]> Pileus Git - ~andy/linux/blob - drivers/net/bonding/bond_main.c
bonding: add retransmit membership reports tunable
[~andy/linux] / 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                 np->real_dev = np->dev = skb->dev;
453                 slave_dev->priv_flags |= IFF_IN_NETPOLL;
454                 netpoll_send_skb(np, skb);
455                 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
456                 np->dev = bond->dev;
457         } else
458 #endif
459                 dev_queue_xmit(skb);
460
461         return 0;
462 }
463
464 /*
465  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
466  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
467  * lock because:
468  * a. This operation is performed in IOCTL context,
469  * b. The operation is protected by the RTNL semaphore in the 8021q code,
470  * c. Holding a lock with BH disabled while directly calling a base driver
471  *    entry point is generally a BAD idea.
472  *
473  * The design of synchronization/protection for this operation in the 8021q
474  * module is good for one or more VLAN devices over a single physical device
475  * and cannot be extended for a teaming solution like bonding, so there is a
476  * potential race condition here where a net device from the vlan group might
477  * be referenced (either by a base driver or the 8021q code) while it is being
478  * removed from the system. However, it turns out we're not making matters
479  * worse, and if it works for regular VLAN usage it will work here too.
480 */
481
482 /**
483  * bond_vlan_rx_register - Propagates registration to slaves
484  * @bond_dev: bonding net device that got called
485  * @grp: vlan group being registered
486  */
487 static void bond_vlan_rx_register(struct net_device *bond_dev,
488                                   struct vlan_group *grp)
489 {
490         struct bonding *bond = netdev_priv(bond_dev);
491         struct slave *slave;
492         int i;
493
494         write_lock(&bond->lock);
495         bond->vlgrp = grp;
496         write_unlock(&bond->lock);
497
498         bond_for_each_slave(bond, slave, i) {
499                 struct net_device *slave_dev = slave->dev;
500                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
501
502                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
503                     slave_ops->ndo_vlan_rx_register) {
504                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
505                 }
506         }
507 }
508
509 /**
510  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
511  * @bond_dev: bonding net device that got called
512  * @vid: vlan id being added
513  */
514 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
515 {
516         struct bonding *bond = netdev_priv(bond_dev);
517         struct slave *slave;
518         int i, res;
519
520         bond_for_each_slave(bond, slave, i) {
521                 struct net_device *slave_dev = slave->dev;
522                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
523
524                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
525                     slave_ops->ndo_vlan_rx_add_vid) {
526                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
527                 }
528         }
529
530         res = bond_add_vlan(bond, vid);
531         if (res) {
532                 pr_err("%s: Error: Failed to add vlan id %d\n",
533                        bond_dev->name, vid);
534         }
535 }
536
537 /**
538  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
539  * @bond_dev: bonding net device that got called
540  * @vid: vlan id being removed
541  */
542 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
543 {
544         struct bonding *bond = netdev_priv(bond_dev);
545         struct slave *slave;
546         struct net_device *vlan_dev;
547         int i, res;
548
549         bond_for_each_slave(bond, slave, i) {
550                 struct net_device *slave_dev = slave->dev;
551                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
552
553                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
554                     slave_ops->ndo_vlan_rx_kill_vid) {
555                         /* Save and then restore vlan_dev in the grp array,
556                          * since the slave's driver might clear it.
557                          */
558                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
559                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
560                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
561                 }
562         }
563
564         res = bond_del_vlan(bond, vid);
565         if (res) {
566                 pr_err("%s: Error: Failed to remove vlan id %d\n",
567                        bond_dev->name, vid);
568         }
569 }
570
571 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
572 {
573         struct vlan_entry *vlan;
574         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
575
576         if (!bond->vlgrp)
577                 return;
578
579         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
580             slave_ops->ndo_vlan_rx_register)
581                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
582
583         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
584             !(slave_ops->ndo_vlan_rx_add_vid))
585                 return;
586
587         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
588                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
589 }
590
591 static void bond_del_vlans_from_slave(struct bonding *bond,
592                                       struct net_device *slave_dev)
593 {
594         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
595         struct vlan_entry *vlan;
596         struct net_device *vlan_dev;
597
598         if (!bond->vlgrp)
599                 return;
600
601         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
602             !(slave_ops->ndo_vlan_rx_kill_vid))
603                 goto unreg;
604
605         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
606                 if (!vlan->vlan_id)
607                         continue;
608                 /* Save and then restore vlan_dev in the grp array,
609                  * since the slave's driver might clear it.
610                  */
611                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
612                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
613                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
614         }
615
616 unreg:
617         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
618             slave_ops->ndo_vlan_rx_register)
619                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
620 }
621
622 /*------------------------------- Link status -------------------------------*/
623
624 /*
625  * Set the carrier state for the master according to the state of its
626  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
627  * do special 802.3ad magic.
628  *
629  * Returns zero if carrier state does not change, nonzero if it does.
630  */
631 static int bond_set_carrier(struct bonding *bond)
632 {
633         struct slave *slave;
634         int i;
635
636         if (bond->slave_cnt == 0)
637                 goto down;
638
639         if (bond->params.mode == BOND_MODE_8023AD)
640                 return bond_3ad_set_carrier(bond);
641
642         bond_for_each_slave(bond, slave, i) {
643                 if (slave->link == BOND_LINK_UP) {
644                         if (!netif_carrier_ok(bond->dev)) {
645                                 netif_carrier_on(bond->dev);
646                                 return 1;
647                         }
648                         return 0;
649                 }
650         }
651
652 down:
653         if (netif_carrier_ok(bond->dev)) {
654                 netif_carrier_off(bond->dev);
655                 return 1;
656         }
657         return 0;
658 }
659
660 /*
661  * Get link speed and duplex from the slave's base driver
662  * using ethtool. If for some reason the call fails or the
663  * values are invalid, fake speed and duplex to 100/Full
664  * and return error.
665  */
666 static int bond_update_speed_duplex(struct slave *slave)
667 {
668         struct net_device *slave_dev = slave->dev;
669         struct ethtool_cmd etool;
670         int res;
671
672         /* Fake speed and duplex */
673         slave->speed = SPEED_100;
674         slave->duplex = DUPLEX_FULL;
675
676         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
677                 return -1;
678
679         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
680         if (res < 0)
681                 return -1;
682
683         switch (etool.speed) {
684         case SPEED_10:
685         case SPEED_100:
686         case SPEED_1000:
687         case SPEED_10000:
688                 break;
689         default:
690                 return -1;
691         }
692
693         switch (etool.duplex) {
694         case DUPLEX_FULL:
695         case DUPLEX_HALF:
696                 break;
697         default:
698                 return -1;
699         }
700
701         slave->speed = etool.speed;
702         slave->duplex = etool.duplex;
703
704         return 0;
705 }
706
707 /*
708  * if <dev> supports MII link status reporting, check its link status.
709  *
710  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
711  * depending upon the setting of the use_carrier parameter.
712  *
713  * Return either BMSR_LSTATUS, meaning that the link is up (or we
714  * can't tell and just pretend it is), or 0, meaning that the link is
715  * down.
716  *
717  * If reporting is non-zero, instead of faking link up, return -1 if
718  * both ETHTOOL and MII ioctls fail (meaning the device does not
719  * support them).  If use_carrier is set, return whatever it says.
720  * It'd be nice if there was a good way to tell if a driver supports
721  * netif_carrier, but there really isn't.
722  */
723 static int bond_check_dev_link(struct bonding *bond,
724                                struct net_device *slave_dev, int reporting)
725 {
726         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
727         int (*ioctl)(struct net_device *, struct ifreq *, int);
728         struct ifreq ifr;
729         struct mii_ioctl_data *mii;
730
731         if (!reporting && !netif_running(slave_dev))
732                 return 0;
733
734         if (bond->params.use_carrier)
735                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
736
737         /* Try to get link status using Ethtool first. */
738         if (slave_dev->ethtool_ops) {
739                 if (slave_dev->ethtool_ops->get_link) {
740                         u32 link;
741
742                         link = slave_dev->ethtool_ops->get_link(slave_dev);
743
744                         return link ? BMSR_LSTATUS : 0;
745                 }
746         }
747
748         /* Ethtool can't be used, fallback to MII ioctls. */
749         ioctl = slave_ops->ndo_do_ioctl;
750         if (ioctl) {
751                 /* TODO: set pointer to correct ioctl on a per team member */
752                 /*       bases to make this more efficient. that is, once  */
753                 /*       we determine the correct ioctl, we will always    */
754                 /*       call it and not the others for that team          */
755                 /*       member.                                           */
756
757                 /*
758                  * We cannot assume that SIOCGMIIPHY will also read a
759                  * register; not all network drivers (e.g., e100)
760                  * support that.
761                  */
762
763                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
764                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
765                 mii = if_mii(&ifr);
766                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
767                         mii->reg_num = MII_BMSR;
768                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
769                                 return mii->val_out & BMSR_LSTATUS;
770                 }
771         }
772
773         /*
774          * If reporting, report that either there's no dev->do_ioctl,
775          * or both SIOCGMIIREG and get_link failed (meaning that we
776          * cannot report link status).  If not reporting, pretend
777          * we're ok.
778          */
779         return reporting ? -1 : BMSR_LSTATUS;
780 }
781
782 /*----------------------------- Multicast list ------------------------------*/
783
784 /*
785  * Push the promiscuity flag down to appropriate slaves
786  */
787 static int bond_set_promiscuity(struct bonding *bond, int inc)
788 {
789         int err = 0;
790         if (USES_PRIMARY(bond->params.mode)) {
791                 /* write lock already acquired */
792                 if (bond->curr_active_slave) {
793                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
794                                                   inc);
795                 }
796         } else {
797                 struct slave *slave;
798                 int i;
799                 bond_for_each_slave(bond, slave, i) {
800                         err = dev_set_promiscuity(slave->dev, inc);
801                         if (err)
802                                 return err;
803                 }
804         }
805         return err;
806 }
807
808 /*
809  * Push the allmulti flag down to all slaves
810  */
811 static int bond_set_allmulti(struct bonding *bond, int inc)
812 {
813         int err = 0;
814         if (USES_PRIMARY(bond->params.mode)) {
815                 /* write lock already acquired */
816                 if (bond->curr_active_slave) {
817                         err = dev_set_allmulti(bond->curr_active_slave->dev,
818                                                inc);
819                 }
820         } else {
821                 struct slave *slave;
822                 int i;
823                 bond_for_each_slave(bond, slave, i) {
824                         err = dev_set_allmulti(slave->dev, inc);
825                         if (err)
826                                 return err;
827                 }
828         }
829         return err;
830 }
831
832 /*
833  * Add a Multicast address to slaves
834  * according to mode
835  */
836 static void bond_mc_add(struct bonding *bond, void *addr)
837 {
838         if (USES_PRIMARY(bond->params.mode)) {
839                 /* write lock already acquired */
840                 if (bond->curr_active_slave)
841                         dev_mc_add(bond->curr_active_slave->dev, addr);
842         } else {
843                 struct slave *slave;
844                 int i;
845
846                 bond_for_each_slave(bond, slave, i)
847                         dev_mc_add(slave->dev, addr);
848         }
849 }
850
851 /*
852  * Remove a multicast address from slave
853  * according to mode
854  */
855 static void bond_mc_del(struct bonding *bond, void *addr)
856 {
857         if (USES_PRIMARY(bond->params.mode)) {
858                 /* write lock already acquired */
859                 if (bond->curr_active_slave)
860                         dev_mc_del(bond->curr_active_slave->dev, addr);
861         } else {
862                 struct slave *slave;
863                 int i;
864                 bond_for_each_slave(bond, slave, i) {
865                         dev_mc_del(slave->dev, addr);
866                 }
867         }
868 }
869
870
871 static void __bond_resend_igmp_join_requests(struct net_device *dev)
872 {
873         struct in_device *in_dev;
874         struct ip_mc_list *im;
875
876         rcu_read_lock();
877         in_dev = __in_dev_get_rcu(dev);
878         if (in_dev) {
879                 for (im = in_dev->mc_list; im; im = im->next)
880                         ip_mc_rejoin_group(im);
881         }
882
883         rcu_read_unlock();
884 }
885
886 /*
887  * Retrieve the list of registered multicast addresses for the bonding
888  * device and retransmit an IGMP JOIN request to the current active
889  * slave.
890  */
891 static void bond_resend_igmp_join_requests(struct bonding *bond)
892 {
893         struct net_device *vlan_dev;
894         struct vlan_entry *vlan;
895
896         read_lock(&bond->lock);
897
898         /* rejoin all groups on bond device */
899         __bond_resend_igmp_join_requests(bond->dev);
900
901         /* rejoin all groups on vlan devices */
902         if (bond->vlgrp) {
903                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
904                         vlan_dev = vlan_group_get_device(bond->vlgrp,
905                                                          vlan->vlan_id);
906                         if (vlan_dev)
907                                 __bond_resend_igmp_join_requests(vlan_dev);
908                 }
909         }
910
911         if (--bond->igmp_retrans > 0)
912                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
913
914         read_unlock(&bond->lock);
915 }
916
917 void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
918 {
919         struct bonding *bond = container_of(work, struct bonding,
920                                                         mcast_work.work);
921         bond_resend_igmp_join_requests(bond);
922 }
923
924 /*
925  * flush all members of flush->mc_list from device dev->mc_list
926  */
927 static void bond_mc_list_flush(struct net_device *bond_dev,
928                                struct net_device *slave_dev)
929 {
930         struct bonding *bond = netdev_priv(bond_dev);
931         struct netdev_hw_addr *ha;
932
933         netdev_for_each_mc_addr(ha, bond_dev)
934                 dev_mc_del(slave_dev, ha->addr);
935
936         if (bond->params.mode == BOND_MODE_8023AD) {
937                 /* del lacpdu mc addr from mc list */
938                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
939
940                 dev_mc_del(slave_dev, lacpdu_multicast);
941         }
942 }
943
944 /*--------------------------- Active slave change ---------------------------*/
945
946 /*
947  * Update the mc list and multicast-related flags for the new and
948  * old active slaves (if any) according to the multicast mode, and
949  * promiscuous flags unconditionally.
950  */
951 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
952                          struct slave *old_active)
953 {
954         struct netdev_hw_addr *ha;
955
956         if (!USES_PRIMARY(bond->params.mode))
957                 /* nothing to do -  mc list is already up-to-date on
958                  * all slaves
959                  */
960                 return;
961
962         if (old_active) {
963                 if (bond->dev->flags & IFF_PROMISC)
964                         dev_set_promiscuity(old_active->dev, -1);
965
966                 if (bond->dev->flags & IFF_ALLMULTI)
967                         dev_set_allmulti(old_active->dev, -1);
968
969                 netdev_for_each_mc_addr(ha, bond->dev)
970                         dev_mc_del(old_active->dev, ha->addr);
971         }
972
973         if (new_active) {
974                 /* FIXME: Signal errors upstream. */
975                 if (bond->dev->flags & IFF_PROMISC)
976                         dev_set_promiscuity(new_active->dev, 1);
977
978                 if (bond->dev->flags & IFF_ALLMULTI)
979                         dev_set_allmulti(new_active->dev, 1);
980
981                 netdev_for_each_mc_addr(ha, bond->dev)
982                         dev_mc_add(new_active->dev, ha->addr);
983         }
984 }
985
986 /*
987  * bond_do_fail_over_mac
988  *
989  * Perform special MAC address swapping for fail_over_mac settings
990  *
991  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
992  */
993 static void bond_do_fail_over_mac(struct bonding *bond,
994                                   struct slave *new_active,
995                                   struct slave *old_active)
996         __releases(&bond->curr_slave_lock)
997         __releases(&bond->lock)
998         __acquires(&bond->lock)
999         __acquires(&bond->curr_slave_lock)
1000 {
1001         u8 tmp_mac[ETH_ALEN];
1002         struct sockaddr saddr;
1003         int rv;
1004
1005         switch (bond->params.fail_over_mac) {
1006         case BOND_FOM_ACTIVE:
1007                 if (new_active)
1008                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1009                                new_active->dev->addr_len);
1010                 break;
1011         case BOND_FOM_FOLLOW:
1012                 /*
1013                  * if new_active && old_active, swap them
1014                  * if just old_active, do nothing (going to no active slave)
1015                  * if just new_active, set new_active to bond's MAC
1016                  */
1017                 if (!new_active)
1018                         return;
1019
1020                 write_unlock_bh(&bond->curr_slave_lock);
1021                 read_unlock(&bond->lock);
1022
1023                 if (old_active) {
1024                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1025                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1026                                ETH_ALEN);
1027                         saddr.sa_family = new_active->dev->type;
1028                 } else {
1029                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1030                         saddr.sa_family = bond->dev->type;
1031                 }
1032
1033                 rv = dev_set_mac_address(new_active->dev, &saddr);
1034                 if (rv) {
1035                         pr_err("%s: Error %d setting MAC of slave %s\n",
1036                                bond->dev->name, -rv, new_active->dev->name);
1037                         goto out;
1038                 }
1039
1040                 if (!old_active)
1041                         goto out;
1042
1043                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1044                 saddr.sa_family = old_active->dev->type;
1045
1046                 rv = dev_set_mac_address(old_active->dev, &saddr);
1047                 if (rv)
1048                         pr_err("%s: Error %d setting MAC of slave %s\n",
1049                                bond->dev->name, -rv, new_active->dev->name);
1050 out:
1051                 read_lock(&bond->lock);
1052                 write_lock_bh(&bond->curr_slave_lock);
1053                 break;
1054         default:
1055                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1056                        bond->dev->name, bond->params.fail_over_mac);
1057                 break;
1058         }
1059
1060 }
1061
1062 static bool bond_should_change_active(struct bonding *bond)
1063 {
1064         struct slave *prim = bond->primary_slave;
1065         struct slave *curr = bond->curr_active_slave;
1066
1067         if (!prim || !curr || curr->link != BOND_LINK_UP)
1068                 return true;
1069         if (bond->force_primary) {
1070                 bond->force_primary = false;
1071                 return true;
1072         }
1073         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1074             (prim->speed < curr->speed ||
1075              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1076                 return false;
1077         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1078                 return false;
1079         return true;
1080 }
1081
1082 /**
1083  * find_best_interface - select the best available slave to be the active one
1084  * @bond: our bonding struct
1085  *
1086  * Warning: Caller must hold curr_slave_lock for writing.
1087  */
1088 static struct slave *bond_find_best_slave(struct bonding *bond)
1089 {
1090         struct slave *new_active, *old_active;
1091         struct slave *bestslave = NULL;
1092         int mintime = bond->params.updelay;
1093         int i;
1094
1095         new_active = bond->curr_active_slave;
1096
1097         if (!new_active) { /* there were no active slaves left */
1098                 if (bond->slave_cnt > 0)   /* found one slave */
1099                         new_active = bond->first_slave;
1100                 else
1101                         return NULL; /* still no slave, return NULL */
1102         }
1103
1104         if ((bond->primary_slave) &&
1105             bond->primary_slave->link == BOND_LINK_UP &&
1106             bond_should_change_active(bond)) {
1107                 new_active = bond->primary_slave;
1108         }
1109
1110         /* remember where to stop iterating over the slaves */
1111         old_active = new_active;
1112
1113         bond_for_each_slave_from(bond, new_active, i, old_active) {
1114                 if (new_active->link == BOND_LINK_UP) {
1115                         return new_active;
1116                 } else if (new_active->link == BOND_LINK_BACK &&
1117                            IS_UP(new_active->dev)) {
1118                         /* link up, but waiting for stabilization */
1119                         if (new_active->delay < mintime) {
1120                                 mintime = new_active->delay;
1121                                 bestslave = new_active;
1122                         }
1123                 }
1124         }
1125
1126         return bestslave;
1127 }
1128
1129 /**
1130  * change_active_interface - change the active slave into the specified one
1131  * @bond: our bonding struct
1132  * @new: the new slave to make the active one
1133  *
1134  * Set the new slave to the bond's settings and unset them on the old
1135  * curr_active_slave.
1136  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1137  *
1138  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1139  * because it is apparently the best available slave we have, even though its
1140  * updelay hasn't timed out yet.
1141  *
1142  * If new_active is not NULL, caller must hold bond->lock for read and
1143  * curr_slave_lock for write_bh.
1144  */
1145 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1146 {
1147         struct slave *old_active = bond->curr_active_slave;
1148
1149         if (old_active == new_active)
1150                 return;
1151
1152         if (new_active) {
1153                 new_active->jiffies = jiffies;
1154
1155                 if (new_active->link == BOND_LINK_BACK) {
1156                         if (USES_PRIMARY(bond->params.mode)) {
1157                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1158                                         bond->dev->name, new_active->dev->name,
1159                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1160                         }
1161
1162                         new_active->delay = 0;
1163                         new_active->link = BOND_LINK_UP;
1164
1165                         if (bond->params.mode == BOND_MODE_8023AD)
1166                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1167
1168                         if (bond_is_lb(bond))
1169                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1170                 } else {
1171                         if (USES_PRIMARY(bond->params.mode)) {
1172                                 pr_info("%s: making interface %s the new active one.\n",
1173                                         bond->dev->name, new_active->dev->name);
1174                         }
1175                 }
1176         }
1177
1178         if (USES_PRIMARY(bond->params.mode))
1179                 bond_mc_swap(bond, new_active, old_active);
1180
1181         if (bond_is_lb(bond)) {
1182                 bond_alb_handle_active_change(bond, new_active);
1183                 if (old_active)
1184                         bond_set_slave_inactive_flags(old_active);
1185                 if (new_active)
1186                         bond_set_slave_active_flags(new_active);
1187         } else {
1188                 bond->curr_active_slave = new_active;
1189         }
1190
1191         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1192                 if (old_active)
1193                         bond_set_slave_inactive_flags(old_active);
1194
1195                 if (new_active) {
1196                         bond_set_slave_active_flags(new_active);
1197
1198                         if (bond->params.fail_over_mac)
1199                                 bond_do_fail_over_mac(bond, new_active,
1200                                                       old_active);
1201
1202                         bond->send_grat_arp = bond->params.num_grat_arp;
1203                         bond_send_gratuitous_arp(bond);
1204
1205                         bond->send_unsol_na = bond->params.num_unsol_na;
1206                         bond_send_unsolicited_na(bond);
1207
1208                         write_unlock_bh(&bond->curr_slave_lock);
1209                         read_unlock(&bond->lock);
1210
1211                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1212
1213                         read_lock(&bond->lock);
1214                         write_lock_bh(&bond->curr_slave_lock);
1215                 }
1216         }
1217
1218         /* resend IGMP joins since active slave has changed or
1219          * all were sent on curr_active_slave */
1220         if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1221             bond->params.mode == BOND_MODE_ROUNDROBIN) {
1222                 bond->igmp_retrans = bond->params.resend_igmp;
1223                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1224         }
1225 }
1226
1227 /**
1228  * bond_select_active_slave - select a new active slave, if needed
1229  * @bond: our bonding struct
1230  *
1231  * This functions should be called when one of the following occurs:
1232  * - The old curr_active_slave has been released or lost its link.
1233  * - The primary_slave has got its link back.
1234  * - A slave has got its link back and there's no old curr_active_slave.
1235  *
1236  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1237  */
1238 void bond_select_active_slave(struct bonding *bond)
1239 {
1240         struct slave *best_slave;
1241         int rv;
1242
1243         best_slave = bond_find_best_slave(bond);
1244         if (best_slave != bond->curr_active_slave) {
1245                 bond_change_active_slave(bond, best_slave);
1246                 rv = bond_set_carrier(bond);
1247                 if (!rv)
1248                         return;
1249
1250                 if (netif_carrier_ok(bond->dev)) {
1251                         pr_info("%s: first active interface up!\n",
1252                                 bond->dev->name);
1253                 } else {
1254                         pr_info("%s: now running without any active interface !\n",
1255                                 bond->dev->name);
1256                 }
1257         }
1258 }
1259
1260 /*--------------------------- slave list handling ---------------------------*/
1261
1262 /*
1263  * This function attaches the slave to the end of list.
1264  *
1265  * bond->lock held for writing by caller.
1266  */
1267 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1268 {
1269         if (bond->first_slave == NULL) { /* attaching the first slave */
1270                 new_slave->next = new_slave;
1271                 new_slave->prev = new_slave;
1272                 bond->first_slave = new_slave;
1273         } else {
1274                 new_slave->next = bond->first_slave;
1275                 new_slave->prev = bond->first_slave->prev;
1276                 new_slave->next->prev = new_slave;
1277                 new_slave->prev->next = new_slave;
1278         }
1279
1280         bond->slave_cnt++;
1281 }
1282
1283 /*
1284  * This function detaches the slave from the list.
1285  * WARNING: no check is made to verify if the slave effectively
1286  * belongs to <bond>.
1287  * Nothing is freed on return, structures are just unchained.
1288  * If any slave pointer in bond was pointing to <slave>,
1289  * it should be changed by the calling function.
1290  *
1291  * bond->lock held for writing by caller.
1292  */
1293 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1294 {
1295         if (slave->next)
1296                 slave->next->prev = slave->prev;
1297
1298         if (slave->prev)
1299                 slave->prev->next = slave->next;
1300
1301         if (bond->first_slave == slave) { /* slave is the first slave */
1302                 if (bond->slave_cnt > 1) { /* there are more slave */
1303                         bond->first_slave = slave->next;
1304                 } else {
1305                         bond->first_slave = NULL; /* slave was the last one */
1306                 }
1307         }
1308
1309         slave->next = NULL;
1310         slave->prev = NULL;
1311         bond->slave_cnt--;
1312 }
1313
1314 #ifdef CONFIG_NET_POLL_CONTROLLER
1315 /*
1316  * You must hold read lock on bond->lock before calling this.
1317  */
1318 static bool slaves_support_netpoll(struct net_device *bond_dev)
1319 {
1320         struct bonding *bond = netdev_priv(bond_dev);
1321         struct slave *slave;
1322         int i = 0;
1323         bool ret = true;
1324
1325         bond_for_each_slave(bond, slave, i) {
1326                 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1327                     !slave->dev->netdev_ops->ndo_poll_controller)
1328                         ret = false;
1329         }
1330         return i != 0 && ret;
1331 }
1332
1333 static void bond_poll_controller(struct net_device *bond_dev)
1334 {
1335         struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1336         if (dev != bond_dev)
1337                 netpoll_poll_dev(dev);
1338 }
1339
1340 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1341 {
1342         struct bonding *bond = netdev_priv(bond_dev);
1343         struct slave *slave;
1344         const struct net_device_ops *ops;
1345         int i;
1346
1347         read_lock(&bond->lock);
1348         bond_dev->npinfo = NULL;
1349         bond_for_each_slave(bond, slave, i) {
1350                 if (slave->dev) {
1351                         ops = slave->dev->netdev_ops;
1352                         if (ops->ndo_netpoll_cleanup)
1353                                 ops->ndo_netpoll_cleanup(slave->dev);
1354                         else
1355                                 slave->dev->npinfo = NULL;
1356                 }
1357         }
1358         read_unlock(&bond->lock);
1359 }
1360
1361 #else
1362
1363 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1364 {
1365 }
1366
1367 #endif
1368
1369 /*---------------------------------- IOCTL ----------------------------------*/
1370
1371 static int bond_sethwaddr(struct net_device *bond_dev,
1372                           struct net_device *slave_dev)
1373 {
1374         pr_debug("bond_dev=%p\n", bond_dev);
1375         pr_debug("slave_dev=%p\n", slave_dev);
1376         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1377         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1378         return 0;
1379 }
1380
1381 #define BOND_VLAN_FEATURES \
1382         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1383          NETIF_F_HW_VLAN_FILTER)
1384
1385 /*
1386  * Compute the common dev->feature set available to all slaves.  Some
1387  * feature bits are managed elsewhere, so preserve those feature bits
1388  * on the master device.
1389  */
1390 static int bond_compute_features(struct bonding *bond)
1391 {
1392         struct slave *slave;
1393         struct net_device *bond_dev = bond->dev;
1394         unsigned long features = bond_dev->features;
1395         unsigned long vlan_features = 0;
1396         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1397                                                 bond_dev->hard_header_len);
1398         int i;
1399
1400         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1401         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1402
1403         if (!bond->first_slave)
1404                 goto done;
1405
1406         features &= ~NETIF_F_ONE_FOR_ALL;
1407
1408         vlan_features = bond->first_slave->dev->vlan_features;
1409         bond_for_each_slave(bond, slave, i) {
1410                 features = netdev_increment_features(features,
1411                                                      slave->dev->features,
1412                                                      NETIF_F_ONE_FOR_ALL);
1413                 vlan_features = netdev_increment_features(vlan_features,
1414                                                         slave->dev->vlan_features,
1415                                                         NETIF_F_ONE_FOR_ALL);
1416                 if (slave->dev->hard_header_len > max_hard_header_len)
1417                         max_hard_header_len = slave->dev->hard_header_len;
1418         }
1419
1420 done:
1421         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1422         bond_dev->features = netdev_fix_features(features, NULL);
1423         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1424         bond_dev->hard_header_len = max_hard_header_len;
1425
1426         return 0;
1427 }
1428
1429 static void bond_setup_by_slave(struct net_device *bond_dev,
1430                                 struct net_device *slave_dev)
1431 {
1432         struct bonding *bond = netdev_priv(bond_dev);
1433
1434         bond_dev->header_ops        = slave_dev->header_ops;
1435
1436         bond_dev->type              = slave_dev->type;
1437         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1438         bond_dev->addr_len          = slave_dev->addr_len;
1439
1440         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1441                 slave_dev->addr_len);
1442         bond->setup_by_slave = 1;
1443 }
1444
1445 /* enslave device <slave> to bond device <master> */
1446 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1447 {
1448         struct bonding *bond = netdev_priv(bond_dev);
1449         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1450         struct slave *new_slave = NULL;
1451         struct netdev_hw_addr *ha;
1452         struct sockaddr addr;
1453         int link_reporting;
1454         int old_features = bond_dev->features;
1455         int res = 0;
1456
1457         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1458                 slave_ops->ndo_do_ioctl == NULL) {
1459                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1460                            bond_dev->name, slave_dev->name);
1461         }
1462
1463         /* bond must be initialized by bond_open() before enslaving */
1464         if (!(bond_dev->flags & IFF_UP)) {
1465                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1466                            bond_dev->name);
1467         }
1468
1469         /* already enslaved */
1470         if (slave_dev->flags & IFF_SLAVE) {
1471                 pr_debug("Error, Device was already enslaved\n");
1472                 return -EBUSY;
1473         }
1474
1475         /* vlan challenged mutual exclusion */
1476         /* no need to lock since we're protected by rtnl_lock */
1477         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1478                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1479                 if (bond->vlgrp) {
1480                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1481                                bond_dev->name, slave_dev->name, bond_dev->name);
1482                         return -EPERM;
1483                 } else {
1484                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1485                                    bond_dev->name, slave_dev->name,
1486                                    slave_dev->name, bond_dev->name);
1487                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1488                 }
1489         } else {
1490                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1491                 if (bond->slave_cnt == 0) {
1492                         /* First slave, and it is not VLAN challenged,
1493                          * so remove the block of adding VLANs over the bond.
1494                          */
1495                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1496                 }
1497         }
1498
1499         /*
1500          * Old ifenslave binaries are no longer supported.  These can
1501          * be identified with moderate accuracy by the state of the slave:
1502          * the current ifenslave will set the interface down prior to
1503          * enslaving it; the old ifenslave will not.
1504          */
1505         if ((slave_dev->flags & IFF_UP)) {
1506                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1507                        slave_dev->name);
1508                 res = -EPERM;
1509                 goto err_undo_flags;
1510         }
1511
1512         /* set bonding device ether type by slave - bonding netdevices are
1513          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1514          * there is a need to override some of the type dependent attribs/funcs.
1515          *
1516          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1517          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1518          */
1519         if (bond->slave_cnt == 0) {
1520                 if (bond_dev->type != slave_dev->type) {
1521                         pr_debug("%s: change device type from %d to %d\n",
1522                                  bond_dev->name,
1523                                  bond_dev->type, slave_dev->type);
1524
1525                         res = netdev_bonding_change(bond_dev,
1526                                                     NETDEV_PRE_TYPE_CHANGE);
1527                         res = notifier_to_errno(res);
1528                         if (res) {
1529                                 pr_err("%s: refused to change device type\n",
1530                                        bond_dev->name);
1531                                 res = -EBUSY;
1532                                 goto err_undo_flags;
1533                         }
1534
1535                         /* Flush unicast and multicast addresses */
1536                         dev_uc_flush(bond_dev);
1537                         dev_mc_flush(bond_dev);
1538
1539                         if (slave_dev->type != ARPHRD_ETHER)
1540                                 bond_setup_by_slave(bond_dev, slave_dev);
1541                         else
1542                                 ether_setup(bond_dev);
1543
1544                         netdev_bonding_change(bond_dev,
1545                                               NETDEV_POST_TYPE_CHANGE);
1546                 }
1547         } else if (bond_dev->type != slave_dev->type) {
1548                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1549                        slave_dev->name,
1550                        slave_dev->type, bond_dev->type);
1551                 res = -EINVAL;
1552                 goto err_undo_flags;
1553         }
1554
1555         if (slave_ops->ndo_set_mac_address == NULL) {
1556                 if (bond->slave_cnt == 0) {
1557                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1558                                    bond_dev->name);
1559                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1560                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1561                         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",
1562                                bond_dev->name);
1563                         res = -EOPNOTSUPP;
1564                         goto err_undo_flags;
1565                 }
1566         }
1567
1568         /* If this is the first slave, then we need to set the master's hardware
1569          * address to be the same as the slave's. */
1570         if (bond->slave_cnt == 0)
1571                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1572                        slave_dev->addr_len);
1573
1574
1575         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1576         if (!new_slave) {
1577                 res = -ENOMEM;
1578                 goto err_undo_flags;
1579         }
1580
1581         /*
1582          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1583          * is set via sysfs or module option if desired.
1584          */
1585         new_slave->queue_id = 0;
1586
1587         /* Save slave's original mtu and then set it to match the bond */
1588         new_slave->original_mtu = slave_dev->mtu;
1589         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1590         if (res) {
1591                 pr_debug("Error %d calling dev_set_mtu\n", res);
1592                 goto err_free;
1593         }
1594
1595         /*
1596          * Save slave's original ("permanent") mac address for modes
1597          * that need it, and for restoring it upon release, and then
1598          * set it to the master's address
1599          */
1600         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1601
1602         if (!bond->params.fail_over_mac) {
1603                 /*
1604                  * Set slave to master's mac address.  The application already
1605                  * set the master's mac address to that of the first slave
1606                  */
1607                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1608                 addr.sa_family = slave_dev->type;
1609                 res = dev_set_mac_address(slave_dev, &addr);
1610                 if (res) {
1611                         pr_debug("Error %d calling set_mac_address\n", res);
1612                         goto err_restore_mtu;
1613                 }
1614         }
1615
1616         res = netdev_set_master(slave_dev, bond_dev);
1617         if (res) {
1618                 pr_debug("Error %d calling netdev_set_master\n", res);
1619                 goto err_restore_mac;
1620         }
1621         /* open the slave since the application closed it */
1622         res = dev_open(slave_dev);
1623         if (res) {
1624                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1625                 goto err_unset_master;
1626         }
1627
1628         new_slave->dev = slave_dev;
1629         slave_dev->priv_flags |= IFF_BONDING;
1630
1631         if (bond_is_lb(bond)) {
1632                 /* bond_alb_init_slave() must be called before all other stages since
1633                  * it might fail and we do not want to have to undo everything
1634                  */
1635                 res = bond_alb_init_slave(bond, new_slave);
1636                 if (res)
1637                         goto err_close;
1638         }
1639
1640         /* If the mode USES_PRIMARY, then the new slave gets the
1641          * master's promisc (and mc) settings only if it becomes the
1642          * curr_active_slave, and that is taken care of later when calling
1643          * bond_change_active()
1644          */
1645         if (!USES_PRIMARY(bond->params.mode)) {
1646                 /* set promiscuity level to new slave */
1647                 if (bond_dev->flags & IFF_PROMISC) {
1648                         res = dev_set_promiscuity(slave_dev, 1);
1649                         if (res)
1650                                 goto err_close;
1651                 }
1652
1653                 /* set allmulti level to new slave */
1654                 if (bond_dev->flags & IFF_ALLMULTI) {
1655                         res = dev_set_allmulti(slave_dev, 1);
1656                         if (res)
1657                                 goto err_close;
1658                 }
1659
1660                 netif_addr_lock_bh(bond_dev);
1661                 /* upload master's mc_list to new slave */
1662                 netdev_for_each_mc_addr(ha, bond_dev)
1663                         dev_mc_add(slave_dev, ha->addr);
1664                 netif_addr_unlock_bh(bond_dev);
1665         }
1666
1667         if (bond->params.mode == BOND_MODE_8023AD) {
1668                 /* add lacpdu mc addr to mc list */
1669                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1670
1671                 dev_mc_add(slave_dev, lacpdu_multicast);
1672         }
1673
1674         bond_add_vlans_on_slave(bond, slave_dev);
1675
1676         write_lock_bh(&bond->lock);
1677
1678         bond_attach_slave(bond, new_slave);
1679
1680         new_slave->delay = 0;
1681         new_slave->link_failure_count = 0;
1682
1683         bond_compute_features(bond);
1684
1685         write_unlock_bh(&bond->lock);
1686
1687         read_lock(&bond->lock);
1688
1689         new_slave->last_arp_rx = jiffies;
1690
1691         if (bond->params.miimon && !bond->params.use_carrier) {
1692                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1693
1694                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1695                         /*
1696                          * miimon is set but a bonded network driver
1697                          * does not support ETHTOOL/MII and
1698                          * arp_interval is not set.  Note: if
1699                          * use_carrier is enabled, we will never go
1700                          * here (because netif_carrier is always
1701                          * supported); thus, we don't need to change
1702                          * the messages for netif_carrier.
1703                          */
1704                         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",
1705                                bond_dev->name, slave_dev->name);
1706                 } else if (link_reporting == -1) {
1707                         /* unable get link status using mii/ethtool */
1708                         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",
1709                                    bond_dev->name, slave_dev->name);
1710                 }
1711         }
1712
1713         /* check for initial state */
1714         if (!bond->params.miimon ||
1715             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1716                 if (bond->params.updelay) {
1717                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1718                         new_slave->link  = BOND_LINK_BACK;
1719                         new_slave->delay = bond->params.updelay;
1720                 } else {
1721                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1722                         new_slave->link  = BOND_LINK_UP;
1723                 }
1724                 new_slave->jiffies = jiffies;
1725         } else {
1726                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1727                 new_slave->link  = BOND_LINK_DOWN;
1728         }
1729
1730         if (bond_update_speed_duplex(new_slave) &&
1731             (new_slave->link != BOND_LINK_DOWN)) {
1732                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1733                            bond_dev->name, new_slave->dev->name);
1734
1735                 if (bond->params.mode == BOND_MODE_8023AD) {
1736                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1737                                    bond_dev->name);
1738                 }
1739         }
1740
1741         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1742                 /* if there is a primary slave, remember it */
1743                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1744                         bond->primary_slave = new_slave;
1745                         bond->force_primary = true;
1746                 }
1747         }
1748
1749         write_lock_bh(&bond->curr_slave_lock);
1750
1751         switch (bond->params.mode) {
1752         case BOND_MODE_ACTIVEBACKUP:
1753                 bond_set_slave_inactive_flags(new_slave);
1754                 bond_select_active_slave(bond);
1755                 break;
1756         case BOND_MODE_8023AD:
1757                 /* in 802.3ad mode, the internal mechanism
1758                  * will activate the slaves in the selected
1759                  * aggregator
1760                  */
1761                 bond_set_slave_inactive_flags(new_slave);
1762                 /* if this is the first slave */
1763                 if (bond->slave_cnt == 1) {
1764                         SLAVE_AD_INFO(new_slave).id = 1;
1765                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1766                          * can be called only after the mac address of the bond is set
1767                          */
1768                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1769                                             bond->params.lacp_fast);
1770                 } else {
1771                         SLAVE_AD_INFO(new_slave).id =
1772                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1773                 }
1774
1775                 bond_3ad_bind_slave(new_slave);
1776                 break;
1777         case BOND_MODE_TLB:
1778         case BOND_MODE_ALB:
1779                 new_slave->state = BOND_STATE_ACTIVE;
1780                 bond_set_slave_inactive_flags(new_slave);
1781                 bond_select_active_slave(bond);
1782                 break;
1783         default:
1784                 pr_debug("This slave is always active in trunk mode\n");
1785
1786                 /* always active in trunk mode */
1787                 new_slave->state = BOND_STATE_ACTIVE;
1788
1789                 /* In trunking mode there is little meaning to curr_active_slave
1790                  * anyway (it holds no special properties of the bond device),
1791                  * so we can change it without calling change_active_interface()
1792                  */
1793                 if (!bond->curr_active_slave)
1794                         bond->curr_active_slave = new_slave;
1795
1796                 break;
1797         } /* switch(bond_mode) */
1798
1799         write_unlock_bh(&bond->curr_slave_lock);
1800
1801         bond_set_carrier(bond);
1802
1803 #ifdef CONFIG_NET_POLL_CONTROLLER
1804         /*
1805          * Netpoll and bonding is broken, make sure it is not initialized
1806          * until it is fixed.
1807          */
1808         if (disable_netpoll) {
1809                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1810         } else {
1811                 if (slaves_support_netpoll(bond_dev)) {
1812                         bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1813                         if (bond_dev->npinfo)
1814                                 slave_dev->npinfo = bond_dev->npinfo;
1815                 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1816                         bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1817                         pr_info("New slave device %s does not support netpoll\n",
1818                                 slave_dev->name);
1819                         pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1820                 }
1821         }
1822 #endif
1823         read_unlock(&bond->lock);
1824
1825         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1826         if (res)
1827                 goto err_close;
1828
1829         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1830                 bond_dev->name, slave_dev->name,
1831                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1832                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1833
1834         /* enslave is successful */
1835         return 0;
1836
1837 /* Undo stages on error */
1838 err_close:
1839         dev_close(slave_dev);
1840
1841 err_unset_master:
1842         netdev_set_master(slave_dev, NULL);
1843
1844 err_restore_mac:
1845         if (!bond->params.fail_over_mac) {
1846                 /* XXX TODO - fom follow mode needs to change master's
1847                  * MAC if this slave's MAC is in use by the bond, or at
1848                  * least print a warning.
1849                  */
1850                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1851                 addr.sa_family = slave_dev->type;
1852                 dev_set_mac_address(slave_dev, &addr);
1853         }
1854
1855 err_restore_mtu:
1856         dev_set_mtu(slave_dev, new_slave->original_mtu);
1857
1858 err_free:
1859         kfree(new_slave);
1860
1861 err_undo_flags:
1862         bond_dev->features = old_features;
1863
1864         return res;
1865 }
1866
1867 /*
1868  * Try to release the slave device <slave> from the bond device <master>
1869  * It is legal to access curr_active_slave without a lock because all the function
1870  * is write-locked.
1871  *
1872  * The rules for slave state should be:
1873  *   for Active/Backup:
1874  *     Active stays on all backups go down
1875  *   for Bonded connections:
1876  *     The first up interface should be left on and all others downed.
1877  */
1878 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1879 {
1880         struct bonding *bond = netdev_priv(bond_dev);
1881         struct slave *slave, *oldcurrent;
1882         struct sockaddr addr;
1883
1884         /* slave is not a slave or master is not master of this slave */
1885         if (!(slave_dev->flags & IFF_SLAVE) ||
1886             (slave_dev->master != bond_dev)) {
1887                 pr_err("%s: Error: cannot release %s.\n",
1888                        bond_dev->name, slave_dev->name);
1889                 return -EINVAL;
1890         }
1891
1892         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1893         write_lock_bh(&bond->lock);
1894
1895         slave = bond_get_slave_by_dev(bond, slave_dev);
1896         if (!slave) {
1897                 /* not a slave of this bond */
1898                 pr_info("%s: %s not enslaved\n",
1899                         bond_dev->name, slave_dev->name);
1900                 write_unlock_bh(&bond->lock);
1901                 return -EINVAL;
1902         }
1903
1904         if (!bond->params.fail_over_mac) {
1905                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1906                     bond->slave_cnt > 1)
1907                         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",
1908                                    bond_dev->name, slave_dev->name,
1909                                    slave->perm_hwaddr,
1910                                    bond_dev->name, slave_dev->name);
1911         }
1912
1913         /* Inform AD package of unbinding of slave. */
1914         if (bond->params.mode == BOND_MODE_8023AD) {
1915                 /* must be called before the slave is
1916                  * detached from the list
1917                  */
1918                 bond_3ad_unbind_slave(slave);
1919         }
1920
1921         pr_info("%s: releasing %s interface %s\n",
1922                 bond_dev->name,
1923                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1924                 slave_dev->name);
1925
1926         oldcurrent = bond->curr_active_slave;
1927
1928         bond->current_arp_slave = NULL;
1929
1930         /* release the slave from its bond */
1931         bond_detach_slave(bond, slave);
1932
1933         bond_compute_features(bond);
1934
1935         if (bond->primary_slave == slave)
1936                 bond->primary_slave = NULL;
1937
1938         if (oldcurrent == slave)
1939                 bond_change_active_slave(bond, NULL);
1940
1941         if (bond_is_lb(bond)) {
1942                 /* Must be called only after the slave has been
1943                  * detached from the list and the curr_active_slave
1944                  * has been cleared (if our_slave == old_current),
1945                  * but before a new active slave is selected.
1946                  */
1947                 write_unlock_bh(&bond->lock);
1948                 bond_alb_deinit_slave(bond, slave);
1949                 write_lock_bh(&bond->lock);
1950         }
1951
1952         if (oldcurrent == slave) {
1953                 /*
1954                  * Note that we hold RTNL over this sequence, so there
1955                  * is no concern that another slave add/remove event
1956                  * will interfere.
1957                  */
1958                 write_unlock_bh(&bond->lock);
1959                 read_lock(&bond->lock);
1960                 write_lock_bh(&bond->curr_slave_lock);
1961
1962                 bond_select_active_slave(bond);
1963
1964                 write_unlock_bh(&bond->curr_slave_lock);
1965                 read_unlock(&bond->lock);
1966                 write_lock_bh(&bond->lock);
1967         }
1968
1969         if (bond->slave_cnt == 0) {
1970                 bond_set_carrier(bond);
1971
1972                 /* if the last slave was removed, zero the mac address
1973                  * of the master so it will be set by the application
1974                  * to the mac address of the first slave
1975                  */
1976                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1977
1978                 if (!bond->vlgrp) {
1979                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1980                 } else {
1981                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1982                                    bond_dev->name, bond_dev->name);
1983                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1984                                    bond_dev->name);
1985                 }
1986         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1987                    !bond_has_challenged_slaves(bond)) {
1988                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1989                         bond_dev->name, slave_dev->name, bond_dev->name);
1990                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1991         }
1992
1993         write_unlock_bh(&bond->lock);
1994
1995         /* must do this from outside any spinlocks */
1996         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1997
1998         bond_del_vlans_from_slave(bond, slave_dev);
1999
2000         /* If the mode USES_PRIMARY, then we should only remove its
2001          * promisc and mc settings if it was the curr_active_slave, but that was
2002          * already taken care of above when we detached the slave
2003          */
2004         if (!USES_PRIMARY(bond->params.mode)) {
2005                 /* unset promiscuity level from slave */
2006                 if (bond_dev->flags & IFF_PROMISC)
2007                         dev_set_promiscuity(slave_dev, -1);
2008
2009                 /* unset allmulti level from slave */
2010                 if (bond_dev->flags & IFF_ALLMULTI)
2011                         dev_set_allmulti(slave_dev, -1);
2012
2013                 /* flush master's mc_list from slave */
2014                 netif_addr_lock_bh(bond_dev);
2015                 bond_mc_list_flush(bond_dev, slave_dev);
2016                 netif_addr_unlock_bh(bond_dev);
2017         }
2018
2019         netdev_set_master(slave_dev, NULL);
2020
2021 #ifdef CONFIG_NET_POLL_CONTROLLER
2022         read_lock_bh(&bond->lock);
2023
2024          /* Make sure netpoll over stays disabled until fixed. */
2025         if (!disable_netpoll)
2026                 if (slaves_support_netpoll(bond_dev))
2027                                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2028         read_unlock_bh(&bond->lock);
2029         if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2030                 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2031         else
2032                 slave_dev->npinfo = NULL;
2033 #endif
2034
2035         /* close slave before restoring its mac address */
2036         dev_close(slave_dev);
2037
2038         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2039                 /* restore original ("permanent") mac address */
2040                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2041                 addr.sa_family = slave_dev->type;
2042                 dev_set_mac_address(slave_dev, &addr);
2043         }
2044
2045         dev_set_mtu(slave_dev, slave->original_mtu);
2046
2047         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2048                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
2049                                    IFF_SLAVE_NEEDARP);
2050
2051         kfree(slave);
2052
2053         return 0;  /* deletion OK */
2054 }
2055
2056 /*
2057 * First release a slave and than destroy the bond if no more slaves are left.
2058 * Must be under rtnl_lock when this function is called.
2059 */
2060 int  bond_release_and_destroy(struct net_device *bond_dev,
2061                               struct net_device *slave_dev)
2062 {
2063         struct bonding *bond = netdev_priv(bond_dev);
2064         int ret;
2065
2066         ret = bond_release(bond_dev, slave_dev);
2067         if ((ret == 0) && (bond->slave_cnt == 0)) {
2068                 pr_info("%s: destroying bond %s.\n",
2069                         bond_dev->name, bond_dev->name);
2070                 unregister_netdevice(bond_dev);
2071         }
2072         return ret;
2073 }
2074
2075 /*
2076  * This function releases all slaves.
2077  */
2078 static int bond_release_all(struct net_device *bond_dev)
2079 {
2080         struct bonding *bond = netdev_priv(bond_dev);
2081         struct slave *slave;
2082         struct net_device *slave_dev;
2083         struct sockaddr addr;
2084
2085         write_lock_bh(&bond->lock);
2086
2087         netif_carrier_off(bond_dev);
2088
2089         if (bond->slave_cnt == 0)
2090                 goto out;
2091
2092         bond->current_arp_slave = NULL;
2093         bond->primary_slave = NULL;
2094         bond_change_active_slave(bond, NULL);
2095
2096         while ((slave = bond->first_slave) != NULL) {
2097                 /* Inform AD package of unbinding of slave
2098                  * before slave is detached from the list.
2099                  */
2100                 if (bond->params.mode == BOND_MODE_8023AD)
2101                         bond_3ad_unbind_slave(slave);
2102
2103                 slave_dev = slave->dev;
2104                 bond_detach_slave(bond, slave);
2105
2106                 /* now that the slave is detached, unlock and perform
2107                  * all the undo steps that should not be called from
2108                  * within a lock.
2109                  */
2110                 write_unlock_bh(&bond->lock);
2111
2112                 if (bond_is_lb(bond)) {
2113                         /* must be called only after the slave
2114                          * has been detached from the list
2115                          */
2116                         bond_alb_deinit_slave(bond, slave);
2117                 }
2118
2119                 bond_compute_features(bond);
2120
2121                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2122                 bond_del_vlans_from_slave(bond, slave_dev);
2123
2124                 /* If the mode USES_PRIMARY, then we should only remove its
2125                  * promisc and mc settings if it was the curr_active_slave, but that was
2126                  * already taken care of above when we detached the slave
2127                  */
2128                 if (!USES_PRIMARY(bond->params.mode)) {
2129                         /* unset promiscuity level from slave */
2130                         if (bond_dev->flags & IFF_PROMISC)
2131                                 dev_set_promiscuity(slave_dev, -1);
2132
2133                         /* unset allmulti level from slave */
2134                         if (bond_dev->flags & IFF_ALLMULTI)
2135                                 dev_set_allmulti(slave_dev, -1);
2136
2137                         /* flush master's mc_list from slave */
2138                         netif_addr_lock_bh(bond_dev);
2139                         bond_mc_list_flush(bond_dev, slave_dev);
2140                         netif_addr_unlock_bh(bond_dev);
2141                 }
2142
2143                 netdev_set_master(slave_dev, NULL);
2144
2145                 /* close slave before restoring its mac address */
2146                 dev_close(slave_dev);
2147
2148                 if (!bond->params.fail_over_mac) {
2149                         /* restore original ("permanent") mac address*/
2150                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2151                         addr.sa_family = slave_dev->type;
2152                         dev_set_mac_address(slave_dev, &addr);
2153                 }
2154
2155                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2156                                            IFF_SLAVE_INACTIVE);
2157
2158                 kfree(slave);
2159
2160                 /* re-acquire the lock before getting the next slave */
2161                 write_lock_bh(&bond->lock);
2162         }
2163
2164         /* zero the mac address of the master so it will be
2165          * set by the application to the mac address of the
2166          * first slave
2167          */
2168         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2169
2170         if (!bond->vlgrp) {
2171                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2172         } else {
2173                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2174                            bond_dev->name, bond_dev->name);
2175                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2176                            bond_dev->name);
2177         }
2178
2179         pr_info("%s: released all slaves\n", bond_dev->name);
2180
2181 out:
2182         write_unlock_bh(&bond->lock);
2183
2184         return 0;
2185 }
2186
2187 /*
2188  * This function changes the active slave to slave <slave_dev>.
2189  * It returns -EINVAL in the following cases.
2190  *  - <slave_dev> is not found in the list.
2191  *  - There is not active slave now.
2192  *  - <slave_dev> is already active.
2193  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2194  *  - <slave_dev> is not running.
2195  * In these cases, this function does nothing.
2196  * In the other cases, current_slave pointer is changed and 0 is returned.
2197  */
2198 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2199 {
2200         struct bonding *bond = netdev_priv(bond_dev);
2201         struct slave *old_active = NULL;
2202         struct slave *new_active = NULL;
2203         int res = 0;
2204
2205         if (!USES_PRIMARY(bond->params.mode))
2206                 return -EINVAL;
2207
2208         /* Verify that master_dev is indeed the master of slave_dev */
2209         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2210                 return -EINVAL;
2211
2212         read_lock(&bond->lock);
2213
2214         read_lock(&bond->curr_slave_lock);
2215         old_active = bond->curr_active_slave;
2216         read_unlock(&bond->curr_slave_lock);
2217
2218         new_active = bond_get_slave_by_dev(bond, slave_dev);
2219
2220         /*
2221          * Changing to the current active: do nothing; return success.
2222          */
2223         if (new_active && (new_active == old_active)) {
2224                 read_unlock(&bond->lock);
2225                 return 0;
2226         }
2227
2228         if ((new_active) &&
2229             (old_active) &&
2230             (new_active->link == BOND_LINK_UP) &&
2231             IS_UP(new_active->dev)) {
2232                 write_lock_bh(&bond->curr_slave_lock);
2233                 bond_change_active_slave(bond, new_active);
2234                 write_unlock_bh(&bond->curr_slave_lock);
2235         } else
2236                 res = -EINVAL;
2237
2238         read_unlock(&bond->lock);
2239
2240         return res;
2241 }
2242
2243 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2244 {
2245         struct bonding *bond = netdev_priv(bond_dev);
2246
2247         info->bond_mode = bond->params.mode;
2248         info->miimon = bond->params.miimon;
2249
2250         read_lock(&bond->lock);
2251         info->num_slaves = bond->slave_cnt;
2252         read_unlock(&bond->lock);
2253
2254         return 0;
2255 }
2256
2257 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2258 {
2259         struct bonding *bond = netdev_priv(bond_dev);
2260         struct slave *slave;
2261         int i, res = -ENODEV;
2262
2263         read_lock(&bond->lock);
2264
2265         bond_for_each_slave(bond, slave, i) {
2266                 if (i == (int)info->slave_id) {
2267                         res = 0;
2268                         strcpy(info->slave_name, slave->dev->name);
2269                         info->link = slave->link;
2270                         info->state = slave->state;
2271                         info->link_failure_count = slave->link_failure_count;
2272                         break;
2273                 }
2274         }
2275
2276         read_unlock(&bond->lock);
2277
2278         return res;
2279 }
2280
2281 /*-------------------------------- Monitoring -------------------------------*/
2282
2283
2284 static int bond_miimon_inspect(struct bonding *bond)
2285 {
2286         struct slave *slave;
2287         int i, link_state, commit = 0;
2288         bool ignore_updelay;
2289
2290         ignore_updelay = !bond->curr_active_slave ? true : false;
2291
2292         bond_for_each_slave(bond, slave, i) {
2293                 slave->new_link = BOND_LINK_NOCHANGE;
2294
2295                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2296
2297                 switch (slave->link) {
2298                 case BOND_LINK_UP:
2299                         if (link_state)
2300                                 continue;
2301
2302                         slave->link = BOND_LINK_FAIL;
2303                         slave->delay = bond->params.downdelay;
2304                         if (slave->delay) {
2305                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2306                                         bond->dev->name,
2307                                         (bond->params.mode ==
2308                                          BOND_MODE_ACTIVEBACKUP) ?
2309                                         ((slave->state == BOND_STATE_ACTIVE) ?
2310                                          "active " : "backup ") : "",
2311                                         slave->dev->name,
2312                                         bond->params.downdelay * bond->params.miimon);
2313                         }
2314                         /*FALLTHRU*/
2315                 case BOND_LINK_FAIL:
2316                         if (link_state) {
2317                                 /*
2318                                  * recovered before downdelay expired
2319                                  */
2320                                 slave->link = BOND_LINK_UP;
2321                                 slave->jiffies = jiffies;
2322                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2323                                         bond->dev->name,
2324                                         (bond->params.downdelay - slave->delay) *
2325                                         bond->params.miimon,
2326                                         slave->dev->name);
2327                                 continue;
2328                         }
2329
2330                         if (slave->delay <= 0) {
2331                                 slave->new_link = BOND_LINK_DOWN;
2332                                 commit++;
2333                                 continue;
2334                         }
2335
2336                         slave->delay--;
2337                         break;
2338
2339                 case BOND_LINK_DOWN:
2340                         if (!link_state)
2341                                 continue;
2342
2343                         slave->link = BOND_LINK_BACK;
2344                         slave->delay = bond->params.updelay;
2345
2346                         if (slave->delay) {
2347                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2348                                         bond->dev->name, slave->dev->name,
2349                                         ignore_updelay ? 0 :
2350                                         bond->params.updelay *
2351                                         bond->params.miimon);
2352                         }
2353                         /*FALLTHRU*/
2354                 case BOND_LINK_BACK:
2355                         if (!link_state) {
2356                                 slave->link = BOND_LINK_DOWN;
2357                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2358                                         bond->dev->name,
2359                                         (bond->params.updelay - slave->delay) *
2360                                         bond->params.miimon,
2361                                         slave->dev->name);
2362
2363                                 continue;
2364                         }
2365
2366                         if (ignore_updelay)
2367                                 slave->delay = 0;
2368
2369                         if (slave->delay <= 0) {
2370                                 slave->new_link = BOND_LINK_UP;
2371                                 commit++;
2372                                 ignore_updelay = false;
2373                                 continue;
2374                         }
2375
2376                         slave->delay--;
2377                         break;
2378                 }
2379         }
2380
2381         return commit;
2382 }
2383
2384 static void bond_miimon_commit(struct bonding *bond)
2385 {
2386         struct slave *slave;
2387         int i;
2388
2389         bond_for_each_slave(bond, slave, i) {
2390                 switch (slave->new_link) {
2391                 case BOND_LINK_NOCHANGE:
2392                         continue;
2393
2394                 case BOND_LINK_UP:
2395                         slave->link = BOND_LINK_UP;
2396                         slave->jiffies = jiffies;
2397
2398                         if (bond->params.mode == BOND_MODE_8023AD) {
2399                                 /* prevent it from being the active one */
2400                                 slave->state = BOND_STATE_BACKUP;
2401                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2402                                 /* make it immediately active */
2403                                 slave->state = BOND_STATE_ACTIVE;
2404                         } else if (slave != bond->primary_slave) {
2405                                 /* prevent it from being the active one */
2406                                 slave->state = BOND_STATE_BACKUP;
2407                         }
2408
2409                         pr_info("%s: link status definitely up for interface %s.\n",
2410                                 bond->dev->name, slave->dev->name);
2411
2412                         /* notify ad that the link status has changed */
2413                         if (bond->params.mode == BOND_MODE_8023AD)
2414                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2415
2416                         if (bond_is_lb(bond))
2417                                 bond_alb_handle_link_change(bond, slave,
2418                                                             BOND_LINK_UP);
2419
2420                         if (!bond->curr_active_slave ||
2421                             (slave == bond->primary_slave))
2422                                 goto do_failover;
2423
2424                         continue;
2425
2426                 case BOND_LINK_DOWN:
2427                         if (slave->link_failure_count < UINT_MAX)
2428                                 slave->link_failure_count++;
2429
2430                         slave->link = BOND_LINK_DOWN;
2431
2432                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2433                             bond->params.mode == BOND_MODE_8023AD)
2434                                 bond_set_slave_inactive_flags(slave);
2435
2436                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2437                                 bond->dev->name, slave->dev->name);
2438
2439                         if (bond->params.mode == BOND_MODE_8023AD)
2440                                 bond_3ad_handle_link_change(slave,
2441                                                             BOND_LINK_DOWN);
2442
2443                         if (bond_is_lb(bond))
2444                                 bond_alb_handle_link_change(bond, slave,
2445                                                             BOND_LINK_DOWN);
2446
2447                         if (slave == bond->curr_active_slave)
2448                                 goto do_failover;
2449
2450                         continue;
2451
2452                 default:
2453                         pr_err("%s: invalid new link %d on slave %s\n",
2454                                bond->dev->name, slave->new_link,
2455                                slave->dev->name);
2456                         slave->new_link = BOND_LINK_NOCHANGE;
2457
2458                         continue;
2459                 }
2460
2461 do_failover:
2462                 ASSERT_RTNL();
2463                 write_lock_bh(&bond->curr_slave_lock);
2464                 bond_select_active_slave(bond);
2465                 write_unlock_bh(&bond->curr_slave_lock);
2466         }
2467
2468         bond_set_carrier(bond);
2469 }
2470
2471 /*
2472  * bond_mii_monitor
2473  *
2474  * Really a wrapper that splits the mii monitor into two phases: an
2475  * inspection, then (if inspection indicates something needs to be done)
2476  * an acquisition of appropriate locks followed by a commit phase to
2477  * implement whatever link state changes are indicated.
2478  */
2479 void bond_mii_monitor(struct work_struct *work)
2480 {
2481         struct bonding *bond = container_of(work, struct bonding,
2482                                             mii_work.work);
2483
2484         read_lock(&bond->lock);
2485         if (bond->kill_timers)
2486                 goto out;
2487
2488         if (bond->slave_cnt == 0)
2489                 goto re_arm;
2490
2491         if (bond->send_grat_arp) {
2492                 read_lock(&bond->curr_slave_lock);
2493                 bond_send_gratuitous_arp(bond);
2494                 read_unlock(&bond->curr_slave_lock);
2495         }
2496
2497         if (bond->send_unsol_na) {
2498                 read_lock(&bond->curr_slave_lock);
2499                 bond_send_unsolicited_na(bond);
2500                 read_unlock(&bond->curr_slave_lock);
2501         }
2502
2503         if (bond_miimon_inspect(bond)) {
2504                 read_unlock(&bond->lock);
2505                 rtnl_lock();
2506                 read_lock(&bond->lock);
2507
2508                 bond_miimon_commit(bond);
2509
2510                 read_unlock(&bond->lock);
2511                 rtnl_unlock();  /* might sleep, hold no other locks */
2512                 read_lock(&bond->lock);
2513         }
2514
2515 re_arm:
2516         if (bond->params.miimon)
2517                 queue_delayed_work(bond->wq, &bond->mii_work,
2518                                    msecs_to_jiffies(bond->params.miimon));
2519 out:
2520         read_unlock(&bond->lock);
2521 }
2522
2523 static __be32 bond_glean_dev_ip(struct net_device *dev)
2524 {
2525         struct in_device *idev;
2526         struct in_ifaddr *ifa;
2527         __be32 addr = 0;
2528
2529         if (!dev)
2530                 return 0;
2531
2532         rcu_read_lock();
2533         idev = __in_dev_get_rcu(dev);
2534         if (!idev)
2535                 goto out;
2536
2537         ifa = idev->ifa_list;
2538         if (!ifa)
2539                 goto out;
2540
2541         addr = ifa->ifa_local;
2542 out:
2543         rcu_read_unlock();
2544         return addr;
2545 }
2546
2547 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2548 {
2549         struct vlan_entry *vlan;
2550
2551         if (ip == bond->master_ip)
2552                 return 1;
2553
2554         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2555                 if (ip == vlan->vlan_ip)
2556                         return 1;
2557         }
2558
2559         return 0;
2560 }
2561
2562 /*
2563  * We go to the (large) trouble of VLAN tagging ARP frames because
2564  * switches in VLAN mode (especially if ports are configured as
2565  * "native" to a VLAN) might not pass non-tagged frames.
2566  */
2567 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2568 {
2569         struct sk_buff *skb;
2570
2571         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2572                  slave_dev->name, dest_ip, src_ip, vlan_id);
2573
2574         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2575                          NULL, slave_dev->dev_addr, NULL);
2576
2577         if (!skb) {
2578                 pr_err("ARP packet allocation failed\n");
2579                 return;
2580         }
2581         if (vlan_id) {
2582                 skb = vlan_put_tag(skb, vlan_id);
2583                 if (!skb) {
2584                         pr_err("failed to insert VLAN tag\n");
2585                         return;
2586                 }
2587         }
2588         arp_xmit(skb);
2589 }
2590
2591
2592 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2593 {
2594         int i, vlan_id, rv;
2595         __be32 *targets = bond->params.arp_targets;
2596         struct vlan_entry *vlan;
2597         struct net_device *vlan_dev;
2598         struct flowi fl;
2599         struct rtable *rt;
2600
2601         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2602                 if (!targets[i])
2603                         break;
2604                 pr_debug("basa: target %x\n", targets[i]);
2605                 if (!bond->vlgrp) {
2606                         pr_debug("basa: empty vlan: arp_send\n");
2607                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2608                                       bond->master_ip, 0);
2609                         continue;
2610                 }
2611
2612                 /*
2613                  * If VLANs are configured, we do a route lookup to
2614                  * determine which VLAN interface would be used, so we
2615                  * can tag the ARP with the proper VLAN tag.
2616                  */
2617                 memset(&fl, 0, sizeof(fl));
2618                 fl.fl4_dst = targets[i];
2619                 fl.fl4_tos = RTO_ONLINK;
2620
2621                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2622                 if (rv) {
2623                         if (net_ratelimit()) {
2624                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2625                                            bond->dev->name, &fl.fl4_dst);
2626                         }
2627                         continue;
2628                 }
2629
2630                 /*
2631                  * This target is not on a VLAN
2632                  */
2633                 if (rt->dst.dev == bond->dev) {
2634                         ip_rt_put(rt);
2635                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2636                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2637                                       bond->master_ip, 0);
2638                         continue;
2639                 }
2640
2641                 vlan_id = 0;
2642                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2643                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2644                         if (vlan_dev == rt->dst.dev) {
2645                                 vlan_id = vlan->vlan_id;
2646                                 pr_debug("basa: vlan match on %s %d\n",
2647                                        vlan_dev->name, vlan_id);
2648                                 break;
2649                         }
2650                 }
2651
2652                 if (vlan_id) {
2653                         ip_rt_put(rt);
2654                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2655                                       vlan->vlan_ip, vlan_id);
2656                         continue;
2657                 }
2658
2659                 if (net_ratelimit()) {
2660                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2661                                    bond->dev->name, &fl.fl4_dst,
2662                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2663                 }
2664                 ip_rt_put(rt);
2665         }
2666 }
2667
2668 /*
2669  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2670  * for each VLAN above us.
2671  *
2672  * Caller must hold curr_slave_lock for read or better
2673  */
2674 static void bond_send_gratuitous_arp(struct bonding *bond)
2675 {
2676         struct slave *slave = bond->curr_active_slave;
2677         struct vlan_entry *vlan;
2678         struct net_device *vlan_dev;
2679
2680         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2681                  bond->dev->name, slave ? slave->dev->name : "NULL");
2682
2683         if (!slave || !bond->send_grat_arp ||
2684             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2685                 return;
2686
2687         bond->send_grat_arp--;
2688
2689         if (bond->master_ip) {
2690                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2691                                 bond->master_ip, 0);
2692         }
2693
2694         if (!bond->vlgrp)
2695                 return;
2696
2697         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2698                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2699                 if (vlan->vlan_ip) {
2700                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2701                                       vlan->vlan_ip, vlan->vlan_id);
2702                 }
2703         }
2704 }
2705
2706 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2707 {
2708         int i;
2709         __be32 *targets = bond->params.arp_targets;
2710
2711         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2712                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2713                          &sip, &tip, i, &targets[i],
2714                          bond_has_this_ip(bond, tip));
2715                 if (sip == targets[i]) {
2716                         if (bond_has_this_ip(bond, tip))
2717                                 slave->last_arp_rx = jiffies;
2718                         return;
2719                 }
2720         }
2721 }
2722
2723 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2724 {
2725         struct arphdr *arp;
2726         struct slave *slave;
2727         struct bonding *bond;
2728         unsigned char *arp_ptr;
2729         __be32 sip, tip;
2730
2731         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2732                 /*
2733                  * When using VLANS and bonding, dev and oriv_dev may be
2734                  * incorrect if the physical interface supports VLAN
2735                  * acceleration.  With this change ARP validation now
2736                  * works for hosts only reachable on the VLAN interface.
2737                  */
2738                 dev = vlan_dev_real_dev(dev);
2739                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2740         }
2741
2742         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2743                 goto out;
2744
2745         bond = netdev_priv(dev);
2746         read_lock(&bond->lock);
2747
2748         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2749                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2750                  orig_dev ? orig_dev->name : "NULL");
2751
2752         slave = bond_get_slave_by_dev(bond, orig_dev);
2753         if (!slave || !slave_do_arp_validate(bond, slave))
2754                 goto out_unlock;
2755
2756         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2757                 goto out_unlock;
2758
2759         arp = arp_hdr(skb);
2760         if (arp->ar_hln != dev->addr_len ||
2761             skb->pkt_type == PACKET_OTHERHOST ||
2762             skb->pkt_type == PACKET_LOOPBACK ||
2763             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2764             arp->ar_pro != htons(ETH_P_IP) ||
2765             arp->ar_pln != 4)
2766                 goto out_unlock;
2767
2768         arp_ptr = (unsigned char *)(arp + 1);
2769         arp_ptr += dev->addr_len;
2770         memcpy(&sip, arp_ptr, 4);
2771         arp_ptr += 4 + dev->addr_len;
2772         memcpy(&tip, arp_ptr, 4);
2773
2774         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2775                  bond->dev->name, slave->dev->name, slave->state,
2776                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2777                  &sip, &tip);
2778
2779         /*
2780          * Backup slaves won't see the ARP reply, but do come through
2781          * here for each ARP probe (so we swap the sip/tip to validate
2782          * the probe).  In a "redundant switch, common router" type of
2783          * configuration, the ARP probe will (hopefully) travel from
2784          * the active, through one switch, the router, then the other
2785          * switch before reaching the backup.
2786          */
2787         if (slave->state == BOND_STATE_ACTIVE)
2788                 bond_validate_arp(bond, slave, sip, tip);
2789         else
2790                 bond_validate_arp(bond, slave, tip, sip);
2791
2792 out_unlock:
2793         read_unlock(&bond->lock);
2794 out:
2795         dev_kfree_skb(skb);
2796         return NET_RX_SUCCESS;
2797 }
2798
2799 /*
2800  * this function is called regularly to monitor each slave's link
2801  * ensuring that traffic is being sent and received when arp monitoring
2802  * is used in load-balancing mode. if the adapter has been dormant, then an
2803  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2804  * arp monitoring in active backup mode.
2805  */
2806 void bond_loadbalance_arp_mon(struct work_struct *work)
2807 {
2808         struct bonding *bond = container_of(work, struct bonding,
2809                                             arp_work.work);
2810         struct slave *slave, *oldcurrent;
2811         int do_failover = 0;
2812         int delta_in_ticks;
2813         int i;
2814
2815         read_lock(&bond->lock);
2816
2817         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2818
2819         if (bond->kill_timers)
2820                 goto out;
2821
2822         if (bond->slave_cnt == 0)
2823                 goto re_arm;
2824
2825         read_lock(&bond->curr_slave_lock);
2826         oldcurrent = bond->curr_active_slave;
2827         read_unlock(&bond->curr_slave_lock);
2828
2829         /* see if any of the previous devices are up now (i.e. they have
2830          * xmt and rcv traffic). the curr_active_slave does not come into
2831          * the picture unless it is null. also, slave->jiffies is not needed
2832          * here because we send an arp on each slave and give a slave as
2833          * long as it needs to get the tx/rx within the delta.
2834          * TODO: what about up/down delay in arp mode? it wasn't here before
2835          *       so it can wait
2836          */
2837         bond_for_each_slave(bond, slave, i) {
2838                 unsigned long trans_start = dev_trans_start(slave->dev);
2839
2840                 if (slave->link != BOND_LINK_UP) {
2841                         if (time_in_range(jiffies,
2842                                 trans_start - delta_in_ticks,
2843                                 trans_start + delta_in_ticks) &&
2844                             time_in_range(jiffies,
2845                                 slave->dev->last_rx - delta_in_ticks,
2846                                 slave->dev->last_rx + delta_in_ticks)) {
2847
2848                                 slave->link  = BOND_LINK_UP;
2849                                 slave->state = BOND_STATE_ACTIVE;
2850
2851                                 /* primary_slave has no meaning in round-robin
2852                                  * mode. the window of a slave being up and
2853                                  * curr_active_slave being null after enslaving
2854                                  * is closed.
2855                                  */
2856                                 if (!oldcurrent) {
2857                                         pr_info("%s: link status definitely up for interface %s, ",
2858                                                 bond->dev->name,
2859                                                 slave->dev->name);
2860                                         do_failover = 1;
2861                                 } else {
2862                                         pr_info("%s: interface %s is now up\n",
2863                                                 bond->dev->name,
2864                                                 slave->dev->name);
2865                                 }
2866                         }
2867                 } else {
2868                         /* slave->link == BOND_LINK_UP */
2869
2870                         /* not all switches will respond to an arp request
2871                          * when the source ip is 0, so don't take the link down
2872                          * if we don't know our ip yet
2873                          */
2874                         if (!time_in_range(jiffies,
2875                                 trans_start - delta_in_ticks,
2876                                 trans_start + 2 * delta_in_ticks) ||
2877                             !time_in_range(jiffies,
2878                                 slave->dev->last_rx - delta_in_ticks,
2879                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2880
2881                                 slave->link  = BOND_LINK_DOWN;
2882                                 slave->state = BOND_STATE_BACKUP;
2883
2884                                 if (slave->link_failure_count < UINT_MAX)
2885                                         slave->link_failure_count++;
2886
2887                                 pr_info("%s: interface %s is now down.\n",
2888                                         bond->dev->name,
2889                                         slave->dev->name);
2890
2891                                 if (slave == oldcurrent)
2892                                         do_failover = 1;
2893                         }
2894                 }
2895
2896                 /* note: if switch is in round-robin mode, all links
2897                  * must tx arp to ensure all links rx an arp - otherwise
2898                  * links may oscillate or not come up at all; if switch is
2899                  * in something like xor mode, there is nothing we can
2900                  * do - all replies will be rx'ed on same link causing slaves
2901                  * to be unstable during low/no traffic periods
2902                  */
2903                 if (IS_UP(slave->dev))
2904                         bond_arp_send_all(bond, slave);
2905         }
2906
2907         if (do_failover) {
2908                 write_lock_bh(&bond->curr_slave_lock);
2909
2910                 bond_select_active_slave(bond);
2911
2912                 write_unlock_bh(&bond->curr_slave_lock);
2913         }
2914
2915 re_arm:
2916         if (bond->params.arp_interval)
2917                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2918 out:
2919         read_unlock(&bond->lock);
2920 }
2921
2922 /*
2923  * Called to inspect slaves for active-backup mode ARP monitor link state
2924  * changes.  Sets new_link in slaves to specify what action should take
2925  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2926  * to link states must be committed.
2927  *
2928  * Called with bond->lock held for read.
2929  */
2930 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2931 {
2932         struct slave *slave;
2933         int i, commit = 0;
2934         unsigned long trans_start;
2935
2936         bond_for_each_slave(bond, slave, i) {
2937                 slave->new_link = BOND_LINK_NOCHANGE;
2938
2939                 if (slave->link != BOND_LINK_UP) {
2940                         if (time_in_range(jiffies,
2941                                 slave_last_rx(bond, slave) - delta_in_ticks,
2942                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2943
2944                                 slave->new_link = BOND_LINK_UP;
2945                                 commit++;
2946                         }
2947
2948                         continue;
2949                 }
2950
2951                 /*
2952                  * Give slaves 2*delta after being enslaved or made
2953                  * active.  This avoids bouncing, as the last receive
2954                  * times need a full ARP monitor cycle to be updated.
2955                  */
2956                 if (time_in_range(jiffies,
2957                                   slave->jiffies - delta_in_ticks,
2958                                   slave->jiffies + 2 * delta_in_ticks))
2959                         continue;
2960
2961                 /*
2962                  * Backup slave is down if:
2963                  * - No current_arp_slave AND
2964                  * - more than 3*delta since last receive AND
2965                  * - the bond has an IP address
2966                  *
2967                  * Note: a non-null current_arp_slave indicates
2968                  * the curr_active_slave went down and we are
2969                  * searching for a new one; under this condition
2970                  * we only take the curr_active_slave down - this
2971                  * gives each slave a chance to tx/rx traffic
2972                  * before being taken out
2973                  */
2974                 if (slave->state == BOND_STATE_BACKUP &&
2975                     !bond->current_arp_slave &&
2976                     !time_in_range(jiffies,
2977                         slave_last_rx(bond, slave) - delta_in_ticks,
2978                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2979
2980                         slave->new_link = BOND_LINK_DOWN;
2981                         commit++;
2982                 }
2983
2984                 /*
2985                  * Active slave is down if:
2986                  * - more than 2*delta since transmitting OR
2987                  * - (more than 2*delta since receive AND
2988                  *    the bond has an IP address)
2989                  */
2990                 trans_start = dev_trans_start(slave->dev);
2991                 if ((slave->state == BOND_STATE_ACTIVE) &&
2992                     (!time_in_range(jiffies,
2993                         trans_start - delta_in_ticks,
2994                         trans_start + 2 * delta_in_ticks) ||
2995                      !time_in_range(jiffies,
2996                         slave_last_rx(bond, slave) - delta_in_ticks,
2997                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2998
2999                         slave->new_link = BOND_LINK_DOWN;
3000                         commit++;
3001                 }
3002         }
3003
3004         return commit;
3005 }
3006
3007 /*
3008  * Called to commit link state changes noted by inspection step of
3009  * active-backup mode ARP monitor.
3010  *
3011  * Called with RTNL and bond->lock for read.
3012  */
3013 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3014 {
3015         struct slave *slave;
3016         int i;
3017         unsigned long trans_start;
3018
3019         bond_for_each_slave(bond, slave, i) {
3020                 switch (slave->new_link) {
3021                 case BOND_LINK_NOCHANGE:
3022                         continue;
3023
3024                 case BOND_LINK_UP:
3025                         trans_start = dev_trans_start(slave->dev);
3026                         if ((!bond->curr_active_slave &&
3027                              time_in_range(jiffies,
3028                                            trans_start - delta_in_ticks,
3029                                            trans_start + delta_in_ticks)) ||
3030                             bond->curr_active_slave != slave) {
3031                                 slave->link = BOND_LINK_UP;
3032                                 bond->current_arp_slave = NULL;
3033
3034                                 pr_info("%s: link status definitely up for interface %s.\n",
3035                                         bond->dev->name, slave->dev->name);
3036
3037                                 if (!bond->curr_active_slave ||
3038                                     (slave == bond->primary_slave))
3039                                         goto do_failover;
3040
3041                         }
3042
3043                         continue;
3044
3045                 case BOND_LINK_DOWN:
3046                         if (slave->link_failure_count < UINT_MAX)
3047                                 slave->link_failure_count++;
3048
3049                         slave->link = BOND_LINK_DOWN;
3050                         bond_set_slave_inactive_flags(slave);
3051
3052                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3053                                 bond->dev->name, slave->dev->name);
3054
3055                         if (slave == bond->curr_active_slave) {
3056                                 bond->current_arp_slave = NULL;
3057                                 goto do_failover;
3058                         }
3059
3060                         continue;
3061
3062                 default:
3063                         pr_err("%s: impossible: new_link %d on slave %s\n",
3064                                bond->dev->name, slave->new_link,
3065                                slave->dev->name);
3066                         continue;
3067                 }
3068
3069 do_failover:
3070                 ASSERT_RTNL();
3071                 write_lock_bh(&bond->curr_slave_lock);
3072                 bond_select_active_slave(bond);
3073                 write_unlock_bh(&bond->curr_slave_lock);
3074         }
3075
3076         bond_set_carrier(bond);
3077 }
3078
3079 /*
3080  * Send ARP probes for active-backup mode ARP monitor.
3081  *
3082  * Called with bond->lock held for read.
3083  */
3084 static void bond_ab_arp_probe(struct bonding *bond)
3085 {
3086         struct slave *slave;
3087         int i;
3088
3089         read_lock(&bond->curr_slave_lock);
3090
3091         if (bond->current_arp_slave && bond->curr_active_slave)
3092                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3093                         bond->current_arp_slave->dev->name,
3094                         bond->curr_active_slave->dev->name);
3095
3096         if (bond->curr_active_slave) {
3097                 bond_arp_send_all(bond, bond->curr_active_slave);
3098                 read_unlock(&bond->curr_slave_lock);
3099                 return;
3100         }
3101
3102         read_unlock(&bond->curr_slave_lock);
3103
3104         /* if we don't have a curr_active_slave, search for the next available
3105          * backup slave from the current_arp_slave and make it the candidate
3106          * for becoming the curr_active_slave
3107          */
3108
3109         if (!bond->current_arp_slave) {
3110                 bond->current_arp_slave = bond->first_slave;
3111                 if (!bond->current_arp_slave)
3112                         return;
3113         }
3114
3115         bond_set_slave_inactive_flags(bond->current_arp_slave);
3116
3117         /* search for next candidate */
3118         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3119                 if (IS_UP(slave->dev)) {
3120                         slave->link = BOND_LINK_BACK;
3121                         bond_set_slave_active_flags(slave);
3122                         bond_arp_send_all(bond, slave);
3123                         slave->jiffies = jiffies;
3124                         bond->current_arp_slave = slave;
3125                         break;
3126                 }
3127
3128                 /* if the link state is up at this point, we
3129                  * mark it down - this can happen if we have
3130                  * simultaneous link failures and
3131                  * reselect_active_interface doesn't make this
3132                  * one the current slave so it is still marked
3133                  * up when it is actually down
3134                  */
3135                 if (slave->link == BOND_LINK_UP) {
3136                         slave->link = BOND_LINK_DOWN;
3137                         if (slave->link_failure_count < UINT_MAX)
3138                                 slave->link_failure_count++;
3139
3140                         bond_set_slave_inactive_flags(slave);
3141
3142                         pr_info("%s: backup interface %s is now down.\n",
3143                                 bond->dev->name, slave->dev->name);
3144                 }
3145         }
3146 }
3147
3148 void bond_activebackup_arp_mon(struct work_struct *work)
3149 {
3150         struct bonding *bond = container_of(work, struct bonding,
3151                                             arp_work.work);
3152         int delta_in_ticks;
3153
3154         read_lock(&bond->lock);
3155
3156         if (bond->kill_timers)
3157                 goto out;
3158
3159         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3160
3161         if (bond->slave_cnt == 0)
3162                 goto re_arm;
3163
3164         if (bond->send_grat_arp) {
3165                 read_lock(&bond->curr_slave_lock);
3166                 bond_send_gratuitous_arp(bond);
3167                 read_unlock(&bond->curr_slave_lock);
3168         }
3169
3170         if (bond->send_unsol_na) {
3171                 read_lock(&bond->curr_slave_lock);
3172                 bond_send_unsolicited_na(bond);
3173                 read_unlock(&bond->curr_slave_lock);
3174         }
3175
3176         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3177                 read_unlock(&bond->lock);
3178                 rtnl_lock();
3179                 read_lock(&bond->lock);
3180
3181                 bond_ab_arp_commit(bond, delta_in_ticks);
3182
3183                 read_unlock(&bond->lock);
3184                 rtnl_unlock();
3185                 read_lock(&bond->lock);
3186         }
3187
3188         bond_ab_arp_probe(bond);
3189
3190 re_arm:
3191         if (bond->params.arp_interval)
3192                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3193 out:
3194         read_unlock(&bond->lock);
3195 }
3196
3197 /*------------------------------ proc/seq_file-------------------------------*/
3198
3199 #ifdef CONFIG_PROC_FS
3200
3201 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3202         __acquires(&dev_base_lock)
3203         __acquires(&bond->lock)
3204 {
3205         struct bonding *bond = seq->private;
3206         loff_t off = 0;
3207         struct slave *slave;
3208         int i;
3209
3210         /* make sure the bond won't be taken away */
3211         read_lock(&dev_base_lock);
3212         read_lock(&bond->lock);
3213
3214         if (*pos == 0)
3215                 return SEQ_START_TOKEN;
3216
3217         bond_for_each_slave(bond, slave, i) {
3218                 if (++off == *pos)
3219                         return slave;
3220         }
3221
3222         return NULL;
3223 }
3224
3225 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3226 {
3227         struct bonding *bond = seq->private;
3228         struct slave *slave = v;
3229
3230         ++*pos;
3231         if (v == SEQ_START_TOKEN)
3232                 return bond->first_slave;
3233
3234         slave = slave->next;
3235
3236         return (slave == bond->first_slave) ? NULL : slave;
3237 }
3238
3239 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3240         __releases(&bond->lock)
3241         __releases(&dev_base_lock)
3242 {
3243         struct bonding *bond = seq->private;
3244
3245         read_unlock(&bond->lock);
3246         read_unlock(&dev_base_lock);
3247 }
3248
3249 static void bond_info_show_master(struct seq_file *seq)
3250 {
3251         struct bonding *bond = seq->private;
3252         struct slave *curr;
3253         int i;
3254
3255         read_lock(&bond->curr_slave_lock);
3256         curr = bond->curr_active_slave;
3257         read_unlock(&bond->curr_slave_lock);
3258
3259         seq_printf(seq, "Bonding Mode: %s",
3260                    bond_mode_name(bond->params.mode));
3261
3262         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3263             bond->params.fail_over_mac)
3264                 seq_printf(seq, " (fail_over_mac %s)",
3265                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3266
3267         seq_printf(seq, "\n");
3268
3269         if (bond->params.mode == BOND_MODE_XOR ||
3270                 bond->params.mode == BOND_MODE_8023AD) {
3271                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3272                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3273                         bond->params.xmit_policy);
3274         }
3275
3276         if (USES_PRIMARY(bond->params.mode)) {
3277                 seq_printf(seq, "Primary Slave: %s",
3278                            (bond->primary_slave) ?
3279                            bond->primary_slave->dev->name : "None");
3280                 if (bond->primary_slave)
3281                         seq_printf(seq, " (primary_reselect %s)",
3282                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3283
3284                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3285                            (curr) ? curr->dev->name : "None");
3286         }
3287
3288         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3289                    "up" : "down");
3290         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3291         seq_printf(seq, "Up Delay (ms): %d\n",
3292                    bond->params.updelay * bond->params.miimon);
3293         seq_printf(seq, "Down Delay (ms): %d\n",
3294                    bond->params.downdelay * bond->params.miimon);
3295
3296
3297         /* ARP information */
3298         if (bond->params.arp_interval > 0) {
3299                 int printed = 0;
3300                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3301                                 bond->params.arp_interval);
3302
3303                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3304
3305                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3306                         if (!bond->params.arp_targets[i])
3307                                 break;
3308                         if (printed)
3309                                 seq_printf(seq, ",");
3310                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3311                         printed = 1;
3312                 }
3313                 seq_printf(seq, "\n");
3314         }
3315
3316         if (bond->params.mode == BOND_MODE_8023AD) {
3317                 struct ad_info ad_info;
3318
3319                 seq_puts(seq, "\n802.3ad info\n");
3320                 seq_printf(seq, "LACP rate: %s\n",
3321                            (bond->params.lacp_fast) ? "fast" : "slow");
3322                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3323                            ad_select_tbl[bond->params.ad_select].modename);
3324
3325                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3326                         seq_printf(seq, "bond %s has no active aggregator\n",
3327                                    bond->dev->name);
3328                 } else {
3329                         seq_printf(seq, "Active Aggregator Info:\n");
3330
3331                         seq_printf(seq, "\tAggregator ID: %d\n",
3332                                    ad_info.aggregator_id);
3333                         seq_printf(seq, "\tNumber of ports: %d\n",
3334                                    ad_info.ports);
3335                         seq_printf(seq, "\tActor Key: %d\n",
3336                                    ad_info.actor_key);
3337                         seq_printf(seq, "\tPartner Key: %d\n",
3338                                    ad_info.partner_key);
3339                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3340                                    ad_info.partner_system);
3341                 }
3342         }
3343 }
3344
3345 static void bond_info_show_slave(struct seq_file *seq,
3346                                  const struct slave *slave)
3347 {
3348         struct bonding *bond = seq->private;
3349
3350         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3351         seq_printf(seq, "MII Status: %s\n",
3352                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3353         seq_printf(seq, "Link Failure Count: %u\n",
3354                    slave->link_failure_count);
3355
3356         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3357
3358         if (bond->params.mode == BOND_MODE_8023AD) {
3359                 const struct aggregator *agg
3360                         = SLAVE_AD_INFO(slave).port.aggregator;
3361
3362                 if (agg)
3363                         seq_printf(seq, "Aggregator ID: %d\n",
3364                                    agg->aggregator_identifier);
3365                 else
3366                         seq_puts(seq, "Aggregator ID: N/A\n");
3367         }
3368         seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3369 }
3370
3371 static int bond_info_seq_show(struct seq_file *seq, void *v)
3372 {
3373         if (v == SEQ_START_TOKEN) {
3374                 seq_printf(seq, "%s\n", version);
3375                 bond_info_show_master(seq);
3376         } else
3377                 bond_info_show_slave(seq, v);
3378
3379         return 0;
3380 }
3381
3382 static const struct seq_operations bond_info_seq_ops = {
3383         .start = bond_info_seq_start,
3384         .next  = bond_info_seq_next,
3385         .stop  = bond_info_seq_stop,
3386         .show  = bond_info_seq_show,
3387 };
3388
3389 static int bond_info_open(struct inode *inode, struct file *file)
3390 {
3391         struct seq_file *seq;
3392         struct proc_dir_entry *proc;
3393         int res;
3394
3395         res = seq_open(file, &bond_info_seq_ops);
3396         if (!res) {
3397                 /* recover the pointer buried in proc_dir_entry data */
3398                 seq = file->private_data;
3399                 proc = PDE(inode);
3400                 seq->private = proc->data;
3401         }
3402
3403         return res;
3404 }
3405
3406 static const struct file_operations bond_info_fops = {
3407         .owner   = THIS_MODULE,
3408         .open    = bond_info_open,
3409         .read    = seq_read,
3410         .llseek  = seq_lseek,
3411         .release = seq_release,
3412 };
3413
3414 static void bond_create_proc_entry(struct bonding *bond)
3415 {
3416         struct net_device *bond_dev = bond->dev;
3417         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3418
3419         if (bn->proc_dir) {
3420                 bond->proc_entry = proc_create_data(bond_dev->name,
3421                                                     S_IRUGO, bn->proc_dir,
3422                                                     &bond_info_fops, bond);
3423                 if (bond->proc_entry == NULL)
3424                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3425                                    DRV_NAME, bond_dev->name);
3426                 else
3427                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3428         }
3429 }
3430
3431 static void bond_remove_proc_entry(struct bonding *bond)
3432 {
3433         struct net_device *bond_dev = bond->dev;
3434         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3435
3436         if (bn->proc_dir && bond->proc_entry) {
3437                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3438                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3439                 bond->proc_entry = NULL;
3440         }
3441 }
3442
3443 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3444  * Caller must hold rtnl_lock.
3445  */
3446 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3447 {
3448         if (!bn->proc_dir) {
3449                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3450                 if (!bn->proc_dir)
3451                         pr_warning("Warning: cannot create /proc/net/%s\n",
3452                                    DRV_NAME);
3453         }
3454 }
3455
3456 /* Destroy the bonding directory under /proc/net, if empty.
3457  * Caller must hold rtnl_lock.
3458  */
3459 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3460 {
3461         if (bn->proc_dir) {
3462                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3463                 bn->proc_dir = NULL;
3464         }
3465 }
3466
3467 #else /* !CONFIG_PROC_FS */
3468
3469 static void bond_create_proc_entry(struct bonding *bond)
3470 {
3471 }
3472
3473 static void bond_remove_proc_entry(struct bonding *bond)
3474 {
3475 }
3476
3477 static inline void bond_create_proc_dir(struct bond_net *bn)
3478 {
3479 }
3480
3481 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3482 {
3483 }
3484
3485 #endif /* CONFIG_PROC_FS */
3486
3487
3488 /*-------------------------- netdev event handling --------------------------*/
3489
3490 /*
3491  * Change device name
3492  */
3493 static int bond_event_changename(struct bonding *bond)
3494 {
3495         bond_remove_proc_entry(bond);
3496         bond_create_proc_entry(bond);
3497
3498         return NOTIFY_DONE;
3499 }
3500
3501 static int bond_master_netdev_event(unsigned long event,
3502                                     struct net_device *bond_dev)
3503 {
3504         struct bonding *event_bond = netdev_priv(bond_dev);
3505
3506         switch (event) {
3507         case NETDEV_CHANGENAME:
3508                 return bond_event_changename(event_bond);
3509         default:
3510                 break;
3511         }
3512
3513         return NOTIFY_DONE;
3514 }
3515
3516 static int bond_slave_netdev_event(unsigned long event,
3517                                    struct net_device *slave_dev)
3518 {
3519         struct net_device *bond_dev = slave_dev->master;
3520         struct bonding *bond = netdev_priv(bond_dev);
3521
3522         switch (event) {
3523         case NETDEV_UNREGISTER:
3524                 if (bond_dev) {
3525                         if (bond->setup_by_slave)
3526                                 bond_release_and_destroy(bond_dev, slave_dev);
3527                         else
3528                                 bond_release(bond_dev, slave_dev);
3529                 }
3530                 break;
3531         case NETDEV_CHANGE:
3532                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3533                         struct slave *slave;
3534
3535                         slave = bond_get_slave_by_dev(bond, slave_dev);
3536                         if (slave) {
3537                                 u16 old_speed = slave->speed;
3538                                 u16 old_duplex = slave->duplex;
3539
3540                                 bond_update_speed_duplex(slave);
3541
3542                                 if (bond_is_lb(bond))
3543                                         break;
3544
3545                                 if (old_speed != slave->speed)
3546                                         bond_3ad_adapter_speed_changed(slave);
3547                                 if (old_duplex != slave->duplex)
3548                                         bond_3ad_adapter_duplex_changed(slave);
3549                         }
3550                 }
3551
3552                 break;
3553         case NETDEV_DOWN:
3554                 /*
3555                  * ... Or is it this?
3556                  */
3557                 break;
3558         case NETDEV_CHANGEMTU:
3559                 /*
3560                  * TODO: Should slaves be allowed to
3561                  * independently alter their MTU?  For
3562                  * an active-backup bond, slaves need
3563                  * not be the same type of device, so
3564                  * MTUs may vary.  For other modes,
3565                  * slaves arguably should have the
3566                  * same MTUs. To do this, we'd need to
3567                  * take over the slave's change_mtu
3568                  * function for the duration of their
3569                  * servitude.
3570                  */
3571                 break;
3572         case NETDEV_CHANGENAME:
3573                 /*
3574                  * TODO: handle changing the primary's name
3575                  */
3576                 break;
3577         case NETDEV_FEAT_CHANGE:
3578                 bond_compute_features(bond);
3579                 break;
3580         default:
3581                 break;
3582         }
3583
3584         return NOTIFY_DONE;
3585 }
3586
3587 /*
3588  * bond_netdev_event: handle netdev notifier chain events.
3589  *
3590  * This function receives events for the netdev chain.  The caller (an
3591  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3592  * locks for us to safely manipulate the slave devices (RTNL lock,
3593  * dev_probe_lock).
3594  */
3595 static int bond_netdev_event(struct notifier_block *this,
3596                              unsigned long event, void *ptr)
3597 {
3598         struct net_device *event_dev = (struct net_device *)ptr;
3599
3600         pr_debug("event_dev: %s, event: %lx\n",
3601                  event_dev ? event_dev->name : "None",
3602                  event);
3603
3604         if (!(event_dev->priv_flags & IFF_BONDING))
3605                 return NOTIFY_DONE;
3606
3607         if (event_dev->flags & IFF_MASTER) {
3608                 pr_debug("IFF_MASTER\n");
3609                 return bond_master_netdev_event(event, event_dev);
3610         }
3611
3612         if (event_dev->flags & IFF_SLAVE) {
3613                 pr_debug("IFF_SLAVE\n");
3614                 return bond_slave_netdev_event(event, event_dev);
3615         }
3616
3617         return NOTIFY_DONE;
3618 }
3619
3620 /*
3621  * bond_inetaddr_event: handle inetaddr notifier chain events.
3622  *
3623  * We keep track of device IPs primarily to use as source addresses in
3624  * ARP monitor probes (rather than spewing out broadcasts all the time).
3625  *
3626  * We track one IP for the main device (if it has one), plus one per VLAN.
3627  */
3628 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3629 {
3630         struct in_ifaddr *ifa = ptr;
3631         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3632         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3633         struct bonding *bond;
3634         struct vlan_entry *vlan;
3635
3636         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3637                 if (bond->dev == event_dev) {
3638                         switch (event) {
3639                         case NETDEV_UP:
3640                                 bond->master_ip = ifa->ifa_local;
3641                                 return NOTIFY_OK;
3642                         case NETDEV_DOWN:
3643                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3644                                 return NOTIFY_OK;
3645                         default:
3646                                 return NOTIFY_DONE;
3647                         }
3648                 }
3649
3650                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3651                         if (!bond->vlgrp)
3652                                 continue;
3653                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3654                         if (vlan_dev == event_dev) {
3655                                 switch (event) {
3656                                 case NETDEV_UP:
3657                                         vlan->vlan_ip = ifa->ifa_local;
3658                                         return NOTIFY_OK;
3659                                 case NETDEV_DOWN:
3660                                         vlan->vlan_ip =
3661                                                 bond_glean_dev_ip(vlan_dev);
3662                                         return NOTIFY_OK;
3663                                 default:
3664                                         return NOTIFY_DONE;
3665                                 }
3666                         }
3667                 }
3668         }
3669         return NOTIFY_DONE;
3670 }
3671
3672 static struct notifier_block bond_netdev_notifier = {
3673         .notifier_call = bond_netdev_event,
3674 };
3675
3676 static struct notifier_block bond_inetaddr_notifier = {
3677         .notifier_call = bond_inetaddr_event,
3678 };
3679
3680 /*-------------------------- Packet type handling ---------------------------*/
3681
3682 /* register to receive lacpdus on a bond */
3683 static void bond_register_lacpdu(struct bonding *bond)
3684 {
3685         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3686
3687         /* initialize packet type */
3688         pk_type->type = PKT_TYPE_LACPDU;
3689         pk_type->dev = bond->dev;
3690         pk_type->func = bond_3ad_lacpdu_recv;
3691
3692         dev_add_pack(pk_type);
3693 }
3694
3695 /* unregister to receive lacpdus on a bond */
3696 static void bond_unregister_lacpdu(struct bonding *bond)
3697 {
3698         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3699 }
3700
3701 void bond_register_arp(struct bonding *bond)
3702 {
3703         struct packet_type *pt = &bond->arp_mon_pt;
3704
3705         if (pt->type)
3706                 return;
3707
3708         pt->type = htons(ETH_P_ARP);
3709         pt->dev = bond->dev;
3710         pt->func = bond_arp_rcv;
3711         dev_add_pack(pt);
3712 }
3713
3714 void bond_unregister_arp(struct bonding *bond)
3715 {
3716         struct packet_type *pt = &bond->arp_mon_pt;
3717
3718         dev_remove_pack(pt);
3719         pt->type = 0;
3720 }
3721
3722 /*---------------------------- Hashing Policies -----------------------------*/
3723
3724 /*
3725  * Hash for the output device based upon layer 2 and layer 3 data. If
3726  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3727  */
3728 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3729 {
3730         struct ethhdr *data = (struct ethhdr *)skb->data;
3731         struct iphdr *iph = ip_hdr(skb);
3732
3733         if (skb->protocol == htons(ETH_P_IP)) {
3734                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3735                         (data->h_dest[5] ^ data->h_source[5])) % count;
3736         }
3737
3738         return (data->h_dest[5] ^ data->h_source[5]) % count;
3739 }
3740
3741 /*
3742  * Hash for the output device based upon layer 3 and layer 4 data. If
3743  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3744  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3745  */
3746 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3747 {
3748         struct ethhdr *data = (struct ethhdr *)skb->data;
3749         struct iphdr *iph = ip_hdr(skb);
3750         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3751         int layer4_xor = 0;
3752
3753         if (skb->protocol == htons(ETH_P_IP)) {
3754                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3755                     (iph->protocol == IPPROTO_TCP ||
3756                      iph->protocol == IPPROTO_UDP)) {
3757                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3758                 }
3759                 return (layer4_xor ^
3760                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3761
3762         }
3763
3764         return (data->h_dest[5] ^ data->h_source[5]) % count;
3765 }
3766
3767 /*
3768  * Hash for the output device based upon layer 2 data
3769  */
3770 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3771 {
3772         struct ethhdr *data = (struct ethhdr *)skb->data;
3773
3774         return (data->h_dest[5] ^ data->h_source[5]) % count;
3775 }
3776
3777 /*-------------------------- Device entry points ----------------------------*/
3778
3779 static int bond_open(struct net_device *bond_dev)
3780 {
3781         struct bonding *bond = netdev_priv(bond_dev);
3782
3783         bond->kill_timers = 0;
3784
3785         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3786
3787         if (bond_is_lb(bond)) {
3788                 /* bond_alb_initialize must be called before the timer
3789                  * is started.
3790                  */
3791                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3792                         /* something went wrong - fail the open operation */
3793                         return -ENOMEM;
3794                 }
3795
3796                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3797                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3798         }
3799
3800         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3801                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3802                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3803         }
3804
3805         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3806                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3807                         INIT_DELAYED_WORK(&bond->arp_work,
3808                                           bond_activebackup_arp_mon);
3809                 else
3810                         INIT_DELAYED_WORK(&bond->arp_work,
3811                                           bond_loadbalance_arp_mon);
3812
3813                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3814                 if (bond->params.arp_validate)
3815                         bond_register_arp(bond);
3816         }
3817
3818         if (bond->params.mode == BOND_MODE_8023AD) {
3819                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3820                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3821                 /* register to receive LACPDUs */
3822                 bond_register_lacpdu(bond);
3823                 bond_3ad_initiate_agg_selection(bond, 1);
3824         }
3825
3826         return 0;
3827 }
3828
3829 static int bond_close(struct net_device *bond_dev)
3830 {
3831         struct bonding *bond = netdev_priv(bond_dev);
3832
3833         if (bond->params.mode == BOND_MODE_8023AD) {
3834                 /* Unregister the receive of LACPDUs */
3835                 bond_unregister_lacpdu(bond);
3836         }
3837
3838         if (bond->params.arp_validate)
3839                 bond_unregister_arp(bond);
3840
3841         write_lock_bh(&bond->lock);
3842
3843         bond->send_grat_arp = 0;
3844         bond->send_unsol_na = 0;
3845
3846         /* signal timers not to re-arm */
3847         bond->kill_timers = 1;
3848
3849         write_unlock_bh(&bond->lock);
3850
3851         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3852                 cancel_delayed_work(&bond->mii_work);
3853         }
3854
3855         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3856                 cancel_delayed_work(&bond->arp_work);
3857         }
3858
3859         switch (bond->params.mode) {
3860         case BOND_MODE_8023AD:
3861                 cancel_delayed_work(&bond->ad_work);
3862                 break;
3863         case BOND_MODE_TLB:
3864         case BOND_MODE_ALB:
3865                 cancel_delayed_work(&bond->alb_work);
3866                 break;
3867         default:
3868                 break;
3869         }
3870
3871         if (delayed_work_pending(&bond->mcast_work))
3872                 cancel_delayed_work(&bond->mcast_work);
3873
3874         if (bond_is_lb(bond)) {
3875                 /* Must be called only after all
3876                  * slaves have been released
3877                  */
3878                 bond_alb_deinitialize(bond);
3879         }
3880
3881         return 0;
3882 }
3883
3884 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3885                                                 struct rtnl_link_stats64 *stats)
3886 {
3887         struct bonding *bond = netdev_priv(bond_dev);
3888         struct rtnl_link_stats64 temp;
3889         struct slave *slave;
3890         int i;
3891
3892         memset(stats, 0, sizeof(*stats));
3893
3894         read_lock_bh(&bond->lock);
3895
3896         bond_for_each_slave(bond, slave, i) {
3897                 const struct rtnl_link_stats64 *sstats =
3898                         dev_get_stats(slave->dev, &temp);
3899
3900                 stats->rx_packets += sstats->rx_packets;
3901                 stats->rx_bytes += sstats->rx_bytes;
3902                 stats->rx_errors += sstats->rx_errors;
3903                 stats->rx_dropped += sstats->rx_dropped;
3904
3905                 stats->tx_packets += sstats->tx_packets;
3906                 stats->tx_bytes += sstats->tx_bytes;
3907                 stats->tx_errors += sstats->tx_errors;
3908                 stats->tx_dropped += sstats->tx_dropped;
3909
3910                 stats->multicast += sstats->multicast;
3911                 stats->collisions += sstats->collisions;
3912
3913                 stats->rx_length_errors += sstats->rx_length_errors;
3914                 stats->rx_over_errors += sstats->rx_over_errors;
3915                 stats->rx_crc_errors += sstats->rx_crc_errors;
3916                 stats->rx_frame_errors += sstats->rx_frame_errors;
3917                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3918                 stats->rx_missed_errors += sstats->rx_missed_errors;
3919
3920                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3921                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3922                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3923                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3924                 stats->tx_window_errors += sstats->tx_window_errors;
3925         }
3926
3927         read_unlock_bh(&bond->lock);
3928
3929         return stats;
3930 }
3931
3932 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3933 {
3934         struct net_device *slave_dev = NULL;
3935         struct ifbond k_binfo;
3936         struct ifbond __user *u_binfo = NULL;
3937         struct ifslave k_sinfo;
3938         struct ifslave __user *u_sinfo = NULL;
3939         struct mii_ioctl_data *mii = NULL;
3940         int res = 0;
3941
3942         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3943
3944         switch (cmd) {
3945         case SIOCGMIIPHY:
3946                 mii = if_mii(ifr);
3947                 if (!mii)
3948                         return -EINVAL;
3949
3950                 mii->phy_id = 0;
3951                 /* Fall Through */
3952         case SIOCGMIIREG:
3953                 /*
3954                  * We do this again just in case we were called by SIOCGMIIREG
3955                  * instead of SIOCGMIIPHY.
3956                  */
3957                 mii = if_mii(ifr);
3958                 if (!mii)
3959                         return -EINVAL;
3960
3961
3962                 if (mii->reg_num == 1) {
3963                         struct bonding *bond = netdev_priv(bond_dev);
3964                         mii->val_out = 0;
3965                         read_lock(&bond->lock);
3966                         read_lock(&bond->curr_slave_lock);
3967                         if (netif_carrier_ok(bond->dev))
3968                                 mii->val_out = BMSR_LSTATUS;
3969
3970                         read_unlock(&bond->curr_slave_lock);
3971                         read_unlock(&bond->lock);
3972                 }
3973
3974                 return 0;
3975         case BOND_INFO_QUERY_OLD:
3976         case SIOCBONDINFOQUERY:
3977                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3978
3979                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3980                         return -EFAULT;
3981
3982                 res = bond_info_query(bond_dev, &k_binfo);
3983                 if (res == 0 &&
3984                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3985                         return -EFAULT;
3986
3987                 return res;
3988         case BOND_SLAVE_INFO_QUERY_OLD:
3989         case SIOCBONDSLAVEINFOQUERY:
3990                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3991
3992                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3993                         return -EFAULT;
3994
3995                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3996                 if (res == 0 &&
3997                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3998                         return -EFAULT;
3999
4000                 return res;
4001         default:
4002                 /* Go on */
4003                 break;
4004         }
4005
4006         if (!capable(CAP_NET_ADMIN))
4007                 return -EPERM;
4008
4009         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4010
4011         pr_debug("slave_dev=%p:\n", slave_dev);
4012
4013         if (!slave_dev)
4014                 res = -ENODEV;
4015         else {
4016                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4017                 switch (cmd) {
4018                 case BOND_ENSLAVE_OLD:
4019                 case SIOCBONDENSLAVE:
4020                         res = bond_enslave(bond_dev, slave_dev);
4021                         break;
4022                 case BOND_RELEASE_OLD:
4023                 case SIOCBONDRELEASE:
4024                         res = bond_release(bond_dev, slave_dev);
4025                         break;
4026                 case BOND_SETHWADDR_OLD:
4027                 case SIOCBONDSETHWADDR:
4028                         res = bond_sethwaddr(bond_dev, slave_dev);
4029                         break;
4030                 case BOND_CHANGE_ACTIVE_OLD:
4031                 case SIOCBONDCHANGEACTIVE:
4032                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4033                         break;
4034                 default:
4035                         res = -EOPNOTSUPP;
4036                 }
4037
4038                 dev_put(slave_dev);
4039         }
4040
4041         return res;
4042 }
4043
4044 static bool bond_addr_in_mc_list(unsigned char *addr,
4045                                  struct netdev_hw_addr_list *list,
4046                                  int addrlen)
4047 {
4048         struct netdev_hw_addr *ha;
4049
4050         netdev_hw_addr_list_for_each(ha, list)
4051                 if (!memcmp(ha->addr, addr, addrlen))
4052                         return true;
4053
4054         return false;
4055 }
4056
4057 static void bond_set_multicast_list(struct net_device *bond_dev)
4058 {
4059         struct bonding *bond = netdev_priv(bond_dev);
4060         struct netdev_hw_addr *ha;
4061         bool found;
4062
4063         /*
4064          * Do promisc before checking multicast_mode
4065          */
4066         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4067                 /*
4068                  * FIXME: Need to handle the error when one of the multi-slaves
4069                  * encounters error.
4070                  */
4071                 bond_set_promiscuity(bond, 1);
4072
4073
4074         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4075                 bond_set_promiscuity(bond, -1);
4076
4077
4078         /* set allmulti flag to slaves */
4079         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4080                 /*
4081                  * FIXME: Need to handle the error when one of the multi-slaves
4082                  * encounters error.
4083                  */
4084                 bond_set_allmulti(bond, 1);
4085
4086
4087         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4088                 bond_set_allmulti(bond, -1);
4089
4090
4091         read_lock(&bond->lock);
4092
4093         bond->flags = bond_dev->flags;
4094
4095         /* looking for addresses to add to slaves' mc list */
4096         netdev_for_each_mc_addr(ha, bond_dev) {
4097                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4098                                              bond_dev->addr_len);
4099                 if (!found)
4100                         bond_mc_add(bond, ha->addr);
4101         }
4102
4103         /* looking for addresses to delete from slaves' list */
4104         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4105                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4106                                              bond_dev->addr_len);
4107                 if (!found)
4108                         bond_mc_del(bond, ha->addr);
4109         }
4110
4111         /* save master's multicast list */
4112         __hw_addr_flush(&bond->mc_list);
4113         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4114                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4115
4116         read_unlock(&bond->lock);
4117 }
4118
4119 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4120 {
4121         struct bonding *bond = netdev_priv(dev);
4122         struct slave *slave = bond->first_slave;
4123
4124         if (slave) {
4125                 const struct net_device_ops *slave_ops
4126                         = slave->dev->netdev_ops;
4127                 if (slave_ops->ndo_neigh_setup)
4128                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4129         }
4130         return 0;
4131 }
4132
4133 /*
4134  * Change the MTU of all of a master's slaves to match the master
4135  */
4136 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4137 {
4138         struct bonding *bond = netdev_priv(bond_dev);
4139         struct slave *slave, *stop_at;
4140         int res = 0;
4141         int i;
4142
4143         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4144                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4145
4146         /* Can't hold bond->lock with bh disabled here since
4147          * some base drivers panic. On the other hand we can't
4148          * hold bond->lock without bh disabled because we'll
4149          * deadlock. The only solution is to rely on the fact
4150          * that we're under rtnl_lock here, and the slaves
4151          * list won't change. This doesn't solve the problem
4152          * of setting the slave's MTU while it is
4153          * transmitting, but the assumption is that the base
4154          * driver can handle that.
4155          *
4156          * TODO: figure out a way to safely iterate the slaves
4157          * list, but without holding a lock around the actual
4158          * call to the base driver.
4159          */
4160
4161         bond_for_each_slave(bond, slave, i) {
4162                 pr_debug("s %p s->p %p c_m %p\n",
4163                          slave,
4164                          slave->prev,
4165                          slave->dev->netdev_ops->ndo_change_mtu);
4166
4167                 res = dev_set_mtu(slave->dev, new_mtu);
4168
4169                 if (res) {
4170                         /* If we failed to set the slave's mtu to the new value
4171                          * we must abort the operation even in ACTIVE_BACKUP
4172                          * mode, because if we allow the backup slaves to have
4173                          * different mtu values than the active slave we'll
4174                          * need to change their mtu when doing a failover. That
4175                          * means changing their mtu from timer context, which
4176                          * is probably not a good idea.
4177                          */
4178                         pr_debug("err %d %s\n", res, slave->dev->name);
4179                         goto unwind;
4180                 }
4181         }
4182
4183         bond_dev->mtu = new_mtu;
4184
4185         return 0;
4186
4187 unwind:
4188         /* unwind from head to the slave that failed */
4189         stop_at = slave;
4190         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4191                 int tmp_res;
4192
4193                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4194                 if (tmp_res) {
4195                         pr_debug("unwind err %d dev %s\n",
4196                                  tmp_res, slave->dev->name);
4197                 }
4198         }
4199
4200         return res;
4201 }
4202
4203 /*
4204  * Change HW address
4205  *
4206  * Note that many devices must be down to change the HW address, and
4207  * downing the master releases all slaves.  We can make bonds full of
4208  * bonding devices to test this, however.
4209  */
4210 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4211 {
4212         struct bonding *bond = netdev_priv(bond_dev);
4213         struct sockaddr *sa = addr, tmp_sa;
4214         struct slave *slave, *stop_at;
4215         int res = 0;
4216         int i;
4217
4218         if (bond->params.mode == BOND_MODE_ALB)
4219                 return bond_alb_set_mac_address(bond_dev, addr);
4220
4221
4222         pr_debug("bond=%p, name=%s\n",
4223                  bond, bond_dev ? bond_dev->name : "None");
4224
4225         /*
4226          * If fail_over_mac is set to active, do nothing and return
4227          * success.  Returning an error causes ifenslave to fail.
4228          */
4229         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4230                 return 0;
4231
4232         if (!is_valid_ether_addr(sa->sa_data))
4233                 return -EADDRNOTAVAIL;
4234
4235         /* Can't hold bond->lock with bh disabled here since
4236          * some base drivers panic. On the other hand we can't
4237          * hold bond->lock without bh disabled because we'll
4238          * deadlock. The only solution is to rely on the fact
4239          * that we're under rtnl_lock here, and the slaves
4240          * list won't change. This doesn't solve the problem
4241          * of setting the slave's hw address while it is
4242          * transmitting, but the assumption is that the base
4243          * driver can handle that.
4244          *
4245          * TODO: figure out a way to safely iterate the slaves
4246          * list, but without holding a lock around the actual
4247          * call to the base driver.
4248          */
4249
4250         bond_for_each_slave(bond, slave, i) {
4251                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4252                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4253
4254                 if (slave_ops->ndo_set_mac_address == NULL) {
4255                         res = -EOPNOTSUPP;
4256                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4257                         goto unwind;
4258                 }
4259
4260                 res = dev_set_mac_address(slave->dev, addr);
4261                 if (res) {
4262                         /* TODO: consider downing the slave
4263                          * and retry ?
4264                          * User should expect communications
4265                          * breakage anyway until ARP finish
4266                          * updating, so...
4267                          */
4268                         pr_debug("err %d %s\n", res, slave->dev->name);
4269                         goto unwind;
4270                 }
4271         }
4272
4273         /* success */
4274         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4275         return 0;
4276
4277 unwind:
4278         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4279         tmp_sa.sa_family = bond_dev->type;
4280
4281         /* unwind from head to the slave that failed */
4282         stop_at = slave;
4283         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4284                 int tmp_res;
4285
4286                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4287                 if (tmp_res) {
4288                         pr_debug("unwind err %d dev %s\n",
4289                                  tmp_res, slave->dev->name);
4290                 }
4291         }
4292
4293         return res;
4294 }
4295
4296 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4297 {
4298         struct bonding *bond = netdev_priv(bond_dev);
4299         struct slave *slave, *start_at;
4300         int i, slave_no, res = 1;
4301         struct iphdr *iph = ip_hdr(skb);
4302
4303         read_lock(&bond->lock);
4304
4305         if (!BOND_IS_OK(bond))
4306                 goto out;
4307         /*
4308          * Start with the curr_active_slave that joined the bond as the
4309          * default for sending IGMP traffic.  For failover purposes one
4310          * needs to maintain some consistency for the interface that will
4311          * send the join/membership reports.  The curr_active_slave found
4312          * will send all of this type of traffic.
4313          */
4314         if ((iph->protocol == IPPROTO_IGMP) &&
4315             (skb->protocol == htons(ETH_P_IP))) {
4316
4317                 read_lock(&bond->curr_slave_lock);
4318                 slave = bond->curr_active_slave;
4319                 read_unlock(&bond->curr_slave_lock);
4320
4321                 if (!slave)
4322                         goto out;
4323         } else {
4324                 /*
4325                  * Concurrent TX may collide on rr_tx_counter; we accept
4326                  * that as being rare enough not to justify using an
4327                  * atomic op here.
4328                  */
4329                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4330
4331                 bond_for_each_slave(bond, slave, i) {
4332                         slave_no--;
4333                         if (slave_no < 0)
4334                                 break;
4335                 }
4336         }
4337
4338         start_at = slave;
4339         bond_for_each_slave_from(bond, slave, i, start_at) {
4340                 if (IS_UP(slave->dev) &&
4341                     (slave->link == BOND_LINK_UP) &&
4342                     (slave->state == BOND_STATE_ACTIVE)) {
4343                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4344                         break;
4345                 }
4346         }
4347
4348 out:
4349         if (res) {
4350                 /* no suitable interface, frame not sent */
4351                 dev_kfree_skb(skb);
4352         }
4353         read_unlock(&bond->lock);
4354         return NETDEV_TX_OK;
4355 }
4356
4357
4358 /*
4359  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4360  * the bond has a usable interface.
4361  */
4362 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4363 {
4364         struct bonding *bond = netdev_priv(bond_dev);
4365         int res = 1;
4366
4367         read_lock(&bond->lock);
4368         read_lock(&bond->curr_slave_lock);
4369
4370         if (!BOND_IS_OK(bond))
4371                 goto out;
4372
4373         if (!bond->curr_active_slave)
4374                 goto out;
4375
4376         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4377
4378 out:
4379         if (res)
4380                 /* no suitable interface, frame not sent */
4381                 dev_kfree_skb(skb);
4382
4383         read_unlock(&bond->curr_slave_lock);
4384         read_unlock(&bond->lock);
4385         return NETDEV_TX_OK;
4386 }
4387
4388 /*
4389  * In bond_xmit_xor() , we determine the output device by using a pre-
4390  * determined xmit_hash_policy(), If the selected device is not enabled,
4391  * find the next active slave.
4392  */
4393 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4394 {
4395         struct bonding *bond = netdev_priv(bond_dev);
4396         struct slave *slave, *start_at;
4397         int slave_no;
4398         int i;
4399         int res = 1;
4400
4401         read_lock(&bond->lock);
4402
4403         if (!BOND_IS_OK(bond))
4404                 goto out;
4405
4406         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4407
4408         bond_for_each_slave(bond, slave, i) {
4409                 slave_no--;
4410                 if (slave_no < 0)
4411                         break;
4412         }
4413
4414         start_at = slave;
4415
4416         bond_for_each_slave_from(bond, slave, i, start_at) {
4417                 if (IS_UP(slave->dev) &&
4418                     (slave->link == BOND_LINK_UP) &&
4419                     (slave->state == BOND_STATE_ACTIVE)) {
4420                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4421                         break;
4422                 }
4423         }
4424
4425 out:
4426         if (res) {
4427                 /* no suitable interface, frame not sent */
4428                 dev_kfree_skb(skb);
4429         }
4430         read_unlock(&bond->lock);
4431         return NETDEV_TX_OK;
4432 }
4433
4434 /*
4435  * in broadcast mode, we send everything to all usable interfaces.
4436  */
4437 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4438 {
4439         struct bonding *bond = netdev_priv(bond_dev);
4440         struct slave *slave, *start_at;
4441         struct net_device *tx_dev = NULL;
4442         int i;
4443         int res = 1;
4444
4445         read_lock(&bond->lock);
4446
4447         if (!BOND_IS_OK(bond))
4448                 goto out;
4449
4450         read_lock(&bond->curr_slave_lock);
4451         start_at = bond->curr_active_slave;
4452         read_unlock(&bond->curr_slave_lock);
4453
4454         if (!start_at)
4455                 goto out;
4456
4457         bond_for_each_slave_from(bond, slave, i, start_at) {
4458                 if (IS_UP(slave->dev) &&
4459                     (slave->link == BOND_LINK_UP) &&
4460                     (slave->state == BOND_STATE_ACTIVE)) {
4461                         if (tx_dev) {
4462                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4463                                 if (!skb2) {
4464                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4465                                                bond_dev->name);
4466                                         continue;
4467                                 }
4468
4469                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4470                                 if (res) {
4471                                         dev_kfree_skb(skb2);
4472                                         continue;
4473                                 }
4474                         }
4475                         tx_dev = slave->dev;
4476                 }
4477         }
4478
4479         if (tx_dev)
4480                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4481
4482 out:
4483         if (res)
4484                 /* no suitable interface, frame not sent */
4485                 dev_kfree_skb(skb);
4486
4487         /* frame sent to all suitable interfaces */
4488         read_unlock(&bond->lock);
4489         return NETDEV_TX_OK;
4490 }
4491
4492 /*------------------------- Device initialization ---------------------------*/
4493
4494 static void bond_set_xmit_hash_policy(struct bonding *bond)
4495 {
4496         switch (bond->params.xmit_policy) {
4497         case BOND_XMIT_POLICY_LAYER23:
4498                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4499                 break;
4500         case BOND_XMIT_POLICY_LAYER34:
4501                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4502                 break;
4503         case BOND_XMIT_POLICY_LAYER2:
4504         default:
4505                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4506                 break;
4507         }
4508 }
4509
4510 /*
4511  * Lookup the slave that corresponds to a qid
4512  */
4513 static inline int bond_slave_override(struct bonding *bond,
4514                                       struct sk_buff *skb)
4515 {
4516         int i, res = 1;
4517         struct slave *slave = NULL;
4518         struct slave *check_slave;
4519
4520         read_lock(&bond->lock);
4521
4522         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4523                 goto out;
4524
4525         /* Find out if any slaves have the same mapping as this skb. */
4526         bond_for_each_slave(bond, check_slave, i) {
4527                 if (check_slave->queue_id == skb->queue_mapping) {
4528                         slave = check_slave;
4529                         break;
4530                 }
4531         }
4532
4533         /* If the slave isn't UP, use default transmit policy. */
4534         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4535             (slave->link == BOND_LINK_UP)) {
4536                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4537         }
4538
4539 out:
4540         read_unlock(&bond->lock);
4541         return res;
4542 }
4543
4544 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4545 {
4546         /*
4547          * This helper function exists to help dev_pick_tx get the correct
4548          * destination queue.  Using a helper function skips the a call to
4549          * skb_tx_hash and will put the skbs in the queue we expect on their
4550          * way down to the bonding driver.
4551          */
4552         return skb->queue_mapping;
4553 }
4554
4555 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4556 {
4557         struct bonding *bond = netdev_priv(dev);
4558
4559         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4560                 if (!bond_slave_override(bond, skb))
4561                         return NETDEV_TX_OK;
4562         }
4563
4564         switch (bond->params.mode) {
4565         case BOND_MODE_ROUNDROBIN:
4566                 return bond_xmit_roundrobin(skb, dev);
4567         case BOND_MODE_ACTIVEBACKUP:
4568                 return bond_xmit_activebackup(skb, dev);
4569         case BOND_MODE_XOR:
4570                 return bond_xmit_xor(skb, dev);
4571         case BOND_MODE_BROADCAST:
4572                 return bond_xmit_broadcast(skb, dev);
4573         case BOND_MODE_8023AD:
4574                 return bond_3ad_xmit_xor(skb, dev);
4575         case BOND_MODE_ALB:
4576         case BOND_MODE_TLB:
4577                 return bond_alb_xmit(skb, dev);
4578         default:
4579                 /* Should never happen, mode already checked */
4580                 pr_err("%s: Error: Unknown bonding mode %d\n",
4581                        dev->name, bond->params.mode);
4582                 WARN_ON_ONCE(1);
4583                 dev_kfree_skb(skb);
4584                 return NETDEV_TX_OK;
4585         }
4586 }
4587
4588
4589 /*
4590  * set bond mode specific net device operations
4591  */
4592 void bond_set_mode_ops(struct bonding *bond, int mode)
4593 {
4594         struct net_device *bond_dev = bond->dev;
4595
4596         switch (mode) {
4597         case BOND_MODE_ROUNDROBIN:
4598                 break;
4599         case BOND_MODE_ACTIVEBACKUP:
4600                 break;
4601         case BOND_MODE_XOR:
4602                 bond_set_xmit_hash_policy(bond);
4603                 break;
4604         case BOND_MODE_BROADCAST:
4605                 break;
4606         case BOND_MODE_8023AD:
4607                 bond_set_master_3ad_flags(bond);
4608                 bond_set_xmit_hash_policy(bond);
4609                 break;
4610         case BOND_MODE_ALB:
4611                 bond_set_master_alb_flags(bond);
4612                 /* FALLTHRU */
4613         case BOND_MODE_TLB:
4614                 break;
4615         default:
4616                 /* Should never happen, mode already checked */
4617                 pr_err("%s: Error: Unknown bonding mode %d\n",
4618                        bond_dev->name, mode);
4619                 break;
4620         }
4621 }
4622
4623 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4624                                     struct ethtool_drvinfo *drvinfo)
4625 {
4626         strncpy(drvinfo->driver, DRV_NAME, 32);
4627         strncpy(drvinfo->version, DRV_VERSION, 32);
4628         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4629 }
4630
4631 static const struct ethtool_ops bond_ethtool_ops = {
4632         .get_drvinfo            = bond_ethtool_get_drvinfo,
4633         .get_link               = ethtool_op_get_link,
4634         .get_tx_csum            = ethtool_op_get_tx_csum,
4635         .get_sg                 = ethtool_op_get_sg,
4636         .get_tso                = ethtool_op_get_tso,
4637         .get_ufo                = ethtool_op_get_ufo,
4638         .get_flags              = ethtool_op_get_flags,
4639 };
4640
4641 static const struct net_device_ops bond_netdev_ops = {
4642         .ndo_init               = bond_init,
4643         .ndo_uninit             = bond_uninit,
4644         .ndo_open               = bond_open,
4645         .ndo_stop               = bond_close,
4646         .ndo_start_xmit         = bond_start_xmit,
4647         .ndo_select_queue       = bond_select_queue,
4648         .ndo_get_stats64        = bond_get_stats,
4649         .ndo_do_ioctl           = bond_do_ioctl,
4650         .ndo_set_multicast_list = bond_set_multicast_list,
4651         .ndo_change_mtu         = bond_change_mtu,
4652         .ndo_set_mac_address    = bond_set_mac_address,
4653         .ndo_neigh_setup        = bond_neigh_setup,
4654         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4655         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4656         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4657 #ifdef CONFIG_NET_POLL_CONTROLLER
4658         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4659         .ndo_poll_controller    = bond_poll_controller,
4660 #endif
4661 };
4662
4663 static void bond_destructor(struct net_device *bond_dev)
4664 {
4665         struct bonding *bond = netdev_priv(bond_dev);
4666         if (bond->wq)
4667                 destroy_workqueue(bond->wq);
4668         free_netdev(bond_dev);
4669 }
4670
4671 static void bond_setup(struct net_device *bond_dev)
4672 {
4673         struct bonding *bond = netdev_priv(bond_dev);
4674
4675         /* initialize rwlocks */
4676         rwlock_init(&bond->lock);
4677         rwlock_init(&bond->curr_slave_lock);
4678
4679         bond->params = bonding_defaults;
4680
4681         /* Initialize pointers */
4682         bond->dev = bond_dev;
4683         INIT_LIST_HEAD(&bond->vlan_list);
4684
4685         /* Initialize the device entry points */
4686         ether_setup(bond_dev);
4687         bond_dev->netdev_ops = &bond_netdev_ops;
4688         bond_dev->ethtool_ops = &bond_ethtool_ops;
4689         bond_set_mode_ops(bond, bond->params.mode);
4690
4691         bond_dev->destructor = bond_destructor;
4692
4693         /* Initialize the device options */
4694         bond_dev->tx_queue_len = 0;
4695         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4696         bond_dev->priv_flags |= IFF_BONDING;
4697         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4698
4699         if (bond->params.arp_interval)
4700                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4701
4702         /* At first, we block adding VLANs. That's the only way to
4703          * prevent problems that occur when adding VLANs over an
4704          * empty bond. The block will be removed once non-challenged
4705          * slaves are enslaved.
4706          */
4707         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4708
4709         /* don't acquire bond device's netif_tx_lock when
4710          * transmitting */
4711         bond_dev->features |= NETIF_F_LLTX;
4712
4713         /* By default, we declare the bond to be fully
4714          * VLAN hardware accelerated capable. Special
4715          * care is taken in the various xmit functions
4716          * when there are slaves that are not hw accel
4717          * capable
4718          */
4719         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4720                                NETIF_F_HW_VLAN_RX |
4721                                NETIF_F_HW_VLAN_FILTER);
4722
4723         /* By default, we enable GRO on bonding devices.
4724          * Actual support requires lowlevel drivers are GRO ready.
4725          */
4726         bond_dev->features |= NETIF_F_GRO;
4727 }
4728
4729 static void bond_work_cancel_all(struct bonding *bond)
4730 {
4731         write_lock_bh(&bond->lock);
4732         bond->kill_timers = 1;
4733         write_unlock_bh(&bond->lock);
4734
4735         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4736                 cancel_delayed_work(&bond->mii_work);
4737
4738         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4739                 cancel_delayed_work(&bond->arp_work);
4740
4741         if (bond->params.mode == BOND_MODE_ALB &&
4742             delayed_work_pending(&bond->alb_work))
4743                 cancel_delayed_work(&bond->alb_work);
4744
4745         if (bond->params.mode == BOND_MODE_8023AD &&
4746             delayed_work_pending(&bond->ad_work))
4747                 cancel_delayed_work(&bond->ad_work);
4748
4749         if (delayed_work_pending(&bond->mcast_work))
4750                 cancel_delayed_work(&bond->mcast_work);
4751 }
4752
4753 /*
4754 * Destroy a bonding device.
4755 * Must be under rtnl_lock when this function is called.
4756 */
4757 static void bond_uninit(struct net_device *bond_dev)
4758 {
4759         struct bonding *bond = netdev_priv(bond_dev);
4760         struct vlan_entry *vlan, *tmp;
4761
4762         bond_netpoll_cleanup(bond_dev);
4763
4764         /* Release the bonded slaves */
4765         bond_release_all(bond_dev);
4766
4767         list_del(&bond->bond_list);
4768
4769         bond_work_cancel_all(bond);
4770
4771         bond_remove_proc_entry(bond);
4772
4773         __hw_addr_flush(&bond->mc_list);
4774
4775         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4776                 list_del(&vlan->vlan_list);
4777                 kfree(vlan);
4778         }
4779 }
4780
4781 /*------------------------- Module initialization ---------------------------*/
4782
4783 /*
4784  * Convert string input module parms.  Accept either the
4785  * number of the mode or its string name.  A bit complicated because
4786  * some mode names are substrings of other names, and calls from sysfs
4787  * may have whitespace in the name (trailing newlines, for example).
4788  */
4789 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4790 {
4791         int modeint = -1, i, rv;
4792         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4793
4794         for (p = (char *)buf; *p; p++)
4795                 if (!(isdigit(*p) || isspace(*p)))
4796                         break;
4797
4798         if (*p)
4799                 rv = sscanf(buf, "%20s", modestr);
4800         else
4801                 rv = sscanf(buf, "%d", &modeint);
4802
4803         if (!rv)
4804                 return -1;
4805
4806         for (i = 0; tbl[i].modename; i++) {
4807                 if (modeint == tbl[i].mode)
4808                         return tbl[i].mode;
4809                 if (strcmp(modestr, tbl[i].modename) == 0)
4810                         return tbl[i].mode;
4811         }
4812
4813         return -1;
4814 }
4815
4816 static int bond_check_params(struct bond_params *params)
4817 {
4818         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4819
4820         /*
4821          * Convert string parameters.
4822          */
4823         if (mode) {
4824                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4825                 if (bond_mode == -1) {
4826                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4827                                mode == NULL ? "NULL" : mode);
4828                         return -EINVAL;
4829                 }
4830         }
4831
4832         if (xmit_hash_policy) {
4833                 if ((bond_mode != BOND_MODE_XOR) &&
4834                     (bond_mode != BOND_MODE_8023AD)) {
4835                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4836                                bond_mode_name(bond_mode));
4837                 } else {
4838                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4839                                                         xmit_hashtype_tbl);
4840                         if (xmit_hashtype == -1) {
4841                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4842                                        xmit_hash_policy == NULL ? "NULL" :
4843                                        xmit_hash_policy);
4844                                 return -EINVAL;
4845                         }
4846                 }
4847         }
4848
4849         if (lacp_rate) {
4850                 if (bond_mode != BOND_MODE_8023AD) {
4851                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4852                                 bond_mode_name(bond_mode));
4853                 } else {
4854                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4855                         if (lacp_fast == -1) {
4856                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4857                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4858                                 return -EINVAL;
4859                         }
4860                 }
4861         }
4862
4863         if (ad_select) {
4864                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4865                 if (params->ad_select == -1) {
4866                         pr_err("Error: Invalid ad_select \"%s\"\n",
4867                                ad_select == NULL ? "NULL" : ad_select);
4868                         return -EINVAL;
4869                 }
4870
4871                 if (bond_mode != BOND_MODE_8023AD) {
4872                         pr_warning("ad_select param only affects 802.3ad mode\n");
4873                 }
4874         } else {
4875                 params->ad_select = BOND_AD_STABLE;
4876         }
4877
4878         if (max_bonds < 0) {
4879                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4880                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4881                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4882         }
4883
4884         if (miimon < 0) {
4885                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4886                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4887                 miimon = BOND_LINK_MON_INTERV;
4888         }
4889
4890         if (updelay < 0) {
4891                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4892                            updelay, INT_MAX);
4893                 updelay = 0;
4894         }
4895
4896         if (downdelay < 0) {
4897                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4898                            downdelay, INT_MAX);
4899                 downdelay = 0;
4900         }
4901
4902         if ((use_carrier != 0) && (use_carrier != 1)) {
4903                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4904                            use_carrier);
4905                 use_carrier = 1;
4906         }
4907
4908         if (num_grat_arp < 0 || num_grat_arp > 255) {
4909                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4910                            num_grat_arp);
4911                 num_grat_arp = 1;
4912         }
4913
4914         if (num_unsol_na < 0 || num_unsol_na > 255) {
4915                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4916                            num_unsol_na);
4917                 num_unsol_na = 1;
4918         }
4919
4920         /* reset values for 802.3ad */
4921         if (bond_mode == BOND_MODE_8023AD) {
4922                 if (!miimon) {
4923                         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");
4924                         pr_warning("Forcing miimon to 100msec\n");
4925                         miimon = 100;
4926                 }
4927         }
4928
4929         if (tx_queues < 1 || tx_queues > 255) {
4930                 pr_warning("Warning: tx_queues (%d) should be between "
4931                            "1 and 255, resetting to %d\n",
4932                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4933                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4934         }
4935
4936         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4937                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4938                            "not of valid value (0/1), so it was set to "
4939                            "0\n", all_slaves_active);
4940                 all_slaves_active = 0;
4941         }
4942
4943         if (resend_igmp < 0 || resend_igmp > 255) {
4944                 pr_warning("Warning: resend_igmp (%d) should be between "
4945                            "0 and 255, resetting to %d\n",
4946                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4947                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4948         }
4949
4950         /* reset values for TLB/ALB */
4951         if ((bond_mode == BOND_MODE_TLB) ||
4952             (bond_mode == BOND_MODE_ALB)) {
4953                 if (!miimon) {
4954                         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");
4955                         pr_warning("Forcing miimon to 100msec\n");
4956                         miimon = 100;
4957                 }
4958         }
4959
4960         if (bond_mode == BOND_MODE_ALB) {
4961                 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",
4962                           updelay);
4963         }
4964
4965         if (!miimon) {
4966                 if (updelay || downdelay) {
4967                         /* just warn the user the up/down delay will have
4968                          * no effect since miimon is zero...
4969                          */
4970                         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",
4971                                    updelay, downdelay);
4972                 }
4973         } else {
4974                 /* don't allow arp monitoring */
4975                 if (arp_interval) {
4976                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4977                                    miimon, arp_interval);
4978                         arp_interval = 0;
4979                 }
4980
4981                 if ((updelay % miimon) != 0) {
4982                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4983                                    updelay, miimon,
4984                                    (updelay / miimon) * miimon);
4985                 }
4986
4987                 updelay /= miimon;
4988
4989                 if ((downdelay % miimon) != 0) {
4990                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4991                                    downdelay, miimon,
4992                                    (downdelay / miimon) * miimon);
4993                 }
4994
4995                 downdelay /= miimon;
4996         }
4997
4998         if (arp_interval < 0) {
4999                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5000                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5001                 arp_interval = BOND_LINK_ARP_INTERV;
5002         }
5003
5004         for (arp_ip_count = 0;
5005              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5006              arp_ip_count++) {
5007                 /* not complete check, but should be good enough to
5008                    catch mistakes */
5009                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5010                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5011                                    arp_ip_target[arp_ip_count]);
5012                         arp_interval = 0;
5013                 } else {
5014                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5015                         arp_target[arp_ip_count] = ip;
5016                 }
5017         }
5018
5019         if (arp_interval && !arp_ip_count) {
5020                 /* don't allow arping if no arp_ip_target given... */
5021                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5022                            arp_interval);
5023                 arp_interval = 0;
5024         }
5025
5026         if (arp_validate) {
5027                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5028                         pr_err("arp_validate only supported in active-backup mode\n");
5029                         return -EINVAL;
5030                 }
5031                 if (!arp_interval) {
5032                         pr_err("arp_validate requires arp_interval\n");
5033                         return -EINVAL;
5034                 }
5035
5036                 arp_validate_value = bond_parse_parm(arp_validate,
5037                                                      arp_validate_tbl);
5038                 if (arp_validate_value == -1) {
5039                         pr_err("Error: invalid arp_validate \"%s\"\n",
5040                                arp_validate == NULL ? "NULL" : arp_validate);
5041                         return -EINVAL;
5042                 }
5043         } else
5044                 arp_validate_value = 0;
5045
5046         if (miimon) {
5047                 pr_info("MII link monitoring set to %d ms\n", miimon);
5048         } else if (arp_interval) {
5049                 int i;
5050
5051                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5052                         arp_interval,
5053                         arp_validate_tbl[arp_validate_value].modename,
5054                         arp_ip_count);
5055
5056                 for (i = 0; i < arp_ip_count; i++)
5057                         pr_info(" %s", arp_ip_target[i]);
5058
5059                 pr_info("\n");
5060
5061         } else if (max_bonds) {
5062                 /* miimon and arp_interval not set, we need one so things
5063                  * work as expected, see bonding.txt for details
5064                  */
5065                 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");
5066         }
5067
5068         if (primary && !USES_PRIMARY(bond_mode)) {
5069                 /* currently, using a primary only makes sense
5070                  * in active backup, TLB or ALB modes
5071                  */
5072                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5073                            primary, bond_mode_name(bond_mode));
5074                 primary = NULL;
5075         }
5076
5077         if (primary && primary_reselect) {
5078                 primary_reselect_value = bond_parse_parm(primary_reselect,
5079                                                          pri_reselect_tbl);
5080                 if (primary_reselect_value == -1) {
5081                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5082                                primary_reselect ==
5083                                         NULL ? "NULL" : primary_reselect);
5084                         return -EINVAL;
5085                 }
5086         } else {
5087                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5088         }
5089
5090         if (fail_over_mac) {
5091                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5092                                                       fail_over_mac_tbl);
5093                 if (fail_over_mac_value == -1) {
5094                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5095                                arp_validate == NULL ? "NULL" : arp_validate);
5096                         return -EINVAL;
5097                 }
5098
5099                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5100                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5101         } else {
5102                 fail_over_mac_value = BOND_FOM_NONE;
5103         }
5104
5105         /* fill params struct with the proper values */
5106         params->mode = bond_mode;
5107         params->xmit_policy = xmit_hashtype;
5108         params->miimon = miimon;
5109         params->num_grat_arp = num_grat_arp;
5110         params->num_unsol_na = num_unsol_na;
5111         params->arp_interval = arp_interval;
5112         params->arp_validate = arp_validate_value;
5113         params->updelay = updelay;
5114         params->downdelay = downdelay;
5115         params->use_carrier = use_carrier;
5116         params->lacp_fast = lacp_fast;
5117         params->primary[0] = 0;
5118         params->primary_reselect = primary_reselect_value;
5119         params->fail_over_mac = fail_over_mac_value;
5120         params->tx_queues = tx_queues;
5121         params->all_slaves_active = all_slaves_active;
5122         params->resend_igmp = resend_igmp;
5123
5124         if (primary) {
5125                 strncpy(params->primary, primary, IFNAMSIZ);
5126                 params->primary[IFNAMSIZ - 1] = 0;
5127         }
5128
5129         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5130
5131         return 0;
5132 }
5133
5134 static struct lock_class_key bonding_netdev_xmit_lock_key;
5135 static struct lock_class_key bonding_netdev_addr_lock_key;
5136
5137 static void bond_set_lockdep_class_one(struct net_device *dev,
5138                                        struct netdev_queue *txq,
5139                                        void *_unused)
5140 {
5141         lockdep_set_class(&txq->_xmit_lock,
5142                           &bonding_netdev_xmit_lock_key);
5143 }
5144
5145 static void bond_set_lockdep_class(struct net_device *dev)
5146 {
5147         lockdep_set_class(&dev->addr_list_lock,
5148                           &bonding_netdev_addr_lock_key);
5149         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5150 }
5151
5152 /*
5153  * Called from registration process
5154  */
5155 static int bond_init(struct net_device *bond_dev)
5156 {
5157         struct bonding *bond = netdev_priv(bond_dev);
5158         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5159
5160         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5161
5162         bond->wq = create_singlethread_workqueue(bond_dev->name);
5163         if (!bond->wq)
5164                 return -ENOMEM;
5165
5166         bond_set_lockdep_class(bond_dev);
5167
5168         netif_carrier_off(bond_dev);
5169
5170         bond_create_proc_entry(bond);
5171         list_add_tail(&bond->bond_list, &bn->dev_list);
5172
5173         bond_prepare_sysfs_group(bond);
5174
5175         __hw_addr_init(&bond->mc_list);
5176         return 0;
5177 }
5178
5179 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5180 {
5181         if (tb[IFLA_ADDRESS]) {
5182                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5183                         return -EINVAL;
5184                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5185                         return -EADDRNOTAVAIL;
5186         }
5187         return 0;
5188 }
5189
5190 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5191         .kind           = "bond",
5192         .priv_size      = sizeof(struct bonding),
5193         .setup          = bond_setup,
5194         .validate       = bond_validate,
5195 };
5196
5197 /* Create a new bond based on the specified name and bonding parameters.
5198  * If name is NULL, obtain a suitable "bond%d" name for us.
5199  * Caller must NOT hold rtnl_lock; we need to release it here before we
5200  * set up our sysfs entries.
5201  */
5202 int bond_create(struct net *net, const char *name)
5203 {
5204         struct net_device *bond_dev;
5205         int res;
5206
5207         rtnl_lock();
5208
5209         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5210                                 bond_setup, tx_queues);
5211         if (!bond_dev) {
5212                 pr_err("%s: eek! can't alloc netdev!\n", name);
5213                 rtnl_unlock();
5214                 return -ENOMEM;
5215         }
5216
5217         dev_net_set(bond_dev, net);
5218         bond_dev->rtnl_link_ops = &bond_link_ops;
5219
5220         if (!name) {
5221                 res = dev_alloc_name(bond_dev, "bond%d");
5222                 if (res < 0)
5223                         goto out;
5224         }
5225
5226         res = register_netdevice(bond_dev);
5227
5228 out:
5229         rtnl_unlock();
5230         if (res < 0)
5231                 bond_destructor(bond_dev);
5232         return res;
5233 }
5234
5235 static int __net_init bond_net_init(struct net *net)
5236 {
5237         struct bond_net *bn = net_generic(net, bond_net_id);
5238
5239         bn->net = net;
5240         INIT_LIST_HEAD(&bn->dev_list);
5241
5242         bond_create_proc_dir(bn);
5243         
5244         return 0;
5245 }
5246
5247 static void __net_exit bond_net_exit(struct net *net)
5248 {
5249         struct bond_net *bn = net_generic(net, bond_net_id);
5250
5251         bond_destroy_proc_dir(bn);
5252 }
5253
5254 static struct pernet_operations bond_net_ops = {
5255         .init = bond_net_init,
5256         .exit = bond_net_exit,
5257         .id   = &bond_net_id,
5258         .size = sizeof(struct bond_net),
5259 };
5260
5261 static int __init bonding_init(void)
5262 {
5263         int i;
5264         int res;
5265
5266         pr_info("%s", version);
5267
5268         res = bond_check_params(&bonding_defaults);
5269         if (res)
5270                 goto out;
5271
5272         res = register_pernet_subsys(&bond_net_ops);
5273         if (res)
5274                 goto out;
5275
5276         res = rtnl_link_register(&bond_link_ops);
5277         if (res)
5278                 goto err_link;
5279
5280         for (i = 0; i < max_bonds; i++) {
5281                 res = bond_create(&init_net, NULL);
5282                 if (res)
5283                         goto err;
5284         }
5285
5286         res = bond_create_sysfs();
5287         if (res)
5288                 goto err;
5289
5290         register_netdevice_notifier(&bond_netdev_notifier);
5291         register_inetaddr_notifier(&bond_inetaddr_notifier);
5292         bond_register_ipv6_notifier();
5293 out:
5294         return res;
5295 err:
5296         rtnl_link_unregister(&bond_link_ops);
5297 err_link:
5298         unregister_pernet_subsys(&bond_net_ops);
5299         goto out;
5300
5301 }
5302
5303 static void __exit bonding_exit(void)
5304 {
5305         unregister_netdevice_notifier(&bond_netdev_notifier);
5306         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5307         bond_unregister_ipv6_notifier();
5308
5309         bond_destroy_sysfs();
5310
5311         rtnl_link_unregister(&bond_link_ops);
5312         unregister_pernet_subsys(&bond_net_ops);
5313 }
5314
5315 module_init(bonding_init);
5316 module_exit(bonding_exit);
5317 MODULE_LICENSE("GPL");
5318 MODULE_VERSION(DRV_VERSION);
5319 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5320 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5321 MODULE_ALIAS_RTNL_LINK("bond");