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