOPNOTSUPP;
3756                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3757                         goto unwind;
3758                 }
3759
3760                 res = dev_set_mac_address(slave->dev, addr);
3761                 if (res) {
3762                         /* TODO: consider downing the slave
3763                          * and retry ?
3764                          * User should expect communications
3765                          * breakage anyway until ARP finish
3766                          * updating, so...
3767                          */
3768                         dprintk("err %d %s\n", res, slave->dev->name);
3769                         goto unwind;
3770                 }
3771         }
3772
3773         /* success */
3774         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3775         return 0;
3776
3777 unwind:
3778         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3779         tmp_sa.sa_family = bond_dev->type;
3780
3781         /* unwind from head to the slave that failed */
3782         stop_at = slave;
3783         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3784                 int tmp_res;
3785
3786                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3787                 if (tmp_res) {
3788                         dprintk("unwind err %d dev %s\n", tmp_res,
3789                                 slave->dev->name);
3790                 }
3791         }
3792
3793         return res;
3794 }
3795
3796 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3797 {
3798         struct bonding *bond = bond_dev->priv;
3799         struct slave *slave, *start_at;
3800         int i;
3801         int res = 1;
3802
3803         read_lock(&bond->lock);
3804
3805         if (!BOND_IS_OK(bond)) {
3806                 goto out;
3807         }
3808
3809         read_lock(&bond->curr_slave_lock);
3810         slave = start_at = bond->curr_active_slave;
3811         read_unlock(&bond->curr_slave_lock);
3812
3813         if (!slave) {
3814                 goto out;
3815         }
3816
3817         bond_for_each_slave_from(bond, slave, i, start_at) {
3818                 if (IS_UP(slave->dev) &&
3819                     (slave->link == BOND_LINK_UP) &&
3820                     (slave->state == BOND_STATE_ACTIVE)) {
3821                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3822
3823                         write_lock(&bond->curr_slave_lock);
3824                         bond->curr_active_slave = slave->next;
3825                         write_unlock(&bond->curr_slave_lock);
3826
3827                         break;
3828                 }
3829         }
3830
3831
3832 out:
3833         if (res) {
3834                 /* no suitable interface, frame not sent */
3835                 dev_kfree_skb(skb);
3836         }
3837         read_unlock(&bond->lock);
3838         return 0;
3839 }
3840
3841 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
3842                                         struct bonding *bond,
3843                                         struct slave *slave)
3844 {
3845         struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
3846         struct ethhdr *eth_data;
3847         u8 *hwaddr;
3848         int res;
3849
3850         if (!skb2) {
3851                 printk(KERN_ERR DRV_NAME ": Error: "
3852                        "bond_activebackup_xmit_copy(): skb_copy() failed\n");
3853                 return;
3854         }
3855
3856         skb2->mac.raw = (unsigned char *)skb2->data;
3857         eth_data = eth_hdr(skb2);
3858
3859         /* Pick an appropriate source MAC address
3860          *      -- use slave's perm MAC addr, unless used by bond
3861          *      -- otherwise, borrow active slave's perm MAC addr
3862          *         since that will not be used
3863          */
3864         hwaddr = slave->perm_hwaddr;
3865         if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
3866                 hwaddr = bond->curr_active_slave->perm_hwaddr;
3867
3868         /* Set source MAC address appropriately */
3869         memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
3870
3871         res = bond_dev_queue_xmit(bond, skb2, slave->dev);
3872         if (res)
3873                 dev_kfree_skb(skb2);
3874
3875         return;
3876 }
3877
3878 /*
3879  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3880  * the bond has a usable interface.
3881  */
3882 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3883 {
3884         struct bonding *bond = bond_dev->priv;
3885         int res = 1;
3886
3887         read_lock(&bond->lock);
3888         read_lock(&bond->curr_slave_lock);
3889
3890         if (!BOND_IS_OK(bond)) {
3891                 goto out;
3892         }
3893
3894         if (!bond->curr_active_slave)
3895                 goto out;
3896
3897         /* Xmit IGMP frames on all slaves to ensure rapid fail-over
3898            for multicast traffic on snooping switches */
3899         if (skb->protocol == __constant_htons(ETH_P_IP) &&
3900             skb->nh.iph->protocol == IPPROTO_IGMP) {
3901                 struct slave *slave, *active_slave;
3902                 int i;
3903
3904                 active_slave = bond->curr_active_slave;
3905                 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
3906                                             active_slave->prev)
3907                         if (IS_UP(slave->dev) &&
3908                             (slave->link == BOND_LINK_UP))
3909                                 bond_activebackup_xmit_copy(skb, bond, slave);
3910         }
3911
3912         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
3913
3914 out:
3915         if (res) {
3916                 /* no suitable interface, frame not sent */
3917                 dev_kfree_skb(skb);
3918         }
3919         read_unlock(&bond->curr_slave_lock);
3920         read_unlock(&bond->lock);
3921         return 0;
3922 }
3923
3924 /*
3925  * In bond_xmit_xor() , we determine the output device by using a pre-
3926  * determined xmit_hash_policy(), If the selected device is not enabled,
3927  * find the next active slave.
3928  */
3929 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3930 {
3931         struct bonding *bond = bond_dev->priv;
3932         struct slave *slave, *start_at;
3933         int slave_no;
3934         int i;
3935         int res = 1;
3936
3937         read_lock(&bond->lock);
3938
3939         if (!BOND_IS_OK(bond)) {
3940                 goto out;
3941         }
3942
3943         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
3944
3945         bond_for_each_slave(bond, slave, i) {
3946                 slave_no--;
3947                 if (slave_no < 0) {
3948                         break;
3949                 }
3950         }
3951
3952         start_at = slave;
3953
3954         bond_for_each_slave_from(bond, slave, i, start_at) {
3955                 if (IS_UP(slave->dev) &&
3956                     (slave->link == BOND_LINK_UP) &&
3957                     (slave->state == BOND_STATE_ACTIVE)) {
3958                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3959                         break;
3960                 }
3961         }
3962
3963 out:
3964         if (res) {
3965                 /* no suitable interface, frame not sent */
3966                 dev_kfree_skb(skb);
3967         }
3968         read_unlock(&bond->lock);
3969         return 0;
3970 }
3971
3972 /*
3973  * in broadcast mode, we send everything to all usable interfaces.
3974  */
3975 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3976 {
3977         struct bonding *bond = bond_dev->priv;
3978         struct slave *slave, *start_at;
3979         struct net_device *tx_dev = NULL;
3980         int i;
3981         int res = 1;
3982
3983         read_lock(&bond->lock);
3984
3985         if (!BOND_IS_OK(bond)) {
3986                 goto out;
3987         }
3988
3989         read_lock(&bond->curr_slave_lock);
3990         start_at = bond->curr_active_slave;
3991         read_unlock(&bond->curr_slave_lock);
3992
3993         if (!start_at) {
3994                 goto out;
3995         }
3996
3997         bond_for_each_slave_from(bond, slave, i, start_at) {
3998                 if (IS_UP(slave->dev) &&
3999                     (slave->link == BOND_LINK_UP) &&
4000                     (slave->state == BOND_STATE_ACTIVE)) {
4001                         if (tx_dev) {
4002                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4003                                 if (!skb2) {
4004                                         printk(KERN_ERR DRV_NAME
4005                                                ": %s: Error: bond_xmit_broadcast(): "
4006                                                "skb_clone() failed\n",
4007                                                bond_dev->name);
4008                                         continue;
4009                                 }
4010
4011                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4012                                 if (res) {
4013                                         dev_kfree_skb(skb2);
4014                                         continue;
4015                                 }
4016                         }
4017                         tx_dev = slave->dev;
4018                 }
4019         }
4020
4021         if (tx_dev) {
4022                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4023         }
4024
4025 out:
4026         if (res) {
4027                 /* no suitable interface, frame not sent */
4028                 dev_kfree_skb(skb);
4029         }
4030         /* frame sent to all suitable interfaces */
4031         read_unlock(&bond->lock);
4032         return 0;
4033 }
4034
4035 /*------------------------- Device initialization ---------------------------*/
4036
4037 /*
4038  * set bond mode specific net device operations
4039  */
4040 void bond_set_mode_ops(struct bonding *bond, int mode)
4041 {
4042         struct net_device *bond_dev = bond->dev;
4043
4044         switch (mode) {
4045         case BOND_MODE_ROUNDROBIN:
4046                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4047                 break;
4048         case BOND_MODE_ACTIVEBACKUP:
4049                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4050                 break;
4051         case BOND_MODE_XOR:
4052                 bond_dev->hard_start_xmit = bond_xmit_xor;
4053                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4054                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4055                 else
4056                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4057                 break;
4058         case BOND_MODE_BROADCAST:
4059                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4060                 break;
4061         case BOND_MODE_8023AD:
4062                 bond_set_master_3ad_flags(bond);
4063                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4064                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4065                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4066                 else
4067                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4068                 break;
4069         case BOND_MODE_ALB:
4070                 bond_set_master_alb_flags(bond);
4071                 /* FALLTHRU */
4072         case BOND_MODE_TLB:
4073                 bond_dev->hard_start_xmit = bond_alb_xmit;
4074                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4075                 break;
4076         default:
4077                 /* Should never happen, mode already checked */
4078                 printk(KERN_ERR DRV_NAME
4079                        ": %s: Error: Unknown bonding mode %d\n",
4080                        bond_dev->name,
4081                        mode);
4082                 break;
4083         }
4084 }
4085
4086 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4087                                     struct ethtool_drvinfo *drvinfo)
4088 {
4089         strncpy(drvinfo->driver, DRV_NAME, 32);
4090         strncpy(drvinfo->version, DRV_VERSION, 32);
4091         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4092 }
4093
4094 static struct ethtool_ops bond_ethtool_ops = {
4095         .get_tx_csum            = ethtool_op_get_tx_csum,
4096         .get_tso                = ethtool_op_get_tso,
4097         .get_ufo                = ethtool_op_get_ufo,
4098         .get_sg                 = ethtool_op_get_sg,
4099         .get_drvinfo            = bond_ethtool_get_drvinfo,
4100 };
4101
4102 /*
4103  * Does not allocate but creates a /proc entry.
4104  * Allowed to fail.
4105  */
4106 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4107 {
4108         struct bonding *bond = bond_dev->priv;
4109
4110         dprintk("Begin bond_init for %s\n", bond_dev->name);
4111
4112         /* initialize rwlocks */
4113         rwlock_init(&bond->lock);
4114         rwlock_init(&bond->curr_slave_lock);
4115
4116         bond->params = *params; /* copy params struct */
4117
4118         /* Initialize pointers */
4119         bond->first_slave = NULL;
4120         bond->curr_active_slave = NULL;
4121         bond->current_arp_slave = NULL;
4122         bond->primary_slave = NULL;
4123         bond->dev = bond_dev;
4124         INIT_LIST_HEAD(&bond->vlan_list);
4125
4126         /* Initialize the device entry points */
4127         bond_dev->open = bond_open;
4128         bond_dev->stop = bond_close;
4129         bond_dev->get_stats = bond_get_stats;
4130         bond_dev->do_ioctl = bond_do_ioctl;
4131         bond_dev->ethtool_ops = &bond_ethtool_ops;
4132         bond_dev->set_multicast_list = bond_set_multicast_list;
4133         bond_dev->change_mtu = bond_change_mtu;
4134         bond_dev->set_mac_address = bond_set_mac_address;
4135
4136         bond_set_mode_ops(bond, bond->params.mode);
4137
4138         bond_dev->destructor = free_netdev;
4139
4140         /* Initialize the device options */
4141         bond_dev->tx_queue_len = 0;
4142         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4143
4144         /* At first, we block adding VLANs. That's the only way to
4145          * prevent problems that occur when adding VLANs over an
4146          * empty bond. The block will be removed once non-challenged
4147          * slaves are enslaved.
4148          */
4149         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4150
4151         /* don't acquire bond device's xmit_lock when 
4152          * transmitting */
4153         bond_dev->features |= NETIF_F_LLTX;
4154
4155         /* By default, we declare the bond to be fully
4156          * VLAN hardware accelerated capable. Special
4157          * care is taken in the various xmit functions
4158          * when there are slaves that are not hw accel
4159          * capable
4160          */
4161         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4162         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4163         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4164         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4165                                NETIF_F_HW_VLAN_RX |
4166                                NETIF_F_HW_VLAN_FILTER);
4167
4168 #ifdef CONFIG_PROC_FS
4169         bond_create_proc_entry(bond);
4170 #endif
4171
4172         list_add_tail(&bond->bond_list, &bond_dev_list);
4173
4174         return 0;
4175 }
4176
4177 /* De-initialize device specific data.
4178  * Caller must hold rtnl_lock.
4179  */
4180 void bond_deinit(struct net_device *bond_dev)
4181 {
4182         struct bonding *bond = bond_dev->priv;
4183
4184         list_del(&bond->bond_list);
4185
4186 #ifdef CONFIG_PROC_FS
4187         bond_remove_proc_entry(bond);
4188 #endif
4189 }
4190
4191 /* Unregister and free all bond devices.
4192  * Caller must hold rtnl_lock.
4193  */
4194 static void bond_free_all(void)
4195 {
4196         struct bonding *bond, *nxt;
4197
4198         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4199                 struct net_device *bond_dev = bond->dev;
4200
4201                 unregister_netdevice(bond_dev);
4202                 bond_deinit(bond_dev);
4203         }
4204
4205 #ifdef CONFIG_PROC_FS
4206         bond_destroy_proc_dir();
4207 #endif
4208 }
4209
4210 /*------------------------- Module initialization ---------------------------*/
4211
4212 /*
4213  * Convert string input module parms.  Accept either the
4214  * number of the mode or its string name.
4215  */
4216 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4217 {
4218         int i;
4219
4220         for (i = 0; tbl[i].modename; i++) {
4221                 if ((isdigit(*mode_arg) &&
4222                      tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4223                     (strncmp(mode_arg, tbl[i].modename,
4224                              strlen(tbl[i].modename)) == 0)) {
4225                         return tbl[i].mode;
4226                 }
4227         }
4228
4229         return -1;
4230 }
4231
4232 static int bond_check_params(struct bond_params *params)
4233 {
4234         /*
4235          * Convert string parameters.
4236          */
4237         if (mode) {
4238                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4239                 if (bond_mode == -1) {
4240                         printk(KERN_ERR DRV_NAME
4241                                ": Error: Invalid bonding mode \"%s\"\n",
4242                                mode == NULL ? "NULL" : mode);
4243                         return -EINVAL;
4244                 }
4245         }
4246
4247         if (xmit_hash_policy) {
4248                 if ((bond_mode != BOND_MODE_XOR) &&
4249                     (bond_mode != BOND_MODE_8023AD)) {
4250                         printk(KERN_INFO DRV_NAME
4251                                ": xor_mode param is irrelevant in mode %s\n",
4252                                bond_mode_name(bond_mode));
4253                 } else {
4254                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4255                                                         xmit_hashtype_tbl);
4256                         if (xmit_hashtype == -1) {
4257                                 printk(KERN_ERR DRV_NAME
4258                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4259                                 xmit_hash_policy == NULL ? "NULL" :
4260                                        xmit_hash_policy);
4261                                 return -EINVAL;
4262                         }
4263                 }
4264         }
4265
4266         if (lacp_rate) {
4267                 if (bond_mode != BOND_MODE_8023AD) {
4268                         printk(KERN_INFO DRV_NAME
4269                                ": lacp_rate param is irrelevant in mode %s\n",
4270                                bond_mode_name(bond_mode));
4271                 } else {
4272                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4273                         if (lacp_fast == -1) {
4274                                 printk(KERN_ERR DRV_NAME
4275                                        ": Error: Invalid lacp rate \"%s\"\n",
4276                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4277                                 return -EINVAL;
4278                         }
4279                 }
4280         }
4281
4282         if (max_bonds < 1 || max_bonds > INT_MAX) {
4283                 printk(KERN_WARNING DRV_NAME
4284                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4285                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4286                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4287                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4288         }
4289
4290         if (miimon < 0) {
4291                 printk(KERN_WARNING DRV_NAME
4292                        ": Warning: miimon module parameter (%d), "
4293                        "not in range 0-%d, so it was reset to %d\n",
4294                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4295                 miimon = BOND_LINK_MON_INTERV;
4296         }
4297
4298         if (updelay < 0) {
4299                 printk(KERN_WARNING DRV_NAME
4300                        ": Warning: updelay module parameter (%d), "
4301                        "not in range 0-%d, so it was reset to 0\n",
4302                        updelay, INT_MAX);
4303                 updelay = 0;
4304         }
4305
4306         if (downdelay < 0) {
4307                 printk(KERN_WARNING DRV_NAME
4308                        ": Warning: downdelay module parameter (%d), "
4309                        "not in range 0-%d, so it was reset to 0\n",
4310                        downdelay, INT_MAX);
4311                 downdelay = 0;
4312         }
4313
4314         if ((use_carrier != 0) && (use_carrier != 1)) {
4315                 printk(KERN_WARNING DRV_NAME
4316                        ": Warning: use_carrier module parameter (%d), "
4317                        "not of valid value (0/1), so it was set to 1\n",
4318                        use_carrier);
4319                 use_carrier = 1;
4320         }
4321
4322         /* reset values for 802.3ad */
4323         if (bond_mode == BOND_MODE_8023AD) {
4324                 if (!miimon) {
4325                         printk(KERN_WARNING DRV_NAME
4326                                ": Warning: miimon must be specified, "
4327                                "otherwise bonding will not detect link "
4328                                "failure, speed and duplex which are "
4329                                "essential for 802.3ad operation\n");
4330                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4331                         miimon = 100;
4332                 }
4333         }
4334
4335         /* reset values for TLB/ALB */
4336         if ((bond_mode == BOND_MODE_TLB) ||
4337             (bond_mode == BOND_MODE_ALB)) {
4338                 if (!miimon) {
4339                         printk(KERN_WARNING DRV_NAME
4340                                ": Warning: miimon must be specified, "
4341                                "otherwise bonding will not detect link "
4342                                "failure and link speed which are essential "
4343                                "for TLB/ALB load balancing\n");
4344                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4345                         miimon = 100;
4346                 }
4347         }
4348
4349         if (bond_mode == BOND_MODE_ALB) {
4350                 printk(KERN_NOTICE DRV_NAME
4351                        ": In ALB mode you might experience client "
4352                        "disconnections upon reconnection of a link if the "
4353                        "bonding module updelay parameter (%d msec) is "
4354                        "incompatible with the forwarding delay time of the "
4355                        "switch\n",
4356                        updelay);
4357         }
4358
4359         if (!miimon) {
4360                 if (updelay || downdelay) {
4361                         /* just warn the user the up/down delay will have
4362                          * no effect since miimon is zero...
4363                          */
4364                         printk(KERN_WARNING DRV_NAME
4365                                ": Warning: miimon module parameter not set "
4366                                "and updelay (%d) or downdelay (%d) module "
4367                                "parameter is set; updelay and downdelay have "
4368                                "no effect unless miimon is set\n",
4369                                updelay, downdelay);
4370                 }
4371         } else {
4372                 /* don't allow arp monitoring */
4373                 if (arp_interval) {
4374                         printk(KERN_WARNING DRV_NAME
4375                                ": Warning: miimon (%d) and arp_interval (%d) "
4376                                "can't be used simultaneously, disabling ARP "
4377                                "monitoring\n",
4378                                miimon, arp_interval);
4379                         arp_interval = 0;
4380                 }
4381
4382                 if ((updelay % miimon) != 0) {
4383                         printk(KERN_WARNING DRV_NAME
4384                                ": Warning: updelay (%d) is not a multiple "
4385                                "of miimon (%d), updelay rounded to %d ms\n",
4386                                updelay, miimon, (updelay / miimon) * miimon);
4387                 }
4388
4389                 updelay /= miimon;
4390
4391                 if ((downdelay % miimon) != 0) {
4392                         printk(KERN_WARNING DRV_NAME
4393                                ": Warning: downdelay (%d) is not a multiple "
4394                                "of miimon (%d), downdelay rounded to %d ms\n",
4395                                downdelay, miimon,
4396                                (downdelay / miimon) * miimon);
4397                 }
4398
4399                 downdelay /= miimon;
4400         }
4401
4402         if (arp_interval < 0) {
4403                 printk(KERN_WARNING DRV_NAME
4404                        ": Warning: arp_interval module parameter (%d) "
4405                        ", not in range 0-%d, so it was reset to %d\n",
4406                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4407                 arp_interval = BOND_LINK_ARP_INTERV;
4408         }
4409
4410         for (arp_ip_count = 0;
4411              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4412              arp_ip_count++) {
4413                 /* not complete check, but should be good enough to
4414                    catch mistakes */
4415                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4416                         printk(KERN_WARNING DRV_NAME
4417                                ": Warning: bad arp_ip_target module parameter "
4418                                "(%s), ARP monitoring will not be performed\n",
4419                                arp_ip_target[arp_ip_count]);
4420                         arp_interval = 0;
4421                 } else {
4422                         u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4423                         arp_target[arp_ip_count] = ip;
4424                 }
4425         }
4426
4427         if (arp_interval && !arp_ip_count) {
4428                 /* don't allow arping if no arp_ip_target given... */
4429                 printk(KERN_WARNING DRV_NAME
4430                        ": Warning: arp_interval module parameter (%d) "
4431                        "specified without providing an arp_ip_target "
4432                        "parameter, arp_interval was reset to 0\n",
4433                        arp_interval);
4434                 arp_interval = 0;
4435         }
4436
4437         if (miimon) {
4438                 printk(KERN_INFO DRV_NAME
4439                        ": MII link monitoring set to %d ms\n",
4440                        miimon);
4441         } else if (arp_interval) {
4442                 int i;
4443
4444                 printk(KERN_INFO DRV_NAME
4445                        ": ARP monitoring set to %d ms with %d target(s):",
4446                        arp_interval, arp_ip_count);
4447
4448                 for (i = 0; i < arp_ip_count; i++)
4449                         printk (" %s", arp_ip_target[i]);
4450
4451                 printk("\n");
4452
4453         } else {
4454                 /* miimon and arp_interval not set, we need one so things
4455                  * work as expected, see bonding.txt for details
4456                  */
4457                 printk(KERN_WARNING DRV_NAME
4458                        ": Warning: either miimon or arp_interval and "
4459                        "arp_ip_target module parameters must be specified, "
4460                        "otherwise bonding will not detect link failures! see "
4461                        "bonding.txt for details.\n");
4462         }
4463
4464         if (primary && !USES_PRIMARY(bond_mode)) {
4465                 /* currently, using a primary only makes sense
4466                  * in active backup, TLB or ALB modes
4467                  */
4468                 printk(KERN_WARNING DRV_NAME
4469                        ": Warning: %s primary device specified but has no "
4470                        "effect in %s mode\n",
4471                        primary, bond_mode_name(bond_mode));
4472                 primary = NULL;
4473         }
4474
4475         /* fill params struct with the proper values */
4476         params->mode = bond_mode;
4477         params->xmit_policy = xmit_hashtype;
4478         params->miimon = miimon;
4479         params->arp_interval = arp_interval;
4480         params->updelay = updelay;
4481         params->downdelay = downdelay;
4482         params->use_carrier = use_carrier;
4483         params->lacp_fast = lacp_fast;
4484         params->primary[0] = 0;
4485
4486         if (primary) {
4487                 strncpy(params->primary, primary, IFNAMSIZ);
4488                 params->primary[IFNAMSIZ - 1] = 0;
4489         }
4490
4491         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4492
4493         return 0;
4494 }
4495
4496 /* Create a new bond based on the specified name and bonding parameters.
4497  * Caller must NOT hold rtnl_lock; we need to release it here before we
4498  * set up our sysfs entries.
4499  */
4500 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4501 {
4502         struct net_device *bond_dev;
4503         int res;
4504
4505         rtnl_lock();
4506         bond_dev = alloc_netdev(sizeof(struct bonding), name, ether_setup);
4507         if (!bond_dev) {
4508                 printk(KERN_ERR DRV_NAME
4509                        ": %s: eek! can't alloc netdev!\n",
4510                        name);
4511                 res = -ENOMEM;
4512                 goto out_rtnl;
4513         }
4514
4515         /* bond_init() must be called after dev_alloc_name() (for the
4516          * /proc files), but before register_netdevice(), because we
4517          * need to set function pointers.
4518          */
4519
4520         res = bond_init(bond_dev, params);
4521         if (res < 0) {
4522                 goto out_netdev;
4523         }
4524
4525         SET_MODULE_OWNER(bond_dev);
4526
4527         res = register_netdevice(bond_dev);
4528         if (res < 0) {
4529                 goto out_bond;
4530         }
4531         if (newbond)
4532                 *newbond = bond_dev->priv;
4533
4534         rtnl_unlock(); /* allows sysfs registration of net device */
4535         res = bond_create_sysfs_entry(bond_dev->priv);
4536         goto done;
4537 out_bond:
4538         bond_deinit(bond_dev);
4539 out_netdev:
4540         free_netdev(bond_dev);
4541 out_rtnl:
4542         rtnl_unlock();
4543 done:
4544         return res;
4545 }
4546
4547 static int __init bonding_init(void)
4548 {
4549         int i;
4550         int res;
4551         char new_bond_name[8];  /* Enough room for 999 bonds at init. */
4552
4553         printk(KERN_INFO "%s", version);
4554
4555         res = bond_check_params(&bonding_defaults);
4556         if (res) {
4557                 goto out;
4558         }
4559
4560 #ifdef CONFIG_PROC_FS
4561         bond_create_proc_dir();
4562 #endif
4563         for (i = 0; i < max_bonds; i++) {
4564                 sprintf(new_bond_name, "bond%d",i);
4565                 res = bond_create(new_bond_name,&bonding_defaults, NULL);
4566                 if (res)
4567                         goto err;
4568         }
4569
4570         res = bond_create_sysfs();
4571         if (res)
4572                 goto err;
4573
4574         register_netdevice_notifier(&bond_netdev_notifier);
4575         register_inetaddr_notifier(&bond_inetaddr_notifier);
4576
4577         goto out;
4578 err:
4579         rtnl_lock();
4580         bond_free_all();
4581         bond_destroy_sysfs();
4582         rtnl_unlock();
4583 out:
4584         return res;
4585
4586 }
4587
4588 static void __exit bonding_exit(void)
4589 {
4590         unregister_netdevice_notifier(&bond_netdev_notifier);
4591         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4592
4593         rtnl_lock();
4594         bond_free_all();
4595         bond_destroy_sysfs();
4596         rtnl_unlock();
4597 }
4598
4599 module_init(bonding_init);
4600 module_exit(bonding_exit);
4601 MODULE_LICENSE("GPL");
4602 MODULE_VERSION(DRV_VERSION);
4603 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4604 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4605 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4606
4607 /*
4608  * Local variables:
4609  *  c-indent-level: 8
4610  *  c-basic-offset: 8
4611  *  tab-width: 8
4612  * End:
4613  */
4614