bonding: documentation and code cleanup for resend_igmp
[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 failover event (alias of num_unsol_na)");
117 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
118 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on failover event (alias of num_grat_arp)");
119 module_param(miimon, int, 0);
120 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
121 module_param(updelay, int, 0);
122 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
123 module_param(downdelay, int, 0);
124 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
125                             "in milliseconds");
126 module_param(use_carrier, int, 0);
127 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
128                               "0 for off, 1 for on (default)");
129 module_param(mode, charp, 0);
130 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
131                        "1 for active-backup, 2 for balance-xor, "
132                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
133                        "6 for balance-alb");
134 module_param(primary, charp, 0);
135 MODULE_PARM_DESC(primary, "Primary network device to use");
136 module_param(primary_reselect, charp, 0);
137 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
138                                    "once it comes up; "
139                                    "0 for always (default), "
140                                    "1 for only if speed of primary is "
141                                    "better, "
142                                    "2 for only on active slave "
143                                    "failure");
144 module_param(lacp_rate, charp, 0);
145 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
146                             "(slow/fast)");
147 module_param(ad_select, charp, 0);
148 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
149 module_param(xmit_hash_policy, charp, 0);
150 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
151                                    ", 1 for layer 3+4");
152 module_param(arp_interval, int, 0);
153 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
154 module_param_array(arp_ip_target, charp, NULL, 0);
155 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
156 module_param(arp_validate, charp, 0);
157 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
158 module_param(fail_over_mac, charp, 0);
159 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
160 module_param(all_slaves_active, int, 0);
161 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
162                                      "by setting active flag for all slaves.  "
163                                      "0 for never (default), 1 for always.");
164 module_param(resend_igmp, int, 0);
165 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
166
167 /*----------------------------- Global variables ----------------------------*/
168
169 #ifdef CONFIG_NET_POLL_CONTROLLER
170 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
171 #endif
172
173 int bond_net_id __read_mostly;
174
175 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
176 static int arp_ip_count;
177 static int bond_mode    = BOND_MODE_ROUNDROBIN;
178 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
179 static int lacp_fast;
180
181 const struct bond_parm_tbl bond_lacp_tbl[] = {
182 {       "slow",         AD_LACP_SLOW},
183 {       "fast",         AD_LACP_FAST},
184 {       NULL,           -1},
185 };
186
187 const struct bond_parm_tbl bond_mode_tbl[] = {
188 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
189 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
190 {       "balance-xor",          BOND_MODE_XOR},
191 {       "broadcast",            BOND_MODE_BROADCAST},
192 {       "802.3ad",              BOND_MODE_8023AD},
193 {       "balance-tlb",          BOND_MODE_TLB},
194 {       "balance-alb",          BOND_MODE_ALB},
195 {       NULL,                   -1},
196 };
197
198 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
199 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
200 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
201 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
202 {       NULL,                   -1},
203 };
204
205 const struct bond_parm_tbl arp_validate_tbl[] = {
206 {       "none",                 BOND_ARP_VALIDATE_NONE},
207 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
208 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
209 {       "all",                  BOND_ARP_VALIDATE_ALL},
210 {       NULL,                   -1},
211 };
212
213 const struct bond_parm_tbl fail_over_mac_tbl[] = {
214 {       "none",                 BOND_FOM_NONE},
215 {       "active",               BOND_FOM_ACTIVE},
216 {       "follow",               BOND_FOM_FOLLOW},
217 {       NULL,                   -1},
218 };
219
220 const struct bond_parm_tbl pri_reselect_tbl[] = {
221 {       "always",               BOND_PRI_RESELECT_ALWAYS},
222 {       "better",               BOND_PRI_RESELECT_BETTER},
223 {       "failure",              BOND_PRI_RESELECT_FAILURE},
224 {       NULL,                   -1},
225 };
226
227 struct bond_parm_tbl ad_select_tbl[] = {
228 {       "stable",       BOND_AD_STABLE},
229 {       "bandwidth",    BOND_AD_BANDWIDTH},
230 {       "count",        BOND_AD_COUNT},
231 {       NULL,           -1},
232 };
233
234 /*-------------------------- Forward declarations ---------------------------*/
235
236 static int bond_init(struct net_device *bond_dev);
237 static void bond_uninit(struct net_device *bond_dev);
238
239 /*---------------------------- General routines -----------------------------*/
240
241 const char *bond_mode_name(int mode)
242 {
243         static const char *names[] = {
244                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
245                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
246                 [BOND_MODE_XOR] = "load balancing (xor)",
247                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
248                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
249                 [BOND_MODE_TLB] = "transmit load balancing",
250                 [BOND_MODE_ALB] = "adaptive load balancing",
251         };
252
253         if (mode < 0 || mode > BOND_MODE_ALB)
254                 return "unknown";
255
256         return names[mode];
257 }
258
259 /*---------------------------------- VLAN -----------------------------------*/
260
261 /**
262  * bond_add_vlan - add a new vlan id on bond
263  * @bond: bond that got the notification
264  * @vlan_id: the vlan id to add
265  *
266  * Returns -ENOMEM if allocation failed.
267  */
268 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
269 {
270         struct vlan_entry *vlan;
271
272         pr_debug("bond: %s, vlan id %d\n",
273                  (bond ? bond->dev->name : "None"), vlan_id);
274
275         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
276         if (!vlan)
277                 return -ENOMEM;
278
279         INIT_LIST_HEAD(&vlan->vlan_list);
280         vlan->vlan_id = vlan_id;
281
282         write_lock_bh(&bond->lock);
283
284         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
285
286         write_unlock_bh(&bond->lock);
287
288         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
289
290         return 0;
291 }
292
293 /**
294  * bond_del_vlan - delete a vlan id from bond
295  * @bond: bond that got the notification
296  * @vlan_id: the vlan id to delete
297  *
298  * returns -ENODEV if @vlan_id was not found in @bond.
299  */
300 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
301 {
302         struct vlan_entry *vlan;
303         int res = -ENODEV;
304
305         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
306
307         block_netpoll_tx();
308         write_lock_bh(&bond->lock);
309
310         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
311                 if (vlan->vlan_id == vlan_id) {
312                         list_del(&vlan->vlan_list);
313
314                         if (bond_is_lb(bond))
315                                 bond_alb_clear_vlan(bond, vlan_id);
316
317                         pr_debug("removed VLAN ID %d from bond %s\n",
318                                  vlan_id, bond->dev->name);
319
320                         kfree(vlan);
321
322                         if (list_empty(&bond->vlan_list) &&
323                             (bond->slave_cnt == 0)) {
324                                 /* Last VLAN removed and no slaves, so
325                                  * restore block on adding VLANs. This will
326                                  * be removed once new slaves that are not
327                                  * VLAN challenged will be added.
328                                  */
329                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
330                         }
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 /**
382  * bond_dev_queue_xmit - Prepare skb for xmit.
383  *
384  * @bond: bond device that got this skb for tx.
385  * @skb: hw accel VLAN tagged skb to transmit
386  * @slave_dev: slave that is supposed to xmit this skbuff
387  */
388 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
389                         struct net_device *slave_dev)
390 {
391         skb->dev = slave_dev;
392         skb->priority = 1;
393         if (unlikely(netpoll_tx_running(slave_dev)))
394                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
395         else
396                 dev_queue_xmit(skb);
397
398         return 0;
399 }
400
401 /*
402  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
403  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
404  * lock because:
405  * a. This operation is performed in IOCTL context,
406  * b. The operation is protected by the RTNL semaphore in the 8021q code,
407  * c. Holding a lock with BH disabled while directly calling a base driver
408  *    entry point is generally a BAD idea.
409  *
410  * The design of synchronization/protection for this operation in the 8021q
411  * module is good for one or more VLAN devices over a single physical device
412  * and cannot be extended for a teaming solution like bonding, so there is a
413  * potential race condition here where a net device from the vlan group might
414  * be referenced (either by a base driver or the 8021q code) while it is being
415  * removed from the system. However, it turns out we're not making matters
416  * worse, and if it works for regular VLAN usage it will work here too.
417 */
418
419 /**
420  * bond_vlan_rx_register - Propagates registration to slaves
421  * @bond_dev: bonding net device that got called
422  * @grp: vlan group being registered
423  */
424 static void bond_vlan_rx_register(struct net_device *bond_dev,
425                                   struct vlan_group *grp)
426 {
427         struct bonding *bond = netdev_priv(bond_dev);
428         struct slave *slave;
429         int i;
430
431         write_lock_bh(&bond->lock);
432         bond->vlgrp = grp;
433         write_unlock_bh(&bond->lock);
434
435         bond_for_each_slave(bond, slave, i) {
436                 struct net_device *slave_dev = slave->dev;
437                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
438
439                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
440                     slave_ops->ndo_vlan_rx_register) {
441                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
442                 }
443         }
444 }
445
446 /**
447  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
448  * @bond_dev: bonding net device that got called
449  * @vid: vlan id being added
450  */
451 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
452 {
453         struct bonding *bond = netdev_priv(bond_dev);
454         struct slave *slave;
455         int i, res;
456
457         bond_for_each_slave(bond, slave, i) {
458                 struct net_device *slave_dev = slave->dev;
459                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
460
461                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
462                     slave_ops->ndo_vlan_rx_add_vid) {
463                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
464                 }
465         }
466
467         res = bond_add_vlan(bond, vid);
468         if (res) {
469                 pr_err("%s: Error: Failed to add vlan id %d\n",
470                        bond_dev->name, vid);
471         }
472 }
473
474 /**
475  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
476  * @bond_dev: bonding net device that got called
477  * @vid: vlan id being removed
478  */
479 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
480 {
481         struct bonding *bond = netdev_priv(bond_dev);
482         struct slave *slave;
483         struct net_device *vlan_dev;
484         int i, res;
485
486         bond_for_each_slave(bond, slave, i) {
487                 struct net_device *slave_dev = slave->dev;
488                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
489
490                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
491                     slave_ops->ndo_vlan_rx_kill_vid) {
492                         /* Save and then restore vlan_dev in the grp array,
493                          * since the slave's driver might clear it.
494                          */
495                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
496                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
497                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
498                 }
499         }
500
501         res = bond_del_vlan(bond, vid);
502         if (res) {
503                 pr_err("%s: Error: Failed to remove vlan id %d\n",
504                        bond_dev->name, vid);
505         }
506 }
507
508 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
509 {
510         struct vlan_entry *vlan;
511         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
512
513         if (!bond->vlgrp)
514                 return;
515
516         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
517             slave_ops->ndo_vlan_rx_register)
518                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
519
520         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
521             !(slave_ops->ndo_vlan_rx_add_vid))
522                 return;
523
524         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
525                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
526 }
527
528 static void bond_del_vlans_from_slave(struct bonding *bond,
529                                       struct net_device *slave_dev)
530 {
531         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
532         struct vlan_entry *vlan;
533         struct net_device *vlan_dev;
534
535         if (!bond->vlgrp)
536                 return;
537
538         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
539             !(slave_ops->ndo_vlan_rx_kill_vid))
540                 goto unreg;
541
542         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
543                 if (!vlan->vlan_id)
544                         continue;
545                 /* Save and then restore vlan_dev in the grp array,
546                  * since the slave's driver might clear it.
547                  */
548                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
549                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
550                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
551         }
552
553 unreg:
554         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
555             slave_ops->ndo_vlan_rx_register)
556                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
557 }
558
559 /*------------------------------- Link status -------------------------------*/
560
561 /*
562  * Set the carrier state for the master according to the state of its
563  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
564  * do special 802.3ad magic.
565  *
566  * Returns zero if carrier state does not change, nonzero if it does.
567  */
568 static int bond_set_carrier(struct bonding *bond)
569 {
570         struct slave *slave;
571         int i;
572
573         if (bond->slave_cnt == 0)
574                 goto down;
575
576         if (bond->params.mode == BOND_MODE_8023AD)
577                 return bond_3ad_set_carrier(bond);
578
579         bond_for_each_slave(bond, slave, i) {
580                 if (slave->link == BOND_LINK_UP) {
581                         if (!netif_carrier_ok(bond->dev)) {
582                                 netif_carrier_on(bond->dev);
583                                 return 1;
584                         }
585                         return 0;
586                 }
587         }
588
589 down:
590         if (netif_carrier_ok(bond->dev)) {
591                 netif_carrier_off(bond->dev);
592                 return 1;
593         }
594         return 0;
595 }
596
597 /*
598  * Get link speed and duplex from the slave's base driver
599  * using ethtool. If for some reason the call fails or the
600  * values are invalid, fake speed and duplex to 100/Full
601  * and return error.
602  */
603 static int bond_update_speed_duplex(struct slave *slave)
604 {
605         struct net_device *slave_dev = slave->dev;
606         struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
607         u32 slave_speed;
608         int res;
609
610         /* Fake speed and duplex */
611         slave->speed = SPEED_100;
612         slave->duplex = DUPLEX_FULL;
613
614         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
615                 return -1;
616
617         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
618         if (res < 0)
619                 return -1;
620
621         slave_speed = ethtool_cmd_speed(&etool);
622         switch (slave_speed) {
623         case SPEED_10:
624         case SPEED_100:
625         case SPEED_1000:
626         case SPEED_10000:
627                 break;
628         default:
629                 return -1;
630         }
631
632         switch (etool.duplex) {
633         case DUPLEX_FULL:
634         case DUPLEX_HALF:
635                 break;
636         default:
637                 return -1;
638         }
639
640         slave->speed = slave_speed;
641         slave->duplex = etool.duplex;
642
643         return 0;
644 }
645
646 /*
647  * if <dev> supports MII link status reporting, check its link status.
648  *
649  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
650  * depending upon the setting of the use_carrier parameter.
651  *
652  * Return either BMSR_LSTATUS, meaning that the link is up (or we
653  * can't tell and just pretend it is), or 0, meaning that the link is
654  * down.
655  *
656  * If reporting is non-zero, instead of faking link up, return -1 if
657  * both ETHTOOL and MII ioctls fail (meaning the device does not
658  * support them).  If use_carrier is set, return whatever it says.
659  * It'd be nice if there was a good way to tell if a driver supports
660  * netif_carrier, but there really isn't.
661  */
662 static int bond_check_dev_link(struct bonding *bond,
663                                struct net_device *slave_dev, int reporting)
664 {
665         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
666         int (*ioctl)(struct net_device *, struct ifreq *, int);
667         struct ifreq ifr;
668         struct mii_ioctl_data *mii;
669
670         if (!reporting && !netif_running(slave_dev))
671                 return 0;
672
673         if (bond->params.use_carrier)
674                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
675
676         /* Try to get link status using Ethtool first. */
677         if (slave_dev->ethtool_ops) {
678                 if (slave_dev->ethtool_ops->get_link) {
679                         u32 link;
680
681                         link = slave_dev->ethtool_ops->get_link(slave_dev);
682
683                         return link ? BMSR_LSTATUS : 0;
684                 }
685         }
686
687         /* Ethtool can't be used, fallback to MII ioctls. */
688         ioctl = slave_ops->ndo_do_ioctl;
689         if (ioctl) {
690                 /* TODO: set pointer to correct ioctl on a per team member */
691                 /*       bases to make this more efficient. that is, once  */
692                 /*       we determine the correct ioctl, we will always    */
693                 /*       call it and not the others for that team          */
694                 /*       member.                                           */
695
696                 /*
697                  * We cannot assume that SIOCGMIIPHY will also read a
698                  * register; not all network drivers (e.g., e100)
699                  * support that.
700                  */
701
702                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
703                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
704                 mii = if_mii(&ifr);
705                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
706                         mii->reg_num = MII_BMSR;
707                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
708                                 return mii->val_out & BMSR_LSTATUS;
709                 }
710         }
711
712         /*
713          * If reporting, report that either there's no dev->do_ioctl,
714          * or both SIOCGMIIREG and get_link failed (meaning that we
715          * cannot report link status).  If not reporting, pretend
716          * we're ok.
717          */
718         return reporting ? -1 : BMSR_LSTATUS;
719 }
720
721 /*----------------------------- Multicast list ------------------------------*/
722
723 /*
724  * Push the promiscuity flag down to appropriate slaves
725  */
726 static int bond_set_promiscuity(struct bonding *bond, int inc)
727 {
728         int err = 0;
729         if (USES_PRIMARY(bond->params.mode)) {
730                 /* write lock already acquired */
731                 if (bond->curr_active_slave) {
732                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
733                                                   inc);
734                 }
735         } else {
736                 struct slave *slave;
737                 int i;
738                 bond_for_each_slave(bond, slave, i) {
739                         err = dev_set_promiscuity(slave->dev, inc);
740                         if (err)
741                                 return err;
742                 }
743         }
744         return err;
745 }
746
747 /*
748  * Push the allmulti flag down to all slaves
749  */
750 static int bond_set_allmulti(struct bonding *bond, int inc)
751 {
752         int err = 0;
753         if (USES_PRIMARY(bond->params.mode)) {
754                 /* write lock already acquired */
755                 if (bond->curr_active_slave) {
756                         err = dev_set_allmulti(bond->curr_active_slave->dev,
757                                                inc);
758                 }
759         } else {
760                 struct slave *slave;
761                 int i;
762                 bond_for_each_slave(bond, slave, i) {
763                         err = dev_set_allmulti(slave->dev, inc);
764                         if (err)
765                                 return err;
766                 }
767         }
768         return err;
769 }
770
771 /*
772  * Add a Multicast address to slaves
773  * according to mode
774  */
775 static void bond_mc_add(struct bonding *bond, void *addr)
776 {
777         if (USES_PRIMARY(bond->params.mode)) {
778                 /* write lock already acquired */
779                 if (bond->curr_active_slave)
780                         dev_mc_add(bond->curr_active_slave->dev, addr);
781         } else {
782                 struct slave *slave;
783                 int i;
784
785                 bond_for_each_slave(bond, slave, i)
786                         dev_mc_add(slave->dev, addr);
787         }
788 }
789
790 /*
791  * Remove a multicast address from slave
792  * according to mode
793  */
794 static void bond_mc_del(struct bonding *bond, void *addr)
795 {
796         if (USES_PRIMARY(bond->params.mode)) {
797                 /* write lock already acquired */
798                 if (bond->curr_active_slave)
799                         dev_mc_del(bond->curr_active_slave->dev, addr);
800         } else {
801                 struct slave *slave;
802                 int i;
803                 bond_for_each_slave(bond, slave, i) {
804                         dev_mc_del(slave->dev, addr);
805                 }
806         }
807 }
808
809
810 static void __bond_resend_igmp_join_requests(struct net_device *dev)
811 {
812         struct in_device *in_dev;
813
814         rcu_read_lock();
815         in_dev = __in_dev_get_rcu(dev);
816         if (in_dev)
817                 ip_mc_rejoin_groups(in_dev);
818         rcu_read_unlock();
819 }
820
821 /*
822  * Retrieve the list of registered multicast addresses for the bonding
823  * device and retransmit an IGMP JOIN request to the current active
824  * slave.
825  */
826 static void bond_resend_igmp_join_requests(struct bonding *bond)
827 {
828         struct net_device *vlan_dev;
829         struct vlan_entry *vlan;
830
831         read_lock(&bond->lock);
832
833         /* rejoin all groups on bond device */
834         __bond_resend_igmp_join_requests(bond->dev);
835
836         /* rejoin all groups on vlan devices */
837         if (bond->vlgrp) {
838                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
839                         vlan_dev = vlan_group_get_device(bond->vlgrp,
840                                                          vlan->vlan_id);
841                         if (vlan_dev)
842                                 __bond_resend_igmp_join_requests(vlan_dev);
843                 }
844         }
845
846         if (--bond->igmp_retrans > 0)
847                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
848
849         read_unlock(&bond->lock);
850 }
851
852 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
853 {
854         struct bonding *bond = container_of(work, struct bonding,
855                                             mcast_work.work);
856         bond_resend_igmp_join_requests(bond);
857 }
858
859 /*
860  * flush all members of flush->mc_list from device dev->mc_list
861  */
862 static void bond_mc_list_flush(struct net_device *bond_dev,
863                                struct net_device *slave_dev)
864 {
865         struct bonding *bond = netdev_priv(bond_dev);
866         struct netdev_hw_addr *ha;
867
868         netdev_for_each_mc_addr(ha, bond_dev)
869                 dev_mc_del(slave_dev, ha->addr);
870
871         if (bond->params.mode == BOND_MODE_8023AD) {
872                 /* del lacpdu mc addr from mc list */
873                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
874
875                 dev_mc_del(slave_dev, lacpdu_multicast);
876         }
877 }
878
879 /*--------------------------- Active slave change ---------------------------*/
880
881 /*
882  * Update the mc list and multicast-related flags for the new and
883  * old active slaves (if any) according to the multicast mode, and
884  * promiscuous flags unconditionally.
885  */
886 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
887                          struct slave *old_active)
888 {
889         struct netdev_hw_addr *ha;
890
891         if (!USES_PRIMARY(bond->params.mode))
892                 /* nothing to do -  mc list is already up-to-date on
893                  * all slaves
894                  */
895                 return;
896
897         if (old_active) {
898                 if (bond->dev->flags & IFF_PROMISC)
899                         dev_set_promiscuity(old_active->dev, -1);
900
901                 if (bond->dev->flags & IFF_ALLMULTI)
902                         dev_set_allmulti(old_active->dev, -1);
903
904                 netdev_for_each_mc_addr(ha, bond->dev)
905                         dev_mc_del(old_active->dev, ha->addr);
906         }
907
908         if (new_active) {
909                 /* FIXME: Signal errors upstream. */
910                 if (bond->dev->flags & IFF_PROMISC)
911                         dev_set_promiscuity(new_active->dev, 1);
912
913                 if (bond->dev->flags & IFF_ALLMULTI)
914                         dev_set_allmulti(new_active->dev, 1);
915
916                 netdev_for_each_mc_addr(ha, bond->dev)
917                         dev_mc_add(new_active->dev, ha->addr);
918         }
919 }
920
921 /*
922  * bond_do_fail_over_mac
923  *
924  * Perform special MAC address swapping for fail_over_mac settings
925  *
926  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
927  */
928 static void bond_do_fail_over_mac(struct bonding *bond,
929                                   struct slave *new_active,
930                                   struct slave *old_active)
931         __releases(&bond->curr_slave_lock)
932         __releases(&bond->lock)
933         __acquires(&bond->lock)
934         __acquires(&bond->curr_slave_lock)
935 {
936         u8 tmp_mac[ETH_ALEN];
937         struct sockaddr saddr;
938         int rv;
939
940         switch (bond->params.fail_over_mac) {
941         case BOND_FOM_ACTIVE:
942                 if (new_active)
943                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
944                                new_active->dev->addr_len);
945                 break;
946         case BOND_FOM_FOLLOW:
947                 /*
948                  * if new_active && old_active, swap them
949                  * if just old_active, do nothing (going to no active slave)
950                  * if just new_active, set new_active to bond's MAC
951                  */
952                 if (!new_active)
953                         return;
954
955                 write_unlock_bh(&bond->curr_slave_lock);
956                 read_unlock(&bond->lock);
957
958                 if (old_active) {
959                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
960                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
961                                ETH_ALEN);
962                         saddr.sa_family = new_active->dev->type;
963                 } else {
964                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
965                         saddr.sa_family = bond->dev->type;
966                 }
967
968                 rv = dev_set_mac_address(new_active->dev, &saddr);
969                 if (rv) {
970                         pr_err("%s: Error %d setting MAC of slave %s\n",
971                                bond->dev->name, -rv, new_active->dev->name);
972                         goto out;
973                 }
974
975                 if (!old_active)
976                         goto out;
977
978                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
979                 saddr.sa_family = old_active->dev->type;
980
981                 rv = dev_set_mac_address(old_active->dev, &saddr);
982                 if (rv)
983                         pr_err("%s: Error %d setting MAC of slave %s\n",
984                                bond->dev->name, -rv, new_active->dev->name);
985 out:
986                 read_lock(&bond->lock);
987                 write_lock_bh(&bond->curr_slave_lock);
988                 break;
989         default:
990                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
991                        bond->dev->name, bond->params.fail_over_mac);
992                 break;
993         }
994
995 }
996
997 static bool bond_should_change_active(struct bonding *bond)
998 {
999         struct slave *prim = bond->primary_slave;
1000         struct slave *curr = bond->curr_active_slave;
1001
1002         if (!prim || !curr || curr->link != BOND_LINK_UP)
1003                 return true;
1004         if (bond->force_primary) {
1005                 bond->force_primary = false;
1006                 return true;
1007         }
1008         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1009             (prim->speed < curr->speed ||
1010              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1011                 return false;
1012         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1013                 return false;
1014         return true;
1015 }
1016
1017 /**
1018  * find_best_interface - select the best available slave to be the active one
1019  * @bond: our bonding struct
1020  *
1021  * Warning: Caller must hold curr_slave_lock for writing.
1022  */
1023 static struct slave *bond_find_best_slave(struct bonding *bond)
1024 {
1025         struct slave *new_active, *old_active;
1026         struct slave *bestslave = NULL;
1027         int mintime = bond->params.updelay;
1028         int i;
1029
1030         new_active = bond->curr_active_slave;
1031
1032         if (!new_active) { /* there were no active slaves left */
1033                 if (bond->slave_cnt > 0)   /* found one slave */
1034                         new_active = bond->first_slave;
1035                 else
1036                         return NULL; /* still no slave, return NULL */
1037         }
1038
1039         if ((bond->primary_slave) &&
1040             bond->primary_slave->link == BOND_LINK_UP &&
1041             bond_should_change_active(bond)) {
1042                 new_active = bond->primary_slave;
1043         }
1044
1045         /* remember where to stop iterating over the slaves */
1046         old_active = new_active;
1047
1048         bond_for_each_slave_from(bond, new_active, i, old_active) {
1049                 if (new_active->link == BOND_LINK_UP) {
1050                         return new_active;
1051                 } else if (new_active->link == BOND_LINK_BACK &&
1052                            IS_UP(new_active->dev)) {
1053                         /* link up, but waiting for stabilization */
1054                         if (new_active->delay < mintime) {
1055                                 mintime = new_active->delay;
1056                                 bestslave = new_active;
1057                         }
1058                 }
1059         }
1060
1061         return bestslave;
1062 }
1063
1064 static bool bond_should_notify_peers(struct bonding *bond)
1065 {
1066         struct slave *slave = bond->curr_active_slave;
1067
1068         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1069                  bond->dev->name, slave ? slave->dev->name : "NULL");
1070
1071         if (!slave || !bond->send_peer_notif ||
1072             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1073                 return false;
1074
1075         bond->send_peer_notif--;
1076         return true;
1077 }
1078
1079 /**
1080  * change_active_interface - change the active slave into the specified one
1081  * @bond: our bonding struct
1082  * @new: the new slave to make the active one
1083  *
1084  * Set the new slave to the bond's settings and unset them on the old
1085  * curr_active_slave.
1086  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1087  *
1088  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1089  * because it is apparently the best available slave we have, even though its
1090  * updelay hasn't timed out yet.
1091  *
1092  * If new_active is not NULL, caller must hold bond->lock for read and
1093  * curr_slave_lock for write_bh.
1094  */
1095 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1096 {
1097         struct slave *old_active = bond->curr_active_slave;
1098
1099         if (old_active == new_active)
1100                 return;
1101
1102         if (new_active) {
1103                 new_active->jiffies = jiffies;
1104
1105                 if (new_active->link == BOND_LINK_BACK) {
1106                         if (USES_PRIMARY(bond->params.mode)) {
1107                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1108                                         bond->dev->name, new_active->dev->name,
1109                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1110                         }
1111
1112                         new_active->delay = 0;
1113                         new_active->link = BOND_LINK_UP;
1114
1115                         if (bond->params.mode == BOND_MODE_8023AD)
1116                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1117
1118                         if (bond_is_lb(bond))
1119                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1120                 } else {
1121                         if (USES_PRIMARY(bond->params.mode)) {
1122                                 pr_info("%s: making interface %s the new active one.\n",
1123                                         bond->dev->name, new_active->dev->name);
1124                         }
1125                 }
1126         }
1127
1128         if (USES_PRIMARY(bond->params.mode))
1129                 bond_mc_swap(bond, new_active, old_active);
1130
1131         if (bond_is_lb(bond)) {
1132                 bond_alb_handle_active_change(bond, new_active);
1133                 if (old_active)
1134                         bond_set_slave_inactive_flags(old_active);
1135                 if (new_active)
1136                         bond_set_slave_active_flags(new_active);
1137         } else {
1138                 bond->curr_active_slave = new_active;
1139         }
1140
1141         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1142                 if (old_active)
1143                         bond_set_slave_inactive_flags(old_active);
1144
1145                 if (new_active) {
1146                         bool should_notify_peers = false;
1147
1148                         bond_set_slave_active_flags(new_active);
1149
1150                         if (bond->params.fail_over_mac)
1151                                 bond_do_fail_over_mac(bond, new_active,
1152                                                       old_active);
1153
1154                         if (netif_running(bond->dev)) {
1155                                 bond->send_peer_notif =
1156                                         bond->params.num_peer_notif;
1157                                 should_notify_peers =
1158                                         bond_should_notify_peers(bond);
1159                         }
1160
1161                         write_unlock_bh(&bond->curr_slave_lock);
1162                         read_unlock(&bond->lock);
1163
1164                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1165                         if (should_notify_peers)
1166                                 netdev_bonding_change(bond->dev,
1167                                                       NETDEV_NOTIFY_PEERS);
1168
1169                         read_lock(&bond->lock);
1170                         write_lock_bh(&bond->curr_slave_lock);
1171                 }
1172         }
1173
1174         /* resend IGMP joins since active slave has changed or
1175          * all were sent on curr_active_slave.
1176          * resend only if bond is brought up with the affected
1177          * bonding modes and the retransmission is enabled */
1178         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1179             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1180              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1181                 bond->igmp_retrans = bond->params.resend_igmp;
1182                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1183         }
1184 }
1185
1186 /**
1187  * bond_select_active_slave - select a new active slave, if needed
1188  * @bond: our bonding struct
1189  *
1190  * This functions should be called when one of the following occurs:
1191  * - The old curr_active_slave has been released or lost its link.
1192  * - The primary_slave has got its link back.
1193  * - A slave has got its link back and there's no old curr_active_slave.
1194  *
1195  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1196  */
1197 void bond_select_active_slave(struct bonding *bond)
1198 {
1199         struct slave *best_slave;
1200         int rv;
1201
1202         best_slave = bond_find_best_slave(bond);
1203         if (best_slave != bond->curr_active_slave) {
1204                 bond_change_active_slave(bond, best_slave);
1205                 rv = bond_set_carrier(bond);
1206                 if (!rv)
1207                         return;
1208
1209                 if (netif_carrier_ok(bond->dev)) {
1210                         pr_info("%s: first active interface up!\n",
1211                                 bond->dev->name);
1212                 } else {
1213                         pr_info("%s: now running without any active interface !\n",
1214                                 bond->dev->name);
1215                 }
1216         }
1217 }
1218
1219 /*--------------------------- slave list handling ---------------------------*/
1220
1221 /*
1222  * This function attaches the slave to the end of list.
1223  *
1224  * bond->lock held for writing by caller.
1225  */
1226 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1227 {
1228         if (bond->first_slave == NULL) { /* attaching the first slave */
1229                 new_slave->next = new_slave;
1230                 new_slave->prev = new_slave;
1231                 bond->first_slave = new_slave;
1232         } else {
1233                 new_slave->next = bond->first_slave;
1234                 new_slave->prev = bond->first_slave->prev;
1235                 new_slave->next->prev = new_slave;
1236                 new_slave->prev->next = new_slave;
1237         }
1238
1239         bond->slave_cnt++;
1240 }
1241
1242 /*
1243  * This function detaches the slave from the list.
1244  * WARNING: no check is made to verify if the slave effectively
1245  * belongs to <bond>.
1246  * Nothing is freed on return, structures are just unchained.
1247  * If any slave pointer in bond was pointing to <slave>,
1248  * it should be changed by the calling function.
1249  *
1250  * bond->lock held for writing by caller.
1251  */
1252 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1253 {
1254         if (slave->next)
1255                 slave->next->prev = slave->prev;
1256
1257         if (slave->prev)
1258                 slave->prev->next = slave->next;
1259
1260         if (bond->first_slave == slave) { /* slave is the first slave */
1261                 if (bond->slave_cnt > 1) { /* there are more slave */
1262                         bond->first_slave = slave->next;
1263                 } else {
1264                         bond->first_slave = NULL; /* slave was the last one */
1265                 }
1266         }
1267
1268         slave->next = NULL;
1269         slave->prev = NULL;
1270         bond->slave_cnt--;
1271 }
1272
1273 #ifdef CONFIG_NET_POLL_CONTROLLER
1274 static inline int slave_enable_netpoll(struct slave *slave)
1275 {
1276         struct netpoll *np;
1277         int err = 0;
1278
1279         np = kzalloc(sizeof(*np), GFP_KERNEL);
1280         err = -ENOMEM;
1281         if (!np)
1282                 goto out;
1283
1284         np->dev = slave->dev;
1285         err = __netpoll_setup(np);
1286         if (err) {
1287                 kfree(np);
1288                 goto out;
1289         }
1290         slave->np = np;
1291 out:
1292         return err;
1293 }
1294 static inline void slave_disable_netpoll(struct slave *slave)
1295 {
1296         struct netpoll *np = slave->np;
1297
1298         if (!np)
1299                 return;
1300
1301         slave->np = NULL;
1302         synchronize_rcu_bh();
1303         __netpoll_cleanup(np);
1304         kfree(np);
1305 }
1306 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1307 {
1308         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1309                 return false;
1310         if (!slave_dev->netdev_ops->ndo_poll_controller)
1311                 return false;
1312         return true;
1313 }
1314
1315 static void bond_poll_controller(struct net_device *bond_dev)
1316 {
1317 }
1318
1319 static void __bond_netpoll_cleanup(struct bonding *bond)
1320 {
1321         struct slave *slave;
1322         int i;
1323
1324         bond_for_each_slave(bond, slave, i)
1325                 if (IS_UP(slave->dev))
1326                         slave_disable_netpoll(slave);
1327 }
1328 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1329 {
1330         struct bonding *bond = netdev_priv(bond_dev);
1331
1332         read_lock(&bond->lock);
1333         __bond_netpoll_cleanup(bond);
1334         read_unlock(&bond->lock);
1335 }
1336
1337 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1338 {
1339         struct bonding *bond = netdev_priv(dev);
1340         struct slave *slave;
1341         int i, err = 0;
1342
1343         read_lock(&bond->lock);
1344         bond_for_each_slave(bond, slave, i) {
1345                 err = slave_enable_netpoll(slave);
1346                 if (err) {
1347                         __bond_netpoll_cleanup(bond);
1348                         break;
1349                 }
1350         }
1351         read_unlock(&bond->lock);
1352         return err;
1353 }
1354
1355 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1356 {
1357         return bond->dev->npinfo;
1358 }
1359
1360 #else
1361 static inline int slave_enable_netpoll(struct slave *slave)
1362 {
1363         return 0;
1364 }
1365 static inline void slave_disable_netpoll(struct slave *slave)
1366 {
1367 }
1368 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1369 {
1370 }
1371 #endif
1372
1373 /*---------------------------------- IOCTL ----------------------------------*/
1374
1375 static int bond_sethwaddr(struct net_device *bond_dev,
1376                           struct net_device *slave_dev)
1377 {
1378         pr_debug("bond_dev=%p\n", bond_dev);
1379         pr_debug("slave_dev=%p\n", slave_dev);
1380         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1381         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1382         return 0;
1383 }
1384
1385 static u32 bond_fix_features(struct net_device *dev, u32 features)
1386 {
1387         struct slave *slave;
1388         struct bonding *bond = netdev_priv(dev);
1389         u32 mask;
1390         int i;
1391
1392         read_lock(&bond->lock);
1393
1394         if (!bond->first_slave) {
1395                 /* Disable adding VLANs to empty bond. But why? --mq */
1396                 features |= NETIF_F_VLAN_CHALLENGED;
1397                 goto out;
1398         }
1399
1400         mask = features;
1401         features &= ~NETIF_F_ONE_FOR_ALL;
1402         features |= NETIF_F_ALL_FOR_ALL;
1403
1404         bond_for_each_slave(bond, slave, i) {
1405                 features = netdev_increment_features(features,
1406                                                      slave->dev->features,
1407                                                      mask);
1408         }
1409
1410 out:
1411         read_unlock(&bond->lock);
1412         return features;
1413 }
1414
1415 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_TX_OFFLOADS | \
1416                                  NETIF_F_SOFT_FEATURES | \
1417                                  NETIF_F_LRO)
1418
1419 static void bond_compute_features(struct bonding *bond)
1420 {
1421         struct slave *slave;
1422         struct net_device *bond_dev = bond->dev;
1423         u32 vlan_features = BOND_VLAN_FEATURES;
1424         unsigned short max_hard_header_len = ETH_HLEN;
1425         int i;
1426
1427         read_lock(&bond->lock);
1428
1429         if (!bond->first_slave)
1430                 goto done;
1431
1432         bond_for_each_slave(bond, slave, i) {
1433                 vlan_features = netdev_increment_features(vlan_features,
1434                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1435
1436                 if (slave->dev->hard_header_len > max_hard_header_len)
1437                         max_hard_header_len = slave->dev->hard_header_len;
1438         }
1439
1440 done:
1441         bond_dev->vlan_features = vlan_features;
1442         bond_dev->hard_header_len = max_hard_header_len;
1443
1444         read_unlock(&bond->lock);
1445
1446         netdev_change_features(bond_dev);
1447 }
1448
1449 static void bond_setup_by_slave(struct net_device *bond_dev,
1450                                 struct net_device *slave_dev)
1451 {
1452         struct bonding *bond = netdev_priv(bond_dev);
1453
1454         bond_dev->header_ops        = slave_dev->header_ops;
1455
1456         bond_dev->type              = slave_dev->type;
1457         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1458         bond_dev->addr_len          = slave_dev->addr_len;
1459
1460         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1461                 slave_dev->addr_len);
1462         bond->setup_by_slave = 1;
1463 }
1464
1465 /* On bonding slaves other than the currently active slave, suppress
1466  * duplicates except for alb non-mcast/bcast.
1467  */
1468 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1469                                             struct slave *slave,
1470                                             struct bonding *bond)
1471 {
1472         if (bond_is_slave_inactive(slave)) {
1473                 if (bond->params.mode == BOND_MODE_ALB &&
1474                     skb->pkt_type != PACKET_BROADCAST &&
1475                     skb->pkt_type != PACKET_MULTICAST)
1476                         return false;
1477                 return true;
1478         }
1479         return false;
1480 }
1481
1482 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1483 {
1484         struct sk_buff *skb = *pskb;
1485         struct slave *slave;
1486         struct bonding *bond;
1487
1488         skb = skb_share_check(skb, GFP_ATOMIC);
1489         if (unlikely(!skb))
1490                 return RX_HANDLER_CONSUMED;
1491
1492         *pskb = skb;
1493
1494         slave = bond_slave_get_rcu(skb->dev);
1495         bond = slave->bond;
1496
1497         if (bond->params.arp_interval)
1498                 slave->dev->last_rx = jiffies;
1499
1500         if (bond->recv_probe) {
1501                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1502
1503                 if (likely(nskb)) {
1504                         bond->recv_probe(nskb, bond, slave);
1505                         dev_kfree_skb(nskb);
1506                 }
1507         }
1508
1509         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1510                 return RX_HANDLER_EXACT;
1511         }
1512
1513         skb->dev = bond->dev;
1514
1515         if (bond->params.mode == BOND_MODE_ALB &&
1516             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1517             skb->pkt_type == PACKET_HOST) {
1518
1519                 if (unlikely(skb_cow_head(skb,
1520                                           skb->data - skb_mac_header(skb)))) {
1521                         kfree_skb(skb);
1522                         return RX_HANDLER_CONSUMED;
1523                 }
1524                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1525         }
1526
1527         return RX_HANDLER_ANOTHER;
1528 }
1529
1530 /* enslave device <slave> to bond device <master> */
1531 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1532 {
1533         struct bonding *bond = netdev_priv(bond_dev);
1534         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1535         struct slave *new_slave = NULL;
1536         struct netdev_hw_addr *ha;
1537         struct sockaddr addr;
1538         int link_reporting;
1539         int res = 0;
1540
1541         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1542                 slave_ops->ndo_do_ioctl == NULL) {
1543                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1544                            bond_dev->name, slave_dev->name);
1545         }
1546
1547         /* already enslaved */
1548         if (slave_dev->flags & IFF_SLAVE) {
1549                 pr_debug("Error, Device was already enslaved\n");
1550                 return -EBUSY;
1551         }
1552
1553         /* vlan challenged mutual exclusion */
1554         /* no need to lock since we're protected by rtnl_lock */
1555         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1556                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1557                 if (bond->vlgrp) {
1558                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1559                                bond_dev->name, slave_dev->name, bond_dev->name);
1560                         return -EPERM;
1561                 } else {
1562                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1563                                    bond_dev->name, slave_dev->name,
1564                                    slave_dev->name, bond_dev->name);
1565                 }
1566         } else {
1567                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1568         }
1569
1570         /*
1571          * Old ifenslave binaries are no longer supported.  These can
1572          * be identified with moderate accuracy by the state of the slave:
1573          * the current ifenslave will set the interface down prior to
1574          * enslaving it; the old ifenslave will not.
1575          */
1576         if ((slave_dev->flags & IFF_UP)) {
1577                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1578                        slave_dev->name);
1579                 res = -EPERM;
1580                 goto err_undo_flags;
1581         }
1582
1583         /* set bonding device ether type by slave - bonding netdevices are
1584          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1585          * there is a need to override some of the type dependent attribs/funcs.
1586          *
1587          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1588          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1589          */
1590         if (bond->slave_cnt == 0) {
1591                 if (bond_dev->type != slave_dev->type) {
1592                         pr_debug("%s: change device type from %d to %d\n",
1593                                  bond_dev->name,
1594                                  bond_dev->type, slave_dev->type);
1595
1596                         res = netdev_bonding_change(bond_dev,
1597                                                     NETDEV_PRE_TYPE_CHANGE);
1598                         res = notifier_to_errno(res);
1599                         if (res) {
1600                                 pr_err("%s: refused to change device type\n",
1601                                        bond_dev->name);
1602                                 res = -EBUSY;
1603                                 goto err_undo_flags;
1604                         }
1605
1606                         /* Flush unicast and multicast addresses */
1607                         dev_uc_flush(bond_dev);
1608                         dev_mc_flush(bond_dev);
1609
1610                         if (slave_dev->type != ARPHRD_ETHER)
1611                                 bond_setup_by_slave(bond_dev, slave_dev);
1612                         else
1613                                 ether_setup(bond_dev);
1614
1615                         netdev_bonding_change(bond_dev,
1616                                               NETDEV_POST_TYPE_CHANGE);
1617                 }
1618         } else if (bond_dev->type != slave_dev->type) {
1619                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1620                        slave_dev->name,
1621                        slave_dev->type, bond_dev->type);
1622                 res = -EINVAL;
1623                 goto err_undo_flags;
1624         }
1625
1626         if (slave_ops->ndo_set_mac_address == NULL) {
1627                 if (bond->slave_cnt == 0) {
1628                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1629                                    bond_dev->name);
1630                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1631                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1632                         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",
1633                                bond_dev->name);
1634                         res = -EOPNOTSUPP;
1635                         goto err_undo_flags;
1636                 }
1637         }
1638
1639         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1640
1641         /* If this is the first slave, then we need to set the master's hardware
1642          * address to be the same as the slave's. */
1643         if (is_zero_ether_addr(bond->dev->dev_addr))
1644                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1645                        slave_dev->addr_len);
1646
1647
1648         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1649         if (!new_slave) {
1650                 res = -ENOMEM;
1651                 goto err_undo_flags;
1652         }
1653
1654         /*
1655          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1656          * is set via sysfs or module option if desired.
1657          */
1658         new_slave->queue_id = 0;
1659
1660         /* Save slave's original mtu and then set it to match the bond */
1661         new_slave->original_mtu = slave_dev->mtu;
1662         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1663         if (res) {
1664                 pr_debug("Error %d calling dev_set_mtu\n", res);
1665                 goto err_free;
1666         }
1667
1668         /*
1669          * Save slave's original ("permanent") mac address for modes
1670          * that need it, and for restoring it upon release, and then
1671          * set it to the master's address
1672          */
1673         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1674
1675         if (!bond->params.fail_over_mac) {
1676                 /*
1677                  * Set slave to master's mac address.  The application already
1678                  * set the master's mac address to that of the first slave
1679                  */
1680                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1681                 addr.sa_family = slave_dev->type;
1682                 res = dev_set_mac_address(slave_dev, &addr);
1683                 if (res) {
1684                         pr_debug("Error %d calling set_mac_address\n", res);
1685                         goto err_restore_mtu;
1686                 }
1687         }
1688
1689         res = netdev_set_bond_master(slave_dev, bond_dev);
1690         if (res) {
1691                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1692                 goto err_restore_mac;
1693         }
1694
1695         /* open the slave since the application closed it */
1696         res = dev_open(slave_dev);
1697         if (res) {
1698                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1699                 goto err_unset_master;
1700         }
1701
1702         new_slave->bond = bond;
1703         new_slave->dev = slave_dev;
1704         slave_dev->priv_flags |= IFF_BONDING;
1705
1706         if (bond_is_lb(bond)) {
1707                 /* bond_alb_init_slave() must be called before all other stages since
1708                  * it might fail and we do not want to have to undo everything
1709                  */
1710                 res = bond_alb_init_slave(bond, new_slave);
1711                 if (res)
1712                         goto err_close;
1713         }
1714
1715         /* If the mode USES_PRIMARY, then the new slave gets the
1716          * master's promisc (and mc) settings only if it becomes the
1717          * curr_active_slave, and that is taken care of later when calling
1718          * bond_change_active()
1719          */
1720         if (!USES_PRIMARY(bond->params.mode)) {
1721                 /* set promiscuity level to new slave */
1722                 if (bond_dev->flags & IFF_PROMISC) {
1723                         res = dev_set_promiscuity(slave_dev, 1);
1724                         if (res)
1725                                 goto err_close;
1726                 }
1727
1728                 /* set allmulti level to new slave */
1729                 if (bond_dev->flags & IFF_ALLMULTI) {
1730                         res = dev_set_allmulti(slave_dev, 1);
1731                         if (res)
1732                                 goto err_close;
1733                 }
1734
1735                 netif_addr_lock_bh(bond_dev);
1736                 /* upload master's mc_list to new slave */
1737                 netdev_for_each_mc_addr(ha, bond_dev)
1738                         dev_mc_add(slave_dev, ha->addr);
1739                 netif_addr_unlock_bh(bond_dev);
1740         }
1741
1742         if (bond->params.mode == BOND_MODE_8023AD) {
1743                 /* add lacpdu mc addr to mc list */
1744                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1745
1746                 dev_mc_add(slave_dev, lacpdu_multicast);
1747         }
1748
1749         bond_add_vlans_on_slave(bond, slave_dev);
1750
1751         write_lock_bh(&bond->lock);
1752
1753         bond_attach_slave(bond, new_slave);
1754
1755         new_slave->delay = 0;
1756         new_slave->link_failure_count = 0;
1757
1758         write_unlock_bh(&bond->lock);
1759
1760         bond_compute_features(bond);
1761
1762         read_lock(&bond->lock);
1763
1764         new_slave->last_arp_rx = jiffies;
1765
1766         if (bond->params.miimon && !bond->params.use_carrier) {
1767                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1768
1769                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1770                         /*
1771                          * miimon is set but a bonded network driver
1772                          * does not support ETHTOOL/MII and
1773                          * arp_interval is not set.  Note: if
1774                          * use_carrier is enabled, we will never go
1775                          * here (because netif_carrier is always
1776                          * supported); thus, we don't need to change
1777                          * the messages for netif_carrier.
1778                          */
1779                         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",
1780                                bond_dev->name, slave_dev->name);
1781                 } else if (link_reporting == -1) {
1782                         /* unable get link status using mii/ethtool */
1783                         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",
1784                                    bond_dev->name, slave_dev->name);
1785                 }
1786         }
1787
1788         /* check for initial state */
1789         if (!bond->params.miimon ||
1790             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1791                 if (bond->params.updelay) {
1792                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1793                         new_slave->link  = BOND_LINK_BACK;
1794                         new_slave->delay = bond->params.updelay;
1795                 } else {
1796                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1797                         new_slave->link  = BOND_LINK_UP;
1798                 }
1799                 new_slave->jiffies = jiffies;
1800         } else {
1801                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1802                 new_slave->link  = BOND_LINK_DOWN;
1803         }
1804
1805         if (bond_update_speed_duplex(new_slave) &&
1806             (new_slave->link != BOND_LINK_DOWN)) {
1807                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1808                            bond_dev->name, new_slave->dev->name);
1809
1810                 if (bond->params.mode == BOND_MODE_8023AD) {
1811                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1812                                    bond_dev->name);
1813                 }
1814         }
1815
1816         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1817                 /* if there is a primary slave, remember it */
1818                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1819                         bond->primary_slave = new_slave;
1820                         bond->force_primary = true;
1821                 }
1822         }
1823
1824         write_lock_bh(&bond->curr_slave_lock);
1825
1826         switch (bond->params.mode) {
1827         case BOND_MODE_ACTIVEBACKUP:
1828                 bond_set_slave_inactive_flags(new_slave);
1829                 bond_select_active_slave(bond);
1830                 break;
1831         case BOND_MODE_8023AD:
1832                 /* in 802.3ad mode, the internal mechanism
1833                  * will activate the slaves in the selected
1834                  * aggregator
1835                  */
1836                 bond_set_slave_inactive_flags(new_slave);
1837                 /* if this is the first slave */
1838                 if (bond->slave_cnt == 1) {
1839                         SLAVE_AD_INFO(new_slave).id = 1;
1840                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1841                          * can be called only after the mac address of the bond is set
1842                          */
1843                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1844                                             bond->params.lacp_fast);
1845                 } else {
1846                         SLAVE_AD_INFO(new_slave).id =
1847                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1848                 }
1849
1850                 bond_3ad_bind_slave(new_slave);
1851                 break;
1852         case BOND_MODE_TLB:
1853         case BOND_MODE_ALB:
1854                 bond_set_active_slave(new_slave);
1855                 bond_set_slave_inactive_flags(new_slave);
1856                 bond_select_active_slave(bond);
1857                 break;
1858         default:
1859                 pr_debug("This slave is always active in trunk mode\n");
1860
1861                 /* always active in trunk mode */
1862                 bond_set_active_slave(new_slave);
1863
1864                 /* In trunking mode there is little meaning to curr_active_slave
1865                  * anyway (it holds no special properties of the bond device),
1866                  * so we can change it without calling change_active_interface()
1867                  */
1868                 if (!bond->curr_active_slave)
1869                         bond->curr_active_slave = new_slave;
1870
1871                 break;
1872         } /* switch(bond_mode) */
1873
1874         write_unlock_bh(&bond->curr_slave_lock);
1875
1876         bond_set_carrier(bond);
1877
1878 #ifdef CONFIG_NET_POLL_CONTROLLER
1879         slave_dev->npinfo = bond_netpoll_info(bond);
1880         if (slave_dev->npinfo) {
1881                 if (slave_enable_netpoll(new_slave)) {
1882                         read_unlock(&bond->lock);
1883                         pr_info("Error, %s: master_dev is using netpoll, "
1884                                  "but new slave device does not support netpoll.\n",
1885                                  bond_dev->name);
1886                         res = -EBUSY;
1887                         goto err_close;
1888                 }
1889         }
1890 #endif
1891
1892         read_unlock(&bond->lock);
1893
1894         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1895         if (res)
1896                 goto err_close;
1897
1898         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1899                                          new_slave);
1900         if (res) {
1901                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1902                 goto err_dest_symlinks;
1903         }
1904
1905         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1906                 bond_dev->name, slave_dev->name,
1907                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1908                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1909
1910         /* enslave is successful */
1911         return 0;
1912
1913 /* Undo stages on error */
1914 err_dest_symlinks:
1915         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1916
1917 err_close:
1918         dev_close(slave_dev);
1919
1920 err_unset_master:
1921         netdev_set_bond_master(slave_dev, NULL);
1922
1923 err_restore_mac:
1924         if (!bond->params.fail_over_mac) {
1925                 /* XXX TODO - fom follow mode needs to change master's
1926                  * MAC if this slave's MAC is in use by the bond, or at
1927                  * least print a warning.
1928                  */
1929                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1930                 addr.sa_family = slave_dev->type;
1931                 dev_set_mac_address(slave_dev, &addr);
1932         }
1933
1934 err_restore_mtu:
1935         dev_set_mtu(slave_dev, new_slave->original_mtu);
1936
1937 err_free:
1938         kfree(new_slave);
1939
1940 err_undo_flags:
1941         bond_compute_features(bond);
1942
1943         return res;
1944 }
1945
1946 /*
1947  * Try to release the slave device <slave> from the bond device <master>
1948  * It is legal to access curr_active_slave without a lock because all the function
1949  * is write-locked.
1950  *
1951  * The rules for slave state should be:
1952  *   for Active/Backup:
1953  *     Active stays on all backups go down
1954  *   for Bonded connections:
1955  *     The first up interface should be left on and all others downed.
1956  */
1957 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1958 {
1959         struct bonding *bond = netdev_priv(bond_dev);
1960         struct slave *slave, *oldcurrent;
1961         struct sockaddr addr;
1962         u32 old_features = bond_dev->features;
1963
1964         /* slave is not a slave or master is not master of this slave */
1965         if (!(slave_dev->flags & IFF_SLAVE) ||
1966             (slave_dev->master != bond_dev)) {
1967                 pr_err("%s: Error: cannot release %s.\n",
1968                        bond_dev->name, slave_dev->name);
1969                 return -EINVAL;
1970         }
1971
1972         block_netpoll_tx();
1973         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1974         write_lock_bh(&bond->lock);
1975
1976         slave = bond_get_slave_by_dev(bond, slave_dev);
1977         if (!slave) {
1978                 /* not a slave of this bond */
1979                 pr_info("%s: %s not enslaved\n",
1980                         bond_dev->name, slave_dev->name);
1981                 write_unlock_bh(&bond->lock);
1982                 unblock_netpoll_tx();
1983                 return -EINVAL;
1984         }
1985
1986         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1987          * for this slave anymore.
1988          */
1989         netdev_rx_handler_unregister(slave_dev);
1990         write_unlock_bh(&bond->lock);
1991         synchronize_net();
1992         write_lock_bh(&bond->lock);
1993
1994         if (!bond->params.fail_over_mac) {
1995                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1996                     bond->slave_cnt > 1)
1997                         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",
1998                                    bond_dev->name, slave_dev->name,
1999                                    slave->perm_hwaddr,
2000                                    bond_dev->name, slave_dev->name);
2001         }
2002
2003         /* Inform AD package of unbinding of slave. */
2004         if (bond->params.mode == BOND_MODE_8023AD) {
2005                 /* must be called before the slave is
2006                  * detached from the list
2007                  */
2008                 bond_3ad_unbind_slave(slave);
2009         }
2010
2011         pr_info("%s: releasing %s interface %s\n",
2012                 bond_dev->name,
2013                 bond_is_active_slave(slave) ? "active" : "backup",
2014                 slave_dev->name);
2015
2016         oldcurrent = bond->curr_active_slave;
2017
2018         bond->current_arp_slave = NULL;
2019
2020         /* release the slave from its bond */
2021         bond_detach_slave(bond, slave);
2022
2023         if (bond->primary_slave == slave)
2024                 bond->primary_slave = NULL;
2025
2026         if (oldcurrent == slave)
2027                 bond_change_active_slave(bond, NULL);
2028
2029         if (bond_is_lb(bond)) {
2030                 /* Must be called only after the slave has been
2031                  * detached from the list and the curr_active_slave
2032                  * has been cleared (if our_slave == old_current),
2033                  * but before a new active slave is selected.
2034                  */
2035                 write_unlock_bh(&bond->lock);
2036                 bond_alb_deinit_slave(bond, slave);
2037                 write_lock_bh(&bond->lock);
2038         }
2039
2040         if (oldcurrent == slave) {
2041                 /*
2042                  * Note that we hold RTNL over this sequence, so there
2043                  * is no concern that another slave add/remove event
2044                  * will interfere.
2045                  */
2046                 write_unlock_bh(&bond->lock);
2047                 read_lock(&bond->lock);
2048                 write_lock_bh(&bond->curr_slave_lock);
2049
2050                 bond_select_active_slave(bond);
2051
2052                 write_unlock_bh(&bond->curr_slave_lock);
2053                 read_unlock(&bond->lock);
2054                 write_lock_bh(&bond->lock);
2055         }
2056
2057         if (bond->slave_cnt == 0) {
2058                 bond_set_carrier(bond);
2059
2060                 /* if the last slave was removed, zero the mac address
2061                  * of the master so it will be set by the application
2062                  * to the mac address of the first slave
2063                  */
2064                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2065
2066                 if (bond->vlgrp) {
2067                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2068                                    bond_dev->name, bond_dev->name);
2069                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2070                                    bond_dev->name);
2071                 }
2072         }
2073
2074         write_unlock_bh(&bond->lock);
2075         unblock_netpoll_tx();
2076
2077         bond_compute_features(bond);
2078         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2079             (old_features & NETIF_F_VLAN_CHALLENGED))
2080                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2081                         bond_dev->name, slave_dev->name, bond_dev->name);
2082
2083         /* must do this from outside any spinlocks */
2084         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2085
2086         bond_del_vlans_from_slave(bond, slave_dev);
2087
2088         /* If the mode USES_PRIMARY, then we should only remove its
2089          * promisc and mc settings if it was the curr_active_slave, but that was
2090          * already taken care of above when we detached the slave
2091          */
2092         if (!USES_PRIMARY(bond->params.mode)) {
2093                 /* unset promiscuity level from slave */
2094                 if (bond_dev->flags & IFF_PROMISC)
2095                         dev_set_promiscuity(slave_dev, -1);
2096
2097                 /* unset allmulti level from slave */
2098                 if (bond_dev->flags & IFF_ALLMULTI)
2099                         dev_set_allmulti(slave_dev, -1);
2100
2101                 /* flush master's mc_list from slave */
2102                 netif_addr_lock_bh(bond_dev);
2103                 bond_mc_list_flush(bond_dev, slave_dev);
2104                 netif_addr_unlock_bh(bond_dev);
2105         }
2106
2107         netdev_set_bond_master(slave_dev, NULL);
2108
2109         slave_disable_netpoll(slave);
2110
2111         /* close slave before restoring its mac address */
2112         dev_close(slave_dev);
2113
2114         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2115                 /* restore original ("permanent") mac address */
2116                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2117                 addr.sa_family = slave_dev->type;
2118                 dev_set_mac_address(slave_dev, &addr);
2119         }
2120
2121         dev_set_mtu(slave_dev, slave->original_mtu);
2122
2123         slave_dev->priv_flags &= ~IFF_BONDING;
2124
2125         kfree(slave);
2126
2127         return 0;  /* deletion OK */
2128 }
2129
2130 /*
2131 * First release a slave and then destroy the bond if no more slaves are left.
2132 * Must be under rtnl_lock when this function is called.
2133 */
2134 static int  bond_release_and_destroy(struct net_device *bond_dev,
2135                                      struct net_device *slave_dev)
2136 {
2137         struct bonding *bond = netdev_priv(bond_dev);
2138         int ret;
2139
2140         ret = bond_release(bond_dev, slave_dev);
2141         if ((ret == 0) && (bond->slave_cnt == 0)) {
2142                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2143                 pr_info("%s: destroying bond %s.\n",
2144                         bond_dev->name, bond_dev->name);
2145                 unregister_netdevice(bond_dev);
2146         }
2147         return ret;
2148 }
2149
2150 /*
2151  * This function releases all slaves.
2152  */
2153 static int bond_release_all(struct net_device *bond_dev)
2154 {
2155         struct bonding *bond = netdev_priv(bond_dev);
2156         struct slave *slave;
2157         struct net_device *slave_dev;
2158         struct sockaddr addr;
2159
2160         write_lock_bh(&bond->lock);
2161
2162         netif_carrier_off(bond_dev);
2163
2164         if (bond->slave_cnt == 0)
2165                 goto out;
2166
2167         bond->current_arp_slave = NULL;
2168         bond->primary_slave = NULL;
2169         bond_change_active_slave(bond, NULL);
2170
2171         while ((slave = bond->first_slave) != NULL) {
2172                 /* Inform AD package of unbinding of slave
2173                  * before slave is detached from the list.
2174                  */
2175                 if (bond->params.mode == BOND_MODE_8023AD)
2176                         bond_3ad_unbind_slave(slave);
2177
2178                 slave_dev = slave->dev;
2179                 bond_detach_slave(bond, slave);
2180
2181                 /* now that the slave is detached, unlock and perform
2182                  * all the undo steps that should not be called from
2183                  * within a lock.
2184                  */
2185                 write_unlock_bh(&bond->lock);
2186
2187                 /* unregister rx_handler early so bond_handle_frame wouldn't
2188                  * be called for this slave anymore.
2189                  */
2190                 netdev_rx_handler_unregister(slave_dev);
2191                 synchronize_net();
2192
2193                 if (bond_is_lb(bond)) {
2194                         /* must be called only after the slave
2195                          * has been detached from the list
2196                          */
2197                         bond_alb_deinit_slave(bond, slave);
2198                 }
2199
2200                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2201                 bond_del_vlans_from_slave(bond, slave_dev);
2202
2203                 /* If the mode USES_PRIMARY, then we should only remove its
2204                  * promisc and mc settings if it was the curr_active_slave, but that was
2205                  * already taken care of above when we detached the slave
2206                  */
2207                 if (!USES_PRIMARY(bond->params.mode)) {
2208                         /* unset promiscuity level from slave */
2209                         if (bond_dev->flags & IFF_PROMISC)
2210                                 dev_set_promiscuity(slave_dev, -1);
2211
2212                         /* unset allmulti level from slave */
2213                         if (bond_dev->flags & IFF_ALLMULTI)
2214                                 dev_set_allmulti(slave_dev, -1);
2215
2216                         /* flush master's mc_list from slave */
2217                         netif_addr_lock_bh(bond_dev);
2218                         bond_mc_list_flush(bond_dev, slave_dev);
2219                         netif_addr_unlock_bh(bond_dev);
2220                 }
2221
2222                 netdev_set_bond_master(slave_dev, NULL);
2223
2224                 slave_disable_netpoll(slave);
2225
2226                 /* close slave before restoring its mac address */
2227                 dev_close(slave_dev);
2228
2229                 if (!bond->params.fail_over_mac) {
2230                         /* restore original ("permanent") mac address*/
2231                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2232                         addr.sa_family = slave_dev->type;
2233                         dev_set_mac_address(slave_dev, &addr);
2234                 }
2235
2236                 kfree(slave);
2237
2238                 /* re-acquire the lock before getting the next slave */
2239                 write_lock_bh(&bond->lock);
2240         }
2241
2242         /* zero the mac address of the master so it will be
2243          * set by the application to the mac address of the
2244          * first slave
2245          */
2246         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2247
2248         if (bond->vlgrp) {
2249                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2250                            bond_dev->name, bond_dev->name);
2251                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2252                            bond_dev->name);
2253         }
2254
2255         pr_info("%s: released all slaves\n", bond_dev->name);
2256
2257 out:
2258         write_unlock_bh(&bond->lock);
2259
2260         bond_compute_features(bond);
2261
2262         return 0;
2263 }
2264
2265 /*
2266  * This function changes the active slave to slave <slave_dev>.
2267  * It returns -EINVAL in the following cases.
2268  *  - <slave_dev> is not found in the list.
2269  *  - There is not active slave now.
2270  *  - <slave_dev> is already active.
2271  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2272  *  - <slave_dev> is not running.
2273  * In these cases, this function does nothing.
2274  * In the other cases, current_slave pointer is changed and 0 is returned.
2275  */
2276 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2277 {
2278         struct bonding *bond = netdev_priv(bond_dev);
2279         struct slave *old_active = NULL;
2280         struct slave *new_active = NULL;
2281         int res = 0;
2282
2283         if (!USES_PRIMARY(bond->params.mode))
2284                 return -EINVAL;
2285
2286         /* Verify that master_dev is indeed the master of slave_dev */
2287         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2288                 return -EINVAL;
2289
2290         read_lock(&bond->lock);
2291
2292         read_lock(&bond->curr_slave_lock);
2293         old_active = bond->curr_active_slave;
2294         read_unlock(&bond->curr_slave_lock);
2295
2296         new_active = bond_get_slave_by_dev(bond, slave_dev);
2297
2298         /*
2299          * Changing to the current active: do nothing; return success.
2300          */
2301         if (new_active && (new_active == old_active)) {
2302                 read_unlock(&bond->lock);
2303                 return 0;
2304         }
2305
2306         if ((new_active) &&
2307             (old_active) &&
2308             (new_active->link == BOND_LINK_UP) &&
2309             IS_UP(new_active->dev)) {
2310                 block_netpoll_tx();
2311                 write_lock_bh(&bond->curr_slave_lock);
2312                 bond_change_active_slave(bond, new_active);
2313                 write_unlock_bh(&bond->curr_slave_lock);
2314                 unblock_netpoll_tx();
2315         } else
2316                 res = -EINVAL;
2317
2318         read_unlock(&bond->lock);
2319
2320         return res;
2321 }
2322
2323 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2324 {
2325         struct bonding *bond = netdev_priv(bond_dev);
2326
2327         info->bond_mode = bond->params.mode;
2328         info->miimon = bond->params.miimon;
2329
2330         read_lock(&bond->lock);
2331         info->num_slaves = bond->slave_cnt;
2332         read_unlock(&bond->lock);
2333
2334         return 0;
2335 }
2336
2337 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2338 {
2339         struct bonding *bond = netdev_priv(bond_dev);
2340         struct slave *slave;
2341         int i, res = -ENODEV;
2342
2343         read_lock(&bond->lock);
2344
2345         bond_for_each_slave(bond, slave, i) {
2346                 if (i == (int)info->slave_id) {
2347                         res = 0;
2348                         strcpy(info->slave_name, slave->dev->name);
2349                         info->link = slave->link;
2350                         info->state = bond_slave_state(slave);
2351                         info->link_failure_count = slave->link_failure_count;
2352                         break;
2353                 }
2354         }
2355
2356         read_unlock(&bond->lock);
2357
2358         return res;
2359 }
2360
2361 /*-------------------------------- Monitoring -------------------------------*/
2362
2363
2364 static int bond_miimon_inspect(struct bonding *bond)
2365 {
2366         struct slave *slave;
2367         int i, link_state, commit = 0;
2368         bool ignore_updelay;
2369
2370         ignore_updelay = !bond->curr_active_slave ? true : false;
2371
2372         bond_for_each_slave(bond, slave, i) {
2373                 slave->new_link = BOND_LINK_NOCHANGE;
2374
2375                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2376
2377                 switch (slave->link) {
2378                 case BOND_LINK_UP:
2379                         if (link_state)
2380                                 continue;
2381
2382                         slave->link = BOND_LINK_FAIL;
2383                         slave->delay = bond->params.downdelay;
2384                         if (slave->delay) {
2385                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2386                                         bond->dev->name,
2387                                         (bond->params.mode ==
2388                                          BOND_MODE_ACTIVEBACKUP) ?
2389                                         (bond_is_active_slave(slave) ?
2390                                          "active " : "backup ") : "",
2391                                         slave->dev->name,
2392                                         bond->params.downdelay * bond->params.miimon);
2393                         }
2394                         /*FALLTHRU*/
2395                 case BOND_LINK_FAIL:
2396                         if (link_state) {
2397                                 /*
2398                                  * recovered before downdelay expired
2399                                  */
2400                                 slave->link = BOND_LINK_UP;
2401                                 slave->jiffies = jiffies;
2402                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2403                                         bond->dev->name,
2404                                         (bond->params.downdelay - slave->delay) *
2405                                         bond->params.miimon,
2406                                         slave->dev->name);
2407                                 continue;
2408                         }
2409
2410                         if (slave->delay <= 0) {
2411                                 slave->new_link = BOND_LINK_DOWN;
2412                                 commit++;
2413                                 continue;
2414                         }
2415
2416                         slave->delay--;
2417                         break;
2418
2419                 case BOND_LINK_DOWN:
2420                         if (!link_state)
2421                                 continue;
2422
2423                         slave->link = BOND_LINK_BACK;
2424                         slave->delay = bond->params.updelay;
2425
2426                         if (slave->delay) {
2427                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2428                                         bond->dev->name, slave->dev->name,
2429                                         ignore_updelay ? 0 :
2430                                         bond->params.updelay *
2431                                         bond->params.miimon);
2432                         }
2433                         /*FALLTHRU*/
2434                 case BOND_LINK_BACK:
2435                         if (!link_state) {
2436                                 slave->link = BOND_LINK_DOWN;
2437                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2438                                         bond->dev->name,
2439                                         (bond->params.updelay - slave->delay) *
2440                                         bond->params.miimon,
2441                                         slave->dev->name);
2442
2443                                 continue;
2444                         }
2445
2446                         if (ignore_updelay)
2447                                 slave->delay = 0;
2448
2449                         if (slave->delay <= 0) {
2450                                 slave->new_link = BOND_LINK_UP;
2451                                 commit++;
2452                                 ignore_updelay = false;
2453                                 continue;
2454                         }
2455
2456                         slave->delay--;
2457                         break;
2458                 }
2459         }
2460
2461         return commit;
2462 }
2463
2464 static void bond_miimon_commit(struct bonding *bond)
2465 {
2466         struct slave *slave;
2467         int i;
2468
2469         bond_for_each_slave(bond, slave, i) {
2470                 switch (slave->new_link) {
2471                 case BOND_LINK_NOCHANGE:
2472                         continue;
2473
2474                 case BOND_LINK_UP:
2475                         slave->link = BOND_LINK_UP;
2476                         slave->jiffies = jiffies;
2477
2478                         if (bond->params.mode == BOND_MODE_8023AD) {
2479                                 /* prevent it from being the active one */
2480                                 bond_set_backup_slave(slave);
2481                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2482                                 /* make it immediately active */
2483                                 bond_set_active_slave(slave);
2484                         } else if (slave != bond->primary_slave) {
2485                                 /* prevent it from being the active one */
2486                                 bond_set_backup_slave(slave);
2487                         }
2488
2489                         bond_update_speed_duplex(slave);
2490
2491                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2492                                 bond->dev->name, slave->dev->name,
2493                                 slave->speed, slave->duplex ? "full" : "half");
2494
2495                         /* notify ad that the link status has changed */
2496                         if (bond->params.mode == BOND_MODE_8023AD)
2497                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2498
2499                         if (bond_is_lb(bond))
2500                                 bond_alb_handle_link_change(bond, slave,
2501                                                             BOND_LINK_UP);
2502
2503                         if (!bond->curr_active_slave ||
2504                             (slave == bond->primary_slave))
2505                                 goto do_failover;
2506
2507                         continue;
2508
2509                 case BOND_LINK_DOWN:
2510                         if (slave->link_failure_count < UINT_MAX)
2511                                 slave->link_failure_count++;
2512
2513                         slave->link = BOND_LINK_DOWN;
2514
2515                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2516                             bond->params.mode == BOND_MODE_8023AD)
2517                                 bond_set_slave_inactive_flags(slave);
2518
2519                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2520                                 bond->dev->name, slave->dev->name);
2521
2522                         if (bond->params.mode == BOND_MODE_8023AD)
2523                                 bond_3ad_handle_link_change(slave,
2524                                                             BOND_LINK_DOWN);
2525
2526                         if (bond_is_lb(bond))
2527                                 bond_alb_handle_link_change(bond, slave,
2528                                                             BOND_LINK_DOWN);
2529
2530                         if (slave == bond->curr_active_slave)
2531                                 goto do_failover;
2532
2533                         continue;
2534
2535                 default:
2536                         pr_err("%s: invalid new link %d on slave %s\n",
2537                                bond->dev->name, slave->new_link,
2538                                slave->dev->name);
2539                         slave->new_link = BOND_LINK_NOCHANGE;
2540
2541                         continue;
2542                 }
2543
2544 do_failover:
2545                 ASSERT_RTNL();
2546                 block_netpoll_tx();
2547                 write_lock_bh(&bond->curr_slave_lock);
2548                 bond_select_active_slave(bond);
2549                 write_unlock_bh(&bond->curr_slave_lock);
2550                 unblock_netpoll_tx();
2551         }
2552
2553         bond_set_carrier(bond);
2554 }
2555
2556 /*
2557  * bond_mii_monitor
2558  *
2559  * Really a wrapper that splits the mii monitor into two phases: an
2560  * inspection, then (if inspection indicates something needs to be done)
2561  * an acquisition of appropriate locks followed by a commit phase to
2562  * implement whatever link state changes are indicated.
2563  */
2564 void bond_mii_monitor(struct work_struct *work)
2565 {
2566         struct bonding *bond = container_of(work, struct bonding,
2567                                             mii_work.work);
2568         bool should_notify_peers = false;
2569
2570         read_lock(&bond->lock);
2571         if (bond->kill_timers)
2572                 goto out;
2573
2574         if (bond->slave_cnt == 0)
2575                 goto re_arm;
2576
2577         should_notify_peers = bond_should_notify_peers(bond);
2578
2579         if (bond_miimon_inspect(bond)) {
2580                 read_unlock(&bond->lock);
2581                 rtnl_lock();
2582                 read_lock(&bond->lock);
2583
2584                 bond_miimon_commit(bond);
2585
2586                 read_unlock(&bond->lock);
2587                 rtnl_unlock();  /* might sleep, hold no other locks */
2588                 read_lock(&bond->lock);
2589         }
2590
2591 re_arm:
2592         if (bond->params.miimon)
2593                 queue_delayed_work(bond->wq, &bond->mii_work,
2594                                    msecs_to_jiffies(bond->params.miimon));
2595 out:
2596         read_unlock(&bond->lock);
2597
2598         if (should_notify_peers) {
2599                 rtnl_lock();
2600                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2601                 rtnl_unlock();
2602         }
2603 }
2604
2605 static __be32 bond_glean_dev_ip(struct net_device *dev)
2606 {
2607         struct in_device *idev;
2608         struct in_ifaddr *ifa;
2609         __be32 addr = 0;
2610
2611         if (!dev)
2612                 return 0;
2613
2614         rcu_read_lock();
2615         idev = __in_dev_get_rcu(dev);
2616         if (!idev)
2617                 goto out;
2618
2619         ifa = idev->ifa_list;
2620         if (!ifa)
2621                 goto out;
2622
2623         addr = ifa->ifa_local;
2624 out:
2625         rcu_read_unlock();
2626         return addr;
2627 }
2628
2629 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2630 {
2631         struct vlan_entry *vlan;
2632
2633         if (ip == bond->master_ip)
2634                 return 1;
2635
2636         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2637                 if (ip == vlan->vlan_ip)
2638                         return 1;
2639         }
2640
2641         return 0;
2642 }
2643
2644 /*
2645  * We go to the (large) trouble of VLAN tagging ARP frames because
2646  * switches in VLAN mode (especially if ports are configured as
2647  * "native" to a VLAN) might not pass non-tagged frames.
2648  */
2649 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2650 {
2651         struct sk_buff *skb;
2652
2653         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2654                  slave_dev->name, dest_ip, src_ip, vlan_id);
2655
2656         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2657                          NULL, slave_dev->dev_addr, NULL);
2658
2659         if (!skb) {
2660                 pr_err("ARP packet allocation failed\n");
2661                 return;
2662         }
2663         if (vlan_id) {
2664                 skb = vlan_put_tag(skb, vlan_id);
2665                 if (!skb) {
2666                         pr_err("failed to insert VLAN tag\n");
2667                         return;
2668                 }
2669         }
2670         arp_xmit(skb);
2671 }
2672
2673
2674 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2675 {
2676         int i, vlan_id;
2677         __be32 *targets = bond->params.arp_targets;
2678         struct vlan_entry *vlan;
2679         struct net_device *vlan_dev;
2680         struct rtable *rt;
2681
2682         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2683                 if (!targets[i])
2684                         break;
2685                 pr_debug("basa: target %x\n", targets[i]);
2686                 if (!bond->vlgrp) {
2687                         pr_debug("basa: empty vlan: arp_send\n");
2688                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2689                                       bond->master_ip, 0);
2690                         continue;
2691                 }
2692
2693                 /*
2694                  * If VLANs are configured, we do a route lookup to
2695                  * determine which VLAN interface would be used, so we
2696                  * can tag the ARP with the proper VLAN tag.
2697                  */
2698                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2699                                      RTO_ONLINK, 0);
2700                 if (IS_ERR(rt)) {
2701                         if (net_ratelimit()) {
2702                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2703                                            bond->dev->name, &targets[i]);
2704                         }
2705                         continue;
2706                 }
2707
2708                 /*
2709                  * This target is not on a VLAN
2710                  */
2711                 if (rt->dst.dev == bond->dev) {
2712                         ip_rt_put(rt);
2713                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2714                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2715                                       bond->master_ip, 0);
2716                         continue;
2717                 }
2718
2719                 vlan_id = 0;
2720                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2721                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2722                         if (vlan_dev == rt->dst.dev) {
2723                                 vlan_id = vlan->vlan_id;
2724                                 pr_debug("basa: vlan match on %s %d\n",
2725                                        vlan_dev->name, vlan_id);
2726                                 break;
2727                         }
2728                 }
2729
2730                 if (vlan_id) {
2731                         ip_rt_put(rt);
2732                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2733                                       vlan->vlan_ip, vlan_id);
2734                         continue;
2735                 }
2736
2737                 if (net_ratelimit()) {
2738                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2739                                    bond->dev->name, &targets[i],
2740                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2741                 }
2742                 ip_rt_put(rt);
2743         }
2744 }
2745
2746 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2747 {
2748         int i;
2749         __be32 *targets = bond->params.arp_targets;
2750
2751         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2752                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2753                          &sip, &tip, i, &targets[i],
2754                          bond_has_this_ip(bond, tip));
2755                 if (sip == targets[i]) {
2756                         if (bond_has_this_ip(bond, tip))
2757                                 slave->last_arp_rx = jiffies;
2758                         return;
2759                 }
2760         }
2761 }
2762
2763 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2764                          struct slave *slave)
2765 {
2766         struct arphdr *arp;
2767         unsigned char *arp_ptr;
2768         __be32 sip, tip;
2769
2770         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2771                 return;
2772
2773         read_lock(&bond->lock);
2774
2775         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2776                  bond->dev->name, skb->dev->name);
2777
2778         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2779                 goto out_unlock;
2780
2781         arp = arp_hdr(skb);
2782         if (arp->ar_hln != bond->dev->addr_len ||
2783             skb->pkt_type == PACKET_OTHERHOST ||
2784             skb->pkt_type == PACKET_LOOPBACK ||
2785             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2786             arp->ar_pro != htons(ETH_P_IP) ||
2787             arp->ar_pln != 4)
2788                 goto out_unlock;
2789
2790         arp_ptr = (unsigned char *)(arp + 1);
2791         arp_ptr += bond->dev->addr_len;
2792         memcpy(&sip, arp_ptr, 4);
2793         arp_ptr += 4 + bond->dev->addr_len;
2794         memcpy(&tip, arp_ptr, 4);
2795
2796         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2797                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2798                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2799                  &sip, &tip);
2800
2801         /*
2802          * Backup slaves won't see the ARP reply, but do come through
2803          * here for each ARP probe (so we swap the sip/tip to validate
2804          * the probe).  In a "redundant switch, common router" type of
2805          * configuration, the ARP probe will (hopefully) travel from
2806          * the active, through one switch, the router, then the other
2807          * switch before reaching the backup.
2808          */
2809         if (bond_is_active_slave(slave))
2810                 bond_validate_arp(bond, slave, sip, tip);
2811         else
2812                 bond_validate_arp(bond, slave, tip, sip);
2813
2814 out_unlock:
2815         read_unlock(&bond->lock);
2816 }
2817
2818 /*
2819  * this function is called regularly to monitor each slave's link
2820  * ensuring that traffic is being sent and received when arp monitoring
2821  * is used in load-balancing mode. if the adapter has been dormant, then an
2822  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2823  * arp monitoring in active backup mode.
2824  */
2825 void bond_loadbalance_arp_mon(struct work_struct *work)
2826 {
2827         struct bonding *bond = container_of(work, struct bonding,
2828                                             arp_work.work);
2829         struct slave *slave, *oldcurrent;
2830         int do_failover = 0;
2831         int delta_in_ticks;
2832         int i;
2833
2834         read_lock(&bond->lock);
2835
2836         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2837
2838         if (bond->kill_timers)
2839                 goto out;
2840
2841         if (bond->slave_cnt == 0)
2842                 goto re_arm;
2843
2844         read_lock(&bond->curr_slave_lock);
2845         oldcurrent = bond->curr_active_slave;
2846         read_unlock(&bond->curr_slave_lock);
2847
2848         /* see if any of the previous devices are up now (i.e. they have
2849          * xmt and rcv traffic). the curr_active_slave does not come into
2850          * the picture unless it is null. also, slave->jiffies is not needed
2851          * here because we send an arp on each slave and give a slave as
2852          * long as it needs to get the tx/rx within the delta.
2853          * TODO: what about up/down delay in arp mode? it wasn't here before
2854          *       so it can wait
2855          */
2856         bond_for_each_slave(bond, slave, i) {
2857                 unsigned long trans_start = dev_trans_start(slave->dev);
2858
2859                 if (slave->link != BOND_LINK_UP) {
2860                         if (time_in_range(jiffies,
2861                                 trans_start - delta_in_ticks,
2862                                 trans_start + delta_in_ticks) &&
2863                             time_in_range(jiffies,
2864                                 slave->dev->last_rx - delta_in_ticks,
2865                                 slave->dev->last_rx + delta_in_ticks)) {
2866
2867                                 slave->link  = BOND_LINK_UP;
2868                                 bond_set_active_slave(slave);
2869
2870                                 /* primary_slave has no meaning in round-robin
2871                                  * mode. the window of a slave being up and
2872                                  * curr_active_slave being null after enslaving
2873                                  * is closed.
2874                                  */
2875                                 if (!oldcurrent) {
2876                                         pr_info("%s: link status definitely up for interface %s, ",
2877                                                 bond->dev->name,
2878                                                 slave->dev->name);
2879                                         do_failover = 1;
2880                                 } else {
2881                                         pr_info("%s: interface %s is now up\n",
2882                                                 bond->dev->name,
2883                                                 slave->dev->name);
2884                                 }
2885                         }
2886                 } else {
2887                         /* slave->link == BOND_LINK_UP */
2888
2889                         /* not all switches will respond to an arp request
2890                          * when the source ip is 0, so don't take the link down
2891                          * if we don't know our ip yet
2892                          */
2893                         if (!time_in_range(jiffies,
2894                                 trans_start - delta_in_ticks,
2895                                 trans_start + 2 * delta_in_ticks) ||
2896                             !time_in_range(jiffies,
2897                                 slave->dev->last_rx - delta_in_ticks,
2898                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2899
2900                                 slave->link  = BOND_LINK_DOWN;
2901                                 bond_set_backup_slave(slave);
2902
2903                                 if (slave->link_failure_count < UINT_MAX)
2904                                         slave->link_failure_count++;
2905
2906                                 pr_info("%s: interface %s is now down.\n",
2907                                         bond->dev->name,
2908                                         slave->dev->name);
2909
2910                                 if (slave == oldcurrent)
2911                                         do_failover = 1;
2912                         }
2913                 }
2914
2915                 /* note: if switch is in round-robin mode, all links
2916                  * must tx arp to ensure all links rx an arp - otherwise
2917                  * links may oscillate or not come up at all; if switch is
2918                  * in something like xor mode, there is nothing we can
2919                  * do - all replies will be rx'ed on same link causing slaves
2920                  * to be unstable during low/no traffic periods
2921                  */
2922                 if (IS_UP(slave->dev))
2923                         bond_arp_send_all(bond, slave);
2924         }
2925
2926         if (do_failover) {
2927                 block_netpoll_tx();
2928                 write_lock_bh(&bond->curr_slave_lock);
2929
2930                 bond_select_active_slave(bond);
2931
2932                 write_unlock_bh(&bond->curr_slave_lock);
2933                 unblock_netpoll_tx();
2934         }
2935
2936 re_arm:
2937         if (bond->params.arp_interval)
2938                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2939 out:
2940         read_unlock(&bond->lock);
2941 }
2942
2943 /*
2944  * Called to inspect slaves for active-backup mode ARP monitor link state
2945  * changes.  Sets new_link in slaves to specify what action should take
2946  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2947  * to link states must be committed.
2948  *
2949  * Called with bond->lock held for read.
2950  */
2951 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2952 {
2953         struct slave *slave;
2954         int i, commit = 0;
2955         unsigned long trans_start;
2956
2957         bond_for_each_slave(bond, slave, i) {
2958                 slave->new_link = BOND_LINK_NOCHANGE;
2959
2960                 if (slave->link != BOND_LINK_UP) {
2961                         if (time_in_range(jiffies,
2962                                 slave_last_rx(bond, slave) - delta_in_ticks,
2963                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2964
2965                                 slave->new_link = BOND_LINK_UP;
2966                                 commit++;
2967                         }
2968
2969                         continue;
2970                 }
2971
2972                 /*
2973                  * Give slaves 2*delta after being enslaved or made
2974                  * active.  This avoids bouncing, as the last receive
2975                  * times need a full ARP monitor cycle to be updated.
2976                  */
2977                 if (time_in_range(jiffies,
2978                                   slave->jiffies - delta_in_ticks,
2979                                   slave->jiffies + 2 * delta_in_ticks))
2980                         continue;
2981
2982                 /*
2983                  * Backup slave is down if:
2984                  * - No current_arp_slave AND
2985                  * - more than 3*delta since last receive AND
2986                  * - the bond has an IP address
2987                  *
2988                  * Note: a non-null current_arp_slave indicates
2989                  * the curr_active_slave went down and we are
2990                  * searching for a new one; under this condition
2991                  * we only take the curr_active_slave down - this
2992                  * gives each slave a chance to tx/rx traffic
2993                  * before being taken out
2994                  */
2995                 if (!bond_is_active_slave(slave) &&
2996                     !bond->current_arp_slave &&
2997                     !time_in_range(jiffies,
2998                         slave_last_rx(bond, slave) - delta_in_ticks,
2999                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3000
3001                         slave->new_link = BOND_LINK_DOWN;
3002                         commit++;
3003                 }
3004
3005                 /*
3006                  * Active slave is down if:
3007                  * - more than 2*delta since transmitting OR
3008                  * - (more than 2*delta since receive AND
3009                  *    the bond has an IP address)
3010                  */
3011                 trans_start = dev_trans_start(slave->dev);
3012                 if (bond_is_active_slave(slave) &&
3013                     (!time_in_range(jiffies,
3014                         trans_start - delta_in_ticks,
3015                         trans_start + 2 * delta_in_ticks) ||
3016                      !time_in_range(jiffies,
3017                         slave_last_rx(bond, slave) - delta_in_ticks,
3018                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3019
3020                         slave->new_link = BOND_LINK_DOWN;
3021                         commit++;
3022                 }
3023         }
3024
3025         return commit;
3026 }
3027
3028 /*
3029  * Called to commit link state changes noted by inspection step of
3030  * active-backup mode ARP monitor.
3031  *
3032  * Called with RTNL and bond->lock for read.
3033  */
3034 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3035 {
3036         struct slave *slave;
3037         int i;
3038         unsigned long trans_start;
3039
3040         bond_for_each_slave(bond, slave, i) {
3041                 switch (slave->new_link) {
3042                 case BOND_LINK_NOCHANGE:
3043                         continue;
3044
3045                 case BOND_LINK_UP:
3046                         trans_start = dev_trans_start(slave->dev);
3047                         if ((!bond->curr_active_slave &&
3048                              time_in_range(jiffies,
3049                                            trans_start - delta_in_ticks,
3050                                            trans_start + delta_in_ticks)) ||
3051                             bond->curr_active_slave != slave) {
3052                                 slave->link = BOND_LINK_UP;
3053                                 bond->current_arp_slave = NULL;
3054
3055                                 pr_info("%s: link status definitely up for interface %s.\n",
3056                                         bond->dev->name, slave->dev->name);
3057
3058                                 if (!bond->curr_active_slave ||
3059                                     (slave == bond->primary_slave))
3060                                         goto do_failover;
3061
3062                         }
3063
3064                         continue;
3065
3066                 case BOND_LINK_DOWN:
3067                         if (slave->link_failure_count < UINT_MAX)
3068                                 slave->link_failure_count++;
3069
3070                         slave->link = BOND_LINK_DOWN;
3071                         bond_set_slave_inactive_flags(slave);
3072
3073                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3074                                 bond->dev->name, slave->dev->name);
3075
3076                         if (slave == bond->curr_active_slave) {
3077                                 bond->current_arp_slave = NULL;
3078                                 goto do_failover;
3079                         }
3080
3081                         continue;
3082
3083                 default:
3084                         pr_err("%s: impossible: new_link %d on slave %s\n",
3085                                bond->dev->name, slave->new_link,
3086                                slave->dev->name);
3087                         continue;
3088                 }
3089
3090 do_failover:
3091                 ASSERT_RTNL();
3092                 block_netpoll_tx();
3093                 write_lock_bh(&bond->curr_slave_lock);
3094                 bond_select_active_slave(bond);
3095                 write_unlock_bh(&bond->curr_slave_lock);
3096                 unblock_netpoll_tx();
3097         }
3098
3099         bond_set_carrier(bond);
3100 }
3101
3102 /*
3103  * Send ARP probes for active-backup mode ARP monitor.
3104  *
3105  * Called with bond->lock held for read.
3106  */
3107 static void bond_ab_arp_probe(struct bonding *bond)
3108 {
3109         struct slave *slave;
3110         int i;
3111
3112         read_lock(&bond->curr_slave_lock);
3113
3114         if (bond->current_arp_slave && bond->curr_active_slave)
3115                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3116                         bond->current_arp_slave->dev->name,
3117                         bond->curr_active_slave->dev->name);
3118
3119         if (bond->curr_active_slave) {
3120                 bond_arp_send_all(bond, bond->curr_active_slave);
3121                 read_unlock(&bond->curr_slave_lock);
3122                 return;
3123         }
3124
3125         read_unlock(&bond->curr_slave_lock);
3126
3127         /* if we don't have a curr_active_slave, search for the next available
3128          * backup slave from the current_arp_slave and make it the candidate
3129          * for becoming the curr_active_slave
3130          */
3131
3132         if (!bond->current_arp_slave) {
3133                 bond->current_arp_slave = bond->first_slave;
3134                 if (!bond->current_arp_slave)
3135                         return;
3136         }
3137
3138         bond_set_slave_inactive_flags(bond->current_arp_slave);
3139
3140         /* search for next candidate */
3141         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3142                 if (IS_UP(slave->dev)) {
3143                         slave->link = BOND_LINK_BACK;
3144                         bond_set_slave_active_flags(slave);
3145                         bond_arp_send_all(bond, slave);
3146                         slave->jiffies = jiffies;
3147                         bond->current_arp_slave = slave;
3148                         break;
3149                 }
3150
3151                 /* if the link state is up at this point, we
3152                  * mark it down - this can happen if we have
3153                  * simultaneous link failures and
3154                  * reselect_active_interface doesn't make this
3155                  * one the current slave so it is still marked
3156                  * up when it is actually down
3157                  */
3158                 if (slave->link == BOND_LINK_UP) {
3159                         slave->link = BOND_LINK_DOWN;
3160                         if (slave->link_failure_count < UINT_MAX)
3161                                 slave->link_failure_count++;
3162
3163                         bond_set_slave_inactive_flags(slave);
3164
3165                         pr_info("%s: backup interface %s is now down.\n",
3166                                 bond->dev->name, slave->dev->name);
3167                 }
3168         }
3169 }
3170
3171 void bond_activebackup_arp_mon(struct work_struct *work)
3172 {
3173         struct bonding *bond = container_of(work, struct bonding,
3174                                             arp_work.work);
3175         bool should_notify_peers = false;
3176         int delta_in_ticks;
3177
3178         read_lock(&bond->lock);
3179
3180         if (bond->kill_timers)
3181                 goto out;
3182
3183         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3184
3185         if (bond->slave_cnt == 0)
3186                 goto re_arm;
3187
3188         should_notify_peers = bond_should_notify_peers(bond);
3189
3190         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3191                 read_unlock(&bond->lock);
3192                 rtnl_lock();
3193                 read_lock(&bond->lock);
3194
3195                 bond_ab_arp_commit(bond, delta_in_ticks);
3196
3197                 read_unlock(&bond->lock);
3198                 rtnl_unlock();
3199                 read_lock(&bond->lock);
3200         }
3201
3202         bond_ab_arp_probe(bond);
3203
3204 re_arm:
3205         if (bond->params.arp_interval)
3206                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3207 out:
3208         read_unlock(&bond->lock);
3209
3210         if (should_notify_peers) {
3211                 rtnl_lock();
3212                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3213                 rtnl_unlock();
3214         }
3215 }
3216
3217 /*-------------------------- netdev event handling --------------------------*/
3218
3219 /*
3220  * Change device name
3221  */
3222 static int bond_event_changename(struct bonding *bond)
3223 {
3224         bond_remove_proc_entry(bond);
3225         bond_create_proc_entry(bond);
3226
3227         bond_debug_reregister(bond);
3228
3229         return NOTIFY_DONE;
3230 }
3231
3232 static int bond_master_netdev_event(unsigned long event,
3233                                     struct net_device *bond_dev)
3234 {
3235         struct bonding *event_bond = netdev_priv(bond_dev);
3236
3237         switch (event) {
3238         case NETDEV_CHANGENAME:
3239                 return bond_event_changename(event_bond);
3240         default:
3241                 break;
3242         }
3243
3244         return NOTIFY_DONE;
3245 }
3246
3247 static int bond_slave_netdev_event(unsigned long event,
3248                                    struct net_device *slave_dev)
3249 {
3250         struct net_device *bond_dev = slave_dev->master;
3251         struct bonding *bond = netdev_priv(bond_dev);
3252
3253         switch (event) {
3254         case NETDEV_UNREGISTER:
3255                 if (bond_dev) {
3256                         if (bond->setup_by_slave)
3257                                 bond_release_and_destroy(bond_dev, slave_dev);
3258                         else
3259                                 bond_release(bond_dev, slave_dev);
3260                 }
3261                 break;
3262         case NETDEV_CHANGE:
3263                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3264                         struct slave *slave;
3265
3266                         slave = bond_get_slave_by_dev(bond, slave_dev);
3267                         if (slave) {
3268                                 u32 old_speed = slave->speed;
3269                                 u8  old_duplex = slave->duplex;
3270
3271                                 bond_update_speed_duplex(slave);
3272
3273                                 if (bond_is_lb(bond))
3274                                         break;
3275
3276                                 if (old_speed != slave->speed)
3277                                         bond_3ad_adapter_speed_changed(slave);
3278                                 if (old_duplex != slave->duplex)
3279                                         bond_3ad_adapter_duplex_changed(slave);
3280                         }
3281                 }
3282
3283                 break;
3284         case NETDEV_DOWN:
3285                 /*
3286                  * ... Or is it this?
3287                  */
3288                 break;
3289         case NETDEV_CHANGEMTU:
3290                 /*
3291                  * TODO: Should slaves be allowed to
3292                  * independently alter their MTU?  For
3293                  * an active-backup bond, slaves need
3294                  * not be the same type of device, so
3295                  * MTUs may vary.  For other modes,
3296                  * slaves arguably should have the
3297                  * same MTUs. To do this, we'd need to
3298                  * take over the slave's change_mtu
3299                  * function for the duration of their
3300                  * servitude.
3301                  */
3302                 break;
3303         case NETDEV_CHANGENAME:
3304                 /*
3305                  * TODO: handle changing the primary's name
3306                  */
3307                 break;
3308         case NETDEV_FEAT_CHANGE:
3309                 bond_compute_features(bond);
3310                 break;
3311         default:
3312                 break;
3313         }
3314
3315         return NOTIFY_DONE;
3316 }
3317
3318 /*
3319  * bond_netdev_event: handle netdev notifier chain events.
3320  *
3321  * This function receives events for the netdev chain.  The caller (an
3322  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3323  * locks for us to safely manipulate the slave devices (RTNL lock,
3324  * dev_probe_lock).
3325  */
3326 static int bond_netdev_event(struct notifier_block *this,
3327                              unsigned long event, void *ptr)
3328 {
3329         struct net_device *event_dev = (struct net_device *)ptr;
3330
3331         pr_debug("event_dev: %s, event: %lx\n",
3332                  event_dev ? event_dev->name : "None",
3333                  event);
3334
3335         if (!(event_dev->priv_flags & IFF_BONDING))
3336                 return NOTIFY_DONE;
3337
3338         if (event_dev->flags & IFF_MASTER) {
3339                 pr_debug("IFF_MASTER\n");
3340                 return bond_master_netdev_event(event, event_dev);
3341         }
3342
3343         if (event_dev->flags & IFF_SLAVE) {
3344                 pr_debug("IFF_SLAVE\n");
3345                 return bond_slave_netdev_event(event, event_dev);
3346         }
3347
3348         return NOTIFY_DONE;
3349 }
3350
3351 /*
3352  * bond_inetaddr_event: handle inetaddr notifier chain events.
3353  *
3354  * We keep track of device IPs primarily to use as source addresses in
3355  * ARP monitor probes (rather than spewing out broadcasts all the time).
3356  *
3357  * We track one IP for the main device (if it has one), plus one per VLAN.
3358  */
3359 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3360 {
3361         struct in_ifaddr *ifa = ptr;
3362         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3363         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3364         struct bonding *bond;
3365         struct vlan_entry *vlan;
3366
3367         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3368                 if (bond->dev == event_dev) {
3369                         switch (event) {
3370                         case NETDEV_UP:
3371                                 bond->master_ip = ifa->ifa_local;
3372                                 return NOTIFY_OK;
3373                         case NETDEV_DOWN:
3374                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3375                                 return NOTIFY_OK;
3376                         default:
3377                                 return NOTIFY_DONE;
3378                         }
3379                 }
3380
3381                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3382                         if (!bond->vlgrp)
3383                                 continue;
3384                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3385                         if (vlan_dev == event_dev) {
3386                                 switch (event) {
3387                                 case NETDEV_UP:
3388                                         vlan->vlan_ip = ifa->ifa_local;
3389                                         return NOTIFY_OK;
3390                                 case NETDEV_DOWN:
3391                                         vlan->vlan_ip =
3392                                                 bond_glean_dev_ip(vlan_dev);
3393                                         return NOTIFY_OK;
3394                                 default:
3395                                         return NOTIFY_DONE;
3396                                 }
3397                         }
3398                 }
3399         }
3400         return NOTIFY_DONE;
3401 }
3402
3403 static struct notifier_block bond_netdev_notifier = {
3404         .notifier_call = bond_netdev_event,
3405 };
3406
3407 static struct notifier_block bond_inetaddr_notifier = {
3408         .notifier_call = bond_inetaddr_event,
3409 };
3410
3411 /*---------------------------- Hashing Policies -----------------------------*/
3412
3413 /*
3414  * Hash for the output device based upon layer 2 and layer 3 data. If
3415  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3416  */
3417 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3418 {
3419         struct ethhdr *data = (struct ethhdr *)skb->data;
3420         struct iphdr *iph = ip_hdr(skb);
3421
3422         if (skb->protocol == htons(ETH_P_IP)) {
3423                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3424                         (data->h_dest[5] ^ data->h_source[5])) % count;
3425         }
3426
3427         return (data->h_dest[5] ^ data->h_source[5]) % count;
3428 }
3429
3430 /*
3431  * Hash for the output device based upon layer 3 and layer 4 data. If
3432  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3433  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3434  */
3435 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3436 {
3437         struct ethhdr *data = (struct ethhdr *)skb->data;
3438         struct iphdr *iph = ip_hdr(skb);
3439         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3440         int layer4_xor = 0;
3441
3442         if (skb->protocol == htons(ETH_P_IP)) {
3443                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3444                     (iph->protocol == IPPROTO_TCP ||
3445                      iph->protocol == IPPROTO_UDP)) {
3446                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3447                 }
3448                 return (layer4_xor ^
3449                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3450
3451         }
3452
3453         return (data->h_dest[5] ^ data->h_source[5]) % count;
3454 }
3455
3456 /*
3457  * Hash for the output device based upon layer 2 data
3458  */
3459 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3460 {
3461         struct ethhdr *data = (struct ethhdr *)skb->data;
3462
3463         return (data->h_dest[5] ^ data->h_source[5]) % count;
3464 }
3465
3466 /*-------------------------- Device entry points ----------------------------*/
3467
3468 static int bond_open(struct net_device *bond_dev)
3469 {
3470         struct bonding *bond = netdev_priv(bond_dev);
3471
3472         bond->kill_timers = 0;
3473
3474         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3475
3476         if (bond_is_lb(bond)) {
3477                 /* bond_alb_initialize must be called before the timer
3478                  * is started.
3479                  */
3480                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3481                         /* something went wrong - fail the open operation */
3482                         return -ENOMEM;
3483                 }
3484
3485                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3486                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3487         }
3488
3489         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3490                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3491                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3492         }
3493
3494         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3495                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3496                         INIT_DELAYED_WORK(&bond->arp_work,
3497                                           bond_activebackup_arp_mon);
3498                 else
3499                         INIT_DELAYED_WORK(&bond->arp_work,
3500                                           bond_loadbalance_arp_mon);
3501
3502                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3503                 if (bond->params.arp_validate)
3504                         bond->recv_probe = bond_arp_rcv;
3505         }
3506
3507         if (bond->params.mode == BOND_MODE_8023AD) {
3508                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3509                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3510                 /* register to receive LACPDUs */
3511                 bond->recv_probe = bond_3ad_lacpdu_recv;
3512                 bond_3ad_initiate_agg_selection(bond, 1);
3513         }
3514
3515         return 0;
3516 }
3517
3518 static int bond_close(struct net_device *bond_dev)
3519 {
3520         struct bonding *bond = netdev_priv(bond_dev);
3521
3522         write_lock_bh(&bond->lock);
3523
3524         bond->send_peer_notif = 0;
3525
3526         /* signal timers not to re-arm */
3527         bond->kill_timers = 1;
3528
3529         write_unlock_bh(&bond->lock);
3530
3531         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3532                 cancel_delayed_work(&bond->mii_work);
3533         }
3534
3535         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3536                 cancel_delayed_work(&bond->arp_work);
3537         }
3538
3539         switch (bond->params.mode) {
3540         case BOND_MODE_8023AD:
3541                 cancel_delayed_work(&bond->ad_work);
3542                 break;
3543         case BOND_MODE_TLB:
3544         case BOND_MODE_ALB:
3545                 cancel_delayed_work(&bond->alb_work);
3546                 break;
3547         default:
3548                 break;
3549         }
3550
3551         if (delayed_work_pending(&bond->mcast_work))
3552                 cancel_delayed_work(&bond->mcast_work);
3553
3554         if (bond_is_lb(bond)) {
3555                 /* Must be called only after all
3556                  * slaves have been released
3557                  */
3558                 bond_alb_deinitialize(bond);
3559         }
3560         bond->recv_probe = NULL;
3561
3562         return 0;
3563 }
3564
3565 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3566                                                 struct rtnl_link_stats64 *stats)
3567 {
3568         struct bonding *bond = netdev_priv(bond_dev);
3569         struct rtnl_link_stats64 temp;
3570         struct slave *slave;
3571         int i;
3572
3573         memset(stats, 0, sizeof(*stats));
3574
3575         read_lock_bh(&bond->lock);
3576
3577         bond_for_each_slave(bond, slave, i) {
3578                 const struct rtnl_link_stats64 *sstats =
3579                         dev_get_stats(slave->dev, &temp);
3580
3581                 stats->rx_packets += sstats->rx_packets;
3582                 stats->rx_bytes += sstats->rx_bytes;
3583                 stats->rx_errors += sstats->rx_errors;
3584                 stats->rx_dropped += sstats->rx_dropped;
3585
3586                 stats->tx_packets += sstats->tx_packets;
3587                 stats->tx_bytes += sstats->tx_bytes;
3588                 stats->tx_errors += sstats->tx_errors;
3589                 stats->tx_dropped += sstats->tx_dropped;
3590
3591                 stats->multicast += sstats->multicast;
3592                 stats->collisions += sstats->collisions;
3593
3594                 stats->rx_length_errors += sstats->rx_length_errors;
3595                 stats->rx_over_errors += sstats->rx_over_errors;
3596                 stats->rx_crc_errors += sstats->rx_crc_errors;
3597                 stats->rx_frame_errors += sstats->rx_frame_errors;
3598                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3599                 stats->rx_missed_errors += sstats->rx_missed_errors;
3600
3601                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3602                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3603                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3604                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3605                 stats->tx_window_errors += sstats->tx_window_errors;
3606         }
3607
3608         read_unlock_bh(&bond->lock);
3609
3610         return stats;
3611 }
3612
3613 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3614 {
3615         struct net_device *slave_dev = NULL;
3616         struct ifbond k_binfo;
3617         struct ifbond __user *u_binfo = NULL;
3618         struct ifslave k_sinfo;
3619         struct ifslave __user *u_sinfo = NULL;
3620         struct mii_ioctl_data *mii = NULL;
3621         int res = 0;
3622
3623         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3624
3625         switch (cmd) {
3626         case SIOCGMIIPHY:
3627                 mii = if_mii(ifr);
3628                 if (!mii)
3629                         return -EINVAL;
3630
3631                 mii->phy_id = 0;
3632                 /* Fall Through */
3633         case SIOCGMIIREG:
3634                 /*
3635                  * We do this again just in case we were called by SIOCGMIIREG
3636                  * instead of SIOCGMIIPHY.
3637                  */
3638                 mii = if_mii(ifr);
3639                 if (!mii)
3640                         return -EINVAL;
3641
3642
3643                 if (mii->reg_num == 1) {
3644                         struct bonding *bond = netdev_priv(bond_dev);
3645                         mii->val_out = 0;
3646                         read_lock(&bond->lock);
3647                         read_lock(&bond->curr_slave_lock);
3648                         if (netif_carrier_ok(bond->dev))
3649                                 mii->val_out = BMSR_LSTATUS;
3650
3651                         read_unlock(&bond->curr_slave_lock);
3652                         read_unlock(&bond->lock);
3653                 }
3654
3655                 return 0;
3656         case BOND_INFO_QUERY_OLD:
3657         case SIOCBONDINFOQUERY:
3658                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3659
3660                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3661                         return -EFAULT;
3662
3663                 res = bond_info_query(bond_dev, &k_binfo);
3664                 if (res == 0 &&
3665                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3666                         return -EFAULT;
3667
3668                 return res;
3669         case BOND_SLAVE_INFO_QUERY_OLD:
3670         case SIOCBONDSLAVEINFOQUERY:
3671                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3672
3673                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3674                         return -EFAULT;
3675
3676                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3677                 if (res == 0 &&
3678                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3679                         return -EFAULT;
3680
3681                 return res;
3682         default:
3683                 /* Go on */
3684                 break;
3685         }
3686
3687         if (!capable(CAP_NET_ADMIN))
3688                 return -EPERM;
3689
3690         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3691
3692         pr_debug("slave_dev=%p:\n", slave_dev);
3693
3694         if (!slave_dev)
3695                 res = -ENODEV;
3696         else {
3697                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3698                 switch (cmd) {
3699                 case BOND_ENSLAVE_OLD:
3700                 case SIOCBONDENSLAVE:
3701                         res = bond_enslave(bond_dev, slave_dev);
3702                         break;
3703                 case BOND_RELEASE_OLD:
3704                 case SIOCBONDRELEASE:
3705                         res = bond_release(bond_dev, slave_dev);
3706                         break;
3707                 case BOND_SETHWADDR_OLD:
3708                 case SIOCBONDSETHWADDR:
3709                         res = bond_sethwaddr(bond_dev, slave_dev);
3710                         break;
3711                 case BOND_CHANGE_ACTIVE_OLD:
3712                 case SIOCBONDCHANGEACTIVE:
3713                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3714                         break;
3715                 default:
3716                         res = -EOPNOTSUPP;
3717                 }
3718
3719                 dev_put(slave_dev);
3720         }
3721
3722         return res;
3723 }
3724
3725 static bool bond_addr_in_mc_list(unsigned char *addr,
3726                                  struct netdev_hw_addr_list *list,
3727                                  int addrlen)
3728 {
3729         struct netdev_hw_addr *ha;
3730
3731         netdev_hw_addr_list_for_each(ha, list)
3732                 if (!memcmp(ha->addr, addr, addrlen))
3733                         return true;
3734
3735         return false;
3736 }
3737
3738 static void bond_set_multicast_list(struct net_device *bond_dev)
3739 {
3740         struct bonding *bond = netdev_priv(bond_dev);
3741         struct netdev_hw_addr *ha;
3742         bool found;
3743
3744         /*
3745          * Do promisc before checking multicast_mode
3746          */
3747         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3748                 /*
3749                  * FIXME: Need to handle the error when one of the multi-slaves
3750                  * encounters error.
3751                  */
3752                 bond_set_promiscuity(bond, 1);
3753
3754
3755         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3756                 bond_set_promiscuity(bond, -1);
3757
3758
3759         /* set allmulti flag to slaves */
3760         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3761                 /*
3762                  * FIXME: Need to handle the error when one of the multi-slaves
3763                  * encounters error.
3764                  */
3765                 bond_set_allmulti(bond, 1);
3766
3767
3768         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3769                 bond_set_allmulti(bond, -1);
3770
3771
3772         read_lock(&bond->lock);
3773
3774         bond->flags = bond_dev->flags;
3775
3776         /* looking for addresses to add to slaves' mc list */
3777         netdev_for_each_mc_addr(ha, bond_dev) {
3778                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3779                                              bond_dev->addr_len);
3780                 if (!found)
3781                         bond_mc_add(bond, ha->addr);
3782         }
3783
3784         /* looking for addresses to delete from slaves' list */
3785         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3786                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3787                                              bond_dev->addr_len);
3788                 if (!found)
3789                         bond_mc_del(bond, ha->addr);
3790         }
3791
3792         /* save master's multicast list */
3793         __hw_addr_flush(&bond->mc_list);
3794         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3795                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3796
3797         read_unlock(&bond->lock);
3798 }
3799
3800 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3801 {
3802         struct bonding *bond = netdev_priv(dev);
3803         struct slave *slave = bond->first_slave;
3804
3805         if (slave) {
3806                 const struct net_device_ops *slave_ops
3807                         = slave->dev->netdev_ops;
3808                 if (slave_ops->ndo_neigh_setup)
3809                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3810         }
3811         return 0;
3812 }
3813
3814 /*
3815  * Change the MTU of all of a master's slaves to match the master
3816  */
3817 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3818 {
3819         struct bonding *bond = netdev_priv(bond_dev);
3820         struct slave *slave, *stop_at;
3821         int res = 0;
3822         int i;
3823
3824         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3825                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3826
3827         /* Can't hold bond->lock with bh disabled here since
3828          * some base drivers panic. On the other hand we can't
3829          * hold bond->lock without bh disabled because we'll
3830          * deadlock. The only solution is to rely on the fact
3831          * that we're under rtnl_lock here, and the slaves
3832          * list won't change. This doesn't solve the problem
3833          * of setting the slave's MTU while it is
3834          * transmitting, but the assumption is that the base
3835          * driver can handle that.
3836          *
3837          * TODO: figure out a way to safely iterate the slaves
3838          * list, but without holding a lock around the actual
3839          * call to the base driver.
3840          */
3841
3842         bond_for_each_slave(bond, slave, i) {
3843                 pr_debug("s %p s->p %p c_m %p\n",
3844                          slave,
3845                          slave->prev,
3846                          slave->dev->netdev_ops->ndo_change_mtu);
3847
3848                 res = dev_set_mtu(slave->dev, new_mtu);
3849
3850                 if (res) {
3851                         /* If we failed to set the slave's mtu to the new value
3852                          * we must abort the operation even in ACTIVE_BACKUP
3853                          * mode, because if we allow the backup slaves to have
3854                          * different mtu values than the active slave we'll
3855                          * need to change their mtu when doing a failover. That
3856                          * means changing their mtu from timer context, which
3857                          * is probably not a good idea.
3858                          */
3859                         pr_debug("err %d %s\n", res, slave->dev->name);
3860                         goto unwind;
3861                 }
3862         }
3863
3864         bond_dev->mtu = new_mtu;
3865
3866         return 0;
3867
3868 unwind:
3869         /* unwind from head to the slave that failed */
3870         stop_at = slave;
3871         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3872                 int tmp_res;
3873
3874                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3875                 if (tmp_res) {
3876                         pr_debug("unwind err %d dev %s\n",
3877                                  tmp_res, slave->dev->name);
3878                 }
3879         }
3880
3881         return res;
3882 }
3883
3884 /*
3885  * Change HW address
3886  *
3887  * Note that many devices must be down to change the HW address, and
3888  * downing the master releases all slaves.  We can make bonds full of
3889  * bonding devices to test this, however.
3890  */
3891 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3892 {
3893         struct bonding *bond = netdev_priv(bond_dev);
3894         struct sockaddr *sa = addr, tmp_sa;
3895         struct slave *slave, *stop_at;
3896         int res = 0;
3897         int i;
3898
3899         if (bond->params.mode == BOND_MODE_ALB)
3900                 return bond_alb_set_mac_address(bond_dev, addr);
3901
3902
3903         pr_debug("bond=%p, name=%s\n",
3904                  bond, bond_dev ? bond_dev->name : "None");
3905
3906         /*
3907          * If fail_over_mac is set to active, do nothing and return
3908          * success.  Returning an error causes ifenslave to fail.
3909          */
3910         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3911                 return 0;
3912
3913         if (!is_valid_ether_addr(sa->sa_data))
3914                 return -EADDRNOTAVAIL;
3915
3916         /* Can't hold bond->lock with bh disabled here since
3917          * some base drivers panic. On the other hand we can't
3918          * hold bond->lock without bh disabled because we'll
3919          * deadlock. The only solution is to rely on the fact
3920          * that we're under rtnl_lock here, and the slaves
3921          * list won't change. This doesn't solve the problem
3922          * of setting the slave's hw address while it is
3923          * transmitting, but the assumption is that the base
3924          * driver can handle that.
3925          *
3926          * TODO: figure out a way to safely iterate the slaves
3927          * list, but without holding a lock around the actual
3928          * call to the base driver.
3929          */
3930
3931         bond_for_each_slave(bond, slave, i) {
3932                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3933                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3934
3935                 if (slave_ops->ndo_set_mac_address == NULL) {
3936                         res = -EOPNOTSUPP;
3937                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3938                         goto unwind;
3939                 }
3940
3941                 res = dev_set_mac_address(slave->dev, addr);
3942                 if (res) {
3943                         /* TODO: consider downing the slave
3944                          * and retry ?
3945                          * User should expect communications
3946                          * breakage anyway until ARP finish
3947                          * updating, so...
3948                          */
3949                         pr_debug("err %d %s\n", res, slave->dev->name);
3950                         goto unwind;
3951                 }
3952         }
3953
3954         /* success */
3955         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3956         return 0;
3957
3958 unwind:
3959         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3960         tmp_sa.sa_family = bond_dev->type;
3961
3962         /* unwind from head to the slave that failed */
3963         stop_at = slave;
3964         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3965                 int tmp_res;
3966
3967                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3968                 if (tmp_res) {
3969                         pr_debug("unwind err %d dev %s\n",
3970                                  tmp_res, slave->dev->name);
3971                 }
3972         }
3973
3974         return res;
3975 }
3976
3977 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3978 {
3979         struct bonding *bond = netdev_priv(bond_dev);
3980         struct slave *slave, *start_at;
3981         int i, slave_no, res = 1;
3982         struct iphdr *iph = ip_hdr(skb);
3983
3984         /*
3985          * Start with the curr_active_slave that joined the bond as the
3986          * default for sending IGMP traffic.  For failover purposes one
3987          * needs to maintain some consistency for the interface that will
3988          * send the join/membership reports.  The curr_active_slave found
3989          * will send all of this type of traffic.
3990          */
3991         if ((iph->protocol == IPPROTO_IGMP) &&
3992             (skb->protocol == htons(ETH_P_IP))) {
3993
3994                 read_lock(&bond->curr_slave_lock);
3995                 slave = bond->curr_active_slave;
3996                 read_unlock(&bond->curr_slave_lock);
3997
3998                 if (!slave)
3999                         goto out;
4000         } else {
4001                 /*
4002                  * Concurrent TX may collide on rr_tx_counter; we accept
4003                  * that as being rare enough not to justify using an
4004                  * atomic op here.
4005                  */
4006                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4007
4008                 bond_for_each_slave(bond, slave, i) {
4009                         slave_no--;
4010                         if (slave_no < 0)
4011                                 break;
4012                 }
4013         }
4014
4015         start_at = slave;
4016         bond_for_each_slave_from(bond, slave, i, start_at) {
4017                 if (IS_UP(slave->dev) &&
4018                     (slave->link == BOND_LINK_UP) &&
4019                     bond_is_active_slave(slave)) {
4020                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4021                         break;
4022                 }
4023         }
4024
4025 out:
4026         if (res) {
4027                 /* no suitable interface, frame not sent */
4028                 dev_kfree_skb(skb);
4029         }
4030
4031         return NETDEV_TX_OK;
4032 }
4033
4034
4035 /*
4036  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4037  * the bond has a usable interface.
4038  */
4039 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4040 {
4041         struct bonding *bond = netdev_priv(bond_dev);
4042         int res = 1;
4043
4044         read_lock(&bond->curr_slave_lock);
4045
4046         if (bond->curr_active_slave)
4047                 res = bond_dev_queue_xmit(bond, skb,
4048                         bond->curr_active_slave->dev);
4049
4050         if (res)
4051                 /* no suitable interface, frame not sent */
4052                 dev_kfree_skb(skb);
4053
4054         read_unlock(&bond->curr_slave_lock);
4055
4056         return NETDEV_TX_OK;
4057 }
4058
4059 /*
4060  * In bond_xmit_xor() , we determine the output device by using a pre-
4061  * determined xmit_hash_policy(), If the selected device is not enabled,
4062  * find the next active slave.
4063  */
4064 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4065 {
4066         struct bonding *bond = netdev_priv(bond_dev);
4067         struct slave *slave, *start_at;
4068         int slave_no;
4069         int i;
4070         int res = 1;
4071
4072         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4073
4074         bond_for_each_slave(bond, slave, i) {
4075                 slave_no--;
4076                 if (slave_no < 0)
4077                         break;
4078         }
4079
4080         start_at = slave;
4081
4082         bond_for_each_slave_from(bond, slave, i, start_at) {
4083                 if (IS_UP(slave->dev) &&
4084                     (slave->link == BOND_LINK_UP) &&
4085                     bond_is_active_slave(slave)) {
4086                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4087                         break;
4088                 }
4089         }
4090
4091         if (res) {
4092                 /* no suitable interface, frame not sent */
4093                 dev_kfree_skb(skb);
4094         }
4095
4096         return NETDEV_TX_OK;
4097 }
4098
4099 /*
4100  * in broadcast mode, we send everything to all usable interfaces.
4101  */
4102 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4103 {
4104         struct bonding *bond = netdev_priv(bond_dev);
4105         struct slave *slave, *start_at;
4106         struct net_device *tx_dev = NULL;
4107         int i;
4108         int res = 1;
4109
4110         read_lock(&bond->curr_slave_lock);
4111         start_at = bond->curr_active_slave;
4112         read_unlock(&bond->curr_slave_lock);
4113
4114         if (!start_at)
4115                 goto out;
4116
4117         bond_for_each_slave_from(bond, slave, i, start_at) {
4118                 if (IS_UP(slave->dev) &&
4119                     (slave->link == BOND_LINK_UP) &&
4120                     bond_is_active_slave(slave)) {
4121                         if (tx_dev) {
4122                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4123                                 if (!skb2) {
4124                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4125                                                bond_dev->name);
4126                                         continue;
4127                                 }
4128
4129                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4130                                 if (res) {
4131                                         dev_kfree_skb(skb2);
4132                                         continue;
4133                                 }
4134                         }
4135                         tx_dev = slave->dev;
4136                 }
4137         }
4138
4139         if (tx_dev)
4140                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4141
4142 out:
4143         if (res)
4144                 /* no suitable interface, frame not sent */
4145                 dev_kfree_skb(skb);
4146
4147         /* frame sent to all suitable interfaces */
4148         return NETDEV_TX_OK;
4149 }
4150
4151 /*------------------------- Device initialization ---------------------------*/
4152
4153 static void bond_set_xmit_hash_policy(struct bonding *bond)
4154 {
4155         switch (bond->params.xmit_policy) {
4156         case BOND_XMIT_POLICY_LAYER23:
4157                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4158                 break;
4159         case BOND_XMIT_POLICY_LAYER34:
4160                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4161                 break;
4162         case BOND_XMIT_POLICY_LAYER2:
4163         default:
4164                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4165                 break;
4166         }
4167 }
4168
4169 /*
4170  * Lookup the slave that corresponds to a qid
4171  */
4172 static inline int bond_slave_override(struct bonding *bond,
4173                                       struct sk_buff *skb)
4174 {
4175         int i, res = 1;
4176         struct slave *slave = NULL;
4177         struct slave *check_slave;
4178
4179         if (!skb->queue_mapping)
4180                 return 1;
4181
4182         /* Find out if any slaves have the same mapping as this skb. */
4183         bond_for_each_slave(bond, check_slave, i) {
4184                 if (check_slave->queue_id == skb->queue_mapping) {
4185                         slave = check_slave;
4186                         break;
4187                 }
4188         }
4189
4190         /* If the slave isn't UP, use default transmit policy. */
4191         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4192             (slave->link == BOND_LINK_UP)) {
4193                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4194         }
4195
4196         return res;
4197 }
4198
4199 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4200 {
4201         /*
4202          * This helper function exists to help dev_pick_tx get the correct
4203          * destination queue.  Using a helper function skips a call to
4204          * skb_tx_hash and will put the skbs in the queue we expect on their
4205          * way down to the bonding driver.
4206          */
4207         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4208
4209         if (unlikely(txq >= dev->real_num_tx_queues)) {
4210                 do {
4211                         txq -= dev->real_num_tx_queues;
4212                 } while (txq >= dev->real_num_tx_queues);
4213         }
4214         return txq;
4215 }
4216
4217 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4218 {
4219         struct bonding *bond = netdev_priv(dev);
4220
4221         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4222                 if (!bond_slave_override(bond, skb))
4223                         return NETDEV_TX_OK;
4224         }
4225
4226         switch (bond->params.mode) {
4227         case BOND_MODE_ROUNDROBIN:
4228                 return bond_xmit_roundrobin(skb, dev);
4229         case BOND_MODE_ACTIVEBACKUP:
4230                 return bond_xmit_activebackup(skb, dev);
4231         case BOND_MODE_XOR:
4232                 return bond_xmit_xor(skb, dev);
4233         case BOND_MODE_BROADCAST:
4234                 return bond_xmit_broadcast(skb, dev);
4235         case BOND_MODE_8023AD:
4236                 return bond_3ad_xmit_xor(skb, dev);
4237         case BOND_MODE_ALB:
4238         case BOND_MODE_TLB:
4239                 return bond_alb_xmit(skb, dev);
4240         default:
4241                 /* Should never happen, mode already checked */
4242                 pr_err("%s: Error: Unknown bonding mode %d\n",
4243                        dev->name, bond->params.mode);
4244                 WARN_ON_ONCE(1);
4245                 dev_kfree_skb(skb);
4246                 return NETDEV_TX_OK;
4247         }
4248 }
4249
4250 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4251 {
4252         struct bonding *bond = netdev_priv(dev);
4253         netdev_tx_t ret = NETDEV_TX_OK;
4254
4255         /*
4256          * If we risk deadlock from transmitting this in the
4257          * netpoll path, tell netpoll to queue the frame for later tx
4258          */
4259         if (is_netpoll_tx_blocked(dev))
4260                 return NETDEV_TX_BUSY;
4261
4262         read_lock(&bond->lock);
4263
4264         if (bond->slave_cnt)
4265                 ret = __bond_start_xmit(skb, dev);
4266         else
4267                 dev_kfree_skb(skb);
4268
4269         read_unlock(&bond->lock);
4270
4271         return ret;
4272 }
4273
4274 /*
4275  * set bond mode specific net device operations
4276  */
4277 void bond_set_mode_ops(struct bonding *bond, int mode)
4278 {
4279         struct net_device *bond_dev = bond->dev;
4280
4281         switch (mode) {
4282         case BOND_MODE_ROUNDROBIN:
4283                 break;
4284         case BOND_MODE_ACTIVEBACKUP:
4285                 break;
4286         case BOND_MODE_XOR:
4287                 bond_set_xmit_hash_policy(bond);
4288                 break;
4289         case BOND_MODE_BROADCAST:
4290                 break;
4291         case BOND_MODE_8023AD:
4292                 bond_set_xmit_hash_policy(bond);
4293                 break;
4294         case BOND_MODE_ALB:
4295                 /* FALLTHRU */
4296         case BOND_MODE_TLB:
4297                 break;
4298         default:
4299                 /* Should never happen, mode already checked */
4300                 pr_err("%s: Error: Unknown bonding mode %d\n",
4301                        bond_dev->name, mode);
4302                 break;
4303         }
4304 }
4305
4306 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4307                                     struct ethtool_drvinfo *drvinfo)
4308 {
4309         strncpy(drvinfo->driver, DRV_NAME, 32);
4310         strncpy(drvinfo->version, DRV_VERSION, 32);
4311         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4312 }
4313
4314 static const struct ethtool_ops bond_ethtool_ops = {
4315         .get_drvinfo            = bond_ethtool_get_drvinfo,
4316         .get_link               = ethtool_op_get_link,
4317 };
4318
4319 static const struct net_device_ops bond_netdev_ops = {
4320         .ndo_init               = bond_init,
4321         .ndo_uninit             = bond_uninit,
4322         .ndo_open               = bond_open,
4323         .ndo_stop               = bond_close,
4324         .ndo_start_xmit         = bond_start_xmit,
4325         .ndo_select_queue       = bond_select_queue,
4326         .ndo_get_stats64        = bond_get_stats,
4327         .ndo_do_ioctl           = bond_do_ioctl,
4328         .ndo_set_multicast_list = bond_set_multicast_list,
4329         .ndo_change_mtu         = bond_change_mtu,
4330         .ndo_set_mac_address    = bond_set_mac_address,
4331         .ndo_neigh_setup        = bond_neigh_setup,
4332         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4333         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4334         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4335 #ifdef CONFIG_NET_POLL_CONTROLLER
4336         .ndo_netpoll_setup      = bond_netpoll_setup,
4337         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4338         .ndo_poll_controller    = bond_poll_controller,
4339 #endif
4340         .ndo_add_slave          = bond_enslave,
4341         .ndo_del_slave          = bond_release,
4342         .ndo_fix_features       = bond_fix_features,
4343 };
4344
4345 static void bond_destructor(struct net_device *bond_dev)
4346 {
4347         struct bonding *bond = netdev_priv(bond_dev);
4348         if (bond->wq)
4349                 destroy_workqueue(bond->wq);
4350         free_netdev(bond_dev);
4351 }
4352
4353 static void bond_setup(struct net_device *bond_dev)
4354 {
4355         struct bonding *bond = netdev_priv(bond_dev);
4356
4357         /* initialize rwlocks */
4358         rwlock_init(&bond->lock);
4359         rwlock_init(&bond->curr_slave_lock);
4360
4361         bond->params = bonding_defaults;
4362
4363         /* Initialize pointers */
4364         bond->dev = bond_dev;
4365         INIT_LIST_HEAD(&bond->vlan_list);
4366
4367         /* Initialize the device entry points */
4368         ether_setup(bond_dev);
4369         bond_dev->netdev_ops = &bond_netdev_ops;
4370         bond_dev->ethtool_ops = &bond_ethtool_ops;
4371         bond_set_mode_ops(bond, bond->params.mode);
4372
4373         bond_dev->destructor = bond_destructor;
4374
4375         /* Initialize the device options */
4376         bond_dev->tx_queue_len = 0;
4377         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4378         bond_dev->priv_flags |= IFF_BONDING;
4379         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4380
4381         /* At first, we block adding VLANs. That's the only way to
4382          * prevent problems that occur when adding VLANs over an
4383          * empty bond. The block will be removed once non-challenged
4384          * slaves are enslaved.
4385          */
4386         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4387
4388         /* don't acquire bond device's netif_tx_lock when
4389          * transmitting */
4390         bond_dev->features |= NETIF_F_LLTX;
4391
4392         /* By default, we declare the bond to be fully
4393          * VLAN hardware accelerated capable. Special
4394          * care is taken in the various xmit functions
4395          * when there are slaves that are not hw accel
4396          * capable
4397          */
4398
4399         bond_dev->hw_features = BOND_VLAN_FEATURES |
4400                                 NETIF_F_HW_VLAN_TX |
4401                                 NETIF_F_HW_VLAN_RX |
4402                                 NETIF_F_HW_VLAN_FILTER;
4403
4404         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4405         bond_dev->features |= bond_dev->hw_features;
4406 }
4407
4408 static void bond_work_cancel_all(struct bonding *bond)
4409 {
4410         write_lock_bh(&bond->lock);
4411         bond->kill_timers = 1;
4412         write_unlock_bh(&bond->lock);
4413
4414         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4415                 cancel_delayed_work(&bond->mii_work);
4416
4417         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4418                 cancel_delayed_work(&bond->arp_work);
4419
4420         if (bond->params.mode == BOND_MODE_ALB &&
4421             delayed_work_pending(&bond->alb_work))
4422                 cancel_delayed_work(&bond->alb_work);
4423
4424         if (bond->params.mode == BOND_MODE_8023AD &&
4425             delayed_work_pending(&bond->ad_work))
4426                 cancel_delayed_work(&bond->ad_work);
4427
4428         if (delayed_work_pending(&bond->mcast_work))
4429                 cancel_delayed_work(&bond->mcast_work);
4430 }
4431
4432 /*
4433 * Destroy a bonding device.
4434 * Must be under rtnl_lock when this function is called.
4435 */
4436 static void bond_uninit(struct net_device *bond_dev)
4437 {
4438         struct bonding *bond = netdev_priv(bond_dev);
4439         struct vlan_entry *vlan, *tmp;
4440
4441         bond_netpoll_cleanup(bond_dev);
4442
4443         /* Release the bonded slaves */
4444         bond_release_all(bond_dev);
4445
4446         list_del(&bond->bond_list);
4447
4448         bond_work_cancel_all(bond);
4449
4450         bond_remove_proc_entry(bond);
4451
4452         bond_debug_unregister(bond);
4453
4454         __hw_addr_flush(&bond->mc_list);
4455
4456         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4457                 list_del(&vlan->vlan_list);
4458                 kfree(vlan);
4459         }
4460 }
4461
4462 /*------------------------- Module initialization ---------------------------*/
4463
4464 /*
4465  * Convert string input module parms.  Accept either the
4466  * number of the mode or its string name.  A bit complicated because
4467  * some mode names are substrings of other names, and calls from sysfs
4468  * may have whitespace in the name (trailing newlines, for example).
4469  */
4470 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4471 {
4472         int modeint = -1, i, rv;
4473         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4474
4475         for (p = (char *)buf; *p; p++)
4476                 if (!(isdigit(*p) || isspace(*p)))
4477                         break;
4478
4479         if (*p)
4480                 rv = sscanf(buf, "%20s", modestr);
4481         else
4482                 rv = sscanf(buf, "%d", &modeint);
4483
4484         if (!rv)
4485                 return -1;
4486
4487         for (i = 0; tbl[i].modename; i++) {
4488                 if (modeint == tbl[i].mode)
4489                         return tbl[i].mode;
4490                 if (strcmp(modestr, tbl[i].modename) == 0)
4491                         return tbl[i].mode;
4492         }
4493
4494         return -1;
4495 }
4496
4497 static int bond_check_params(struct bond_params *params)
4498 {
4499         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4500
4501         /*
4502          * Convert string parameters.
4503          */
4504         if (mode) {
4505                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4506                 if (bond_mode == -1) {
4507                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4508                                mode == NULL ? "NULL" : mode);
4509                         return -EINVAL;
4510                 }
4511         }
4512
4513         if (xmit_hash_policy) {
4514                 if ((bond_mode != BOND_MODE_XOR) &&
4515                     (bond_mode != BOND_MODE_8023AD)) {
4516                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4517                                bond_mode_name(bond_mode));
4518                 } else {
4519                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4520                                                         xmit_hashtype_tbl);
4521                         if (xmit_hashtype == -1) {
4522                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4523                                        xmit_hash_policy == NULL ? "NULL" :
4524                                        xmit_hash_policy);
4525                                 return -EINVAL;
4526                         }
4527                 }
4528         }
4529
4530         if (lacp_rate) {
4531                 if (bond_mode != BOND_MODE_8023AD) {
4532                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4533                                 bond_mode_name(bond_mode));
4534                 } else {
4535                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4536                         if (lacp_fast == -1) {
4537                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4538                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4539                                 return -EINVAL;
4540                         }
4541                 }
4542         }
4543
4544         if (ad_select) {
4545                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4546                 if (params->ad_select == -1) {
4547                         pr_err("Error: Invalid ad_select \"%s\"\n",
4548                                ad_select == NULL ? "NULL" : ad_select);
4549                         return -EINVAL;
4550                 }
4551
4552                 if (bond_mode != BOND_MODE_8023AD) {
4553                         pr_warning("ad_select param only affects 802.3ad mode\n");
4554                 }
4555         } else {
4556                 params->ad_select = BOND_AD_STABLE;
4557         }
4558
4559         if (max_bonds < 0) {
4560                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4561                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4562                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4563         }
4564
4565         if (miimon < 0) {
4566                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4567                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4568                 miimon = BOND_LINK_MON_INTERV;
4569         }
4570
4571         if (updelay < 0) {
4572                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4573                            updelay, INT_MAX);
4574                 updelay = 0;
4575         }
4576
4577         if (downdelay < 0) {
4578                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4579                            downdelay, INT_MAX);
4580                 downdelay = 0;
4581         }
4582
4583         if ((use_carrier != 0) && (use_carrier != 1)) {
4584                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4585                            use_carrier);
4586                 use_carrier = 1;
4587         }
4588
4589         if (num_peer_notif < 0 || num_peer_notif > 255) {
4590                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4591                            num_peer_notif);
4592                 num_peer_notif = 1;
4593         }
4594
4595         /* reset values for 802.3ad */
4596         if (bond_mode == BOND_MODE_8023AD) {
4597                 if (!miimon) {
4598                         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");
4599                         pr_warning("Forcing miimon to 100msec\n");
4600                         miimon = 100;
4601                 }
4602         }
4603
4604         if (tx_queues < 1 || tx_queues > 255) {
4605                 pr_warning("Warning: tx_queues (%d) should be between "
4606                            "1 and 255, resetting to %d\n",
4607                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4608                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4609         }
4610
4611         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4612                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4613                            "not of valid value (0/1), so it was set to "
4614                            "0\n", all_slaves_active);
4615                 all_slaves_active = 0;
4616         }
4617
4618         if (resend_igmp < 0 || resend_igmp > 255) {
4619                 pr_warning("Warning: resend_igmp (%d) should be between "
4620                            "0 and 255, resetting to %d\n",
4621                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4622                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4623         }
4624
4625         /* reset values for TLB/ALB */
4626         if ((bond_mode == BOND_MODE_TLB) ||
4627             (bond_mode == BOND_MODE_ALB)) {
4628                 if (!miimon) {
4629                         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");
4630                         pr_warning("Forcing miimon to 100msec\n");
4631                         miimon = 100;
4632                 }
4633         }
4634
4635         if (bond_mode == BOND_MODE_ALB) {
4636                 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",
4637                           updelay);
4638         }
4639
4640         if (!miimon) {
4641                 if (updelay || downdelay) {
4642                         /* just warn the user the up/down delay will have
4643                          * no effect since miimon is zero...
4644                          */
4645                         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",
4646                                    updelay, downdelay);
4647                 }
4648         } else {
4649                 /* don't allow arp monitoring */
4650                 if (arp_interval) {
4651                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4652                                    miimon, arp_interval);
4653                         arp_interval = 0;
4654                 }
4655
4656                 if ((updelay % miimon) != 0) {
4657                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4658                                    updelay, miimon,
4659                                    (updelay / miimon) * miimon);
4660                 }
4661
4662                 updelay /= miimon;
4663
4664                 if ((downdelay % miimon) != 0) {
4665                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4666                                    downdelay, miimon,
4667                                    (downdelay / miimon) * miimon);
4668                 }
4669
4670                 downdelay /= miimon;
4671         }
4672
4673         if (arp_interval < 0) {
4674                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4675                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4676                 arp_interval = BOND_LINK_ARP_INTERV;
4677         }
4678
4679         for (arp_ip_count = 0;
4680              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4681              arp_ip_count++) {
4682                 /* not complete check, but should be good enough to
4683                    catch mistakes */
4684                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4685                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4686                                    arp_ip_target[arp_ip_count]);
4687                         arp_interval = 0;
4688                 } else {
4689                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4690                         arp_target[arp_ip_count] = ip;
4691                 }
4692         }
4693
4694         if (arp_interval && !arp_ip_count) {
4695                 /* don't allow arping if no arp_ip_target given... */
4696                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4697                            arp_interval);
4698                 arp_interval = 0;
4699         }
4700
4701         if (arp_validate) {
4702                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4703                         pr_err("arp_validate only supported in active-backup mode\n");
4704                         return -EINVAL;
4705                 }
4706                 if (!arp_interval) {
4707                         pr_err("arp_validate requires arp_interval\n");
4708                         return -EINVAL;
4709                 }
4710
4711                 arp_validate_value = bond_parse_parm(arp_validate,
4712                                                      arp_validate_tbl);
4713                 if (arp_validate_value == -1) {
4714                         pr_err("Error: invalid arp_validate \"%s\"\n",
4715                                arp_validate == NULL ? "NULL" : arp_validate);
4716                         return -EINVAL;
4717                 }
4718         } else
4719                 arp_validate_value = 0;
4720
4721         if (miimon) {
4722                 pr_info("MII link monitoring set to %d ms\n", miimon);
4723         } else if (arp_interval) {
4724                 int i;
4725
4726                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4727                         arp_interval,
4728                         arp_validate_tbl[arp_validate_value].modename,
4729                         arp_ip_count);
4730
4731                 for (i = 0; i < arp_ip_count; i++)
4732                         pr_info(" %s", arp_ip_target[i]);
4733
4734                 pr_info("\n");
4735
4736         } else if (max_bonds) {
4737                 /* miimon and arp_interval not set, we need one so things
4738                  * work as expected, see bonding.txt for details
4739                  */
4740                 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");
4741         }
4742
4743         if (primary && !USES_PRIMARY(bond_mode)) {
4744                 /* currently, using a primary only makes sense
4745                  * in active backup, TLB or ALB modes
4746                  */
4747                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4748                            primary, bond_mode_name(bond_mode));
4749                 primary = NULL;
4750         }
4751
4752         if (primary && primary_reselect) {
4753                 primary_reselect_value = bond_parse_parm(primary_reselect,
4754                                                          pri_reselect_tbl);
4755                 if (primary_reselect_value == -1) {
4756                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4757                                primary_reselect ==
4758                                         NULL ? "NULL" : primary_reselect);
4759                         return -EINVAL;
4760                 }
4761         } else {
4762                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4763         }
4764
4765         if (fail_over_mac) {
4766                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4767                                                       fail_over_mac_tbl);
4768                 if (fail_over_mac_value == -1) {
4769                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4770                                arp_validate == NULL ? "NULL" : arp_validate);
4771                         return -EINVAL;
4772                 }
4773
4774                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4775                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4776         } else {
4777                 fail_over_mac_value = BOND_FOM_NONE;
4778         }
4779
4780         /* fill params struct with the proper values */
4781         params->mode = bond_mode;
4782         params->xmit_policy = xmit_hashtype;
4783         params->miimon = miimon;
4784         params->num_peer_notif = num_peer_notif;
4785         params->arp_interval = arp_interval;
4786         params->arp_validate = arp_validate_value;
4787         params->updelay = updelay;
4788         params->downdelay = downdelay;
4789         params->use_carrier = use_carrier;
4790         params->lacp_fast = lacp_fast;
4791         params->primary[0] = 0;
4792         params->primary_reselect = primary_reselect_value;
4793         params->fail_over_mac = fail_over_mac_value;
4794         params->tx_queues = tx_queues;
4795         params->all_slaves_active = all_slaves_active;
4796         params->resend_igmp = resend_igmp;
4797
4798         if (primary) {
4799                 strncpy(params->primary, primary, IFNAMSIZ);
4800                 params->primary[IFNAMSIZ - 1] = 0;
4801         }
4802
4803         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4804
4805         return 0;
4806 }
4807
4808 static struct lock_class_key bonding_netdev_xmit_lock_key;
4809 static struct lock_class_key bonding_netdev_addr_lock_key;
4810
4811 static void bond_set_lockdep_class_one(struct net_device *dev,
4812                                        struct netdev_queue *txq,
4813                                        void *_unused)
4814 {
4815         lockdep_set_class(&txq->_xmit_lock,
4816                           &bonding_netdev_xmit_lock_key);
4817 }
4818
4819 static void bond_set_lockdep_class(struct net_device *dev)
4820 {
4821         lockdep_set_class(&dev->addr_list_lock,
4822                           &bonding_netdev_addr_lock_key);
4823         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4824 }
4825
4826 /*
4827  * Called from registration process
4828  */
4829 static int bond_init(struct net_device *bond_dev)
4830 {
4831         struct bonding *bond = netdev_priv(bond_dev);
4832         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4833         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4834
4835         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4836
4837         /*
4838          * Initialize locks that may be required during
4839          * en/deslave operations.  All of the bond_open work
4840          * (of which this is part) should really be moved to
4841          * a phase prior to dev_open
4842          */
4843         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4844         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4845
4846         bond->wq = create_singlethread_workqueue(bond_dev->name);
4847         if (!bond->wq)
4848                 return -ENOMEM;
4849
4850         bond_set_lockdep_class(bond_dev);
4851
4852         bond_create_proc_entry(bond);
4853         list_add_tail(&bond->bond_list, &bn->dev_list);
4854
4855         bond_prepare_sysfs_group(bond);
4856
4857         bond_debug_register(bond);
4858
4859         __hw_addr_init(&bond->mc_list);
4860         return 0;
4861 }
4862
4863 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4864 {
4865         if (tb[IFLA_ADDRESS]) {
4866                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4867                         return -EINVAL;
4868                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4869                         return -EADDRNOTAVAIL;
4870         }
4871         return 0;
4872 }
4873
4874 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4875         .kind           = "bond",
4876         .priv_size      = sizeof(struct bonding),
4877         .setup          = bond_setup,
4878         .validate       = bond_validate,
4879 };
4880
4881 /* Create a new bond based on the specified name and bonding parameters.
4882  * If name is NULL, obtain a suitable "bond%d" name for us.
4883  * Caller must NOT hold rtnl_lock; we need to release it here before we
4884  * set up our sysfs entries.
4885  */
4886 int bond_create(struct net *net, const char *name)
4887 {
4888         struct net_device *bond_dev;
4889         int res;
4890
4891         rtnl_lock();
4892
4893         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4894                                    name ? name : "bond%d",
4895                                    bond_setup, tx_queues);
4896         if (!bond_dev) {
4897                 pr_err("%s: eek! can't alloc netdev!\n", name);
4898                 rtnl_unlock();
4899                 return -ENOMEM;
4900         }
4901
4902         dev_net_set(bond_dev, net);
4903         bond_dev->rtnl_link_ops = &bond_link_ops;
4904
4905         res = register_netdevice(bond_dev);
4906
4907         netif_carrier_off(bond_dev);
4908
4909         rtnl_unlock();
4910         if (res < 0)
4911                 bond_destructor(bond_dev);
4912         return res;
4913 }
4914
4915 static int __net_init bond_net_init(struct net *net)
4916 {
4917         struct bond_net *bn = net_generic(net, bond_net_id);
4918
4919         bn->net = net;
4920         INIT_LIST_HEAD(&bn->dev_list);
4921
4922         bond_create_proc_dir(bn);
4923         
4924         return 0;
4925 }
4926
4927 static void __net_exit bond_net_exit(struct net *net)
4928 {
4929         struct bond_net *bn = net_generic(net, bond_net_id);
4930
4931         bond_destroy_proc_dir(bn);
4932 }
4933
4934 static struct pernet_operations bond_net_ops = {
4935         .init = bond_net_init,
4936         .exit = bond_net_exit,
4937         .id   = &bond_net_id,
4938         .size = sizeof(struct bond_net),
4939 };
4940
4941 static int __init bonding_init(void)
4942 {
4943         int i;
4944         int res;
4945
4946         pr_info("%s", bond_version);
4947
4948         res = bond_check_params(&bonding_defaults);
4949         if (res)
4950                 goto out;
4951
4952         res = register_pernet_subsys(&bond_net_ops);
4953         if (res)
4954                 goto out;
4955
4956         res = rtnl_link_register(&bond_link_ops);
4957         if (res)
4958                 goto err_link;
4959
4960         bond_create_debugfs();
4961
4962         for (i = 0; i < max_bonds; i++) {
4963                 res = bond_create(&init_net, NULL);
4964                 if (res)
4965                         goto err;
4966         }
4967
4968         res = bond_create_sysfs();
4969         if (res)
4970                 goto err;
4971
4972         register_netdevice_notifier(&bond_netdev_notifier);
4973         register_inetaddr_notifier(&bond_inetaddr_notifier);
4974 out:
4975         return res;
4976 err:
4977         rtnl_link_unregister(&bond_link_ops);
4978 err_link:
4979         unregister_pernet_subsys(&bond_net_ops);
4980         goto out;
4981
4982 }
4983
4984 static void __exit bonding_exit(void)
4985 {
4986         unregister_netdevice_notifier(&bond_netdev_notifier);
4987         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4988
4989         bond_destroy_sysfs();
4990         bond_destroy_debugfs();
4991
4992         rtnl_link_unregister(&bond_link_ops);
4993         unregister_pernet_subsys(&bond_net_ops);
4994
4995 #ifdef CONFIG_NET_POLL_CONTROLLER
4996         /*
4997          * Make sure we don't have an imbalance on our netpoll blocking
4998          */
4999         WARN_ON(atomic_read(&netpoll_block_tx));
5000 #endif
5001 }
5002
5003 module_init(bonding_init);
5004 module_exit(bonding_exit);
5005 MODULE_LICENSE("GPL");
5006 MODULE_VERSION(DRV_VERSION);
5007 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5008 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5009 MODULE_ALIAS_RTNL_LINK("bond");