Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6.git] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <linux/io.h>
57 #include <asm/system.h>
58 #include <asm/dma.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
66 #include <net/sock.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
70 #include <net/arp.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <linux/preempt.h>
77 #include <net/route.h>
78 #include <net/net_namespace.h>
79 #include <net/netns/generic.h>
80 #include "bonding.h"
81 #include "bond_3ad.h"
82 #include "bond_alb.h"
83
84 /*---------------------------- Module parameters ----------------------------*/
85
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV    0
88 #define BOND_LINK_ARP_INTERV    0
89
90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
93 static int miimon       = BOND_LINK_MON_INTERV;
94 static int updelay;
95 static int downdelay;
96 static int use_carrier  = 1;
97 static char *mode;
98 static char *primary;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static char *ad_select;
102 static char *xmit_hash_policy;
103 static int arp_interval = BOND_LINK_ARP_INTERV;
104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
105 static char *arp_validate;
106 static char *fail_over_mac;
107 static int all_slaves_active = 0;
108 static struct bond_params bonding_defaults;
109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
110
111 module_param(max_bonds, int, 0);
112 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
113 module_param(tx_queues, int, 0);
114 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
115 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
116 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
117                                "failover event (alias of num_unsol_na)");
118 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
119 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
120                                "failover event (alias of num_grat_arp)");
121 module_param(miimon, int, 0);
122 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
123 module_param(updelay, int, 0);
124 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
125 module_param(downdelay, int, 0);
126 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
127                             "in milliseconds");
128 module_param(use_carrier, int, 0);
129 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
130                               "0 for off, 1 for on (default)");
131 module_param(mode, charp, 0);
132 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
133                        "1 for active-backup, 2 for balance-xor, "
134                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
135                        "6 for balance-alb");
136 module_param(primary, charp, 0);
137 MODULE_PARM_DESC(primary, "Primary network device to use");
138 module_param(primary_reselect, charp, 0);
139 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
140                                    "once it comes up; "
141                                    "0 for always (default), "
142                                    "1 for only if speed of primary is "
143                                    "better, "
144                                    "2 for only on active slave "
145                                    "failure");
146 module_param(lacp_rate, charp, 0);
147 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
148                             "0 for slow, 1 for fast");
149 module_param(ad_select, charp, 0);
150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
151                             "0 for stable (default), 1 for bandwidth, "
152                             "2 for count");
153 module_param(xmit_hash_policy, charp, 0);
154 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
155                                    "0 for layer 2 (default), 1 for layer 3+4, "
156                                    "2 for layer 2+3");
157 module_param(arp_interval, int, 0);
158 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
159 module_param_array(arp_ip_target, charp, NULL, 0);
160 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
161 module_param(arp_validate, charp, 0);
162 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
163                                "0 for none (default), 1 for active, "
164                                "2 for backup, 3 for all");
165 module_param(fail_over_mac, charp, 0);
166 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
167                                 "the same MAC; 0 for none (default), "
168                                 "1 for active, 2 for follow");
169 module_param(all_slaves_active, int, 0);
170 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
171                                      "by setting active flag for all slaves; "
172                                      "0 for never (default), 1 for always.");
173 module_param(resend_igmp, int, 0);
174 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
175                               "link failure");
176
177 /*----------------------------- Global variables ----------------------------*/
178
179 #ifdef CONFIG_NET_POLL_CONTROLLER
180 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
181 #endif
182
183 int bond_net_id __read_mostly;
184
185 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
186 static int arp_ip_count;
187 static int bond_mode    = BOND_MODE_ROUNDROBIN;
188 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
189 static int lacp_fast;
190
191 const struct bond_parm_tbl bond_lacp_tbl[] = {
192 {       "slow",         AD_LACP_SLOW},
193 {       "fast",         AD_LACP_FAST},
194 {       NULL,           -1},
195 };
196
197 const struct bond_parm_tbl bond_mode_tbl[] = {
198 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
199 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
200 {       "balance-xor",          BOND_MODE_XOR},
201 {       "broadcast",            BOND_MODE_BROADCAST},
202 {       "802.3ad",              BOND_MODE_8023AD},
203 {       "balance-tlb",          BOND_MODE_TLB},
204 {       "balance-alb",          BOND_MODE_ALB},
205 {       NULL,                   -1},
206 };
207
208 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
209 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
210 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
211 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
212 {       NULL,                   -1},
213 };
214
215 const struct bond_parm_tbl arp_validate_tbl[] = {
216 {       "none",                 BOND_ARP_VALIDATE_NONE},
217 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
218 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
219 {       "all",                  BOND_ARP_VALIDATE_ALL},
220 {       NULL,                   -1},
221 };
222
223 const struct bond_parm_tbl fail_over_mac_tbl[] = {
224 {       "none",                 BOND_FOM_NONE},
225 {       "active",               BOND_FOM_ACTIVE},
226 {       "follow",               BOND_FOM_FOLLOW},
227 {       NULL,                   -1},
228 };
229
230 const struct bond_parm_tbl pri_reselect_tbl[] = {
231 {       "always",               BOND_PRI_RESELECT_ALWAYS},
232 {       "better",               BOND_PRI_RESELECT_BETTER},
233 {       "failure",              BOND_PRI_RESELECT_FAILURE},
234 {       NULL,                   -1},
235 };
236
237 struct bond_parm_tbl ad_select_tbl[] = {
238 {       "stable",       BOND_AD_STABLE},
239 {       "bandwidth",    BOND_AD_BANDWIDTH},
240 {       "count",        BOND_AD_COUNT},
241 {       NULL,           -1},
242 };
243
244 /*-------------------------- Forward declarations ---------------------------*/
245
246 static int bond_init(struct net_device *bond_dev);
247 static void bond_uninit(struct net_device *bond_dev);
248
249 /*---------------------------- General routines -----------------------------*/
250
251 const char *bond_mode_name(int mode)
252 {
253         static const char *names[] = {
254                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
255                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
256                 [BOND_MODE_XOR] = "load balancing (xor)",
257                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
258                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
259                 [BOND_MODE_TLB] = "transmit load balancing",
260                 [BOND_MODE_ALB] = "adaptive load balancing",
261         };
262
263         if (mode < 0 || mode > BOND_MODE_ALB)
264                 return "unknown";
265
266         return names[mode];
267 }
268
269 /*---------------------------------- VLAN -----------------------------------*/
270
271 /**
272  * bond_add_vlan - add a new vlan id on bond
273  * @bond: bond that got the notification
274  * @vlan_id: the vlan id to add
275  *
276  * Returns -ENOMEM if allocation failed.
277  */
278 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
279 {
280         struct vlan_entry *vlan;
281
282         pr_debug("bond: %s, vlan id %d\n",
283                  (bond ? bond->dev->name : "None"), vlan_id);
284
285         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
286         if (!vlan)
287                 return -ENOMEM;
288
289         INIT_LIST_HEAD(&vlan->vlan_list);
290         vlan->vlan_id = vlan_id;
291
292         write_lock_bh(&bond->lock);
293
294         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
295
296         write_unlock_bh(&bond->lock);
297
298         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
299
300         return 0;
301 }
302
303 /**
304  * bond_del_vlan - delete a vlan id from bond
305  * @bond: bond that got the notification
306  * @vlan_id: the vlan id to delete
307  *
308  * returns -ENODEV if @vlan_id was not found in @bond.
309  */
310 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
311 {
312         struct vlan_entry *vlan;
313         int res = -ENODEV;
314
315         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
316
317         block_netpoll_tx();
318         write_lock_bh(&bond->lock);
319
320         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
321                 if (vlan->vlan_id == vlan_id) {
322                         list_del(&vlan->vlan_list);
323
324                         if (bond_is_lb(bond))
325                                 bond_alb_clear_vlan(bond, vlan_id);
326
327                         pr_debug("removed VLAN ID %d from bond %s\n",
328                                  vlan_id, bond->dev->name);
329
330                         kfree(vlan);
331
332                         res = 0;
333                         goto out;
334                 }
335         }
336
337         pr_debug("couldn't find VLAN ID %d in bond %s\n",
338                  vlan_id, bond->dev->name);
339
340 out:
341         write_unlock_bh(&bond->lock);
342         unblock_netpoll_tx();
343         return res;
344 }
345
346 /**
347  * bond_next_vlan - safely skip to the next item in the vlans list.
348  * @bond: the bond we're working on
349  * @curr: item we're advancing from
350  *
351  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
352  * or @curr->next otherwise (even if it is @curr itself again).
353  *
354  * Caller must hold bond->lock
355  */
356 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
357 {
358         struct vlan_entry *next, *last;
359
360         if (list_empty(&bond->vlan_list))
361                 return NULL;
362
363         if (!curr) {
364                 next = list_entry(bond->vlan_list.next,
365                                   struct vlan_entry, vlan_list);
366         } else {
367                 last = list_entry(bond->vlan_list.prev,
368                                   struct vlan_entry, vlan_list);
369                 if (last == curr) {
370                         next = list_entry(bond->vlan_list.next,
371                                           struct vlan_entry, vlan_list);
372                 } else {
373                         next = list_entry(curr->vlan_list.next,
374                                           struct vlan_entry, vlan_list);
375                 }
376         }
377
378         return next;
379 }
380
381 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
382
383 /**
384  * bond_dev_queue_xmit - Prepare skb for xmit.
385  *
386  * @bond: bond device that got this skb for tx.
387  * @skb: hw accel VLAN tagged skb to transmit
388  * @slave_dev: slave that is supposed to xmit this skbuff
389  */
390 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
391                         struct net_device *slave_dev)
392 {
393         skb->dev = slave_dev;
394         skb->priority = 1;
395
396         skb->queue_mapping = bond_queue_mapping(skb);
397
398         if (unlikely(netpoll_tx_running(slave_dev)))
399                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
400         else
401                 dev_queue_xmit(skb);
402
403         return 0;
404 }
405
406 /*
407  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
408  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
409  * lock because:
410  * a. This operation is performed in IOCTL context,
411  * b. The operation is protected by the RTNL semaphore in the 8021q code,
412  * c. Holding a lock with BH disabled while directly calling a base driver
413  *    entry point is generally a BAD idea.
414  *
415  * The design of synchronization/protection for this operation in the 8021q
416  * module is good for one or more VLAN devices over a single physical device
417  * and cannot be extended for a teaming solution like bonding, so there is a
418  * potential race condition here where a net device from the vlan group might
419  * be referenced (either by a base driver or the 8021q code) while it is being
420  * removed from the system. However, it turns out we're not making matters
421  * worse, and if it works for regular VLAN usage it will work here too.
422 */
423
424 /**
425  * bond_vlan_rx_register - Propagates registration to slaves
426  * @bond_dev: bonding net device that got called
427  * @grp: vlan group being registered
428  */
429 static void bond_vlan_rx_register(struct net_device *bond_dev,
430                                   struct vlan_group *grp)
431 {
432         struct bonding *bond = netdev_priv(bond_dev);
433         struct slave *slave;
434         int i;
435
436         write_lock_bh(&bond->lock);
437         bond->vlgrp = grp;
438         write_unlock_bh(&bond->lock);
439
440         bond_for_each_slave(bond, slave, i) {
441                 struct net_device *slave_dev = slave->dev;
442                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
443
444                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
445                     slave_ops->ndo_vlan_rx_register) {
446                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
447                 }
448         }
449 }
450
451 /**
452  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
453  * @bond_dev: bonding net device that got called
454  * @vid: vlan id being added
455  */
456 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
457 {
458         struct bonding *bond = netdev_priv(bond_dev);
459         struct slave *slave;
460         int i, res;
461
462         bond_for_each_slave(bond, slave, i) {
463                 struct net_device *slave_dev = slave->dev;
464                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
465
466                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
467                     slave_ops->ndo_vlan_rx_add_vid) {
468                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
469                 }
470         }
471
472         res = bond_add_vlan(bond, vid);
473         if (res) {
474                 pr_err("%s: Error: Failed to add vlan id %d\n",
475                        bond_dev->name, vid);
476         }
477 }
478
479 /**
480  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
481  * @bond_dev: bonding net device that got called
482  * @vid: vlan id being removed
483  */
484 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
485 {
486         struct bonding *bond = netdev_priv(bond_dev);
487         struct slave *slave;
488         struct net_device *vlan_dev;
489         int i, res;
490
491         bond_for_each_slave(bond, slave, i) {
492                 struct net_device *slave_dev = slave->dev;
493                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
494
495                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
496                     slave_ops->ndo_vlan_rx_kill_vid) {
497                         /* Save and then restore vlan_dev in the grp array,
498                          * since the slave's driver might clear it.
499                          */
500                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
501                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
502                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
503                 }
504         }
505
506         res = bond_del_vlan(bond, vid);
507         if (res) {
508                 pr_err("%s: Error: Failed to remove vlan id %d\n",
509                        bond_dev->name, vid);
510         }
511 }
512
513 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
514 {
515         struct vlan_entry *vlan;
516         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
517
518         if (!bond->vlgrp)
519                 return;
520
521         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
522             slave_ops->ndo_vlan_rx_register)
523                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
524
525         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
526             !(slave_ops->ndo_vlan_rx_add_vid))
527                 return;
528
529         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
530                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
531 }
532
533 static void bond_del_vlans_from_slave(struct bonding *bond,
534                                       struct net_device *slave_dev)
535 {
536         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
537         struct vlan_entry *vlan;
538         struct net_device *vlan_dev;
539
540         if (!bond->vlgrp)
541                 return;
542
543         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
544             !(slave_ops->ndo_vlan_rx_kill_vid))
545                 goto unreg;
546
547         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
548                 if (!vlan->vlan_id)
549                         continue;
550                 /* Save and then restore vlan_dev in the grp array,
551                  * since the slave's driver might clear it.
552                  */
553                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
554                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
555                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
556         }
557
558 unreg:
559         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
560             slave_ops->ndo_vlan_rx_register)
561                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
562 }
563
564 /*------------------------------- Link status -------------------------------*/
565
566 /*
567  * Set the carrier state for the master according to the state of its
568  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
569  * do special 802.3ad magic.
570  *
571  * Returns zero if carrier state does not change, nonzero if it does.
572  */
573 static int bond_set_carrier(struct bonding *bond)
574 {
575         struct slave *slave;
576         int i;
577
578         if (bond->slave_cnt == 0)
579                 goto down;
580
581         if (bond->params.mode == BOND_MODE_8023AD)
582                 return bond_3ad_set_carrier(bond);
583
584         bond_for_each_slave(bond, slave, i) {
585                 if (slave->link == BOND_LINK_UP) {
586                         if (!netif_carrier_ok(bond->dev)) {
587                                 netif_carrier_on(bond->dev);
588                                 return 1;
589                         }
590                         return 0;
591                 }
592         }
593
594 down:
595         if (netif_carrier_ok(bond->dev)) {
596                 netif_carrier_off(bond->dev);
597                 return 1;
598         }
599         return 0;
600 }
601
602 /*
603  * Get link speed and duplex from the slave's base driver
604  * using ethtool. If for some reason the call fails or the
605  * values are invalid, fake speed and duplex to 100/Full
606  * and return error.
607  */
608 static int bond_update_speed_duplex(struct slave *slave)
609 {
610         struct net_device *slave_dev = slave->dev;
611         struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
612         u32 slave_speed;
613         int res;
614
615         /* Fake speed and duplex */
616         slave->speed = SPEED_100;
617         slave->duplex = DUPLEX_FULL;
618
619         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
620                 return -1;
621
622         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
623         if (res < 0)
624                 return -1;
625
626         slave_speed = ethtool_cmd_speed(&etool);
627         if (slave_speed == 0 || slave_speed == ((__u32) -1))
628                 return -1;
629
630         switch (etool.duplex) {
631         case DUPLEX_FULL:
632         case DUPLEX_HALF:
633                 break;
634         default:
635                 return -1;
636         }
637
638         slave->speed = slave_speed;
639         slave->duplex = etool.duplex;
640
641         return 0;
642 }
643
644 /*
645  * if <dev> supports MII link status reporting, check its link status.
646  *
647  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
648  * depending upon the setting of the use_carrier parameter.
649  *
650  * Return either BMSR_LSTATUS, meaning that the link is up (or we
651  * can't tell and just pretend it is), or 0, meaning that the link is
652  * down.
653  *
654  * If reporting is non-zero, instead of faking link up, return -1 if
655  * both ETHTOOL and MII ioctls fail (meaning the device does not
656  * support them).  If use_carrier is set, return whatever it says.
657  * It'd be nice if there was a good way to tell if a driver supports
658  * netif_carrier, but there really isn't.
659  */
660 static int bond_check_dev_link(struct bonding *bond,
661                                struct net_device *slave_dev, int reporting)
662 {
663         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
664         int (*ioctl)(struct net_device *, struct ifreq *, int);
665         struct ifreq ifr;
666         struct mii_ioctl_data *mii;
667
668         if (!reporting && !netif_running(slave_dev))
669                 return 0;
670
671         if (bond->params.use_carrier)
672                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
673
674         /* Try to get link status using Ethtool first. */
675         if (slave_dev->ethtool_ops) {
676                 if (slave_dev->ethtool_ops->get_link) {
677                         u32 link;
678
679                         link = slave_dev->ethtool_ops->get_link(slave_dev);
680
681                         return link ? BMSR_LSTATUS : 0;
682                 }
683         }
684
685         /* Ethtool can't be used, fallback to MII ioctls. */
686         ioctl = slave_ops->ndo_do_ioctl;
687         if (ioctl) {
688                 /* TODO: set pointer to correct ioctl on a per team member */
689                 /*       bases to make this more efficient. that is, once  */
690                 /*       we determine the correct ioctl, we will always    */
691                 /*       call it and not the others for that team          */
692                 /*       member.                                           */
693
694                 /*
695                  * We cannot assume that SIOCGMIIPHY will also read a
696                  * register; not all network drivers (e.g., e100)
697                  * support that.
698                  */
699
700                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
701                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
702                 mii = if_mii(&ifr);
703                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
704                         mii->reg_num = MII_BMSR;
705                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
706                                 return mii->val_out & BMSR_LSTATUS;
707                 }
708         }
709
710         /*
711          * If reporting, report that either there's no dev->do_ioctl,
712          * or both SIOCGMIIREG and get_link failed (meaning that we
713          * cannot report link status).  If not reporting, pretend
714          * we're ok.
715          */
716         return reporting ? -1 : BMSR_LSTATUS;
717 }
718
719 /*----------------------------- Multicast list ------------------------------*/
720
721 /*
722  * Push the promiscuity flag down to appropriate slaves
723  */
724 static int bond_set_promiscuity(struct bonding *bond, int inc)
725 {
726         int err = 0;
727         if (USES_PRIMARY(bond->params.mode)) {
728                 /* write lock already acquired */
729                 if (bond->curr_active_slave) {
730                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
731                                                   inc);
732                 }
733         } else {
734                 struct slave *slave;
735                 int i;
736                 bond_for_each_slave(bond, slave, i) {
737                         err = dev_set_promiscuity(slave->dev, inc);
738                         if (err)
739                                 return err;
740                 }
741         }
742         return err;
743 }
744
745 /*
746  * Push the allmulti flag down to all slaves
747  */
748 static int bond_set_allmulti(struct bonding *bond, int inc)
749 {
750         int err = 0;
751         if (USES_PRIMARY(bond->params.mode)) {
752                 /* write lock already acquired */
753                 if (bond->curr_active_slave) {
754                         err = dev_set_allmulti(bond->curr_active_slave->dev,
755                                                inc);
756                 }
757         } else {
758                 struct slave *slave;
759                 int i;
760                 bond_for_each_slave(bond, slave, i) {
761                         err = dev_set_allmulti(slave->dev, inc);
762                         if (err)
763                                 return err;
764                 }
765         }
766         return err;
767 }
768
769 /*
770  * Add a Multicast address to slaves
771  * according to mode
772  */
773 static void bond_mc_add(struct bonding *bond, void *addr)
774 {
775         if (USES_PRIMARY(bond->params.mode)) {
776                 /* write lock already acquired */
777                 if (bond->curr_active_slave)
778                         dev_mc_add(bond->curr_active_slave->dev, addr);
779         } else {
780                 struct slave *slave;
781                 int i;
782
783                 bond_for_each_slave(bond, slave, i)
784                         dev_mc_add(slave->dev, addr);
785         }
786 }
787
788 /*
789  * Remove a multicast address from slave
790  * according to mode
791  */
792 static void bond_mc_del(struct bonding *bond, void *addr)
793 {
794         if (USES_PRIMARY(bond->params.mode)) {
795                 /* write lock already acquired */
796                 if (bond->curr_active_slave)
797                         dev_mc_del(bond->curr_active_slave->dev, addr);
798         } else {
799                 struct slave *slave;
800                 int i;
801                 bond_for_each_slave(bond, slave, i) {
802                         dev_mc_del(slave->dev, addr);
803                 }
804         }
805 }
806
807
808 static void __bond_resend_igmp_join_requests(struct net_device *dev)
809 {
810         struct in_device *in_dev;
811
812         rcu_read_lock();
813         in_dev = __in_dev_get_rcu(dev);
814         if (in_dev)
815                 ip_mc_rejoin_groups(in_dev);
816         rcu_read_unlock();
817 }
818
819 /*
820  * Retrieve the list of registered multicast addresses for the bonding
821  * device and retransmit an IGMP JOIN request to the current active
822  * slave.
823  */
824 static void bond_resend_igmp_join_requests(struct bonding *bond)
825 {
826         struct net_device *vlan_dev;
827         struct vlan_entry *vlan;
828
829         read_lock(&bond->lock);
830
831         /* rejoin all groups on bond device */
832         __bond_resend_igmp_join_requests(bond->dev);
833
834         /* rejoin all groups on vlan devices */
835         if (bond->vlgrp) {
836                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
837                         vlan_dev = vlan_group_get_device(bond->vlgrp,
838                                                          vlan->vlan_id);
839                         if (vlan_dev)
840                                 __bond_resend_igmp_join_requests(vlan_dev);
841                 }
842         }
843
844         if (--bond->igmp_retrans > 0)
845                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
846
847         read_unlock(&bond->lock);
848 }
849
850 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
851 {
852         struct bonding *bond = container_of(work, struct bonding,
853                                             mcast_work.work);
854         bond_resend_igmp_join_requests(bond);
855 }
856
857 /*
858  * flush all members of flush->mc_list from device dev->mc_list
859  */
860 static void bond_mc_list_flush(struct net_device *bond_dev,
861                                struct net_device *slave_dev)
862 {
863         struct bonding *bond = netdev_priv(bond_dev);
864         struct netdev_hw_addr *ha;
865
866         netdev_for_each_mc_addr(ha, bond_dev)
867                 dev_mc_del(slave_dev, ha->addr);
868
869         if (bond->params.mode == BOND_MODE_8023AD) {
870                 /* del lacpdu mc addr from mc list */
871                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
872
873                 dev_mc_del(slave_dev, lacpdu_multicast);
874         }
875 }
876
877 /*--------------------------- Active slave change ---------------------------*/
878
879 /*
880  * Update the mc list and multicast-related flags for the new and
881  * old active slaves (if any) according to the multicast mode, and
882  * promiscuous flags unconditionally.
883  */
884 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
885                          struct slave *old_active)
886 {
887         struct netdev_hw_addr *ha;
888
889         if (!USES_PRIMARY(bond->params.mode))
890                 /* nothing to do -  mc list is already up-to-date on
891                  * all slaves
892                  */
893                 return;
894
895         if (old_active) {
896                 if (bond->dev->flags & IFF_PROMISC)
897                         dev_set_promiscuity(old_active->dev, -1);
898
899                 if (bond->dev->flags & IFF_ALLMULTI)
900                         dev_set_allmulti(old_active->dev, -1);
901
902                 netdev_for_each_mc_addr(ha, bond->dev)
903                         dev_mc_del(old_active->dev, ha->addr);
904         }
905
906         if (new_active) {
907                 /* FIXME: Signal errors upstream. */
908                 if (bond->dev->flags & IFF_PROMISC)
909                         dev_set_promiscuity(new_active->dev, 1);
910
911                 if (bond->dev->flags & IFF_ALLMULTI)
912                         dev_set_allmulti(new_active->dev, 1);
913
914                 netdev_for_each_mc_addr(ha, bond->dev)
915                         dev_mc_add(new_active->dev, ha->addr);
916         }
917 }
918
919 /*
920  * bond_do_fail_over_mac
921  *
922  * Perform special MAC address swapping for fail_over_mac settings
923  *
924  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
925  */
926 static void bond_do_fail_over_mac(struct bonding *bond,
927                                   struct slave *new_active,
928                                   struct slave *old_active)
929         __releases(&bond->curr_slave_lock)
930         __releases(&bond->lock)
931         __acquires(&bond->lock)
932         __acquires(&bond->curr_slave_lock)
933 {
934         u8 tmp_mac[ETH_ALEN];
935         struct sockaddr saddr;
936         int rv;
937
938         switch (bond->params.fail_over_mac) {
939         case BOND_FOM_ACTIVE:
940                 if (new_active)
941                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
942                                new_active->dev->addr_len);
943                 break;
944         case BOND_FOM_FOLLOW:
945                 /*
946                  * if new_active && old_active, swap them
947                  * if just old_active, do nothing (going to no active slave)
948                  * if just new_active, set new_active to bond's MAC
949                  */
950                 if (!new_active)
951                         return;
952
953                 write_unlock_bh(&bond->curr_slave_lock);
954                 read_unlock(&bond->lock);
955
956                 if (old_active) {
957                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
958                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
959                                ETH_ALEN);
960                         saddr.sa_family = new_active->dev->type;
961                 } else {
962                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
963                         saddr.sa_family = bond->dev->type;
964                 }
965
966                 rv = dev_set_mac_address(new_active->dev, &saddr);
967                 if (rv) {
968                         pr_err("%s: Error %d setting MAC of slave %s\n",
969                                bond->dev->name, -rv, new_active->dev->name);
970                         goto out;
971                 }
972
973                 if (!old_active)
974                         goto out;
975
976                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
977                 saddr.sa_family = old_active->dev->type;
978
979                 rv = dev_set_mac_address(old_active->dev, &saddr);
980                 if (rv)
981                         pr_err("%s: Error %d setting MAC of slave %s\n",
982                                bond->dev->name, -rv, new_active->dev->name);
983 out:
984                 read_lock(&bond->lock);
985                 write_lock_bh(&bond->curr_slave_lock);
986                 break;
987         default:
988                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
989                        bond->dev->name, bond->params.fail_over_mac);
990                 break;
991         }
992
993 }
994
995 static bool bond_should_change_active(struct bonding *bond)
996 {
997         struct slave *prim = bond->primary_slave;
998         struct slave *curr = bond->curr_active_slave;
999
1000         if (!prim || !curr || curr->link != BOND_LINK_UP)
1001                 return true;
1002         if (bond->force_primary) {
1003                 bond->force_primary = false;
1004                 return true;
1005         }
1006         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1007             (prim->speed < curr->speed ||
1008              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1009                 return false;
1010         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1011                 return false;
1012         return true;
1013 }
1014
1015 /**
1016  * find_best_interface - select the best available slave to be the active one
1017  * @bond: our bonding struct
1018  *
1019  * Warning: Caller must hold curr_slave_lock for writing.
1020  */
1021 static struct slave *bond_find_best_slave(struct bonding *bond)
1022 {
1023         struct slave *new_active, *old_active;
1024         struct slave *bestslave = NULL;
1025         int mintime = bond->params.updelay;
1026         int i;
1027
1028         new_active = bond->curr_active_slave;
1029
1030         if (!new_active) { /* there were no active slaves left */
1031                 if (bond->slave_cnt > 0)   /* found one slave */
1032                         new_active = bond->first_slave;
1033                 else
1034                         return NULL; /* still no slave, return NULL */
1035         }
1036
1037         if ((bond->primary_slave) &&
1038             bond->primary_slave->link == BOND_LINK_UP &&
1039             bond_should_change_active(bond)) {
1040                 new_active = bond->primary_slave;
1041         }
1042
1043         /* remember where to stop iterating over the slaves */
1044         old_active = new_active;
1045
1046         bond_for_each_slave_from(bond, new_active, i, old_active) {
1047                 if (new_active->link == BOND_LINK_UP) {
1048                         return new_active;
1049                 } else if (new_active->link == BOND_LINK_BACK &&
1050                            IS_UP(new_active->dev)) {
1051                         /* link up, but waiting for stabilization */
1052                         if (new_active->delay < mintime) {
1053                                 mintime = new_active->delay;
1054                                 bestslave = new_active;
1055                         }
1056                 }
1057         }
1058
1059         return bestslave;
1060 }
1061
1062 static bool bond_should_notify_peers(struct bonding *bond)
1063 {
1064         struct slave *slave = bond->curr_active_slave;
1065
1066         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1067                  bond->dev->name, slave ? slave->dev->name : "NULL");
1068
1069         if (!slave || !bond->send_peer_notif ||
1070             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1071                 return false;
1072
1073         bond->send_peer_notif--;
1074         return true;
1075 }
1076
1077 /**
1078  * change_active_interface - change the active slave into the specified one
1079  * @bond: our bonding struct
1080  * @new: the new slave to make the active one
1081  *
1082  * Set the new slave to the bond's settings and unset them on the old
1083  * curr_active_slave.
1084  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1085  *
1086  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1087  * because it is apparently the best available slave we have, even though its
1088  * updelay hasn't timed out yet.
1089  *
1090  * If new_active is not NULL, caller must hold bond->lock for read and
1091  * curr_slave_lock for write_bh.
1092  */
1093 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1094 {
1095         struct slave *old_active = bond->curr_active_slave;
1096
1097         if (old_active == new_active)
1098                 return;
1099
1100         if (new_active) {
1101                 new_active->jiffies = jiffies;
1102
1103                 if (new_active->link == BOND_LINK_BACK) {
1104                         if (USES_PRIMARY(bond->params.mode)) {
1105                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1106                                         bond->dev->name, new_active->dev->name,
1107                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1108                         }
1109
1110                         new_active->delay = 0;
1111                         new_active->link = BOND_LINK_UP;
1112
1113                         if (bond->params.mode == BOND_MODE_8023AD)
1114                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1115
1116                         if (bond_is_lb(bond))
1117                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1118                 } else {
1119                         if (USES_PRIMARY(bond->params.mode)) {
1120                                 pr_info("%s: making interface %s the new active one.\n",
1121                                         bond->dev->name, new_active->dev->name);
1122                         }
1123                 }
1124         }
1125
1126         if (USES_PRIMARY(bond->params.mode))
1127                 bond_mc_swap(bond, new_active, old_active);
1128
1129         if (bond_is_lb(bond)) {
1130                 bond_alb_handle_active_change(bond, new_active);
1131                 if (old_active)
1132                         bond_set_slave_inactive_flags(old_active);
1133                 if (new_active)
1134                         bond_set_slave_active_flags(new_active);
1135         } else {
1136                 bond->curr_active_slave = new_active;
1137         }
1138
1139         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1140                 if (old_active)
1141                         bond_set_slave_inactive_flags(old_active);
1142
1143                 if (new_active) {
1144                         bool should_notify_peers = false;
1145
1146                         bond_set_slave_active_flags(new_active);
1147
1148                         if (bond->params.fail_over_mac)
1149                                 bond_do_fail_over_mac(bond, new_active,
1150                                                       old_active);
1151
1152                         if (netif_running(bond->dev)) {
1153                                 bond->send_peer_notif =
1154                                         bond->params.num_peer_notif;
1155                                 should_notify_peers =
1156                                         bond_should_notify_peers(bond);
1157                         }
1158
1159                         write_unlock_bh(&bond->curr_slave_lock);
1160                         read_unlock(&bond->lock);
1161
1162                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1163                         if (should_notify_peers)
1164                                 netdev_bonding_change(bond->dev,
1165                                                       NETDEV_NOTIFY_PEERS);
1166
1167                         read_lock(&bond->lock);
1168                         write_lock_bh(&bond->curr_slave_lock);
1169                 }
1170         }
1171
1172         /* resend IGMP joins since active slave has changed or
1173          * all were sent on curr_active_slave.
1174          * resend only if bond is brought up with the affected
1175          * bonding modes and the retransmission is enabled */
1176         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1177             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1178              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1179                 bond->igmp_retrans = bond->params.resend_igmp;
1180                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1181         }
1182 }
1183
1184 /**
1185  * bond_select_active_slave - select a new active slave, if needed
1186  * @bond: our bonding struct
1187  *
1188  * This functions should be called when one of the following occurs:
1189  * - The old curr_active_slave has been released or lost its link.
1190  * - The primary_slave has got its link back.
1191  * - A slave has got its link back and there's no old curr_active_slave.
1192  *
1193  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1194  */
1195 void bond_select_active_slave(struct bonding *bond)
1196 {
1197         struct slave *best_slave;
1198         int rv;
1199
1200         best_slave = bond_find_best_slave(bond);
1201         if (best_slave != bond->curr_active_slave) {
1202                 bond_change_active_slave(bond, best_slave);
1203                 rv = bond_set_carrier(bond);
1204                 if (!rv)
1205                         return;
1206
1207                 if (netif_carrier_ok(bond->dev)) {
1208                         pr_info("%s: first active interface up!\n",
1209                                 bond->dev->name);
1210                 } else {
1211                         pr_info("%s: now running without any active interface !\n",
1212                                 bond->dev->name);
1213                 }
1214         }
1215 }
1216
1217 /*--------------------------- slave list handling ---------------------------*/
1218
1219 /*
1220  * This function attaches the slave to the end of list.
1221  *
1222  * bond->lock held for writing by caller.
1223  */
1224 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1225 {
1226         if (bond->first_slave == NULL) { /* attaching the first slave */
1227                 new_slave->next = new_slave;
1228                 new_slave->prev = new_slave;
1229                 bond->first_slave = new_slave;
1230         } else {
1231                 new_slave->next = bond->first_slave;
1232                 new_slave->prev = bond->first_slave->prev;
1233                 new_slave->next->prev = new_slave;
1234                 new_slave->prev->next = new_slave;
1235         }
1236
1237         bond->slave_cnt++;
1238 }
1239
1240 /*
1241  * This function detaches the slave from the list.
1242  * WARNING: no check is made to verify if the slave effectively
1243  * belongs to <bond>.
1244  * Nothing is freed on return, structures are just unchained.
1245  * If any slave pointer in bond was pointing to <slave>,
1246  * it should be changed by the calling function.
1247  *
1248  * bond->lock held for writing by caller.
1249  */
1250 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1251 {
1252         if (slave->next)
1253                 slave->next->prev = slave->prev;
1254
1255         if (slave->prev)
1256                 slave->prev->next = slave->next;
1257
1258         if (bond->first_slave == slave) { /* slave is the first slave */
1259                 if (bond->slave_cnt > 1) { /* there are more slave */
1260                         bond->first_slave = slave->next;
1261                 } else {
1262                         bond->first_slave = NULL; /* slave was the last one */
1263                 }
1264         }
1265
1266         slave->next = NULL;
1267         slave->prev = NULL;
1268         bond->slave_cnt--;
1269 }
1270
1271 #ifdef CONFIG_NET_POLL_CONTROLLER
1272 static inline int slave_enable_netpoll(struct slave *slave)
1273 {
1274         struct netpoll *np;
1275         int err = 0;
1276
1277         np = kzalloc(sizeof(*np), GFP_KERNEL);
1278         err = -ENOMEM;
1279         if (!np)
1280                 goto out;
1281
1282         np->dev = slave->dev;
1283         strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1284         err = __netpoll_setup(np);
1285         if (err) {
1286                 kfree(np);
1287                 goto out;
1288         }
1289         slave->np = np;
1290 out:
1291         return err;
1292 }
1293 static inline void slave_disable_netpoll(struct slave *slave)
1294 {
1295         struct netpoll *np = slave->np;
1296
1297         if (!np)
1298                 return;
1299
1300         slave->np = NULL;
1301         synchronize_rcu_bh();
1302         __netpoll_cleanup(np);
1303         kfree(np);
1304 }
1305 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1306 {
1307         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1308                 return false;
1309         if (!slave_dev->netdev_ops->ndo_poll_controller)
1310                 return false;
1311         return true;
1312 }
1313
1314 static void bond_poll_controller(struct net_device *bond_dev)
1315 {
1316 }
1317
1318 static void __bond_netpoll_cleanup(struct bonding *bond)
1319 {
1320         struct slave *slave;
1321         int i;
1322
1323         bond_for_each_slave(bond, slave, i)
1324                 if (IS_UP(slave->dev))
1325                         slave_disable_netpoll(slave);
1326 }
1327 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1328 {
1329         struct bonding *bond = netdev_priv(bond_dev);
1330
1331         read_lock(&bond->lock);
1332         __bond_netpoll_cleanup(bond);
1333         read_unlock(&bond->lock);
1334 }
1335
1336 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1337 {
1338         struct bonding *bond = netdev_priv(dev);
1339         struct slave *slave;
1340         int i, err = 0;
1341
1342         read_lock(&bond->lock);
1343         bond_for_each_slave(bond, slave, i) {
1344                 err = slave_enable_netpoll(slave);
1345                 if (err) {
1346                         __bond_netpoll_cleanup(bond);
1347                         break;
1348                 }
1349         }
1350         read_unlock(&bond->lock);
1351         return err;
1352 }
1353
1354 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1355 {
1356         return bond->dev->npinfo;
1357 }
1358
1359 #else
1360 static inline int slave_enable_netpoll(struct slave *slave)
1361 {
1362         return 0;
1363 }
1364 static inline void slave_disable_netpoll(struct slave *slave)
1365 {
1366 }
1367 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1368 {
1369 }
1370 #endif
1371
1372 /*---------------------------------- IOCTL ----------------------------------*/
1373
1374 static int bond_sethwaddr(struct net_device *bond_dev,
1375                           struct net_device *slave_dev)
1376 {
1377         pr_debug("bond_dev=%p\n", bond_dev);
1378         pr_debug("slave_dev=%p\n", slave_dev);
1379         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1380         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1381         return 0;
1382 }
1383
1384 static u32 bond_fix_features(struct net_device *dev, u32 features)
1385 {
1386         struct slave *slave;
1387         struct bonding *bond = netdev_priv(dev);
1388         u32 mask;
1389         int i;
1390
1391         read_lock(&bond->lock);
1392
1393         if (!bond->first_slave) {
1394                 /* Disable adding VLANs to empty bond. But why? --mq */
1395                 features |= NETIF_F_VLAN_CHALLENGED;
1396                 goto out;
1397         }
1398
1399         mask = features;
1400         features &= ~NETIF_F_ONE_FOR_ALL;
1401         features |= NETIF_F_ALL_FOR_ALL;
1402
1403         bond_for_each_slave(bond, slave, i) {
1404                 features = netdev_increment_features(features,
1405                                                      slave->dev->features,
1406                                                      mask);
1407         }
1408
1409 out:
1410         read_unlock(&bond->lock);
1411         return features;
1412 }
1413
1414 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_TX_OFFLOADS | \
1415                                  NETIF_F_SOFT_FEATURES | \
1416                                  NETIF_F_LRO)
1417
1418 static void bond_compute_features(struct bonding *bond)
1419 {
1420         struct slave *slave;
1421         struct net_device *bond_dev = bond->dev;
1422         u32 vlan_features = BOND_VLAN_FEATURES;
1423         unsigned short max_hard_header_len = ETH_HLEN;
1424         int i;
1425
1426         read_lock(&bond->lock);
1427
1428         if (!bond->first_slave)
1429                 goto done;
1430
1431         bond_for_each_slave(bond, slave, i) {
1432                 vlan_features = netdev_increment_features(vlan_features,
1433                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1434
1435                 if (slave->dev->hard_header_len > max_hard_header_len)
1436                         max_hard_header_len = slave->dev->hard_header_len;
1437         }
1438
1439 done:
1440         bond_dev->vlan_features = vlan_features;
1441         bond_dev->hard_header_len = max_hard_header_len;
1442
1443         read_unlock(&bond->lock);
1444
1445         netdev_change_features(bond_dev);
1446 }
1447
1448 static void bond_setup_by_slave(struct net_device *bond_dev,
1449                                 struct net_device *slave_dev)
1450 {
1451         struct bonding *bond = netdev_priv(bond_dev);
1452
1453         bond_dev->header_ops        = slave_dev->header_ops;
1454
1455         bond_dev->type              = slave_dev->type;
1456         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1457         bond_dev->addr_len          = slave_dev->addr_len;
1458
1459         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1460                 slave_dev->addr_len);
1461         bond->setup_by_slave = 1;
1462 }
1463
1464 /* On bonding slaves other than the currently active slave, suppress
1465  * duplicates except for alb non-mcast/bcast.
1466  */
1467 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1468                                             struct slave *slave,
1469                                             struct bonding *bond)
1470 {
1471         if (bond_is_slave_inactive(slave)) {
1472                 if (bond->params.mode == BOND_MODE_ALB &&
1473                     skb->pkt_type != PACKET_BROADCAST &&
1474                     skb->pkt_type != PACKET_MULTICAST)
1475                         return false;
1476                 return true;
1477         }
1478         return false;
1479 }
1480
1481 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1482 {
1483         struct sk_buff *skb = *pskb;
1484         struct slave *slave;
1485         struct bonding *bond;
1486
1487         skb = skb_share_check(skb, GFP_ATOMIC);
1488         if (unlikely(!skb))
1489                 return RX_HANDLER_CONSUMED;
1490
1491         *pskb = skb;
1492
1493         slave = bond_slave_get_rcu(skb->dev);
1494         bond = slave->bond;
1495
1496         if (bond->params.arp_interval)
1497                 slave->dev->last_rx = jiffies;
1498
1499         if (bond->recv_probe) {
1500                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1501
1502                 if (likely(nskb)) {
1503                         bond->recv_probe(nskb, bond, slave);
1504                         dev_kfree_skb(nskb);
1505                 }
1506         }
1507
1508         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1509                 return RX_HANDLER_EXACT;
1510         }
1511
1512         skb->dev = bond->dev;
1513
1514         if (bond->params.mode == BOND_MODE_ALB &&
1515             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1516             skb->pkt_type == PACKET_HOST) {
1517
1518                 if (unlikely(skb_cow_head(skb,
1519                                           skb->data - skb_mac_header(skb)))) {
1520                         kfree_skb(skb);
1521                         return RX_HANDLER_CONSUMED;
1522                 }
1523                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1524         }
1525
1526         return RX_HANDLER_ANOTHER;
1527 }
1528
1529 /* enslave device <slave> to bond device <master> */
1530 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1531 {
1532         struct bonding *bond = netdev_priv(bond_dev);
1533         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1534         struct slave *new_slave = NULL;
1535         struct netdev_hw_addr *ha;
1536         struct sockaddr addr;
1537         int link_reporting;
1538         int res = 0;
1539
1540         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1541                 slave_ops->ndo_do_ioctl == NULL) {
1542                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1543                            bond_dev->name, slave_dev->name);
1544         }
1545
1546         /* already enslaved */
1547         if (slave_dev->flags & IFF_SLAVE) {
1548                 pr_debug("Error, Device was already enslaved\n");
1549                 return -EBUSY;
1550         }
1551
1552         /* vlan challenged mutual exclusion */
1553         /* no need to lock since we're protected by rtnl_lock */
1554         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1555                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1556                 if (bond->vlgrp) {
1557                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1558                                bond_dev->name, slave_dev->name, bond_dev->name);
1559                         return -EPERM;
1560                 } else {
1561                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1562                                    bond_dev->name, slave_dev->name,
1563                                    slave_dev->name, bond_dev->name);
1564                 }
1565         } else {
1566                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1567         }
1568
1569         /*
1570          * Old ifenslave binaries are no longer supported.  These can
1571          * be identified with moderate accuracy by the state of the slave:
1572          * the current ifenslave will set the interface down prior to
1573          * enslaving it; the old ifenslave will not.
1574          */
1575         if ((slave_dev->flags & IFF_UP)) {
1576                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1577                        slave_dev->name);
1578                 res = -EPERM;
1579                 goto err_undo_flags;
1580         }
1581
1582         /* set bonding device ether type by slave - bonding netdevices are
1583          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1584          * there is a need to override some of the type dependent attribs/funcs.
1585          *
1586          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1587          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1588          */
1589         if (bond->slave_cnt == 0) {
1590                 if (bond_dev->type != slave_dev->type) {
1591                         pr_debug("%s: change device type from %d to %d\n",
1592                                  bond_dev->name,
1593                                  bond_dev->type, slave_dev->type);
1594
1595                         res = netdev_bonding_change(bond_dev,
1596                                                     NETDEV_PRE_TYPE_CHANGE);
1597                         res = notifier_to_errno(res);
1598                         if (res) {
1599                                 pr_err("%s: refused to change device type\n",
1600                                        bond_dev->name);
1601                                 res = -EBUSY;
1602                                 goto err_undo_flags;
1603                         }
1604
1605                         /* Flush unicast and multicast addresses */
1606                         dev_uc_flush(bond_dev);
1607                         dev_mc_flush(bond_dev);
1608
1609                         if (slave_dev->type != ARPHRD_ETHER)
1610                                 bond_setup_by_slave(bond_dev, slave_dev);
1611                         else
1612                                 ether_setup(bond_dev);
1613
1614                         netdev_bonding_change(bond_dev,
1615                                               NETDEV_POST_TYPE_CHANGE);
1616                 }
1617         } else if (bond_dev->type != slave_dev->type) {
1618                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1619                        slave_dev->name,
1620                        slave_dev->type, bond_dev->type);
1621                 res = -EINVAL;
1622                 goto err_undo_flags;
1623         }
1624
1625         if (slave_ops->ndo_set_mac_address == NULL) {
1626                 if (bond->slave_cnt == 0) {
1627                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1628                                    bond_dev->name);
1629                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1630                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1631                         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",
1632                                bond_dev->name);
1633                         res = -EOPNOTSUPP;
1634                         goto err_undo_flags;
1635                 }
1636         }
1637
1638         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1639
1640         /* If this is the first slave, then we need to set the master's hardware
1641          * address to be the same as the slave's. */
1642         if (is_zero_ether_addr(bond->dev->dev_addr))
1643                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1644                        slave_dev->addr_len);
1645
1646
1647         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1648         if (!new_slave) {
1649                 res = -ENOMEM;
1650                 goto err_undo_flags;
1651         }
1652
1653         /*
1654          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1655          * is set via sysfs or module option if desired.
1656          */
1657         new_slave->queue_id = 0;
1658
1659         /* Save slave's original mtu and then set it to match the bond */
1660         new_slave->original_mtu = slave_dev->mtu;
1661         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1662         if (res) {
1663                 pr_debug("Error %d calling dev_set_mtu\n", res);
1664                 goto err_free;
1665         }
1666
1667         /*
1668          * Save slave's original ("permanent") mac address for modes
1669          * that need it, and for restoring it upon release, and then
1670          * set it to the master's address
1671          */
1672         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1673
1674         if (!bond->params.fail_over_mac) {
1675                 /*
1676                  * Set slave to master's mac address.  The application already
1677                  * set the master's mac address to that of the first slave
1678                  */
1679                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1680                 addr.sa_family = slave_dev->type;
1681                 res = dev_set_mac_address(slave_dev, &addr);
1682                 if (res) {
1683                         pr_debug("Error %d calling set_mac_address\n", res);
1684                         goto err_restore_mtu;
1685                 }
1686         }
1687
1688         res = netdev_set_bond_master(slave_dev, bond_dev);
1689         if (res) {
1690                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1691                 goto err_restore_mac;
1692         }
1693
1694         /* open the slave since the application closed it */
1695         res = dev_open(slave_dev);
1696         if (res) {
1697                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1698                 goto err_unset_master;
1699         }
1700
1701         new_slave->bond = bond;
1702         new_slave->dev = slave_dev;
1703         slave_dev->priv_flags |= IFF_BONDING;
1704
1705         if (bond_is_lb(bond)) {
1706                 /* bond_alb_init_slave() must be called before all other stages since
1707                  * it might fail and we do not want to have to undo everything
1708                  */
1709                 res = bond_alb_init_slave(bond, new_slave);
1710                 if (res)
1711                         goto err_close;
1712         }
1713
1714         /* If the mode USES_PRIMARY, then the new slave gets the
1715          * master's promisc (and mc) settings only if it becomes the
1716          * curr_active_slave, and that is taken care of later when calling
1717          * bond_change_active()
1718          */
1719         if (!USES_PRIMARY(bond->params.mode)) {
1720                 /* set promiscuity level to new slave */
1721                 if (bond_dev->flags & IFF_PROMISC) {
1722                         res = dev_set_promiscuity(slave_dev, 1);
1723                         if (res)
1724                                 goto err_close;
1725                 }
1726
1727                 /* set allmulti level to new slave */
1728                 if (bond_dev->flags & IFF_ALLMULTI) {
1729                         res = dev_set_allmulti(slave_dev, 1);
1730                         if (res)
1731                                 goto err_close;
1732                 }
1733
1734                 netif_addr_lock_bh(bond_dev);
1735                 /* upload master's mc_list to new slave */
1736                 netdev_for_each_mc_addr(ha, bond_dev)
1737                         dev_mc_add(slave_dev, ha->addr);
1738                 netif_addr_unlock_bh(bond_dev);
1739         }
1740
1741         if (bond->params.mode == BOND_MODE_8023AD) {
1742                 /* add lacpdu mc addr to mc list */
1743                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1744
1745                 dev_mc_add(slave_dev, lacpdu_multicast);
1746         }
1747
1748         bond_add_vlans_on_slave(bond, slave_dev);
1749
1750         write_lock_bh(&bond->lock);
1751
1752         bond_attach_slave(bond, new_slave);
1753
1754         new_slave->delay = 0;
1755         new_slave->link_failure_count = 0;
1756
1757         write_unlock_bh(&bond->lock);
1758
1759         bond_compute_features(bond);
1760
1761         read_lock(&bond->lock);
1762
1763         new_slave->last_arp_rx = jiffies;
1764
1765         if (bond->params.miimon && !bond->params.use_carrier) {
1766                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1767
1768                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1769                         /*
1770                          * miimon is set but a bonded network driver
1771                          * does not support ETHTOOL/MII and
1772                          * arp_interval is not set.  Note: if
1773                          * use_carrier is enabled, we will never go
1774                          * here (because netif_carrier is always
1775                          * supported); thus, we don't need to change
1776                          * the messages for netif_carrier.
1777                          */
1778                         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",
1779                                bond_dev->name, slave_dev->name);
1780                 } else if (link_reporting == -1) {
1781                         /* unable get link status using mii/ethtool */
1782                         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",
1783                                    bond_dev->name, slave_dev->name);
1784                 }
1785         }
1786
1787         /* check for initial state */
1788         if (!bond->params.miimon ||
1789             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1790                 if (bond->params.updelay) {
1791                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1792                         new_slave->link  = BOND_LINK_BACK;
1793                         new_slave->delay = bond->params.updelay;
1794                 } else {
1795                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1796                         new_slave->link  = BOND_LINK_UP;
1797                 }
1798                 new_slave->jiffies = jiffies;
1799         } else {
1800                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1801                 new_slave->link  = BOND_LINK_DOWN;
1802         }
1803
1804         if (bond_update_speed_duplex(new_slave) &&
1805             (new_slave->link != BOND_LINK_DOWN)) {
1806                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1807                            bond_dev->name, new_slave->dev->name);
1808
1809                 if (bond->params.mode == BOND_MODE_8023AD) {
1810                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1811                                    bond_dev->name);
1812                 }
1813         }
1814
1815         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1816                 /* if there is a primary slave, remember it */
1817                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1818                         bond->primary_slave = new_slave;
1819                         bond->force_primary = true;
1820                 }
1821         }
1822
1823         write_lock_bh(&bond->curr_slave_lock);
1824
1825         switch (bond->params.mode) {
1826         case BOND_MODE_ACTIVEBACKUP:
1827                 bond_set_slave_inactive_flags(new_slave);
1828                 bond_select_active_slave(bond);
1829                 break;
1830         case BOND_MODE_8023AD:
1831                 /* in 802.3ad mode, the internal mechanism
1832                  * will activate the slaves in the selected
1833                  * aggregator
1834                  */
1835                 bond_set_slave_inactive_flags(new_slave);
1836                 /* if this is the first slave */
1837                 if (bond->slave_cnt == 1) {
1838                         SLAVE_AD_INFO(new_slave).id = 1;
1839                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1840                          * can be called only after the mac address of the bond is set
1841                          */
1842                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1843                 } else {
1844                         SLAVE_AD_INFO(new_slave).id =
1845                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1846                 }
1847
1848                 bond_3ad_bind_slave(new_slave);
1849                 break;
1850         case BOND_MODE_TLB:
1851         case BOND_MODE_ALB:
1852                 bond_set_active_slave(new_slave);
1853                 bond_set_slave_inactive_flags(new_slave);
1854                 bond_select_active_slave(bond);
1855                 break;
1856         default:
1857                 pr_debug("This slave is always active in trunk mode\n");
1858
1859                 /* always active in trunk mode */
1860                 bond_set_active_slave(new_slave);
1861
1862                 /* In trunking mode there is little meaning to curr_active_slave
1863                  * anyway (it holds no special properties of the bond device),
1864                  * so we can change it without calling change_active_interface()
1865                  */
1866                 if (!bond->curr_active_slave)
1867                         bond->curr_active_slave = new_slave;
1868
1869                 break;
1870         } /* switch(bond_mode) */
1871
1872         write_unlock_bh(&bond->curr_slave_lock);
1873
1874         bond_set_carrier(bond);
1875
1876 #ifdef CONFIG_NET_POLL_CONTROLLER
1877         slave_dev->npinfo = bond_netpoll_info(bond);
1878         if (slave_dev->npinfo) {
1879                 if (slave_enable_netpoll(new_slave)) {
1880                         read_unlock(&bond->lock);
1881                         pr_info("Error, %s: master_dev is using netpoll, "
1882                                  "but new slave device does not support netpoll.\n",
1883                                  bond_dev->name);
1884                         res = -EBUSY;
1885                         goto err_close;
1886                 }
1887         }
1888 #endif
1889
1890         read_unlock(&bond->lock);
1891
1892         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1893         if (res)
1894                 goto err_close;
1895
1896         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1897                                          new_slave);
1898         if (res) {
1899                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1900                 goto err_dest_symlinks;
1901         }
1902
1903         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1904                 bond_dev->name, slave_dev->name,
1905                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1906                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1907
1908         /* enslave is successful */
1909         return 0;
1910
1911 /* Undo stages on error */
1912 err_dest_symlinks:
1913         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1914
1915 err_close:
1916         dev_close(slave_dev);
1917
1918 err_unset_master:
1919         netdev_set_bond_master(slave_dev, NULL);
1920
1921 err_restore_mac:
1922         if (!bond->params.fail_over_mac) {
1923                 /* XXX TODO - fom follow mode needs to change master's
1924                  * MAC if this slave's MAC is in use by the bond, or at
1925                  * least print a warning.
1926                  */
1927                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1928                 addr.sa_family = slave_dev->type;
1929                 dev_set_mac_address(slave_dev, &addr);
1930         }
1931
1932 err_restore_mtu:
1933         dev_set_mtu(slave_dev, new_slave->original_mtu);
1934
1935 err_free:
1936         kfree(new_slave);
1937
1938 err_undo_flags:
1939         bond_compute_features(bond);
1940
1941         return res;
1942 }
1943
1944 /*
1945  * Try to release the slave device <slave> from the bond device <master>
1946  * It is legal to access curr_active_slave without a lock because all the function
1947  * is write-locked.
1948  *
1949  * The rules for slave state should be:
1950  *   for Active/Backup:
1951  *     Active stays on all backups go down
1952  *   for Bonded connections:
1953  *     The first up interface should be left on and all others downed.
1954  */
1955 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1956 {
1957         struct bonding *bond = netdev_priv(bond_dev);
1958         struct slave *slave, *oldcurrent;
1959         struct sockaddr addr;
1960         u32 old_features = bond_dev->features;
1961
1962         /* slave is not a slave or master is not master of this slave */
1963         if (!(slave_dev->flags & IFF_SLAVE) ||
1964             (slave_dev->master != bond_dev)) {
1965                 pr_err("%s: Error: cannot release %s.\n",
1966                        bond_dev->name, slave_dev->name);
1967                 return -EINVAL;
1968         }
1969
1970         block_netpoll_tx();
1971         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1972         write_lock_bh(&bond->lock);
1973
1974         slave = bond_get_slave_by_dev(bond, slave_dev);
1975         if (!slave) {
1976                 /* not a slave of this bond */
1977                 pr_info("%s: %s not enslaved\n",
1978                         bond_dev->name, slave_dev->name);
1979                 write_unlock_bh(&bond->lock);
1980                 unblock_netpoll_tx();
1981                 return -EINVAL;
1982         }
1983
1984         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1985          * for this slave anymore.
1986          */
1987         netdev_rx_handler_unregister(slave_dev);
1988         write_unlock_bh(&bond->lock);
1989         synchronize_net();
1990         write_lock_bh(&bond->lock);
1991
1992         if (!bond->params.fail_over_mac) {
1993                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1994                     bond->slave_cnt > 1)
1995                         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",
1996                                    bond_dev->name, slave_dev->name,
1997                                    slave->perm_hwaddr,
1998                                    bond_dev->name, slave_dev->name);
1999         }
2000
2001         /* Inform AD package of unbinding of slave. */
2002         if (bond->params.mode == BOND_MODE_8023AD) {
2003                 /* must be called before the slave is
2004                  * detached from the list
2005                  */
2006                 bond_3ad_unbind_slave(slave);
2007         }
2008
2009         pr_info("%s: releasing %s interface %s\n",
2010                 bond_dev->name,
2011                 bond_is_active_slave(slave) ? "active" : "backup",
2012                 slave_dev->name);
2013
2014         oldcurrent = bond->curr_active_slave;
2015
2016         bond->current_arp_slave = NULL;
2017
2018         /* release the slave from its bond */
2019         bond_detach_slave(bond, slave);
2020
2021         if (bond->primary_slave == slave)
2022                 bond->primary_slave = NULL;
2023
2024         if (oldcurrent == slave)
2025                 bond_change_active_slave(bond, NULL);
2026
2027         if (bond_is_lb(bond)) {
2028                 /* Must be called only after the slave has been
2029                  * detached from the list and the curr_active_slave
2030                  * has been cleared (if our_slave == old_current),
2031                  * but before a new active slave is selected.
2032                  */
2033                 write_unlock_bh(&bond->lock);
2034                 bond_alb_deinit_slave(bond, slave);
2035                 write_lock_bh(&bond->lock);
2036         }
2037
2038         if (oldcurrent == slave) {
2039                 /*
2040                  * Note that we hold RTNL over this sequence, so there
2041                  * is no concern that another slave add/remove event
2042                  * will interfere.
2043                  */
2044                 write_unlock_bh(&bond->lock);
2045                 read_lock(&bond->lock);
2046                 write_lock_bh(&bond->curr_slave_lock);
2047
2048                 bond_select_active_slave(bond);
2049
2050                 write_unlock_bh(&bond->curr_slave_lock);
2051                 read_unlock(&bond->lock);
2052                 write_lock_bh(&bond->lock);
2053         }
2054
2055         if (bond->slave_cnt == 0) {
2056                 bond_set_carrier(bond);
2057
2058                 /* if the last slave was removed, zero the mac address
2059                  * of the master so it will be set by the application
2060                  * to the mac address of the first slave
2061                  */
2062                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2063
2064                 if (bond->vlgrp) {
2065                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2066                                    bond_dev->name, bond_dev->name);
2067                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2068                                    bond_dev->name);
2069                 }
2070         }
2071
2072         write_unlock_bh(&bond->lock);
2073         unblock_netpoll_tx();
2074
2075         bond_compute_features(bond);
2076         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2077             (old_features & NETIF_F_VLAN_CHALLENGED))
2078                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2079                         bond_dev->name, slave_dev->name, bond_dev->name);
2080
2081         /* must do this from outside any spinlocks */
2082         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2083
2084         bond_del_vlans_from_slave(bond, slave_dev);
2085
2086         /* If the mode USES_PRIMARY, then we should only remove its
2087          * promisc and mc settings if it was the curr_active_slave, but that was
2088          * already taken care of above when we detached the slave
2089          */
2090         if (!USES_PRIMARY(bond->params.mode)) {
2091                 /* unset promiscuity level from slave */
2092                 if (bond_dev->flags & IFF_PROMISC)
2093                         dev_set_promiscuity(slave_dev, -1);
2094
2095                 /* unset allmulti level from slave */
2096                 if (bond_dev->flags & IFF_ALLMULTI)
2097                         dev_set_allmulti(slave_dev, -1);
2098
2099                 /* flush master's mc_list from slave */
2100                 netif_addr_lock_bh(bond_dev);
2101                 bond_mc_list_flush(bond_dev, slave_dev);
2102                 netif_addr_unlock_bh(bond_dev);
2103         }
2104
2105         netdev_set_bond_master(slave_dev, NULL);
2106
2107         slave_disable_netpoll(slave);
2108
2109         /* close slave before restoring its mac address */
2110         dev_close(slave_dev);
2111
2112         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2113                 /* restore original ("permanent") mac address */
2114                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2115                 addr.sa_family = slave_dev->type;
2116                 dev_set_mac_address(slave_dev, &addr);
2117         }
2118
2119         dev_set_mtu(slave_dev, slave->original_mtu);
2120
2121         slave_dev->priv_flags &= ~IFF_BONDING;
2122
2123         kfree(slave);
2124
2125         return 0;  /* deletion OK */
2126 }
2127
2128 /*
2129 * First release a slave and then destroy the bond if no more slaves are left.
2130 * Must be under rtnl_lock when this function is called.
2131 */
2132 static int  bond_release_and_destroy(struct net_device *bond_dev,
2133                                      struct net_device *slave_dev)
2134 {
2135         struct bonding *bond = netdev_priv(bond_dev);
2136         int ret;
2137
2138         ret = bond_release(bond_dev, slave_dev);
2139         if ((ret == 0) && (bond->slave_cnt == 0)) {
2140                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2141                 pr_info("%s: destroying bond %s.\n",
2142                         bond_dev->name, bond_dev->name);
2143                 unregister_netdevice(bond_dev);
2144         }
2145         return ret;
2146 }
2147
2148 /*
2149  * This function releases all slaves.
2150  */
2151 static int bond_release_all(struct net_device *bond_dev)
2152 {
2153         struct bonding *bond = netdev_priv(bond_dev);
2154         struct slave *slave;
2155         struct net_device *slave_dev;
2156         struct sockaddr addr;
2157
2158         write_lock_bh(&bond->lock);
2159
2160         netif_carrier_off(bond_dev);
2161
2162         if (bond->slave_cnt == 0)
2163                 goto out;
2164
2165         bond->current_arp_slave = NULL;
2166         bond->primary_slave = NULL;
2167         bond_change_active_slave(bond, NULL);
2168
2169         while ((slave = bond->first_slave) != NULL) {
2170                 /* Inform AD package of unbinding of slave
2171                  * before slave is detached from the list.
2172                  */
2173                 if (bond->params.mode == BOND_MODE_8023AD)
2174                         bond_3ad_unbind_slave(slave);
2175
2176                 slave_dev = slave->dev;
2177                 bond_detach_slave(bond, slave);
2178
2179                 /* now that the slave is detached, unlock and perform
2180                  * all the undo steps that should not be called from
2181                  * within a lock.
2182                  */
2183                 write_unlock_bh(&bond->lock);
2184
2185                 /* unregister rx_handler early so bond_handle_frame wouldn't
2186                  * be called for this slave anymore.
2187                  */
2188                 netdev_rx_handler_unregister(slave_dev);
2189                 synchronize_net();
2190
2191                 if (bond_is_lb(bond)) {
2192                         /* must be called only after the slave
2193                          * has been detached from the list
2194                          */
2195                         bond_alb_deinit_slave(bond, slave);
2196                 }
2197
2198                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2199                 bond_del_vlans_from_slave(bond, slave_dev);
2200
2201                 /* If the mode USES_PRIMARY, then we should only remove its
2202                  * promisc and mc settings if it was the curr_active_slave, but that was
2203                  * already taken care of above when we detached the slave
2204                  */
2205                 if (!USES_PRIMARY(bond->params.mode)) {
2206                         /* unset promiscuity level from slave */
2207                         if (bond_dev->flags & IFF_PROMISC)
2208                                 dev_set_promiscuity(slave_dev, -1);
2209
2210                         /* unset allmulti level from slave */
2211                         if (bond_dev->flags & IFF_ALLMULTI)
2212                                 dev_set_allmulti(slave_dev, -1);
2213
2214                         /* flush master's mc_list from slave */
2215                         netif_addr_lock_bh(bond_dev);
2216                         bond_mc_list_flush(bond_dev, slave_dev);
2217                         netif_addr_unlock_bh(bond_dev);
2218                 }
2219
2220                 netdev_set_bond_master(slave_dev, NULL);
2221
2222                 slave_disable_netpoll(slave);
2223
2224                 /* close slave before restoring its mac address */
2225                 dev_close(slave_dev);
2226
2227                 if (!bond->params.fail_over_mac) {
2228                         /* restore original ("permanent") mac address*/
2229                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2230                         addr.sa_family = slave_dev->type;
2231                         dev_set_mac_address(slave_dev, &addr);
2232                 }
2233
2234                 kfree(slave);
2235
2236                 /* re-acquire the lock before getting the next slave */
2237                 write_lock_bh(&bond->lock);
2238         }
2239
2240         /* zero the mac address of the master so it will be
2241          * set by the application to the mac address of the
2242          * first slave
2243          */
2244         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2245
2246         if (bond->vlgrp) {
2247                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2248                            bond_dev->name, bond_dev->name);
2249                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2250                            bond_dev->name);
2251         }
2252
2253         pr_info("%s: released all slaves\n", bond_dev->name);
2254
2255 out:
2256         write_unlock_bh(&bond->lock);
2257
2258         bond_compute_features(bond);
2259
2260         return 0;
2261 }
2262
2263 /*
2264  * This function changes the active slave to slave <slave_dev>.
2265  * It returns -EINVAL in the following cases.
2266  *  - <slave_dev> is not found in the list.
2267  *  - There is not active slave now.
2268  *  - <slave_dev> is already active.
2269  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2270  *  - <slave_dev> is not running.
2271  * In these cases, this function does nothing.
2272  * In the other cases, current_slave pointer is changed and 0 is returned.
2273  */
2274 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2275 {
2276         struct bonding *bond = netdev_priv(bond_dev);
2277         struct slave *old_active = NULL;
2278         struct slave *new_active = NULL;
2279         int res = 0;
2280
2281         if (!USES_PRIMARY(bond->params.mode))
2282                 return -EINVAL;
2283
2284         /* Verify that master_dev is indeed the master of slave_dev */
2285         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2286                 return -EINVAL;
2287
2288         read_lock(&bond->lock);
2289
2290         read_lock(&bond->curr_slave_lock);
2291         old_active = bond->curr_active_slave;
2292         read_unlock(&bond->curr_slave_lock);
2293
2294         new_active = bond_get_slave_by_dev(bond, slave_dev);
2295
2296         /*
2297          * Changing to the current active: do nothing; return success.
2298          */
2299         if (new_active && (new_active == old_active)) {
2300                 read_unlock(&bond->lock);
2301                 return 0;
2302         }
2303
2304         if ((new_active) &&
2305             (old_active) &&
2306             (new_active->link == BOND_LINK_UP) &&
2307             IS_UP(new_active->dev)) {
2308                 block_netpoll_tx();
2309                 write_lock_bh(&bond->curr_slave_lock);
2310                 bond_change_active_slave(bond, new_active);
2311                 write_unlock_bh(&bond->curr_slave_lock);
2312                 unblock_netpoll_tx();
2313         } else
2314                 res = -EINVAL;
2315
2316         read_unlock(&bond->lock);
2317
2318         return res;
2319 }
2320
2321 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2322 {
2323         struct bonding *bond = netdev_priv(bond_dev);
2324
2325         info->bond_mode = bond->params.mode;
2326         info->miimon = bond->params.miimon;
2327
2328         read_lock(&bond->lock);
2329         info->num_slaves = bond->slave_cnt;
2330         read_unlock(&bond->lock);
2331
2332         return 0;
2333 }
2334
2335 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2336 {
2337         struct bonding *bond = netdev_priv(bond_dev);
2338         struct slave *slave;
2339         int i, res = -ENODEV;
2340
2341         read_lock(&bond->lock);
2342
2343         bond_for_each_slave(bond, slave, i) {
2344                 if (i == (int)info->slave_id) {
2345                         res = 0;
2346                         strcpy(info->slave_name, slave->dev->name);
2347                         info->link = slave->link;
2348                         info->state = bond_slave_state(slave);
2349                         info->link_failure_count = slave->link_failure_count;
2350                         break;
2351                 }
2352         }
2353
2354         read_unlock(&bond->lock);
2355
2356         return res;
2357 }
2358
2359 /*-------------------------------- Monitoring -------------------------------*/
2360
2361
2362 static int bond_miimon_inspect(struct bonding *bond)
2363 {
2364         struct slave *slave;
2365         int i, link_state, commit = 0;
2366         bool ignore_updelay;
2367
2368         ignore_updelay = !bond->curr_active_slave ? true : false;
2369
2370         bond_for_each_slave(bond, slave, i) {
2371                 slave->new_link = BOND_LINK_NOCHANGE;
2372
2373                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2374
2375                 switch (slave->link) {
2376                 case BOND_LINK_UP:
2377                         if (link_state)
2378                                 continue;
2379
2380                         slave->link = BOND_LINK_FAIL;
2381                         slave->delay = bond->params.downdelay;
2382                         if (slave->delay) {
2383                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2384                                         bond->dev->name,
2385                                         (bond->params.mode ==
2386                                          BOND_MODE_ACTIVEBACKUP) ?
2387                                         (bond_is_active_slave(slave) ?
2388                                          "active " : "backup ") : "",
2389                                         slave->dev->name,
2390                                         bond->params.downdelay * bond->params.miimon);
2391                         }
2392                         /*FALLTHRU*/
2393                 case BOND_LINK_FAIL:
2394                         if (link_state) {
2395                                 /*
2396                                  * recovered before downdelay expired
2397                                  */
2398                                 slave->link = BOND_LINK_UP;
2399                                 slave->jiffies = jiffies;
2400                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2401                                         bond->dev->name,
2402                                         (bond->params.downdelay - slave->delay) *
2403                                         bond->params.miimon,
2404                                         slave->dev->name);
2405                                 continue;
2406                         }
2407
2408                         if (slave->delay <= 0) {
2409                                 slave->new_link = BOND_LINK_DOWN;
2410                                 commit++;
2411                                 continue;
2412                         }
2413
2414                         slave->delay--;
2415                         break;
2416
2417                 case BOND_LINK_DOWN:
2418                         if (!link_state)
2419                                 continue;
2420
2421                         slave->link = BOND_LINK_BACK;
2422                         slave->delay = bond->params.updelay;
2423
2424                         if (slave->delay) {
2425                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2426                                         bond->dev->name, slave->dev->name,
2427                                         ignore_updelay ? 0 :
2428                                         bond->params.updelay *
2429                                         bond->params.miimon);
2430                         }
2431                         /*FALLTHRU*/
2432                 case BOND_LINK_BACK:
2433                         if (!link_state) {
2434                                 slave->link = BOND_LINK_DOWN;
2435                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2436                                         bond->dev->name,
2437                                         (bond->params.updelay - slave->delay) *
2438                                         bond->params.miimon,
2439                                         slave->dev->name);
2440
2441                                 continue;
2442                         }
2443
2444                         if (ignore_updelay)
2445                                 slave->delay = 0;
2446
2447                         if (slave->delay <= 0) {
2448                                 slave->new_link = BOND_LINK_UP;
2449                                 commit++;
2450                                 ignore_updelay = false;
2451                                 continue;
2452                         }
2453
2454                         slave->delay--;
2455                         break;
2456                 }
2457         }
2458
2459         return commit;
2460 }
2461
2462 static void bond_miimon_commit(struct bonding *bond)
2463 {
2464         struct slave *slave;
2465         int i;
2466
2467         bond_for_each_slave(bond, slave, i) {
2468                 switch (slave->new_link) {
2469                 case BOND_LINK_NOCHANGE:
2470                         continue;
2471
2472                 case BOND_LINK_UP:
2473                         slave->link = BOND_LINK_UP;
2474                         slave->jiffies = jiffies;
2475
2476                         if (bond->params.mode == BOND_MODE_8023AD) {
2477                                 /* prevent it from being the active one */
2478                                 bond_set_backup_slave(slave);
2479                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2480                                 /* make it immediately active */
2481                                 bond_set_active_slave(slave);
2482                         } else if (slave != bond->primary_slave) {
2483                                 /* prevent it from being the active one */
2484                                 bond_set_backup_slave(slave);
2485                         }
2486
2487                         bond_update_speed_duplex(slave);
2488
2489                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2490                                 bond->dev->name, slave->dev->name,
2491                                 slave->speed, slave->duplex ? "full" : "half");
2492
2493                         /* notify ad that the link status has changed */
2494                         if (bond->params.mode == BOND_MODE_8023AD)
2495                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2496
2497                         if (bond_is_lb(bond))
2498                                 bond_alb_handle_link_change(bond, slave,
2499                                                             BOND_LINK_UP);
2500
2501                         if (!bond->curr_active_slave ||
2502                             (slave == bond->primary_slave))
2503                                 goto do_failover;
2504
2505                         continue;
2506
2507                 case BOND_LINK_DOWN:
2508                         if (slave->link_failure_count < UINT_MAX)
2509                                 slave->link_failure_count++;
2510
2511                         slave->link = BOND_LINK_DOWN;
2512
2513                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2514                             bond->params.mode == BOND_MODE_8023AD)
2515                                 bond_set_slave_inactive_flags(slave);
2516
2517                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2518                                 bond->dev->name, slave->dev->name);
2519
2520                         if (bond->params.mode == BOND_MODE_8023AD)
2521                                 bond_3ad_handle_link_change(slave,
2522                                                             BOND_LINK_DOWN);
2523
2524                         if (bond_is_lb(bond))
2525                                 bond_alb_handle_link_change(bond, slave,
2526                                                             BOND_LINK_DOWN);
2527
2528                         if (slave == bond->curr_active_slave)
2529                                 goto do_failover;
2530
2531                         continue;
2532
2533                 default:
2534                         pr_err("%s: invalid new link %d on slave %s\n",
2535                                bond->dev->name, slave->new_link,
2536                                slave->dev->name);
2537                         slave->new_link = BOND_LINK_NOCHANGE;
2538
2539                         continue;
2540                 }
2541
2542 do_failover:
2543                 ASSERT_RTNL();
2544                 block_netpoll_tx();
2545                 write_lock_bh(&bond->curr_slave_lock);
2546                 bond_select_active_slave(bond);
2547                 write_unlock_bh(&bond->curr_slave_lock);
2548                 unblock_netpoll_tx();
2549         }
2550
2551         bond_set_carrier(bond);
2552 }
2553
2554 /*
2555  * bond_mii_monitor
2556  *
2557  * Really a wrapper that splits the mii monitor into two phases: an
2558  * inspection, then (if inspection indicates something needs to be done)
2559  * an acquisition of appropriate locks followed by a commit phase to
2560  * implement whatever link state changes are indicated.
2561  */
2562 void bond_mii_monitor(struct work_struct *work)
2563 {
2564         struct bonding *bond = container_of(work, struct bonding,
2565                                             mii_work.work);
2566         bool should_notify_peers = false;
2567
2568         read_lock(&bond->lock);
2569         if (bond->kill_timers)
2570                 goto out;
2571
2572         if (bond->slave_cnt == 0)
2573                 goto re_arm;
2574
2575         should_notify_peers = bond_should_notify_peers(bond);
2576
2577         if (bond_miimon_inspect(bond)) {
2578                 read_unlock(&bond->lock);
2579                 rtnl_lock();
2580                 read_lock(&bond->lock);
2581
2582                 bond_miimon_commit(bond);
2583
2584                 read_unlock(&bond->lock);
2585                 rtnl_unlock();  /* might sleep, hold no other locks */
2586                 read_lock(&bond->lock);
2587         }
2588
2589 re_arm:
2590         if (bond->params.miimon)
2591                 queue_delayed_work(bond->wq, &bond->mii_work,
2592                                    msecs_to_jiffies(bond->params.miimon));
2593 out:
2594         read_unlock(&bond->lock);
2595
2596         if (should_notify_peers) {
2597                 rtnl_lock();
2598                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2599                 rtnl_unlock();
2600         }
2601 }
2602
2603 static __be32 bond_glean_dev_ip(struct net_device *dev)
2604 {
2605         struct in_device *idev;
2606         struct in_ifaddr *ifa;
2607         __be32 addr = 0;
2608
2609         if (!dev)
2610                 return 0;
2611
2612         rcu_read_lock();
2613         idev = __in_dev_get_rcu(dev);
2614         if (!idev)
2615                 goto out;
2616
2617         ifa = idev->ifa_list;
2618         if (!ifa)
2619                 goto out;
2620
2621         addr = ifa->ifa_local;
2622 out:
2623         rcu_read_unlock();
2624         return addr;
2625 }
2626
2627 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2628 {
2629         struct vlan_entry *vlan;
2630
2631         if (ip == bond->master_ip)
2632                 return 1;
2633
2634         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2635                 if (ip == vlan->vlan_ip)
2636                         return 1;
2637         }
2638
2639         return 0;
2640 }
2641
2642 /*
2643  * We go to the (large) trouble of VLAN tagging ARP frames because
2644  * switches in VLAN mode (especially if ports are configured as
2645  * "native" to a VLAN) might not pass non-tagged frames.
2646  */
2647 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2648 {
2649         struct sk_buff *skb;
2650
2651         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2652                  slave_dev->name, dest_ip, src_ip, vlan_id);
2653
2654         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2655                          NULL, slave_dev->dev_addr, NULL);
2656
2657         if (!skb) {
2658                 pr_err("ARP packet allocation failed\n");
2659                 return;
2660         }
2661         if (vlan_id) {
2662                 skb = vlan_put_tag(skb, vlan_id);
2663                 if (!skb) {
2664                         pr_err("failed to insert VLAN tag\n");
2665                         return;
2666                 }
2667         }
2668         arp_xmit(skb);
2669 }
2670
2671
2672 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2673 {
2674         int i, vlan_id;
2675         __be32 *targets = bond->params.arp_targets;
2676         struct vlan_entry *vlan;
2677         struct net_device *vlan_dev;
2678         struct rtable *rt;
2679
2680         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2681                 if (!targets[i])
2682                         break;
2683                 pr_debug("basa: target %x\n", targets[i]);
2684                 if (!bond->vlgrp) {
2685                         pr_debug("basa: empty vlan: arp_send\n");
2686                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2687                                       bond->master_ip, 0);
2688                         continue;
2689                 }
2690
2691                 /*
2692                  * If VLANs are configured, we do a route lookup to
2693                  * determine which VLAN interface would be used, so we
2694                  * can tag the ARP with the proper VLAN tag.
2695                  */
2696                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2697                                      RTO_ONLINK, 0);
2698                 if (IS_ERR(rt)) {
2699                         if (net_ratelimit()) {
2700                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2701                                            bond->dev->name, &targets[i]);
2702                         }
2703                         continue;
2704                 }
2705
2706                 /*
2707                  * This target is not on a VLAN
2708                  */
2709                 if (rt->dst.dev == bond->dev) {
2710                         ip_rt_put(rt);
2711                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2712                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2713                                       bond->master_ip, 0);
2714                         continue;
2715                 }
2716
2717                 vlan_id = 0;
2718                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2719                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2720                         if (vlan_dev == rt->dst.dev) {
2721                                 vlan_id = vlan->vlan_id;
2722                                 pr_debug("basa: vlan match on %s %d\n",
2723                                        vlan_dev->name, vlan_id);
2724                                 break;
2725                         }
2726                 }
2727
2728                 if (vlan_id) {
2729                         ip_rt_put(rt);
2730                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2731                                       vlan->vlan_ip, vlan_id);
2732                         continue;
2733                 }
2734
2735                 if (net_ratelimit()) {
2736                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2737                                    bond->dev->name, &targets[i],
2738                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2739                 }
2740                 ip_rt_put(rt);
2741         }
2742 }
2743
2744 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2745 {
2746         int i;
2747         __be32 *targets = bond->params.arp_targets;
2748
2749         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2750                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2751                          &sip, &tip, i, &targets[i],
2752                          bond_has_this_ip(bond, tip));
2753                 if (sip == targets[i]) {
2754                         if (bond_has_this_ip(bond, tip))
2755                                 slave->last_arp_rx = jiffies;
2756                         return;
2757                 }
2758         }
2759 }
2760
2761 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2762                          struct slave *slave)
2763 {
2764         struct arphdr *arp;
2765         unsigned char *arp_ptr;
2766         __be32 sip, tip;
2767
2768         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2769                 return;
2770
2771         read_lock(&bond->lock);
2772
2773         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2774                  bond->dev->name, skb->dev->name);
2775
2776         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2777                 goto out_unlock;
2778
2779         arp = arp_hdr(skb);
2780         if (arp->ar_hln != bond->dev->addr_len ||
2781             skb->pkt_type == PACKET_OTHERHOST ||
2782             skb->pkt_type == PACKET_LOOPBACK ||
2783             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2784             arp->ar_pro != htons(ETH_P_IP) ||
2785             arp->ar_pln != 4)
2786                 goto out_unlock;
2787
2788         arp_ptr = (unsigned char *)(arp + 1);
2789         arp_ptr += bond->dev->addr_len;
2790         memcpy(&sip, arp_ptr, 4);
2791         arp_ptr += 4 + bond->dev->addr_len;
2792         memcpy(&tip, arp_ptr, 4);
2793
2794         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2795                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2796                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2797                  &sip, &tip);
2798
2799         /*
2800          * Backup slaves won't see the ARP reply, but do come through
2801          * here for each ARP probe (so we swap the sip/tip to validate
2802          * the probe).  In a "redundant switch, common router" type of
2803          * configuration, the ARP probe will (hopefully) travel from
2804          * the active, through one switch, the router, then the other
2805          * switch before reaching the backup.
2806          */
2807         if (bond_is_active_slave(slave))
2808                 bond_validate_arp(bond, slave, sip, tip);
2809         else
2810                 bond_validate_arp(bond, slave, tip, sip);
2811
2812 out_unlock:
2813         read_unlock(&bond->lock);
2814 }
2815
2816 /*
2817  * this function is called regularly to monitor each slave's link
2818  * ensuring that traffic is being sent and received when arp monitoring
2819  * is used in load-balancing mode. if the adapter has been dormant, then an
2820  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2821  * arp monitoring in active backup mode.
2822  */
2823 void bond_loadbalance_arp_mon(struct work_struct *work)
2824 {
2825         struct bonding *bond = container_of(work, struct bonding,
2826                                             arp_work.work);
2827         struct slave *slave, *oldcurrent;
2828         int do_failover = 0;
2829         int delta_in_ticks;
2830         int i;
2831
2832         read_lock(&bond->lock);
2833
2834         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2835
2836         if (bond->kill_timers)
2837                 goto out;
2838
2839         if (bond->slave_cnt == 0)
2840                 goto re_arm;
2841
2842         read_lock(&bond->curr_slave_lock);
2843         oldcurrent = bond->curr_active_slave;
2844         read_unlock(&bond->curr_slave_lock);
2845
2846         /* see if any of the previous devices are up now (i.e. they have
2847          * xmt and rcv traffic). the curr_active_slave does not come into
2848          * the picture unless it is null. also, slave->jiffies is not needed
2849          * here because we send an arp on each slave and give a slave as
2850          * long as it needs to get the tx/rx within the delta.
2851          * TODO: what about up/down delay in arp mode? it wasn't here before
2852          *       so it can wait
2853          */
2854         bond_for_each_slave(bond, slave, i) {
2855                 unsigned long trans_start = dev_trans_start(slave->dev);
2856
2857                 if (slave->link != BOND_LINK_UP) {
2858                         if (time_in_range(jiffies,
2859                                 trans_start - delta_in_ticks,
2860                                 trans_start + delta_in_ticks) &&
2861                             time_in_range(jiffies,
2862                                 slave->dev->last_rx - delta_in_ticks,
2863                                 slave->dev->last_rx + delta_in_ticks)) {
2864
2865                                 slave->link  = BOND_LINK_UP;
2866                                 bond_set_active_slave(slave);
2867
2868                                 /* primary_slave has no meaning in round-robin
2869                                  * mode. the window of a slave being up and
2870                                  * curr_active_slave being null after enslaving
2871                                  * is closed.
2872                                  */
2873                                 if (!oldcurrent) {
2874                                         pr_info("%s: link status definitely up for interface %s, ",
2875                                                 bond->dev->name,
2876                                                 slave->dev->name);
2877                                         do_failover = 1;
2878                                 } else {
2879                                         pr_info("%s: interface %s is now up\n",
2880                                                 bond->dev->name,
2881                                                 slave->dev->name);
2882                                 }
2883                         }
2884                 } else {
2885                         /* slave->link == BOND_LINK_UP */
2886
2887                         /* not all switches will respond to an arp request
2888                          * when the source ip is 0, so don't take the link down
2889                          * if we don't know our ip yet
2890                          */
2891                         if (!time_in_range(jiffies,
2892                                 trans_start - delta_in_ticks,
2893                                 trans_start + 2 * delta_in_ticks) ||
2894                             !time_in_range(jiffies,
2895                                 slave->dev->last_rx - delta_in_ticks,
2896                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2897
2898                                 slave->link  = BOND_LINK_DOWN;
2899                                 bond_set_backup_slave(slave);
2900
2901                                 if (slave->link_failure_count < UINT_MAX)
2902                                         slave->link_failure_count++;
2903
2904                                 pr_info("%s: interface %s is now down.\n",
2905                                         bond->dev->name,
2906                                         slave->dev->name);
2907
2908                                 if (slave == oldcurrent)
2909                                         do_failover = 1;
2910                         }
2911                 }
2912
2913                 /* note: if switch is in round-robin mode, all links
2914                  * must tx arp to ensure all links rx an arp - otherwise
2915                  * links may oscillate or not come up at all; if switch is
2916                  * in something like xor mode, there is nothing we can
2917                  * do - all replies will be rx'ed on same link causing slaves
2918                  * to be unstable during low/no traffic periods
2919                  */
2920                 if (IS_UP(slave->dev))
2921                         bond_arp_send_all(bond, slave);
2922         }
2923
2924         if (do_failover) {
2925                 block_netpoll_tx();
2926                 write_lock_bh(&bond->curr_slave_lock);
2927
2928                 bond_select_active_slave(bond);
2929
2930                 write_unlock_bh(&bond->curr_slave_lock);
2931                 unblock_netpoll_tx();
2932         }
2933
2934 re_arm:
2935         if (bond->params.arp_interval)
2936                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2937 out:
2938         read_unlock(&bond->lock);
2939 }
2940
2941 /*
2942  * Called to inspect slaves for active-backup mode ARP monitor link state
2943  * changes.  Sets new_link in slaves to specify what action should take
2944  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2945  * to link states must be committed.
2946  *
2947  * Called with bond->lock held for read.
2948  */
2949 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2950 {
2951         struct slave *slave;
2952         int i, commit = 0;
2953         unsigned long trans_start;
2954
2955         bond_for_each_slave(bond, slave, i) {
2956                 slave->new_link = BOND_LINK_NOCHANGE;
2957
2958                 if (slave->link != BOND_LINK_UP) {
2959                         if (time_in_range(jiffies,
2960                                 slave_last_rx(bond, slave) - delta_in_ticks,
2961                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2962
2963                                 slave->new_link = BOND_LINK_UP;
2964                                 commit++;
2965                         }
2966
2967                         continue;
2968                 }
2969
2970                 /*
2971                  * Give slaves 2*delta after being enslaved or made
2972                  * active.  This avoids bouncing, as the last receive
2973                  * times need a full ARP monitor cycle to be updated.
2974                  */
2975                 if (time_in_range(jiffies,
2976                                   slave->jiffies - delta_in_ticks,
2977                                   slave->jiffies + 2 * delta_in_ticks))
2978                         continue;
2979
2980                 /*
2981                  * Backup slave is down if:
2982                  * - No current_arp_slave AND
2983                  * - more than 3*delta since last receive AND
2984                  * - the bond has an IP address
2985                  *
2986                  * Note: a non-null current_arp_slave indicates
2987                  * the curr_active_slave went down and we are
2988                  * searching for a new one; under this condition
2989                  * we only take the curr_active_slave down - this
2990                  * gives each slave a chance to tx/rx traffic
2991                  * before being taken out
2992                  */
2993                 if (!bond_is_active_slave(slave) &&
2994                     !bond->current_arp_slave &&
2995                     !time_in_range(jiffies,
2996                         slave_last_rx(bond, slave) - delta_in_ticks,
2997                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2998
2999                         slave->new_link = BOND_LINK_DOWN;
3000                         commit++;
3001                 }
3002
3003                 /*
3004                  * Active slave is down if:
3005                  * - more than 2*delta since transmitting OR
3006                  * - (more than 2*delta since receive AND
3007                  *    the bond has an IP address)
3008                  */
3009                 trans_start = dev_trans_start(slave->dev);
3010                 if (bond_is_active_slave(slave) &&
3011                     (!time_in_range(jiffies,
3012                         trans_start - delta_in_ticks,
3013                         trans_start + 2 * delta_in_ticks) ||
3014                      !time_in_range(jiffies,
3015                         slave_last_rx(bond, slave) - delta_in_ticks,
3016                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3017
3018                         slave->new_link = BOND_LINK_DOWN;
3019                         commit++;
3020                 }
3021         }
3022
3023         return commit;
3024 }
3025
3026 /*
3027  * Called to commit link state changes noted by inspection step of
3028  * active-backup mode ARP monitor.
3029  *
3030  * Called with RTNL and bond->lock for read.
3031  */
3032 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3033 {
3034         struct slave *slave;
3035         int i;
3036         unsigned long trans_start;
3037
3038         bond_for_each_slave(bond, slave, i) {
3039                 switch (slave->new_link) {
3040                 case BOND_LINK_NOCHANGE:
3041                         continue;
3042
3043                 case BOND_LINK_UP:
3044                         trans_start = dev_trans_start(slave->dev);
3045                         if ((!bond->curr_active_slave &&
3046                              time_in_range(jiffies,
3047                                            trans_start - delta_in_ticks,
3048                                            trans_start + delta_in_ticks)) ||
3049                             bond->curr_active_slave != slave) {
3050                                 slave->link = BOND_LINK_UP;
3051                                 bond->current_arp_slave = NULL;
3052
3053                                 pr_info("%s: link status definitely up for interface %s.\n",
3054                                         bond->dev->name, slave->dev->name);
3055
3056                                 if (!bond->curr_active_slave ||
3057                                     (slave == bond->primary_slave))
3058                                         goto do_failover;
3059
3060                         }
3061
3062                         continue;
3063
3064                 case BOND_LINK_DOWN:
3065                         if (slave->link_failure_count < UINT_MAX)
3066                                 slave->link_failure_count++;
3067
3068                         slave->link = BOND_LINK_DOWN;
3069                         bond_set_slave_inactive_flags(slave);
3070
3071                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3072                                 bond->dev->name, slave->dev->name);
3073
3074                         if (slave == bond->curr_active_slave) {
3075                                 bond->current_arp_slave = NULL;
3076                                 goto do_failover;
3077                         }
3078
3079                         continue;
3080
3081                 default:
3082                         pr_err("%s: impossible: new_link %d on slave %s\n",
3083                                bond->dev->name, slave->new_link,
3084                                slave->dev->name);
3085                         continue;
3086                 }
3087
3088 do_failover:
3089                 ASSERT_RTNL();
3090                 block_netpoll_tx();
3091                 write_lock_bh(&bond->curr_slave_lock);
3092                 bond_select_active_slave(bond);
3093                 write_unlock_bh(&bond->curr_slave_lock);
3094                 unblock_netpoll_tx();
3095         }
3096
3097         bond_set_carrier(bond);
3098 }
3099
3100 /*
3101  * Send ARP probes for active-backup mode ARP monitor.
3102  *
3103  * Called with bond->lock held for read.
3104  */
3105 static void bond_ab_arp_probe(struct bonding *bond)
3106 {
3107         struct slave *slave;
3108         int i;
3109
3110         read_lock(&bond->curr_slave_lock);
3111
3112         if (bond->current_arp_slave && bond->curr_active_slave)
3113                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3114                         bond->current_arp_slave->dev->name,
3115                         bond->curr_active_slave->dev->name);
3116
3117         if (bond->curr_active_slave) {
3118                 bond_arp_send_all(bond, bond->curr_active_slave);
3119                 read_unlock(&bond->curr_slave_lock);
3120                 return;
3121         }
3122
3123         read_unlock(&bond->curr_slave_lock);
3124
3125         /* if we don't have a curr_active_slave, search for the next available
3126          * backup slave from the current_arp_slave and make it the candidate
3127          * for becoming the curr_active_slave
3128          */
3129
3130         if (!bond->current_arp_slave) {
3131                 bond->current_arp_slave = bond->first_slave;
3132                 if (!bond->current_arp_slave)
3133                         return;
3134         }
3135
3136         bond_set_slave_inactive_flags(bond->current_arp_slave);
3137
3138         /* search for next candidate */
3139         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3140                 if (IS_UP(slave->dev)) {
3141                         slave->link = BOND_LINK_BACK;
3142                         bond_set_slave_active_flags(slave);
3143                         bond_arp_send_all(bond, slave);
3144                         slave->jiffies = jiffies;
3145                         bond->current_arp_slave = slave;
3146                         break;
3147                 }
3148
3149                 /* if the link state is up at this point, we
3150                  * mark it down - this can happen if we have
3151                  * simultaneous link failures and
3152                  * reselect_active_interface doesn't make this
3153                  * one the current slave so it is still marked
3154                  * up when it is actually down
3155                  */
3156                 if (slave->link == BOND_LINK_UP) {
3157                         slave->link = BOND_LINK_DOWN;
3158                         if (slave->link_failure_count < UINT_MAX)
3159                                 slave->link_failure_count++;
3160
3161                         bond_set_slave_inactive_flags(slave);
3162
3163                         pr_info("%s: backup interface %s is now down.\n",
3164                                 bond->dev->name, slave->dev->name);
3165                 }
3166         }
3167 }
3168
3169 void bond_activebackup_arp_mon(struct work_struct *work)
3170 {
3171         struct bonding *bond = container_of(work, struct bonding,
3172                                             arp_work.work);
3173         bool should_notify_peers = false;
3174         int delta_in_ticks;
3175
3176         read_lock(&bond->lock);
3177
3178         if (bond->kill_timers)
3179                 goto out;
3180
3181         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3182
3183         if (bond->slave_cnt == 0)
3184                 goto re_arm;
3185
3186         should_notify_peers = bond_should_notify_peers(bond);
3187
3188         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3189                 read_unlock(&bond->lock);
3190                 rtnl_lock();
3191                 read_lock(&bond->lock);
3192
3193                 bond_ab_arp_commit(bond, delta_in_ticks);
3194
3195                 read_unlock(&bond->lock);
3196                 rtnl_unlock();
3197                 read_lock(&bond->lock);
3198         }
3199
3200         bond_ab_arp_probe(bond);
3201
3202 re_arm:
3203         if (bond->params.arp_interval)
3204                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3205 out:
3206         read_unlock(&bond->lock);
3207
3208         if (should_notify_peers) {
3209                 rtnl_lock();
3210                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3211                 rtnl_unlock();
3212         }
3213 }
3214
3215 /*-------------------------- netdev event handling --------------------------*/
3216
3217 /*
3218  * Change device name
3219  */
3220 static int bond_event_changename(struct bonding *bond)
3221 {
3222         bond_remove_proc_entry(bond);
3223         bond_create_proc_entry(bond);
3224
3225         bond_debug_reregister(bond);
3226
3227         return NOTIFY_DONE;
3228 }
3229
3230 static int bond_master_netdev_event(unsigned long event,
3231                                     struct net_device *bond_dev)
3232 {
3233         struct bonding *event_bond = netdev_priv(bond_dev);
3234
3235         switch (event) {
3236         case NETDEV_CHANGENAME:
3237                 return bond_event_changename(event_bond);
3238         default:
3239                 break;
3240         }
3241
3242         return NOTIFY_DONE;
3243 }
3244
3245 static int bond_slave_netdev_event(unsigned long event,
3246                                    struct net_device *slave_dev)
3247 {
3248         struct net_device *bond_dev = slave_dev->master;
3249         struct bonding *bond = netdev_priv(bond_dev);
3250
3251         switch (event) {
3252         case NETDEV_UNREGISTER:
3253                 if (bond_dev) {
3254                         if (bond->setup_by_slave)
3255                                 bond_release_and_destroy(bond_dev, slave_dev);
3256                         else
3257                                 bond_release(bond_dev, slave_dev);
3258                 }
3259                 break;
3260         case NETDEV_CHANGE:
3261                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3262                         struct slave *slave;
3263
3264                         slave = bond_get_slave_by_dev(bond, slave_dev);
3265                         if (slave) {
3266                                 u32 old_speed = slave->speed;
3267                                 u8  old_duplex = slave->duplex;
3268
3269                                 bond_update_speed_duplex(slave);
3270
3271                                 if (bond_is_lb(bond))
3272                                         break;
3273
3274                                 if (old_speed != slave->speed)
3275                                         bond_3ad_adapter_speed_changed(slave);
3276                                 if (old_duplex != slave->duplex)
3277                                         bond_3ad_adapter_duplex_changed(slave);
3278                         }
3279                 }
3280
3281                 break;
3282         case NETDEV_DOWN:
3283                 /*
3284                  * ... Or is it this?
3285                  */
3286                 break;
3287         case NETDEV_CHANGEMTU:
3288                 /*
3289                  * TODO: Should slaves be allowed to
3290                  * independently alter their MTU?  For
3291                  * an active-backup bond, slaves need
3292                  * not be the same type of device, so
3293                  * MTUs may vary.  For other modes,
3294                  * slaves arguably should have the
3295                  * same MTUs. To do this, we'd need to
3296                  * take over the slave's change_mtu
3297                  * function for the duration of their
3298                  * servitude.
3299                  */
3300                 break;
3301         case NETDEV_CHANGENAME:
3302                 /*
3303                  * TODO: handle changing the primary's name
3304                  */
3305                 break;
3306         case NETDEV_FEAT_CHANGE:
3307                 bond_compute_features(bond);
3308                 break;
3309         default:
3310                 break;
3311         }
3312
3313         return NOTIFY_DONE;
3314 }
3315
3316 /*
3317  * bond_netdev_event: handle netdev notifier chain events.
3318  *
3319  * This function receives events for the netdev chain.  The caller (an
3320  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3321  * locks for us to safely manipulate the slave devices (RTNL lock,
3322  * dev_probe_lock).
3323  */
3324 static int bond_netdev_event(struct notifier_block *this,
3325                              unsigned long event, void *ptr)
3326 {
3327         struct net_device *event_dev = (struct net_device *)ptr;
3328
3329         pr_debug("event_dev: %s, event: %lx\n",
3330                  event_dev ? event_dev->name : "None",
3331                  event);
3332
3333         if (!(event_dev->priv_flags & IFF_BONDING))
3334                 return NOTIFY_DONE;
3335
3336         if (event_dev->flags & IFF_MASTER) {
3337                 pr_debug("IFF_MASTER\n");
3338                 return bond_master_netdev_event(event, event_dev);
3339         }
3340
3341         if (event_dev->flags & IFF_SLAVE) {
3342                 pr_debug("IFF_SLAVE\n");
3343                 return bond_slave_netdev_event(event, event_dev);
3344         }
3345
3346         return NOTIFY_DONE;
3347 }
3348
3349 /*
3350  * bond_inetaddr_event: handle inetaddr notifier chain events.
3351  *
3352  * We keep track of device IPs primarily to use as source addresses in
3353  * ARP monitor probes (rather than spewing out broadcasts all the time).
3354  *
3355  * We track one IP for the main device (if it has one), plus one per VLAN.
3356  */
3357 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3358 {
3359         struct in_ifaddr *ifa = ptr;
3360         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3361         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3362         struct bonding *bond;
3363         struct vlan_entry *vlan;
3364
3365         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3366                 if (bond->dev == event_dev) {
3367                         switch (event) {
3368                         case NETDEV_UP:
3369                                 bond->master_ip = ifa->ifa_local;
3370                                 return NOTIFY_OK;
3371                         case NETDEV_DOWN:
3372                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3373                                 return NOTIFY_OK;
3374                         default:
3375                                 return NOTIFY_DONE;
3376                         }
3377                 }
3378
3379                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3380                         if (!bond->vlgrp)
3381                                 continue;
3382                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3383                         if (vlan_dev == event_dev) {
3384                                 switch (event) {
3385                                 case NETDEV_UP:
3386                                         vlan->vlan_ip = ifa->ifa_local;
3387                                         return NOTIFY_OK;
3388                                 case NETDEV_DOWN:
3389                                         vlan->vlan_ip =
3390                                                 bond_glean_dev_ip(vlan_dev);
3391                                         return NOTIFY_OK;
3392                                 default:
3393                                         return NOTIFY_DONE;
3394                                 }
3395                         }
3396                 }
3397         }
3398         return NOTIFY_DONE;
3399 }
3400
3401 static struct notifier_block bond_netdev_notifier = {
3402         .notifier_call = bond_netdev_event,
3403 };
3404
3405 static struct notifier_block bond_inetaddr_notifier = {
3406         .notifier_call = bond_inetaddr_event,
3407 };
3408
3409 /*---------------------------- Hashing Policies -----------------------------*/
3410
3411 /*
3412  * Hash for the output device based upon layer 2 and layer 3 data. If
3413  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3414  */
3415 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3416 {
3417         struct ethhdr *data = (struct ethhdr *)skb->data;
3418         struct iphdr *iph = ip_hdr(skb);
3419
3420         if (skb->protocol == htons(ETH_P_IP)) {
3421                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3422                         (data->h_dest[5] ^ data->h_source[5])) % count;
3423         }
3424
3425         return (data->h_dest[5] ^ data->h_source[5]) % count;
3426 }
3427
3428 /*
3429  * Hash for the output device based upon layer 3 and layer 4 data. If
3430  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3431  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3432  */
3433 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3434 {
3435         struct ethhdr *data = (struct ethhdr *)skb->data;
3436         struct iphdr *iph = ip_hdr(skb);
3437         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3438         int layer4_xor = 0;
3439
3440         if (skb->protocol == htons(ETH_P_IP)) {
3441                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3442                     (iph->protocol == IPPROTO_TCP ||
3443                      iph->protocol == IPPROTO_UDP)) {
3444                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3445                 }
3446                 return (layer4_xor ^
3447                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3448
3449         }
3450
3451         return (data->h_dest[5] ^ data->h_source[5]) % count;
3452 }
3453
3454 /*
3455  * Hash for the output device based upon layer 2 data
3456  */
3457 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3458 {
3459         struct ethhdr *data = (struct ethhdr *)skb->data;
3460
3461         return (data->h_dest[5] ^ data->h_source[5]) % count;
3462 }
3463
3464 /*-------------------------- Device entry points ----------------------------*/
3465
3466 static int bond_open(struct net_device *bond_dev)
3467 {
3468         struct bonding *bond = netdev_priv(bond_dev);
3469
3470         bond->kill_timers = 0;
3471
3472         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3473
3474         if (bond_is_lb(bond)) {
3475                 /* bond_alb_initialize must be called before the timer
3476                  * is started.
3477                  */
3478                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3479                         /* something went wrong - fail the open operation */
3480                         return -ENOMEM;
3481                 }
3482
3483                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3484                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3485         }
3486
3487         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3488                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3489                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3490         }
3491
3492         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3493                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3494                         INIT_DELAYED_WORK(&bond->arp_work,
3495                                           bond_activebackup_arp_mon);
3496                 else
3497                         INIT_DELAYED_WORK(&bond->arp_work,
3498                                           bond_loadbalance_arp_mon);
3499
3500                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3501                 if (bond->params.arp_validate)
3502                         bond->recv_probe = bond_arp_rcv;
3503         }
3504
3505         if (bond->params.mode == BOND_MODE_8023AD) {
3506                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3507                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3508                 /* register to receive LACPDUs */
3509                 bond->recv_probe = bond_3ad_lacpdu_recv;
3510                 bond_3ad_initiate_agg_selection(bond, 1);
3511         }
3512
3513         return 0;
3514 }
3515
3516 static int bond_close(struct net_device *bond_dev)
3517 {
3518         struct bonding *bond = netdev_priv(bond_dev);
3519
3520         write_lock_bh(&bond->lock);
3521
3522         bond->send_peer_notif = 0;
3523
3524         /* signal timers not to re-arm */
3525         bond->kill_timers = 1;
3526
3527         write_unlock_bh(&bond->lock);
3528
3529         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3530                 cancel_delayed_work(&bond->mii_work);
3531         }
3532
3533         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3534                 cancel_delayed_work(&bond->arp_work);
3535         }
3536
3537         switch (bond->params.mode) {
3538         case BOND_MODE_8023AD:
3539                 cancel_delayed_work(&bond->ad_work);
3540                 break;
3541         case BOND_MODE_TLB:
3542         case BOND_MODE_ALB:
3543                 cancel_delayed_work(&bond->alb_work);
3544                 break;
3545         default:
3546                 break;
3547         }
3548
3549         if (delayed_work_pending(&bond->mcast_work))
3550                 cancel_delayed_work(&bond->mcast_work);
3551
3552         if (bond_is_lb(bond)) {
3553                 /* Must be called only after all
3554                  * slaves have been released
3555                  */
3556                 bond_alb_deinitialize(bond);
3557         }
3558         bond->recv_probe = NULL;
3559
3560         return 0;
3561 }
3562
3563 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3564                                                 struct rtnl_link_stats64 *stats)
3565 {
3566         struct bonding *bond = netdev_priv(bond_dev);
3567         struct rtnl_link_stats64 temp;
3568         struct slave *slave;
3569         int i;
3570
3571         memset(stats, 0, sizeof(*stats));
3572
3573         read_lock_bh(&bond->lock);
3574
3575         bond_for_each_slave(bond, slave, i) {
3576                 const struct rtnl_link_stats64 *sstats =
3577                         dev_get_stats(slave->dev, &temp);
3578
3579                 stats->rx_packets += sstats->rx_packets;
3580                 stats->rx_bytes += sstats->rx_bytes;
3581                 stats->rx_errors += sstats->rx_errors;
3582                 stats->rx_dropped += sstats->rx_dropped;
3583
3584                 stats->tx_packets += sstats->tx_packets;
3585                 stats->tx_bytes += sstats->tx_bytes;
3586                 stats->tx_errors += sstats->tx_errors;
3587                 stats->tx_dropped += sstats->tx_dropped;
3588
3589                 stats->multicast += sstats->multicast;
3590                 stats->collisions += sstats->collisions;
3591
3592                 stats->rx_length_errors += sstats->rx_length_errors;
3593                 stats->rx_over_errors += sstats->rx_over_errors;
3594                 stats->rx_crc_errors += sstats->rx_crc_errors;
3595                 stats->rx_frame_errors += sstats->rx_frame_errors;
3596                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3597                 stats->rx_missed_errors += sstats->rx_missed_errors;
3598
3599                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3600                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3601                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3602                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3603                 stats->tx_window_errors += sstats->tx_window_errors;
3604         }
3605
3606         read_unlock_bh(&bond->lock);
3607
3608         return stats;
3609 }
3610
3611 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3612 {
3613         struct net_device *slave_dev = NULL;
3614         struct ifbond k_binfo;
3615         struct ifbond __user *u_binfo = NULL;
3616         struct ifslave k_sinfo;
3617         struct ifslave __user *u_sinfo = NULL;
3618         struct mii_ioctl_data *mii = NULL;
3619         int res = 0;
3620
3621         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3622
3623         switch (cmd) {
3624         case SIOCGMIIPHY:
3625                 mii = if_mii(ifr);
3626                 if (!mii)
3627                         return -EINVAL;
3628
3629                 mii->phy_id = 0;
3630                 /* Fall Through */
3631         case SIOCGMIIREG:
3632                 /*
3633                  * We do this again just in case we were called by SIOCGMIIREG
3634                  * instead of SIOCGMIIPHY.
3635                  */
3636                 mii = if_mii(ifr);
3637                 if (!mii)
3638                         return -EINVAL;
3639
3640
3641                 if (mii->reg_num == 1) {
3642                         struct bonding *bond = netdev_priv(bond_dev);
3643                         mii->val_out = 0;
3644                         read_lock(&bond->lock);
3645                         read_lock(&bond->curr_slave_lock);
3646                         if (netif_carrier_ok(bond->dev))
3647                                 mii->val_out = BMSR_LSTATUS;
3648
3649                         read_unlock(&bond->curr_slave_lock);
3650                         read_unlock(&bond->lock);
3651                 }
3652
3653                 return 0;
3654         case BOND_INFO_QUERY_OLD:
3655         case SIOCBONDINFOQUERY:
3656                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3657
3658                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3659                         return -EFAULT;
3660
3661                 res = bond_info_query(bond_dev, &k_binfo);
3662                 if (res == 0 &&
3663                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3664                         return -EFAULT;
3665
3666                 return res;
3667         case BOND_SLAVE_INFO_QUERY_OLD:
3668         case SIOCBONDSLAVEINFOQUERY:
3669                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3670
3671                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3672                         return -EFAULT;
3673
3674                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3675                 if (res == 0 &&
3676                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3677                         return -EFAULT;
3678
3679                 return res;
3680         default:
3681                 /* Go on */
3682                 break;
3683         }
3684
3685         if (!capable(CAP_NET_ADMIN))
3686                 return -EPERM;
3687
3688         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3689
3690         pr_debug("slave_dev=%p:\n", slave_dev);
3691
3692         if (!slave_dev)
3693                 res = -ENODEV;
3694         else {
3695                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3696                 switch (cmd) {
3697                 case BOND_ENSLAVE_OLD:
3698                 case SIOCBONDENSLAVE:
3699                         res = bond_enslave(bond_dev, slave_dev);
3700                         break;
3701                 case BOND_RELEASE_OLD:
3702                 case SIOCBONDRELEASE:
3703                         res = bond_release(bond_dev, slave_dev);
3704                         break;
3705                 case BOND_SETHWADDR_OLD:
3706                 case SIOCBONDSETHWADDR:
3707                         res = bond_sethwaddr(bond_dev, slave_dev);
3708                         break;
3709                 case BOND_CHANGE_ACTIVE_OLD:
3710                 case SIOCBONDCHANGEACTIVE:
3711                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3712                         break;
3713                 default:
3714                         res = -EOPNOTSUPP;
3715                 }
3716
3717                 dev_put(slave_dev);
3718         }
3719
3720         return res;
3721 }
3722
3723 static bool bond_addr_in_mc_list(unsigned char *addr,
3724                                  struct netdev_hw_addr_list *list,
3725                                  int addrlen)
3726 {
3727         struct netdev_hw_addr *ha;
3728
3729         netdev_hw_addr_list_for_each(ha, list)
3730                 if (!memcmp(ha->addr, addr, addrlen))
3731                         return true;
3732
3733         return false;
3734 }
3735
3736 static void bond_set_multicast_list(struct net_device *bond_dev)
3737 {
3738         struct bonding *bond = netdev_priv(bond_dev);
3739         struct netdev_hw_addr *ha;
3740         bool found;
3741
3742         /*
3743          * Do promisc before checking multicast_mode
3744          */
3745         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3746                 /*
3747                  * FIXME: Need to handle the error when one of the multi-slaves
3748                  * encounters error.
3749                  */
3750                 bond_set_promiscuity(bond, 1);
3751
3752
3753         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3754                 bond_set_promiscuity(bond, -1);
3755
3756
3757         /* set allmulti flag to slaves */
3758         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3759                 /*
3760                  * FIXME: Need to handle the error when one of the multi-slaves
3761                  * encounters error.
3762                  */
3763                 bond_set_allmulti(bond, 1);
3764
3765
3766         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3767                 bond_set_allmulti(bond, -1);
3768
3769
3770         read_lock(&bond->lock);
3771
3772         bond->flags = bond_dev->flags;
3773
3774         /* looking for addresses to add to slaves' mc list */
3775         netdev_for_each_mc_addr(ha, bond_dev) {
3776                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3777                                              bond_dev->addr_len);
3778                 if (!found)
3779                         bond_mc_add(bond, ha->addr);
3780         }
3781
3782         /* looking for addresses to delete from slaves' list */
3783         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3784                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3785                                              bond_dev->addr_len);
3786                 if (!found)
3787                         bond_mc_del(bond, ha->addr);
3788         }
3789
3790         /* save master's multicast list */
3791         __hw_addr_flush(&bond->mc_list);
3792         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3793                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3794
3795         read_unlock(&bond->lock);
3796 }
3797
3798 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3799 {
3800         struct bonding *bond = netdev_priv(dev);
3801         struct slave *slave = bond->first_slave;
3802
3803         if (slave) {
3804                 const struct net_device_ops *slave_ops
3805                         = slave->dev->netdev_ops;
3806                 if (slave_ops->ndo_neigh_setup)
3807                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3808         }
3809         return 0;
3810 }
3811
3812 /*
3813  * Change the MTU of all of a master's slaves to match the master
3814  */
3815 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3816 {
3817         struct bonding *bond = netdev_priv(bond_dev);
3818         struct slave *slave, *stop_at;
3819         int res = 0;
3820         int i;
3821
3822         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3823                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3824
3825         /* Can't hold bond->lock with bh disabled here since
3826          * some base drivers panic. On the other hand we can't
3827          * hold bond->lock without bh disabled because we'll
3828          * deadlock. The only solution is to rely on the fact
3829          * that we're under rtnl_lock here, and the slaves
3830          * list won't change. This doesn't solve the problem
3831          * of setting the slave's MTU while it is
3832          * transmitting, but the assumption is that the base
3833          * driver can handle that.
3834          *
3835          * TODO: figure out a way to safely iterate the slaves
3836          * list, but without holding a lock around the actual
3837          * call to the base driver.
3838          */
3839
3840         bond_for_each_slave(bond, slave, i) {
3841                 pr_debug("s %p s->p %p c_m %p\n",
3842                          slave,
3843                          slave->prev,
3844                          slave->dev->netdev_ops->ndo_change_mtu);
3845
3846                 res = dev_set_mtu(slave->dev, new_mtu);
3847
3848                 if (res) {
3849                         /* If we failed to set the slave's mtu to the new value
3850                          * we must abort the operation even in ACTIVE_BACKUP
3851                          * mode, because if we allow the backup slaves to have
3852                          * different mtu values than the active slave we'll
3853                          * need to change their mtu when doing a failover. That
3854                          * means changing their mtu from timer context, which
3855                          * is probably not a good idea.
3856                          */
3857                         pr_debug("err %d %s\n", res, slave->dev->name);
3858                         goto unwind;
3859                 }
3860         }
3861
3862         bond_dev->mtu = new_mtu;
3863
3864         return 0;
3865
3866 unwind:
3867         /* unwind from head to the slave that failed */
3868         stop_at = slave;
3869         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3870                 int tmp_res;
3871
3872                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3873                 if (tmp_res) {
3874                         pr_debug("unwind err %d dev %s\n",
3875                                  tmp_res, slave->dev->name);
3876                 }
3877         }
3878
3879         return res;
3880 }
3881
3882 /*
3883  * Change HW address
3884  *
3885  * Note that many devices must be down to change the HW address, and
3886  * downing the master releases all slaves.  We can make bonds full of
3887  * bonding devices to test this, however.
3888  */
3889 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3890 {
3891         struct bonding *bond = netdev_priv(bond_dev);
3892         struct sockaddr *sa = addr, tmp_sa;
3893         struct slave *slave, *stop_at;
3894         int res = 0;
3895         int i;
3896
3897         if (bond->params.mode == BOND_MODE_ALB)
3898                 return bond_alb_set_mac_address(bond_dev, addr);
3899
3900
3901         pr_debug("bond=%p, name=%s\n",
3902                  bond, bond_dev ? bond_dev->name : "None");
3903
3904         /*
3905          * If fail_over_mac is set to active, do nothing and return
3906          * success.  Returning an error causes ifenslave to fail.
3907          */
3908         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3909                 return 0;
3910
3911         if (!is_valid_ether_addr(sa->sa_data))
3912                 return -EADDRNOTAVAIL;
3913
3914         /* Can't hold bond->lock with bh disabled here since
3915          * some base drivers panic. On the other hand we can't
3916          * hold bond->lock without bh disabled because we'll
3917          * deadlock. The only solution is to rely on the fact
3918          * that we're under rtnl_lock here, and the slaves
3919          * list won't change. This doesn't solve the problem
3920          * of setting the slave's hw address while it is
3921          * transmitting, but the assumption is that the base
3922          * driver can handle that.
3923          *
3924          * TODO: figure out a way to safely iterate the slaves
3925          * list, but without holding a lock around the actual
3926          * call to the base driver.
3927          */
3928
3929         bond_for_each_slave(bond, slave, i) {
3930                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3931                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3932
3933                 if (slave_ops->ndo_set_mac_address == NULL) {
3934                         res = -EOPNOTSUPP;
3935                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3936                         goto unwind;
3937                 }
3938
3939                 res = dev_set_mac_address(slave->dev, addr);
3940                 if (res) {
3941                         /* TODO: consider downing the slave
3942                          * and retry ?
3943                          * User should expect communications
3944                          * breakage anyway until ARP finish
3945                          * updating, so...
3946                          */
3947                         pr_debug("err %d %s\n", res, slave->dev->name);
3948                         goto unwind;
3949                 }
3950         }
3951
3952         /* success */
3953         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3954         return 0;
3955
3956 unwind:
3957         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3958         tmp_sa.sa_family = bond_dev->type;
3959
3960         /* unwind from head to the slave that failed */
3961         stop_at = slave;
3962         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3963                 int tmp_res;
3964
3965                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3966                 if (tmp_res) {
3967                         pr_debug("unwind err %d dev %s\n",
3968                                  tmp_res, slave->dev->name);
3969                 }
3970         }
3971
3972         return res;
3973 }
3974
3975 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3976 {
3977         struct bonding *bond = netdev_priv(bond_dev);
3978         struct slave *slave, *start_at;
3979         int i, slave_no, res = 1;
3980         struct iphdr *iph = ip_hdr(skb);
3981
3982         /*
3983          * Start with the curr_active_slave that joined the bond as the
3984          * default for sending IGMP traffic.  For failover purposes one
3985          * needs to maintain some consistency for the interface that will
3986          * send the join/membership reports.  The curr_active_slave found
3987          * will send all of this type of traffic.
3988          */
3989         if ((iph->protocol == IPPROTO_IGMP) &&
3990             (skb->protocol == htons(ETH_P_IP))) {
3991
3992                 read_lock(&bond->curr_slave_lock);
3993                 slave = bond->curr_active_slave;
3994                 read_unlock(&bond->curr_slave_lock);
3995
3996                 if (!slave)
3997                         goto out;
3998         } else {
3999                 /*
4000                  * Concurrent TX may collide on rr_tx_counter; we accept
4001                  * that as being rare enough not to justify using an
4002                  * atomic op here.
4003                  */
4004                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4005
4006                 bond_for_each_slave(bond, slave, i) {
4007                         slave_no--;
4008                         if (slave_no < 0)
4009                                 break;
4010                 }
4011         }
4012
4013         start_at = slave;
4014         bond_for_each_slave_from(bond, slave, i, start_at) {
4015                 if (IS_UP(slave->dev) &&
4016                     (slave->link == BOND_LINK_UP) &&
4017                     bond_is_active_slave(slave)) {
4018                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4019                         break;
4020                 }
4021         }
4022
4023 out:
4024         if (res) {
4025                 /* no suitable interface, frame not sent */
4026                 dev_kfree_skb(skb);
4027         }
4028
4029         return NETDEV_TX_OK;
4030 }
4031
4032
4033 /*
4034  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4035  * the bond has a usable interface.
4036  */
4037 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4038 {
4039         struct bonding *bond = netdev_priv(bond_dev);
4040         int res = 1;
4041
4042         read_lock(&bond->curr_slave_lock);
4043
4044         if (bond->curr_active_slave)
4045                 res = bond_dev_queue_xmit(bond, skb,
4046                         bond->curr_active_slave->dev);
4047
4048         if (res)
4049                 /* no suitable interface, frame not sent */
4050                 dev_kfree_skb(skb);
4051
4052         read_unlock(&bond->curr_slave_lock);
4053
4054         return NETDEV_TX_OK;
4055 }
4056
4057 /*
4058  * In bond_xmit_xor() , we determine the output device by using a pre-
4059  * determined xmit_hash_policy(), If the selected device is not enabled,
4060  * find the next active slave.
4061  */
4062 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4063 {
4064         struct bonding *bond = netdev_priv(bond_dev);
4065         struct slave *slave, *start_at;
4066         int slave_no;
4067         int i;
4068         int res = 1;
4069
4070         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4071
4072         bond_for_each_slave(bond, slave, i) {
4073                 slave_no--;
4074                 if (slave_no < 0)
4075                         break;
4076         }
4077
4078         start_at = slave;
4079
4080         bond_for_each_slave_from(bond, slave, i, start_at) {
4081                 if (IS_UP(slave->dev) &&
4082                     (slave->link == BOND_LINK_UP) &&
4083                     bond_is_active_slave(slave)) {
4084                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4085                         break;
4086                 }
4087         }
4088
4089         if (res) {
4090                 /* no suitable interface, frame not sent */
4091                 dev_kfree_skb(skb);
4092         }
4093
4094         return NETDEV_TX_OK;
4095 }
4096
4097 /*
4098  * in broadcast mode, we send everything to all usable interfaces.
4099  */
4100 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4101 {
4102         struct bonding *bond = netdev_priv(bond_dev);
4103         struct slave *slave, *start_at;
4104         struct net_device *tx_dev = NULL;
4105         int i;
4106         int res = 1;
4107
4108         read_lock(&bond->curr_slave_lock);
4109         start_at = bond->curr_active_slave;
4110         read_unlock(&bond->curr_slave_lock);
4111
4112         if (!start_at)
4113                 goto out;
4114
4115         bond_for_each_slave_from(bond, slave, i, start_at) {
4116                 if (IS_UP(slave->dev) &&
4117                     (slave->link == BOND_LINK_UP) &&
4118                     bond_is_active_slave(slave)) {
4119                         if (tx_dev) {
4120                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4121                                 if (!skb2) {
4122                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4123                                                bond_dev->name);
4124                                         continue;
4125                                 }
4126
4127                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4128                                 if (res) {
4129                                         dev_kfree_skb(skb2);
4130                                         continue;
4131                                 }
4132                         }
4133                         tx_dev = slave->dev;
4134                 }
4135         }
4136
4137         if (tx_dev)
4138                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4139
4140 out:
4141         if (res)
4142                 /* no suitable interface, frame not sent */
4143                 dev_kfree_skb(skb);
4144
4145         /* frame sent to all suitable interfaces */
4146         return NETDEV_TX_OK;
4147 }
4148
4149 /*------------------------- Device initialization ---------------------------*/
4150
4151 static void bond_set_xmit_hash_policy(struct bonding *bond)
4152 {
4153         switch (bond->params.xmit_policy) {
4154         case BOND_XMIT_POLICY_LAYER23:
4155                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4156                 break;
4157         case BOND_XMIT_POLICY_LAYER34:
4158                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4159                 break;
4160         case BOND_XMIT_POLICY_LAYER2:
4161         default:
4162                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4163                 break;
4164         }
4165 }
4166
4167 /*
4168  * Lookup the slave that corresponds to a qid
4169  */
4170 static inline int bond_slave_override(struct bonding *bond,
4171                                       struct sk_buff *skb)
4172 {
4173         int i, res = 1;
4174         struct slave *slave = NULL;
4175         struct slave *check_slave;
4176
4177         if (!skb->queue_mapping)
4178                 return 1;
4179
4180         /* Find out if any slaves have the same mapping as this skb. */
4181         bond_for_each_slave(bond, check_slave, i) {
4182                 if (check_slave->queue_id == skb->queue_mapping) {
4183                         slave = check_slave;
4184                         break;
4185                 }
4186         }
4187
4188         /* If the slave isn't UP, use default transmit policy. */
4189         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4190             (slave->link == BOND_LINK_UP)) {
4191                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4192         }
4193
4194         return res;
4195 }
4196
4197
4198 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4199 {
4200         /*
4201          * This helper function exists to help dev_pick_tx get the correct
4202          * destination queue.  Using a helper function skips a call to
4203          * skb_tx_hash and will put the skbs in the queue we expect on their
4204          * way down to the bonding driver.
4205          */
4206         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4207
4208         /*
4209          * Save the original txq to restore before passing to the driver
4210          */
4211         bond_queue_mapping(skb) = skb->queue_mapping;
4212
4213         if (unlikely(txq >= dev->real_num_tx_queues)) {
4214                 do {
4215                         txq -= dev->real_num_tx_queues;
4216                 } while (txq >= dev->real_num_tx_queues);
4217         }
4218         return txq;
4219 }
4220
4221 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4222 {
4223         struct bonding *bond = netdev_priv(dev);
4224
4225         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4226                 if (!bond_slave_override(bond, skb))
4227                         return NETDEV_TX_OK;
4228         }
4229
4230         switch (bond->params.mode) {
4231         case BOND_MODE_ROUNDROBIN:
4232                 return bond_xmit_roundrobin(skb, dev);
4233         case BOND_MODE_ACTIVEBACKUP:
4234                 return bond_xmit_activebackup(skb, dev);
4235         case BOND_MODE_XOR:
4236                 return bond_xmit_xor(skb, dev);
4237         case BOND_MODE_BROADCAST:
4238                 return bond_xmit_broadcast(skb, dev);
4239         case BOND_MODE_8023AD:
4240                 return bond_3ad_xmit_xor(skb, dev);
4241         case BOND_MODE_ALB:
4242         case BOND_MODE_TLB:
4243                 return bond_alb_xmit(skb, dev);
4244         default:
4245                 /* Should never happen, mode already checked */
4246                 pr_err("%s: Error: Unknown bonding mode %d\n",
4247                        dev->name, bond->params.mode);
4248                 WARN_ON_ONCE(1);
4249                 dev_kfree_skb(skb);
4250                 return NETDEV_TX_OK;
4251         }
4252 }
4253
4254 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4255 {
4256         struct bonding *bond = netdev_priv(dev);
4257         netdev_tx_t ret = NETDEV_TX_OK;
4258
4259         /*
4260          * If we risk deadlock from transmitting this in the
4261          * netpoll path, tell netpoll to queue the frame for later tx
4262          */
4263         if (is_netpoll_tx_blocked(dev))
4264                 return NETDEV_TX_BUSY;
4265
4266         read_lock(&bond->lock);
4267
4268         if (bond->slave_cnt)
4269                 ret = __bond_start_xmit(skb, dev);
4270         else
4271                 dev_kfree_skb(skb);
4272
4273         read_unlock(&bond->lock);
4274
4275         return ret;
4276 }
4277
4278 /*
4279  * set bond mode specific net device operations
4280  */
4281 void bond_set_mode_ops(struct bonding *bond, int mode)
4282 {
4283         struct net_device *bond_dev = bond->dev;
4284
4285         switch (mode) {
4286         case BOND_MODE_ROUNDROBIN:
4287                 break;
4288         case BOND_MODE_ACTIVEBACKUP:
4289                 break;
4290         case BOND_MODE_XOR:
4291                 bond_set_xmit_hash_policy(bond);
4292                 break;
4293         case BOND_MODE_BROADCAST:
4294                 break;
4295         case BOND_MODE_8023AD:
4296                 bond_set_xmit_hash_policy(bond);
4297                 break;
4298         case BOND_MODE_ALB:
4299                 /* FALLTHRU */
4300         case BOND_MODE_TLB:
4301                 break;
4302         default:
4303                 /* Should never happen, mode already checked */
4304                 pr_err("%s: Error: Unknown bonding mode %d\n",
4305                        bond_dev->name, mode);
4306                 break;
4307         }
4308 }
4309
4310 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4311                                     struct ethtool_drvinfo *drvinfo)
4312 {
4313         strncpy(drvinfo->driver, DRV_NAME, 32);
4314         strncpy(drvinfo->version, DRV_VERSION, 32);
4315         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4316 }
4317
4318 static const struct ethtool_ops bond_ethtool_ops = {
4319         .get_drvinfo            = bond_ethtool_get_drvinfo,
4320         .get_link               = ethtool_op_get_link,
4321 };
4322
4323 static const struct net_device_ops bond_netdev_ops = {
4324         .ndo_init               = bond_init,
4325         .ndo_uninit             = bond_uninit,
4326         .ndo_open               = bond_open,
4327         .ndo_stop               = bond_close,
4328         .ndo_start_xmit         = bond_start_xmit,
4329         .ndo_select_queue       = bond_select_queue,
4330         .ndo_get_stats64        = bond_get_stats,
4331         .ndo_do_ioctl           = bond_do_ioctl,
4332         .ndo_set_multicast_list = bond_set_multicast_list,
4333         .ndo_change_mtu         = bond_change_mtu,
4334         .ndo_set_mac_address    = bond_set_mac_address,
4335         .ndo_neigh_setup        = bond_neigh_setup,
4336         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4337         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4338         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4339 #ifdef CONFIG_NET_POLL_CONTROLLER
4340         .ndo_netpoll_setup      = bond_netpoll_setup,
4341         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4342         .ndo_poll_controller    = bond_poll_controller,
4343 #endif
4344         .ndo_add_slave          = bond_enslave,
4345         .ndo_del_slave          = bond_release,
4346         .ndo_fix_features       = bond_fix_features,
4347 };
4348
4349 static void bond_destructor(struct net_device *bond_dev)
4350 {
4351         struct bonding *bond = netdev_priv(bond_dev);
4352         if (bond->wq)
4353                 destroy_workqueue(bond->wq);
4354         free_netdev(bond_dev);
4355 }
4356
4357 static void bond_setup(struct net_device *bond_dev)
4358 {
4359         struct bonding *bond = netdev_priv(bond_dev);
4360
4361         /* initialize rwlocks */
4362         rwlock_init(&bond->lock);
4363         rwlock_init(&bond->curr_slave_lock);
4364
4365         bond->params = bonding_defaults;
4366
4367         /* Initialize pointers */
4368         bond->dev = bond_dev;
4369         INIT_LIST_HEAD(&bond->vlan_list);
4370
4371         /* Initialize the device entry points */
4372         ether_setup(bond_dev);
4373         bond_dev->netdev_ops = &bond_netdev_ops;
4374         bond_dev->ethtool_ops = &bond_ethtool_ops;
4375         bond_set_mode_ops(bond, bond->params.mode);
4376
4377         bond_dev->destructor = bond_destructor;
4378
4379         /* Initialize the device options */
4380         bond_dev->tx_queue_len = 0;
4381         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4382         bond_dev->priv_flags |= IFF_BONDING;
4383         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4384
4385         /* At first, we block adding VLANs. That's the only way to
4386          * prevent problems that occur when adding VLANs over an
4387          * empty bond. The block will be removed once non-challenged
4388          * slaves are enslaved.
4389          */
4390         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4391
4392         /* don't acquire bond device's netif_tx_lock when
4393          * transmitting */
4394         bond_dev->features |= NETIF_F_LLTX;
4395
4396         /* By default, we declare the bond to be fully
4397          * VLAN hardware accelerated capable. Special
4398          * care is taken in the various xmit functions
4399          * when there are slaves that are not hw accel
4400          * capable
4401          */
4402
4403         bond_dev->hw_features = BOND_VLAN_FEATURES |
4404                                 NETIF_F_HW_VLAN_TX |
4405                                 NETIF_F_HW_VLAN_RX |
4406                                 NETIF_F_HW_VLAN_FILTER;
4407
4408         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4409         bond_dev->features |= bond_dev->hw_features;
4410 }
4411
4412 static void bond_work_cancel_all(struct bonding *bond)
4413 {
4414         write_lock_bh(&bond->lock);
4415         bond->kill_timers = 1;
4416         write_unlock_bh(&bond->lock);
4417
4418         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4419                 cancel_delayed_work(&bond->mii_work);
4420
4421         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4422                 cancel_delayed_work(&bond->arp_work);
4423
4424         if (bond->params.mode == BOND_MODE_ALB &&
4425             delayed_work_pending(&bond->alb_work))
4426                 cancel_delayed_work(&bond->alb_work);
4427
4428         if (bond->params.mode == BOND_MODE_8023AD &&
4429             delayed_work_pending(&bond->ad_work))
4430                 cancel_delayed_work(&bond->ad_work);
4431
4432         if (delayed_work_pending(&bond->mcast_work))
4433                 cancel_delayed_work(&bond->mcast_work);
4434 }
4435
4436 /*
4437 * Destroy a bonding device.
4438 * Must be under rtnl_lock when this function is called.
4439 */
4440 static void bond_uninit(struct net_device *bond_dev)
4441 {
4442         struct bonding *bond = netdev_priv(bond_dev);
4443         struct vlan_entry *vlan, *tmp;
4444
4445         bond_netpoll_cleanup(bond_dev);
4446
4447         /* Release the bonded slaves */
4448         bond_release_all(bond_dev);
4449
4450         list_del(&bond->bond_list);
4451
4452         bond_work_cancel_all(bond);
4453
4454         bond_remove_proc_entry(bond);
4455
4456         bond_debug_unregister(bond);
4457
4458         __hw_addr_flush(&bond->mc_list);
4459
4460         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4461                 list_del(&vlan->vlan_list);
4462                 kfree(vlan);
4463         }
4464 }
4465
4466 /*------------------------- Module initialization ---------------------------*/
4467
4468 /*
4469  * Convert string input module parms.  Accept either the
4470  * number of the mode or its string name.  A bit complicated because
4471  * some mode names are substrings of other names, and calls from sysfs
4472  * may have whitespace in the name (trailing newlines, for example).
4473  */
4474 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4475 {
4476         int modeint = -1, i, rv;
4477         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4478
4479         for (p = (char *)buf; *p; p++)
4480                 if (!(isdigit(*p) || isspace(*p)))
4481                         break;
4482
4483         if (*p)
4484                 rv = sscanf(buf, "%20s", modestr);
4485         else
4486                 rv = sscanf(buf, "%d", &modeint);
4487
4488         if (!rv)
4489                 return -1;
4490
4491         for (i = 0; tbl[i].modename; i++) {
4492                 if (modeint == tbl[i].mode)
4493                         return tbl[i].mode;
4494                 if (strcmp(modestr, tbl[i].modename) == 0)
4495                         return tbl[i].mode;
4496         }
4497
4498         return -1;
4499 }
4500
4501 static int bond_check_params(struct bond_params *params)
4502 {
4503         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4504
4505         /*
4506          * Convert string parameters.
4507          */
4508         if (mode) {
4509                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4510                 if (bond_mode == -1) {
4511                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4512                                mode == NULL ? "NULL" : mode);
4513                         return -EINVAL;
4514                 }
4515         }
4516
4517         if (xmit_hash_policy) {
4518                 if ((bond_mode != BOND_MODE_XOR) &&
4519                     (bond_mode != BOND_MODE_8023AD)) {
4520                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4521                                bond_mode_name(bond_mode));
4522                 } else {
4523                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4524                                                         xmit_hashtype_tbl);
4525                         if (xmit_hashtype == -1) {
4526                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4527                                        xmit_hash_policy == NULL ? "NULL" :
4528                                        xmit_hash_policy);
4529                                 return -EINVAL;
4530                         }
4531                 }
4532         }
4533
4534         if (lacp_rate) {
4535                 if (bond_mode != BOND_MODE_8023AD) {
4536                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4537                                 bond_mode_name(bond_mode));
4538                 } else {
4539                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4540                         if (lacp_fast == -1) {
4541                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4542                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4543                                 return -EINVAL;
4544                         }
4545                 }
4546         }
4547
4548         if (ad_select) {
4549                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4550                 if (params->ad_select == -1) {
4551                         pr_err("Error: Invalid ad_select \"%s\"\n",
4552                                ad_select == NULL ? "NULL" : ad_select);
4553                         return -EINVAL;
4554                 }
4555
4556                 if (bond_mode != BOND_MODE_8023AD) {
4557                         pr_warning("ad_select param only affects 802.3ad mode\n");
4558                 }
4559         } else {
4560                 params->ad_select = BOND_AD_STABLE;
4561         }
4562
4563         if (max_bonds < 0) {
4564                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4565                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4566                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4567         }
4568
4569         if (miimon < 0) {
4570                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4571                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4572                 miimon = BOND_LINK_MON_INTERV;
4573         }
4574
4575         if (updelay < 0) {
4576                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4577                            updelay, INT_MAX);
4578                 updelay = 0;
4579         }
4580
4581         if (downdelay < 0) {
4582                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4583                            downdelay, INT_MAX);
4584                 downdelay = 0;
4585         }
4586
4587         if ((use_carrier != 0) && (use_carrier != 1)) {
4588                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4589                            use_carrier);
4590                 use_carrier = 1;
4591         }
4592
4593         if (num_peer_notif < 0 || num_peer_notif > 255) {
4594                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4595                            num_peer_notif);
4596                 num_peer_notif = 1;
4597         }
4598
4599         /* reset values for 802.3ad */
4600         if (bond_mode == BOND_MODE_8023AD) {
4601                 if (!miimon) {
4602                         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");
4603                         pr_warning("Forcing miimon to 100msec\n");
4604                         miimon = 100;
4605                 }
4606         }
4607
4608         if (tx_queues < 1 || tx_queues > 255) {
4609                 pr_warning("Warning: tx_queues (%d) should be between "
4610                            "1 and 255, resetting to %d\n",
4611                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4612                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4613         }
4614
4615         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4616                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4617                            "not of valid value (0/1), so it was set to "
4618                            "0\n", all_slaves_active);
4619                 all_slaves_active = 0;
4620         }
4621
4622         if (resend_igmp < 0 || resend_igmp > 255) {
4623                 pr_warning("Warning: resend_igmp (%d) should be between "
4624                            "0 and 255, resetting to %d\n",
4625                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4626                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4627         }
4628
4629         /* reset values for TLB/ALB */
4630         if ((bond_mode == BOND_MODE_TLB) ||
4631             (bond_mode == BOND_MODE_ALB)) {
4632                 if (!miimon) {
4633                         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");
4634                         pr_warning("Forcing miimon to 100msec\n");
4635                         miimon = 100;
4636                 }
4637         }
4638
4639         if (bond_mode == BOND_MODE_ALB) {
4640                 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",
4641                           updelay);
4642         }
4643
4644         if (!miimon) {
4645                 if (updelay || downdelay) {
4646                         /* just warn the user the up/down delay will have
4647                          * no effect since miimon is zero...
4648                          */
4649                         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",
4650                                    updelay, downdelay);
4651                 }
4652         } else {
4653                 /* don't allow arp monitoring */
4654                 if (arp_interval) {
4655                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4656                                    miimon, arp_interval);
4657                         arp_interval = 0;
4658                 }
4659
4660                 if ((updelay % miimon) != 0) {
4661                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4662                                    updelay, miimon,
4663                                    (updelay / miimon) * miimon);
4664                 }
4665
4666                 updelay /= miimon;
4667
4668                 if ((downdelay % miimon) != 0) {
4669                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4670                                    downdelay, miimon,
4671                                    (downdelay / miimon) * miimon);
4672                 }
4673
4674                 downdelay /= miimon;
4675         }
4676
4677         if (arp_interval < 0) {
4678                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4679                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4680                 arp_interval = BOND_LINK_ARP_INTERV;
4681         }
4682
4683         for (arp_ip_count = 0;
4684              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4685              arp_ip_count++) {
4686                 /* not complete check, but should be good enough to
4687                    catch mistakes */
4688                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4689                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4690                                    arp_ip_target[arp_ip_count]);
4691                         arp_interval = 0;
4692                 } else {
4693                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4694                         arp_target[arp_ip_count] = ip;
4695                 }
4696         }
4697
4698         if (arp_interval && !arp_ip_count) {
4699                 /* don't allow arping if no arp_ip_target given... */
4700                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4701                            arp_interval);
4702                 arp_interval = 0;
4703         }
4704
4705         if (arp_validate) {
4706                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4707                         pr_err("arp_validate only supported in active-backup mode\n");
4708                         return -EINVAL;
4709                 }
4710                 if (!arp_interval) {
4711                         pr_err("arp_validate requires arp_interval\n");
4712                         return -EINVAL;
4713                 }
4714
4715                 arp_validate_value = bond_parse_parm(arp_validate,
4716                                                      arp_validate_tbl);
4717                 if (arp_validate_value == -1) {
4718                         pr_err("Error: invalid arp_validate \"%s\"\n",
4719                                arp_validate == NULL ? "NULL" : arp_validate);
4720                         return -EINVAL;
4721                 }
4722         } else
4723                 arp_validate_value = 0;
4724
4725         if (miimon) {
4726                 pr_info("MII link monitoring set to %d ms\n", miimon);
4727         } else if (arp_interval) {
4728                 int i;
4729
4730                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4731                         arp_interval,
4732                         arp_validate_tbl[arp_validate_value].modename,
4733                         arp_ip_count);
4734
4735                 for (i = 0; i < arp_ip_count; i++)
4736                         pr_info(" %s", arp_ip_target[i]);
4737
4738                 pr_info("\n");
4739
4740         } else if (max_bonds) {
4741                 /* miimon and arp_interval not set, we need one so things
4742                  * work as expected, see bonding.txt for details
4743                  */
4744                 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");
4745         }
4746
4747         if (primary && !USES_PRIMARY(bond_mode)) {
4748                 /* currently, using a primary only makes sense
4749                  * in active backup, TLB or ALB modes
4750                  */
4751                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4752                            primary, bond_mode_name(bond_mode));
4753                 primary = NULL;
4754         }
4755
4756         if (primary && primary_reselect) {
4757                 primary_reselect_value = bond_parse_parm(primary_reselect,
4758                                                          pri_reselect_tbl);
4759                 if (primary_reselect_value == -1) {
4760                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4761                                primary_reselect ==
4762                                         NULL ? "NULL" : primary_reselect);
4763                         return -EINVAL;
4764                 }
4765         } else {
4766                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4767         }
4768
4769         if (fail_over_mac) {
4770                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4771                                                       fail_over_mac_tbl);
4772                 if (fail_over_mac_value == -1) {
4773                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4774                                arp_validate == NULL ? "NULL" : arp_validate);
4775                         return -EINVAL;
4776                 }
4777
4778                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4779                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4780         } else {
4781                 fail_over_mac_value = BOND_FOM_NONE;
4782         }
4783
4784         /* fill params struct with the proper values */
4785         params->mode = bond_mode;
4786         params->xmit_policy = xmit_hashtype;
4787         params->miimon = miimon;
4788         params->num_peer_notif = num_peer_notif;
4789         params->arp_interval = arp_interval;
4790         params->arp_validate = arp_validate_value;
4791         params->updelay = updelay;
4792         params->downdelay = downdelay;
4793         params->use_carrier = use_carrier;
4794         params->lacp_fast = lacp_fast;
4795         params->primary[0] = 0;
4796         params->primary_reselect = primary_reselect_value;
4797         params->fail_over_mac = fail_over_mac_value;
4798         params->tx_queues = tx_queues;
4799         params->all_slaves_active = all_slaves_active;
4800         params->resend_igmp = resend_igmp;
4801
4802         if (primary) {
4803                 strncpy(params->primary, primary, IFNAMSIZ);
4804                 params->primary[IFNAMSIZ - 1] = 0;
4805         }
4806
4807         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4808
4809         return 0;
4810 }
4811
4812 static struct lock_class_key bonding_netdev_xmit_lock_key;
4813 static struct lock_class_key bonding_netdev_addr_lock_key;
4814
4815 static void bond_set_lockdep_class_one(struct net_device *dev,
4816                                        struct netdev_queue *txq,
4817                                        void *_unused)
4818 {
4819         lockdep_set_class(&txq->_xmit_lock,
4820                           &bonding_netdev_xmit_lock_key);
4821 }
4822
4823 static void bond_set_lockdep_class(struct net_device *dev)
4824 {
4825         lockdep_set_class(&dev->addr_list_lock,
4826                           &bonding_netdev_addr_lock_key);
4827         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4828 }
4829
4830 /*
4831  * Called from registration process
4832  */
4833 static int bond_init(struct net_device *bond_dev)
4834 {
4835         struct bonding *bond = netdev_priv(bond_dev);
4836         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4837         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4838
4839         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4840
4841         /*
4842          * Initialize locks that may be required during
4843          * en/deslave operations.  All of the bond_open work
4844          * (of which this is part) should really be moved to
4845          * a phase prior to dev_open
4846          */
4847         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4848         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4849
4850         bond->wq = create_singlethread_workqueue(bond_dev->name);
4851         if (!bond->wq)
4852                 return -ENOMEM;
4853
4854         bond_set_lockdep_class(bond_dev);
4855
4856         bond_create_proc_entry(bond);
4857         list_add_tail(&bond->bond_list, &bn->dev_list);
4858
4859         bond_prepare_sysfs_group(bond);
4860
4861         bond_debug_register(bond);
4862
4863         __hw_addr_init(&bond->mc_list);
4864         return 0;
4865 }
4866
4867 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4868 {
4869         if (tb[IFLA_ADDRESS]) {
4870                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4871                         return -EINVAL;
4872                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4873                         return -EADDRNOTAVAIL;
4874         }
4875         return 0;
4876 }
4877
4878 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4879         .kind           = "bond",
4880         .priv_size      = sizeof(struct bonding),
4881         .setup          = bond_setup,
4882         .validate       = bond_validate,
4883 };
4884
4885 /* Create a new bond based on the specified name and bonding parameters.
4886  * If name is NULL, obtain a suitable "bond%d" name for us.
4887  * Caller must NOT hold rtnl_lock; we need to release it here before we
4888  * set up our sysfs entries.
4889  */
4890 int bond_create(struct net *net, const char *name)
4891 {
4892         struct net_device *bond_dev;
4893         int res;
4894
4895         rtnl_lock();
4896
4897         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4898                                    name ? name : "bond%d",
4899                                    bond_setup, tx_queues);
4900         if (!bond_dev) {
4901                 pr_err("%s: eek! can't alloc netdev!\n", name);
4902                 rtnl_unlock();
4903                 return -ENOMEM;
4904         }
4905
4906         dev_net_set(bond_dev, net);
4907         bond_dev->rtnl_link_ops = &bond_link_ops;
4908
4909         res = register_netdevice(bond_dev);
4910
4911         netif_carrier_off(bond_dev);
4912
4913         rtnl_unlock();
4914         if (res < 0)
4915                 bond_destructor(bond_dev);
4916         return res;
4917 }
4918
4919 static int __net_init bond_net_init(struct net *net)
4920 {
4921         struct bond_net *bn = net_generic(net, bond_net_id);
4922
4923         bn->net = net;
4924         INIT_LIST_HEAD(&bn->dev_list);
4925
4926         bond_create_proc_dir(bn);
4927         
4928         return 0;
4929 }
4930
4931 static void __net_exit bond_net_exit(struct net *net)
4932 {
4933         struct bond_net *bn = net_generic(net, bond_net_id);
4934
4935         bond_destroy_proc_dir(bn);
4936 }
4937
4938 static struct pernet_operations bond_net_ops = {
4939         .init = bond_net_init,
4940         .exit = bond_net_exit,
4941         .id   = &bond_net_id,
4942         .size = sizeof(struct bond_net),
4943 };
4944
4945 static int __init bonding_init(void)
4946 {
4947         int i;
4948         int res;
4949
4950         pr_info("%s", bond_version);
4951
4952         res = bond_check_params(&bonding_defaults);
4953         if (res)
4954                 goto out;
4955
4956         res = register_pernet_subsys(&bond_net_ops);
4957         if (res)
4958                 goto out;
4959
4960         res = rtnl_link_register(&bond_link_ops);
4961         if (res)
4962                 goto err_link;
4963
4964         bond_create_debugfs();
4965
4966         for (i = 0; i < max_bonds; i++) {
4967                 res = bond_create(&init_net, NULL);
4968                 if (res)
4969                         goto err;
4970         }
4971
4972         res = bond_create_sysfs();
4973         if (res)
4974                 goto err;
4975
4976         register_netdevice_notifier(&bond_netdev_notifier);
4977         register_inetaddr_notifier(&bond_inetaddr_notifier);
4978 out:
4979         return res;
4980 err:
4981         rtnl_link_unregister(&bond_link_ops);
4982 err_link:
4983         unregister_pernet_subsys(&bond_net_ops);
4984         goto out;
4985
4986 }
4987
4988 static void __exit bonding_exit(void)
4989 {
4990         unregister_netdevice_notifier(&bond_netdev_notifier);
4991         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4992
4993         bond_destroy_sysfs();
4994         bond_destroy_debugfs();
4995
4996         rtnl_link_unregister(&bond_link_ops);
4997         unregister_pernet_subsys(&bond_net_ops);
4998
4999 #ifdef CONFIG_NET_POLL_CONTROLLER
5000         /*
5001          * Make sure we don't have an imbalance on our netpoll blocking
5002          */
5003         WARN_ON(atomic_read(&netpoll_block_tx));
5004 #endif
5005 }
5006
5007 module_init(bonding_init);
5008 module_exit(bonding_exit);
5009 MODULE_LICENSE("GPL");
5010 MODULE_VERSION(DRV_VERSION);
5011 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5012 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5013 MODULE_ALIAS_RTNL_LINK("bond");