bonding: Fix bonding drivers improper modification of netpoll structure
[linux-2.6.git] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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 <linux/io.h>
57 #include <asm/system.h>
58 #include <asm/dma.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/netpoll.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
73 #include <net/arp.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <linux/jiffies.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
82 #include "bonding.h"
83 #include "bond_3ad.h"
84 #include "bond_alb.h"
85
86 /*---------------------------- Module parameters ----------------------------*/
87
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV    0
90 #define BOND_LINK_ARP_INTERV    0
91
92 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
93 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
94 static int num_grat_arp = 1;
95 static int num_unsol_na = 1;
96 static int miimon       = BOND_LINK_MON_INTERV;
97 static int updelay;
98 static int downdelay;
99 static int use_carrier  = 1;
100 static char *mode;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static char *ad_select;
105 static char *xmit_hash_policy;
106 static int arp_interval = BOND_LINK_ARP_INTERV;
107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
108 static char *arp_validate;
109 static char *fail_over_mac;
110 static int all_slaves_active = 0;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113
114 module_param(max_bonds, int, 0);
115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
116 module_param(tx_queues, int, 0);
117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
118 module_param(num_grat_arp, int, 0644);
119 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
120 module_param(num_unsol_na, int, 0644);
121 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
122 module_param(miimon, int, 0);
123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124 module_param(updelay, int, 0);
125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126 module_param(downdelay, int, 0);
127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128                             "in milliseconds");
129 module_param(use_carrier, int, 0);
130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131                               "0 for off, 1 for on (default)");
132 module_param(mode, charp, 0);
133 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
134                        "1 for active-backup, 2 for balance-xor, "
135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136                        "6 for balance-alb");
137 module_param(primary, charp, 0);
138 MODULE_PARM_DESC(primary, "Primary network device to use");
139 module_param(primary_reselect, charp, 0);
140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141                                    "once it comes up; "
142                                    "0 for always (default), "
143                                    "1 for only if speed of primary is "
144                                    "better, "
145                                    "2 for only on active slave "
146                                    "failure");
147 module_param(lacp_rate, charp, 0);
148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
149                             "(slow/fast)");
150 module_param(ad_select, charp, 0);
151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
152 module_param(xmit_hash_policy, charp, 0);
153 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
154                                    ", 1 for layer 3+4");
155 module_param(arp_interval, int, 0);
156 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
157 module_param_array(arp_ip_target, charp, NULL, 0);
158 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
159 module_param(arp_validate, charp, 0);
160 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
161 module_param(fail_over_mac, charp, 0);
162 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
163 module_param(all_slaves_active, int, 0);
164 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
165                                      "by setting active flag for all slaves.  "
166                                      "0 for never (default), 1 for always.");
167 module_param(resend_igmp, int, 0);
168 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
169
170 /*----------------------------- Global variables ----------------------------*/
171
172 static const char * const version =
173         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
174
175 int bond_net_id __read_mostly;
176
177 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
178 static int arp_ip_count;
179 static int bond_mode    = BOND_MODE_ROUNDROBIN;
180 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
181 static int lacp_fast;
182 #ifdef CONFIG_NET_POLL_CONTROLLER
183 static int disable_netpoll = 1;
184 #endif
185
186 const struct bond_parm_tbl bond_lacp_tbl[] = {
187 {       "slow",         AD_LACP_SLOW},
188 {       "fast",         AD_LACP_FAST},
189 {       NULL,           -1},
190 };
191
192 const struct bond_parm_tbl bond_mode_tbl[] = {
193 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
194 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
195 {       "balance-xor",          BOND_MODE_XOR},
196 {       "broadcast",            BOND_MODE_BROADCAST},
197 {       "802.3ad",              BOND_MODE_8023AD},
198 {       "balance-tlb",          BOND_MODE_TLB},
199 {       "balance-alb",          BOND_MODE_ALB},
200 {       NULL,                   -1},
201 };
202
203 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
204 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
205 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
206 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
207 {       NULL,                   -1},
208 };
209
210 const struct bond_parm_tbl arp_validate_tbl[] = {
211 {       "none",                 BOND_ARP_VALIDATE_NONE},
212 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
213 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
214 {       "all",                  BOND_ARP_VALIDATE_ALL},
215 {       NULL,                   -1},
216 };
217
218 const struct bond_parm_tbl fail_over_mac_tbl[] = {
219 {       "none",                 BOND_FOM_NONE},
220 {       "active",               BOND_FOM_ACTIVE},
221 {       "follow",               BOND_FOM_FOLLOW},
222 {       NULL,                   -1},
223 };
224
225 const struct bond_parm_tbl pri_reselect_tbl[] = {
226 {       "always",               BOND_PRI_RESELECT_ALWAYS},
227 {       "better",               BOND_PRI_RESELECT_BETTER},
228 {       "failure",              BOND_PRI_RESELECT_FAILURE},
229 {       NULL,                   -1},
230 };
231
232 struct bond_parm_tbl ad_select_tbl[] = {
233 {       "stable",       BOND_AD_STABLE},
234 {       "bandwidth",    BOND_AD_BANDWIDTH},
235 {       "count",        BOND_AD_COUNT},
236 {       NULL,           -1},
237 };
238
239 /*-------------------------- Forward declarations ---------------------------*/
240
241 static void bond_send_gratuitous_arp(struct bonding *bond);
242 static int bond_init(struct net_device *bond_dev);
243 static void bond_uninit(struct net_device *bond_dev);
244
245 /*---------------------------- General routines -----------------------------*/
246
247 static const char *bond_mode_name(int mode)
248 {
249         static const char *names[] = {
250                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
251                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
252                 [BOND_MODE_XOR] = "load balancing (xor)",
253                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
254                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
255                 [BOND_MODE_TLB] = "transmit load balancing",
256                 [BOND_MODE_ALB] = "adaptive load balancing",
257         };
258
259         if (mode < 0 || mode > BOND_MODE_ALB)
260                 return "unknown";
261
262         return names[mode];
263 }
264
265 /*---------------------------------- VLAN -----------------------------------*/
266
267 /**
268  * bond_add_vlan - add a new vlan id on bond
269  * @bond: bond that got the notification
270  * @vlan_id: the vlan id to add
271  *
272  * Returns -ENOMEM if allocation failed.
273  */
274 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
275 {
276         struct vlan_entry *vlan;
277
278         pr_debug("bond: %s, vlan id %d\n",
279                  (bond ? bond->dev->name : "None"), vlan_id);
280
281         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
282         if (!vlan)
283                 return -ENOMEM;
284
285         INIT_LIST_HEAD(&vlan->vlan_list);
286         vlan->vlan_id = vlan_id;
287
288         write_lock_bh(&bond->lock);
289
290         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
291
292         write_unlock_bh(&bond->lock);
293
294         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
295
296         return 0;
297 }
298
299 /**
300  * bond_del_vlan - delete a vlan id from bond
301  * @bond: bond that got the notification
302  * @vlan_id: the vlan id to delete
303  *
304  * returns -ENODEV if @vlan_id was not found in @bond.
305  */
306 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
307 {
308         struct vlan_entry *vlan;
309         int res = -ENODEV;
310
311         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
312
313         write_lock_bh(&bond->lock);
314
315         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
316                 if (vlan->vlan_id == vlan_id) {
317                         list_del(&vlan->vlan_list);
318
319                         if (bond_is_lb(bond))
320                                 bond_alb_clear_vlan(bond, vlan_id);
321
322                         pr_debug("removed VLAN ID %d from bond %s\n",
323                                  vlan_id, bond->dev->name);
324
325                         kfree(vlan);
326
327                         if (list_empty(&bond->vlan_list) &&
328                             (bond->slave_cnt == 0)) {
329                                 /* Last VLAN removed and no slaves, so
330                                  * restore block on adding VLANs. This will
331                                  * be removed once new slaves that are not
332                                  * VLAN challenged will be added.
333                                  */
334                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
335                         }
336
337                         res = 0;
338                         goto out;
339                 }
340         }
341
342         pr_debug("couldn't find VLAN ID %d in bond %s\n",
343                  vlan_id, bond->dev->name);
344
345 out:
346         write_unlock_bh(&bond->lock);
347         return res;
348 }
349
350 /**
351  * bond_has_challenged_slaves
352  * @bond: the bond we're working on
353  *
354  * Searches the slave list. Returns 1 if a vlan challenged slave
355  * was found, 0 otherwise.
356  *
357  * Assumes bond->lock is held.
358  */
359 static int bond_has_challenged_slaves(struct bonding *bond)
360 {
361         struct slave *slave;
362         int i;
363
364         bond_for_each_slave(bond, slave, i) {
365                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
366                         pr_debug("found VLAN challenged slave - %s\n",
367                                  slave->dev->name);
368                         return 1;
369                 }
370         }
371
372         pr_debug("no VLAN challenged slaves found\n");
373         return 0;
374 }
375
376 /**
377  * bond_next_vlan - safely skip to the next item in the vlans list.
378  * @bond: the bond we're working on
379  * @curr: item we're advancing from
380  *
381  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
382  * or @curr->next otherwise (even if it is @curr itself again).
383  *
384  * Caller must hold bond->lock
385  */
386 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
387 {
388         struct vlan_entry *next, *last;
389
390         if (list_empty(&bond->vlan_list))
391                 return NULL;
392
393         if (!curr) {
394                 next = list_entry(bond->vlan_list.next,
395                                   struct vlan_entry, vlan_list);
396         } else {
397                 last = list_entry(bond->vlan_list.prev,
398                                   struct vlan_entry, vlan_list);
399                 if (last == curr) {
400                         next = list_entry(bond->vlan_list.next,
401                                           struct vlan_entry, vlan_list);
402                 } else {
403                         next = list_entry(curr->vlan_list.next,
404                                           struct vlan_entry, vlan_list);
405                 }
406         }
407
408         return next;
409 }
410
411 /**
412  * bond_dev_queue_xmit - Prepare skb for xmit.
413  *
414  * @bond: bond device that got this skb for tx.
415  * @skb: hw accel VLAN tagged skb to transmit
416  * @slave_dev: slave that is supposed to xmit this skbuff
417  *
418  * When the bond gets an skb to transmit that is
419  * already hardware accelerated VLAN tagged, and it
420  * needs to relay this skb to a slave that is not
421  * hw accel capable, the skb needs to be "unaccelerated",
422  * i.e. strip the hwaccel tag and re-insert it as part
423  * of the payload.
424  */
425 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
426                         struct net_device *slave_dev)
427 {
428         unsigned short uninitialized_var(vlan_id);
429
430         /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
431         if (!list_empty(&bond->vlan_list) &&
432             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
433             vlan_get_tag(skb, &vlan_id) == 0) {
434                 skb->dev = slave_dev;
435                 skb = vlan_put_tag(skb, vlan_id);
436                 if (!skb) {
437                         /* vlan_put_tag() frees the skb in case of error,
438                          * so return success here so the calling functions
439                          * won't attempt to free is again.
440                          */
441                         return 0;
442                 }
443         } else {
444                 skb->dev = slave_dev;
445         }
446
447         skb->priority = 1;
448 #ifdef CONFIG_NET_POLL_CONTROLLER
449         if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
450                 struct netpoll *np = bond->dev->npinfo->netpoll;
451                 slave_dev->npinfo = bond->dev->npinfo;
452                 slave_dev->priv_flags |= IFF_IN_NETPOLL;
453                 netpoll_send_skb_on_dev(np, skb, slave_dev);
454                 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
455         } else
456 #endif
457                 dev_queue_xmit(skb);
458
459         return 0;
460 }
461
462 /*
463  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
464  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
465  * lock because:
466  * a. This operation is performed in IOCTL context,
467  * b. The operation is protected by the RTNL semaphore in the 8021q code,
468  * c. Holding a lock with BH disabled while directly calling a base driver
469  *    entry point is generally a BAD idea.
470  *
471  * The design of synchronization/protection for this operation in the 8021q
472  * module is good for one or more VLAN devices over a single physical device
473  * and cannot be extended for a teaming solution like bonding, so there is a
474  * potential race condition here where a net device from the vlan group might
475  * be referenced (either by a base driver or the 8021q code) while it is being
476  * removed from the system. However, it turns out we're not making matters
477  * worse, and if it works for regular VLAN usage it will work here too.
478 */
479
480 /**
481  * bond_vlan_rx_register - Propagates registration to slaves
482  * @bond_dev: bonding net device that got called
483  * @grp: vlan group being registered
484  */
485 static void bond_vlan_rx_register(struct net_device *bond_dev,
486                                   struct vlan_group *grp)
487 {
488         struct bonding *bond = netdev_priv(bond_dev);
489         struct slave *slave;
490         int i;
491
492         write_lock(&bond->lock);
493         bond->vlgrp = grp;
494         write_unlock(&bond->lock);
495
496         bond_for_each_slave(bond, slave, i) {
497                 struct net_device *slave_dev = slave->dev;
498                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
499
500                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
501                     slave_ops->ndo_vlan_rx_register) {
502                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
503                 }
504         }
505 }
506
507 /**
508  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
509  * @bond_dev: bonding net device that got called
510  * @vid: vlan id being added
511  */
512 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
513 {
514         struct bonding *bond = netdev_priv(bond_dev);
515         struct slave *slave;
516         int i, res;
517
518         bond_for_each_slave(bond, slave, i) {
519                 struct net_device *slave_dev = slave->dev;
520                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
521
522                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
523                     slave_ops->ndo_vlan_rx_add_vid) {
524                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
525                 }
526         }
527
528         res = bond_add_vlan(bond, vid);
529         if (res) {
530                 pr_err("%s: Error: Failed to add vlan id %d\n",
531                        bond_dev->name, vid);
532         }
533 }
534
535 /**
536  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
537  * @bond_dev: bonding net device that got called
538  * @vid: vlan id being removed
539  */
540 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
541 {
542         struct bonding *bond = netdev_priv(bond_dev);
543         struct slave *slave;
544         struct net_device *vlan_dev;
545         int i, res;
546
547         bond_for_each_slave(bond, slave, i) {
548                 struct net_device *slave_dev = slave->dev;
549                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
550
551                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
552                     slave_ops->ndo_vlan_rx_kill_vid) {
553                         /* Save and then restore vlan_dev in the grp array,
554                          * since the slave's driver might clear it.
555                          */
556                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
557                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
558                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
559                 }
560         }
561
562         res = bond_del_vlan(bond, vid);
563         if (res) {
564                 pr_err("%s: Error: Failed to remove vlan id %d\n",
565                        bond_dev->name, vid);
566         }
567 }
568
569 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
570 {
571         struct vlan_entry *vlan;
572         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
573
574         if (!bond->vlgrp)
575                 return;
576
577         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
578             slave_ops->ndo_vlan_rx_register)
579                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
580
581         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
582             !(slave_ops->ndo_vlan_rx_add_vid))
583                 return;
584
585         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
586                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
587 }
588
589 static void bond_del_vlans_from_slave(struct bonding *bond,
590                                       struct net_device *slave_dev)
591 {
592         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
593         struct vlan_entry *vlan;
594         struct net_device *vlan_dev;
595
596         if (!bond->vlgrp)
597                 return;
598
599         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
600             !(slave_ops->ndo_vlan_rx_kill_vid))
601                 goto unreg;
602
603         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
604                 if (!vlan->vlan_id)
605                         continue;
606                 /* Save and then restore vlan_dev in the grp array,
607                  * since the slave's driver might clear it.
608                  */
609                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
610                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
611                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
612         }
613
614 unreg:
615         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
616             slave_ops->ndo_vlan_rx_register)
617                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
618 }
619
620 /*------------------------------- Link status -------------------------------*/
621
622 /*
623  * Set the carrier state for the master according to the state of its
624  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
625  * do special 802.3ad magic.
626  *
627  * Returns zero if carrier state does not change, nonzero if it does.
628  */
629 static int bond_set_carrier(struct bonding *bond)
630 {
631         struct slave *slave;
632         int i;
633
634         if (bond->slave_cnt == 0)
635                 goto down;
636
637         if (bond->params.mode == BOND_MODE_8023AD)
638                 return bond_3ad_set_carrier(bond);
639
640         bond_for_each_slave(bond, slave, i) {
641                 if (slave->link == BOND_LINK_UP) {
642                         if (!netif_carrier_ok(bond->dev)) {
643                                 netif_carrier_on(bond->dev);
644                                 return 1;
645                         }
646                         return 0;
647                 }
648         }
649
650 down:
651         if (netif_carrier_ok(bond->dev)) {
652                 netif_carrier_off(bond->dev);
653                 return 1;
654         }
655         return 0;
656 }
657
658 /*
659  * Get link speed and duplex from the slave's base driver
660  * using ethtool. If for some reason the call fails or the
661  * values are invalid, fake speed and duplex to 100/Full
662  * and return error.
663  */
664 static int bond_update_speed_duplex(struct slave *slave)
665 {
666         struct net_device *slave_dev = slave->dev;
667         struct ethtool_cmd etool;
668         int res;
669
670         /* Fake speed and duplex */
671         slave->speed = SPEED_100;
672         slave->duplex = DUPLEX_FULL;
673
674         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
675                 return -1;
676
677         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
678         if (res < 0)
679                 return -1;
680
681         switch (etool.speed) {
682         case SPEED_10:
683         case SPEED_100:
684         case SPEED_1000:
685         case SPEED_10000:
686                 break;
687         default:
688                 return -1;
689         }
690
691         switch (etool.duplex) {
692         case DUPLEX_FULL:
693         case DUPLEX_HALF:
694                 break;
695         default:
696                 return -1;
697         }
698
699         slave->speed = etool.speed;
700         slave->duplex = etool.duplex;
701
702         return 0;
703 }
704
705 /*
706  * if <dev> supports MII link status reporting, check its link status.
707  *
708  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
709  * depending upon the setting of the use_carrier parameter.
710  *
711  * Return either BMSR_LSTATUS, meaning that the link is up (or we
712  * can't tell and just pretend it is), or 0, meaning that the link is
713  * down.
714  *
715  * If reporting is non-zero, instead of faking link up, return -1 if
716  * both ETHTOOL and MII ioctls fail (meaning the device does not
717  * support them).  If use_carrier is set, return whatever it says.
718  * It'd be nice if there was a good way to tell if a driver supports
719  * netif_carrier, but there really isn't.
720  */
721 static int bond_check_dev_link(struct bonding *bond,
722                                struct net_device *slave_dev, int reporting)
723 {
724         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
725         int (*ioctl)(struct net_device *, struct ifreq *, int);
726         struct ifreq ifr;
727         struct mii_ioctl_data *mii;
728
729         if (!reporting && !netif_running(slave_dev))
730                 return 0;
731
732         if (bond->params.use_carrier)
733                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
734
735         /* Try to get link status using Ethtool first. */
736         if (slave_dev->ethtool_ops) {
737                 if (slave_dev->ethtool_ops->get_link) {
738                         u32 link;
739
740                         link = slave_dev->ethtool_ops->get_link(slave_dev);
741
742                         return link ? BMSR_LSTATUS : 0;
743                 }
744         }
745
746         /* Ethtool can't be used, fallback to MII ioctls. */
747         ioctl = slave_ops->ndo_do_ioctl;
748         if (ioctl) {
749                 /* TODO: set pointer to correct ioctl on a per team member */
750                 /*       bases to make this more efficient. that is, once  */
751                 /*       we determine the correct ioctl, we will always    */
752                 /*       call it and not the others for that team          */
753                 /*       member.                                           */
754
755                 /*
756                  * We cannot assume that SIOCGMIIPHY will also read a
757                  * register; not all network drivers (e.g., e100)
758                  * support that.
759                  */
760
761                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
762                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
763                 mii = if_mii(&ifr);
764                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
765                         mii->reg_num = MII_BMSR;
766                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
767                                 return mii->val_out & BMSR_LSTATUS;
768                 }
769         }
770
771         /*
772          * If reporting, report that either there's no dev->do_ioctl,
773          * or both SIOCGMIIREG and get_link failed (meaning that we
774          * cannot report link status).  If not reporting, pretend
775          * we're ok.
776          */
777         return reporting ? -1 : BMSR_LSTATUS;
778 }
779
780 /*----------------------------- Multicast list ------------------------------*/
781
782 /*
783  * Push the promiscuity flag down to appropriate slaves
784  */
785 static int bond_set_promiscuity(struct bonding *bond, int inc)
786 {
787         int err = 0;
788         if (USES_PRIMARY(bond->params.mode)) {
789                 /* write lock already acquired */
790                 if (bond->curr_active_slave) {
791                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
792                                                   inc);
793                 }
794         } else {
795                 struct slave *slave;
796                 int i;
797                 bond_for_each_slave(bond, slave, i) {
798                         err = dev_set_promiscuity(slave->dev, inc);
799                         if (err)
800                                 return err;
801                 }
802         }
803         return err;
804 }
805
806 /*
807  * Push the allmulti flag down to all slaves
808  */
809 static int bond_set_allmulti(struct bonding *bond, int inc)
810 {
811         int err = 0;
812         if (USES_PRIMARY(bond->params.mode)) {
813                 /* write lock already acquired */
814                 if (bond->curr_active_slave) {
815                         err = dev_set_allmulti(bond->curr_active_slave->dev,
816                                                inc);
817                 }
818         } else {
819                 struct slave *slave;
820                 int i;
821                 bond_for_each_slave(bond, slave, i) {
822                         err = dev_set_allmulti(slave->dev, inc);
823                         if (err)
824                                 return err;
825                 }
826         }
827         return err;
828 }
829
830 /*
831  * Add a Multicast address to slaves
832  * according to mode
833  */
834 static void bond_mc_add(struct bonding *bond, void *addr)
835 {
836         if (USES_PRIMARY(bond->params.mode)) {
837                 /* write lock already acquired */
838                 if (bond->curr_active_slave)
839                         dev_mc_add(bond->curr_active_slave->dev, addr);
840         } else {
841                 struct slave *slave;
842                 int i;
843
844                 bond_for_each_slave(bond, slave, i)
845                         dev_mc_add(slave->dev, addr);
846         }
847 }
848
849 /*
850  * Remove a multicast address from slave
851  * according to mode
852  */
853 static void bond_mc_del(struct bonding *bond, void *addr)
854 {
855         if (USES_PRIMARY(bond->params.mode)) {
856                 /* write lock already acquired */
857                 if (bond->curr_active_slave)
858                         dev_mc_del(bond->curr_active_slave->dev, addr);
859         } else {
860                 struct slave *slave;
861                 int i;
862                 bond_for_each_slave(bond, slave, i) {
863                         dev_mc_del(slave->dev, addr);
864                 }
865         }
866 }
867
868
869 static void __bond_resend_igmp_join_requests(struct net_device *dev)
870 {
871         struct in_device *in_dev;
872         struct ip_mc_list *im;
873
874         rcu_read_lock();
875         in_dev = __in_dev_get_rcu(dev);
876         if (in_dev) {
877                 for (im = in_dev->mc_list; im; im = im->next)
878                         ip_mc_rejoin_group(im);
879         }
880
881         rcu_read_unlock();
882 }
883
884 /*
885  * Retrieve the list of registered multicast addresses for the bonding
886  * device and retransmit an IGMP JOIN request to the current active
887  * slave.
888  */
889 static void bond_resend_igmp_join_requests(struct bonding *bond)
890 {
891         struct net_device *vlan_dev;
892         struct vlan_entry *vlan;
893
894         read_lock(&bond->lock);
895
896         /* rejoin all groups on bond device */
897         __bond_resend_igmp_join_requests(bond->dev);
898
899         /* rejoin all groups on vlan devices */
900         if (bond->vlgrp) {
901                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
902                         vlan_dev = vlan_group_get_device(bond->vlgrp,
903                                                          vlan->vlan_id);
904                         if (vlan_dev)
905                                 __bond_resend_igmp_join_requests(vlan_dev);
906                 }
907         }
908
909         if (--bond->igmp_retrans > 0)
910                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
911
912         read_unlock(&bond->lock);
913 }
914
915 void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
916 {
917         struct bonding *bond = container_of(work, struct bonding,
918                                                         mcast_work.work);
919         bond_resend_igmp_join_requests(bond);
920 }
921
922 /*
923  * flush all members of flush->mc_list from device dev->mc_list
924  */
925 static void bond_mc_list_flush(struct net_device *bond_dev,
926                                struct net_device *slave_dev)
927 {
928         struct bonding *bond = netdev_priv(bond_dev);
929         struct netdev_hw_addr *ha;
930
931         netdev_for_each_mc_addr(ha, bond_dev)
932                 dev_mc_del(slave_dev, ha->addr);
933
934         if (bond->params.mode == BOND_MODE_8023AD) {
935                 /* del lacpdu mc addr from mc list */
936                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
937
938                 dev_mc_del(slave_dev, lacpdu_multicast);
939         }
940 }
941
942 /*--------------------------- Active slave change ---------------------------*/
943
944 /*
945  * Update the mc list and multicast-related flags for the new and
946  * old active slaves (if any) according to the multicast mode, and
947  * promiscuous flags unconditionally.
948  */
949 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
950                          struct slave *old_active)
951 {
952         struct netdev_hw_addr *ha;
953
954         if (!USES_PRIMARY(bond->params.mode))
955                 /* nothing to do -  mc list is already up-to-date on
956                  * all slaves
957                  */
958                 return;
959
960         if (old_active) {
961                 if (bond->dev->flags & IFF_PROMISC)
962                         dev_set_promiscuity(old_active->dev, -1);
963
964                 if (bond->dev->flags & IFF_ALLMULTI)
965                         dev_set_allmulti(old_active->dev, -1);
966
967                 netdev_for_each_mc_addr(ha, bond->dev)
968                         dev_mc_del(old_active->dev, ha->addr);
969         }
970
971         if (new_active) {
972                 /* FIXME: Signal errors upstream. */
973                 if (bond->dev->flags & IFF_PROMISC)
974                         dev_set_promiscuity(new_active->dev, 1);
975
976                 if (bond->dev->flags & IFF_ALLMULTI)
977                         dev_set_allmulti(new_active->dev, 1);
978
979                 netdev_for_each_mc_addr(ha, bond->dev)
980                         dev_mc_add(new_active->dev, ha->addr);
981         }
982 }
983
984 /*
985  * bond_do_fail_over_mac
986  *
987  * Perform special MAC address swapping for fail_over_mac settings
988  *
989  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
990  */
991 static void bond_do_fail_over_mac(struct bonding *bond,
992                                   struct slave *new_active,
993                                   struct slave *old_active)
994         __releases(&bond->curr_slave_lock)
995         __releases(&bond->lock)
996         __acquires(&bond->lock)
997         __acquires(&bond->curr_slave_lock)
998 {
999         u8 tmp_mac[ETH_ALEN];
1000         struct sockaddr saddr;
1001         int rv;
1002
1003         switch (bond->params.fail_over_mac) {
1004         case BOND_FOM_ACTIVE:
1005                 if (new_active)
1006                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1007                                new_active->dev->addr_len);
1008                 break;
1009         case BOND_FOM_FOLLOW:
1010                 /*
1011                  * if new_active && old_active, swap them
1012                  * if just old_active, do nothing (going to no active slave)
1013                  * if just new_active, set new_active to bond's MAC
1014                  */
1015                 if (!new_active)
1016                         return;
1017
1018                 write_unlock_bh(&bond->curr_slave_lock);
1019                 read_unlock(&bond->lock);
1020
1021                 if (old_active) {
1022                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1023                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1024                                ETH_ALEN);
1025                         saddr.sa_family = new_active->dev->type;
1026                 } else {
1027                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1028                         saddr.sa_family = bond->dev->type;
1029                 }
1030
1031                 rv = dev_set_mac_address(new_active->dev, &saddr);
1032                 if (rv) {
1033                         pr_err("%s: Error %d setting MAC of slave %s\n",
1034                                bond->dev->name, -rv, new_active->dev->name);
1035                         goto out;
1036                 }
1037
1038                 if (!old_active)
1039                         goto out;
1040
1041                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1042                 saddr.sa_family = old_active->dev->type;
1043
1044                 rv = dev_set_mac_address(old_active->dev, &saddr);
1045                 if (rv)
1046                         pr_err("%s: Error %d setting MAC of slave %s\n",
1047                                bond->dev->name, -rv, new_active->dev->name);
1048 out:
1049                 read_lock(&bond->lock);
1050                 write_lock_bh(&bond->curr_slave_lock);
1051                 break;
1052         default:
1053                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1054                        bond->dev->name, bond->params.fail_over_mac);
1055                 break;
1056         }
1057
1058 }
1059
1060 static bool bond_should_change_active(struct bonding *bond)
1061 {
1062         struct slave *prim = bond->primary_slave;
1063         struct slave *curr = bond->curr_active_slave;
1064
1065         if (!prim || !curr || curr->link != BOND_LINK_UP)
1066                 return true;
1067         if (bond->force_primary) {
1068                 bond->force_primary = false;
1069                 return true;
1070         }
1071         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1072             (prim->speed < curr->speed ||
1073              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1074                 return false;
1075         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1076                 return false;
1077         return true;
1078 }
1079
1080 /**
1081  * find_best_interface - select the best available slave to be the active one
1082  * @bond: our bonding struct
1083  *
1084  * Warning: Caller must hold curr_slave_lock for writing.
1085  */
1086 static struct slave *bond_find_best_slave(struct bonding *bond)
1087 {
1088         struct slave *new_active, *old_active;
1089         struct slave *bestslave = NULL;
1090         int mintime = bond->params.updelay;
1091         int i;
1092
1093         new_active = bond->curr_active_slave;
1094
1095         if (!new_active) { /* there were no active slaves left */
1096                 if (bond->slave_cnt > 0)   /* found one slave */
1097                         new_active = bond->first_slave;
1098                 else
1099                         return NULL; /* still no slave, return NULL */
1100         }
1101
1102         if ((bond->primary_slave) &&
1103             bond->primary_slave->link == BOND_LINK_UP &&
1104             bond_should_change_active(bond)) {
1105                 new_active = bond->primary_slave;
1106         }
1107
1108         /* remember where to stop iterating over the slaves */
1109         old_active = new_active;
1110
1111         bond_for_each_slave_from(bond, new_active, i, old_active) {
1112                 if (new_active->link == BOND_LINK_UP) {
1113                         return new_active;
1114                 } else if (new_active->link == BOND_LINK_BACK &&
1115                            IS_UP(new_active->dev)) {
1116                         /* link up, but waiting for stabilization */
1117                         if (new_active->delay < mintime) {
1118                                 mintime = new_active->delay;
1119                                 bestslave = new_active;
1120                         }
1121                 }
1122         }
1123
1124         return bestslave;
1125 }
1126
1127 /**
1128  * change_active_interface - change the active slave into the specified one
1129  * @bond: our bonding struct
1130  * @new: the new slave to make the active one
1131  *
1132  * Set the new slave to the bond's settings and unset them on the old
1133  * curr_active_slave.
1134  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1135  *
1136  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1137  * because it is apparently the best available slave we have, even though its
1138  * updelay hasn't timed out yet.
1139  *
1140  * If new_active is not NULL, caller must hold bond->lock for read and
1141  * curr_slave_lock for write_bh.
1142  */
1143 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1144 {
1145         struct slave *old_active = bond->curr_active_slave;
1146
1147         if (old_active == new_active)
1148                 return;
1149
1150         if (new_active) {
1151                 new_active->jiffies = jiffies;
1152
1153                 if (new_active->link == BOND_LINK_BACK) {
1154                         if (USES_PRIMARY(bond->params.mode)) {
1155                                 pr_info("%s: making interface %s the new 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                         if (bond_is_lb(bond))
1167                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1168                 } else {
1169                         if (USES_PRIMARY(bond->params.mode)) {
1170                                 pr_info("%s: making interface %s the new active one.\n",
1171                                         bond->dev->name, new_active->dev->name);
1172                         }
1173                 }
1174         }
1175
1176         if (USES_PRIMARY(bond->params.mode))
1177                 bond_mc_swap(bond, new_active, old_active);
1178
1179         if (bond_is_lb(bond)) {
1180                 bond_alb_handle_active_change(bond, new_active);
1181                 if (old_active)
1182                         bond_set_slave_inactive_flags(old_active);
1183                 if (new_active)
1184                         bond_set_slave_active_flags(new_active);
1185         } else {
1186                 bond->curr_active_slave = new_active;
1187         }
1188
1189         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1190                 if (old_active)
1191                         bond_set_slave_inactive_flags(old_active);
1192
1193                 if (new_active) {
1194                         bond_set_slave_active_flags(new_active);
1195
1196                         if (bond->params.fail_over_mac)
1197                                 bond_do_fail_over_mac(bond, new_active,
1198                                                       old_active);
1199
1200                         bond->send_grat_arp = bond->params.num_grat_arp;
1201                         bond_send_gratuitous_arp(bond);
1202
1203                         bond->send_unsol_na = bond->params.num_unsol_na;
1204                         bond_send_unsolicited_na(bond);
1205
1206                         write_unlock_bh(&bond->curr_slave_lock);
1207                         read_unlock(&bond->lock);
1208
1209                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1210
1211                         read_lock(&bond->lock);
1212                         write_lock_bh(&bond->curr_slave_lock);
1213                 }
1214         }
1215
1216         /* resend IGMP joins since active slave has changed or
1217          * all were sent on curr_active_slave */
1218         if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1219             bond->params.mode == BOND_MODE_ROUNDROBIN) {
1220                 bond->igmp_retrans = bond->params.resend_igmp;
1221                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1222         }
1223 }
1224
1225 /**
1226  * bond_select_active_slave - select a new active slave, if needed
1227  * @bond: our bonding struct
1228  *
1229  * This functions should be called when one of the following occurs:
1230  * - The old curr_active_slave has been released or lost its link.
1231  * - The primary_slave has got its link back.
1232  * - A slave has got its link back and there's no old curr_active_slave.
1233  *
1234  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1235  */
1236 void bond_select_active_slave(struct bonding *bond)
1237 {
1238         struct slave *best_slave;
1239         int rv;
1240
1241         best_slave = bond_find_best_slave(bond);
1242         if (best_slave != bond->curr_active_slave) {
1243                 bond_change_active_slave(bond, best_slave);
1244                 rv = bond_set_carrier(bond);
1245                 if (!rv)
1246                         return;
1247
1248                 if (netif_carrier_ok(bond->dev)) {
1249                         pr_info("%s: first active interface up!\n",
1250                                 bond->dev->name);
1251                 } else {
1252                         pr_info("%s: now running without any active interface !\n",
1253                                 bond->dev->name);
1254                 }
1255         }
1256 }
1257
1258 /*--------------------------- slave list handling ---------------------------*/
1259
1260 /*
1261  * This function attaches the slave to the end of list.
1262  *
1263  * bond->lock held for writing by caller.
1264  */
1265 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1266 {
1267         if (bond->first_slave == NULL) { /* attaching the first slave */
1268                 new_slave->next = new_slave;
1269                 new_slave->prev = new_slave;
1270                 bond->first_slave = new_slave;
1271         } else {
1272                 new_slave->next = bond->first_slave;
1273                 new_slave->prev = bond->first_slave->prev;
1274                 new_slave->next->prev = new_slave;
1275                 new_slave->prev->next = new_slave;
1276         }
1277
1278         bond->slave_cnt++;
1279 }
1280
1281 /*
1282  * This function detaches the slave from the list.
1283  * WARNING: no check is made to verify if the slave effectively
1284  * belongs to <bond>.
1285  * Nothing is freed on return, structures are just unchained.
1286  * If any slave pointer in bond was pointing to <slave>,
1287  * it should be changed by the calling function.
1288  *
1289  * bond->lock held for writing by caller.
1290  */
1291 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1292 {
1293         if (slave->next)
1294                 slave->next->prev = slave->prev;
1295
1296         if (slave->prev)
1297                 slave->prev->next = slave->next;
1298
1299         if (bond->first_slave == slave) { /* slave is the first slave */
1300                 if (bond->slave_cnt > 1) { /* there are more slave */
1301                         bond->first_slave = slave->next;
1302                 } else {
1303                         bond->first_slave = NULL; /* slave was the last one */
1304                 }
1305         }
1306
1307         slave->next = NULL;
1308         slave->prev = NULL;
1309         bond->slave_cnt--;
1310 }
1311
1312 #ifdef CONFIG_NET_POLL_CONTROLLER
1313 /*
1314  * You must hold read lock on bond->lock before calling this.
1315  */
1316 static bool slaves_support_netpoll(struct net_device *bond_dev)
1317 {
1318         struct bonding *bond = netdev_priv(bond_dev);
1319         struct slave *slave;
1320         int i = 0;
1321         bool ret = true;
1322
1323         bond_for_each_slave(bond, slave, i) {
1324                 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1325                     !slave->dev->netdev_ops->ndo_poll_controller)
1326                         ret = false;
1327         }
1328         return i != 0 && ret;
1329 }
1330
1331 static void bond_poll_controller(struct net_device *bond_dev)
1332 {
1333         struct bonding *bond = netdev_priv(bond_dev);
1334         struct slave *slave;
1335         int i;
1336
1337         bond_for_each_slave(bond, slave, i) {
1338                 if (slave->dev && IS_UP(slave->dev))
1339                         netpoll_poll_dev(slave->dev);
1340         }
1341 }
1342
1343 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1344 {
1345         struct bonding *bond = netdev_priv(bond_dev);
1346         struct slave *slave;
1347         const struct net_device_ops *ops;
1348         int i;
1349
1350         read_lock(&bond->lock);
1351         bond_dev->npinfo = NULL;
1352         bond_for_each_slave(bond, slave, i) {
1353                 if (slave->dev) {
1354                         ops = slave->dev->netdev_ops;
1355                         if (ops->ndo_netpoll_cleanup)
1356                                 ops->ndo_netpoll_cleanup(slave->dev);
1357                         else
1358                                 slave->dev->npinfo = NULL;
1359                 }
1360         }
1361         read_unlock(&bond->lock);
1362 }
1363
1364 #else
1365
1366 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1367 {
1368 }
1369
1370 #endif
1371
1372 /*---------------------------------- IOCTL ----------------------------------*/
1373
1374 static int bond_sethwaddr(struct net_device *bond_dev,
1375                           struct net_device *slave_dev)
1376 {
1377         pr_debug("bond_dev=%p\n", bond_dev);
1378         pr_debug("slave_dev=%p\n", slave_dev);
1379         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1380         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1381         return 0;
1382 }
1383
1384 #define BOND_VLAN_FEATURES \
1385         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1386          NETIF_F_HW_VLAN_FILTER)
1387
1388 /*
1389  * Compute the common dev->feature set available to all slaves.  Some
1390  * feature bits are managed elsewhere, so preserve those feature bits
1391  * on the master device.
1392  */
1393 static int bond_compute_features(struct bonding *bond)
1394 {
1395         struct slave *slave;
1396         struct net_device *bond_dev = bond->dev;
1397         unsigned long features = bond_dev->features;
1398         unsigned long vlan_features = 0;
1399         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1400                                                 bond_dev->hard_header_len);
1401         int i;
1402
1403         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1404         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1405
1406         if (!bond->first_slave)
1407                 goto done;
1408
1409         features &= ~NETIF_F_ONE_FOR_ALL;
1410
1411         vlan_features = bond->first_slave->dev->vlan_features;
1412         bond_for_each_slave(bond, slave, i) {
1413                 features = netdev_increment_features(features,
1414                                                      slave->dev->features,
1415                                                      NETIF_F_ONE_FOR_ALL);
1416                 vlan_features = netdev_increment_features(vlan_features,
1417                                                         slave->dev->vlan_features,
1418                                                         NETIF_F_ONE_FOR_ALL);
1419                 if (slave->dev->hard_header_len > max_hard_header_len)
1420                         max_hard_header_len = slave->dev->hard_header_len;
1421         }
1422
1423 done:
1424         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1425         bond_dev->features = netdev_fix_features(features, NULL);
1426         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1427         bond_dev->hard_header_len = max_hard_header_len;
1428
1429         return 0;
1430 }
1431
1432 static void bond_setup_by_slave(struct net_device *bond_dev,
1433                                 struct net_device *slave_dev)
1434 {
1435         struct bonding *bond = netdev_priv(bond_dev);
1436
1437         bond_dev->header_ops        = slave_dev->header_ops;
1438
1439         bond_dev->type              = slave_dev->type;
1440         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1441         bond_dev->addr_len          = slave_dev->addr_len;
1442
1443         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1444                 slave_dev->addr_len);
1445         bond->setup_by_slave = 1;
1446 }
1447
1448 /* enslave device <slave> to bond device <master> */
1449 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1450 {
1451         struct bonding *bond = netdev_priv(bond_dev);
1452         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1453         struct slave *new_slave = NULL;
1454         struct netdev_hw_addr *ha;
1455         struct sockaddr addr;
1456         int link_reporting;
1457         int old_features = bond_dev->features;
1458         int res = 0;
1459
1460         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1461                 slave_ops->ndo_do_ioctl == NULL) {
1462                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1463                            bond_dev->name, slave_dev->name);
1464         }
1465
1466         /* bond must be initialized by bond_open() before enslaving */
1467         if (!(bond_dev->flags & IFF_UP)) {
1468                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1469                            bond_dev->name);
1470         }
1471
1472         /* already enslaved */
1473         if (slave_dev->flags & IFF_SLAVE) {
1474                 pr_debug("Error, Device was already enslaved\n");
1475                 return -EBUSY;
1476         }
1477
1478         /* vlan challenged mutual exclusion */
1479         /* no need to lock since we're protected by rtnl_lock */
1480         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1481                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1482                 if (bond->vlgrp) {
1483                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1484                                bond_dev->name, slave_dev->name, bond_dev->name);
1485                         return -EPERM;
1486                 } else {
1487                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1488                                    bond_dev->name, slave_dev->name,
1489                                    slave_dev->name, bond_dev->name);
1490                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1491                 }
1492         } else {
1493                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1494                 if (bond->slave_cnt == 0) {
1495                         /* First slave, and it is not VLAN challenged,
1496                          * so remove the block of adding VLANs over the bond.
1497                          */
1498                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1499                 }
1500         }
1501
1502         /*
1503          * Old ifenslave binaries are no longer supported.  These can
1504          * be identified with moderate accuracy by the state of the slave:
1505          * the current ifenslave will set the interface down prior to
1506          * enslaving it; the old ifenslave will not.
1507          */
1508         if ((slave_dev->flags & IFF_UP)) {
1509                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1510                        slave_dev->name);
1511                 res = -EPERM;
1512                 goto err_undo_flags;
1513         }
1514
1515         /* set bonding device ether type by slave - bonding netdevices are
1516          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1517          * there is a need to override some of the type dependent attribs/funcs.
1518          *
1519          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1520          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1521          */
1522         if (bond->slave_cnt == 0) {
1523                 if (bond_dev->type != slave_dev->type) {
1524                         pr_debug("%s: change device type from %d to %d\n",
1525                                  bond_dev->name,
1526                                  bond_dev->type, slave_dev->type);
1527
1528                         res = netdev_bonding_change(bond_dev,
1529                                                     NETDEV_PRE_TYPE_CHANGE);
1530                         res = notifier_to_errno(res);
1531                         if (res) {
1532                                 pr_err("%s: refused to change device type\n",
1533                                        bond_dev->name);
1534                                 res = -EBUSY;
1535                                 goto err_undo_flags;
1536                         }
1537
1538                         /* Flush unicast and multicast addresses */
1539                         dev_uc_flush(bond_dev);
1540                         dev_mc_flush(bond_dev);
1541
1542                         if (slave_dev->type != ARPHRD_ETHER)
1543                                 bond_setup_by_slave(bond_dev, slave_dev);
1544                         else
1545                                 ether_setup(bond_dev);
1546
1547                         netdev_bonding_change(bond_dev,
1548                                               NETDEV_POST_TYPE_CHANGE);
1549                 }
1550         } else if (bond_dev->type != slave_dev->type) {
1551                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1552                        slave_dev->name,
1553                        slave_dev->type, bond_dev->type);
1554                 res = -EINVAL;
1555                 goto err_undo_flags;
1556         }
1557
1558         if (slave_ops->ndo_set_mac_address == NULL) {
1559                 if (bond->slave_cnt == 0) {
1560                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1561                                    bond_dev->name);
1562                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1563                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1564                         pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1565                                bond_dev->name);
1566                         res = -EOPNOTSUPP;
1567                         goto err_undo_flags;
1568                 }
1569         }
1570
1571         /* If this is the first slave, then we need to set the master's hardware
1572          * address to be the same as the slave's. */
1573         if (bond->slave_cnt == 0)
1574                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1575                        slave_dev->addr_len);
1576
1577
1578         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1579         if (!new_slave) {
1580                 res = -ENOMEM;
1581                 goto err_undo_flags;
1582         }
1583
1584         /*
1585          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1586          * is set via sysfs or module option if desired.
1587          */
1588         new_slave->queue_id = 0;
1589
1590         /* Save slave's original mtu and then set it to match the bond */
1591         new_slave->original_mtu = slave_dev->mtu;
1592         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1593         if (res) {
1594                 pr_debug("Error %d calling dev_set_mtu\n", res);
1595                 goto err_free;
1596         }
1597
1598         /*
1599          * Save slave's original ("permanent") mac address for modes
1600          * that need it, and for restoring it upon release, and then
1601          * set it to the master's address
1602          */
1603         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1604
1605         if (!bond->params.fail_over_mac) {
1606                 /*
1607                  * Set slave to master's mac address.  The application already
1608                  * set the master's mac address to that of the first slave
1609                  */
1610                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1611                 addr.sa_family = slave_dev->type;
1612                 res = dev_set_mac_address(slave_dev, &addr);
1613                 if (res) {
1614                         pr_debug("Error %d calling set_mac_address\n", res);
1615                         goto err_restore_mtu;
1616                 }
1617         }
1618
1619         res = netdev_set_master(slave_dev, bond_dev);
1620         if (res) {
1621                 pr_debug("Error %d calling netdev_set_master\n", res);
1622                 goto err_restore_mac;
1623         }
1624         /* open the slave since the application closed it */
1625         res = dev_open(slave_dev);
1626         if (res) {
1627                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1628                 goto err_unset_master;
1629         }
1630
1631         new_slave->dev = slave_dev;
1632         slave_dev->priv_flags |= IFF_BONDING;
1633
1634         if (bond_is_lb(bond)) {
1635                 /* bond_alb_init_slave() must be called before all other stages since
1636                  * it might fail and we do not want to have to undo everything
1637                  */
1638                 res = bond_alb_init_slave(bond, new_slave);
1639                 if (res)
1640                         goto err_close;
1641         }
1642
1643         /* If the mode USES_PRIMARY, then the new slave gets the
1644          * master's promisc (and mc) settings only if it becomes the
1645          * curr_active_slave, and that is taken care of later when calling
1646          * bond_change_active()
1647          */
1648         if (!USES_PRIMARY(bond->params.mode)) {
1649                 /* set promiscuity level to new slave */
1650                 if (bond_dev->flags & IFF_PROMISC) {
1651                         res = dev_set_promiscuity(slave_dev, 1);
1652                         if (res)
1653                                 goto err_close;
1654                 }
1655
1656                 /* set allmulti level to new slave */
1657                 if (bond_dev->flags & IFF_ALLMULTI) {
1658                         res = dev_set_allmulti(slave_dev, 1);
1659                         if (res)
1660                                 goto err_close;
1661                 }
1662
1663                 netif_addr_lock_bh(bond_dev);
1664                 /* upload master's mc_list to new slave */
1665                 netdev_for_each_mc_addr(ha, bond_dev)
1666                         dev_mc_add(slave_dev, ha->addr);
1667                 netif_addr_unlock_bh(bond_dev);
1668         }
1669
1670         if (bond->params.mode == BOND_MODE_8023AD) {
1671                 /* add lacpdu mc addr to mc list */
1672                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1673
1674                 dev_mc_add(slave_dev, lacpdu_multicast);
1675         }
1676
1677         bond_add_vlans_on_slave(bond, slave_dev);
1678
1679         write_lock_bh(&bond->lock);
1680
1681         bond_attach_slave(bond, new_slave);
1682
1683         new_slave->delay = 0;
1684         new_slave->link_failure_count = 0;
1685
1686         bond_compute_features(bond);
1687
1688         write_unlock_bh(&bond->lock);
1689
1690         read_lock(&bond->lock);
1691
1692         new_slave->last_arp_rx = jiffies;
1693
1694         if (bond->params.miimon && !bond->params.use_carrier) {
1695                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1696
1697                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1698                         /*
1699                          * miimon is set but a bonded network driver
1700                          * does not support ETHTOOL/MII and
1701                          * arp_interval is not set.  Note: if
1702                          * use_carrier is enabled, we will never go
1703                          * here (because netif_carrier is always
1704                          * supported); thus, we don't need to change
1705                          * the messages for netif_carrier.
1706                          */
1707                         pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1708                                bond_dev->name, slave_dev->name);
1709                 } else if (link_reporting == -1) {
1710                         /* unable get link status using mii/ethtool */
1711                         pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1712                                    bond_dev->name, slave_dev->name);
1713                 }
1714         }
1715
1716         /* check for initial state */
1717         if (!bond->params.miimon ||
1718             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1719                 if (bond->params.updelay) {
1720                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1721                         new_slave->link  = BOND_LINK_BACK;
1722                         new_slave->delay = bond->params.updelay;
1723                 } else {
1724                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1725                         new_slave->link  = BOND_LINK_UP;
1726                 }
1727                 new_slave->jiffies = jiffies;
1728         } else {
1729                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1730                 new_slave->link  = BOND_LINK_DOWN;
1731         }
1732
1733         if (bond_update_speed_duplex(new_slave) &&
1734             (new_slave->link != BOND_LINK_DOWN)) {
1735                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1736                            bond_dev->name, new_slave->dev->name);
1737
1738                 if (bond->params.mode == BOND_MODE_8023AD) {
1739                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1740                                    bond_dev->name);
1741                 }
1742         }
1743
1744         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1745                 /* if there is a primary slave, remember it */
1746                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1747                         bond->primary_slave = new_slave;
1748                         bond->force_primary = true;
1749                 }
1750         }
1751
1752         write_lock_bh(&bond->curr_slave_lock);
1753
1754         switch (bond->params.mode) {
1755         case BOND_MODE_ACTIVEBACKUP:
1756                 bond_set_slave_inactive_flags(new_slave);
1757                 bond_select_active_slave(bond);
1758                 break;
1759         case BOND_MODE_8023AD:
1760                 /* in 802.3ad mode, the internal mechanism
1761                  * will activate the slaves in the selected
1762                  * aggregator
1763                  */
1764                 bond_set_slave_inactive_flags(new_slave);
1765                 /* if this is the first slave */
1766                 if (bond->slave_cnt == 1) {
1767                         SLAVE_AD_INFO(new_slave).id = 1;
1768                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1769                          * can be called only after the mac address of the bond is set
1770                          */
1771                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1772                                             bond->params.lacp_fast);
1773                 } else {
1774                         SLAVE_AD_INFO(new_slave).id =
1775                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1776                 }
1777
1778                 bond_3ad_bind_slave(new_slave);
1779                 break;
1780         case BOND_MODE_TLB:
1781         case BOND_MODE_ALB:
1782                 new_slave->state = BOND_STATE_ACTIVE;
1783                 bond_set_slave_inactive_flags(new_slave);
1784                 bond_select_active_slave(bond);
1785                 break;
1786         default:
1787                 pr_debug("This slave is always active in trunk mode\n");
1788
1789                 /* always active in trunk mode */
1790                 new_slave->state = BOND_STATE_ACTIVE;
1791
1792                 /* In trunking mode there is little meaning to curr_active_slave
1793                  * anyway (it holds no special properties of the bond device),
1794                  * so we can change it without calling change_active_interface()
1795                  */
1796                 if (!bond->curr_active_slave)
1797                         bond->curr_active_slave = new_slave;
1798
1799                 break;
1800         } /* switch(bond_mode) */
1801
1802         write_unlock_bh(&bond->curr_slave_lock);
1803
1804         bond_set_carrier(bond);
1805
1806 #ifdef CONFIG_NET_POLL_CONTROLLER
1807         /*
1808          * Netpoll and bonding is broken, make sure it is not initialized
1809          * until it is fixed.
1810          */
1811         if (disable_netpoll) {
1812                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1813         } else {
1814                 if (slaves_support_netpoll(bond_dev)) {
1815                         bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1816                         if (bond_dev->npinfo)
1817                                 slave_dev->npinfo = bond_dev->npinfo;
1818                 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1819                         bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1820                         pr_info("New slave device %s does not support netpoll\n",
1821                                 slave_dev->name);
1822                         pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1823                 }
1824         }
1825 #endif
1826         read_unlock(&bond->lock);
1827
1828         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1829         if (res)
1830                 goto err_close;
1831
1832         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1833                 bond_dev->name, slave_dev->name,
1834                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1835                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1836
1837         /* enslave is successful */
1838         return 0;
1839
1840 /* Undo stages on error */
1841 err_close:
1842         dev_close(slave_dev);
1843
1844 err_unset_master:
1845         netdev_set_master(slave_dev, NULL);
1846
1847 err_restore_mac:
1848         if (!bond->params.fail_over_mac) {
1849                 /* XXX TODO - fom follow mode needs to change master's
1850                  * MAC if this slave's MAC is in use by the bond, or at
1851                  * least print a warning.
1852                  */
1853                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1854                 addr.sa_family = slave_dev->type;
1855                 dev_set_mac_address(slave_dev, &addr);
1856         }
1857
1858 err_restore_mtu:
1859         dev_set_mtu(slave_dev, new_slave->original_mtu);
1860
1861 err_free:
1862         kfree(new_slave);
1863
1864 err_undo_flags:
1865         bond_dev->features = old_features;
1866
1867         return res;
1868 }
1869
1870 /*
1871  * Try to release the slave device <slave> from the bond device <master>
1872  * It is legal to access curr_active_slave without a lock because all the function
1873  * is write-locked.
1874  *
1875  * The rules for slave state should be:
1876  *   for Active/Backup:
1877  *     Active stays on all backups go down
1878  *   for Bonded connections:
1879  *     The first up interface should be left on and all others downed.
1880  */
1881 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1882 {
1883         struct bonding *bond = netdev_priv(bond_dev);
1884         struct slave *slave, *oldcurrent;
1885         struct sockaddr addr;
1886
1887         /* slave is not a slave or master is not master of this slave */
1888         if (!(slave_dev->flags & IFF_SLAVE) ||
1889             (slave_dev->master != bond_dev)) {
1890                 pr_err("%s: Error: cannot release %s.\n",
1891                        bond_dev->name, slave_dev->name);
1892                 return -EINVAL;
1893         }
1894
1895         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1896         write_lock_bh(&bond->lock);
1897
1898         slave = bond_get_slave_by_dev(bond, slave_dev);
1899         if (!slave) {
1900                 /* not a slave of this bond */
1901                 pr_info("%s: %s not enslaved\n",
1902                         bond_dev->name, slave_dev->name);
1903                 write_unlock_bh(&bond->lock);
1904                 return -EINVAL;
1905         }
1906
1907         if (!bond->params.fail_over_mac) {
1908                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1909                     bond->slave_cnt > 1)
1910                         pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1911                                    bond_dev->name, slave_dev->name,
1912                                    slave->perm_hwaddr,
1913                                    bond_dev->name, slave_dev->name);
1914         }
1915
1916         /* Inform AD package of unbinding of slave. */
1917         if (bond->params.mode == BOND_MODE_8023AD) {
1918                 /* must be called before the slave is
1919                  * detached from the list
1920                  */
1921                 bond_3ad_unbind_slave(slave);
1922         }
1923
1924         pr_info("%s: releasing %s interface %s\n",
1925                 bond_dev->name,
1926                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1927                 slave_dev->name);
1928
1929         oldcurrent = bond->curr_active_slave;
1930
1931         bond->current_arp_slave = NULL;
1932
1933         /* release the slave from its bond */
1934         bond_detach_slave(bond, slave);
1935
1936         bond_compute_features(bond);
1937
1938         if (bond->primary_slave == slave)
1939                 bond->primary_slave = NULL;
1940
1941         if (oldcurrent == slave)
1942                 bond_change_active_slave(bond, NULL);
1943
1944         if (bond_is_lb(bond)) {
1945                 /* Must be called only after the slave has been
1946                  * detached from the list and the curr_active_slave
1947                  * has been cleared (if our_slave == old_current),
1948                  * but before a new active slave is selected.
1949                  */
1950                 write_unlock_bh(&bond->lock);
1951                 bond_alb_deinit_slave(bond, slave);
1952                 write_lock_bh(&bond->lock);
1953         }
1954
1955         if (oldcurrent == slave) {
1956                 /*
1957                  * Note that we hold RTNL over this sequence, so there
1958                  * is no concern that another slave add/remove event
1959                  * will interfere.
1960                  */
1961                 write_unlock_bh(&bond->lock);
1962                 read_lock(&bond->lock);
1963                 write_lock_bh(&bond->curr_slave_lock);
1964
1965                 bond_select_active_slave(bond);
1966
1967                 write_unlock_bh(&bond->curr_slave_lock);
1968                 read_unlock(&bond->lock);
1969                 write_lock_bh(&bond->lock);
1970         }
1971
1972         if (bond->slave_cnt == 0) {
1973                 bond_set_carrier(bond);
1974
1975                 /* if the last slave was removed, zero the mac address
1976                  * of the master so it will be set by the application
1977                  * to the mac address of the first slave
1978                  */
1979                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1980
1981                 if (!bond->vlgrp) {
1982                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1983                 } else {
1984                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1985                                    bond_dev->name, bond_dev->name);
1986                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1987                                    bond_dev->name);
1988                 }
1989         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1990                    !bond_has_challenged_slaves(bond)) {
1991                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1992                         bond_dev->name, slave_dev->name, bond_dev->name);
1993                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1994         }
1995
1996         write_unlock_bh(&bond->lock);
1997
1998         /* must do this from outside any spinlocks */
1999         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2000
2001         bond_del_vlans_from_slave(bond, slave_dev);
2002
2003         /* If the mode USES_PRIMARY, then we should only remove its
2004          * promisc and mc settings if it was the curr_active_slave, but that was
2005          * already taken care of above when we detached the slave
2006          */
2007         if (!USES_PRIMARY(bond->params.mode)) {
2008                 /* unset promiscuity level from slave */
2009                 if (bond_dev->flags & IFF_PROMISC)
2010                         dev_set_promiscuity(slave_dev, -1);
2011
2012                 /* unset allmulti level from slave */
2013                 if (bond_dev->flags & IFF_ALLMULTI)
2014                         dev_set_allmulti(slave_dev, -1);
2015
2016                 /* flush master's mc_list from slave */
2017                 netif_addr_lock_bh(bond_dev);
2018                 bond_mc_list_flush(bond_dev, slave_dev);
2019                 netif_addr_unlock_bh(bond_dev);
2020         }
2021
2022         netdev_set_master(slave_dev, NULL);
2023
2024 #ifdef CONFIG_NET_POLL_CONTROLLER
2025         read_lock_bh(&bond->lock);
2026
2027          /* Make sure netpoll over stays disabled until fixed. */
2028         if (!disable_netpoll)
2029                 if (slaves_support_netpoll(bond_dev))
2030                                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2031         read_unlock_bh(&bond->lock);
2032         if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2033                 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2034         else
2035                 slave_dev->npinfo = NULL;
2036 #endif
2037
2038         /* close slave before restoring its mac address */
2039         dev_close(slave_dev);
2040
2041         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2042                 /* restore original ("permanent") mac address */
2043                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2044                 addr.sa_family = slave_dev->type;
2045                 dev_set_mac_address(slave_dev, &addr);
2046         }
2047
2048         dev_set_mtu(slave_dev, slave->original_mtu);
2049
2050         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2051                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
2052                                    IFF_SLAVE_NEEDARP);
2053
2054         kfree(slave);
2055
2056         return 0;  /* deletion OK */
2057 }
2058
2059 /*
2060 * First release a slave and than destroy the bond if no more slaves are left.
2061 * Must be under rtnl_lock when this function is called.
2062 */
2063 int  bond_release_and_destroy(struct net_device *bond_dev,
2064                               struct net_device *slave_dev)
2065 {
2066         struct bonding *bond = netdev_priv(bond_dev);
2067         int ret;
2068
2069         ret = bond_release(bond_dev, slave_dev);
2070         if ((ret == 0) && (bond->slave_cnt == 0)) {
2071                 pr_info("%s: destroying bond %s.\n",
2072                         bond_dev->name, bond_dev->name);
2073                 unregister_netdevice(bond_dev);
2074         }
2075         return ret;
2076 }
2077
2078 /*
2079  * This function releases all slaves.
2080  */
2081 static int bond_release_all(struct net_device *bond_dev)
2082 {
2083         struct bonding *bond = netdev_priv(bond_dev);
2084         struct slave *slave;
2085         struct net_device *slave_dev;
2086         struct sockaddr addr;
2087
2088         write_lock_bh(&bond->lock);
2089
2090         netif_carrier_off(bond_dev);
2091
2092         if (bond->slave_cnt == 0)
2093                 goto out;
2094
2095         bond->current_arp_slave = NULL;
2096         bond->primary_slave = NULL;
2097         bond_change_active_slave(bond, NULL);
2098
2099         while ((slave = bond->first_slave) != NULL) {
2100                 /* Inform AD package of unbinding of slave
2101                  * before slave is detached from the list.
2102                  */
2103                 if (bond->params.mode == BOND_MODE_8023AD)
2104                         bond_3ad_unbind_slave(slave);
2105
2106                 slave_dev = slave->dev;
2107                 bond_detach_slave(bond, slave);
2108
2109                 /* now that the slave is detached, unlock and perform
2110                  * all the undo steps that should not be called from
2111                  * within a lock.
2112                  */
2113                 write_unlock_bh(&bond->lock);
2114
2115                 if (bond_is_lb(bond)) {
2116                         /* must be called only after the slave
2117                          * has been detached from the list
2118                          */
2119                         bond_alb_deinit_slave(bond, slave);
2120                 }
2121
2122                 bond_compute_features(bond);
2123
2124                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2125                 bond_del_vlans_from_slave(bond, slave_dev);
2126
2127                 /* If the mode USES_PRIMARY, then we should only remove its
2128                  * promisc and mc settings if it was the curr_active_slave, but that was
2129                  * already taken care of above when we detached the slave
2130                  */
2131                 if (!USES_PRIMARY(bond->params.mode)) {
2132                         /* unset promiscuity level from slave */
2133                         if (bond_dev->flags & IFF_PROMISC)
2134                                 dev_set_promiscuity(slave_dev, -1);
2135
2136                         /* unset allmulti level from slave */
2137                         if (bond_dev->flags & IFF_ALLMULTI)
2138                                 dev_set_allmulti(slave_dev, -1);
2139
2140                         /* flush master's mc_list from slave */
2141                         netif_addr_lock_bh(bond_dev);
2142                         bond_mc_list_flush(bond_dev, slave_dev);
2143                         netif_addr_unlock_bh(bond_dev);
2144                 }
2145
2146                 netdev_set_master(slave_dev, NULL);
2147
2148                 /* close slave before restoring its mac address */
2149                 dev_close(slave_dev);
2150
2151                 if (!bond->params.fail_over_mac) {
2152                         /* restore original ("permanent") mac address*/
2153                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2154                         addr.sa_family = slave_dev->type;
2155                         dev_set_mac_address(slave_dev, &addr);
2156                 }
2157
2158                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2159                                            IFF_SLAVE_INACTIVE);
2160
2161                 kfree(slave);
2162
2163                 /* re-acquire the lock before getting the next slave */
2164                 write_lock_bh(&bond->lock);
2165         }
2166
2167         /* zero the mac address of the master so it will be
2168          * set by the application to the mac address of the
2169          * first slave
2170          */
2171         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2172
2173         if (!bond->vlgrp) {
2174                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2175         } else {
2176                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2177                            bond_dev->name, bond_dev->name);
2178                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2179                            bond_dev->name);
2180         }
2181
2182         pr_info("%s: released all slaves\n", bond_dev->name);
2183
2184 out:
2185         write_unlock_bh(&bond->lock);
2186
2187         return 0;
2188 }
2189
2190 /*
2191  * This function changes the active slave to slave <slave_dev>.
2192  * It returns -EINVAL in the following cases.
2193  *  - <slave_dev> is not found in the list.
2194  *  - There is not active slave now.
2195  *  - <slave_dev> is already active.
2196  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2197  *  - <slave_dev> is not running.
2198  * In these cases, this function does nothing.
2199  * In the other cases, current_slave pointer is changed and 0 is returned.
2200  */
2201 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2202 {
2203         struct bonding *bond = netdev_priv(bond_dev);
2204         struct slave *old_active = NULL;
2205         struct slave *new_active = NULL;
2206         int res = 0;
2207
2208         if (!USES_PRIMARY(bond->params.mode))
2209                 return -EINVAL;
2210
2211         /* Verify that master_dev is indeed the master of slave_dev */
2212         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2213                 return -EINVAL;
2214
2215         read_lock(&bond->lock);
2216
2217         read_lock(&bond->curr_slave_lock);
2218         old_active = bond->curr_active_slave;
2219         read_unlock(&bond->curr_slave_lock);
2220
2221         new_active = bond_get_slave_by_dev(bond, slave_dev);
2222
2223         /*
2224          * Changing to the current active: do nothing; return success.
2225          */
2226         if (new_active && (new_active == old_active)) {
2227                 read_unlock(&bond->lock);
2228                 return 0;
2229         }
2230
2231         if ((new_active) &&
2232             (old_active) &&
2233             (new_active->link == BOND_LINK_UP) &&
2234             IS_UP(new_active->dev)) {
2235                 write_lock_bh(&bond->curr_slave_lock);
2236                 bond_change_active_slave(bond, new_active);
2237                 write_unlock_bh(&bond->curr_slave_lock);
2238         } else
2239                 res = -EINVAL;
2240
2241         read_unlock(&bond->lock);
2242
2243         return res;
2244 }
2245
2246 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2247 {
2248         struct bonding *bond = netdev_priv(bond_dev);
2249
2250         info->bond_mode = bond->params.mode;
2251         info->miimon = bond->params.miimon;
2252
2253         read_lock(&bond->lock);
2254         info->num_slaves = bond->slave_cnt;
2255         read_unlock(&bond->lock);
2256
2257         return 0;
2258 }
2259
2260 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2261 {
2262         struct bonding *bond = netdev_priv(bond_dev);
2263         struct slave *slave;
2264         int i, res = -ENODEV;
2265
2266         read_lock(&bond->lock);
2267
2268         bond_for_each_slave(bond, slave, i) {
2269                 if (i == (int)info->slave_id) {
2270                         res = 0;
2271                         strcpy(info->slave_name, slave->dev->name);
2272                         info->link = slave->link;
2273                         info->state = slave->state;
2274                         info->link_failure_count = slave->link_failure_count;
2275                         break;
2276                 }
2277         }
2278
2279         read_unlock(&bond->lock);
2280
2281         return res;
2282 }
2283
2284 /*-------------------------------- Monitoring -------------------------------*/
2285
2286
2287 static int bond_miimon_inspect(struct bonding *bond)
2288 {
2289         struct slave *slave;
2290         int i, link_state, commit = 0;
2291         bool ignore_updelay;
2292
2293         ignore_updelay = !bond->curr_active_slave ? true : false;
2294
2295         bond_for_each_slave(bond, slave, i) {
2296                 slave->new_link = BOND_LINK_NOCHANGE;
2297
2298                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2299
2300                 switch (slave->link) {
2301                 case BOND_LINK_UP:
2302                         if (link_state)
2303                                 continue;
2304
2305                         slave->link = BOND_LINK_FAIL;
2306                         slave->delay = bond->params.downdelay;
2307                         if (slave->delay) {
2308                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2309                                         bond->dev->name,
2310                                         (bond->params.mode ==
2311                                          BOND_MODE_ACTIVEBACKUP) ?
2312                                         ((slave->state == BOND_STATE_ACTIVE) ?
2313                                          "active " : "backup ") : "",
2314                                         slave->dev->name,
2315                                         bond->params.downdelay * bond->params.miimon);
2316                         }
2317                         /*FALLTHRU*/
2318                 case BOND_LINK_FAIL:
2319                         if (link_state) {
2320                                 /*
2321                                  * recovered before downdelay expired
2322                                  */
2323                                 slave->link = BOND_LINK_UP;
2324                                 slave->jiffies = jiffies;
2325                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2326                                         bond->dev->name,
2327                                         (bond->params.downdelay - slave->delay) *
2328                                         bond->params.miimon,
2329                                         slave->dev->name);
2330                                 continue;
2331                         }
2332
2333                         if (slave->delay <= 0) {
2334                                 slave->new_link = BOND_LINK_DOWN;
2335                                 commit++;
2336                                 continue;
2337                         }
2338
2339                         slave->delay--;
2340                         break;
2341
2342                 case BOND_LINK_DOWN:
2343                         if (!link_state)
2344                                 continue;
2345
2346                         slave->link = BOND_LINK_BACK;
2347                         slave->delay = bond->params.updelay;
2348
2349                         if (slave->delay) {
2350                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2351                                         bond->dev->name, slave->dev->name,
2352                                         ignore_updelay ? 0 :
2353                                         bond->params.updelay *
2354                                         bond->params.miimon);
2355                         }
2356                         /*FALLTHRU*/
2357                 case BOND_LINK_BACK:
2358                         if (!link_state) {
2359                                 slave->link = BOND_LINK_DOWN;
2360                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2361                                         bond->dev->name,
2362                                         (bond->params.updelay - slave->delay) *
2363                                         bond->params.miimon,
2364                                         slave->dev->name);
2365
2366                                 continue;
2367                         }
2368
2369                         if (ignore_updelay)
2370                                 slave->delay = 0;
2371
2372                         if (slave->delay <= 0) {
2373                                 slave->new_link = BOND_LINK_UP;
2374                                 commit++;
2375                                 ignore_updelay = false;
2376                                 continue;
2377                         }
2378
2379                         slave->delay--;
2380                         break;
2381                 }
2382         }
2383
2384         return commit;
2385 }
2386
2387 static void bond_miimon_commit(struct bonding *bond)
2388 {
2389         struct slave *slave;
2390         int i;
2391
2392         bond_for_each_slave(bond, slave, i) {
2393                 switch (slave->new_link) {
2394                 case BOND_LINK_NOCHANGE:
2395                         continue;
2396
2397                 case BOND_LINK_UP:
2398                         slave->link = BOND_LINK_UP;
2399                         slave->jiffies = jiffies;
2400
2401                         if (bond->params.mode == BOND_MODE_8023AD) {
2402                                 /* prevent it from being the active one */
2403                                 slave->state = BOND_STATE_BACKUP;
2404                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2405                                 /* make it immediately active */
2406                                 slave->state = BOND_STATE_ACTIVE;
2407                         } else if (slave != bond->primary_slave) {
2408                                 /* prevent it from being the active one */
2409                                 slave->state = BOND_STATE_BACKUP;
2410                         }
2411
2412                         bond_update_speed_duplex(slave);
2413
2414                         pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2415                                 bond->dev->name, slave->dev->name,
2416                                 slave->speed, slave->duplex ? "full" : "half");
2417
2418                         /* notify ad that the link status has changed */
2419                         if (bond->params.mode == BOND_MODE_8023AD)
2420                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2421
2422                         if (bond_is_lb(bond))
2423                                 bond_alb_handle_link_change(bond, slave,
2424                                                             BOND_LINK_UP);
2425
2426                         if (!bond->curr_active_slave ||
2427                             (slave == bond->primary_slave))
2428                                 goto do_failover;
2429
2430                         continue;
2431
2432                 case BOND_LINK_DOWN:
2433                         if (slave->link_failure_count < UINT_MAX)
2434                                 slave->link_failure_count++;
2435
2436                         slave->link = BOND_LINK_DOWN;
2437
2438                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2439                             bond->params.mode == BOND_MODE_8023AD)
2440                                 bond_set_slave_inactive_flags(slave);
2441
2442                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2443                                 bond->dev->name, slave->dev->name);
2444
2445                         if (bond->params.mode == BOND_MODE_8023AD)
2446                                 bond_3ad_handle_link_change(slave,
2447                                                             BOND_LINK_DOWN);
2448
2449                         if (bond_is_lb(bond))
2450                                 bond_alb_handle_link_change(bond, slave,
2451                                                             BOND_LINK_DOWN);
2452
2453                         if (slave == bond->curr_active_slave)
2454                                 goto do_failover;
2455
2456                         continue;
2457
2458                 default:
2459                         pr_err("%s: invalid new link %d on slave %s\n",
2460                                bond->dev->name, slave->new_link,
2461                                slave->dev->name);
2462                         slave->new_link = BOND_LINK_NOCHANGE;
2463
2464                         continue;
2465                 }
2466
2467 do_failover:
2468                 ASSERT_RTNL();
2469                 write_lock_bh(&bond->curr_slave_lock);
2470                 bond_select_active_slave(bond);
2471                 write_unlock_bh(&bond->curr_slave_lock);
2472         }
2473
2474         bond_set_carrier(bond);
2475 }
2476
2477 /*
2478  * bond_mii_monitor
2479  *
2480  * Really a wrapper that splits the mii monitor into two phases: an
2481  * inspection, then (if inspection indicates something needs to be done)
2482  * an acquisition of appropriate locks followed by a commit phase to
2483  * implement whatever link state changes are indicated.
2484  */
2485 void bond_mii_monitor(struct work_struct *work)
2486 {
2487         struct bonding *bond = container_of(work, struct bonding,
2488                                             mii_work.work);
2489
2490         read_lock(&bond->lock);
2491         if (bond->kill_timers)
2492                 goto out;
2493
2494         if (bond->slave_cnt == 0)
2495                 goto re_arm;
2496
2497         if (bond->send_grat_arp) {
2498                 read_lock(&bond->curr_slave_lock);
2499                 bond_send_gratuitous_arp(bond);
2500                 read_unlock(&bond->curr_slave_lock);
2501         }
2502
2503         if (bond->send_unsol_na) {
2504                 read_lock(&bond->curr_slave_lock);
2505                 bond_send_unsolicited_na(bond);
2506                 read_unlock(&bond->curr_slave_lock);
2507         }
2508
2509         if (bond_miimon_inspect(bond)) {
2510                 read_unlock(&bond->lock);
2511                 rtnl_lock();
2512                 read_lock(&bond->lock);
2513
2514                 bond_miimon_commit(bond);
2515
2516                 read_unlock(&bond->lock);
2517                 rtnl_unlock();  /* might sleep, hold no other locks */
2518                 read_lock(&bond->lock);
2519         }
2520
2521 re_arm:
2522         if (bond->params.miimon)
2523                 queue_delayed_work(bond->wq, &bond->mii_work,
2524                                    msecs_to_jiffies(bond->params.miimon));
2525 out:
2526         read_unlock(&bond->lock);
2527 }
2528
2529 static __be32 bond_glean_dev_ip(struct net_device *dev)
2530 {
2531         struct in_device *idev;
2532         struct in_ifaddr *ifa;
2533         __be32 addr = 0;
2534
2535         if (!dev)
2536                 return 0;
2537
2538         rcu_read_lock();
2539         idev = __in_dev_get_rcu(dev);
2540         if (!idev)
2541                 goto out;
2542
2543         ifa = idev->ifa_list;
2544         if (!ifa)
2545                 goto out;
2546
2547         addr = ifa->ifa_local;
2548 out:
2549         rcu_read_unlock();
2550         return addr;
2551 }
2552
2553 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2554 {
2555         struct vlan_entry *vlan;
2556
2557         if (ip == bond->master_ip)
2558                 return 1;
2559
2560         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2561                 if (ip == vlan->vlan_ip)
2562                         return 1;
2563         }
2564
2565         return 0;
2566 }
2567
2568 /*
2569  * We go to the (large) trouble of VLAN tagging ARP frames because
2570  * switches in VLAN mode (especially if ports are configured as
2571  * "native" to a VLAN) might not pass non-tagged frames.
2572  */
2573 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2574 {
2575         struct sk_buff *skb;
2576
2577         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2578                  slave_dev->name, dest_ip, src_ip, vlan_id);
2579
2580         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2581                          NULL, slave_dev->dev_addr, NULL);
2582
2583         if (!skb) {
2584                 pr_err("ARP packet allocation failed\n");
2585                 return;
2586         }
2587         if (vlan_id) {
2588                 skb = vlan_put_tag(skb, vlan_id);
2589                 if (!skb) {
2590                         pr_err("failed to insert VLAN tag\n");
2591                         return;
2592                 }
2593         }
2594         arp_xmit(skb);
2595 }
2596
2597
2598 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2599 {
2600         int i, vlan_id, rv;
2601         __be32 *targets = bond->params.arp_targets;
2602         struct vlan_entry *vlan;
2603         struct net_device *vlan_dev;
2604         struct flowi fl;
2605         struct rtable *rt;
2606
2607         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2608                 if (!targets[i])
2609                         break;
2610                 pr_debug("basa: target %x\n", targets[i]);
2611                 if (!bond->vlgrp) {
2612                         pr_debug("basa: empty vlan: arp_send\n");
2613                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2614                                       bond->master_ip, 0);
2615                         continue;
2616                 }
2617
2618                 /*
2619                  * If VLANs are configured, we do a route lookup to
2620                  * determine which VLAN interface would be used, so we
2621                  * can tag the ARP with the proper VLAN tag.
2622                  */
2623                 memset(&fl, 0, sizeof(fl));
2624                 fl.fl4_dst = targets[i];
2625                 fl.fl4_tos = RTO_ONLINK;
2626
2627                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2628                 if (rv) {
2629                         if (net_ratelimit()) {
2630                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2631                                            bond->dev->name, &fl.fl4_dst);
2632                         }
2633                         continue;
2634                 }
2635
2636                 /*
2637                  * This target is not on a VLAN
2638                  */
2639                 if (rt->dst.dev == bond->dev) {
2640                         ip_rt_put(rt);
2641                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2642                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2643                                       bond->master_ip, 0);
2644                         continue;
2645                 }
2646
2647                 vlan_id = 0;
2648                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2649                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2650                         if (vlan_dev == rt->dst.dev) {
2651                                 vlan_id = vlan->vlan_id;
2652                                 pr_debug("basa: vlan match on %s %d\n",
2653                                        vlan_dev->name, vlan_id);
2654                                 break;
2655                         }
2656                 }
2657
2658                 if (vlan_id) {
2659                         ip_rt_put(rt);
2660                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2661                                       vlan->vlan_ip, vlan_id);
2662                         continue;
2663                 }
2664
2665                 if (net_ratelimit()) {
2666                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2667                                    bond->dev->name, &fl.fl4_dst,
2668                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2669                 }
2670                 ip_rt_put(rt);
2671         }
2672 }
2673
2674 /*
2675  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2676  * for each VLAN above us.
2677  *
2678  * Caller must hold curr_slave_lock for read or better
2679  */
2680 static void bond_send_gratuitous_arp(struct bonding *bond)
2681 {
2682         struct slave *slave = bond->curr_active_slave;
2683         struct vlan_entry *vlan;
2684         struct net_device *vlan_dev;
2685
2686         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2687                  bond->dev->name, slave ? slave->dev->name : "NULL");
2688
2689         if (!slave || !bond->send_grat_arp ||
2690             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2691                 return;
2692
2693         bond->send_grat_arp--;
2694
2695         if (bond->master_ip) {
2696                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2697                                 bond->master_ip, 0);
2698         }
2699
2700         if (!bond->vlgrp)
2701                 return;
2702
2703         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2704                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2705                 if (vlan->vlan_ip) {
2706                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2707                                       vlan->vlan_ip, vlan->vlan_id);
2708                 }
2709         }
2710 }
2711
2712 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2713 {
2714         int i;
2715         __be32 *targets = bond->params.arp_targets;
2716
2717         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2718                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2719                          &sip, &tip, i, &targets[i],
2720                          bond_has_this_ip(bond, tip));
2721                 if (sip == targets[i]) {
2722                         if (bond_has_this_ip(bond, tip))
2723                                 slave->last_arp_rx = jiffies;
2724                         return;
2725                 }
2726         }
2727 }
2728
2729 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2730 {
2731         struct arphdr *arp;
2732         struct slave *slave;
2733         struct bonding *bond;
2734         unsigned char *arp_ptr;
2735         __be32 sip, tip;
2736
2737         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2738                 /*
2739                  * When using VLANS and bonding, dev and oriv_dev may be
2740                  * incorrect if the physical interface supports VLAN
2741                  * acceleration.  With this change ARP validation now
2742                  * works for hosts only reachable on the VLAN interface.
2743                  */
2744                 dev = vlan_dev_real_dev(dev);
2745                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2746         }
2747
2748         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2749                 goto out;
2750
2751         bond = netdev_priv(dev);
2752         read_lock(&bond->lock);
2753
2754         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2755                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2756                  orig_dev ? orig_dev->name : "NULL");
2757
2758         slave = bond_get_slave_by_dev(bond, orig_dev);
2759         if (!slave || !slave_do_arp_validate(bond, slave))
2760                 goto out_unlock;
2761
2762         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2763                 goto out_unlock;
2764
2765         arp = arp_hdr(skb);
2766         if (arp->ar_hln != dev->addr_len ||
2767             skb->pkt_type == PACKET_OTHERHOST ||
2768             skb->pkt_type == PACKET_LOOPBACK ||
2769             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2770             arp->ar_pro != htons(ETH_P_IP) ||
2771             arp->ar_pln != 4)
2772                 goto out_unlock;
2773
2774         arp_ptr = (unsigned char *)(arp + 1);
2775         arp_ptr += dev->addr_len;
2776         memcpy(&sip, arp_ptr, 4);
2777         arp_ptr += 4 + dev->addr_len;
2778         memcpy(&tip, arp_ptr, 4);
2779
2780         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2781                  bond->dev->name, slave->dev->name, slave->state,
2782                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2783                  &sip, &tip);
2784
2785         /*
2786          * Backup slaves won't see the ARP reply, but do come through
2787          * here for each ARP probe (so we swap the sip/tip to validate
2788          * the probe).  In a "redundant switch, common router" type of
2789          * configuration, the ARP probe will (hopefully) travel from
2790          * the active, through one switch, the router, then the other
2791          * switch before reaching the backup.
2792          */
2793         if (slave->state == BOND_STATE_ACTIVE)
2794                 bond_validate_arp(bond, slave, sip, tip);
2795         else
2796                 bond_validate_arp(bond, slave, tip, sip);
2797
2798 out_unlock:
2799         read_unlock(&bond->lock);
2800 out:
2801         dev_kfree_skb(skb);
2802         return NET_RX_SUCCESS;
2803 }
2804
2805 /*
2806  * this function is called regularly to monitor each slave's link
2807  * ensuring that traffic is being sent and received when arp monitoring
2808  * is used in load-balancing mode. if the adapter has been dormant, then an
2809  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2810  * arp monitoring in active backup mode.
2811  */
2812 void bond_loadbalance_arp_mon(struct work_struct *work)
2813 {
2814         struct bonding *bond = container_of(work, struct bonding,
2815                                             arp_work.work);
2816         struct slave *slave, *oldcurrent;
2817         int do_failover = 0;
2818         int delta_in_ticks;
2819         int i;
2820
2821         read_lock(&bond->lock);
2822
2823         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2824
2825         if (bond->kill_timers)
2826                 goto out;
2827
2828         if (bond->slave_cnt == 0)
2829                 goto re_arm;
2830
2831         read_lock(&bond->curr_slave_lock);
2832         oldcurrent = bond->curr_active_slave;
2833         read_unlock(&bond->curr_slave_lock);
2834
2835         /* see if any of the previous devices are up now (i.e. they have
2836          * xmt and rcv traffic). the curr_active_slave does not come into
2837          * the picture unless it is null. also, slave->jiffies is not needed
2838          * here because we send an arp on each slave and give a slave as
2839          * long as it needs to get the tx/rx within the delta.
2840          * TODO: what about up/down delay in arp mode? it wasn't here before
2841          *       so it can wait
2842          */
2843         bond_for_each_slave(bond, slave, i) {
2844                 unsigned long trans_start = dev_trans_start(slave->dev);
2845
2846                 if (slave->link != BOND_LINK_UP) {
2847                         if (time_in_range(jiffies,
2848                                 trans_start - delta_in_ticks,
2849                                 trans_start + delta_in_ticks) &&
2850                             time_in_range(jiffies,
2851                                 slave->dev->last_rx - delta_in_ticks,
2852                                 slave->dev->last_rx + delta_in_ticks)) {
2853
2854                                 slave->link  = BOND_LINK_UP;
2855                                 slave->state = BOND_STATE_ACTIVE;
2856
2857                                 /* primary_slave has no meaning in round-robin
2858                                  * mode. the window of a slave being up and
2859                                  * curr_active_slave being null after enslaving
2860                                  * is closed.
2861                                  */
2862                                 if (!oldcurrent) {
2863                                         pr_info("%s: link status definitely up for interface %s, ",
2864                                                 bond->dev->name,
2865                                                 slave->dev->name);
2866                                         do_failover = 1;
2867                                 } else {
2868                                         pr_info("%s: interface %s is now up\n",
2869                                                 bond->dev->name,
2870                                                 slave->dev->name);
2871                                 }
2872                         }
2873                 } else {
2874                         /* slave->link == BOND_LINK_UP */
2875
2876                         /* not all switches will respond to an arp request
2877                          * when the source ip is 0, so don't take the link down
2878                          * if we don't know our ip yet
2879                          */
2880                         if (!time_in_range(jiffies,
2881                                 trans_start - delta_in_ticks,
2882                                 trans_start + 2 * delta_in_ticks) ||
2883                             !time_in_range(jiffies,
2884                                 slave->dev->last_rx - delta_in_ticks,
2885                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2886
2887                                 slave->link  = BOND_LINK_DOWN;
2888                                 slave->state = BOND_STATE_BACKUP;
2889
2890                                 if (slave->link_failure_count < UINT_MAX)
2891                                         slave->link_failure_count++;
2892
2893                                 pr_info("%s: interface %s is now down.\n",
2894                                         bond->dev->name,
2895                                         slave->dev->name);
2896
2897                                 if (slave == oldcurrent)
2898                                         do_failover = 1;
2899                         }
2900                 }
2901
2902                 /* note: if switch is in round-robin mode, all links
2903                  * must tx arp to ensure all links rx an arp - otherwise
2904                  * links may oscillate or not come up at all; if switch is
2905                  * in something like xor mode, there is nothing we can
2906                  * do - all replies will be rx'ed on same link causing slaves
2907                  * to be unstable during low/no traffic periods
2908                  */
2909                 if (IS_UP(slave->dev))
2910                         bond_arp_send_all(bond, slave);
2911         }
2912
2913         if (do_failover) {
2914                 write_lock_bh(&bond->curr_slave_lock);
2915
2916                 bond_select_active_slave(bond);
2917
2918                 write_unlock_bh(&bond->curr_slave_lock);
2919         }
2920
2921 re_arm:
2922         if (bond->params.arp_interval)
2923                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2924 out:
2925         read_unlock(&bond->lock);
2926 }
2927
2928 /*
2929  * Called to inspect slaves for active-backup mode ARP monitor link state
2930  * changes.  Sets new_link in slaves to specify what action should take
2931  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2932  * to link states must be committed.
2933  *
2934  * Called with bond->lock held for read.
2935  */
2936 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2937 {
2938         struct slave *slave;
2939         int i, commit = 0;
2940         unsigned long trans_start;
2941
2942         bond_for_each_slave(bond, slave, i) {
2943                 slave->new_link = BOND_LINK_NOCHANGE;
2944
2945                 if (slave->link != BOND_LINK_UP) {
2946                         if (time_in_range(jiffies,
2947                                 slave_last_rx(bond, slave) - delta_in_ticks,
2948                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2949
2950                                 slave->new_link = BOND_LINK_UP;
2951                                 commit++;
2952                         }
2953
2954                         continue;
2955                 }
2956
2957                 /*
2958                  * Give slaves 2*delta after being enslaved or made
2959                  * active.  This avoids bouncing, as the last receive
2960                  * times need a full ARP monitor cycle to be updated.
2961                  */
2962                 if (time_in_range(jiffies,
2963                                   slave->jiffies - delta_in_ticks,
2964                                   slave->jiffies + 2 * delta_in_ticks))
2965                         continue;
2966
2967                 /*
2968                  * Backup slave is down if:
2969                  * - No current_arp_slave AND
2970                  * - more than 3*delta since last receive AND
2971                  * - the bond has an IP address
2972                  *
2973                  * Note: a non-null current_arp_slave indicates
2974                  * the curr_active_slave went down and we are
2975                  * searching for a new one; under this condition
2976                  * we only take the curr_active_slave down - this
2977                  * gives each slave a chance to tx/rx traffic
2978                  * before being taken out
2979                  */
2980                 if (slave->state == BOND_STATE_BACKUP &&
2981                     !bond->current_arp_slave &&
2982                     !time_in_range(jiffies,
2983                         slave_last_rx(bond, slave) - delta_in_ticks,
2984                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2985
2986                         slave->new_link = BOND_LINK_DOWN;
2987                         commit++;
2988                 }
2989
2990                 /*
2991                  * Active slave is down if:
2992                  * - more than 2*delta since transmitting OR
2993                  * - (more than 2*delta since receive AND
2994                  *    the bond has an IP address)
2995                  */
2996                 trans_start = dev_trans_start(slave->dev);
2997                 if ((slave->state == BOND_STATE_ACTIVE) &&
2998                     (!time_in_range(jiffies,
2999                         trans_start - delta_in_ticks,
3000                         trans_start + 2 * delta_in_ticks) ||
3001                      !time_in_range(jiffies,
3002                         slave_last_rx(bond, slave) - delta_in_ticks,
3003                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3004
3005                         slave->new_link = BOND_LINK_DOWN;
3006                         commit++;
3007                 }
3008         }
3009
3010         return commit;
3011 }
3012
3013 /*
3014  * Called to commit link state changes noted by inspection step of
3015  * active-backup mode ARP monitor.
3016  *
3017  * Called with RTNL and bond->lock for read.
3018  */
3019 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3020 {
3021         struct slave *slave;
3022         int i;
3023         unsigned long trans_start;
3024
3025         bond_for_each_slave(bond, slave, i) {
3026                 switch (slave->new_link) {
3027                 case BOND_LINK_NOCHANGE:
3028                         continue;
3029
3030                 case BOND_LINK_UP:
3031                         trans_start = dev_trans_start(slave->dev);
3032                         if ((!bond->curr_active_slave &&
3033                              time_in_range(jiffies,
3034                                            trans_start - delta_in_ticks,
3035                                            trans_start + delta_in_ticks)) ||
3036                             bond->curr_active_slave != slave) {
3037                                 slave->link = BOND_LINK_UP;
3038                                 bond->current_arp_slave = NULL;
3039
3040                                 pr_info("%s: link status definitely up for interface %s.\n",
3041                                         bond->dev->name, slave->dev->name);
3042
3043                                 if (!bond->curr_active_slave ||
3044                                     (slave == bond->primary_slave))
3045                                         goto do_failover;
3046
3047                         }
3048
3049                         continue;
3050
3051                 case BOND_LINK_DOWN:
3052                         if (slave->link_failure_count < UINT_MAX)
3053                                 slave->link_failure_count++;
3054
3055                         slave->link = BOND_LINK_DOWN;
3056                         bond_set_slave_inactive_flags(slave);
3057
3058                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3059                                 bond->dev->name, slave->dev->name);
3060
3061                         if (slave == bond->curr_active_slave) {
3062                                 bond->current_arp_slave = NULL;
3063                                 goto do_failover;
3064                         }
3065
3066                         continue;
3067
3068                 default:
3069                         pr_err("%s: impossible: new_link %d on slave %s\n",
3070                                bond->dev->name, slave->new_link,
3071                                slave->dev->name);
3072                         continue;
3073                 }
3074
3075 do_failover:
3076                 ASSERT_RTNL();
3077                 write_lock_bh(&bond->curr_slave_lock);
3078                 bond_select_active_slave(bond);
3079                 write_unlock_bh(&bond->curr_slave_lock);
3080         }
3081
3082         bond_set_carrier(bond);
3083 }
3084
3085 /*
3086  * Send ARP probes for active-backup mode ARP monitor.
3087  *
3088  * Called with bond->lock held for read.
3089  */
3090 static void bond_ab_arp_probe(struct bonding *bond)
3091 {
3092         struct slave *slave;
3093         int i;
3094
3095         read_lock(&bond->curr_slave_lock);
3096
3097         if (bond->current_arp_slave && bond->curr_active_slave)
3098                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3099                         bond->current_arp_slave->dev->name,
3100                         bond->curr_active_slave->dev->name);
3101
3102         if (bond->curr_active_slave) {
3103                 bond_arp_send_all(bond, bond->curr_active_slave);
3104                 read_unlock(&bond->curr_slave_lock);
3105                 return;
3106         }
3107
3108         read_unlock(&bond->curr_slave_lock);
3109
3110         /* if we don't have a curr_active_slave, search for the next available
3111          * backup slave from the current_arp_slave and make it the candidate
3112          * for becoming the curr_active_slave
3113          */
3114
3115         if (!bond->current_arp_slave) {
3116                 bond->current_arp_slave = bond->first_slave;
3117                 if (!bond->current_arp_slave)
3118                         return;
3119         }
3120
3121         bond_set_slave_inactive_flags(bond->current_arp_slave);
3122
3123         /* search for next candidate */
3124         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3125                 if (IS_UP(slave->dev)) {
3126                         slave->link = BOND_LINK_BACK;
3127                         bond_set_slave_active_flags(slave);
3128                         bond_arp_send_all(bond, slave);
3129                         slave->jiffies = jiffies;
3130                         bond->current_arp_slave = slave;
3131                         break;
3132                 }
3133
3134                 /* if the link state is up at this point, we
3135                  * mark it down - this can happen if we have
3136                  * simultaneous link failures and
3137                  * reselect_active_interface doesn't make this
3138                  * one the current slave so it is still marked
3139                  * up when it is actually down
3140                  */
3141                 if (slave->link == BOND_LINK_UP) {
3142                         slave->link = BOND_LINK_DOWN;
3143                         if (slave->link_failure_count < UINT_MAX)
3144                                 slave->link_failure_count++;
3145
3146                         bond_set_slave_inactive_flags(slave);
3147
3148                         pr_info("%s: backup interface %s is now down.\n",
3149                                 bond->dev->name, slave->dev->name);
3150                 }
3151         }
3152 }
3153
3154 void bond_activebackup_arp_mon(struct work_struct *work)
3155 {
3156         struct bonding *bond = container_of(work, struct bonding,
3157                                             arp_work.work);
3158         int delta_in_ticks;
3159
3160         read_lock(&bond->lock);
3161
3162         if (bond->kill_timers)
3163                 goto out;
3164
3165         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3166
3167         if (bond->slave_cnt == 0)
3168                 goto re_arm;
3169
3170         if (bond->send_grat_arp) {
3171                 read_lock(&bond->curr_slave_lock);
3172                 bond_send_gratuitous_arp(bond);
3173                 read_unlock(&bond->curr_slave_lock);
3174         }
3175
3176         if (bond->send_unsol_na) {
3177                 read_lock(&bond->curr_slave_lock);
3178                 bond_send_unsolicited_na(bond);
3179                 read_unlock(&bond->curr_slave_lock);
3180         }
3181
3182         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3183                 read_unlock(&bond->lock);
3184                 rtnl_lock();
3185                 read_lock(&bond->lock);
3186
3187                 bond_ab_arp_commit(bond, delta_in_ticks);
3188
3189                 read_unlock(&bond->lock);
3190                 rtnl_unlock();
3191                 read_lock(&bond->lock);
3192         }
3193
3194         bond_ab_arp_probe(bond);
3195
3196 re_arm:
3197         if (bond->params.arp_interval)
3198                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3199 out:
3200         read_unlock(&bond->lock);
3201 }
3202
3203 /*------------------------------ proc/seq_file-------------------------------*/
3204
3205 #ifdef CONFIG_PROC_FS
3206
3207 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3208         __acquires(&dev_base_lock)
3209         __acquires(&bond->lock)
3210 {
3211         struct bonding *bond = seq->private;
3212         loff_t off = 0;
3213         struct slave *slave;
3214         int i;
3215
3216         /* make sure the bond won't be taken away */
3217         read_lock(&dev_base_lock);
3218         read_lock(&bond->lock);
3219
3220         if (*pos == 0)
3221                 return SEQ_START_TOKEN;
3222
3223         bond_for_each_slave(bond, slave, i) {
3224                 if (++off == *pos)
3225                         return slave;
3226         }
3227
3228         return NULL;
3229 }
3230
3231 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3232 {
3233         struct bonding *bond = seq->private;
3234         struct slave *slave = v;
3235
3236         ++*pos;
3237         if (v == SEQ_START_TOKEN)
3238                 return bond->first_slave;
3239
3240         slave = slave->next;
3241
3242         return (slave == bond->first_slave) ? NULL : slave;
3243 }
3244
3245 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3246         __releases(&bond->lock)
3247         __releases(&dev_base_lock)
3248 {
3249         struct bonding *bond = seq->private;
3250
3251         read_unlock(&bond->lock);
3252         read_unlock(&dev_base_lock);
3253 }
3254
3255 static void bond_info_show_master(struct seq_file *seq)
3256 {
3257         struct bonding *bond = seq->private;
3258         struct slave *curr;
3259         int i;
3260
3261         read_lock(&bond->curr_slave_lock);
3262         curr = bond->curr_active_slave;
3263         read_unlock(&bond->curr_slave_lock);
3264
3265         seq_printf(seq, "Bonding Mode: %s",
3266                    bond_mode_name(bond->params.mode));
3267
3268         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3269             bond->params.fail_over_mac)
3270                 seq_printf(seq, " (fail_over_mac %s)",
3271                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3272
3273         seq_printf(seq, "\n");
3274
3275         if (bond->params.mode == BOND_MODE_XOR ||
3276                 bond->params.mode == BOND_MODE_8023AD) {
3277                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3278                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3279                         bond->params.xmit_policy);
3280         }
3281
3282         if (USES_PRIMARY(bond->params.mode)) {
3283                 seq_printf(seq, "Primary Slave: %s",
3284                            (bond->primary_slave) ?
3285                            bond->primary_slave->dev->name : "None");
3286                 if (bond->primary_slave)
3287                         seq_printf(seq, " (primary_reselect %s)",
3288                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3289
3290                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3291                            (curr) ? curr->dev->name : "None");
3292         }
3293
3294         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3295                    "up" : "down");
3296         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3297         seq_printf(seq, "Up Delay (ms): %d\n",
3298                    bond->params.updelay * bond->params.miimon);
3299         seq_printf(seq, "Down Delay (ms): %d\n",
3300                    bond->params.downdelay * bond->params.miimon);
3301
3302
3303         /* ARP information */
3304         if (bond->params.arp_interval > 0) {
3305                 int printed = 0;
3306                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3307                                 bond->params.arp_interval);
3308
3309                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3310
3311                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3312                         if (!bond->params.arp_targets[i])
3313                                 break;
3314                         if (printed)
3315                                 seq_printf(seq, ",");
3316                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3317                         printed = 1;
3318                 }
3319                 seq_printf(seq, "\n");
3320         }
3321
3322         if (bond->params.mode == BOND_MODE_8023AD) {
3323                 struct ad_info ad_info;
3324
3325                 seq_puts(seq, "\n802.3ad info\n");
3326                 seq_printf(seq, "LACP rate: %s\n",
3327                            (bond->params.lacp_fast) ? "fast" : "slow");
3328                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3329                            ad_select_tbl[bond->params.ad_select].modename);
3330
3331                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3332                         seq_printf(seq, "bond %s has no active aggregator\n",
3333                                    bond->dev->name);
3334                 } else {
3335                         seq_printf(seq, "Active Aggregator Info:\n");
3336
3337                         seq_printf(seq, "\tAggregator ID: %d\n",
3338                                    ad_info.aggregator_id);
3339                         seq_printf(seq, "\tNumber of ports: %d\n",
3340                                    ad_info.ports);
3341                         seq_printf(seq, "\tActor Key: %d\n",
3342                                    ad_info.actor_key);
3343                         seq_printf(seq, "\tPartner Key: %d\n",
3344                                    ad_info.partner_key);
3345                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3346                                    ad_info.partner_system);
3347                 }
3348         }
3349 }
3350
3351 static void bond_info_show_slave(struct seq_file *seq,
3352                                  const struct slave *slave)
3353 {
3354         struct bonding *bond = seq->private;
3355
3356         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3357         seq_printf(seq, "MII Status: %s\n",
3358                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3359         seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3360         seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3361         seq_printf(seq, "Link Failure Count: %u\n",
3362                    slave->link_failure_count);
3363
3364         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3365
3366         if (bond->params.mode == BOND_MODE_8023AD) {
3367                 const struct aggregator *agg
3368                         = SLAVE_AD_INFO(slave).port.aggregator;
3369
3370                 if (agg)
3371                         seq_printf(seq, "Aggregator ID: %d\n",
3372                                    agg->aggregator_identifier);
3373                 else
3374                         seq_puts(seq, "Aggregator ID: N/A\n");
3375         }
3376         seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3377 }
3378
3379 static int bond_info_seq_show(struct seq_file *seq, void *v)
3380 {
3381         if (v == SEQ_START_TOKEN) {
3382                 seq_printf(seq, "%s\n", version);
3383                 bond_info_show_master(seq);
3384         } else
3385                 bond_info_show_slave(seq, v);
3386
3387         return 0;
3388 }
3389
3390 static const struct seq_operations bond_info_seq_ops = {
3391         .start = bond_info_seq_start,
3392         .next  = bond_info_seq_next,
3393         .stop  = bond_info_seq_stop,
3394         .show  = bond_info_seq_show,
3395 };
3396
3397 static int bond_info_open(struct inode *inode, struct file *file)
3398 {
3399         struct seq_file *seq;
3400         struct proc_dir_entry *proc;
3401         int res;
3402
3403         res = seq_open(file, &bond_info_seq_ops);
3404         if (!res) {
3405                 /* recover the pointer buried in proc_dir_entry data */
3406                 seq = file->private_data;
3407                 proc = PDE(inode);
3408                 seq->private = proc->data;
3409         }
3410
3411         return res;
3412 }
3413
3414 static const struct file_operations bond_info_fops = {
3415         .owner   = THIS_MODULE,
3416         .open    = bond_info_open,
3417         .read    = seq_read,
3418         .llseek  = seq_lseek,
3419         .release = seq_release,
3420 };
3421
3422 static void bond_create_proc_entry(struct bonding *bond)
3423 {
3424         struct net_device *bond_dev = bond->dev;
3425         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3426
3427         if (bn->proc_dir) {
3428                 bond->proc_entry = proc_create_data(bond_dev->name,
3429                                                     S_IRUGO, bn->proc_dir,
3430                                                     &bond_info_fops, bond);
3431                 if (bond->proc_entry == NULL)
3432                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3433                                    DRV_NAME, bond_dev->name);
3434                 else
3435                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3436         }
3437 }
3438
3439 static void bond_remove_proc_entry(struct bonding *bond)
3440 {
3441         struct net_device *bond_dev = bond->dev;
3442         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3443
3444         if (bn->proc_dir && bond->proc_entry) {
3445                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3446                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3447                 bond->proc_entry = NULL;
3448         }
3449 }
3450
3451 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3452  * Caller must hold rtnl_lock.
3453  */
3454 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3455 {
3456         if (!bn->proc_dir) {
3457                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3458                 if (!bn->proc_dir)
3459                         pr_warning("Warning: cannot create /proc/net/%s\n",
3460                                    DRV_NAME);
3461         }
3462 }
3463
3464 /* Destroy the bonding directory under /proc/net, if empty.
3465  * Caller must hold rtnl_lock.
3466  */
3467 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3468 {
3469         if (bn->proc_dir) {
3470                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3471                 bn->proc_dir = NULL;
3472         }
3473 }
3474
3475 #else /* !CONFIG_PROC_FS */
3476
3477 static void bond_create_proc_entry(struct bonding *bond)
3478 {
3479 }
3480
3481 static void bond_remove_proc_entry(struct bonding *bond)
3482 {
3483 }
3484
3485 static inline void bond_create_proc_dir(struct bond_net *bn)
3486 {
3487 }
3488
3489 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3490 {
3491 }
3492
3493 #endif /* CONFIG_PROC_FS */
3494
3495
3496 /*-------------------------- netdev event handling --------------------------*/
3497
3498 /*
3499  * Change device name
3500  */
3501 static int bond_event_changename(struct bonding *bond)
3502 {
3503         bond_remove_proc_entry(bond);
3504         bond_create_proc_entry(bond);
3505
3506         return NOTIFY_DONE;
3507 }
3508
3509 static int bond_master_netdev_event(unsigned long event,
3510                                     struct net_device *bond_dev)
3511 {
3512         struct bonding *event_bond = netdev_priv(bond_dev);
3513
3514         switch (event) {
3515         case NETDEV_CHANGENAME:
3516                 return bond_event_changename(event_bond);
3517         default:
3518                 break;
3519         }
3520
3521         return NOTIFY_DONE;
3522 }
3523
3524 static int bond_slave_netdev_event(unsigned long event,
3525                                    struct net_device *slave_dev)
3526 {
3527         struct net_device *bond_dev = slave_dev->master;
3528         struct bonding *bond = netdev_priv(bond_dev);
3529
3530         switch (event) {
3531         case NETDEV_UNREGISTER:
3532                 if (bond_dev) {
3533                         if (bond->setup_by_slave)
3534                                 bond_release_and_destroy(bond_dev, slave_dev);
3535                         else
3536                                 bond_release(bond_dev, slave_dev);
3537                 }
3538                 break;
3539         case NETDEV_CHANGE:
3540                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3541                         struct slave *slave;
3542
3543                         slave = bond_get_slave_by_dev(bond, slave_dev);
3544                         if (slave) {
3545                                 u16 old_speed = slave->speed;
3546                                 u16 old_duplex = slave->duplex;
3547
3548                                 bond_update_speed_duplex(slave);
3549
3550                                 if (bond_is_lb(bond))
3551                                         break;
3552
3553                                 if (old_speed != slave->speed)
3554                                         bond_3ad_adapter_speed_changed(slave);
3555                                 if (old_duplex != slave->duplex)
3556                                         bond_3ad_adapter_duplex_changed(slave);
3557                         }
3558                 }
3559
3560                 break;
3561         case NETDEV_DOWN:
3562                 /*
3563                  * ... Or is it this?
3564                  */
3565                 break;
3566         case NETDEV_CHANGEMTU:
3567                 /*
3568                  * TODO: Should slaves be allowed to
3569                  * independently alter their MTU?  For
3570                  * an active-backup bond, slaves need
3571                  * not be the same type of device, so
3572                  * MTUs may vary.  For other modes,
3573                  * slaves arguably should have the
3574                  * same MTUs. To do this, we'd need to
3575                  * take over the slave's change_mtu
3576                  * function for the duration of their
3577                  * servitude.
3578                  */
3579                 break;
3580         case NETDEV_CHANGENAME:
3581                 /*
3582                  * TODO: handle changing the primary's name
3583                  */
3584                 break;
3585         case NETDEV_FEAT_CHANGE:
3586                 bond_compute_features(bond);
3587                 break;
3588         default:
3589                 break;
3590         }
3591
3592         return NOTIFY_DONE;
3593 }
3594
3595 /*
3596  * bond_netdev_event: handle netdev notifier chain events.
3597  *
3598  * This function receives events for the netdev chain.  The caller (an
3599  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3600  * locks for us to safely manipulate the slave devices (RTNL lock,
3601  * dev_probe_lock).
3602  */
3603 static int bond_netdev_event(struct notifier_block *this,
3604                              unsigned long event, void *ptr)
3605 {
3606         struct net_device *event_dev = (struct net_device *)ptr;
3607
3608         pr_debug("event_dev: %s, event: %lx\n",
3609                  event_dev ? event_dev->name : "None",
3610                  event);
3611
3612         if (!(event_dev->priv_flags & IFF_BONDING))
3613                 return NOTIFY_DONE;
3614
3615         if (event_dev->flags & IFF_MASTER) {
3616                 pr_debug("IFF_MASTER\n");
3617                 return bond_master_netdev_event(event, event_dev);
3618         }
3619
3620         if (event_dev->flags & IFF_SLAVE) {
3621                 pr_debug("IFF_SLAVE\n");
3622                 return bond_slave_netdev_event(event, event_dev);
3623         }
3624
3625         return NOTIFY_DONE;
3626 }
3627
3628 /*
3629  * bond_inetaddr_event: handle inetaddr notifier chain events.
3630  *
3631  * We keep track of device IPs primarily to use as source addresses in
3632  * ARP monitor probes (rather than spewing out broadcasts all the time).
3633  *
3634  * We track one IP for the main device (if it has one), plus one per VLAN.
3635  */
3636 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3637 {
3638         struct in_ifaddr *ifa = ptr;
3639         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3640         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3641         struct bonding *bond;
3642         struct vlan_entry *vlan;
3643
3644         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3645                 if (bond->dev == event_dev) {
3646                         switch (event) {
3647                         case NETDEV_UP:
3648                                 bond->master_ip = ifa->ifa_local;
3649                                 return NOTIFY_OK;
3650                         case NETDEV_DOWN:
3651                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3652                                 return NOTIFY_OK;
3653                         default:
3654                                 return NOTIFY_DONE;
3655                         }
3656                 }
3657
3658                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3659                         if (!bond->vlgrp)
3660                                 continue;
3661                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3662                         if (vlan_dev == event_dev) {
3663                                 switch (event) {
3664                                 case NETDEV_UP:
3665                                         vlan->vlan_ip = ifa->ifa_local;
3666                                         return NOTIFY_OK;
3667                                 case NETDEV_DOWN:
3668                                         vlan->vlan_ip =
3669                                                 bond_glean_dev_ip(vlan_dev);
3670                                         return NOTIFY_OK;
3671                                 default:
3672                                         return NOTIFY_DONE;
3673                                 }
3674                         }
3675                 }
3676         }
3677         return NOTIFY_DONE;
3678 }
3679
3680 static struct notifier_block bond_netdev_notifier = {
3681         .notifier_call = bond_netdev_event,
3682 };
3683
3684 static struct notifier_block bond_inetaddr_notifier = {
3685         .notifier_call = bond_inetaddr_event,
3686 };
3687
3688 /*-------------------------- Packet type handling ---------------------------*/
3689
3690 /* register to receive lacpdus on a bond */
3691 static void bond_register_lacpdu(struct bonding *bond)
3692 {
3693         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3694
3695         /* initialize packet type */
3696         pk_type->type = PKT_TYPE_LACPDU;
3697         pk_type->dev = bond->dev;
3698         pk_type->func = bond_3ad_lacpdu_recv;
3699
3700         dev_add_pack(pk_type);
3701 }
3702
3703 /* unregister to receive lacpdus on a bond */
3704 static void bond_unregister_lacpdu(struct bonding *bond)
3705 {
3706         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3707 }
3708
3709 void bond_register_arp(struct bonding *bond)
3710 {
3711         struct packet_type *pt = &bond->arp_mon_pt;
3712
3713         if (pt->type)
3714                 return;
3715
3716         pt->type = htons(ETH_P_ARP);
3717         pt->dev = bond->dev;
3718         pt->func = bond_arp_rcv;
3719         dev_add_pack(pt);
3720 }
3721
3722 void bond_unregister_arp(struct bonding *bond)
3723 {
3724         struct packet_type *pt = &bond->arp_mon_pt;
3725
3726         dev_remove_pack(pt);
3727         pt->type = 0;
3728 }
3729
3730 /*---------------------------- Hashing Policies -----------------------------*/
3731
3732 /*
3733  * Hash for the output device based upon layer 2 and layer 3 data. If
3734  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3735  */
3736 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3737 {
3738         struct ethhdr *data = (struct ethhdr *)skb->data;
3739         struct iphdr *iph = ip_hdr(skb);
3740
3741         if (skb->protocol == htons(ETH_P_IP)) {
3742                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3743                         (data->h_dest[5] ^ data->h_source[5])) % count;
3744         }
3745
3746         return (data->h_dest[5] ^ data->h_source[5]) % count;
3747 }
3748
3749 /*
3750  * Hash for the output device based upon layer 3 and layer 4 data. If
3751  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3752  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3753  */
3754 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3755 {
3756         struct ethhdr *data = (struct ethhdr *)skb->data;
3757         struct iphdr *iph = ip_hdr(skb);
3758         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3759         int layer4_xor = 0;
3760
3761         if (skb->protocol == htons(ETH_P_IP)) {
3762                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3763                     (iph->protocol == IPPROTO_TCP ||
3764                      iph->protocol == IPPROTO_UDP)) {
3765                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3766                 }
3767                 return (layer4_xor ^
3768                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3769
3770         }
3771
3772         return (data->h_dest[5] ^ data->h_source[5]) % count;
3773 }
3774
3775 /*
3776  * Hash for the output device based upon layer 2 data
3777  */
3778 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3779 {
3780         struct ethhdr *data = (struct ethhdr *)skb->data;
3781
3782         return (data->h_dest[5] ^ data->h_source[5]) % count;
3783 }
3784
3785 /*-------------------------- Device entry points ----------------------------*/
3786
3787 static int bond_open(struct net_device *bond_dev)
3788 {
3789         struct bonding *bond = netdev_priv(bond_dev);
3790
3791         bond->kill_timers = 0;
3792
3793         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3794
3795         if (bond_is_lb(bond)) {
3796                 /* bond_alb_initialize must be called before the timer
3797                  * is started.
3798                  */
3799                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3800                         /* something went wrong - fail the open operation */
3801                         return -ENOMEM;
3802                 }
3803
3804                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3805                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3806         }
3807
3808         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3809                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3810                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3811         }
3812
3813         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3814                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3815                         INIT_DELAYED_WORK(&bond->arp_work,
3816                                           bond_activebackup_arp_mon);
3817                 else
3818                         INIT_DELAYED_WORK(&bond->arp_work,
3819                                           bond_loadbalance_arp_mon);
3820
3821                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3822                 if (bond->params.arp_validate)
3823                         bond_register_arp(bond);
3824         }
3825
3826         if (bond->params.mode == BOND_MODE_8023AD) {
3827                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3828                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3829                 /* register to receive LACPDUs */
3830                 bond_register_lacpdu(bond);
3831                 bond_3ad_initiate_agg_selection(bond, 1);
3832         }
3833
3834         return 0;
3835 }
3836
3837 static int bond_close(struct net_device *bond_dev)
3838 {
3839         struct bonding *bond = netdev_priv(bond_dev);
3840
3841         if (bond->params.mode == BOND_MODE_8023AD) {
3842                 /* Unregister the receive of LACPDUs */
3843                 bond_unregister_lacpdu(bond);
3844         }
3845
3846         if (bond->params.arp_validate)
3847                 bond_unregister_arp(bond);
3848
3849         write_lock_bh(&bond->lock);
3850
3851         bond->send_grat_arp = 0;
3852         bond->send_unsol_na = 0;
3853
3854         /* signal timers not to re-arm */
3855         bond->kill_timers = 1;
3856
3857         write_unlock_bh(&bond->lock);
3858
3859         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3860                 cancel_delayed_work(&bond->mii_work);
3861         }
3862
3863         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3864                 cancel_delayed_work(&bond->arp_work);
3865         }
3866
3867         switch (bond->params.mode) {
3868         case BOND_MODE_8023AD:
3869                 cancel_delayed_work(&bond->ad_work);
3870                 break;
3871         case BOND_MODE_TLB:
3872         case BOND_MODE_ALB:
3873                 cancel_delayed_work(&bond->alb_work);
3874                 break;
3875         default:
3876                 break;
3877         }
3878
3879         if (delayed_work_pending(&bond->mcast_work))
3880                 cancel_delayed_work(&bond->mcast_work);
3881
3882         if (bond_is_lb(bond)) {
3883                 /* Must be called only after all
3884                  * slaves have been released
3885                  */
3886                 bond_alb_deinitialize(bond);
3887         }
3888
3889         return 0;
3890 }
3891
3892 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3893                                                 struct rtnl_link_stats64 *stats)
3894 {
3895         struct bonding *bond = netdev_priv(bond_dev);
3896         struct rtnl_link_stats64 temp;
3897         struct slave *slave;
3898         int i;
3899
3900         memset(stats, 0, sizeof(*stats));
3901
3902         read_lock_bh(&bond->lock);
3903
3904         bond_for_each_slave(bond, slave, i) {
3905                 const struct rtnl_link_stats64 *sstats =
3906                         dev_get_stats(slave->dev, &temp);
3907
3908                 stats->rx_packets += sstats->rx_packets;
3909                 stats->rx_bytes += sstats->rx_bytes;
3910                 stats->rx_errors += sstats->rx_errors;
3911                 stats->rx_dropped += sstats->rx_dropped;
3912
3913                 stats->tx_packets += sstats->tx_packets;
3914                 stats->tx_bytes += sstats->tx_bytes;
3915                 stats->tx_errors += sstats->tx_errors;
3916                 stats->tx_dropped += sstats->tx_dropped;
3917
3918                 stats->multicast += sstats->multicast;
3919                 stats->collisions += sstats->collisions;
3920
3921                 stats->rx_length_errors += sstats->rx_length_errors;
3922                 stats->rx_over_errors += sstats->rx_over_errors;
3923                 stats->rx_crc_errors += sstats->rx_crc_errors;
3924                 stats->rx_frame_errors += sstats->rx_frame_errors;
3925                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3926                 stats->rx_missed_errors += sstats->rx_missed_errors;
3927
3928                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3929                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3930                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3931                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3932                 stats->tx_window_errors += sstats->tx_window_errors;
3933         }
3934
3935         read_unlock_bh(&bond->lock);
3936
3937         return stats;
3938 }
3939
3940 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3941 {
3942         struct net_device *slave_dev = NULL;
3943         struct ifbond k_binfo;
3944         struct ifbond __user *u_binfo = NULL;
3945         struct ifslave k_sinfo;
3946         struct ifslave __user *u_sinfo = NULL;
3947         struct mii_ioctl_data *mii = NULL;
3948         int res = 0;
3949
3950         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3951
3952         switch (cmd) {
3953         case SIOCGMIIPHY:
3954                 mii = if_mii(ifr);
3955                 if (!mii)
3956                         return -EINVAL;
3957
3958                 mii->phy_id = 0;
3959                 /* Fall Through */
3960         case SIOCGMIIREG:
3961                 /*
3962                  * We do this again just in case we were called by SIOCGMIIREG
3963                  * instead of SIOCGMIIPHY.
3964                  */
3965                 mii = if_mii(ifr);
3966                 if (!mii)
3967                         return -EINVAL;
3968
3969
3970                 if (mii->reg_num == 1) {
3971                         struct bonding *bond = netdev_priv(bond_dev);
3972                         mii->val_out = 0;
3973                         read_lock(&bond->lock);
3974                         read_lock(&bond->curr_slave_lock);
3975                         if (netif_carrier_ok(bond->dev))
3976                                 mii->val_out = BMSR_LSTATUS;
3977
3978                         read_unlock(&bond->curr_slave_lock);
3979                         read_unlock(&bond->lock);
3980                 }
3981
3982                 return 0;
3983         case BOND_INFO_QUERY_OLD:
3984         case SIOCBONDINFOQUERY:
3985                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3986
3987                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3988                         return -EFAULT;
3989
3990                 res = bond_info_query(bond_dev, &k_binfo);
3991                 if (res == 0 &&
3992                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3993                         return -EFAULT;
3994
3995                 return res;
3996         case BOND_SLAVE_INFO_QUERY_OLD:
3997         case SIOCBONDSLAVEINFOQUERY:
3998                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3999
4000                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4001                         return -EFAULT;
4002
4003                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4004                 if (res == 0 &&
4005                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4006                         return -EFAULT;
4007
4008                 return res;
4009         default:
4010                 /* Go on */
4011                 break;
4012         }
4013
4014         if (!capable(CAP_NET_ADMIN))
4015                 return -EPERM;
4016
4017         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4018
4019         pr_debug("slave_dev=%p:\n", slave_dev);
4020
4021         if (!slave_dev)
4022                 res = -ENODEV;
4023         else {
4024                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4025                 switch (cmd) {
4026                 case BOND_ENSLAVE_OLD:
4027                 case SIOCBONDENSLAVE:
4028                         res = bond_enslave(bond_dev, slave_dev);
4029                         break;
4030                 case BOND_RELEASE_OLD:
4031                 case SIOCBONDRELEASE:
4032                         res = bond_release(bond_dev, slave_dev);
4033                         break;
4034                 case BOND_SETHWADDR_OLD:
4035                 case SIOCBONDSETHWADDR:
4036                         res = bond_sethwaddr(bond_dev, slave_dev);
4037                         break;
4038                 case BOND_CHANGE_ACTIVE_OLD:
4039                 case SIOCBONDCHANGEACTIVE:
4040                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4041                         break;
4042                 default:
4043                         res = -EOPNOTSUPP;
4044                 }
4045
4046                 dev_put(slave_dev);
4047         }
4048
4049         return res;
4050 }
4051
4052 static bool bond_addr_in_mc_list(unsigned char *addr,
4053                                  struct netdev_hw_addr_list *list,
4054                                  int addrlen)
4055 {
4056         struct netdev_hw_addr *ha;
4057
4058         netdev_hw_addr_list_for_each(ha, list)
4059                 if (!memcmp(ha->addr, addr, addrlen))
4060                         return true;
4061
4062         return false;
4063 }
4064
4065 static void bond_set_multicast_list(struct net_device *bond_dev)
4066 {
4067         struct bonding *bond = netdev_priv(bond_dev);
4068         struct netdev_hw_addr *ha;
4069         bool found;
4070
4071         /*
4072          * Do promisc before checking multicast_mode
4073          */
4074         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4075                 /*
4076                  * FIXME: Need to handle the error when one of the multi-slaves
4077                  * encounters error.
4078                  */
4079                 bond_set_promiscuity(bond, 1);
4080
4081
4082         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4083                 bond_set_promiscuity(bond, -1);
4084
4085
4086         /* set allmulti flag to slaves */
4087         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4088                 /*
4089                  * FIXME: Need to handle the error when one of the multi-slaves
4090                  * encounters error.
4091                  */
4092                 bond_set_allmulti(bond, 1);
4093
4094
4095         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4096                 bond_set_allmulti(bond, -1);
4097
4098
4099         read_lock(&bond->lock);
4100
4101         bond->flags = bond_dev->flags;
4102
4103         /* looking for addresses to add to slaves' mc list */
4104         netdev_for_each_mc_addr(ha, bond_dev) {
4105                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4106                                              bond_dev->addr_len);
4107                 if (!found)
4108                         bond_mc_add(bond, ha->addr);
4109         }
4110
4111         /* looking for addresses to delete from slaves' list */
4112         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4113                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4114                                              bond_dev->addr_len);
4115                 if (!found)
4116                         bond_mc_del(bond, ha->addr);
4117         }
4118
4119         /* save master's multicast list */
4120         __hw_addr_flush(&bond->mc_list);
4121         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4122                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4123
4124         read_unlock(&bond->lock);
4125 }
4126
4127 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4128 {
4129         struct bonding *bond = netdev_priv(dev);
4130         struct slave *slave = bond->first_slave;
4131
4132         if (slave) {
4133                 const struct net_device_ops *slave_ops
4134                         = slave->dev->netdev_ops;
4135                 if (slave_ops->ndo_neigh_setup)
4136                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4137         }
4138         return 0;
4139 }
4140
4141 /*
4142  * Change the MTU of all of a master's slaves to match the master
4143  */
4144 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4145 {
4146         struct bonding *bond = netdev_priv(bond_dev);
4147         struct slave *slave, *stop_at;
4148         int res = 0;
4149         int i;
4150
4151         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4152                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4153
4154         /* Can't hold bond->lock with bh disabled here since
4155          * some base drivers panic. On the other hand we can't
4156          * hold bond->lock without bh disabled because we'll
4157          * deadlock. The only solution is to rely on the fact
4158          * that we're under rtnl_lock here, and the slaves
4159          * list won't change. This doesn't solve the problem
4160          * of setting the slave's MTU while it is
4161          * transmitting, but the assumption is that the base
4162          * driver can handle that.
4163          *
4164          * TODO: figure out a way to safely iterate the slaves
4165          * list, but without holding a lock around the actual
4166          * call to the base driver.
4167          */
4168
4169         bond_for_each_slave(bond, slave, i) {
4170                 pr_debug("s %p s->p %p c_m %p\n",
4171                          slave,
4172                          slave->prev,
4173                          slave->dev->netdev_ops->ndo_change_mtu);
4174
4175                 res = dev_set_mtu(slave->dev, new_mtu);
4176
4177                 if (res) {
4178                         /* If we failed to set the slave's mtu to the new value
4179                          * we must abort the operation even in ACTIVE_BACKUP
4180                          * mode, because if we allow the backup slaves to have
4181                          * different mtu values than the active slave we'll
4182                          * need to change their mtu when doing a failover. That
4183                          * means changing their mtu from timer context, which
4184                          * is probably not a good idea.
4185                          */
4186                         pr_debug("err %d %s\n", res, slave->dev->name);
4187                         goto unwind;
4188                 }
4189         }
4190
4191         bond_dev->mtu = new_mtu;
4192
4193         return 0;
4194
4195 unwind:
4196         /* unwind from head to the slave that failed */
4197         stop_at = slave;
4198         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4199                 int tmp_res;
4200
4201                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4202                 if (tmp_res) {
4203                         pr_debug("unwind err %d dev %s\n",
4204                                  tmp_res, slave->dev->name);
4205                 }
4206         }
4207
4208         return res;
4209 }
4210
4211 /*
4212  * Change HW address
4213  *
4214  * Note that many devices must be down to change the HW address, and
4215  * downing the master releases all slaves.  We can make bonds full of
4216  * bonding devices to test this, however.
4217  */
4218 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4219 {
4220         struct bonding *bond = netdev_priv(bond_dev);
4221         struct sockaddr *sa = addr, tmp_sa;
4222         struct slave *slave, *stop_at;
4223         int res = 0;
4224         int i;
4225
4226         if (bond->params.mode == BOND_MODE_ALB)
4227                 return bond_alb_set_mac_address(bond_dev, addr);
4228
4229
4230         pr_debug("bond=%p, name=%s\n",
4231                  bond, bond_dev ? bond_dev->name : "None");
4232
4233         /*
4234          * If fail_over_mac is set to active, do nothing and return
4235          * success.  Returning an error causes ifenslave to fail.
4236          */
4237         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4238                 return 0;
4239
4240         if (!is_valid_ether_addr(sa->sa_data))
4241                 return -EADDRNOTAVAIL;
4242
4243         /* Can't hold bond->lock with bh disabled here since
4244          * some base drivers panic. On the other hand we can't
4245          * hold bond->lock without bh disabled because we'll
4246          * deadlock. The only solution is to rely on the fact
4247          * that we're under rtnl_lock here, and the slaves
4248          * list won't change. This doesn't solve the problem
4249          * of setting the slave's hw address while it is
4250          * transmitting, but the assumption is that the base
4251          * driver can handle that.
4252          *
4253          * TODO: figure out a way to safely iterate the slaves
4254          * list, but without holding a lock around the actual
4255          * call to the base driver.
4256          */
4257
4258         bond_for_each_slave(bond, slave, i) {
4259                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4260                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4261
4262                 if (slave_ops->ndo_set_mac_address == NULL) {
4263                         res = -EOPNOTSUPP;
4264                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4265                         goto unwind;
4266                 }
4267
4268                 res = dev_set_mac_address(slave->dev, addr);
4269                 if (res) {
4270                         /* TODO: consider downing the slave
4271                          * and retry ?
4272                          * User should expect communications
4273                          * breakage anyway until ARP finish
4274                          * updating, so...
4275                          */
4276                         pr_debug("err %d %s\n", res, slave->dev->name);
4277                         goto unwind;
4278                 }
4279         }
4280
4281         /* success */
4282         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4283         return 0;
4284
4285 unwind:
4286         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4287         tmp_sa.sa_family = bond_dev->type;
4288
4289         /* unwind from head to the slave that failed */
4290         stop_at = slave;
4291         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4292                 int tmp_res;
4293
4294                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4295                 if (tmp_res) {
4296                         pr_debug("unwind err %d dev %s\n",
4297                                  tmp_res, slave->dev->name);
4298                 }
4299         }
4300
4301         return res;
4302 }
4303
4304 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4305 {
4306         struct bonding *bond = netdev_priv(bond_dev);
4307         struct slave *slave, *start_at;
4308         int i, slave_no, res = 1;
4309         struct iphdr *iph = ip_hdr(skb);
4310
4311         read_lock(&bond->lock);
4312
4313         if (!BOND_IS_OK(bond))
4314                 goto out;
4315         /*
4316          * Start with the curr_active_slave that joined the bond as the
4317          * default for sending IGMP traffic.  For failover purposes one
4318          * needs to maintain some consistency for the interface that will
4319          * send the join/membership reports.  The curr_active_slave found
4320          * will send all of this type of traffic.
4321          */
4322         if ((iph->protocol == IPPROTO_IGMP) &&
4323             (skb->protocol == htons(ETH_P_IP))) {
4324
4325                 read_lock(&bond->curr_slave_lock);
4326                 slave = bond->curr_active_slave;
4327                 read_unlock(&bond->curr_slave_lock);
4328
4329                 if (!slave)
4330                         goto out;
4331         } else {
4332                 /*
4333                  * Concurrent TX may collide on rr_tx_counter; we accept
4334                  * that as being rare enough not to justify using an
4335                  * atomic op here.
4336                  */
4337                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4338
4339                 bond_for_each_slave(bond, slave, i) {
4340                         slave_no--;
4341                         if (slave_no < 0)
4342                                 break;
4343                 }
4344         }
4345
4346         start_at = slave;
4347         bond_for_each_slave_from(bond, slave, i, start_at) {
4348                 if (IS_UP(slave->dev) &&
4349                     (slave->link == BOND_LINK_UP) &&
4350                     (slave->state == BOND_STATE_ACTIVE)) {
4351                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4352                         break;
4353                 }
4354         }
4355
4356 out:
4357         if (res) {
4358                 /* no suitable interface, frame not sent */
4359                 dev_kfree_skb(skb);
4360         }
4361         read_unlock(&bond->lock);
4362         return NETDEV_TX_OK;
4363 }
4364
4365
4366 /*
4367  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4368  * the bond has a usable interface.
4369  */
4370 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4371 {
4372         struct bonding *bond = netdev_priv(bond_dev);
4373         int res = 1;
4374
4375         read_lock(&bond->lock);
4376         read_lock(&bond->curr_slave_lock);
4377
4378         if (!BOND_IS_OK(bond))
4379                 goto out;
4380
4381         if (!bond->curr_active_slave)
4382                 goto out;
4383
4384         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4385
4386 out:
4387         if (res)
4388                 /* no suitable interface, frame not sent */
4389                 dev_kfree_skb(skb);
4390
4391         read_unlock(&bond->curr_slave_lock);
4392         read_unlock(&bond->lock);
4393         return NETDEV_TX_OK;
4394 }
4395
4396 /*
4397  * In bond_xmit_xor() , we determine the output device by using a pre-
4398  * determined xmit_hash_policy(), If the selected device is not enabled,
4399  * find the next active slave.
4400  */
4401 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4402 {
4403         struct bonding *bond = netdev_priv(bond_dev);
4404         struct slave *slave, *start_at;
4405         int slave_no;
4406         int i;
4407         int res = 1;
4408
4409         read_lock(&bond->lock);
4410
4411         if (!BOND_IS_OK(bond))
4412                 goto out;
4413
4414         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4415
4416         bond_for_each_slave(bond, slave, i) {
4417                 slave_no--;
4418                 if (slave_no < 0)
4419                         break;
4420         }
4421
4422         start_at = slave;
4423
4424         bond_for_each_slave_from(bond, slave, i, start_at) {
4425                 if (IS_UP(slave->dev) &&
4426                     (slave->link == BOND_LINK_UP) &&
4427                     (slave->state == BOND_STATE_ACTIVE)) {
4428                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4429                         break;
4430                 }
4431         }
4432
4433 out:
4434         if (res) {
4435                 /* no suitable interface, frame not sent */
4436                 dev_kfree_skb(skb);
4437         }
4438         read_unlock(&bond->lock);
4439         return NETDEV_TX_OK;
4440 }
4441
4442 /*
4443  * in broadcast mode, we send everything to all usable interfaces.
4444  */
4445 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4446 {
4447         struct bonding *bond = netdev_priv(bond_dev);
4448         struct slave *slave, *start_at;
4449         struct net_device *tx_dev = NULL;
4450         int i;
4451         int res = 1;
4452
4453         read_lock(&bond->lock);
4454
4455         if (!BOND_IS_OK(bond))
4456                 goto out;
4457
4458         read_lock(&bond->curr_slave_lock);
4459         start_at = bond->curr_active_slave;
4460         read_unlock(&bond->curr_slave_lock);
4461
4462         if (!start_at)
4463                 goto out;
4464
4465         bond_for_each_slave_from(bond, slave, i, start_at) {
4466                 if (IS_UP(slave->dev) &&
4467                     (slave->link == BOND_LINK_UP) &&
4468                     (slave->state == BOND_STATE_ACTIVE)) {
4469                         if (tx_dev) {
4470                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4471                                 if (!skb2) {
4472                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4473                                                bond_dev->name);
4474                                         continue;
4475                                 }
4476
4477                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4478                                 if (res) {
4479                                         dev_kfree_skb(skb2);
4480                                         continue;
4481                                 }
4482                         }
4483                         tx_dev = slave->dev;
4484                 }
4485         }
4486
4487         if (tx_dev)
4488                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4489
4490 out:
4491         if (res)
4492                 /* no suitable interface, frame not sent */
4493                 dev_kfree_skb(skb);
4494
4495         /* frame sent to all suitable interfaces */
4496         read_unlock(&bond->lock);
4497         return NETDEV_TX_OK;
4498 }
4499
4500 /*------------------------- Device initialization ---------------------------*/
4501
4502 static void bond_set_xmit_hash_policy(struct bonding *bond)
4503 {
4504         switch (bond->params.xmit_policy) {
4505         case BOND_XMIT_POLICY_LAYER23:
4506                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4507                 break;
4508         case BOND_XMIT_POLICY_LAYER34:
4509                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4510                 break;
4511         case BOND_XMIT_POLICY_LAYER2:
4512         default:
4513                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4514                 break;
4515         }
4516 }
4517
4518 /*
4519  * Lookup the slave that corresponds to a qid
4520  */
4521 static inline int bond_slave_override(struct bonding *bond,
4522                                       struct sk_buff *skb)
4523 {
4524         int i, res = 1;
4525         struct slave *slave = NULL;
4526         struct slave *check_slave;
4527
4528         read_lock(&bond->lock);
4529
4530         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4531                 goto out;
4532
4533         /* Find out if any slaves have the same mapping as this skb. */
4534         bond_for_each_slave(bond, check_slave, i) {
4535                 if (check_slave->queue_id == skb->queue_mapping) {
4536                         slave = check_slave;
4537                         break;
4538                 }
4539         }
4540
4541         /* If the slave isn't UP, use default transmit policy. */
4542         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4543             (slave->link == BOND_LINK_UP)) {
4544                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4545         }
4546
4547 out:
4548         read_unlock(&bond->lock);
4549         return res;
4550 }
4551
4552 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4553 {
4554         /*
4555          * This helper function exists to help dev_pick_tx get the correct
4556          * destination queue.  Using a helper function skips the a call to
4557          * skb_tx_hash and will put the skbs in the queue we expect on their
4558          * way down to the bonding driver.
4559          */
4560         return skb->queue_mapping;
4561 }
4562
4563 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4564 {
4565         struct bonding *bond = netdev_priv(dev);
4566
4567         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4568                 if (!bond_slave_override(bond, skb))
4569                         return NETDEV_TX_OK;
4570         }
4571
4572         switch (bond->params.mode) {
4573         case BOND_MODE_ROUNDROBIN:
4574                 return bond_xmit_roundrobin(skb, dev);
4575         case BOND_MODE_ACTIVEBACKUP:
4576                 return bond_xmit_activebackup(skb, dev);
4577         case BOND_MODE_XOR:
4578                 return bond_xmit_xor(skb, dev);
4579         case BOND_MODE_BROADCAST:
4580                 return bond_xmit_broadcast(skb, dev);
4581         case BOND_MODE_8023AD:
4582                 return bond_3ad_xmit_xor(skb, dev);
4583         case BOND_MODE_ALB:
4584         case BOND_MODE_TLB:
4585                 return bond_alb_xmit(skb, dev);
4586         default:
4587                 /* Should never happen, mode already checked */
4588                 pr_err("%s: Error: Unknown bonding mode %d\n",
4589                        dev->name, bond->params.mode);
4590                 WARN_ON_ONCE(1);
4591                 dev_kfree_skb(skb);
4592                 return NETDEV_TX_OK;
4593         }
4594 }
4595
4596
4597 /*
4598  * set bond mode specific net device operations
4599  */
4600 void bond_set_mode_ops(struct bonding *bond, int mode)
4601 {
4602         struct net_device *bond_dev = bond->dev;
4603
4604         switch (mode) {
4605         case BOND_MODE_ROUNDROBIN:
4606                 break;
4607         case BOND_MODE_ACTIVEBACKUP:
4608                 break;
4609         case BOND_MODE_XOR:
4610                 bond_set_xmit_hash_policy(bond);
4611                 break;
4612         case BOND_MODE_BROADCAST:
4613                 break;
4614         case BOND_MODE_8023AD:
4615                 bond_set_master_3ad_flags(bond);
4616                 bond_set_xmit_hash_policy(bond);
4617                 break;
4618         case BOND_MODE_ALB:
4619                 bond_set_master_alb_flags(bond);
4620                 /* FALLTHRU */
4621         case BOND_MODE_TLB:
4622                 break;
4623         default:
4624                 /* Should never happen, mode already checked */
4625                 pr_err("%s: Error: Unknown bonding mode %d\n",
4626                        bond_dev->name, mode);
4627                 break;
4628         }
4629 }
4630
4631 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4632                                     struct ethtool_drvinfo *drvinfo)
4633 {
4634         strncpy(drvinfo->driver, DRV_NAME, 32);
4635         strncpy(drvinfo->version, DRV_VERSION, 32);
4636         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4637 }
4638
4639 static const struct ethtool_ops bond_ethtool_ops = {
4640         .get_drvinfo            = bond_ethtool_get_drvinfo,
4641         .get_link               = ethtool_op_get_link,
4642         .get_tx_csum            = ethtool_op_get_tx_csum,
4643         .get_sg                 = ethtool_op_get_sg,
4644         .get_tso                = ethtool_op_get_tso,
4645         .get_ufo                = ethtool_op_get_ufo,
4646         .get_flags              = ethtool_op_get_flags,
4647 };
4648
4649 static const struct net_device_ops bond_netdev_ops = {
4650         .ndo_init               = bond_init,
4651         .ndo_uninit             = bond_uninit,
4652         .ndo_open               = bond_open,
4653         .ndo_stop               = bond_close,
4654         .ndo_start_xmit         = bond_start_xmit,
4655         .ndo_select_queue       = bond_select_queue,
4656         .ndo_get_stats64        = bond_get_stats,
4657         .ndo_do_ioctl           = bond_do_ioctl,
4658         .ndo_set_multicast_list = bond_set_multicast_list,
4659         .ndo_change_mtu         = bond_change_mtu,
4660         .ndo_set_mac_address    = bond_set_mac_address,
4661         .ndo_neigh_setup        = bond_neigh_setup,
4662         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4663         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4664         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4665 #ifdef CONFIG_NET_POLL_CONTROLLER
4666         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4667         .ndo_poll_controller    = bond_poll_controller,
4668 #endif
4669 };
4670
4671 static void bond_destructor(struct net_device *bond_dev)
4672 {
4673         struct bonding *bond = netdev_priv(bond_dev);
4674         if (bond->wq)
4675                 destroy_workqueue(bond->wq);
4676         free_netdev(bond_dev);
4677 }
4678
4679 static void bond_setup(struct net_device *bond_dev)
4680 {
4681         struct bonding *bond = netdev_priv(bond_dev);
4682
4683         /* initialize rwlocks */
4684         rwlock_init(&bond->lock);
4685         rwlock_init(&bond->curr_slave_lock);
4686
4687         bond->params = bonding_defaults;
4688
4689         /* Initialize pointers */
4690         bond->dev = bond_dev;
4691         INIT_LIST_HEAD(&bond->vlan_list);
4692
4693         /* Initialize the device entry points */
4694         ether_setup(bond_dev);
4695         bond_dev->netdev_ops = &bond_netdev_ops;
4696         bond_dev->ethtool_ops = &bond_ethtool_ops;
4697         bond_set_mode_ops(bond, bond->params.mode);
4698
4699         bond_dev->destructor = bond_destructor;
4700
4701         /* Initialize the device options */
4702         bond_dev->tx_queue_len = 0;
4703         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4704         bond_dev->priv_flags |= IFF_BONDING;
4705         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4706
4707         if (bond->params.arp_interval)
4708                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4709
4710         /* At first, we block adding VLANs. That's the only way to
4711          * prevent problems that occur when adding VLANs over an
4712          * empty bond. The block will be removed once non-challenged
4713          * slaves are enslaved.
4714          */
4715         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4716
4717         /* don't acquire bond device's netif_tx_lock when
4718          * transmitting */
4719         bond_dev->features |= NETIF_F_LLTX;
4720
4721         /* By default, we declare the bond to be fully
4722          * VLAN hardware accelerated capable. Special
4723          * care is taken in the various xmit functions
4724          * when there are slaves that are not hw accel
4725          * capable
4726          */
4727         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4728                                NETIF_F_HW_VLAN_RX |
4729                                NETIF_F_HW_VLAN_FILTER);
4730
4731         /* By default, we enable GRO on bonding devices.
4732          * Actual support requires lowlevel drivers are GRO ready.
4733          */
4734         bond_dev->features |= NETIF_F_GRO;
4735 }
4736
4737 static void bond_work_cancel_all(struct bonding *bond)
4738 {
4739         write_lock_bh(&bond->lock);
4740         bond->kill_timers = 1;
4741         write_unlock_bh(&bond->lock);
4742
4743         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4744                 cancel_delayed_work(&bond->mii_work);
4745
4746         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4747                 cancel_delayed_work(&bond->arp_work);
4748
4749         if (bond->params.mode == BOND_MODE_ALB &&
4750             delayed_work_pending(&bond->alb_work))
4751                 cancel_delayed_work(&bond->alb_work);
4752
4753         if (bond->params.mode == BOND_MODE_8023AD &&
4754             delayed_work_pending(&bond->ad_work))
4755                 cancel_delayed_work(&bond->ad_work);
4756
4757         if (delayed_work_pending(&bond->mcast_work))
4758                 cancel_delayed_work(&bond->mcast_work);
4759 }
4760
4761 /*
4762 * Destroy a bonding device.
4763 * Must be under rtnl_lock when this function is called.
4764 */
4765 static void bond_uninit(struct net_device *bond_dev)
4766 {
4767         struct bonding *bond = netdev_priv(bond_dev);
4768         struct vlan_entry *vlan, *tmp;
4769
4770         bond_netpoll_cleanup(bond_dev);
4771
4772         /* Release the bonded slaves */
4773         bond_release_all(bond_dev);
4774
4775         list_del(&bond->bond_list);
4776
4777         bond_work_cancel_all(bond);
4778
4779         bond_remove_proc_entry(bond);
4780
4781         __hw_addr_flush(&bond->mc_list);
4782
4783         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4784                 list_del(&vlan->vlan_list);
4785                 kfree(vlan);
4786         }
4787 }
4788
4789 /*------------------------- Module initialization ---------------------------*/
4790
4791 /*
4792  * Convert string input module parms.  Accept either the
4793  * number of the mode or its string name.  A bit complicated because
4794  * some mode names are substrings of other names, and calls from sysfs
4795  * may have whitespace in the name (trailing newlines, for example).
4796  */
4797 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4798 {
4799         int modeint = -1, i, rv;
4800         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4801
4802         for (p = (char *)buf; *p; p++)
4803                 if (!(isdigit(*p) || isspace(*p)))
4804                         break;
4805
4806         if (*p)
4807                 rv = sscanf(buf, "%20s", modestr);
4808         else
4809                 rv = sscanf(buf, "%d", &modeint);
4810
4811         if (!rv)
4812                 return -1;
4813
4814         for (i = 0; tbl[i].modename; i++) {
4815                 if (modeint == tbl[i].mode)
4816                         return tbl[i].mode;
4817                 if (strcmp(modestr, tbl[i].modename) == 0)
4818                         return tbl[i].mode;
4819         }
4820
4821         return -1;
4822 }
4823
4824 static int bond_check_params(struct bond_params *params)
4825 {
4826         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4827
4828         /*
4829          * Convert string parameters.
4830          */
4831         if (mode) {
4832                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4833                 if (bond_mode == -1) {
4834                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4835                                mode == NULL ? "NULL" : mode);
4836                         return -EINVAL;
4837                 }
4838         }
4839
4840         if (xmit_hash_policy) {
4841                 if ((bond_mode != BOND_MODE_XOR) &&
4842                     (bond_mode != BOND_MODE_8023AD)) {
4843                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4844                                bond_mode_name(bond_mode));
4845                 } else {
4846                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4847                                                         xmit_hashtype_tbl);
4848                         if (xmit_hashtype == -1) {
4849                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4850                                        xmit_hash_policy == NULL ? "NULL" :
4851                                        xmit_hash_policy);
4852                                 return -EINVAL;
4853                         }
4854                 }
4855         }
4856
4857         if (lacp_rate) {
4858                 if (bond_mode != BOND_MODE_8023AD) {
4859                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4860                                 bond_mode_name(bond_mode));
4861                 } else {
4862                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4863                         if (lacp_fast == -1) {
4864                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4865                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4866                                 return -EINVAL;
4867                         }
4868                 }
4869         }
4870
4871         if (ad_select) {
4872                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4873                 if (params->ad_select == -1) {
4874                         pr_err("Error: Invalid ad_select \"%s\"\n",
4875                                ad_select == NULL ? "NULL" : ad_select);
4876                         return -EINVAL;
4877                 }
4878
4879                 if (bond_mode != BOND_MODE_8023AD) {
4880                         pr_warning("ad_select param only affects 802.3ad mode\n");
4881                 }
4882         } else {
4883                 params->ad_select = BOND_AD_STABLE;
4884         }
4885
4886         if (max_bonds < 0) {
4887                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4888                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4889                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4890         }
4891
4892         if (miimon < 0) {
4893                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4894                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4895                 miimon = BOND_LINK_MON_INTERV;
4896         }
4897
4898         if (updelay < 0) {
4899                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4900                            updelay, INT_MAX);
4901                 updelay = 0;
4902         }
4903
4904         if (downdelay < 0) {
4905                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4906                            downdelay, INT_MAX);
4907                 downdelay = 0;
4908         }
4909
4910         if ((use_carrier != 0) && (use_carrier != 1)) {
4911                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4912                            use_carrier);
4913                 use_carrier = 1;
4914         }
4915
4916         if (num_grat_arp < 0 || num_grat_arp > 255) {
4917                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4918                            num_grat_arp);
4919                 num_grat_arp = 1;
4920         }
4921
4922         if (num_unsol_na < 0 || num_unsol_na > 255) {
4923                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4924                            num_unsol_na);
4925                 num_unsol_na = 1;
4926         }
4927
4928         /* reset values for 802.3ad */
4929         if (bond_mode == BOND_MODE_8023AD) {
4930                 if (!miimon) {
4931                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4932                         pr_warning("Forcing miimon to 100msec\n");
4933                         miimon = 100;
4934                 }
4935         }
4936
4937         if (tx_queues < 1 || tx_queues > 255) {
4938                 pr_warning("Warning: tx_queues (%d) should be between "
4939                            "1 and 255, resetting to %d\n",
4940                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4941                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4942         }
4943
4944         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4945                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4946                            "not of valid value (0/1), so it was set to "
4947                            "0\n", all_slaves_active);
4948                 all_slaves_active = 0;
4949         }
4950
4951         if (resend_igmp < 0 || resend_igmp > 255) {
4952                 pr_warning("Warning: resend_igmp (%d) should be between "
4953                            "0 and 255, resetting to %d\n",
4954                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4955                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4956         }
4957
4958         /* reset values for TLB/ALB */
4959         if ((bond_mode == BOND_MODE_TLB) ||
4960             (bond_mode == BOND_MODE_ALB)) {
4961                 if (!miimon) {
4962                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4963                         pr_warning("Forcing miimon to 100msec\n");
4964                         miimon = 100;
4965                 }
4966         }
4967
4968         if (bond_mode == BOND_MODE_ALB) {
4969                 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4970                           updelay);
4971         }
4972
4973         if (!miimon) {
4974                 if (updelay || downdelay) {
4975                         /* just warn the user the up/down delay will have
4976                          * no effect since miimon is zero...
4977                          */
4978                         pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4979                                    updelay, downdelay);
4980                 }
4981         } else {
4982                 /* don't allow arp monitoring */
4983                 if (arp_interval) {
4984                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4985                                    miimon, arp_interval);
4986                         arp_interval = 0;
4987                 }
4988
4989                 if ((updelay % miimon) != 0) {
4990                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4991                                    updelay, miimon,
4992                                    (updelay / miimon) * miimon);
4993                 }
4994
4995                 updelay /= miimon;
4996
4997                 if ((downdelay % miimon) != 0) {
4998                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4999                                    downdelay, miimon,
5000                                    (downdelay / miimon) * miimon);
5001                 }
5002
5003                 downdelay /= miimon;
5004         }
5005
5006         if (arp_interval < 0) {
5007                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5008                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5009                 arp_interval = BOND_LINK_ARP_INTERV;
5010         }
5011
5012         for (arp_ip_count = 0;
5013              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5014              arp_ip_count++) {
5015                 /* not complete check, but should be good enough to
5016                    catch mistakes */
5017                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5018                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5019                                    arp_ip_target[arp_ip_count]);
5020                         arp_interval = 0;
5021                 } else {
5022                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5023                         arp_target[arp_ip_count] = ip;
5024                 }
5025         }
5026
5027         if (arp_interval && !arp_ip_count) {
5028                 /* don't allow arping if no arp_ip_target given... */
5029                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5030                            arp_interval);
5031                 arp_interval = 0;
5032         }
5033
5034         if (arp_validate) {
5035                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5036                         pr_err("arp_validate only supported in active-backup mode\n");
5037                         return -EINVAL;
5038                 }
5039                 if (!arp_interval) {
5040                         pr_err("arp_validate requires arp_interval\n");
5041                         return -EINVAL;
5042                 }
5043
5044                 arp_validate_value = bond_parse_parm(arp_validate,
5045                                                      arp_validate_tbl);
5046                 if (arp_validate_value == -1) {
5047                         pr_err("Error: invalid arp_validate \"%s\"\n",
5048                                arp_validate == NULL ? "NULL" : arp_validate);
5049                         return -EINVAL;
5050                 }
5051         } else
5052                 arp_validate_value = 0;
5053
5054         if (miimon) {
5055                 pr_info("MII link monitoring set to %d ms\n", miimon);
5056         } else if (arp_interval) {
5057                 int i;
5058
5059                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5060                         arp_interval,
5061                         arp_validate_tbl[arp_validate_value].modename,
5062                         arp_ip_count);
5063
5064                 for (i = 0; i < arp_ip_count; i++)
5065                         pr_info(" %s", arp_ip_target[i]);
5066
5067                 pr_info("\n");
5068
5069         } else if (max_bonds) {
5070                 /* miimon and arp_interval not set, we need one so things
5071                  * work as expected, see bonding.txt for details
5072                  */
5073                 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
5074         }
5075
5076         if (primary && !USES_PRIMARY(bond_mode)) {
5077                 /* currently, using a primary only makes sense
5078                  * in active backup, TLB or ALB modes
5079                  */
5080                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5081                            primary, bond_mode_name(bond_mode));
5082                 primary = NULL;
5083         }
5084
5085         if (primary && primary_reselect) {
5086                 primary_reselect_value = bond_parse_parm(primary_reselect,
5087                                                          pri_reselect_tbl);
5088                 if (primary_reselect_value == -1) {
5089                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5090                                primary_reselect ==
5091                                         NULL ? "NULL" : primary_reselect);
5092                         return -EINVAL;
5093                 }
5094         } else {
5095                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5096         }
5097
5098         if (fail_over_mac) {
5099                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5100                                                       fail_over_mac_tbl);
5101                 if (fail_over_mac_value == -1) {
5102                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5103                                arp_validate == NULL ? "NULL" : arp_validate);
5104                         return -EINVAL;
5105                 }
5106
5107                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5108                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5109         } else {
5110                 fail_over_mac_value = BOND_FOM_NONE;
5111         }
5112
5113         /* fill params struct with the proper values */
5114         params->mode = bond_mode;
5115         params->xmit_policy = xmit_hashtype;
5116         params->miimon = miimon;
5117         params->num_grat_arp = num_grat_arp;
5118         params->num_unsol_na = num_unsol_na;
5119         params->arp_interval = arp_interval;
5120         params->arp_validate = arp_validate_value;
5121         params->updelay = updelay;
5122         params->downdelay = downdelay;
5123         params->use_carrier = use_carrier;
5124         params->lacp_fast = lacp_fast;
5125         params->primary[0] = 0;
5126         params->primary_reselect = primary_reselect_value;
5127         params->fail_over_mac = fail_over_mac_value;
5128         params->tx_queues = tx_queues;
5129         params->all_slaves_active = all_slaves_active;
5130         params->resend_igmp = resend_igmp;
5131
5132         if (primary) {
5133                 strncpy(params->primary, primary, IFNAMSIZ);
5134                 params->primary[IFNAMSIZ - 1] = 0;
5135         }
5136
5137         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5138
5139         return 0;
5140 }
5141
5142 static struct lock_class_key bonding_netdev_xmit_lock_key;
5143 static struct lock_class_key bonding_netdev_addr_lock_key;
5144
5145 static void bond_set_lockdep_class_one(struct net_device *dev,
5146                                        struct netdev_queue *txq,
5147                                        void *_unused)
5148 {
5149         lockdep_set_class(&txq->_xmit_lock,
5150                           &bonding_netdev_xmit_lock_key);
5151 }
5152
5153 static void bond_set_lockdep_class(struct net_device *dev)
5154 {
5155         lockdep_set_class(&dev->addr_list_lock,
5156                           &bonding_netdev_addr_lock_key);
5157         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5158 }
5159
5160 /*
5161  * Called from registration process
5162  */
5163 static int bond_init(struct net_device *bond_dev)
5164 {
5165         struct bonding *bond = netdev_priv(bond_dev);
5166         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5167
5168         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5169
5170         bond->wq = create_singlethread_workqueue(bond_dev->name);
5171         if (!bond->wq)
5172                 return -ENOMEM;
5173
5174         bond_set_lockdep_class(bond_dev);
5175
5176         netif_carrier_off(bond_dev);
5177
5178         bond_create_proc_entry(bond);
5179         list_add_tail(&bond->bond_list, &bn->dev_list);
5180
5181         bond_prepare_sysfs_group(bond);
5182
5183         __hw_addr_init(&bond->mc_list);
5184         return 0;
5185 }
5186
5187 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5188 {
5189         if (tb[IFLA_ADDRESS]) {
5190                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5191                         return -EINVAL;
5192                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5193                         return -EADDRNOTAVAIL;
5194         }
5195         return 0;
5196 }
5197
5198 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5199         .kind           = "bond",
5200         .priv_size      = sizeof(struct bonding),
5201         .setup          = bond_setup,
5202         .validate       = bond_validate,
5203 };
5204
5205 /* Create a new bond based on the specified name and bonding parameters.
5206  * If name is NULL, obtain a suitable "bond%d" name for us.
5207  * Caller must NOT hold rtnl_lock; we need to release it here before we
5208  * set up our sysfs entries.
5209  */
5210 int bond_create(struct net *net, const char *name)
5211 {
5212         struct net_device *bond_dev;
5213         int res;
5214
5215         rtnl_lock();
5216
5217         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5218                                 bond_setup, tx_queues);
5219         if (!bond_dev) {
5220                 pr_err("%s: eek! can't alloc netdev!\n", name);
5221                 rtnl_unlock();
5222                 return -ENOMEM;
5223         }
5224
5225         dev_net_set(bond_dev, net);
5226         bond_dev->rtnl_link_ops = &bond_link_ops;
5227
5228         if (!name) {
5229                 res = dev_alloc_name(bond_dev, "bond%d");
5230                 if (res < 0)
5231                         goto out;
5232         } else {
5233                 /*
5234                  * If we're given a name to register
5235                  * we need to ensure that its not already
5236                  * registered
5237                  */
5238                 res = -EEXIST;
5239                 if (__dev_get_by_name(net, name) != NULL)
5240                         goto out;
5241         }
5242
5243         res = register_netdevice(bond_dev);
5244
5245 out:
5246         rtnl_unlock();
5247         if (res < 0)
5248                 bond_destructor(bond_dev);
5249         return res;
5250 }
5251
5252 static int __net_init bond_net_init(struct net *net)
5253 {
5254         struct bond_net *bn = net_generic(net, bond_net_id);
5255
5256         bn->net = net;
5257         INIT_LIST_HEAD(&bn->dev_list);
5258
5259         bond_create_proc_dir(bn);
5260         
5261         return 0;
5262 }
5263
5264 static void __net_exit bond_net_exit(struct net *net)
5265 {
5266         struct bond_net *bn = net_generic(net, bond_net_id);
5267
5268         bond_destroy_proc_dir(bn);
5269 }
5270
5271 static struct pernet_operations bond_net_ops = {
5272         .init = bond_net_init,
5273         .exit = bond_net_exit,
5274         .id   = &bond_net_id,
5275         .size = sizeof(struct bond_net),
5276 };
5277
5278 static int __init bonding_init(void)
5279 {
5280         int i;
5281         int res;
5282
5283         pr_info("%s", version);
5284
5285         res = bond_check_params(&bonding_defaults);
5286         if (res)
5287                 goto out;
5288
5289         res = register_pernet_subsys(&bond_net_ops);
5290         if (res)
5291                 goto out;
5292
5293         res = rtnl_link_register(&bond_link_ops);
5294         if (res)
5295                 goto err_link;
5296
5297         for (i = 0; i < max_bonds; i++) {
5298                 res = bond_create(&init_net, NULL);
5299                 if (res)
5300                         goto err;
5301         }
5302
5303         res = bond_create_sysfs();
5304         if (res)
5305                 goto err;
5306
5307         register_netdevice_notifier(&bond_netdev_notifier);
5308         register_inetaddr_notifier(&bond_inetaddr_notifier);
5309         bond_register_ipv6_notifier();
5310 out:
5311         return res;
5312 err:
5313         rtnl_link_unregister(&bond_link_ops);
5314 err_link:
5315         unregister_pernet_subsys(&bond_net_ops);
5316         goto out;
5317
5318 }
5319
5320 static void __exit bonding_exit(void)
5321 {
5322         unregister_netdevice_notifier(&bond_netdev_notifier);
5323         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5324         bond_unregister_ipv6_notifier();
5325
5326         bond_destroy_sysfs();
5327
5328         rtnl_link_unregister(&bond_link_ops);
5329         unregister_pernet_subsys(&bond_net_ops);
5330 }
5331
5332 module_init(bonding_init);
5333 module_exit(bonding_exit);
5334 MODULE_LICENSE("GPL");
5335 MODULE_VERSION(DRV_VERSION);
5336 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5337 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5338 MODULE_ALIAS_RTNL_LINK("bond");