bonding: use ethtool for link checking first
[linux-2.6.git] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
41 #include <linux/in.h>
42 #include <net/ip.h>
43 #include <linux/ip.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
54 #include <asm/system.h>
55 #include <asm/io.h>
56 #include <asm/dma.h>
57 #include <asm/uaccess.h>
58 #include <linux/errno.h>
59 #include <linux/netdevice.h>
60 #include <linux/inetdevice.h>
61 #include <linux/igmp.h>
62 #include <linux/etherdevice.h>
63 #include <linux/skbuff.h>
64 #include <net/sock.h>
65 #include <linux/rtnetlink.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
70 #include <net/arp.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include "bonding.h"
79 #include "bond_3ad.h"
80 #include "bond_alb.h"
81
82 /*---------------------------- Module parameters ----------------------------*/
83
84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
85 #define BOND_LINK_MON_INTERV    0
86 #define BOND_LINK_ARP_INTERV    0
87
88 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
89 static int num_grat_arp = 1;
90 static int num_unsol_na = 1;
91 static int miimon       = BOND_LINK_MON_INTERV;
92 static int updelay      = 0;
93 static int downdelay    = 0;
94 static int use_carrier  = 1;
95 static char *mode       = NULL;
96 static char *primary    = NULL;
97 static char *lacp_rate  = NULL;
98 static char *ad_select  = NULL;
99 static char *xmit_hash_policy = NULL;
100 static int arp_interval = BOND_LINK_ARP_INTERV;
101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
102 static char *arp_validate = NULL;
103 static char *fail_over_mac = NULL;
104 struct bond_params bonding_defaults;
105
106 module_param(max_bonds, int, 0);
107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
108 module_param(num_grat_arp, int, 0644);
109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
110 module_param(num_unsol_na, int, 0644);
111 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
112 module_param(miimon, int, 0);
113 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
114 module_param(updelay, int, 0);
115 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
116 module_param(downdelay, int, 0);
117 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
118                             "in milliseconds");
119 module_param(use_carrier, int, 0);
120 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
121                               "0 for off, 1 for on (default)");
122 module_param(mode, charp, 0);
123 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
124                        "1 for active-backup, 2 for balance-xor, "
125                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
126                        "6 for balance-alb");
127 module_param(primary, charp, 0);
128 MODULE_PARM_DESC(primary, "Primary network device to use");
129 module_param(lacp_rate, charp, 0);
130 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
131                             "(slow/fast)");
132 module_param(ad_select, charp, 0);
133 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
134 module_param(xmit_hash_policy, charp, 0);
135 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
136                                    ", 1 for layer 3+4");
137 module_param(arp_interval, int, 0);
138 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
139 module_param_array(arp_ip_target, charp, NULL, 0);
140 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
141 module_param(arp_validate, charp, 0);
142 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
143 module_param(fail_over_mac, charp, 0);
144 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
145
146 /*----------------------------- Global variables ----------------------------*/
147
148 static const char * const version =
149         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
150
151 LIST_HEAD(bond_dev_list);
152
153 #ifdef CONFIG_PROC_FS
154 static struct proc_dir_entry *bond_proc_dir = NULL;
155 #endif
156
157 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
158 static int arp_ip_count = 0;
159 static int bond_mode    = BOND_MODE_ROUNDROBIN;
160 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
161 static int lacp_fast    = 0;
162
163
164 const struct bond_parm_tbl bond_lacp_tbl[] = {
165 {       "slow",         AD_LACP_SLOW},
166 {       "fast",         AD_LACP_FAST},
167 {       NULL,           -1},
168 };
169
170 const struct bond_parm_tbl bond_mode_tbl[] = {
171 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
172 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
173 {       "balance-xor",          BOND_MODE_XOR},
174 {       "broadcast",            BOND_MODE_BROADCAST},
175 {       "802.3ad",              BOND_MODE_8023AD},
176 {       "balance-tlb",          BOND_MODE_TLB},
177 {       "balance-alb",          BOND_MODE_ALB},
178 {       NULL,                   -1},
179 };
180
181 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
182 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
183 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
184 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
185 {       NULL,                   -1},
186 };
187
188 const struct bond_parm_tbl arp_validate_tbl[] = {
189 {       "none",                 BOND_ARP_VALIDATE_NONE},
190 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
191 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
192 {       "all",                  BOND_ARP_VALIDATE_ALL},
193 {       NULL,                   -1},
194 };
195
196 const struct bond_parm_tbl fail_over_mac_tbl[] = {
197 {       "none",                 BOND_FOM_NONE},
198 {       "active",               BOND_FOM_ACTIVE},
199 {       "follow",               BOND_FOM_FOLLOW},
200 {       NULL,                   -1},
201 };
202
203 struct bond_parm_tbl ad_select_tbl[] = {
204 {       "stable",       BOND_AD_STABLE},
205 {       "bandwidth",    BOND_AD_BANDWIDTH},
206 {       "count",        BOND_AD_COUNT},
207 {       NULL,           -1},
208 };
209
210 /*-------------------------- Forward declarations ---------------------------*/
211
212 static void bond_send_gratuitous_arp(struct bonding *bond);
213 static void bond_deinit(struct net_device *bond_dev);
214
215 /*---------------------------- General routines -----------------------------*/
216
217 static const char *bond_mode_name(int mode)
218 {
219         static const char *names[] = {
220                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222                 [BOND_MODE_XOR] = "load balancing (xor)",
223                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
224                 [BOND_MODE_8023AD]= "IEEE 802.3ad Dynamic link aggregation",
225                 [BOND_MODE_TLB] = "transmit load balancing",
226                 [BOND_MODE_ALB] = "adaptive load balancing",
227         };
228
229         if (mode < 0 || mode > BOND_MODE_ALB)
230                 return "unknown";
231
232         return names[mode];
233 }
234
235 /*---------------------------------- VLAN -----------------------------------*/
236
237 /**
238  * bond_add_vlan - add a new vlan id on bond
239  * @bond: bond that got the notification
240  * @vlan_id: the vlan id to add
241  *
242  * Returns -ENOMEM if allocation failed.
243  */
244 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
245 {
246         struct vlan_entry *vlan;
247
248         pr_debug("bond: %s, vlan id %d\n",
249                 (bond ? bond->dev->name: "None"), vlan_id);
250
251         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
252         if (!vlan) {
253                 return -ENOMEM;
254         }
255
256         INIT_LIST_HEAD(&vlan->vlan_list);
257         vlan->vlan_id = vlan_id;
258
259         write_lock_bh(&bond->lock);
260
261         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
262
263         write_unlock_bh(&bond->lock);
264
265         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
266
267         return 0;
268 }
269
270 /**
271  * bond_del_vlan - delete a vlan id from bond
272  * @bond: bond that got the notification
273  * @vlan_id: the vlan id to delete
274  *
275  * returns -ENODEV if @vlan_id was not found in @bond.
276  */
277 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
278 {
279         struct vlan_entry *vlan;
280         int res = -ENODEV;
281
282         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
283
284         write_lock_bh(&bond->lock);
285
286         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
287                 if (vlan->vlan_id == vlan_id) {
288                         list_del(&vlan->vlan_list);
289
290                         if (bond_is_lb(bond))
291                                 bond_alb_clear_vlan(bond, vlan_id);
292
293                         pr_debug("removed VLAN ID %d from bond %s\n", vlan_id,
294                                 bond->dev->name);
295
296                         kfree(vlan);
297
298                         if (list_empty(&bond->vlan_list) &&
299                             (bond->slave_cnt == 0)) {
300                                 /* Last VLAN removed and no slaves, so
301                                  * restore block on adding VLANs. This will
302                                  * be removed once new slaves that are not
303                                  * VLAN challenged will be added.
304                                  */
305                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
306                         }
307
308                         res = 0;
309                         goto out;
310                 }
311         }
312
313         pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id,
314                 bond->dev->name);
315
316 out:
317         write_unlock_bh(&bond->lock);
318         return res;
319 }
320
321 /**
322  * bond_has_challenged_slaves
323  * @bond: the bond we're working on
324  *
325  * Searches the slave list. Returns 1 if a vlan challenged slave
326  * was found, 0 otherwise.
327  *
328  * Assumes bond->lock is held.
329  */
330 static int bond_has_challenged_slaves(struct bonding *bond)
331 {
332         struct slave *slave;
333         int i;
334
335         bond_for_each_slave(bond, slave, i) {
336                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
337                         pr_debug("found VLAN challenged slave - %s\n",
338                                 slave->dev->name);
339                         return 1;
340                 }
341         }
342
343         pr_debug("no VLAN challenged slaves found\n");
344         return 0;
345 }
346
347 /**
348  * bond_next_vlan - safely skip to the next item in the vlans list.
349  * @bond: the bond we're working on
350  * @curr: item we're advancing from
351  *
352  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
353  * or @curr->next otherwise (even if it is @curr itself again).
354  * 
355  * Caller must hold bond->lock
356  */
357 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
358 {
359         struct vlan_entry *next, *last;
360
361         if (list_empty(&bond->vlan_list)) {
362                 return NULL;
363         }
364
365         if (!curr) {
366                 next = list_entry(bond->vlan_list.next,
367                                   struct vlan_entry, vlan_list);
368         } else {
369                 last = list_entry(bond->vlan_list.prev,
370                                   struct vlan_entry, vlan_list);
371                 if (last == curr) {
372                         next = list_entry(bond->vlan_list.next,
373                                           struct vlan_entry, vlan_list);
374                 } else {
375                         next = list_entry(curr->vlan_list.next,
376                                           struct vlan_entry, vlan_list);
377                 }
378         }
379
380         return next;
381 }
382
383 /**
384  * bond_dev_queue_xmit - Prepare skb for xmit.
385  * 
386  * @bond: bond device that got this skb for tx.
387  * @skb: hw accel VLAN tagged skb to transmit
388  * @slave_dev: slave that is supposed to xmit this skbuff
389  * 
390  * When the bond gets an skb to transmit that is
391  * already hardware accelerated VLAN tagged, and it
392  * needs to relay this skb to a slave that is not
393  * hw accel capable, the skb needs to be "unaccelerated",
394  * i.e. strip the hwaccel tag and re-insert it as part
395  * of the payload.
396  */
397 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
398 {
399         unsigned short uninitialized_var(vlan_id);
400
401         if (!list_empty(&bond->vlan_list) &&
402             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
403             vlan_get_tag(skb, &vlan_id) == 0) {
404                 skb->dev = slave_dev;
405                 skb = vlan_put_tag(skb, vlan_id);
406                 if (!skb) {
407                         /* vlan_put_tag() frees the skb in case of error,
408                          * so return success here so the calling functions
409                          * won't attempt to free is again.
410                          */
411                         return 0;
412                 }
413         } else {
414                 skb->dev = slave_dev;
415         }
416
417         skb->priority = 1;
418         dev_queue_xmit(skb);
419
420         return 0;
421 }
422
423 /*
424  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
425  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
426  * lock because:
427  * a. This operation is performed in IOCTL context,
428  * b. The operation is protected by the RTNL semaphore in the 8021q code,
429  * c. Holding a lock with BH disabled while directly calling a base driver
430  *    entry point is generally a BAD idea.
431  * 
432  * The design of synchronization/protection for this operation in the 8021q
433  * module is good for one or more VLAN devices over a single physical device
434  * and cannot be extended for a teaming solution like bonding, so there is a
435  * potential race condition here where a net device from the vlan group might
436  * be referenced (either by a base driver or the 8021q code) while it is being
437  * removed from the system. However, it turns out we're not making matters
438  * worse, and if it works for regular VLAN usage it will work here too.
439 */
440
441 /**
442  * bond_vlan_rx_register - Propagates registration to slaves
443  * @bond_dev: bonding net device that got called
444  * @grp: vlan group being registered
445  */
446 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
447 {
448         struct bonding *bond = netdev_priv(bond_dev);
449         struct slave *slave;
450         int i;
451
452         bond->vlgrp = grp;
453
454         bond_for_each_slave(bond, slave, i) {
455                 struct net_device *slave_dev = slave->dev;
456                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
457
458                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
459                     slave_ops->ndo_vlan_rx_register) {
460                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
461                 }
462         }
463 }
464
465 /**
466  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
467  * @bond_dev: bonding net device that got called
468  * @vid: vlan id being added
469  */
470 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
471 {
472         struct bonding *bond = netdev_priv(bond_dev);
473         struct slave *slave;
474         int i, res;
475
476         bond_for_each_slave(bond, slave, i) {
477                 struct net_device *slave_dev = slave->dev;
478                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
479
480                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
481                     slave_ops->ndo_vlan_rx_add_vid) {
482                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
483                 }
484         }
485
486         res = bond_add_vlan(bond, vid);
487         if (res) {
488                 printk(KERN_ERR DRV_NAME
489                        ": %s: Error: Failed to add vlan id %d\n",
490                        bond_dev->name, vid);
491         }
492 }
493
494 /**
495  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
496  * @bond_dev: bonding net device that got called
497  * @vid: vlan id being removed
498  */
499 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
500 {
501         struct bonding *bond = netdev_priv(bond_dev);
502         struct slave *slave;
503         struct net_device *vlan_dev;
504         int i, res;
505
506         bond_for_each_slave(bond, slave, i) {
507                 struct net_device *slave_dev = slave->dev;
508                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
509
510                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
511                     slave_ops->ndo_vlan_rx_kill_vid) {
512                         /* Save and then restore vlan_dev in the grp array,
513                          * since the slave's driver might clear it.
514                          */
515                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
516                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
517                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
518                 }
519         }
520
521         res = bond_del_vlan(bond, vid);
522         if (res) {
523                 printk(KERN_ERR DRV_NAME
524                        ": %s: Error: Failed to remove vlan id %d\n",
525                        bond_dev->name, vid);
526         }
527 }
528
529 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
530 {
531         struct vlan_entry *vlan;
532         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
533
534         write_lock_bh(&bond->lock);
535
536         if (list_empty(&bond->vlan_list))
537                 goto out;
538
539         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
540             slave_ops->ndo_vlan_rx_register)
541                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
542
543         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
544             !(slave_ops->ndo_vlan_rx_add_vid))
545                 goto out;
546
547         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
548                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
549
550 out:
551         write_unlock_bh(&bond->lock);
552 }
553
554 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
555 {
556         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
557         struct vlan_entry *vlan;
558         struct net_device *vlan_dev;
559
560         write_lock_bh(&bond->lock);
561
562         if (list_empty(&bond->vlan_list))
563                 goto out;
564
565         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
566             !(slave_ops->ndo_vlan_rx_kill_vid))
567                 goto unreg;
568
569         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
570                 /* Save and then restore vlan_dev in the grp array,
571                  * since the slave's driver might clear it.
572                  */
573                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
574                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
575                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
576         }
577
578 unreg:
579         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
580             slave_ops->ndo_vlan_rx_register)
581                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
582
583 out:
584         write_unlock_bh(&bond->lock);
585 }
586
587 /*------------------------------- Link status -------------------------------*/
588
589 /*
590  * Set the carrier state for the master according to the state of its
591  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
592  * do special 802.3ad magic.
593  *
594  * Returns zero if carrier state does not change, nonzero if it does.
595  */
596 static int bond_set_carrier(struct bonding *bond)
597 {
598         struct slave *slave;
599         int i;
600
601         if (bond->slave_cnt == 0)
602                 goto down;
603
604         if (bond->params.mode == BOND_MODE_8023AD)
605                 return bond_3ad_set_carrier(bond);
606
607         bond_for_each_slave(bond, slave, i) {
608                 if (slave->link == BOND_LINK_UP) {
609                         if (!netif_carrier_ok(bond->dev)) {
610                                 netif_carrier_on(bond->dev);
611                                 return 1;
612                         }
613                         return 0;
614                 }
615         }
616
617 down:
618         if (netif_carrier_ok(bond->dev)) {
619                 netif_carrier_off(bond->dev);
620                 return 1;
621         }
622         return 0;
623 }
624
625 /*
626  * Get link speed and duplex from the slave's base driver
627  * using ethtool. If for some reason the call fails or the
628  * values are invalid, fake speed and duplex to 100/Full
629  * and return error.
630  */
631 static int bond_update_speed_duplex(struct slave *slave)
632 {
633         struct net_device *slave_dev = slave->dev;
634         struct ethtool_cmd etool;
635         int res;
636
637         /* Fake speed and duplex */
638         slave->speed = SPEED_100;
639         slave->duplex = DUPLEX_FULL;
640
641         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
642                 return -1;
643
644         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
645         if (res < 0)
646                 return -1;
647
648         switch (etool.speed) {
649         case SPEED_10:
650         case SPEED_100:
651         case SPEED_1000:
652         case SPEED_10000:
653                 break;
654         default:
655                 return -1;
656         }
657
658         switch (etool.duplex) {
659         case DUPLEX_FULL:
660         case DUPLEX_HALF:
661                 break;
662         default:
663                 return -1;
664         }
665
666         slave->speed = etool.speed;
667         slave->duplex = etool.duplex;
668
669         return 0;
670 }
671
672 /*
673  * if <dev> supports MII link status reporting, check its link status.
674  *
675  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
676  * depening upon the setting of the use_carrier parameter.
677  *
678  * Return either BMSR_LSTATUS, meaning that the link is up (or we
679  * can't tell and just pretend it is), or 0, meaning that the link is
680  * down.
681  *
682  * If reporting is non-zero, instead of faking link up, return -1 if
683  * both ETHTOOL and MII ioctls fail (meaning the device does not
684  * support them).  If use_carrier is set, return whatever it says.
685  * It'd be nice if there was a good way to tell if a driver supports
686  * netif_carrier, but there really isn't.
687  */
688 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
689 {
690         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
691         static int (* ioctl)(struct net_device *, struct ifreq *, int);
692         struct ifreq ifr;
693         struct mii_ioctl_data *mii;
694
695         if (bond->params.use_carrier)
696                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
697
698         /* Try to get link status using Ethtool first. */
699         if (slave_dev->ethtool_ops) {
700                 if (slave_dev->ethtool_ops->get_link) {
701                         u32 link;
702
703                         link = slave_dev->ethtool_ops->get_link(slave_dev);
704
705                         return link ? BMSR_LSTATUS : 0;
706                 }
707         }
708
709         /* Ethtool can't be used, fallback to MII ioclts. */
710         ioctl = slave_ops->ndo_do_ioctl;
711         if (ioctl) {
712                 /* TODO: set pointer to correct ioctl on a per team member */
713                 /*       bases to make this more efficient. that is, once  */
714                 /*       we determine the correct ioctl, we will always    */
715                 /*       call it and not the others for that team          */
716                 /*       member.                                           */
717
718                 /*
719                  * We cannot assume that SIOCGMIIPHY will also read a
720                  * register; not all network drivers (e.g., e100)
721                  * support that.
722                  */
723
724                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
725                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
726                 mii = if_mii(&ifr);
727                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
728                         mii->reg_num = MII_BMSR;
729                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
730                                 return (mii->val_out & BMSR_LSTATUS);
731                         }
732                 }
733         }
734
735         /*
736          * If reporting, report that either there's no dev->do_ioctl,
737          * or both SIOCGMIIREG and get_link failed (meaning that we
738          * cannot report link status).  If not reporting, pretend
739          * we're ok.
740          */
741         return (reporting ? -1 : BMSR_LSTATUS);
742 }
743
744 /*----------------------------- Multicast list ------------------------------*/
745
746 /*
747  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
748  */
749 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
750 {
751         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
752                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
753 }
754
755 /*
756  * returns dmi entry if found, NULL otherwise
757  */
758 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
759 {
760         struct dev_mc_list *idmi;
761
762         for (idmi = mc_list; idmi; idmi = idmi->next) {
763                 if (bond_is_dmi_same(dmi, idmi)) {
764                         return idmi;
765                 }
766         }
767
768         return NULL;
769 }
770
771 /*
772  * Push the promiscuity flag down to appropriate slaves
773  */
774 static int bond_set_promiscuity(struct bonding *bond, int inc)
775 {
776         int err = 0;
777         if (USES_PRIMARY(bond->params.mode)) {
778                 /* write lock already acquired */
779                 if (bond->curr_active_slave) {
780                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
781                                                   inc);
782                 }
783         } else {
784                 struct slave *slave;
785                 int i;
786                 bond_for_each_slave(bond, slave, i) {
787                         err = dev_set_promiscuity(slave->dev, inc);
788                         if (err)
789                                 return err;
790                 }
791         }
792         return err;
793 }
794
795 /*
796  * Push the allmulti flag down to all slaves
797  */
798 static int bond_set_allmulti(struct bonding *bond, int inc)
799 {
800         int err = 0;
801         if (USES_PRIMARY(bond->params.mode)) {
802                 /* write lock already acquired */
803                 if (bond->curr_active_slave) {
804                         err = dev_set_allmulti(bond->curr_active_slave->dev,
805                                                inc);
806                 }
807         } else {
808                 struct slave *slave;
809                 int i;
810                 bond_for_each_slave(bond, slave, i) {
811                         err = dev_set_allmulti(slave->dev, inc);
812                         if (err)
813                                 return err;
814                 }
815         }
816         return err;
817 }
818
819 /*
820  * Add a Multicast address to slaves
821  * according to mode
822  */
823 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
824 {
825         if (USES_PRIMARY(bond->params.mode)) {
826                 /* write lock already acquired */
827                 if (bond->curr_active_slave) {
828                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
829                 }
830         } else {
831                 struct slave *slave;
832                 int i;
833                 bond_for_each_slave(bond, slave, i) {
834                         dev_mc_add(slave->dev, addr, alen, 0);
835                 }
836         }
837 }
838
839 /*
840  * Remove a multicast address from slave
841  * according to mode
842  */
843 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
844 {
845         if (USES_PRIMARY(bond->params.mode)) {
846                 /* write lock already acquired */
847                 if (bond->curr_active_slave) {
848                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
849                 }
850         } else {
851                 struct slave *slave;
852                 int i;
853                 bond_for_each_slave(bond, slave, i) {
854                         dev_mc_delete(slave->dev, addr, alen, 0);
855                 }
856         }
857 }
858
859
860 /*
861  * Retrieve the list of registered multicast addresses for the bonding
862  * device and retransmit an IGMP JOIN request to the current active
863  * slave.
864  */
865 static void bond_resend_igmp_join_requests(struct bonding *bond)
866 {
867         struct in_device *in_dev;
868         struct ip_mc_list *im;
869
870         rcu_read_lock();
871         in_dev = __in_dev_get_rcu(bond->dev);
872         if (in_dev) {
873                 for (im = in_dev->mc_list; im; im = im->next) {
874                         ip_mc_rejoin_group(im);
875                 }
876         }
877
878         rcu_read_unlock();
879 }
880
881 /*
882  * Totally destroys the mc_list in bond
883  */
884 static void bond_mc_list_destroy(struct bonding *bond)
885 {
886         struct dev_mc_list *dmi;
887
888         dmi = bond->mc_list;
889         while (dmi) {
890                 bond->mc_list = dmi->next;
891                 kfree(dmi);
892                 dmi = bond->mc_list;
893         }
894         bond->mc_list = NULL;
895 }
896
897 /*
898  * Copy all the Multicast addresses from src to the bonding device dst
899  */
900 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
901                              gfp_t gfp_flag)
902 {
903         struct dev_mc_list *dmi, *new_dmi;
904
905         for (dmi = mc_list; dmi; dmi = dmi->next) {
906                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
907
908                 if (!new_dmi) {
909                         /* FIXME: Potential memory leak !!! */
910                         return -ENOMEM;
911                 }
912
913                 new_dmi->next = bond->mc_list;
914                 bond->mc_list = new_dmi;
915                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
916                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
917                 new_dmi->dmi_users = dmi->dmi_users;
918                 new_dmi->dmi_gusers = dmi->dmi_gusers;
919         }
920
921         return 0;
922 }
923
924 /*
925  * flush all members of flush->mc_list from device dev->mc_list
926  */
927 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
928 {
929         struct bonding *bond = netdev_priv(bond_dev);
930         struct dev_mc_list *dmi;
931
932         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
933                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
934         }
935
936         if (bond->params.mode == BOND_MODE_8023AD) {
937                 /* del lacpdu mc addr from mc list */
938                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
939
940                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
941         }
942 }
943
944 /*--------------------------- Active slave change ---------------------------*/
945
946 /*
947  * Update the mc list and multicast-related flags for the new and
948  * old active slaves (if any) according to the multicast mode, and
949  * promiscuous flags unconditionally.
950  */
951 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
952 {
953         struct dev_mc_list *dmi;
954
955         if (!USES_PRIMARY(bond->params.mode)) {
956                 /* nothing to do -  mc list is already up-to-date on
957                  * all slaves
958                  */
959                 return;
960         }
961
962         if (old_active) {
963                 if (bond->dev->flags & IFF_PROMISC) {
964                         dev_set_promiscuity(old_active->dev, -1);
965                 }
966
967                 if (bond->dev->flags & IFF_ALLMULTI) {
968                         dev_set_allmulti(old_active->dev, -1);
969                 }
970
971                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
972                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
973                 }
974         }
975
976         if (new_active) {
977                 /* FIXME: Signal errors upstream. */
978                 if (bond->dev->flags & IFF_PROMISC) {
979                         dev_set_promiscuity(new_active->dev, 1);
980                 }
981
982                 if (bond->dev->flags & IFF_ALLMULTI) {
983                         dev_set_allmulti(new_active->dev, 1);
984                 }
985
986                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
987                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
988                 }
989                 bond_resend_igmp_join_requests(bond);
990         }
991 }
992
993 /*
994  * bond_do_fail_over_mac
995  *
996  * Perform special MAC address swapping for fail_over_mac settings
997  *
998  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
999  */
1000 static void bond_do_fail_over_mac(struct bonding *bond,
1001                                   struct slave *new_active,
1002                                   struct slave *old_active)
1003         __releases(&bond->curr_slave_lock)
1004         __releases(&bond->lock)
1005         __acquires(&bond->lock)
1006         __acquires(&bond->curr_slave_lock)
1007 {
1008         u8 tmp_mac[ETH_ALEN];
1009         struct sockaddr saddr;
1010         int rv;
1011
1012         switch (bond->params.fail_over_mac) {
1013         case BOND_FOM_ACTIVE:
1014                 if (new_active)
1015                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1016                                new_active->dev->addr_len);
1017                 break;
1018         case BOND_FOM_FOLLOW:
1019                 /*
1020                  * if new_active && old_active, swap them
1021                  * if just old_active, do nothing (going to no active slave)
1022                  * if just new_active, set new_active to bond's MAC
1023                  */
1024                 if (!new_active)
1025                         return;
1026
1027                 write_unlock_bh(&bond->curr_slave_lock);
1028                 read_unlock(&bond->lock);
1029
1030                 if (old_active) {
1031                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1032                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1033                                ETH_ALEN);
1034                         saddr.sa_family = new_active->dev->type;
1035                 } else {
1036                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1037                         saddr.sa_family = bond->dev->type;
1038                 }
1039
1040                 rv = dev_set_mac_address(new_active->dev, &saddr);
1041                 if (rv) {
1042                         printk(KERN_ERR DRV_NAME
1043                                ": %s: Error %d setting MAC of slave %s\n",
1044                                bond->dev->name, -rv, new_active->dev->name);
1045                         goto out;
1046                 }
1047
1048                 if (!old_active)
1049                         goto out;
1050
1051                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1052                 saddr.sa_family = old_active->dev->type;
1053
1054                 rv = dev_set_mac_address(old_active->dev, &saddr);
1055                 if (rv)
1056                         printk(KERN_ERR DRV_NAME
1057                                ": %s: Error %d setting MAC of slave %s\n",
1058                                bond->dev->name, -rv, new_active->dev->name);
1059 out:
1060                 read_lock(&bond->lock);
1061                 write_lock_bh(&bond->curr_slave_lock);
1062                 break;
1063         default:
1064                 printk(KERN_ERR DRV_NAME
1065                        ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1066                        bond->dev->name, bond->params.fail_over_mac);
1067                 break;
1068         }
1069
1070 }
1071
1072
1073 /**
1074  * find_best_interface - select the best available slave to be the active one
1075  * @bond: our bonding struct
1076  *
1077  * Warning: Caller must hold curr_slave_lock for writing.
1078  */
1079 static struct slave *bond_find_best_slave(struct bonding *bond)
1080 {
1081         struct slave *new_active, *old_active;
1082         struct slave *bestslave = NULL;
1083         int mintime = bond->params.updelay;
1084         int i;
1085
1086         new_active = old_active = bond->curr_active_slave;
1087
1088         if (!new_active) { /* there were no active slaves left */
1089                 if (bond->slave_cnt > 0) {  /* found one slave */
1090                         new_active = bond->first_slave;
1091                 } else {
1092                         return NULL; /* still no slave, return NULL */
1093                 }
1094         }
1095
1096         /* first try the primary link; if arping, a link must tx/rx traffic
1097          * before it can be considered the curr_active_slave - also, we would skip
1098          * slaves between the curr_active_slave and primary_slave that may be up
1099          * and able to arp
1100          */
1101         if ((bond->primary_slave) &&
1102             (!bond->params.arp_interval) &&
1103             (IS_UP(bond->primary_slave->dev))) {
1104                 new_active = bond->primary_slave;
1105         }
1106
1107         /* remember where to stop iterating over the slaves */
1108         old_active = new_active;
1109
1110         bond_for_each_slave_from(bond, new_active, i, old_active) {
1111                 if (IS_UP(new_active->dev)) {
1112                         if (new_active->link == BOND_LINK_UP) {
1113                                 return new_active;
1114                         } else if (new_active->link == BOND_LINK_BACK) {
1115                                 /* link up, but waiting for stabilization */
1116                                 if (new_active->delay < mintime) {
1117                                         mintime = new_active->delay;
1118                                         bestslave = new_active;
1119                                 }
1120                         }
1121                 }
1122         }
1123
1124         return bestslave;
1125 }
1126
1127 /**
1128  * change_active_interface - change the active slave into the specified one
1129  * @bond: our bonding struct
1130  * @new: the new slave to make the active one
1131  *
1132  * Set the new slave to the bond's settings and unset them on the old
1133  * curr_active_slave.
1134  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1135  *
1136  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1137  * because it is apparently the best available slave we have, even though its
1138  * updelay hasn't timed out yet.
1139  *
1140  * If new_active is not NULL, caller must hold bond->lock for read and
1141  * curr_slave_lock for write_bh.
1142  */
1143 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1144 {
1145         struct slave *old_active = bond->curr_active_slave;
1146
1147         if (old_active == new_active) {
1148                 return;
1149         }
1150
1151         if (new_active) {
1152                 new_active->jiffies = jiffies;
1153
1154                 if (new_active->link == BOND_LINK_BACK) {
1155                         if (USES_PRIMARY(bond->params.mode)) {
1156                                 printk(KERN_INFO DRV_NAME
1157                                        ": %s: making interface %s the new "
1158                                        "active one %d ms earlier.\n",
1159                                        bond->dev->name, new_active->dev->name,
1160                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1161                         }
1162
1163                         new_active->delay = 0;
1164                         new_active->link = BOND_LINK_UP;
1165
1166                         if (bond->params.mode == BOND_MODE_8023AD) {
1167                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1168                         }
1169
1170                         if (bond_is_lb(bond))
1171                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1172                 } else {
1173                         if (USES_PRIMARY(bond->params.mode)) {
1174                                 printk(KERN_INFO DRV_NAME
1175                                        ": %s: making interface %s the new "
1176                                        "active one.\n",
1177                                        bond->dev->name, new_active->dev->name);
1178                         }
1179                 }
1180         }
1181
1182         if (USES_PRIMARY(bond->params.mode)) {
1183                 bond_mc_swap(bond, new_active, old_active);
1184         }
1185
1186         if (bond_is_lb(bond)) {
1187                 bond_alb_handle_active_change(bond, new_active);
1188                 if (old_active)
1189                         bond_set_slave_inactive_flags(old_active);
1190                 if (new_active)
1191                         bond_set_slave_active_flags(new_active);
1192         } else {
1193                 bond->curr_active_slave = new_active;
1194         }
1195
1196         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1197                 if (old_active) {
1198                         bond_set_slave_inactive_flags(old_active);
1199                 }
1200
1201                 if (new_active) {
1202                         bond_set_slave_active_flags(new_active);
1203
1204                         if (bond->params.fail_over_mac)
1205                                 bond_do_fail_over_mac(bond, new_active,
1206                                                       old_active);
1207
1208                         bond->send_grat_arp = bond->params.num_grat_arp;
1209                         bond_send_gratuitous_arp(bond);
1210
1211                         bond->send_unsol_na = bond->params.num_unsol_na;
1212                         bond_send_unsolicited_na(bond);
1213
1214                         write_unlock_bh(&bond->curr_slave_lock);
1215                         read_unlock(&bond->lock);
1216
1217                         netdev_bonding_change(bond->dev);
1218
1219                         read_lock(&bond->lock);
1220                         write_lock_bh(&bond->curr_slave_lock);
1221                 }
1222         }
1223 }
1224
1225 /**
1226  * bond_select_active_slave - select a new active slave, if needed
1227  * @bond: our bonding struct
1228  *
1229  * This functions shoud be called when one of the following occurs:
1230  * - The old curr_active_slave has been released or lost its link.
1231  * - The primary_slave has got its link back.
1232  * - A slave has got its link back and there's no old curr_active_slave.
1233  *
1234  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1235  */
1236 void bond_select_active_slave(struct bonding *bond)
1237 {
1238         struct slave *best_slave;
1239         int rv;
1240
1241         best_slave = bond_find_best_slave(bond);
1242         if (best_slave != bond->curr_active_slave) {
1243                 bond_change_active_slave(bond, best_slave);
1244                 rv = bond_set_carrier(bond);
1245                 if (!rv)
1246                         return;
1247
1248                 if (netif_carrier_ok(bond->dev)) {
1249                         printk(KERN_INFO DRV_NAME
1250                                ": %s: first active interface up!\n",
1251                                bond->dev->name);
1252                 } else {
1253                         printk(KERN_INFO DRV_NAME ": %s: "
1254                                "now running without any active interface !\n",
1255                                bond->dev->name);
1256                 }
1257         }
1258 }
1259
1260 /*--------------------------- slave list handling ---------------------------*/
1261
1262 /*
1263  * This function attaches the slave to the end of list.
1264  *
1265  * bond->lock held for writing by caller.
1266  */
1267 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1268 {
1269         if (bond->first_slave == NULL) { /* attaching the first slave */
1270                 new_slave->next = new_slave;
1271                 new_slave->prev = new_slave;
1272                 bond->first_slave = new_slave;
1273         } else {
1274                 new_slave->next = bond->first_slave;
1275                 new_slave->prev = bond->first_slave->prev;
1276                 new_slave->next->prev = new_slave;
1277                 new_slave->prev->next = new_slave;
1278         }
1279
1280         bond->slave_cnt++;
1281 }
1282
1283 /*
1284  * This function detaches the slave from the list.
1285  * WARNING: no check is made to verify if the slave effectively
1286  * belongs to <bond>.
1287  * Nothing is freed on return, structures are just unchained.
1288  * If any slave pointer in bond was pointing to <slave>,
1289  * it should be changed by the calling function.
1290  *
1291  * bond->lock held for writing by caller.
1292  */
1293 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1294 {
1295         if (slave->next) {
1296                 slave->next->prev = slave->prev;
1297         }
1298
1299         if (slave->prev) {
1300                 slave->prev->next = slave->next;
1301         }
1302
1303         if (bond->first_slave == slave) { /* slave is the first slave */
1304                 if (bond->slave_cnt > 1) { /* there are more slave */
1305                         bond->first_slave = slave->next;
1306                 } else {
1307                         bond->first_slave = NULL; /* slave was the last one */
1308                 }
1309         }
1310
1311         slave->next = NULL;
1312         slave->prev = NULL;
1313         bond->slave_cnt--;
1314 }
1315
1316 /*---------------------------------- IOCTL ----------------------------------*/
1317
1318 static int bond_sethwaddr(struct net_device *bond_dev,
1319                           struct net_device *slave_dev)
1320 {
1321         pr_debug("bond_dev=%p\n", bond_dev);
1322         pr_debug("slave_dev=%p\n", slave_dev);
1323         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1324         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1325         return 0;
1326 }
1327
1328 #define BOND_VLAN_FEATURES \
1329         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1330          NETIF_F_HW_VLAN_FILTER)
1331
1332 /* 
1333  * Compute the common dev->feature set available to all slaves.  Some
1334  * feature bits are managed elsewhere, so preserve those feature bits
1335  * on the master device.
1336  */
1337 static int bond_compute_features(struct bonding *bond)
1338 {
1339         struct slave *slave;
1340         struct net_device *bond_dev = bond->dev;
1341         unsigned long features = bond_dev->features;
1342         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1343                                                 bond_dev->hard_header_len);
1344         int i;
1345
1346         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1347         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1348
1349         if (!bond->first_slave)
1350                 goto done;
1351
1352         features &= ~NETIF_F_ONE_FOR_ALL;
1353
1354         bond_for_each_slave(bond, slave, i) {
1355                 features = netdev_increment_features(features,
1356                                                      slave->dev->features,
1357                                                      NETIF_F_ONE_FOR_ALL);
1358                 if (slave->dev->hard_header_len > max_hard_header_len)
1359                         max_hard_header_len = slave->dev->hard_header_len;
1360         }
1361
1362 done:
1363         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1364         bond_dev->features = netdev_fix_features(features, NULL);
1365         bond_dev->hard_header_len = max_hard_header_len;
1366
1367         return 0;
1368 }
1369
1370 static void bond_setup_by_slave(struct net_device *bond_dev,
1371                                 struct net_device *slave_dev)
1372 {
1373         struct bonding *bond = netdev_priv(bond_dev);
1374
1375         bond_dev->header_ops        = slave_dev->header_ops;
1376
1377         bond_dev->type              = slave_dev->type;
1378         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1379         bond_dev->addr_len          = slave_dev->addr_len;
1380
1381         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1382                 slave_dev->addr_len);
1383         bond->setup_by_slave = 1;
1384 }
1385
1386 /* enslave device <slave> to bond device <master> */
1387 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1388 {
1389         struct bonding *bond = netdev_priv(bond_dev);
1390         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1391         struct slave *new_slave = NULL;
1392         struct dev_mc_list *dmi;
1393         struct sockaddr addr;
1394         int link_reporting;
1395         int old_features = bond_dev->features;
1396         int res = 0;
1397
1398         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1399                 slave_ops->ndo_do_ioctl == NULL) {
1400                 printk(KERN_WARNING DRV_NAME
1401                        ": %s: Warning: no link monitoring support for %s\n",
1402                        bond_dev->name, slave_dev->name);
1403         }
1404
1405         /* bond must be initialized by bond_open() before enslaving */
1406         if (!(bond_dev->flags & IFF_UP)) {
1407                 printk(KERN_WARNING DRV_NAME
1408                         " %s: master_dev is not up in bond_enslave\n",
1409                         bond_dev->name);
1410         }
1411
1412         /* already enslaved */
1413         if (slave_dev->flags & IFF_SLAVE) {
1414                 pr_debug("Error, Device was already enslaved\n");
1415                 return -EBUSY;
1416         }
1417
1418         /* vlan challenged mutual exclusion */
1419         /* no need to lock since we're protected by rtnl_lock */
1420         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1421                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1422                 if (!list_empty(&bond->vlan_list)) {
1423                         printk(KERN_ERR DRV_NAME
1424                                ": %s: Error: cannot enslave VLAN "
1425                                "challenged slave %s on VLAN enabled "
1426                                "bond %s\n", bond_dev->name, slave_dev->name,
1427                                bond_dev->name);
1428                         return -EPERM;
1429                 } else {
1430                         printk(KERN_WARNING DRV_NAME
1431                                ": %s: Warning: enslaved VLAN challenged "
1432                                "slave %s. Adding VLANs will be blocked as "
1433                                "long as %s is part of bond %s\n",
1434                                bond_dev->name, slave_dev->name, slave_dev->name,
1435                                bond_dev->name);
1436                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1437                 }
1438         } else {
1439                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1440                 if (bond->slave_cnt == 0) {
1441                         /* First slave, and it is not VLAN challenged,
1442                          * so remove the block of adding VLANs over the bond.
1443                          */
1444                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1445                 }
1446         }
1447
1448         /*
1449          * Old ifenslave binaries are no longer supported.  These can
1450          * be identified with moderate accurary by the state of the slave:
1451          * the current ifenslave will set the interface down prior to
1452          * enslaving it; the old ifenslave will not.
1453          */
1454         if ((slave_dev->flags & IFF_UP)) {
1455                 printk(KERN_ERR DRV_NAME ": %s is up. "
1456                        "This may be due to an out of date ifenslave.\n",
1457                        slave_dev->name);
1458                 res = -EPERM;
1459                 goto err_undo_flags;
1460         }
1461
1462         /* set bonding device ether type by slave - bonding netdevices are
1463          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1464          * there is a need to override some of the type dependent attribs/funcs.
1465          *
1466          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1467          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1468          */
1469         if (bond->slave_cnt == 0) {
1470                 if (slave_dev->type != ARPHRD_ETHER)
1471                         bond_setup_by_slave(bond_dev, slave_dev);
1472         } else if (bond_dev->type != slave_dev->type) {
1473                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1474                         "from other slaves (%d), can not enslave it.\n",
1475                         slave_dev->name,
1476                         slave_dev->type, bond_dev->type);
1477                         res = -EINVAL;
1478                         goto err_undo_flags;
1479         }
1480
1481         if (slave_ops->ndo_set_mac_address == NULL) {
1482                 if (bond->slave_cnt == 0) {
1483                         printk(KERN_WARNING DRV_NAME
1484                                ": %s: Warning: The first slave device "
1485                                "specified does not support setting the MAC "
1486                                "address. Setting fail_over_mac to active.",
1487                                bond_dev->name);
1488                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1489                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1490                         printk(KERN_ERR DRV_NAME
1491                                 ": %s: Error: The slave device specified "
1492                                 "does not support setting the MAC address, "
1493                                 "but fail_over_mac is not set to active.\n"
1494                                 , bond_dev->name);
1495                         res = -EOPNOTSUPP;
1496                         goto err_undo_flags;
1497                 }
1498         }
1499
1500         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1501         if (!new_slave) {
1502                 res = -ENOMEM;
1503                 goto err_undo_flags;
1504         }
1505
1506         /* save slave's original flags before calling
1507          * netdev_set_master and dev_open
1508          */
1509         new_slave->original_flags = slave_dev->flags;
1510
1511         /*
1512          * Save slave's original ("permanent") mac address for modes
1513          * that need it, and for restoring it upon release, and then
1514          * set it to the master's address
1515          */
1516         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1517
1518         if (!bond->params.fail_over_mac) {
1519                 /*
1520                  * Set slave to master's mac address.  The application already
1521                  * set the master's mac address to that of the first slave
1522                  */
1523                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1524                 addr.sa_family = slave_dev->type;
1525                 res = dev_set_mac_address(slave_dev, &addr);
1526                 if (res) {
1527                         pr_debug("Error %d calling set_mac_address\n", res);
1528                         goto err_free;
1529                 }
1530         }
1531
1532         res = netdev_set_master(slave_dev, bond_dev);
1533         if (res) {
1534                 pr_debug("Error %d calling netdev_set_master\n", res);
1535                 goto err_restore_mac;
1536         }
1537         /* open the slave since the application closed it */
1538         res = dev_open(slave_dev);
1539         if (res) {
1540                 pr_debug("Openning slave %s failed\n", slave_dev->name);
1541                 goto err_unset_master;
1542         }
1543
1544         new_slave->dev = slave_dev;
1545         slave_dev->priv_flags |= IFF_BONDING;
1546
1547         if (bond_is_lb(bond)) {
1548                 /* bond_alb_init_slave() must be called before all other stages since
1549                  * it might fail and we do not want to have to undo everything
1550                  */
1551                 res = bond_alb_init_slave(bond, new_slave);
1552                 if (res) {
1553                         goto err_close;
1554                 }
1555         }
1556
1557         /* If the mode USES_PRIMARY, then the new slave gets the
1558          * master's promisc (and mc) settings only if it becomes the
1559          * curr_active_slave, and that is taken care of later when calling
1560          * bond_change_active()
1561          */
1562         if (!USES_PRIMARY(bond->params.mode)) {
1563                 /* set promiscuity level to new slave */
1564                 if (bond_dev->flags & IFF_PROMISC) {
1565                         res = dev_set_promiscuity(slave_dev, 1);
1566                         if (res)
1567                                 goto err_close;
1568                 }
1569
1570                 /* set allmulti level to new slave */
1571                 if (bond_dev->flags & IFF_ALLMULTI) {
1572                         res = dev_set_allmulti(slave_dev, 1);
1573                         if (res)
1574                                 goto err_close;
1575                 }
1576
1577                 netif_addr_lock_bh(bond_dev);
1578                 /* upload master's mc_list to new slave */
1579                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1580                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1581                 }
1582                 netif_addr_unlock_bh(bond_dev);
1583         }
1584
1585         if (bond->params.mode == BOND_MODE_8023AD) {
1586                 /* add lacpdu mc addr to mc list */
1587                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1588
1589                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1590         }
1591
1592         bond_add_vlans_on_slave(bond, slave_dev);
1593
1594         write_lock_bh(&bond->lock);
1595
1596         bond_attach_slave(bond, new_slave);
1597
1598         new_slave->delay = 0;
1599         new_slave->link_failure_count = 0;
1600
1601         bond_compute_features(bond);
1602
1603         write_unlock_bh(&bond->lock);
1604
1605         read_lock(&bond->lock);
1606
1607         new_slave->last_arp_rx = jiffies;
1608
1609         if (bond->params.miimon && !bond->params.use_carrier) {
1610                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1611
1612                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1613                         /*
1614                          * miimon is set but a bonded network driver
1615                          * does not support ETHTOOL/MII and
1616                          * arp_interval is not set.  Note: if
1617                          * use_carrier is enabled, we will never go
1618                          * here (because netif_carrier is always
1619                          * supported); thus, we don't need to change
1620                          * the messages for netif_carrier.
1621                          */
1622                         printk(KERN_WARNING DRV_NAME
1623                                ": %s: Warning: MII and ETHTOOL support not "
1624                                "available for interface %s, and "
1625                                "arp_interval/arp_ip_target module parameters "
1626                                "not specified, thus bonding will not detect "
1627                                "link failures! see bonding.txt for details.\n",
1628                                bond_dev->name, slave_dev->name);
1629                 } else if (link_reporting == -1) {
1630                         /* unable get link status using mii/ethtool */
1631                         printk(KERN_WARNING DRV_NAME
1632                                ": %s: Warning: can't get link status from "
1633                                "interface %s; the network driver associated "
1634                                "with this interface does not support MII or "
1635                                "ETHTOOL link status reporting, thus miimon "
1636                                "has no effect on this interface.\n",
1637                                bond_dev->name, slave_dev->name);
1638                 }
1639         }
1640
1641         /* check for initial state */
1642         if (!bond->params.miimon ||
1643             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1644                 if (bond->params.updelay) {
1645                         pr_debug("Initial state of slave_dev is "
1646                                 "BOND_LINK_BACK\n");
1647                         new_slave->link  = BOND_LINK_BACK;
1648                         new_slave->delay = bond->params.updelay;
1649                 } else {
1650                         pr_debug("Initial state of slave_dev is "
1651                                 "BOND_LINK_UP\n");
1652                         new_slave->link  = BOND_LINK_UP;
1653                 }
1654                 new_slave->jiffies = jiffies;
1655         } else {
1656                 pr_debug("Initial state of slave_dev is "
1657                         "BOND_LINK_DOWN\n");
1658                 new_slave->link  = BOND_LINK_DOWN;
1659         }
1660
1661         if (bond_update_speed_duplex(new_slave) &&
1662             (new_slave->link != BOND_LINK_DOWN)) {
1663                 printk(KERN_WARNING DRV_NAME
1664                        ": %s: Warning: failed to get speed and duplex from %s, "
1665                        "assumed to be 100Mb/sec and Full.\n",
1666                        bond_dev->name, new_slave->dev->name);
1667
1668                 if (bond->params.mode == BOND_MODE_8023AD) {
1669                         printk(KERN_WARNING DRV_NAME
1670                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1671                                "support in base driver for proper aggregator "
1672                                "selection.\n", bond_dev->name);
1673                 }
1674         }
1675
1676         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1677                 /* if there is a primary slave, remember it */
1678                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1679                         bond->primary_slave = new_slave;
1680                 }
1681         }
1682
1683         write_lock_bh(&bond->curr_slave_lock);
1684
1685         switch (bond->params.mode) {
1686         case BOND_MODE_ACTIVEBACKUP:
1687                 bond_set_slave_inactive_flags(new_slave);
1688                 bond_select_active_slave(bond);
1689                 break;
1690         case BOND_MODE_8023AD:
1691                 /* in 802.3ad mode, the internal mechanism
1692                  * will activate the slaves in the selected
1693                  * aggregator
1694                  */
1695                 bond_set_slave_inactive_flags(new_slave);
1696                 /* if this is the first slave */
1697                 if (bond->slave_cnt == 1) {
1698                         SLAVE_AD_INFO(new_slave).id = 1;
1699                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1700                          * can be called only after the mac address of the bond is set
1701                          */
1702                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1703                                             bond->params.lacp_fast);
1704                 } else {
1705                         SLAVE_AD_INFO(new_slave).id =
1706                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1707                 }
1708
1709                 bond_3ad_bind_slave(new_slave);
1710                 break;
1711         case BOND_MODE_TLB:
1712         case BOND_MODE_ALB:
1713                 new_slave->state = BOND_STATE_ACTIVE;
1714                 bond_set_slave_inactive_flags(new_slave);
1715                 bond_select_active_slave(bond);
1716                 break;
1717         default:
1718                 pr_debug("This slave is always active in trunk mode\n");
1719
1720                 /* always active in trunk mode */
1721                 new_slave->state = BOND_STATE_ACTIVE;
1722
1723                 /* In trunking mode there is little meaning to curr_active_slave
1724                  * anyway (it holds no special properties of the bond device),
1725                  * so we can change it without calling change_active_interface()
1726                  */
1727                 if (!bond->curr_active_slave) {
1728                         bond->curr_active_slave = new_slave;
1729                 }
1730                 break;
1731         } /* switch(bond_mode) */
1732
1733         write_unlock_bh(&bond->curr_slave_lock);
1734
1735         bond_set_carrier(bond);
1736
1737         read_unlock(&bond->lock);
1738
1739         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1740         if (res)
1741                 goto err_close;
1742
1743         printk(KERN_INFO DRV_NAME
1744                ": %s: enslaving %s as a%s interface with a%s link.\n",
1745                bond_dev->name, slave_dev->name,
1746                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1747                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1748
1749         /* enslave is successful */
1750         return 0;
1751
1752 /* Undo stages on error */
1753 err_close:
1754         dev_close(slave_dev);
1755
1756 err_unset_master:
1757         netdev_set_master(slave_dev, NULL);
1758
1759 err_restore_mac:
1760         if (!bond->params.fail_over_mac) {
1761                 /* XXX TODO - fom follow mode needs to change master's
1762                  * MAC if this slave's MAC is in use by the bond, or at
1763                  * least print a warning.
1764                  */
1765                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1766                 addr.sa_family = slave_dev->type;
1767                 dev_set_mac_address(slave_dev, &addr);
1768         }
1769
1770 err_free:
1771         kfree(new_slave);
1772
1773 err_undo_flags:
1774         bond_dev->features = old_features;
1775  
1776         return res;
1777 }
1778
1779 /*
1780  * Try to release the slave device <slave> from the bond device <master>
1781  * It is legal to access curr_active_slave without a lock because all the function
1782  * is write-locked.
1783  *
1784  * The rules for slave state should be:
1785  *   for Active/Backup:
1786  *     Active stays on all backups go down
1787  *   for Bonded connections:
1788  *     The first up interface should be left on and all others downed.
1789  */
1790 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1791 {
1792         struct bonding *bond = netdev_priv(bond_dev);
1793         struct slave *slave, *oldcurrent;
1794         struct sockaddr addr;
1795         int mac_addr_differ;
1796
1797         /* slave is not a slave or master is not master of this slave */
1798         if (!(slave_dev->flags & IFF_SLAVE) ||
1799             (slave_dev->master != bond_dev)) {
1800                 printk(KERN_ERR DRV_NAME
1801                        ": %s: Error: cannot release %s.\n",
1802                        bond_dev->name, slave_dev->name);
1803                 return -EINVAL;
1804         }
1805
1806         write_lock_bh(&bond->lock);
1807
1808         slave = bond_get_slave_by_dev(bond, slave_dev);
1809         if (!slave) {
1810                 /* not a slave of this bond */
1811                 printk(KERN_INFO DRV_NAME
1812                        ": %s: %s not enslaved\n",
1813                        bond_dev->name, slave_dev->name);
1814                 write_unlock_bh(&bond->lock);
1815                 return -EINVAL;
1816         }
1817
1818         if (!bond->params.fail_over_mac) {
1819                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1820                                          ETH_ALEN);
1821                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1822                         printk(KERN_WARNING DRV_NAME
1823                                ": %s: Warning: the permanent HWaddr of %s - "
1824                                "%pM - is still in use by %s. "
1825                                "Set the HWaddr of %s to a different address "
1826                                "to avoid conflicts.\n",
1827                                bond_dev->name, slave_dev->name,
1828                                slave->perm_hwaddr,
1829                                bond_dev->name, slave_dev->name);
1830         }
1831
1832         /* Inform AD package of unbinding of slave. */
1833         if (bond->params.mode == BOND_MODE_8023AD) {
1834                 /* must be called before the slave is
1835                  * detached from the list
1836                  */
1837                 bond_3ad_unbind_slave(slave);
1838         }
1839
1840         printk(KERN_INFO DRV_NAME
1841                ": %s: releasing %s interface %s\n",
1842                bond_dev->name,
1843                (slave->state == BOND_STATE_ACTIVE)
1844                ? "active" : "backup",
1845                slave_dev->name);
1846
1847         oldcurrent = bond->curr_active_slave;
1848
1849         bond->current_arp_slave = NULL;
1850
1851         /* release the slave from its bond */
1852         bond_detach_slave(bond, slave);
1853
1854         bond_compute_features(bond);
1855
1856         if (bond->primary_slave == slave) {
1857                 bond->primary_slave = NULL;
1858         }
1859
1860         if (oldcurrent == slave) {
1861                 bond_change_active_slave(bond, NULL);
1862         }
1863
1864         if (bond_is_lb(bond)) {
1865                 /* Must be called only after the slave has been
1866                  * detached from the list and the curr_active_slave
1867                  * has been cleared (if our_slave == old_current),
1868                  * but before a new active slave is selected.
1869                  */
1870                 write_unlock_bh(&bond->lock);
1871                 bond_alb_deinit_slave(bond, slave);
1872                 write_lock_bh(&bond->lock);
1873         }
1874
1875         if (oldcurrent == slave) {
1876                 /*
1877                  * Note that we hold RTNL over this sequence, so there
1878                  * is no concern that another slave add/remove event
1879                  * will interfere.
1880                  */
1881                 write_unlock_bh(&bond->lock);
1882                 read_lock(&bond->lock);
1883                 write_lock_bh(&bond->curr_slave_lock);
1884
1885                 bond_select_active_slave(bond);
1886
1887                 write_unlock_bh(&bond->curr_slave_lock);
1888                 read_unlock(&bond->lock);
1889                 write_lock_bh(&bond->lock);
1890         }
1891
1892         if (bond->slave_cnt == 0) {
1893                 bond_set_carrier(bond);
1894
1895                 /* if the last slave was removed, zero the mac address
1896                  * of the master so it will be set by the application
1897                  * to the mac address of the first slave
1898                  */
1899                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1900
1901                 if (list_empty(&bond->vlan_list)) {
1902                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1903                 } else {
1904                         printk(KERN_WARNING DRV_NAME
1905                                ": %s: Warning: clearing HW address of %s while it "
1906                                "still has VLANs.\n",
1907                                bond_dev->name, bond_dev->name);
1908                         printk(KERN_WARNING DRV_NAME
1909                                ": %s: When re-adding slaves, make sure the bond's "
1910                                "HW address matches its VLANs'.\n",
1911                                bond_dev->name);
1912                 }
1913         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1914                    !bond_has_challenged_slaves(bond)) {
1915                 printk(KERN_INFO DRV_NAME
1916                        ": %s: last VLAN challenged slave %s "
1917                        "left bond %s. VLAN blocking is removed\n",
1918                        bond_dev->name, slave_dev->name, bond_dev->name);
1919                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1920         }
1921
1922         write_unlock_bh(&bond->lock);
1923
1924         /* must do this from outside any spinlocks */
1925         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1926
1927         bond_del_vlans_from_slave(bond, slave_dev);
1928
1929         /* If the mode USES_PRIMARY, then we should only remove its
1930          * promisc and mc settings if it was the curr_active_slave, but that was
1931          * already taken care of above when we detached the slave
1932          */
1933         if (!USES_PRIMARY(bond->params.mode)) {
1934                 /* unset promiscuity level from slave */
1935                 if (bond_dev->flags & IFF_PROMISC) {
1936                         dev_set_promiscuity(slave_dev, -1);
1937                 }
1938
1939                 /* unset allmulti level from slave */
1940                 if (bond_dev->flags & IFF_ALLMULTI) {
1941                         dev_set_allmulti(slave_dev, -1);
1942                 }
1943
1944                 /* flush master's mc_list from slave */
1945                 netif_addr_lock_bh(bond_dev);
1946                 bond_mc_list_flush(bond_dev, slave_dev);
1947                 netif_addr_unlock_bh(bond_dev);
1948         }
1949
1950         netdev_set_master(slave_dev, NULL);
1951
1952         /* close slave before restoring its mac address */
1953         dev_close(slave_dev);
1954
1955         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1956                 /* restore original ("permanent") mac address */
1957                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1958                 addr.sa_family = slave_dev->type;
1959                 dev_set_mac_address(slave_dev, &addr);
1960         }
1961
1962         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1963                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1964                                    IFF_SLAVE_NEEDARP);
1965
1966         kfree(slave);
1967
1968         return 0;  /* deletion OK */
1969 }
1970
1971 /*
1972 * Destroy a bonding device.
1973 * Must be under rtnl_lock when this function is called.
1974 */
1975 void bond_destroy(struct bonding *bond)
1976 {
1977         bond_deinit(bond->dev);
1978         bond_destroy_sysfs_entry(bond);
1979         unregister_netdevice(bond->dev);
1980 }
1981
1982 static void bond_destructor(struct net_device *bond_dev)
1983 {
1984         struct bonding *bond = netdev_priv(bond_dev);
1985
1986         if (bond->wq)
1987                 destroy_workqueue(bond->wq);
1988
1989         netif_addr_lock_bh(bond_dev);
1990         bond_mc_list_destroy(bond);
1991         netif_addr_unlock_bh(bond_dev);
1992
1993         free_netdev(bond_dev);
1994 }
1995
1996 /*
1997 * First release a slave and than destroy the bond if no more slaves iare left.
1998 * Must be under rtnl_lock when this function is called.
1999 */
2000 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2001 {
2002         struct bonding *bond = netdev_priv(bond_dev);
2003         int ret;
2004
2005         ret = bond_release(bond_dev, slave_dev);
2006         if ((ret == 0) && (bond->slave_cnt == 0)) {
2007                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2008                        bond_dev->name, bond_dev->name);
2009                 bond_destroy(bond);
2010         }
2011         return ret;
2012 }
2013
2014 /*
2015  * This function releases all slaves.
2016  */
2017 static int bond_release_all(struct net_device *bond_dev)
2018 {
2019         struct bonding *bond = netdev_priv(bond_dev);
2020         struct slave *slave;
2021         struct net_device *slave_dev;
2022         struct sockaddr addr;
2023
2024         write_lock_bh(&bond->lock);
2025
2026         netif_carrier_off(bond_dev);
2027
2028         if (bond->slave_cnt == 0) {
2029                 goto out;
2030         }
2031
2032         bond->current_arp_slave = NULL;
2033         bond->primary_slave = NULL;
2034         bond_change_active_slave(bond, NULL);
2035
2036         while ((slave = bond->first_slave) != NULL) {
2037                 /* Inform AD package of unbinding of slave
2038                  * before slave is detached from the list.
2039                  */
2040                 if (bond->params.mode == BOND_MODE_8023AD) {
2041                         bond_3ad_unbind_slave(slave);
2042                 }
2043
2044                 slave_dev = slave->dev;
2045                 bond_detach_slave(bond, slave);
2046
2047                 /* now that the slave is detached, unlock and perform
2048                  * all the undo steps that should not be called from
2049                  * within a lock.
2050                  */
2051                 write_unlock_bh(&bond->lock);
2052
2053                 if (bond_is_lb(bond)) {
2054                         /* must be called only after the slave
2055                          * has been detached from the list
2056                          */
2057                         bond_alb_deinit_slave(bond, slave);
2058                 }
2059
2060                 bond_compute_features(bond);
2061
2062                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2063                 bond_del_vlans_from_slave(bond, slave_dev);
2064
2065                 /* If the mode USES_PRIMARY, then we should only remove its
2066                  * promisc and mc settings if it was the curr_active_slave, but that was
2067                  * already taken care of above when we detached the slave
2068                  */
2069                 if (!USES_PRIMARY(bond->params.mode)) {
2070                         /* unset promiscuity level from slave */
2071                         if (bond_dev->flags & IFF_PROMISC) {
2072                                 dev_set_promiscuity(slave_dev, -1);
2073                         }
2074
2075                         /* unset allmulti level from slave */
2076                         if (bond_dev->flags & IFF_ALLMULTI) {
2077                                 dev_set_allmulti(slave_dev, -1);
2078                         }
2079
2080                         /* flush master's mc_list from slave */
2081                         netif_addr_lock_bh(bond_dev);
2082                         bond_mc_list_flush(bond_dev, slave_dev);
2083                         netif_addr_unlock_bh(bond_dev);
2084                 }
2085
2086                 netdev_set_master(slave_dev, NULL);
2087
2088                 /* close slave before restoring its mac address */
2089                 dev_close(slave_dev);
2090
2091                 if (!bond->params.fail_over_mac) {
2092                         /* restore original ("permanent") mac address*/
2093                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2094                         addr.sa_family = slave_dev->type;
2095                         dev_set_mac_address(slave_dev, &addr);
2096                 }
2097
2098                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2099                                            IFF_SLAVE_INACTIVE);
2100
2101                 kfree(slave);
2102
2103                 /* re-acquire the lock before getting the next slave */
2104                 write_lock_bh(&bond->lock);
2105         }
2106
2107         /* zero the mac address of the master so it will be
2108          * set by the application to the mac address of the
2109          * first slave
2110          */
2111         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2112
2113         if (list_empty(&bond->vlan_list)) {
2114                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2115         } else {
2116                 printk(KERN_WARNING DRV_NAME
2117                        ": %s: Warning: clearing HW address of %s while it "
2118                        "still has VLANs.\n",
2119                        bond_dev->name, bond_dev->name);
2120                 printk(KERN_WARNING DRV_NAME
2121                        ": %s: When re-adding slaves, make sure the bond's "
2122                        "HW address matches its VLANs'.\n",
2123                        bond_dev->name);
2124         }
2125
2126         printk(KERN_INFO DRV_NAME
2127                ": %s: released all slaves\n",
2128                bond_dev->name);
2129
2130 out:
2131         write_unlock_bh(&bond->lock);
2132
2133         return 0;
2134 }
2135
2136 /*
2137  * This function changes the active slave to slave <slave_dev>.
2138  * It returns -EINVAL in the following cases.
2139  *  - <slave_dev> is not found in the list.
2140  *  - There is not active slave now.
2141  *  - <slave_dev> is already active.
2142  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2143  *  - <slave_dev> is not running.
2144  * In these cases, this fuction does nothing.
2145  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2146  */
2147 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2148 {
2149         struct bonding *bond = netdev_priv(bond_dev);
2150         struct slave *old_active = NULL;
2151         struct slave *new_active = NULL;
2152         int res = 0;
2153
2154         if (!USES_PRIMARY(bond->params.mode)) {
2155                 return -EINVAL;
2156         }
2157
2158         /* Verify that master_dev is indeed the master of slave_dev */
2159         if (!(slave_dev->flags & IFF_SLAVE) ||
2160             (slave_dev->master != bond_dev)) {
2161                 return -EINVAL;
2162         }
2163
2164         read_lock(&bond->lock);
2165
2166         read_lock(&bond->curr_slave_lock);
2167         old_active = bond->curr_active_slave;
2168         read_unlock(&bond->curr_slave_lock);
2169
2170         new_active = bond_get_slave_by_dev(bond, slave_dev);
2171
2172         /*
2173          * Changing to the current active: do nothing; return success.
2174          */
2175         if (new_active && (new_active == old_active)) {
2176                 read_unlock(&bond->lock);
2177                 return 0;
2178         }
2179
2180         if ((new_active) &&
2181             (old_active) &&
2182             (new_active->link == BOND_LINK_UP) &&
2183             IS_UP(new_active->dev)) {
2184                 write_lock_bh(&bond->curr_slave_lock);
2185                 bond_change_active_slave(bond, new_active);
2186                 write_unlock_bh(&bond->curr_slave_lock);
2187         } else {
2188                 res = -EINVAL;
2189         }
2190
2191         read_unlock(&bond->lock);
2192
2193         return res;
2194 }
2195
2196 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2197 {
2198         struct bonding *bond = netdev_priv(bond_dev);
2199
2200         info->bond_mode = bond->params.mode;
2201         info->miimon = bond->params.miimon;
2202
2203         read_lock(&bond->lock);
2204         info->num_slaves = bond->slave_cnt;
2205         read_unlock(&bond->lock);
2206
2207         return 0;
2208 }
2209
2210 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2211 {
2212         struct bonding *bond = netdev_priv(bond_dev);
2213         struct slave *slave;
2214         int i, found = 0;
2215
2216         if (info->slave_id < 0) {
2217                 return -ENODEV;
2218         }
2219
2220         read_lock(&bond->lock);
2221
2222         bond_for_each_slave(bond, slave, i) {
2223                 if (i == (int)info->slave_id) {
2224                         found = 1;
2225                         break;
2226                 }
2227         }
2228
2229         read_unlock(&bond->lock);
2230
2231         if (found) {
2232                 strcpy(info->slave_name, slave->dev->name);
2233                 info->link = slave->link;
2234                 info->state = slave->state;
2235                 info->link_failure_count = slave->link_failure_count;
2236         } else {
2237                 return -ENODEV;
2238         }
2239
2240         return 0;
2241 }
2242
2243 /*-------------------------------- Monitoring -------------------------------*/
2244
2245
2246 static int bond_miimon_inspect(struct bonding *bond)
2247 {
2248         struct slave *slave;
2249         int i, link_state, commit = 0;
2250
2251         bond_for_each_slave(bond, slave, i) {
2252                 slave->new_link = BOND_LINK_NOCHANGE;
2253
2254                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2255
2256                 switch (slave->link) {
2257                 case BOND_LINK_UP:
2258                         if (link_state)
2259                                 continue;
2260
2261                         slave->link = BOND_LINK_FAIL;
2262                         slave->delay = bond->params.downdelay;
2263                         if (slave->delay) {
2264                                 printk(KERN_INFO DRV_NAME
2265                                        ": %s: link status down for %s"
2266                                        "interface %s, disabling it in %d ms.\n",
2267                                        bond->dev->name,
2268                                        (bond->params.mode ==
2269                                         BOND_MODE_ACTIVEBACKUP) ?
2270                                        ((slave->state == BOND_STATE_ACTIVE) ?
2271                                         "active " : "backup ") : "",
2272                                        slave->dev->name,
2273                                        bond->params.downdelay * bond->params.miimon);
2274                         }
2275                         /*FALLTHRU*/
2276                 case BOND_LINK_FAIL:
2277                         if (link_state) {
2278                                 /*
2279                                  * recovered before downdelay expired
2280                                  */
2281                                 slave->link = BOND_LINK_UP;
2282                                 slave->jiffies = jiffies;
2283                                 printk(KERN_INFO DRV_NAME
2284                                        ": %s: link status up again after %d "
2285                                        "ms for interface %s.\n",
2286                                        bond->dev->name,
2287                                        (bond->params.downdelay - slave->delay) *
2288                                        bond->params.miimon,
2289                                        slave->dev->name);
2290                                 continue;
2291                         }
2292
2293                         if (slave->delay <= 0) {
2294                                 slave->new_link = BOND_LINK_DOWN;
2295                                 commit++;
2296                                 continue;
2297                         }
2298
2299                         slave->delay--;
2300                         break;
2301
2302                 case BOND_LINK_DOWN:
2303                         if (!link_state)
2304                                 continue;
2305
2306                         slave->link = BOND_LINK_BACK;
2307                         slave->delay = bond->params.updelay;
2308
2309                         if (slave->delay) {
2310                                 printk(KERN_INFO DRV_NAME
2311                                        ": %s: link status up for "
2312                                        "interface %s, enabling it in %d ms.\n",
2313                                        bond->dev->name, slave->dev->name,
2314                                        bond->params.updelay *
2315                                        bond->params.miimon);
2316                         }
2317                         /*FALLTHRU*/
2318                 case BOND_LINK_BACK:
2319                         if (!link_state) {
2320                                 slave->link = BOND_LINK_DOWN;
2321                                 printk(KERN_INFO DRV_NAME
2322                                        ": %s: link status down again after %d "
2323                                        "ms for interface %s.\n",
2324                                        bond->dev->name,
2325                                        (bond->params.updelay - slave->delay) *
2326                                        bond->params.miimon,
2327                                        slave->dev->name);
2328
2329                                 continue;
2330                         }
2331
2332                         if (slave->delay <= 0) {
2333                                 slave->new_link = BOND_LINK_UP;
2334                                 commit++;
2335                                 continue;
2336                         }
2337
2338                         slave->delay--;
2339                         break;
2340                 }
2341         }
2342
2343         return commit;
2344 }
2345
2346 static void bond_miimon_commit(struct bonding *bond)
2347 {
2348         struct slave *slave;
2349         int i;
2350
2351         bond_for_each_slave(bond, slave, i) {
2352                 switch (slave->new_link) {
2353                 case BOND_LINK_NOCHANGE:
2354                         continue;
2355
2356                 case BOND_LINK_UP:
2357                         slave->link = BOND_LINK_UP;
2358                         slave->jiffies = jiffies;
2359
2360                         if (bond->params.mode == BOND_MODE_8023AD) {
2361                                 /* prevent it from being the active one */
2362                                 slave->state = BOND_STATE_BACKUP;
2363                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2364                                 /* make it immediately active */
2365                                 slave->state = BOND_STATE_ACTIVE;
2366                         } else if (slave != bond->primary_slave) {
2367                                 /* prevent it from being the active one */
2368                                 slave->state = BOND_STATE_BACKUP;
2369                         }
2370
2371                         printk(KERN_INFO DRV_NAME
2372                                ": %s: link status definitely "
2373                                "up for interface %s.\n",
2374                                bond->dev->name, slave->dev->name);
2375
2376                         /* notify ad that the link status has changed */
2377                         if (bond->params.mode == BOND_MODE_8023AD)
2378                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2379
2380                         if (bond_is_lb(bond))
2381                                 bond_alb_handle_link_change(bond, slave,
2382                                                             BOND_LINK_UP);
2383
2384                         if (!bond->curr_active_slave ||
2385                             (slave == bond->primary_slave))
2386                                 goto do_failover;
2387
2388                         continue;
2389
2390                 case BOND_LINK_DOWN:
2391                         if (slave->link_failure_count < UINT_MAX)
2392                                 slave->link_failure_count++;
2393
2394                         slave->link = BOND_LINK_DOWN;
2395
2396                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2397                             bond->params.mode == BOND_MODE_8023AD)
2398                                 bond_set_slave_inactive_flags(slave);
2399
2400                         printk(KERN_INFO DRV_NAME
2401                                ": %s: link status definitely down for "
2402                                "interface %s, disabling it\n",
2403                                bond->dev->name, slave->dev->name);
2404
2405                         if (bond->params.mode == BOND_MODE_8023AD)
2406                                 bond_3ad_handle_link_change(slave,
2407                                                             BOND_LINK_DOWN);
2408
2409                         if (bond->params.mode == BOND_MODE_TLB ||
2410                             bond->params.mode == BOND_MODE_ALB)
2411                                 bond_alb_handle_link_change(bond, slave,
2412                                                             BOND_LINK_DOWN);
2413
2414                         if (slave == bond->curr_active_slave)
2415                                 goto do_failover;
2416
2417                         continue;
2418
2419                 default:
2420                         printk(KERN_ERR DRV_NAME
2421                                ": %s: invalid new link %d on slave %s\n",
2422                                bond->dev->name, slave->new_link,
2423                                slave->dev->name);
2424                         slave->new_link = BOND_LINK_NOCHANGE;
2425
2426                         continue;
2427                 }
2428
2429 do_failover:
2430                 ASSERT_RTNL();
2431                 write_lock_bh(&bond->curr_slave_lock);
2432                 bond_select_active_slave(bond);
2433                 write_unlock_bh(&bond->curr_slave_lock);
2434         }
2435
2436         bond_set_carrier(bond);
2437 }
2438
2439 /*
2440  * bond_mii_monitor
2441  *
2442  * Really a wrapper that splits the mii monitor into two phases: an
2443  * inspection, then (if inspection indicates something needs to be done)
2444  * an acquisition of appropriate locks followed by a commit phase to
2445  * implement whatever link state changes are indicated.
2446  */
2447 void bond_mii_monitor(struct work_struct *work)
2448 {
2449         struct bonding *bond = container_of(work, struct bonding,
2450                                             mii_work.work);
2451
2452         read_lock(&bond->lock);
2453         if (bond->kill_timers)
2454                 goto out;
2455
2456         if (bond->slave_cnt == 0)
2457                 goto re_arm;
2458
2459         if (bond->send_grat_arp) {
2460                 read_lock(&bond->curr_slave_lock);
2461                 bond_send_gratuitous_arp(bond);
2462                 read_unlock(&bond->curr_slave_lock);
2463         }
2464
2465         if (bond->send_unsol_na) {
2466                 read_lock(&bond->curr_slave_lock);
2467                 bond_send_unsolicited_na(bond);
2468                 read_unlock(&bond->curr_slave_lock);
2469         }
2470
2471         if (bond_miimon_inspect(bond)) {
2472                 read_unlock(&bond->lock);
2473                 rtnl_lock();
2474                 read_lock(&bond->lock);
2475
2476                 bond_miimon_commit(bond);
2477
2478                 read_unlock(&bond->lock);
2479                 rtnl_unlock();  /* might sleep, hold no other locks */
2480                 read_lock(&bond->lock);
2481         }
2482
2483 re_arm:
2484         if (bond->params.miimon)
2485                 queue_delayed_work(bond->wq, &bond->mii_work,
2486                                    msecs_to_jiffies(bond->params.miimon));
2487 out:
2488         read_unlock(&bond->lock);
2489 }
2490
2491 static __be32 bond_glean_dev_ip(struct net_device *dev)
2492 {
2493         struct in_device *idev;
2494         struct in_ifaddr *ifa;
2495         __be32 addr = 0;
2496
2497         if (!dev)
2498                 return 0;
2499
2500         rcu_read_lock();
2501         idev = __in_dev_get_rcu(dev);
2502         if (!idev)
2503                 goto out;
2504
2505         ifa = idev->ifa_list;
2506         if (!ifa)
2507                 goto out;
2508
2509         addr = ifa->ifa_local;
2510 out:
2511         rcu_read_unlock();
2512         return addr;
2513 }
2514
2515 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2516 {
2517         struct vlan_entry *vlan;
2518
2519         if (ip == bond->master_ip)
2520                 return 1;
2521
2522         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2523                 if (ip == vlan->vlan_ip)
2524                         return 1;
2525         }
2526
2527         return 0;
2528 }
2529
2530 /*
2531  * We go to the (large) trouble of VLAN tagging ARP frames because
2532  * switches in VLAN mode (especially if ports are configured as
2533  * "native" to a VLAN) might not pass non-tagged frames.
2534  */
2535 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2536 {
2537         struct sk_buff *skb;
2538
2539         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2540                slave_dev->name, dest_ip, src_ip, vlan_id);
2541                
2542         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2543                          NULL, slave_dev->dev_addr, NULL);
2544
2545         if (!skb) {
2546                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2547                 return;
2548         }
2549         if (vlan_id) {
2550                 skb = vlan_put_tag(skb, vlan_id);
2551                 if (!skb) {
2552                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2553                         return;
2554                 }
2555         }
2556         arp_xmit(skb);
2557 }
2558
2559
2560 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2561 {
2562         int i, vlan_id, rv;
2563         __be32 *targets = bond->params.arp_targets;
2564         struct vlan_entry *vlan;
2565         struct net_device *vlan_dev;
2566         struct flowi fl;
2567         struct rtable *rt;
2568
2569         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2570                 if (!targets[i])
2571                         break;
2572                 pr_debug("basa: target %x\n", targets[i]);
2573                 if (list_empty(&bond->vlan_list)) {
2574                         pr_debug("basa: empty vlan: arp_send\n");
2575                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2576                                       bond->master_ip, 0);
2577                         continue;
2578                 }
2579
2580                 /*
2581                  * If VLANs are configured, we do a route lookup to
2582                  * determine which VLAN interface would be used, so we
2583                  * can tag the ARP with the proper VLAN tag.
2584                  */
2585                 memset(&fl, 0, sizeof(fl));
2586                 fl.fl4_dst = targets[i];
2587                 fl.fl4_tos = RTO_ONLINK;
2588
2589                 rv = ip_route_output_key(&init_net, &rt, &fl);
2590                 if (rv) {
2591                         if (net_ratelimit()) {
2592                                 printk(KERN_WARNING DRV_NAME
2593                              ": %s: no route to arp_ip_target %pI4\n",
2594                                        bond->dev->name, &fl.fl4_dst);
2595                         }
2596                         continue;
2597                 }
2598
2599                 /*
2600                  * This target is not on a VLAN
2601                  */
2602                 if (rt->u.dst.dev == bond->dev) {
2603                         ip_rt_put(rt);
2604                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2605                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2606                                       bond->master_ip, 0);
2607                         continue;
2608                 }
2609
2610                 vlan_id = 0;
2611                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2612                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2613                         if (vlan_dev == rt->u.dst.dev) {
2614                                 vlan_id = vlan->vlan_id;
2615                                 pr_debug("basa: vlan match on %s %d\n",
2616                                        vlan_dev->name, vlan_id);
2617                                 break;
2618                         }
2619                 }
2620
2621                 if (vlan_id) {
2622                         ip_rt_put(rt);
2623                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2624                                       vlan->vlan_ip, vlan_id);
2625                         continue;
2626                 }
2627
2628                 if (net_ratelimit()) {
2629                         printk(KERN_WARNING DRV_NAME
2630                ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2631                                bond->dev->name, &fl.fl4_dst,
2632                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2633                 }
2634                 ip_rt_put(rt);
2635         }
2636 }
2637
2638 /*
2639  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2640  * for each VLAN above us.
2641  *
2642  * Caller must hold curr_slave_lock for read or better
2643  */
2644 static void bond_send_gratuitous_arp(struct bonding *bond)
2645 {
2646         struct slave *slave = bond->curr_active_slave;
2647         struct vlan_entry *vlan;
2648         struct net_device *vlan_dev;
2649
2650         pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2651                                 slave ? slave->dev->name : "NULL");
2652
2653         if (!slave || !bond->send_grat_arp ||
2654             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2655                 return;
2656
2657         bond->send_grat_arp--;
2658
2659         if (bond->master_ip) {
2660                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2661                                 bond->master_ip, 0);
2662         }
2663
2664         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2665                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2666                 if (vlan->vlan_ip) {
2667                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2668                                       vlan->vlan_ip, vlan->vlan_id);
2669                 }
2670         }
2671 }
2672
2673 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2674 {
2675         int i;
2676         __be32 *targets = bond->params.arp_targets;
2677
2678         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2679                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2680                         &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2681                 if (sip == targets[i]) {
2682                         if (bond_has_this_ip(bond, tip))
2683                                 slave->last_arp_rx = jiffies;
2684                         return;
2685                 }
2686         }
2687 }
2688
2689 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2690 {
2691         struct arphdr *arp;
2692         struct slave *slave;
2693         struct bonding *bond;
2694         unsigned char *arp_ptr;
2695         __be32 sip, tip;
2696
2697         if (dev_net(dev) != &init_net)
2698                 goto out;
2699
2700         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2701                 goto out;
2702
2703         bond = netdev_priv(dev);
2704         read_lock(&bond->lock);
2705
2706         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2707                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2708                 orig_dev ? orig_dev->name : "NULL");
2709
2710         slave = bond_get_slave_by_dev(bond, orig_dev);
2711         if (!slave || !slave_do_arp_validate(bond, slave))
2712                 goto out_unlock;
2713
2714         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2715                 goto out_unlock;
2716
2717         arp = arp_hdr(skb);
2718         if (arp->ar_hln != dev->addr_len ||
2719             skb->pkt_type == PACKET_OTHERHOST ||
2720             skb->pkt_type == PACKET_LOOPBACK ||
2721             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2722             arp->ar_pro != htons(ETH_P_IP) ||
2723             arp->ar_pln != 4)
2724                 goto out_unlock;
2725
2726         arp_ptr = (unsigned char *)(arp + 1);
2727         arp_ptr += dev->addr_len;
2728         memcpy(&sip, arp_ptr, 4);
2729         arp_ptr += 4 + dev->addr_len;
2730         memcpy(&tip, arp_ptr, 4);
2731
2732         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2733                 bond->dev->name, slave->dev->name, slave->state,
2734                 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2735                 &sip, &tip);
2736
2737         /*
2738          * Backup slaves won't see the ARP reply, but do come through
2739          * here for each ARP probe (so we swap the sip/tip to validate
2740          * the probe).  In a "redundant switch, common router" type of
2741          * configuration, the ARP probe will (hopefully) travel from
2742          * the active, through one switch, the router, then the other
2743          * switch before reaching the backup.
2744          */
2745         if (slave->state == BOND_STATE_ACTIVE)
2746                 bond_validate_arp(bond, slave, sip, tip);
2747         else
2748                 bond_validate_arp(bond, slave, tip, sip);
2749
2750 out_unlock:
2751         read_unlock(&bond->lock);
2752 out:
2753         dev_kfree_skb(skb);
2754         return NET_RX_SUCCESS;
2755 }
2756
2757 /*
2758  * this function is called regularly to monitor each slave's link
2759  * ensuring that traffic is being sent and received when arp monitoring
2760  * is used in load-balancing mode. if the adapter has been dormant, then an
2761  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2762  * arp monitoring in active backup mode.
2763  */
2764 void bond_loadbalance_arp_mon(struct work_struct *work)
2765 {
2766         struct bonding *bond = container_of(work, struct bonding,
2767                                             arp_work.work);
2768         struct slave *slave, *oldcurrent;
2769         int do_failover = 0;
2770         int delta_in_ticks;
2771         int i;
2772
2773         read_lock(&bond->lock);
2774
2775         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2776
2777         if (bond->kill_timers) {
2778                 goto out;
2779         }
2780
2781         if (bond->slave_cnt == 0) {
2782                 goto re_arm;
2783         }
2784
2785         read_lock(&bond->curr_slave_lock);
2786         oldcurrent = bond->curr_active_slave;
2787         read_unlock(&bond->curr_slave_lock);
2788
2789         /* see if any of the previous devices are up now (i.e. they have
2790          * xmt and rcv traffic). the curr_active_slave does not come into
2791          * the picture unless it is null. also, slave->jiffies is not needed
2792          * here because we send an arp on each slave and give a slave as
2793          * long as it needs to get the tx/rx within the delta.
2794          * TODO: what about up/down delay in arp mode? it wasn't here before
2795          *       so it can wait
2796          */
2797         bond_for_each_slave(bond, slave, i) {
2798                 if (slave->link != BOND_LINK_UP) {
2799                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2800                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2801
2802                                 slave->link  = BOND_LINK_UP;
2803                                 slave->state = BOND_STATE_ACTIVE;
2804
2805                                 /* primary_slave has no meaning in round-robin
2806                                  * mode. the window of a slave being up and
2807                                  * curr_active_slave being null after enslaving
2808                                  * is closed.
2809                                  */
2810                                 if (!oldcurrent) {
2811                                         printk(KERN_INFO DRV_NAME
2812                                                ": %s: link status definitely "
2813                                                "up for interface %s, ",
2814                                                bond->dev->name,
2815                                                slave->dev->name);
2816                                         do_failover = 1;
2817                                 } else {
2818                                         printk(KERN_INFO DRV_NAME
2819                                                ": %s: interface %s is now up\n",
2820                                                bond->dev->name,
2821                                                slave->dev->name);
2822                                 }
2823                         }
2824                 } else {
2825                         /* slave->link == BOND_LINK_UP */
2826
2827                         /* not all switches will respond to an arp request
2828                          * when the source ip is 0, so don't take the link down
2829                          * if we don't know our ip yet
2830                          */
2831                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2832                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2833
2834                                 slave->link  = BOND_LINK_DOWN;
2835                                 slave->state = BOND_STATE_BACKUP;
2836
2837                                 if (slave->link_failure_count < UINT_MAX) {
2838                                         slave->link_failure_count++;
2839                                 }
2840
2841                                 printk(KERN_INFO DRV_NAME
2842                                        ": %s: interface %s is now down.\n",
2843                                        bond->dev->name,
2844                                        slave->dev->name);
2845
2846                                 if (slave == oldcurrent) {
2847                                         do_failover = 1;
2848                                 }
2849                         }
2850                 }
2851
2852                 /* note: if switch is in round-robin mode, all links
2853                  * must tx arp to ensure all links rx an arp - otherwise
2854                  * links may oscillate or not come up at all; if switch is
2855                  * in something like xor mode, there is nothing we can
2856                  * do - all replies will be rx'ed on same link causing slaves
2857                  * to be unstable during low/no traffic periods
2858                  */
2859                 if (IS_UP(slave->dev)) {
2860                         bond_arp_send_all(bond, slave);
2861                 }
2862         }
2863
2864         if (do_failover) {
2865                 write_lock_bh(&bond->curr_slave_lock);
2866
2867                 bond_select_active_slave(bond);
2868
2869                 write_unlock_bh(&bond->curr_slave_lock);
2870         }
2871
2872 re_arm:
2873         if (bond->params.arp_interval)
2874                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2875 out:
2876         read_unlock(&bond->lock);
2877 }
2878
2879 /*
2880  * Called to inspect slaves for active-backup mode ARP monitor link state
2881  * changes.  Sets new_link in slaves to specify what action should take
2882  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2883  * to link states must be committed.
2884  *
2885  * Called with bond->lock held for read.
2886  */
2887 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2888 {
2889         struct slave *slave;
2890         int i, commit = 0;
2891
2892         bond_for_each_slave(bond, slave, i) {
2893                 slave->new_link = BOND_LINK_NOCHANGE;
2894
2895                 if (slave->link != BOND_LINK_UP) {
2896                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2897                                            delta_in_ticks)) {
2898                                 slave->new_link = BOND_LINK_UP;
2899                                 commit++;
2900                         }
2901
2902                         continue;
2903                 }
2904
2905                 /*
2906                  * Give slaves 2*delta after being enslaved or made
2907                  * active.  This avoids bouncing, as the last receive
2908                  * times need a full ARP monitor cycle to be updated.
2909                  */
2910                 if (!time_after_eq(jiffies, slave->jiffies +
2911                                    2 * delta_in_ticks))
2912                         continue;
2913
2914                 /*
2915                  * Backup slave is down if:
2916                  * - No current_arp_slave AND
2917                  * - more than 3*delta since last receive AND
2918                  * - the bond has an IP address
2919                  *
2920                  * Note: a non-null current_arp_slave indicates
2921                  * the curr_active_slave went down and we are
2922                  * searching for a new one; under this condition
2923                  * we only take the curr_active_slave down - this
2924                  * gives each slave a chance to tx/rx traffic
2925                  * before being taken out
2926                  */
2927                 if (slave->state == BOND_STATE_BACKUP &&
2928                     !bond->current_arp_slave &&
2929                     time_after(jiffies, slave_last_rx(bond, slave) +
2930                                3 * delta_in_ticks)) {
2931                         slave->new_link = BOND_LINK_DOWN;
2932                         commit++;
2933                 }
2934
2935                 /*
2936                  * Active slave is down if:
2937                  * - more than 2*delta since transmitting OR
2938                  * - (more than 2*delta since receive AND
2939                  *    the bond has an IP address)
2940                  */
2941                 if ((slave->state == BOND_STATE_ACTIVE) &&
2942                     (time_after_eq(jiffies, slave->dev->trans_start +
2943                                     2 * delta_in_ticks) ||
2944                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2945                                      + 2 * delta_in_ticks)))) {
2946                         slave->new_link = BOND_LINK_DOWN;
2947                         commit++;
2948                 }
2949         }
2950
2951         read_lock(&bond->curr_slave_lock);
2952
2953         /*
2954          * Trigger a commit if the primary option setting has changed.
2955          */
2956         if (bond->primary_slave &&
2957             (bond->primary_slave != bond->curr_active_slave) &&
2958             (bond->primary_slave->link == BOND_LINK_UP))
2959                 commit++;
2960
2961         read_unlock(&bond->curr_slave_lock);
2962
2963         return commit;
2964 }
2965
2966 /*
2967  * Called to commit link state changes noted by inspection step of
2968  * active-backup mode ARP monitor.
2969  *
2970  * Called with RTNL and bond->lock for read.
2971  */
2972 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2973 {
2974         struct slave *slave;
2975         int i;
2976
2977         bond_for_each_slave(bond, slave, i) {
2978                 switch (slave->new_link) {
2979                 case BOND_LINK_NOCHANGE:
2980                         continue;
2981
2982                 case BOND_LINK_UP:
2983                         write_lock_bh(&bond->curr_slave_lock);
2984
2985                         if (!bond->curr_active_slave &&
2986                             time_before_eq(jiffies, slave->dev->trans_start +
2987                                            delta_in_ticks)) {
2988                                 slave->link = BOND_LINK_UP;
2989                                 bond_change_active_slave(bond, slave);
2990                                 bond->current_arp_slave = NULL;
2991
2992                                 printk(KERN_INFO DRV_NAME
2993                                        ": %s: %s is up and now the "
2994                                        "active interface\n",
2995                                        bond->dev->name, slave->dev->name);
2996
2997                         } else if (bond->curr_active_slave != slave) {
2998                                 /* this slave has just come up but we
2999                                  * already have a current slave; this can
3000                                  * also happen if bond_enslave adds a new
3001                                  * slave that is up while we are searching
3002                                  * for a new slave
3003                                  */
3004                                 slave->link = BOND_LINK_UP;
3005                                 bond_set_slave_inactive_flags(slave);
3006                                 bond->current_arp_slave = NULL;
3007
3008                                 printk(KERN_INFO DRV_NAME
3009                                        ": %s: backup interface %s is now up\n",
3010                                        bond->dev->name, slave->dev->name);
3011                         }
3012
3013                         write_unlock_bh(&bond->curr_slave_lock);
3014
3015                         break;
3016
3017                 case BOND_LINK_DOWN:
3018                         if (slave->link_failure_count < UINT_MAX)
3019                                 slave->link_failure_count++;
3020
3021                         slave->link = BOND_LINK_DOWN;
3022
3023                         if (slave == bond->curr_active_slave) {
3024                                 printk(KERN_INFO DRV_NAME
3025                                        ": %s: link status down for active "
3026                                        "interface %s, disabling it\n",
3027                                        bond->dev->name, slave->dev->name);
3028
3029                                 bond_set_slave_inactive_flags(slave);
3030
3031                                 write_lock_bh(&bond->curr_slave_lock);
3032
3033                                 bond_select_active_slave(bond);
3034                                 if (bond->curr_active_slave)
3035                                         bond->curr_active_slave->jiffies =
3036                                                 jiffies;
3037
3038                                 write_unlock_bh(&bond->curr_slave_lock);
3039
3040                                 bond->current_arp_slave = NULL;
3041
3042                         } else if (slave->state == BOND_STATE_BACKUP) {
3043                                 printk(KERN_INFO DRV_NAME
3044                                        ": %s: backup interface %s is now down\n",
3045                                        bond->dev->name, slave->dev->name);
3046
3047                                 bond_set_slave_inactive_flags(slave);
3048                         }
3049                         break;
3050
3051                 default:
3052                         printk(KERN_ERR DRV_NAME
3053                                ": %s: impossible: new_link %d on slave %s\n",
3054                                bond->dev->name, slave->new_link,
3055                                slave->dev->name);
3056                 }
3057         }
3058
3059         /*
3060          * No race with changes to primary via sysfs, as we hold rtnl.
3061          */
3062         if (bond->primary_slave &&
3063             (bond->primary_slave != bond->curr_active_slave) &&
3064             (bond->primary_slave->link == BOND_LINK_UP)) {
3065                 write_lock_bh(&bond->curr_slave_lock);
3066                 bond_change_active_slave(bond, bond->primary_slave);
3067                 write_unlock_bh(&bond->curr_slave_lock);
3068         }
3069
3070         bond_set_carrier(bond);
3071 }
3072
3073 /*
3074  * Send ARP probes for active-backup mode ARP monitor.
3075  *
3076  * Called with bond->lock held for read.
3077  */
3078 static void bond_ab_arp_probe(struct bonding *bond)
3079 {
3080         struct slave *slave;
3081         int i;
3082
3083         read_lock(&bond->curr_slave_lock);
3084
3085         if (bond->current_arp_slave && bond->curr_active_slave)
3086                 printk("PROBE: c_arp %s && cas %s BAD\n",
3087                        bond->current_arp_slave->dev->name,
3088                        bond->curr_active_slave->dev->name);
3089
3090         if (bond->curr_active_slave) {
3091                 bond_arp_send_all(bond, bond->curr_active_slave);
3092                 read_unlock(&bond->curr_slave_lock);
3093                 return;
3094         }
3095
3096         read_unlock(&bond->curr_slave_lock);
3097
3098         /* if we don't have a curr_active_slave, search for the next available
3099          * backup slave from the current_arp_slave and make it the candidate
3100          * for becoming the curr_active_slave
3101          */
3102
3103         if (!bond->current_arp_slave) {
3104                 bond->current_arp_slave = bond->first_slave;
3105                 if (!bond->current_arp_slave)
3106                         return;
3107         }
3108
3109         bond_set_slave_inactive_flags(bond->current_arp_slave);
3110
3111         /* search for next candidate */
3112         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3113                 if (IS_UP(slave->dev)) {
3114                         slave->link = BOND_LINK_BACK;
3115                         bond_set_slave_active_flags(slave);
3116                         bond_arp_send_all(bond, slave);
3117                         slave->jiffies = jiffies;
3118                         bond->current_arp_slave = slave;
3119                         break;
3120                 }
3121
3122                 /* if the link state is up at this point, we
3123                  * mark it down - this can happen if we have
3124                  * simultaneous link failures and
3125                  * reselect_active_interface doesn't make this
3126                  * one the current slave so it is still marked
3127                  * up when it is actually down
3128                  */
3129                 if (slave->link == BOND_LINK_UP) {
3130                         slave->link = BOND_LINK_DOWN;
3131                         if (slave->link_failure_count < UINT_MAX)
3132                                 slave->link_failure_count++;
3133
3134                         bond_set_slave_inactive_flags(slave);
3135
3136                         printk(KERN_INFO DRV_NAME
3137                                ": %s: backup interface %s is now down.\n",
3138                                bond->dev->name, slave->dev->name);
3139                 }
3140         }
3141 }
3142
3143 void bond_activebackup_arp_mon(struct work_struct *work)
3144 {
3145         struct bonding *bond = container_of(work, struct bonding,
3146                                             arp_work.work);
3147         int delta_in_ticks;
3148
3149         read_lock(&bond->lock);
3150
3151         if (bond->kill_timers)
3152                 goto out;
3153
3154         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3155
3156         if (bond->slave_cnt == 0)
3157                 goto re_arm;
3158
3159         if (bond->send_grat_arp) {
3160                 read_lock(&bond->curr_slave_lock);
3161                 bond_send_gratuitous_arp(bond);
3162                 read_unlock(&bond->curr_slave_lock);
3163         }
3164
3165         if (bond->send_unsol_na) {
3166                 read_lock(&bond->curr_slave_lock);
3167                 bond_send_unsolicited_na(bond);
3168                 read_unlock(&bond->curr_slave_lock);
3169         }
3170
3171         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3172                 read_unlock(&bond->lock);
3173                 rtnl_lock();
3174                 read_lock(&bond->lock);
3175
3176                 bond_ab_arp_commit(bond, delta_in_ticks);
3177
3178                 read_unlock(&bond->lock);
3179                 rtnl_unlock();
3180                 read_lock(&bond->lock);
3181         }
3182
3183         bond_ab_arp_probe(bond);
3184
3185 re_arm:
3186         if (bond->params.arp_interval) {
3187                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3188         }
3189 out:
3190         read_unlock(&bond->lock);
3191 }
3192
3193 /*------------------------------ proc/seq_file-------------------------------*/
3194
3195 #ifdef CONFIG_PROC_FS
3196
3197 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3198         __acquires(&dev_base_lock)
3199         __acquires(&bond->lock)
3200 {
3201         struct bonding *bond = seq->private;
3202         loff_t off = 0;
3203         struct slave *slave;
3204         int i;
3205
3206         /* make sure the bond won't be taken away */
3207         read_lock(&dev_base_lock);
3208         read_lock(&bond->lock);
3209
3210         if (*pos == 0) {
3211                 return SEQ_START_TOKEN;
3212         }
3213
3214         bond_for_each_slave(bond, slave, i) {
3215                 if (++off == *pos) {
3216                         return slave;
3217                 }
3218         }
3219
3220         return NULL;
3221 }
3222
3223 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3224 {
3225         struct bonding *bond = seq->private;
3226         struct slave *slave = v;
3227
3228         ++*pos;
3229         if (v == SEQ_START_TOKEN) {
3230                 return bond->first_slave;
3231         }
3232
3233         slave = slave->next;
3234
3235         return (slave == bond->first_slave) ? NULL : slave;
3236 }
3237
3238 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3239         __releases(&bond->lock)
3240         __releases(&dev_base_lock)
3241 {
3242         struct bonding *bond = seq->private;
3243
3244         read_unlock(&bond->lock);
3245         read_unlock(&dev_base_lock);
3246 }
3247
3248 static void bond_info_show_master(struct seq_file *seq)
3249 {
3250         struct bonding *bond = seq->private;
3251         struct slave *curr;
3252         int i;
3253
3254         read_lock(&bond->curr_slave_lock);
3255         curr = bond->curr_active_slave;
3256         read_unlock(&bond->curr_slave_lock);
3257
3258         seq_printf(seq, "Bonding Mode: %s",
3259                    bond_mode_name(bond->params.mode));
3260
3261         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3262             bond->params.fail_over_mac)
3263                 seq_printf(seq, " (fail_over_mac %s)",
3264                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3265
3266         seq_printf(seq, "\n");
3267
3268         if (bond->params.mode == BOND_MODE_XOR ||
3269                 bond->params.mode == BOND_MODE_8023AD) {
3270                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3271                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3272                         bond->params.xmit_policy);
3273         }
3274
3275         if (USES_PRIMARY(bond->params.mode)) {
3276                 seq_printf(seq, "Primary Slave: %s\n",
3277                            (bond->primary_slave) ?
3278                            bond->primary_slave->dev->name : "None");
3279
3280                 seq_printf(seq, "Currently Active Slave: %s\n",
3281                            (curr) ? curr->dev->name : "None");
3282         }
3283
3284         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3285                    "up" : "down");
3286         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3287         seq_printf(seq, "Up Delay (ms): %d\n",
3288                    bond->params.updelay * bond->params.miimon);
3289         seq_printf(seq, "Down Delay (ms): %d\n",
3290                    bond->params.downdelay * bond->params.miimon);
3291
3292
3293         /* ARP information */
3294         if(bond->params.arp_interval > 0) {
3295                 int printed=0;
3296                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3297                                 bond->params.arp_interval);
3298
3299                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3300
3301                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3302                         if (!bond->params.arp_targets[i])
3303                                 break;
3304                         if (printed)
3305                                 seq_printf(seq, ",");
3306                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3307                         printed = 1;
3308                 }
3309                 seq_printf(seq, "\n");
3310         }
3311
3312         if (bond->params.mode == BOND_MODE_8023AD) {
3313                 struct ad_info ad_info;
3314
3315                 seq_puts(seq, "\n802.3ad info\n");
3316                 seq_printf(seq, "LACP rate: %s\n",
3317                            (bond->params.lacp_fast) ? "fast" : "slow");
3318                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3319                            ad_select_tbl[bond->params.ad_select].modename);
3320
3321                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3322                         seq_printf(seq, "bond %s has no active aggregator\n",
3323                                    bond->dev->name);
3324                 } else {
3325                         seq_printf(seq, "Active Aggregator Info:\n");
3326
3327                         seq_printf(seq, "\tAggregator ID: %d\n",
3328                                    ad_info.aggregator_id);
3329                         seq_printf(seq, "\tNumber of ports: %d\n",
3330                                    ad_info.ports);
3331                         seq_printf(seq, "\tActor Key: %d\n",
3332                                    ad_info.actor_key);
3333                         seq_printf(seq, "\tPartner Key: %d\n",
3334                                    ad_info.partner_key);
3335                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3336                                    ad_info.partner_system);
3337                 }
3338         }
3339 }
3340
3341 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3342 {
3343         struct bonding *bond = seq->private;
3344
3345         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3346         seq_printf(seq, "MII Status: %s\n",
3347                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3348         seq_printf(seq, "Link Failure Count: %u\n",
3349                    slave->link_failure_count);
3350
3351         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3352
3353         if (bond->params.mode == BOND_MODE_8023AD) {
3354                 const struct aggregator *agg
3355                         = SLAVE_AD_INFO(slave).port.aggregator;
3356
3357                 if (agg) {
3358                         seq_printf(seq, "Aggregator ID: %d\n",
3359                                    agg->aggregator_identifier);
3360                 } else {
3361                         seq_puts(seq, "Aggregator ID: N/A\n");
3362                 }
3363         }
3364 }
3365
3366 static int bond_info_seq_show(struct seq_file *seq, void *v)
3367 {
3368         if (v == SEQ_START_TOKEN) {
3369                 seq_printf(seq, "%s\n", version);
3370                 bond_info_show_master(seq);
3371         } else {
3372                 bond_info_show_slave(seq, v);
3373         }
3374
3375         return 0;
3376 }
3377
3378 static const struct seq_operations bond_info_seq_ops = {
3379         .start = bond_info_seq_start,
3380         .next  = bond_info_seq_next,
3381         .stop  = bond_info_seq_stop,
3382         .show  = bond_info_seq_show,
3383 };
3384
3385 static int bond_info_open(struct inode *inode, struct file *file)
3386 {
3387         struct seq_file *seq;
3388         struct proc_dir_entry *proc;
3389         int res;
3390
3391         res = seq_open(file, &bond_info_seq_ops);
3392         if (!res) {
3393                 /* recover the pointer buried in proc_dir_entry data */
3394                 seq = file->private_data;
3395                 proc = PDE(inode);
3396                 seq->private = proc->data;
3397         }
3398
3399         return res;
3400 }
3401
3402 static const struct file_operations bond_info_fops = {
3403         .owner   = THIS_MODULE,
3404         .open    = bond_info_open,
3405         .read    = seq_read,
3406         .llseek  = seq_lseek,
3407         .release = seq_release,
3408 };
3409
3410 static int bond_create_proc_entry(struct bonding *bond)
3411 {
3412         struct net_device *bond_dev = bond->dev;
3413
3414         if (bond_proc_dir) {
3415                 bond->proc_entry = proc_create_data(bond_dev->name,
3416                                                     S_IRUGO, bond_proc_dir,
3417                                                     &bond_info_fops, bond);
3418                 if (bond->proc_entry == NULL) {
3419                         printk(KERN_WARNING DRV_NAME
3420                                ": Warning: Cannot create /proc/net/%s/%s\n",
3421                                DRV_NAME, bond_dev->name);
3422                 } else {
3423                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3424                 }
3425         }
3426
3427         return 0;
3428 }
3429
3430 static void bond_remove_proc_entry(struct bonding *bond)
3431 {
3432         if (bond_proc_dir && bond->proc_entry) {
3433                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3434                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3435                 bond->proc_entry = NULL;
3436         }
3437 }
3438
3439 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3440  * Caller must hold rtnl_lock.
3441  */
3442 static void bond_create_proc_dir(void)
3443 {
3444         if (!bond_proc_dir) {
3445                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3446                 if (!bond_proc_dir)
3447                         printk(KERN_WARNING DRV_NAME
3448                                 ": Warning: cannot create /proc/net/%s\n",
3449                                 DRV_NAME);
3450         }
3451 }
3452
3453 /* Destroy the bonding directory under /proc/net, if empty.
3454  * Caller must hold rtnl_lock.
3455  */
3456 static void bond_destroy_proc_dir(void)
3457 {
3458         if (bond_proc_dir) {
3459                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3460                 bond_proc_dir = NULL;
3461         }
3462 }
3463 #endif /* CONFIG_PROC_FS */
3464
3465 /*-------------------------- netdev event handling --------------------------*/
3466
3467 /*
3468  * Change device name
3469  */
3470 static int bond_event_changename(struct bonding *bond)
3471 {
3472 #ifdef CONFIG_PROC_FS
3473         bond_remove_proc_entry(bond);
3474         bond_create_proc_entry(bond);
3475 #endif
3476         down_write(&(bonding_rwsem));
3477         bond_destroy_sysfs_entry(bond);
3478         bond_create_sysfs_entry(bond);
3479         up_write(&(bonding_rwsem));
3480         return NOTIFY_DONE;
3481 }
3482
3483 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3484 {
3485         struct bonding *event_bond = netdev_priv(bond_dev);
3486
3487         switch (event) {
3488         case NETDEV_CHANGENAME:
3489                 return bond_event_changename(event_bond);
3490         case NETDEV_UNREGISTER:
3491                 bond_release_all(event_bond->dev);
3492                 break;
3493         default:
3494                 break;
3495         }
3496
3497         return NOTIFY_DONE;
3498 }
3499
3500 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3501 {
3502         struct net_device *bond_dev = slave_dev->master;
3503         struct bonding *bond = netdev_priv(bond_dev);
3504
3505         switch (event) {
3506         case NETDEV_UNREGISTER:
3507                 if (bond_dev) {
3508                         if (bond->setup_by_slave)
3509                                 bond_release_and_destroy(bond_dev, slave_dev);
3510                         else
3511                                 bond_release(bond_dev, slave_dev);
3512                 }
3513                 break;
3514         case NETDEV_CHANGE:
3515                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3516                         struct slave *slave;
3517
3518                         slave = bond_get_slave_by_dev(bond, slave_dev);
3519                         if (slave) {
3520                                 u16 old_speed = slave->speed;
3521                                 u16 old_duplex = slave->duplex;
3522
3523                                 bond_update_speed_duplex(slave);
3524
3525                                 if (bond_is_lb(bond))
3526                                         break;
3527
3528                                 if (old_speed != slave->speed)
3529                                         bond_3ad_adapter_speed_changed(slave);
3530                                 if (old_duplex != slave->duplex)
3531                                         bond_3ad_adapter_duplex_changed(slave);
3532                         }
3533                 }
3534
3535                 break;
3536         case NETDEV_DOWN:
3537                 /*
3538                  * ... Or is it this?
3539                  */
3540                 break;
3541         case NETDEV_CHANGEMTU:
3542                 /*
3543                  * TODO: Should slaves be allowed to
3544                  * independently alter their MTU?  For
3545                  * an active-backup bond, slaves need
3546                  * not be the same type of device, so
3547                  * MTUs may vary.  For other modes,
3548                  * slaves arguably should have the
3549                  * same MTUs. To do this, we'd need to
3550                  * take over the slave's change_mtu
3551                  * function for the duration of their
3552                  * servitude.
3553                  */
3554                 break;
3555         case NETDEV_CHANGENAME:
3556                 /*
3557                  * TODO: handle changing the primary's name
3558                  */
3559                 break;
3560         case NETDEV_FEAT_CHANGE:
3561                 bond_compute_features(bond);
3562                 break;
3563         default:
3564                 break;
3565         }
3566
3567         return NOTIFY_DONE;
3568 }
3569
3570 /*
3571  * bond_netdev_event: handle netdev notifier chain events.
3572  *
3573  * This function receives events for the netdev chain.  The caller (an
3574  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3575  * locks for us to safely manipulate the slave devices (RTNL lock,
3576  * dev_probe_lock).
3577  */
3578 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3579 {
3580         struct net_device *event_dev = (struct net_device *)ptr;
3581
3582         if (dev_net(event_dev) != &init_net)
3583                 return NOTIFY_DONE;
3584
3585         pr_debug("event_dev: %s, event: %lx\n",
3586                 (event_dev ? event_dev->name : "None"),
3587                 event);
3588
3589         if (!(event_dev->priv_flags & IFF_BONDING))
3590                 return NOTIFY_DONE;
3591
3592         if (event_dev->flags & IFF_MASTER) {
3593                 pr_debug("IFF_MASTER\n");
3594                 return bond_master_netdev_event(event, event_dev);
3595         }
3596
3597         if (event_dev->flags & IFF_SLAVE) {
3598                 pr_debug("IFF_SLAVE\n");
3599                 return bond_slave_netdev_event(event, event_dev);
3600         }
3601
3602         return NOTIFY_DONE;
3603 }
3604
3605 /*
3606  * bond_inetaddr_event: handle inetaddr notifier chain events.
3607  *
3608  * We keep track of device IPs primarily to use as source addresses in
3609  * ARP monitor probes (rather than spewing out broadcasts all the time).
3610  *
3611  * We track one IP for the main device (if it has one), plus one per VLAN.
3612  */
3613 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3614 {
3615         struct in_ifaddr *ifa = ptr;
3616         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3617         struct bonding *bond;
3618         struct vlan_entry *vlan;
3619
3620         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3621                 return NOTIFY_DONE;
3622
3623         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3624                 if (bond->dev == event_dev) {
3625                         switch (event) {
3626                         case NETDEV_UP:
3627                                 bond->master_ip = ifa->ifa_local;
3628                                 return NOTIFY_OK;
3629                         case NETDEV_DOWN:
3630                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3631                                 return NOTIFY_OK;
3632                         default:
3633                                 return NOTIFY_DONE;
3634                         }
3635                 }
3636
3637                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3638                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3639                         if (vlan_dev == event_dev) {
3640                                 switch (event) {
3641                                 case NETDEV_UP:
3642                                         vlan->vlan_ip = ifa->ifa_local;
3643                                         return NOTIFY_OK;
3644                                 case NETDEV_DOWN:
3645                                         vlan->vlan_ip =
3646                                                 bond_glean_dev_ip(vlan_dev);
3647                                         return NOTIFY_OK;
3648                                 default:
3649                                         return NOTIFY_DONE;
3650                                 }
3651                         }
3652                 }
3653         }
3654         return NOTIFY_DONE;
3655 }
3656
3657 static struct notifier_block bond_netdev_notifier = {
3658         .notifier_call = bond_netdev_event,
3659 };
3660
3661 static struct notifier_block bond_inetaddr_notifier = {
3662         .notifier_call = bond_inetaddr_event,
3663 };
3664
3665 /*-------------------------- Packet type handling ---------------------------*/
3666
3667 /* register to receive lacpdus on a bond */
3668 static void bond_register_lacpdu(struct bonding *bond)
3669 {
3670         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3671
3672         /* initialize packet type */
3673         pk_type->type = PKT_TYPE_LACPDU;
3674         pk_type->dev = bond->dev;
3675         pk_type->func = bond_3ad_lacpdu_recv;
3676
3677         dev_add_pack(pk_type);
3678 }
3679
3680 /* unregister to receive lacpdus on a bond */
3681 static void bond_unregister_lacpdu(struct bonding *bond)
3682 {
3683         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3684 }
3685
3686 void bond_register_arp(struct bonding *bond)
3687 {
3688         struct packet_type *pt = &bond->arp_mon_pt;
3689
3690         if (pt->type)
3691                 return;
3692
3693         pt->type = htons(ETH_P_ARP);
3694         pt->dev = bond->dev;
3695         pt->func = bond_arp_rcv;
3696         dev_add_pack(pt);
3697 }
3698
3699 void bond_unregister_arp(struct bonding *bond)
3700 {
3701         struct packet_type *pt = &bond->arp_mon_pt;
3702
3703         dev_remove_pack(pt);
3704         pt->type = 0;
3705 }
3706
3707 /*---------------------------- Hashing Policies -----------------------------*/
3708
3709 /*
3710  * Hash for the output device based upon layer 2 and layer 3 data. If
3711  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3712  */
3713 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3714                                      struct net_device *bond_dev, int count)
3715 {
3716         struct ethhdr *data = (struct ethhdr *)skb->data;
3717         struct iphdr *iph = ip_hdr(skb);
3718
3719         if (skb->protocol == htons(ETH_P_IP)) {
3720                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3721                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3722         }
3723
3724         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3725 }
3726
3727 /*
3728  * Hash for the output device based upon layer 3 and layer 4 data. If
3729  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3730  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3731  */
3732 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3733                                     struct net_device *bond_dev, int count)
3734 {
3735         struct ethhdr *data = (struct ethhdr *)skb->data;
3736         struct iphdr *iph = ip_hdr(skb);
3737         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3738         int layer4_xor = 0;
3739
3740         if (skb->protocol == htons(ETH_P_IP)) {
3741                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3742                     (iph->protocol == IPPROTO_TCP ||
3743                      iph->protocol == IPPROTO_UDP)) {
3744                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3745                 }
3746                 return (layer4_xor ^
3747                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3748
3749         }
3750
3751         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3752 }
3753
3754 /*
3755  * Hash for the output device based upon layer 2 data
3756  */
3757 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3758                                    struct net_device *bond_dev, int count)
3759 {
3760         struct ethhdr *data = (struct ethhdr *)skb->data;
3761
3762         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3763 }
3764
3765 /*-------------------------- Device entry points ----------------------------*/
3766
3767 static int bond_open(struct net_device *bond_dev)
3768 {
3769         struct bonding *bond = netdev_priv(bond_dev);
3770
3771         bond->kill_timers = 0;
3772
3773         if (bond_is_lb(bond)) {
3774                 /* bond_alb_initialize must be called before the timer
3775                  * is started.
3776                  */
3777                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3778                         /* something went wrong - fail the open operation */
3779                         return -1;
3780                 }
3781
3782                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3783                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3784         }
3785
3786         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3787                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3788                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3789         }
3790
3791         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3792                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3793                         INIT_DELAYED_WORK(&bond->arp_work,
3794                                           bond_activebackup_arp_mon);
3795                 else
3796                         INIT_DELAYED_WORK(&bond->arp_work,
3797                                           bond_loadbalance_arp_mon);
3798
3799                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3800                 if (bond->params.arp_validate)
3801                         bond_register_arp(bond);
3802         }
3803
3804         if (bond->params.mode == BOND_MODE_8023AD) {
3805                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3806                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3807                 /* register to receive LACPDUs */
3808                 bond_register_lacpdu(bond);
3809                 bond_3ad_initiate_agg_selection(bond, 1);
3810         }
3811
3812         return 0;
3813 }
3814
3815 static int bond_close(struct net_device *bond_dev)
3816 {
3817         struct bonding *bond = netdev_priv(bond_dev);
3818
3819         if (bond->params.mode == BOND_MODE_8023AD) {
3820                 /* Unregister the receive of LACPDUs */
3821                 bond_unregister_lacpdu(bond);
3822         }
3823
3824         if (bond->params.arp_validate)
3825                 bond_unregister_arp(bond);
3826
3827         write_lock_bh(&bond->lock);
3828
3829         bond->send_grat_arp = 0;
3830         bond->send_unsol_na = 0;
3831
3832         /* signal timers not to re-arm */
3833         bond->kill_timers = 1;
3834
3835         write_unlock_bh(&bond->lock);
3836
3837         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3838                 cancel_delayed_work(&bond->mii_work);
3839         }
3840
3841         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3842                 cancel_delayed_work(&bond->arp_work);
3843         }
3844
3845         switch (bond->params.mode) {
3846         case BOND_MODE_8023AD:
3847                 cancel_delayed_work(&bond->ad_work);
3848                 break;
3849         case BOND_MODE_TLB:
3850         case BOND_MODE_ALB:
3851                 cancel_delayed_work(&bond->alb_work);
3852                 break;
3853         default:
3854                 break;
3855         }
3856
3857
3858         if (bond_is_lb(bond)) {
3859                 /* Must be called only after all
3860                  * slaves have been released
3861                  */
3862                 bond_alb_deinitialize(bond);
3863         }
3864
3865         return 0;
3866 }
3867
3868 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3869 {
3870         struct bonding *bond = netdev_priv(bond_dev);
3871         struct net_device_stats *stats = &bond->stats;
3872         struct net_device_stats local_stats;
3873         struct slave *slave;
3874         int i;
3875
3876         memset(&local_stats, 0, sizeof(struct net_device_stats));
3877
3878         read_lock_bh(&bond->lock);
3879
3880         bond_for_each_slave(bond, slave, i) {
3881                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3882
3883                 local_stats.rx_packets += sstats->rx_packets;
3884                 local_stats.rx_bytes += sstats->rx_bytes;
3885                 local_stats.rx_errors += sstats->rx_errors;
3886                 local_stats.rx_dropped += sstats->rx_dropped;
3887
3888                 local_stats.tx_packets += sstats->tx_packets;
3889                 local_stats.tx_bytes += sstats->tx_bytes;
3890                 local_stats.tx_errors += sstats->tx_errors;
3891                 local_stats.tx_dropped += sstats->tx_dropped;
3892
3893                 local_stats.multicast += sstats->multicast;
3894                 local_stats.collisions += sstats->collisions;
3895
3896                 local_stats.rx_length_errors += sstats->rx_length_errors;
3897                 local_stats.rx_over_errors += sstats->rx_over_errors;
3898                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3899                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3900                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3901                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3902
3903                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3904                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3905                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3906                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3907                 local_stats.tx_window_errors += sstats->tx_window_errors;
3908         }
3909
3910         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3911
3912         read_unlock_bh(&bond->lock);
3913
3914         return stats;
3915 }
3916
3917 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3918 {
3919         struct net_device *slave_dev = NULL;
3920         struct ifbond k_binfo;
3921         struct ifbond __user *u_binfo = NULL;
3922         struct ifslave k_sinfo;
3923         struct ifslave __user *u_sinfo = NULL;
3924         struct mii_ioctl_data *mii = NULL;
3925         int res = 0;
3926
3927         pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3928                 bond_dev->name, cmd);
3929
3930         switch (cmd) {
3931         case SIOCGMIIPHY:
3932                 mii = if_mii(ifr);
3933                 if (!mii) {
3934                         return -EINVAL;
3935                 }
3936                 mii->phy_id = 0;
3937                 /* Fall Through */
3938         case SIOCGMIIREG:
3939                 /*
3940                  * We do this again just in case we were called by SIOCGMIIREG
3941                  * instead of SIOCGMIIPHY.
3942                  */
3943                 mii = if_mii(ifr);
3944                 if (!mii) {
3945                         return -EINVAL;
3946                 }
3947
3948                 if (mii->reg_num == 1) {
3949                         struct bonding *bond = netdev_priv(bond_dev);
3950                         mii->val_out = 0;
3951                         read_lock(&bond->lock);
3952                         read_lock(&bond->curr_slave_lock);
3953                         if (netif_carrier_ok(bond->dev)) {
3954                                 mii->val_out = BMSR_LSTATUS;
3955                         }
3956                         read_unlock(&bond->curr_slave_lock);
3957                         read_unlock(&bond->lock);
3958                 }
3959
3960                 return 0;
3961         case BOND_INFO_QUERY_OLD:
3962         case SIOCBONDINFOQUERY:
3963                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3964
3965                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3966                         return -EFAULT;
3967                 }
3968
3969                 res = bond_info_query(bond_dev, &k_binfo);
3970                 if (res == 0) {
3971                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3972                                 return -EFAULT;
3973                         }
3974                 }
3975
3976                 return res;
3977         case BOND_SLAVE_INFO_QUERY_OLD:
3978         case SIOCBONDSLAVEINFOQUERY:
3979                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3980
3981                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3982                         return -EFAULT;
3983                 }
3984
3985                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3986                 if (res == 0) {
3987                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3988                                 return -EFAULT;
3989                         }
3990                 }
3991
3992                 return res;
3993         default:
3994                 /* Go on */
3995                 break;
3996         }
3997
3998         if (!capable(CAP_NET_ADMIN)) {
3999                 return -EPERM;
4000         }
4001
4002         down_write(&(bonding_rwsem));
4003         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
4004
4005         pr_debug("slave_dev=%p: \n", slave_dev);
4006
4007         if (!slave_dev) {
4008                 res = -ENODEV;
4009         } else {
4010                 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
4011                 switch (cmd) {
4012                 case BOND_ENSLAVE_OLD:
4013                 case SIOCBONDENSLAVE:
4014                         res = bond_enslave(bond_dev, slave_dev);
4015                         break;
4016                 case BOND_RELEASE_OLD:
4017                 case SIOCBONDRELEASE:
4018                         res = bond_release(bond_dev, slave_dev);
4019                         break;
4020                 case BOND_SETHWADDR_OLD:
4021                 case SIOCBONDSETHWADDR:
4022                         res = bond_sethwaddr(bond_dev, slave_dev);
4023                         break;
4024                 case BOND_CHANGE_ACTIVE_OLD:
4025                 case SIOCBONDCHANGEACTIVE:
4026                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4027                         break;
4028                 default:
4029                         res = -EOPNOTSUPP;
4030                 }
4031
4032                 dev_put(slave_dev);
4033         }
4034
4035         up_write(&(bonding_rwsem));
4036         return res;
4037 }
4038
4039 static void bond_set_multicast_list(struct net_device *bond_dev)
4040 {
4041         struct bonding *bond = netdev_priv(bond_dev);
4042         struct dev_mc_list *dmi;
4043
4044         /*
4045          * Do promisc before checking multicast_mode
4046          */
4047         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4048                 /*
4049                  * FIXME: Need to handle the error when one of the multi-slaves
4050                  * encounters error.
4051                  */
4052                 bond_set_promiscuity(bond, 1);
4053         }
4054
4055         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4056                 bond_set_promiscuity(bond, -1);
4057         }
4058
4059         /* set allmulti flag to slaves */
4060         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4061                 /*
4062                  * FIXME: Need to handle the error when one of the multi-slaves
4063                  * encounters error.
4064                  */
4065                 bond_set_allmulti(bond, 1);
4066         }
4067
4068         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4069                 bond_set_allmulti(bond, -1);
4070         }
4071
4072         read_lock(&bond->lock);
4073
4074         bond->flags = bond_dev->flags;
4075
4076         /* looking for addresses to add to slaves' mc list */
4077         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4078                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4079                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4080                 }
4081         }
4082
4083         /* looking for addresses to delete from slaves' list */
4084         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4085                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4086                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4087                 }
4088         }
4089
4090         /* save master's multicast list */
4091         bond_mc_list_destroy(bond);
4092         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4093
4094         read_unlock(&bond->lock);
4095 }
4096
4097 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4098 {
4099         struct bonding *bond = netdev_priv(dev);
4100         struct slave *slave = bond->first_slave;
4101
4102         if (slave) {
4103                 const struct net_device_ops *slave_ops
4104                         = slave->dev->netdev_ops;
4105                 if (slave_ops->ndo_neigh_setup)
4106                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4107         }
4108         return 0;
4109 }
4110
4111 /*
4112  * Change the MTU of all of a master's slaves to match the master
4113  */
4114 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4115 {
4116         struct bonding *bond = netdev_priv(bond_dev);
4117         struct slave *slave, *stop_at;
4118         int res = 0;
4119         int i;
4120
4121         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4122                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4123
4124         /* Can't hold bond->lock with bh disabled here since
4125          * some base drivers panic. On the other hand we can't
4126          * hold bond->lock without bh disabled because we'll
4127          * deadlock. The only solution is to rely on the fact
4128          * that we're under rtnl_lock here, and the slaves
4129          * list won't change. This doesn't solve the problem
4130          * of setting the slave's MTU while it is
4131          * transmitting, but the assumption is that the base
4132          * driver can handle that.
4133          *
4134          * TODO: figure out a way to safely iterate the slaves
4135          * list, but without holding a lock around the actual
4136          * call to the base driver.
4137          */
4138
4139         bond_for_each_slave(bond, slave, i) {
4140                 pr_debug("s %p s->p %p c_m %p\n", slave,
4141                         slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
4142
4143                 res = dev_set_mtu(slave->dev, new_mtu);
4144
4145                 if (res) {
4146                         /* If we failed to set the slave's mtu to the new value
4147                          * we must abort the operation even in ACTIVE_BACKUP
4148                          * mode, because if we allow the backup slaves to have
4149                          * different mtu values than the active slave we'll
4150                          * need to change their mtu when doing a failover. That
4151                          * means changing their mtu from timer context, which
4152                          * is probably not a good idea.
4153                          */
4154                         pr_debug("err %d %s\n", res, slave->dev->name);
4155                         goto unwind;
4156                 }
4157         }
4158
4159         bond_dev->mtu = new_mtu;
4160
4161         return 0;
4162
4163 unwind:
4164         /* unwind from head to the slave that failed */
4165         stop_at = slave;
4166         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4167                 int tmp_res;
4168
4169                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4170                 if (tmp_res) {
4171                         pr_debug("unwind err %d dev %s\n", tmp_res,
4172                                 slave->dev->name);
4173                 }
4174         }
4175
4176         return res;
4177 }
4178
4179 /*
4180  * Change HW address
4181  *
4182  * Note that many devices must be down to change the HW address, and
4183  * downing the master releases all slaves.  We can make bonds full of
4184  * bonding devices to test this, however.
4185  */
4186 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4187 {
4188         struct bonding *bond = netdev_priv(bond_dev);
4189         struct sockaddr *sa = addr, tmp_sa;
4190         struct slave *slave, *stop_at;
4191         int res = 0;
4192         int i;
4193
4194         if (bond->params.mode == BOND_MODE_ALB)
4195                 return bond_alb_set_mac_address(bond_dev, addr);
4196
4197
4198         pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4199
4200         /*
4201          * If fail_over_mac is set to active, do nothing and return
4202          * success.  Returning an error causes ifenslave to fail.
4203          */
4204         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4205                 return 0;
4206
4207         if (!is_valid_ether_addr(sa->sa_data)) {
4208                 return -EADDRNOTAVAIL;
4209         }
4210
4211         /* Can't hold bond->lock with bh disabled here since
4212          * some base drivers panic. On the other hand we can't
4213          * hold bond->lock without bh disabled because we'll
4214          * deadlock. The only solution is to rely on the fact
4215          * that we're under rtnl_lock here, and the slaves
4216          * list won't change. This doesn't solve the problem
4217          * of setting the slave's hw address while it is
4218          * transmitting, but the assumption is that the base
4219          * driver can handle that.
4220          *
4221          * TODO: figure out a way to safely iterate the slaves
4222          * list, but without holding a lock around the actual
4223          * call to the base driver.
4224          */
4225
4226         bond_for_each_slave(bond, slave, i) {
4227                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4228                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4229
4230                 if (slave_ops->ndo_set_mac_address == NULL) {
4231                         res = -EOPNOTSUPP;
4232                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4233                         goto unwind;
4234                 }
4235
4236                 res = dev_set_mac_address(slave->dev, addr);
4237                 if (res) {
4238                         /* TODO: consider downing the slave
4239                          * and retry ?
4240                          * User should expect communications
4241                          * breakage anyway until ARP finish
4242                          * updating, so...
4243                          */
4244                         pr_debug("err %d %s\n", res, slave->dev->name);
4245                         goto unwind;
4246                 }
4247         }
4248
4249         /* success */
4250         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4251         return 0;
4252
4253 unwind:
4254         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4255         tmp_sa.sa_family = bond_dev->type;
4256
4257         /* unwind from head to the slave that failed */
4258         stop_at = slave;
4259         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4260                 int tmp_res;
4261
4262                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4263                 if (tmp_res) {
4264                         pr_debug("unwind err %d dev %s\n", tmp_res,
4265                                 slave->dev->name);
4266                 }
4267         }
4268
4269         return res;
4270 }
4271
4272 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4273 {
4274         struct bonding *bond = netdev_priv(bond_dev);
4275         struct slave *slave, *start_at;
4276         int i, slave_no, res = 1;
4277
4278         read_lock(&bond->lock);
4279
4280         if (!BOND_IS_OK(bond)) {
4281                 goto out;
4282         }
4283
4284         /*
4285          * Concurrent TX may collide on rr_tx_counter; we accept that
4286          * as being rare enough not to justify using an atomic op here
4287          */
4288         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4289
4290         bond_for_each_slave(bond, slave, i) {
4291                 slave_no--;
4292                 if (slave_no < 0) {
4293                         break;
4294                 }
4295         }
4296
4297         start_at = slave;
4298         bond_for_each_slave_from(bond, slave, i, start_at) {
4299                 if (IS_UP(slave->dev) &&
4300                     (slave->link == BOND_LINK_UP) &&
4301                     (slave->state == BOND_STATE_ACTIVE)) {
4302                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4303                         break;
4304                 }
4305         }
4306
4307 out:
4308         if (res) {
4309                 /* no suitable interface, frame not sent */
4310                 dev_kfree_skb(skb);
4311         }
4312         read_unlock(&bond->lock);
4313         return 0;
4314 }
4315
4316
4317 /*
4318  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4319  * the bond has a usable interface.
4320  */
4321 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4322 {
4323         struct bonding *bond = netdev_priv(bond_dev);
4324         int res = 1;
4325
4326         read_lock(&bond->lock);
4327         read_lock(&bond->curr_slave_lock);
4328
4329         if (!BOND_IS_OK(bond)) {
4330                 goto out;
4331         }
4332
4333         if (!bond->curr_active_slave)
4334                 goto out;
4335
4336         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4337
4338 out:
4339         if (res) {
4340                 /* no suitable interface, frame not sent */
4341                 dev_kfree_skb(skb);
4342         }
4343         read_unlock(&bond->curr_slave_lock);
4344         read_unlock(&bond->lock);
4345         return 0;
4346 }
4347
4348 /*
4349  * In bond_xmit_xor() , we determine the output device by using a pre-
4350  * determined xmit_hash_policy(), If the selected device is not enabled,
4351  * find the next active slave.
4352  */
4353 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4354 {
4355         struct bonding *bond = netdev_priv(bond_dev);
4356         struct slave *slave, *start_at;
4357         int slave_no;
4358         int i;
4359         int res = 1;
4360
4361         read_lock(&bond->lock);
4362
4363         if (!BOND_IS_OK(bond)) {
4364                 goto out;
4365         }
4366
4367         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4368
4369         bond_for_each_slave(bond, slave, i) {
4370                 slave_no--;
4371                 if (slave_no < 0) {
4372                         break;
4373                 }
4374         }
4375
4376         start_at = slave;
4377
4378         bond_for_each_slave_from(bond, slave, i, start_at) {
4379                 if (IS_UP(slave->dev) &&
4380                     (slave->link == BOND_LINK_UP) &&
4381                     (slave->state == BOND_STATE_ACTIVE)) {
4382                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4383                         break;
4384                 }
4385         }
4386
4387 out:
4388         if (res) {
4389                 /* no suitable interface, frame not sent */
4390                 dev_kfree_skb(skb);
4391         }
4392         read_unlock(&bond->lock);
4393         return 0;
4394 }
4395
4396 /*
4397  * in broadcast mode, we send everything to all usable interfaces.
4398  */
4399 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4400 {
4401         struct bonding *bond = netdev_priv(bond_dev);
4402         struct slave *slave, *start_at;
4403         struct net_device *tx_dev = NULL;
4404         int i;
4405         int res = 1;
4406
4407         read_lock(&bond->lock);
4408
4409         if (!BOND_IS_OK(bond)) {
4410                 goto out;
4411         }
4412
4413         read_lock(&bond->curr_slave_lock);
4414         start_at = bond->curr_active_slave;
4415         read_unlock(&bond->curr_slave_lock);
4416
4417         if (!start_at) {
4418                 goto out;
4419         }
4420
4421         bond_for_each_slave_from(bond, slave, i, start_at) {
4422                 if (IS_UP(slave->dev) &&
4423                     (slave->link == BOND_LINK_UP) &&
4424                     (slave->state == BOND_STATE_ACTIVE)) {
4425                         if (tx_dev) {
4426                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4427                                 if (!skb2) {
4428                                         printk(KERN_ERR DRV_NAME
4429                                                ": %s: Error: bond_xmit_broadcast(): "
4430                                                "skb_clone() failed\n",
4431                                                bond_dev->name);
4432                                         continue;
4433                                 }
4434
4435                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4436                                 if (res) {
4437                                         dev_kfree_skb(skb2);
4438                                         continue;
4439                                 }
4440                         }
4441                         tx_dev = slave->dev;
4442                 }
4443         }
4444
4445         if (tx_dev) {
4446                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4447         }
4448
4449 out:
4450         if (res) {
4451                 /* no suitable interface, frame not sent */
4452                 dev_kfree_skb(skb);
4453         }
4454         /* frame sent to all suitable interfaces */
4455         read_unlock(&bond->lock);
4456         return 0;
4457 }
4458
4459 /*------------------------- Device initialization ---------------------------*/
4460
4461 static void bond_set_xmit_hash_policy(struct bonding *bond)
4462 {
4463         switch (bond->params.xmit_policy) {
4464         case BOND_XMIT_POLICY_LAYER23:
4465                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4466                 break;
4467         case BOND_XMIT_POLICY_LAYER34:
4468                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4469                 break;
4470         case BOND_XMIT_POLICY_LAYER2:
4471         default:
4472                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4473                 break;
4474         }
4475 }
4476
4477 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4478 {
4479         const struct bonding *bond = netdev_priv(dev);
4480
4481         switch (bond->params.mode) {
4482         case BOND_MODE_ROUNDROBIN:
4483                 return bond_xmit_roundrobin(skb, dev);
4484         case BOND_MODE_ACTIVEBACKUP:
4485                 return bond_xmit_activebackup(skb, dev);
4486         case BOND_MODE_XOR:
4487                 return bond_xmit_xor(skb, dev);
4488         case BOND_MODE_BROADCAST:
4489                 return bond_xmit_broadcast(skb, dev);
4490         case BOND_MODE_8023AD:
4491                 return bond_3ad_xmit_xor(skb, dev);
4492         case BOND_MODE_ALB:
4493         case BOND_MODE_TLB:
4494                 return bond_alb_xmit(skb, dev);
4495         default:
4496                 /* Should never happen, mode already checked */
4497                 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4498                      dev->name, bond->params.mode);
4499                 WARN_ON_ONCE(1);
4500                 dev_kfree_skb(skb);
4501                 return NETDEV_TX_OK;
4502         }
4503 }
4504
4505
4506 /*
4507  * set bond mode specific net device operations
4508  */
4509 void bond_set_mode_ops(struct bonding *bond, int mode)
4510 {
4511         struct net_device *bond_dev = bond->dev;
4512
4513         switch (mode) {
4514         case BOND_MODE_ROUNDROBIN:
4515                 break;
4516         case BOND_MODE_ACTIVEBACKUP:
4517                 break;
4518         case BOND_MODE_XOR:
4519                 bond_set_xmit_hash_policy(bond);
4520                 break;
4521         case BOND_MODE_BROADCAST:
4522                 break;
4523         case BOND_MODE_8023AD:
4524                 bond_set_master_3ad_flags(bond);
4525                 bond_set_xmit_hash_policy(bond);
4526                 break;
4527         case BOND_MODE_ALB:
4528                 bond_set_master_alb_flags(bond);
4529                 /* FALLTHRU */
4530         case BOND_MODE_TLB:
4531                 break;
4532         default:
4533                 /* Should never happen, mode already checked */
4534                 printk(KERN_ERR DRV_NAME
4535                        ": %s: Error: Unknown bonding mode %d\n",
4536                        bond_dev->name,
4537                        mode);
4538                 break;
4539         }
4540 }
4541
4542 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4543                                     struct ethtool_drvinfo *drvinfo)
4544 {
4545         strncpy(drvinfo->driver, DRV_NAME, 32);
4546         strncpy(drvinfo->version, DRV_VERSION, 32);
4547         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4548 }
4549
4550 static const struct ethtool_ops bond_ethtool_ops = {
4551         .get_drvinfo            = bond_ethtool_get_drvinfo,
4552         .get_link               = ethtool_op_get_link,
4553         .get_tx_csum            = ethtool_op_get_tx_csum,
4554         .get_sg                 = ethtool_op_get_sg,
4555         .get_tso                = ethtool_op_get_tso,
4556         .get_ufo                = ethtool_op_get_ufo,
4557         .get_flags              = ethtool_op_get_flags,
4558 };
4559
4560 static const struct net_device_ops bond_netdev_ops = {
4561         .ndo_open               = bond_open,
4562         .ndo_stop               = bond_close,
4563         .ndo_start_xmit         = bond_start_xmit,
4564         .ndo_get_stats          = bond_get_stats,
4565         .ndo_do_ioctl           = bond_do_ioctl,
4566         .ndo_set_multicast_list = bond_set_multicast_list,
4567         .ndo_change_mtu         = bond_change_mtu,
4568         .ndo_set_mac_address    = bond_set_mac_address,
4569         .ndo_neigh_setup        = bond_neigh_setup,
4570         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4571         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4572         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4573 };
4574
4575 /*
4576  * Does not allocate but creates a /proc entry.
4577  * Allowed to fail.
4578  */
4579 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4580 {
4581         struct bonding *bond = netdev_priv(bond_dev);
4582
4583         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4584
4585         /* initialize rwlocks */
4586         rwlock_init(&bond->lock);
4587         rwlock_init(&bond->curr_slave_lock);
4588
4589         bond->params = *params; /* copy params struct */
4590
4591         bond->wq = create_singlethread_workqueue(bond_dev->name);
4592         if (!bond->wq)
4593                 return -ENOMEM;
4594
4595         /* Initialize pointers */
4596         bond->first_slave = NULL;
4597         bond->curr_active_slave = NULL;
4598         bond->current_arp_slave = NULL;
4599         bond->primary_slave = NULL;
4600         bond->dev = bond_dev;
4601         bond->send_grat_arp = 0;
4602         bond->send_unsol_na = 0;
4603         bond->setup_by_slave = 0;
4604         INIT_LIST_HEAD(&bond->vlan_list);
4605
4606         /* Initialize the device entry points */
4607         bond_dev->netdev_ops = &bond_netdev_ops;
4608         bond_dev->ethtool_ops = &bond_ethtool_ops;
4609         bond_set_mode_ops(bond, bond->params.mode);
4610
4611         bond_dev->destructor = bond_destructor;
4612
4613         /* Initialize the device options */
4614         bond_dev->tx_queue_len = 0;
4615         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4616         bond_dev->priv_flags |= IFF_BONDING;
4617         if (bond->params.arp_interval)
4618                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4619
4620         /* At first, we block adding VLANs. That's the only way to
4621          * prevent problems that occur when adding VLANs over an
4622          * empty bond. The block will be removed once non-challenged
4623          * slaves are enslaved.
4624          */
4625         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4626
4627         /* don't acquire bond device's netif_tx_lock when
4628          * transmitting */
4629         bond_dev->features |= NETIF_F_LLTX;
4630
4631         /* By default, we declare the bond to be fully
4632          * VLAN hardware accelerated capable. Special
4633          * care is taken in the various xmit functions
4634          * when there are slaves that are not hw accel
4635          * capable
4636          */
4637         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4638                                NETIF_F_HW_VLAN_RX |
4639                                NETIF_F_HW_VLAN_FILTER);
4640
4641 #ifdef CONFIG_PROC_FS
4642         bond_create_proc_entry(bond);
4643 #endif
4644         list_add_tail(&bond->bond_list, &bond_dev_list);
4645
4646         return 0;
4647 }
4648
4649 static void bond_work_cancel_all(struct bonding *bond)
4650 {
4651         write_lock_bh(&bond->lock);
4652         bond->kill_timers = 1;
4653         write_unlock_bh(&bond->lock);
4654
4655         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4656                 cancel_delayed_work(&bond->mii_work);
4657
4658         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4659                 cancel_delayed_work(&bond->arp_work);
4660
4661         if (bond->params.mode == BOND_MODE_ALB &&
4662             delayed_work_pending(&bond->alb_work))
4663                 cancel_delayed_work(&bond->alb_work);
4664
4665         if (bond->params.mode == BOND_MODE_8023AD &&
4666             delayed_work_pending(&bond->ad_work))
4667                 cancel_delayed_work(&bond->ad_work);
4668 }
4669
4670 /* De-initialize device specific data.
4671  * Caller must hold rtnl_lock.
4672  */
4673 static void bond_deinit(struct net_device *bond_dev)
4674 {
4675         struct bonding *bond = netdev_priv(bond_dev);
4676
4677         list_del(&bond->bond_list);
4678
4679         bond_work_cancel_all(bond);
4680
4681 #ifdef CONFIG_PROC_FS
4682         bond_remove_proc_entry(bond);
4683 #endif
4684 }
4685
4686 /* Unregister and free all bond devices.
4687  * Caller must hold rtnl_lock.
4688  */
4689 static void bond_free_all(void)
4690 {
4691         struct bonding *bond, *nxt;
4692
4693         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4694                 struct net_device *bond_dev = bond->dev;
4695
4696                 bond_work_cancel_all(bond);
4697                 /* Release the bonded slaves */
4698                 bond_release_all(bond_dev);
4699                 bond_destroy(bond);
4700         }
4701
4702 #ifdef CONFIG_PROC_FS
4703         bond_destroy_proc_dir();
4704 #endif
4705 }
4706
4707 /*------------------------- Module initialization ---------------------------*/
4708
4709 /*
4710  * Convert string input module parms.  Accept either the
4711  * number of the mode or its string name.  A bit complicated because
4712  * some mode names are substrings of other names, and calls from sysfs
4713  * may have whitespace in the name (trailing newlines, for example).
4714  */
4715 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4716 {
4717         int modeint = -1, i, rv;
4718         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4719
4720         for (p = (char *)buf; *p; p++)
4721                 if (!(isdigit(*p) || isspace(*p)))
4722                         break;
4723
4724         if (*p)
4725                 rv = sscanf(buf, "%20s", modestr);
4726         else
4727                 rv = sscanf(buf, "%d", &modeint);
4728
4729         if (!rv)
4730                 return -1;
4731
4732         for (i = 0; tbl[i].modename; i++) {
4733                 if (modeint == tbl[i].mode)
4734                         return tbl[i].mode;
4735                 if (strcmp(modestr, tbl[i].modename) == 0)
4736                         return tbl[i].mode;
4737         }
4738
4739         return -1;
4740 }
4741
4742 static int bond_check_params(struct bond_params *params)
4743 {
4744         int arp_validate_value, fail_over_mac_value;
4745
4746         /*
4747          * Convert string parameters.
4748          */
4749         if (mode) {
4750                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4751                 if (bond_mode == -1) {
4752                         printk(KERN_ERR DRV_NAME
4753                                ": Error: Invalid bonding mode \"%s\"\n",
4754                                mode == NULL ? "NULL" : mode);
4755                         return -EINVAL;
4756                 }
4757         }
4758
4759         if (xmit_hash_policy) {
4760                 if ((bond_mode != BOND_MODE_XOR) &&
4761                     (bond_mode != BOND_MODE_8023AD)) {
4762                         printk(KERN_INFO DRV_NAME
4763                                ": xor_mode param is irrelevant in mode %s\n",
4764                                bond_mode_name(bond_mode));
4765                 } else {
4766                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4767                                                         xmit_hashtype_tbl);
4768                         if (xmit_hashtype == -1) {
4769                                 printk(KERN_ERR DRV_NAME
4770                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4771                                 xmit_hash_policy == NULL ? "NULL" :
4772                                        xmit_hash_policy);
4773                                 return -EINVAL;
4774                         }
4775                 }
4776         }
4777
4778         if (lacp_rate) {
4779                 if (bond_mode != BOND_MODE_8023AD) {
4780                         printk(KERN_INFO DRV_NAME
4781                                ": lacp_rate param is irrelevant in mode %s\n",
4782                                bond_mode_name(bond_mode));
4783                 } else {
4784                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4785                         if (lacp_fast == -1) {
4786                                 printk(KERN_ERR DRV_NAME
4787                                        ": Error: Invalid lacp rate \"%s\"\n",
4788                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4789                                 return -EINVAL;
4790                         }
4791                 }
4792         }
4793
4794         if (ad_select) {
4795                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4796                 if (params->ad_select == -1) {
4797                         printk(KERN_ERR DRV_NAME
4798                                ": Error: Invalid ad_select \"%s\"\n",
4799                                ad_select == NULL ? "NULL" : ad_select);
4800                         return -EINVAL;
4801                 }
4802
4803                 if (bond_mode != BOND_MODE_8023AD) {
4804                         printk(KERN_WARNING DRV_NAME
4805                                ": ad_select param only affects 802.3ad mode\n");
4806                 }
4807         } else {
4808                 params->ad_select = BOND_AD_STABLE;
4809         }
4810
4811         if (max_bonds < 0 || max_bonds > INT_MAX) {
4812                 printk(KERN_WARNING DRV_NAME
4813                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4814                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4815                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4816                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4817         }
4818
4819         if (miimon < 0) {
4820                 printk(KERN_WARNING DRV_NAME
4821                        ": Warning: miimon module parameter (%d), "
4822                        "not in range 0-%d, so it was reset to %d\n",
4823                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4824                 miimon = BOND_LINK_MON_INTERV;
4825         }
4826
4827         if (updelay < 0) {
4828                 printk(KERN_WARNING DRV_NAME
4829                        ": Warning: updelay module parameter (%d), "
4830                        "not in range 0-%d, so it was reset to 0\n",
4831                        updelay, INT_MAX);
4832                 updelay = 0;
4833         }
4834
4835         if (downdelay < 0) {
4836                 printk(KERN_WARNING DRV_NAME
4837                        ": Warning: downdelay module parameter (%d), "
4838                        "not in range 0-%d, so it was reset to 0\n",
4839                        downdelay, INT_MAX);
4840                 downdelay = 0;
4841         }
4842
4843         if ((use_carrier != 0) && (use_carrier != 1)) {
4844                 printk(KERN_WARNING DRV_NAME
4845                        ": Warning: use_carrier module parameter (%d), "
4846                        "not of valid value (0/1), so it was set to 1\n",
4847                        use_carrier);
4848                 use_carrier = 1;
4849         }
4850
4851         if (num_grat_arp < 0 || num_grat_arp > 255) {
4852                 printk(KERN_WARNING DRV_NAME
4853                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4854                        "was reset to 1 \n", num_grat_arp);
4855                 num_grat_arp = 1;
4856         }
4857
4858         if (num_unsol_na < 0 || num_unsol_na > 255) {
4859                 printk(KERN_WARNING DRV_NAME
4860                        ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4861                        "was reset to 1 \n", num_unsol_na);
4862                 num_unsol_na = 1;
4863         }
4864
4865         /* reset values for 802.3ad */
4866         if (bond_mode == BOND_MODE_8023AD) {
4867                 if (!miimon) {
4868                         printk(KERN_WARNING DRV_NAME
4869                                ": Warning: miimon must be specified, "
4870                                "otherwise bonding will not detect link "
4871                                "failure, speed and duplex which are "
4872                                "essential for 802.3ad operation\n");
4873                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4874                         miimon = 100;
4875                 }
4876         }
4877
4878         /* reset values for TLB/ALB */
4879         if ((bond_mode == BOND_MODE_TLB) ||
4880             (bond_mode == BOND_MODE_ALB)) {
4881                 if (!miimon) {
4882                         printk(KERN_WARNING DRV_NAME
4883                                ": Warning: miimon must be specified, "
4884                                "otherwise bonding will not detect link "
4885                                "failure and link speed which are essential "
4886                                "for TLB/ALB load balancing\n");
4887                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4888                         miimon = 100;
4889                 }
4890         }
4891
4892         if (bond_mode == BOND_MODE_ALB) {
4893                 printk(KERN_NOTICE DRV_NAME
4894                        ": In ALB mode you might experience client "
4895                        "disconnections upon reconnection of a link if the "
4896                        "bonding module updelay parameter (%d msec) is "
4897                        "incompatible with the forwarding delay time of the "
4898                        "switch\n",
4899                        updelay);
4900         }
4901
4902         if (!miimon) {
4903                 if (updelay || downdelay) {
4904                         /* just warn the user the up/down delay will have
4905                          * no effect since miimon is zero...
4906                          */
4907                         printk(KERN_WARNING DRV_NAME
4908                                ": Warning: miimon module parameter not set "
4909                                "and updelay (%d) or downdelay (%d) module "
4910                                "parameter is set; updelay and downdelay have "
4911                                "no effect unless miimon is set\n",
4912                                updelay, downdelay);
4913                 }
4914         } else {
4915                 /* don't allow arp monitoring */
4916                 if (arp_interval) {
4917                         printk(KERN_WARNING DRV_NAME
4918                                ": Warning: miimon (%d) and arp_interval (%d) "
4919                                "can't be used simultaneously, disabling ARP "
4920                                "monitoring\n",
4921                                miimon, arp_interval);
4922                         arp_interval = 0;
4923                 }
4924
4925                 if ((updelay % miimon) != 0) {
4926                         printk(KERN_WARNING DRV_NAME
4927                                ": Warning: updelay (%d) is not a multiple "
4928                                "of miimon (%d), updelay rounded to %d ms\n",
4929                                updelay, miimon, (updelay / miimon) * miimon);
4930                 }
4931
4932                 updelay /= miimon;
4933
4934                 if ((downdelay % miimon) != 0) {
4935                         printk(KERN_WARNING DRV_NAME
4936                                ": Warning: downdelay (%d) is not a multiple "
4937                                "of miimon (%d), downdelay rounded to %d ms\n",
4938                                downdelay, miimon,
4939                                (downdelay / miimon) * miimon);
4940                 }
4941
4942                 downdelay /= miimon;
4943         }
4944
4945         if (arp_interval < 0) {
4946                 printk(KERN_WARNING DRV_NAME
4947                        ": Warning: arp_interval module parameter (%d) "
4948                        ", not in range 0-%d, so it was reset to %d\n",
4949                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4950                 arp_interval = BOND_LINK_ARP_INTERV;
4951         }
4952
4953         for (arp_ip_count = 0;
4954              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4955              arp_ip_count++) {
4956                 /* not complete check, but should be good enough to
4957                    catch mistakes */
4958                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4959                         printk(KERN_WARNING DRV_NAME
4960                                ": Warning: bad arp_ip_target module parameter "
4961                                "(%s), ARP monitoring will not be performed\n",
4962                                arp_ip_target[arp_ip_count]);
4963                         arp_interval = 0;
4964                 } else {
4965                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4966                         arp_target[arp_ip_count] = ip;
4967                 }
4968         }
4969
4970         if (arp_interval && !arp_ip_count) {
4971                 /* don't allow arping if no arp_ip_target given... */
4972                 printk(KERN_WARNING DRV_NAME
4973                        ": Warning: arp_interval module parameter (%d) "
4974                        "specified without providing an arp_ip_target "
4975                        "parameter, arp_interval was reset to 0\n",
4976                        arp_interval);
4977                 arp_interval = 0;
4978         }
4979
4980         if (arp_validate) {
4981                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4982                         printk(KERN_ERR DRV_NAME
4983                ": arp_validate only supported in active-backup mode\n");
4984                         return -EINVAL;
4985                 }
4986                 if (!arp_interval) {
4987                         printk(KERN_ERR DRV_NAME
4988                                ": arp_validate requires arp_interval\n");
4989                         return -EINVAL;
4990                 }
4991
4992                 arp_validate_value = bond_parse_parm(arp_validate,
4993                                                      arp_validate_tbl);
4994                 if (arp_validate_value == -1) {
4995                         printk(KERN_ERR DRV_NAME
4996                                ": Error: invalid arp_validate \"%s\"\n",
4997                                arp_validate == NULL ? "NULL" : arp_validate);
4998                         return -EINVAL;
4999                 }
5000         } else
5001                 arp_validate_value = 0;
5002
5003         if (miimon) {
5004                 printk(KERN_INFO DRV_NAME
5005                        ": MII link monitoring set to %d ms\n",
5006                        miimon);
5007         } else if (arp_interval) {
5008                 int i;
5009
5010                 printk(KERN_INFO DRV_NAME
5011                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5012                        arp_interval,
5013                        arp_validate_tbl[arp_validate_value].modename,
5014                        arp_ip_count);
5015
5016                 for (i = 0; i < arp_ip_count; i++)
5017                         printk (" %s", arp_ip_target[i]);
5018
5019                 printk("\n");
5020
5021         } else if (max_bonds) {
5022                 /* miimon and arp_interval not set, we need one so things
5023                  * work as expected, see bonding.txt for details
5024                  */
5025                 printk(KERN_WARNING DRV_NAME
5026                        ": Warning: either miimon or arp_interval and "
5027                        "arp_ip_target module parameters must be specified, "
5028                        "otherwise bonding will not detect link failures! see "
5029                        "bonding.txt for details.\n");
5030         }
5031
5032         if (primary && !USES_PRIMARY(bond_mode)) {
5033                 /* currently, using a primary only makes sense
5034                  * in active backup, TLB or ALB modes
5035                  */
5036                 printk(KERN_WARNING DRV_NAME
5037                        ": Warning: %s primary device specified but has no "
5038                        "effect in %s mode\n",
5039                        primary, bond_mode_name(bond_mode));
5040                 primary = NULL;
5041         }
5042
5043         if (fail_over_mac) {
5044                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5045                                                       fail_over_mac_tbl);
5046                 if (fail_over_mac_value == -1) {
5047                         printk(KERN_ERR DRV_NAME
5048                                ": Error: invalid fail_over_mac \"%s\"\n",
5049                                arp_validate == NULL ? "NULL" : arp_validate);
5050                         return -EINVAL;
5051                 }
5052
5053                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5054                         printk(KERN_WARNING DRV_NAME
5055                                ": Warning: fail_over_mac only affects "
5056                                "active-backup mode.\n");
5057         } else {
5058                 fail_over_mac_value = BOND_FOM_NONE;
5059         }
5060
5061         /* fill params struct with the proper values */
5062         params->mode = bond_mode;
5063         params->xmit_policy = xmit_hashtype;
5064         params->miimon = miimon;
5065         params->num_grat_arp = num_grat_arp;
5066         params->num_unsol_na = num_unsol_na;
5067         params->arp_interval = arp_interval;
5068         params->arp_validate = arp_validate_value;
5069         params->updelay = updelay;
5070         params->downdelay = downdelay;
5071         params->use_carrier = use_carrier;
5072         params->lacp_fast = lacp_fast;
5073         params->primary[0] = 0;
5074         params->fail_over_mac = fail_over_mac_value;
5075
5076         if (primary) {
5077                 strncpy(params->primary, primary, IFNAMSIZ);
5078                 params->primary[IFNAMSIZ - 1] = 0;
5079         }
5080
5081         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5082
5083         return 0;
5084 }
5085
5086 static struct lock_class_key bonding_netdev_xmit_lock_key;
5087 static struct lock_class_key bonding_netdev_addr_lock_key;
5088
5089 static void bond_set_lockdep_class_one(struct net_device *dev,
5090                                        struct netdev_queue *txq,
5091                                        void *_unused)
5092 {
5093         lockdep_set_class(&txq->_xmit_lock,
5094                           &bonding_netdev_xmit_lock_key);
5095 }
5096
5097 static void bond_set_lockdep_class(struct net_device *dev)
5098 {
5099         lockdep_set_class(&dev->addr_list_lock,
5100                           &bonding_netdev_addr_lock_key);
5101         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5102 }
5103
5104 /* Create a new bond based on the specified name and bonding parameters.
5105  * If name is NULL, obtain a suitable "bond%d" name for us.
5106  * Caller must NOT hold rtnl_lock; we need to release it here before we
5107  * set up our sysfs entries.
5108  */
5109 int bond_create(char *name, struct bond_params *params)
5110 {
5111         struct net_device *bond_dev;
5112         struct bonding *bond;
5113         int res;
5114
5115         rtnl_lock();
5116         down_write(&bonding_rwsem);
5117
5118         /* Check to see if the bond already exists. */
5119         if (name) {
5120                 list_for_each_entry(bond, &bond_dev_list, bond_list)
5121                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5122                                 printk(KERN_ERR DRV_NAME
5123                                ": cannot add bond %s; it already exists\n",
5124                                        name);
5125                                 res = -EPERM;
5126                                 goto out_rtnl;
5127                         }
5128         }
5129
5130         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5131                                 ether_setup);
5132         if (!bond_dev) {
5133                 printk(KERN_ERR DRV_NAME
5134                        ": %s: eek! can't alloc netdev!\n",
5135                        name);
5136                 res = -ENOMEM;
5137                 goto out_rtnl;
5138         }
5139
5140         if (!name) {
5141                 res = dev_alloc_name(bond_dev, "bond%d");
5142                 if (res < 0)
5143                         goto out_netdev;
5144         }
5145
5146         /* bond_init() must be called after dev_alloc_name() (for the
5147          * /proc files), but before register_netdevice(), because we
5148          * need to set function pointers.
5149          */
5150
5151         res = bond_init(bond_dev, params);
5152         if (res < 0) {
5153                 goto out_netdev;
5154         }
5155
5156         res = register_netdevice(bond_dev);
5157         if (res < 0) {
5158                 goto out_bond;
5159         }
5160
5161         bond_set_lockdep_class(bond_dev);
5162
5163         netif_carrier_off(bond_dev);
5164
5165         up_write(&bonding_rwsem);
5166         rtnl_unlock(); /* allows sysfs registration of net device */
5167         res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5168         if (res < 0) {
5169                 rtnl_lock();
5170                 down_write(&bonding_rwsem);
5171                 bond_deinit(bond_dev);
5172                 unregister_netdevice(bond_dev);
5173                 goto out_rtnl;
5174         }
5175
5176         return 0;
5177
5178 out_bond:
5179         bond_deinit(bond_dev);
5180 out_netdev:
5181         free_netdev(bond_dev);
5182 out_rtnl:
5183         up_write(&bonding_rwsem);
5184         rtnl_unlock();
5185         return res;
5186 }
5187
5188 static int __init bonding_init(void)
5189 {
5190         int i;
5191         int res;
5192         struct bonding *bond;
5193
5194         printk(KERN_INFO "%s", version);
5195
5196         res = bond_check_params(&bonding_defaults);
5197         if (res) {
5198                 goto out;
5199         }
5200
5201 #ifdef CONFIG_PROC_FS
5202         bond_create_proc_dir();
5203 #endif
5204
5205         init_rwsem(&bonding_rwsem);
5206
5207         for (i = 0; i < max_bonds; i++) {
5208                 res = bond_create(NULL, &bonding_defaults);
5209                 if (res)
5210                         goto err;
5211         }
5212
5213         res = bond_create_sysfs();
5214         if (res)
5215                 goto err;
5216
5217         register_netdevice_notifier(&bond_netdev_notifier);
5218         register_inetaddr_notifier(&bond_inetaddr_notifier);
5219         bond_register_ipv6_notifier();
5220
5221         goto out;
5222 err:
5223         list_for_each_entry(bond, &bond_dev_list, bond_list) {
5224                 bond_work_cancel_all(bond);
5225                 destroy_workqueue(bond->wq);
5226         }
5227
5228         bond_destroy_sysfs();
5229
5230         rtnl_lock();
5231         bond_free_all();
5232         rtnl_unlock();
5233 out:
5234         return res;
5235
5236 }
5237
5238 static void __exit bonding_exit(void)
5239 {
5240         unregister_netdevice_notifier(&bond_netdev_notifier);
5241         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5242         bond_unregister_ipv6_notifier();
5243
5244         bond_destroy_sysfs();
5245
5246         rtnl_lock();
5247         bond_free_all();
5248         rtnl_unlock();
5249 }
5250
5251 module_init(bonding_init);
5252 module_exit(bonding_exit);
5253 MODULE_LICENSE("GPL");
5254 MODULE_VERSION(DRV_VERSION);
5255 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5256 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5257 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5258
5259 /*
5260  * Local variables:
5261  *  c-indent-level: 8
5262  *  c-basic-offset: 8
5263  *  tab-width: 8
5264  * End:
5265  */
5266