71efff323941e28fc5886d9b6800bdd438de972e
[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 int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval = BOND_LINK_ARP_INTERV;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *fail_over_mac;
108 static int all_slaves_active = 0;
109 static struct bond_params bonding_defaults;
110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
111
112 module_param(max_bonds, int, 0);
113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
114 module_param(tx_queues, int, 0);
115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
118                                "failover event (alias of num_unsol_na)");
119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
121                                "failover event (alias of num_grat_arp)");
122 module_param(miimon, int, 0);
123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124 module_param(updelay, int, 0);
125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126 module_param(downdelay, int, 0);
127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128                             "in milliseconds");
129 module_param(use_carrier, int, 0);
130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131                               "0 for off, 1 for on (default)");
132 module_param(mode, charp, 0);
133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
134                        "1 for active-backup, 2 for balance-xor, "
135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136                        "6 for balance-alb");
137 module_param(primary, charp, 0);
138 MODULE_PARM_DESC(primary, "Primary network device to use");
139 module_param(primary_reselect, charp, 0);
140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141                                    "once it comes up; "
142                                    "0 for always (default), "
143                                    "1 for only if speed of primary is "
144                                    "better, "
145                                    "2 for only on active slave "
146                                    "failure");
147 module_param(lacp_rate, charp, 0);
148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
149                             "0 for slow, 1 for fast");
150 module_param(ad_select, charp, 0);
151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
152                             "0 for stable (default), 1 for bandwidth, "
153                             "2 for count");
154 module_param(min_links, int, 0);
155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
156
157 module_param(xmit_hash_policy, charp, 0);
158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
159                                    "0 for layer 2 (default), 1 for layer 3+4, "
160                                    "2 for layer 2+3");
161 module_param(arp_interval, int, 0);
162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
163 module_param_array(arp_ip_target, charp, NULL, 0);
164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
165 module_param(arp_validate, charp, 0);
166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
167                                "0 for none (default), 1 for active, "
168                                "2 for backup, 3 for all");
169 module_param(fail_over_mac, charp, 0);
170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
171                                 "the same MAC; 0 for none (default), "
172                                 "1 for active, 2 for follow");
173 module_param(all_slaves_active, int, 0);
174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
175                                      "by setting active flag for all slaves; "
176                                      "0 for never (default), 1 for always.");
177 module_param(resend_igmp, int, 0);
178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
179                               "link failure");
180
181 /*----------------------------- Global variables ----------------------------*/
182
183 #ifdef CONFIG_NET_POLL_CONTROLLER
184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
185 #endif
186
187 int bond_net_id __read_mostly;
188
189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
190 static int arp_ip_count;
191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
193 static int lacp_fast;
194
195 const struct bond_parm_tbl bond_lacp_tbl[] = {
196 {       "slow",         AD_LACP_SLOW},
197 {       "fast",         AD_LACP_FAST},
198 {       NULL,           -1},
199 };
200
201 const struct bond_parm_tbl bond_mode_tbl[] = {
202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
204 {       "balance-xor",          BOND_MODE_XOR},
205 {       "broadcast",            BOND_MODE_BROADCAST},
206 {       "802.3ad",              BOND_MODE_8023AD},
207 {       "balance-tlb",          BOND_MODE_TLB},
208 {       "balance-alb",          BOND_MODE_ALB},
209 {       NULL,                   -1},
210 };
211
212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
216 {       NULL,                   -1},
217 };
218
219 const struct bond_parm_tbl arp_validate_tbl[] = {
220 {       "none",                 BOND_ARP_VALIDATE_NONE},
221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
223 {       "all",                  BOND_ARP_VALIDATE_ALL},
224 {       NULL,                   -1},
225 };
226
227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
228 {       "none",                 BOND_FOM_NONE},
229 {       "active",               BOND_FOM_ACTIVE},
230 {       "follow",               BOND_FOM_FOLLOW},
231 {       NULL,                   -1},
232 };
233
234 const struct bond_parm_tbl pri_reselect_tbl[] = {
235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
236 {       "better",               BOND_PRI_RESELECT_BETTER},
237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
238 {       NULL,                   -1},
239 };
240
241 struct bond_parm_tbl ad_select_tbl[] = {
242 {       "stable",       BOND_AD_STABLE},
243 {       "bandwidth",    BOND_AD_BANDWIDTH},
244 {       "count",        BOND_AD_COUNT},
245 {       NULL,           -1},
246 };
247
248 /*-------------------------- Forward declarations ---------------------------*/
249
250 static int bond_init(struct net_device *bond_dev);
251 static void bond_uninit(struct net_device *bond_dev);
252
253 /*---------------------------- General routines -----------------------------*/
254
255 const char *bond_mode_name(int mode)
256 {
257         static const char *names[] = {
258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
260                 [BOND_MODE_XOR] = "load balancing (xor)",
261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
263                 [BOND_MODE_TLB] = "transmit load balancing",
264                 [BOND_MODE_ALB] = "adaptive load balancing",
265         };
266
267         if (mode < 0 || mode > BOND_MODE_ALB)
268                 return "unknown";
269
270         return names[mode];
271 }
272
273 /*---------------------------------- VLAN -----------------------------------*/
274
275 /**
276  * bond_add_vlan - add a new vlan id on bond
277  * @bond: bond that got the notification
278  * @vlan_id: the vlan id to add
279  *
280  * Returns -ENOMEM if allocation failed.
281  */
282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
283 {
284         struct vlan_entry *vlan;
285
286         pr_debug("bond: %s, vlan id %d\n",
287                  (bond ? bond->dev->name : "None"), vlan_id);
288
289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
290         if (!vlan)
291                 return -ENOMEM;
292
293         INIT_LIST_HEAD(&vlan->vlan_list);
294         vlan->vlan_id = vlan_id;
295
296         write_lock_bh(&bond->lock);
297
298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
299
300         write_unlock_bh(&bond->lock);
301
302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
303
304         return 0;
305 }
306
307 /**
308  * bond_del_vlan - delete a vlan id from bond
309  * @bond: bond that got the notification
310  * @vlan_id: the vlan id to delete
311  *
312  * returns -ENODEV if @vlan_id was not found in @bond.
313  */
314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
315 {
316         struct vlan_entry *vlan;
317         int res = -ENODEV;
318
319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
320
321         block_netpoll_tx();
322         write_lock_bh(&bond->lock);
323
324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
325                 if (vlan->vlan_id == vlan_id) {
326                         list_del(&vlan->vlan_list);
327
328                         if (bond_is_lb(bond))
329                                 bond_alb_clear_vlan(bond, vlan_id);
330
331                         pr_debug("removed VLAN ID %d from bond %s\n",
332                                  vlan_id, bond->dev->name);
333
334                         kfree(vlan);
335
336                         res = 0;
337                         goto out;
338                 }
339         }
340
341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
342                  vlan_id, bond->dev->name);
343
344 out:
345         write_unlock_bh(&bond->lock);
346         unblock_netpoll_tx();
347         return res;
348 }
349
350 /**
351  * bond_next_vlan - safely skip to the next item in the vlans list.
352  * @bond: the bond we're working on
353  * @curr: item we're advancing from
354  *
355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356  * or @curr->next otherwise (even if it is @curr itself again).
357  *
358  * Caller must hold bond->lock
359  */
360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
361 {
362         struct vlan_entry *next, *last;
363
364         if (list_empty(&bond->vlan_list))
365                 return NULL;
366
367         if (!curr) {
368                 next = list_entry(bond->vlan_list.next,
369                                   struct vlan_entry, vlan_list);
370         } else {
371                 last = list_entry(bond->vlan_list.prev,
372                                   struct vlan_entry, vlan_list);
373                 if (last == curr) {
374                         next = list_entry(bond->vlan_list.next,
375                                           struct vlan_entry, vlan_list);
376                 } else {
377                         next = list_entry(curr->vlan_list.next,
378                                           struct vlan_entry, vlan_list);
379                 }
380         }
381
382         return next;
383 }
384
385 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
386
387 /**
388  * bond_dev_queue_xmit - Prepare skb for xmit.
389  *
390  * @bond: bond device that got this skb for tx.
391  * @skb: hw accel VLAN tagged skb to transmit
392  * @slave_dev: slave that is supposed to xmit this skbuff
393  */
394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
395                         struct net_device *slave_dev)
396 {
397         skb->dev = slave_dev;
398         skb->priority = 1;
399
400         skb->queue_mapping = bond_queue_mapping(skb);
401
402         if (unlikely(netpoll_tx_running(slave_dev)))
403                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
404         else
405                 dev_queue_xmit(skb);
406
407         return 0;
408 }
409
410 /*
411  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
412  * We don't protect the slave list iteration with a lock because:
413  * a. This operation is performed in IOCTL context,
414  * b. The operation is protected by the RTNL semaphore in the 8021q code,
415  * c. Holding a lock with BH disabled while directly calling a base driver
416  *    entry point is generally a BAD idea.
417  *
418  * The design of synchronization/protection for this operation in the 8021q
419  * module is good for one or more VLAN devices over a single physical device
420  * and cannot be extended for a teaming solution like bonding, so there is a
421  * potential race condition here where a net device from the vlan group might
422  * be referenced (either by a base driver or the 8021q code) while it is being
423  * removed from the system. However, it turns out we're not making matters
424  * worse, and if it works for regular VLAN usage it will work here too.
425 */
426
427 /**
428  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
429  * @bond_dev: bonding net device that got called
430  * @vid: vlan id being added
431  */
432 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
433 {
434         struct bonding *bond = netdev_priv(bond_dev);
435         struct slave *slave;
436         int i, res;
437
438         bond_for_each_slave(bond, slave, i) {
439                 struct net_device *slave_dev = slave->dev;
440                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441
442                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
443                     slave_ops->ndo_vlan_rx_add_vid) {
444                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
445                 }
446         }
447
448         res = bond_add_vlan(bond, vid);
449         if (res) {
450                 pr_err("%s: Error: Failed to add vlan id %d\n",
451                        bond_dev->name, vid);
452         }
453 }
454
455 /**
456  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
457  * @bond_dev: bonding net device that got called
458  * @vid: vlan id being removed
459  */
460 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
461 {
462         struct bonding *bond = netdev_priv(bond_dev);
463         struct slave *slave;
464         int i, res;
465
466         bond_for_each_slave(bond, slave, i) {
467                 struct net_device *slave_dev = slave->dev;
468                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
469
470                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
471                     slave_ops->ndo_vlan_rx_kill_vid) {
472                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
473                 }
474         }
475
476         res = bond_del_vlan(bond, vid);
477         if (res) {
478                 pr_err("%s: Error: Failed to remove vlan id %d\n",
479                        bond_dev->name, vid);
480         }
481 }
482
483 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
484 {
485         struct vlan_entry *vlan;
486         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
487
488         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
489             !(slave_ops->ndo_vlan_rx_add_vid))
490                 return;
491
492         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
493                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
494 }
495
496 static void bond_del_vlans_from_slave(struct bonding *bond,
497                                       struct net_device *slave_dev)
498 {
499         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
500         struct vlan_entry *vlan;
501
502         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
503             !(slave_ops->ndo_vlan_rx_kill_vid))
504                 return;
505
506         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
507                 if (!vlan->vlan_id)
508                         continue;
509                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
510         }
511 }
512
513 /*------------------------------- Link status -------------------------------*/
514
515 /*
516  * Set the carrier state for the master according to the state of its
517  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
518  * do special 802.3ad magic.
519  *
520  * Returns zero if carrier state does not change, nonzero if it does.
521  */
522 static int bond_set_carrier(struct bonding *bond)
523 {
524         struct slave *slave;
525         int i;
526
527         if (bond->slave_cnt == 0)
528                 goto down;
529
530         if (bond->params.mode == BOND_MODE_8023AD)
531                 return bond_3ad_set_carrier(bond);
532
533         bond_for_each_slave(bond, slave, i) {
534                 if (slave->link == BOND_LINK_UP) {
535                         if (!netif_carrier_ok(bond->dev)) {
536                                 netif_carrier_on(bond->dev);
537                                 return 1;
538                         }
539                         return 0;
540                 }
541         }
542
543 down:
544         if (netif_carrier_ok(bond->dev)) {
545                 netif_carrier_off(bond->dev);
546                 return 1;
547         }
548         return 0;
549 }
550
551 /*
552  * Get link speed and duplex from the slave's base driver
553  * using ethtool. If for some reason the call fails or the
554  * values are invalid, fake speed and duplex to 100/Full
555  * and return error.
556  */
557 static int bond_update_speed_duplex(struct slave *slave)
558 {
559         struct net_device *slave_dev = slave->dev;
560         struct ethtool_cmd ecmd;
561         u32 slave_speed;
562         int res;
563
564         /* Fake speed and duplex */
565         slave->speed = SPEED_100;
566         slave->duplex = DUPLEX_FULL;
567
568         res = __ethtool_get_settings(slave_dev, &ecmd);
569         if (res < 0)
570                 return -1;
571
572         slave_speed = ethtool_cmd_speed(&ecmd);
573         if (slave_speed == 0 || slave_speed == ((__u32) -1))
574                 return -1;
575
576         switch (ecmd.duplex) {
577         case DUPLEX_FULL:
578         case DUPLEX_HALF:
579                 break;
580         default:
581                 return -1;
582         }
583
584         slave->speed = slave_speed;
585         slave->duplex = ecmd.duplex;
586
587         return 0;
588 }
589
590 /*
591  * if <dev> supports MII link status reporting, check its link status.
592  *
593  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
594  * depending upon the setting of the use_carrier parameter.
595  *
596  * Return either BMSR_LSTATUS, meaning that the link is up (or we
597  * can't tell and just pretend it is), or 0, meaning that the link is
598  * down.
599  *
600  * If reporting is non-zero, instead of faking link up, return -1 if
601  * both ETHTOOL and MII ioctls fail (meaning the device does not
602  * support them).  If use_carrier is set, return whatever it says.
603  * It'd be nice if there was a good way to tell if a driver supports
604  * netif_carrier, but there really isn't.
605  */
606 static int bond_check_dev_link(struct bonding *bond,
607                                struct net_device *slave_dev, int reporting)
608 {
609         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
610         int (*ioctl)(struct net_device *, struct ifreq *, int);
611         struct ifreq ifr;
612         struct mii_ioctl_data *mii;
613
614         if (!reporting && !netif_running(slave_dev))
615                 return 0;
616
617         if (bond->params.use_carrier)
618                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
619
620         /* Try to get link status using Ethtool first. */
621         if (slave_dev->ethtool_ops) {
622                 if (slave_dev->ethtool_ops->get_link) {
623                         u32 link;
624
625                         link = slave_dev->ethtool_ops->get_link(slave_dev);
626
627                         return link ? BMSR_LSTATUS : 0;
628                 }
629         }
630
631         /* Ethtool can't be used, fallback to MII ioctls. */
632         ioctl = slave_ops->ndo_do_ioctl;
633         if (ioctl) {
634                 /* TODO: set pointer to correct ioctl on a per team member */
635                 /*       bases to make this more efficient. that is, once  */
636                 /*       we determine the correct ioctl, we will always    */
637                 /*       call it and not the others for that team          */
638                 /*       member.                                           */
639
640                 /*
641                  * We cannot assume that SIOCGMIIPHY will also read a
642                  * register; not all network drivers (e.g., e100)
643                  * support that.
644                  */
645
646                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
647                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
648                 mii = if_mii(&ifr);
649                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
650                         mii->reg_num = MII_BMSR;
651                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
652                                 return mii->val_out & BMSR_LSTATUS;
653                 }
654         }
655
656         /*
657          * If reporting, report that either there's no dev->do_ioctl,
658          * or both SIOCGMIIREG and get_link failed (meaning that we
659          * cannot report link status).  If not reporting, pretend
660          * we're ok.
661          */
662         return reporting ? -1 : BMSR_LSTATUS;
663 }
664
665 /*----------------------------- Multicast list ------------------------------*/
666
667 /*
668  * Push the promiscuity flag down to appropriate slaves
669  */
670 static int bond_set_promiscuity(struct bonding *bond, int inc)
671 {
672         int err = 0;
673         if (USES_PRIMARY(bond->params.mode)) {
674                 /* write lock already acquired */
675                 if (bond->curr_active_slave) {
676                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
677                                                   inc);
678                 }
679         } else {
680                 struct slave *slave;
681                 int i;
682                 bond_for_each_slave(bond, slave, i) {
683                         err = dev_set_promiscuity(slave->dev, inc);
684                         if (err)
685                                 return err;
686                 }
687         }
688         return err;
689 }
690
691 /*
692  * Push the allmulti flag down to all slaves
693  */
694 static int bond_set_allmulti(struct bonding *bond, int inc)
695 {
696         int err = 0;
697         if (USES_PRIMARY(bond->params.mode)) {
698                 /* write lock already acquired */
699                 if (bond->curr_active_slave) {
700                         err = dev_set_allmulti(bond->curr_active_slave->dev,
701                                                inc);
702                 }
703         } else {
704                 struct slave *slave;
705                 int i;
706                 bond_for_each_slave(bond, slave, i) {
707                         err = dev_set_allmulti(slave->dev, inc);
708                         if (err)
709                                 return err;
710                 }
711         }
712         return err;
713 }
714
715 /*
716  * Add a Multicast address to slaves
717  * according to mode
718  */
719 static void bond_mc_add(struct bonding *bond, void *addr)
720 {
721         if (USES_PRIMARY(bond->params.mode)) {
722                 /* write lock already acquired */
723                 if (bond->curr_active_slave)
724                         dev_mc_add(bond->curr_active_slave->dev, addr);
725         } else {
726                 struct slave *slave;
727                 int i;
728
729                 bond_for_each_slave(bond, slave, i)
730                         dev_mc_add(slave->dev, addr);
731         }
732 }
733
734 /*
735  * Remove a multicast address from slave
736  * according to mode
737  */
738 static void bond_mc_del(struct bonding *bond, void *addr)
739 {
740         if (USES_PRIMARY(bond->params.mode)) {
741                 /* write lock already acquired */
742                 if (bond->curr_active_slave)
743                         dev_mc_del(bond->curr_active_slave->dev, addr);
744         } else {
745                 struct slave *slave;
746                 int i;
747                 bond_for_each_slave(bond, slave, i) {
748                         dev_mc_del(slave->dev, addr);
749                 }
750         }
751 }
752
753
754 static void __bond_resend_igmp_join_requests(struct net_device *dev)
755 {
756         struct in_device *in_dev;
757
758         rcu_read_lock();
759         in_dev = __in_dev_get_rcu(dev);
760         if (in_dev)
761                 ip_mc_rejoin_groups(in_dev);
762         rcu_read_unlock();
763 }
764
765 /*
766  * Retrieve the list of registered multicast addresses for the bonding
767  * device and retransmit an IGMP JOIN request to the current active
768  * slave.
769  */
770 static void bond_resend_igmp_join_requests(struct bonding *bond)
771 {
772         struct net_device *vlan_dev;
773         struct vlan_entry *vlan;
774
775         read_lock(&bond->lock);
776
777         if (bond->kill_timers)
778                 goto out;
779
780         /* rejoin all groups on bond device */
781         __bond_resend_igmp_join_requests(bond->dev);
782
783         /* rejoin all groups on vlan devices */
784         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
785                 rcu_read_lock();
786                 vlan_dev = __vlan_find_dev_deep(bond->dev,
787                                                 vlan->vlan_id);
788                 rcu_read_unlock();
789                 if (vlan_dev)
790                         __bond_resend_igmp_join_requests(vlan_dev);
791         }
792
793         if ((--bond->igmp_retrans > 0) && !bond->kill_timers)
794                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
795 out:
796         read_unlock(&bond->lock);
797 }
798
799 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
800 {
801         struct bonding *bond = container_of(work, struct bonding,
802                                             mcast_work.work);
803         bond_resend_igmp_join_requests(bond);
804 }
805
806 /*
807  * flush all members of flush->mc_list from device dev->mc_list
808  */
809 static void bond_mc_list_flush(struct net_device *bond_dev,
810                                struct net_device *slave_dev)
811 {
812         struct bonding *bond = netdev_priv(bond_dev);
813         struct netdev_hw_addr *ha;
814
815         netdev_for_each_mc_addr(ha, bond_dev)
816                 dev_mc_del(slave_dev, ha->addr);
817
818         if (bond->params.mode == BOND_MODE_8023AD) {
819                 /* del lacpdu mc addr from mc list */
820                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
821
822                 dev_mc_del(slave_dev, lacpdu_multicast);
823         }
824 }
825
826 /*--------------------------- Active slave change ---------------------------*/
827
828 /*
829  * Update the mc list and multicast-related flags for the new and
830  * old active slaves (if any) according to the multicast mode, and
831  * promiscuous flags unconditionally.
832  */
833 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
834                          struct slave *old_active)
835 {
836         struct netdev_hw_addr *ha;
837
838         if (!USES_PRIMARY(bond->params.mode))
839                 /* nothing to do -  mc list is already up-to-date on
840                  * all slaves
841                  */
842                 return;
843
844         if (old_active) {
845                 if (bond->dev->flags & IFF_PROMISC)
846                         dev_set_promiscuity(old_active->dev, -1);
847
848                 if (bond->dev->flags & IFF_ALLMULTI)
849                         dev_set_allmulti(old_active->dev, -1);
850
851                 netdev_for_each_mc_addr(ha, bond->dev)
852                         dev_mc_del(old_active->dev, ha->addr);
853         }
854
855         if (new_active) {
856                 /* FIXME: Signal errors upstream. */
857                 if (bond->dev->flags & IFF_PROMISC)
858                         dev_set_promiscuity(new_active->dev, 1);
859
860                 if (bond->dev->flags & IFF_ALLMULTI)
861                         dev_set_allmulti(new_active->dev, 1);
862
863                 netdev_for_each_mc_addr(ha, bond->dev)
864                         dev_mc_add(new_active->dev, ha->addr);
865         }
866 }
867
868 /*
869  * bond_do_fail_over_mac
870  *
871  * Perform special MAC address swapping for fail_over_mac settings
872  *
873  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
874  */
875 static void bond_do_fail_over_mac(struct bonding *bond,
876                                   struct slave *new_active,
877                                   struct slave *old_active)
878         __releases(&bond->curr_slave_lock)
879         __releases(&bond->lock)
880         __acquires(&bond->lock)
881         __acquires(&bond->curr_slave_lock)
882 {
883         u8 tmp_mac[ETH_ALEN];
884         struct sockaddr saddr;
885         int rv;
886
887         switch (bond->params.fail_over_mac) {
888         case BOND_FOM_ACTIVE:
889                 if (new_active)
890                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
891                                new_active->dev->addr_len);
892                 break;
893         case BOND_FOM_FOLLOW:
894                 /*
895                  * if new_active && old_active, swap them
896                  * if just old_active, do nothing (going to no active slave)
897                  * if just new_active, set new_active to bond's MAC
898                  */
899                 if (!new_active)
900                         return;
901
902                 write_unlock_bh(&bond->curr_slave_lock);
903                 read_unlock(&bond->lock);
904
905                 if (old_active) {
906                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
907                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
908                                ETH_ALEN);
909                         saddr.sa_family = new_active->dev->type;
910                 } else {
911                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
912                         saddr.sa_family = bond->dev->type;
913                 }
914
915                 rv = dev_set_mac_address(new_active->dev, &saddr);
916                 if (rv) {
917                         pr_err("%s: Error %d setting MAC of slave %s\n",
918                                bond->dev->name, -rv, new_active->dev->name);
919                         goto out;
920                 }
921
922                 if (!old_active)
923                         goto out;
924
925                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
926                 saddr.sa_family = old_active->dev->type;
927
928                 rv = dev_set_mac_address(old_active->dev, &saddr);
929                 if (rv)
930                         pr_err("%s: Error %d setting MAC of slave %s\n",
931                                bond->dev->name, -rv, new_active->dev->name);
932 out:
933                 read_lock(&bond->lock);
934                 write_lock_bh(&bond->curr_slave_lock);
935                 break;
936         default:
937                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
938                        bond->dev->name, bond->params.fail_over_mac);
939                 break;
940         }
941
942 }
943
944 static bool bond_should_change_active(struct bonding *bond)
945 {
946         struct slave *prim = bond->primary_slave;
947         struct slave *curr = bond->curr_active_slave;
948
949         if (!prim || !curr || curr->link != BOND_LINK_UP)
950                 return true;
951         if (bond->force_primary) {
952                 bond->force_primary = false;
953                 return true;
954         }
955         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
956             (prim->speed < curr->speed ||
957              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
958                 return false;
959         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
960                 return false;
961         return true;
962 }
963
964 /**
965  * find_best_interface - select the best available slave to be the active one
966  * @bond: our bonding struct
967  *
968  * Warning: Caller must hold curr_slave_lock for writing.
969  */
970 static struct slave *bond_find_best_slave(struct bonding *bond)
971 {
972         struct slave *new_active, *old_active;
973         struct slave *bestslave = NULL;
974         int mintime = bond->params.updelay;
975         int i;
976
977         new_active = bond->curr_active_slave;
978
979         if (!new_active) { /* there were no active slaves left */
980                 if (bond->slave_cnt > 0)   /* found one slave */
981                         new_active = bond->first_slave;
982                 else
983                         return NULL; /* still no slave, return NULL */
984         }
985
986         if ((bond->primary_slave) &&
987             bond->primary_slave->link == BOND_LINK_UP &&
988             bond_should_change_active(bond)) {
989                 new_active = bond->primary_slave;
990         }
991
992         /* remember where to stop iterating over the slaves */
993         old_active = new_active;
994
995         bond_for_each_slave_from(bond, new_active, i, old_active) {
996                 if (new_active->link == BOND_LINK_UP) {
997                         return new_active;
998                 } else if (new_active->link == BOND_LINK_BACK &&
999                            IS_UP(new_active->dev)) {
1000                         /* link up, but waiting for stabilization */
1001                         if (new_active->delay < mintime) {
1002                                 mintime = new_active->delay;
1003                                 bestslave = new_active;
1004                         }
1005                 }
1006         }
1007
1008         return bestslave;
1009 }
1010
1011 static bool bond_should_notify_peers(struct bonding *bond)
1012 {
1013         struct slave *slave = bond->curr_active_slave;
1014
1015         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1016                  bond->dev->name, slave ? slave->dev->name : "NULL");
1017
1018         if (!slave || !bond->send_peer_notif ||
1019             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1020                 return false;
1021
1022         bond->send_peer_notif--;
1023         return true;
1024 }
1025
1026 /**
1027  * change_active_interface - change the active slave into the specified one
1028  * @bond: our bonding struct
1029  * @new: the new slave to make the active one
1030  *
1031  * Set the new slave to the bond's settings and unset them on the old
1032  * curr_active_slave.
1033  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1034  *
1035  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1036  * because it is apparently the best available slave we have, even though its
1037  * updelay hasn't timed out yet.
1038  *
1039  * If new_active is not NULL, caller must hold bond->lock for read and
1040  * curr_slave_lock for write_bh.
1041  */
1042 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1043 {
1044         struct slave *old_active = bond->curr_active_slave;
1045
1046         if (old_active == new_active)
1047                 return;
1048
1049         if (new_active) {
1050                 new_active->jiffies = jiffies;
1051
1052                 if (new_active->link == BOND_LINK_BACK) {
1053                         if (USES_PRIMARY(bond->params.mode)) {
1054                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1055                                         bond->dev->name, new_active->dev->name,
1056                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1057                         }
1058
1059                         new_active->delay = 0;
1060                         new_active->link = BOND_LINK_UP;
1061
1062                         if (bond->params.mode == BOND_MODE_8023AD)
1063                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1064
1065                         if (bond_is_lb(bond))
1066                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1067                 } else {
1068                         if (USES_PRIMARY(bond->params.mode)) {
1069                                 pr_info("%s: making interface %s the new active one.\n",
1070                                         bond->dev->name, new_active->dev->name);
1071                         }
1072                 }
1073         }
1074
1075         if (USES_PRIMARY(bond->params.mode))
1076                 bond_mc_swap(bond, new_active, old_active);
1077
1078         if (bond_is_lb(bond)) {
1079                 bond_alb_handle_active_change(bond, new_active);
1080                 if (old_active)
1081                         bond_set_slave_inactive_flags(old_active);
1082                 if (new_active)
1083                         bond_set_slave_active_flags(new_active);
1084         } else {
1085                 bond->curr_active_slave = new_active;
1086         }
1087
1088         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1089                 if (old_active)
1090                         bond_set_slave_inactive_flags(old_active);
1091
1092                 if (new_active) {
1093                         bool should_notify_peers = false;
1094
1095                         bond_set_slave_active_flags(new_active);
1096
1097                         if (bond->params.fail_over_mac)
1098                                 bond_do_fail_over_mac(bond, new_active,
1099                                                       old_active);
1100
1101                         if (netif_running(bond->dev)) {
1102                                 bond->send_peer_notif =
1103                                         bond->params.num_peer_notif;
1104                                 should_notify_peers =
1105                                         bond_should_notify_peers(bond);
1106                         }
1107
1108                         write_unlock_bh(&bond->curr_slave_lock);
1109                         read_unlock(&bond->lock);
1110
1111                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1112                         if (should_notify_peers)
1113                                 netdev_bonding_change(bond->dev,
1114                                                       NETDEV_NOTIFY_PEERS);
1115
1116                         read_lock(&bond->lock);
1117                         write_lock_bh(&bond->curr_slave_lock);
1118                 }
1119         }
1120
1121         /* resend IGMP joins since active slave has changed or
1122          * all were sent on curr_active_slave.
1123          * resend only if bond is brought up with the affected
1124          * bonding modes and the retransmission is enabled */
1125         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1126             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1127              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1128                 bond->igmp_retrans = bond->params.resend_igmp;
1129                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1130         }
1131 }
1132
1133 /**
1134  * bond_select_active_slave - select a new active slave, if needed
1135  * @bond: our bonding struct
1136  *
1137  * This functions should be called when one of the following occurs:
1138  * - The old curr_active_slave has been released or lost its link.
1139  * - The primary_slave has got its link back.
1140  * - A slave has got its link back and there's no old curr_active_slave.
1141  *
1142  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1143  */
1144 void bond_select_active_slave(struct bonding *bond)
1145 {
1146         struct slave *best_slave;
1147         int rv;
1148
1149         best_slave = bond_find_best_slave(bond);
1150         if (best_slave != bond->curr_active_slave) {
1151                 bond_change_active_slave(bond, best_slave);
1152                 rv = bond_set_carrier(bond);
1153                 if (!rv)
1154                         return;
1155
1156                 if (netif_carrier_ok(bond->dev)) {
1157                         pr_info("%s: first active interface up!\n",
1158                                 bond->dev->name);
1159                 } else {
1160                         pr_info("%s: now running without any active interface !\n",
1161                                 bond->dev->name);
1162                 }
1163         }
1164 }
1165
1166 /*--------------------------- slave list handling ---------------------------*/
1167
1168 /*
1169  * This function attaches the slave to the end of list.
1170  *
1171  * bond->lock held for writing by caller.
1172  */
1173 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1174 {
1175         if (bond->first_slave == NULL) { /* attaching the first slave */
1176                 new_slave->next = new_slave;
1177                 new_slave->prev = new_slave;
1178                 bond->first_slave = new_slave;
1179         } else {
1180                 new_slave->next = bond->first_slave;
1181                 new_slave->prev = bond->first_slave->prev;
1182                 new_slave->next->prev = new_slave;
1183                 new_slave->prev->next = new_slave;
1184         }
1185
1186         bond->slave_cnt++;
1187 }
1188
1189 /*
1190  * This function detaches the slave from the list.
1191  * WARNING: no check is made to verify if the slave effectively
1192  * belongs to <bond>.
1193  * Nothing is freed on return, structures are just unchained.
1194  * If any slave pointer in bond was pointing to <slave>,
1195  * it should be changed by the calling function.
1196  *
1197  * bond->lock held for writing by caller.
1198  */
1199 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1200 {
1201         if (slave->next)
1202                 slave->next->prev = slave->prev;
1203
1204         if (slave->prev)
1205                 slave->prev->next = slave->next;
1206
1207         if (bond->first_slave == slave) { /* slave is the first slave */
1208                 if (bond->slave_cnt > 1) { /* there are more slave */
1209                         bond->first_slave = slave->next;
1210                 } else {
1211                         bond->first_slave = NULL; /* slave was the last one */
1212                 }
1213         }
1214
1215         slave->next = NULL;
1216         slave->prev = NULL;
1217         bond->slave_cnt--;
1218 }
1219
1220 #ifdef CONFIG_NET_POLL_CONTROLLER
1221 static inline int slave_enable_netpoll(struct slave *slave)
1222 {
1223         struct netpoll *np;
1224         int err = 0;
1225
1226         np = kzalloc(sizeof(*np), GFP_KERNEL);
1227         err = -ENOMEM;
1228         if (!np)
1229                 goto out;
1230
1231         np->dev = slave->dev;
1232         strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1233         err = __netpoll_setup(np);
1234         if (err) {
1235                 kfree(np);
1236                 goto out;
1237         }
1238         slave->np = np;
1239 out:
1240         return err;
1241 }
1242 static inline void slave_disable_netpoll(struct slave *slave)
1243 {
1244         struct netpoll *np = slave->np;
1245
1246         if (!np)
1247                 return;
1248
1249         slave->np = NULL;
1250         synchronize_rcu_bh();
1251         __netpoll_cleanup(np);
1252         kfree(np);
1253 }
1254 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1255 {
1256         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1257                 return false;
1258         if (!slave_dev->netdev_ops->ndo_poll_controller)
1259                 return false;
1260         return true;
1261 }
1262
1263 static void bond_poll_controller(struct net_device *bond_dev)
1264 {
1265 }
1266
1267 static void __bond_netpoll_cleanup(struct bonding *bond)
1268 {
1269         struct slave *slave;
1270         int i;
1271
1272         bond_for_each_slave(bond, slave, i)
1273                 if (IS_UP(slave->dev))
1274                         slave_disable_netpoll(slave);
1275 }
1276 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1277 {
1278         struct bonding *bond = netdev_priv(bond_dev);
1279
1280         read_lock(&bond->lock);
1281         __bond_netpoll_cleanup(bond);
1282         read_unlock(&bond->lock);
1283 }
1284
1285 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1286 {
1287         struct bonding *bond = netdev_priv(dev);
1288         struct slave *slave;
1289         int i, err = 0;
1290
1291         read_lock(&bond->lock);
1292         bond_for_each_slave(bond, slave, i) {
1293                 err = slave_enable_netpoll(slave);
1294                 if (err) {
1295                         __bond_netpoll_cleanup(bond);
1296                         break;
1297                 }
1298         }
1299         read_unlock(&bond->lock);
1300         return err;
1301 }
1302
1303 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1304 {
1305         return bond->dev->npinfo;
1306 }
1307
1308 #else
1309 static inline int slave_enable_netpoll(struct slave *slave)
1310 {
1311         return 0;
1312 }
1313 static inline void slave_disable_netpoll(struct slave *slave)
1314 {
1315 }
1316 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1317 {
1318 }
1319 #endif
1320
1321 /*---------------------------------- IOCTL ----------------------------------*/
1322
1323 static int bond_sethwaddr(struct net_device *bond_dev,
1324                           struct net_device *slave_dev)
1325 {
1326         pr_debug("bond_dev=%p\n", bond_dev);
1327         pr_debug("slave_dev=%p\n", slave_dev);
1328         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1329         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1330         return 0;
1331 }
1332
1333 static u32 bond_fix_features(struct net_device *dev, u32 features)
1334 {
1335         struct slave *slave;
1336         struct bonding *bond = netdev_priv(dev);
1337         u32 mask;
1338         int i;
1339
1340         read_lock(&bond->lock);
1341
1342         if (!bond->first_slave) {
1343                 /* Disable adding VLANs to empty bond. But why? --mq */
1344                 features |= NETIF_F_VLAN_CHALLENGED;
1345                 goto out;
1346         }
1347
1348         mask = features;
1349         features &= ~NETIF_F_ONE_FOR_ALL;
1350         features |= NETIF_F_ALL_FOR_ALL;
1351
1352         bond_for_each_slave(bond, slave, i) {
1353                 features = netdev_increment_features(features,
1354                                                      slave->dev->features,
1355                                                      mask);
1356         }
1357
1358 out:
1359         read_unlock(&bond->lock);
1360         return features;
1361 }
1362
1363 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1364                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1365                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1366
1367 static void bond_compute_features(struct bonding *bond)
1368 {
1369         struct slave *slave;
1370         struct net_device *bond_dev = bond->dev;
1371         u32 vlan_features = BOND_VLAN_FEATURES;
1372         unsigned short max_hard_header_len = ETH_HLEN;
1373         int i;
1374
1375         read_lock(&bond->lock);
1376
1377         if (!bond->first_slave)
1378                 goto done;
1379
1380         bond_for_each_slave(bond, slave, i) {
1381                 vlan_features = netdev_increment_features(vlan_features,
1382                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1383
1384                 if (slave->dev->hard_header_len > max_hard_header_len)
1385                         max_hard_header_len = slave->dev->hard_header_len;
1386         }
1387
1388 done:
1389         bond_dev->vlan_features = vlan_features;
1390         bond_dev->hard_header_len = max_hard_header_len;
1391
1392         read_unlock(&bond->lock);
1393
1394         netdev_change_features(bond_dev);
1395 }
1396
1397 static void bond_setup_by_slave(struct net_device *bond_dev,
1398                                 struct net_device *slave_dev)
1399 {
1400         struct bonding *bond = netdev_priv(bond_dev);
1401
1402         bond_dev->header_ops        = slave_dev->header_ops;
1403
1404         bond_dev->type              = slave_dev->type;
1405         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1406         bond_dev->addr_len          = slave_dev->addr_len;
1407
1408         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1409                 slave_dev->addr_len);
1410         bond->setup_by_slave = 1;
1411 }
1412
1413 /* On bonding slaves other than the currently active slave, suppress
1414  * duplicates except for alb non-mcast/bcast.
1415  */
1416 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1417                                             struct slave *slave,
1418                                             struct bonding *bond)
1419 {
1420         if (bond_is_slave_inactive(slave)) {
1421                 if (bond->params.mode == BOND_MODE_ALB &&
1422                     skb->pkt_type != PACKET_BROADCAST &&
1423                     skb->pkt_type != PACKET_MULTICAST)
1424                         return false;
1425                 return true;
1426         }
1427         return false;
1428 }
1429
1430 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1431 {
1432         struct sk_buff *skb = *pskb;
1433         struct slave *slave;
1434         struct bonding *bond;
1435         void (*recv_probe)(struct sk_buff *, struct bonding *,
1436                                 struct slave *);
1437
1438         skb = skb_share_check(skb, GFP_ATOMIC);
1439         if (unlikely(!skb))
1440                 return RX_HANDLER_CONSUMED;
1441
1442         *pskb = skb;
1443
1444         slave = bond_slave_get_rcu(skb->dev);
1445         bond = slave->bond;
1446
1447         if (bond->params.arp_interval)
1448                 slave->dev->last_rx = jiffies;
1449
1450         recv_probe = ACCESS_ONCE(bond->recv_probe);
1451         if (recv_probe) {
1452                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1453
1454                 if (likely(nskb)) {
1455                         recv_probe(nskb, bond, slave);
1456                         dev_kfree_skb(nskb);
1457                 }
1458         }
1459
1460         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1461                 return RX_HANDLER_EXACT;
1462         }
1463
1464         skb->dev = bond->dev;
1465
1466         if (bond->params.mode == BOND_MODE_ALB &&
1467             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1468             skb->pkt_type == PACKET_HOST) {
1469
1470                 if (unlikely(skb_cow_head(skb,
1471                                           skb->data - skb_mac_header(skb)))) {
1472                         kfree_skb(skb);
1473                         return RX_HANDLER_CONSUMED;
1474                 }
1475                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1476         }
1477
1478         return RX_HANDLER_ANOTHER;
1479 }
1480
1481 /* enslave device <slave> to bond device <master> */
1482 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1483 {
1484         struct bonding *bond = netdev_priv(bond_dev);
1485         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1486         struct slave *new_slave = NULL;
1487         struct netdev_hw_addr *ha;
1488         struct sockaddr addr;
1489         int link_reporting;
1490         int res = 0;
1491
1492         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1493                 slave_ops->ndo_do_ioctl == NULL) {
1494                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1495                            bond_dev->name, slave_dev->name);
1496         }
1497
1498         /* already enslaved */
1499         if (slave_dev->flags & IFF_SLAVE) {
1500                 pr_debug("Error, Device was already enslaved\n");
1501                 return -EBUSY;
1502         }
1503
1504         /* vlan challenged mutual exclusion */
1505         /* no need to lock since we're protected by rtnl_lock */
1506         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1507                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1508                 if (bond_vlan_used(bond)) {
1509                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1510                                bond_dev->name, slave_dev->name, bond_dev->name);
1511                         return -EPERM;
1512                 } else {
1513                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1514                                    bond_dev->name, slave_dev->name,
1515                                    slave_dev->name, bond_dev->name);
1516                 }
1517         } else {
1518                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1519         }
1520
1521         /*
1522          * Old ifenslave binaries are no longer supported.  These can
1523          * be identified with moderate accuracy by the state of the slave:
1524          * the current ifenslave will set the interface down prior to
1525          * enslaving it; the old ifenslave will not.
1526          */
1527         if ((slave_dev->flags & IFF_UP)) {
1528                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1529                        slave_dev->name);
1530                 res = -EPERM;
1531                 goto err_undo_flags;
1532         }
1533
1534         /* set bonding device ether type by slave - bonding netdevices are
1535          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1536          * there is a need to override some of the type dependent attribs/funcs.
1537          *
1538          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1539          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1540          */
1541         if (bond->slave_cnt == 0) {
1542                 if (bond_dev->type != slave_dev->type) {
1543                         pr_debug("%s: change device type from %d to %d\n",
1544                                  bond_dev->name,
1545                                  bond_dev->type, slave_dev->type);
1546
1547                         res = netdev_bonding_change(bond_dev,
1548                                                     NETDEV_PRE_TYPE_CHANGE);
1549                         res = notifier_to_errno(res);
1550                         if (res) {
1551                                 pr_err("%s: refused to change device type\n",
1552                                        bond_dev->name);
1553                                 res = -EBUSY;
1554                                 goto err_undo_flags;
1555                         }
1556
1557                         /* Flush unicast and multicast addresses */
1558                         dev_uc_flush(bond_dev);
1559                         dev_mc_flush(bond_dev);
1560
1561                         if (slave_dev->type != ARPHRD_ETHER)
1562                                 bond_setup_by_slave(bond_dev, slave_dev);
1563                         else {
1564                                 ether_setup(bond_dev);
1565                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1566                         }
1567
1568                         netdev_bonding_change(bond_dev,
1569                                               NETDEV_POST_TYPE_CHANGE);
1570                 }
1571         } else if (bond_dev->type != slave_dev->type) {
1572                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1573                        slave_dev->name,
1574                        slave_dev->type, bond_dev->type);
1575                 res = -EINVAL;
1576                 goto err_undo_flags;
1577         }
1578
1579         if (slave_ops->ndo_set_mac_address == NULL) {
1580                 if (bond->slave_cnt == 0) {
1581                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1582                                    bond_dev->name);
1583                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1584                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1585                         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",
1586                                bond_dev->name);
1587                         res = -EOPNOTSUPP;
1588                         goto err_undo_flags;
1589                 }
1590         }
1591
1592         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1593
1594         /* If this is the first slave, then we need to set the master's hardware
1595          * address to be the same as the slave's. */
1596         if (is_zero_ether_addr(bond->dev->dev_addr))
1597                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1598                        slave_dev->addr_len);
1599
1600
1601         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1602         if (!new_slave) {
1603                 res = -ENOMEM;
1604                 goto err_undo_flags;
1605         }
1606
1607         /*
1608          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1609          * is set via sysfs or module option if desired.
1610          */
1611         new_slave->queue_id = 0;
1612
1613         /* Save slave's original mtu and then set it to match the bond */
1614         new_slave->original_mtu = slave_dev->mtu;
1615         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1616         if (res) {
1617                 pr_debug("Error %d calling dev_set_mtu\n", res);
1618                 goto err_free;
1619         }
1620
1621         /*
1622          * Save slave's original ("permanent") mac address for modes
1623          * that need it, and for restoring it upon release, and then
1624          * set it to the master's address
1625          */
1626         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1627
1628         if (!bond->params.fail_over_mac) {
1629                 /*
1630                  * Set slave to master's mac address.  The application already
1631                  * set the master's mac address to that of the first slave
1632                  */
1633                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1634                 addr.sa_family = slave_dev->type;
1635                 res = dev_set_mac_address(slave_dev, &addr);
1636                 if (res) {
1637                         pr_debug("Error %d calling set_mac_address\n", res);
1638                         goto err_restore_mtu;
1639                 }
1640         }
1641
1642         res = netdev_set_bond_master(slave_dev, bond_dev);
1643         if (res) {
1644                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1645                 goto err_restore_mac;
1646         }
1647
1648         /* open the slave since the application closed it */
1649         res = dev_open(slave_dev);
1650         if (res) {
1651                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1652                 goto err_unset_master;
1653         }
1654
1655         new_slave->bond = bond;
1656         new_slave->dev = slave_dev;
1657         slave_dev->priv_flags |= IFF_BONDING;
1658
1659         if (bond_is_lb(bond)) {
1660                 /* bond_alb_init_slave() must be called before all other stages since
1661                  * it might fail and we do not want to have to undo everything
1662                  */
1663                 res = bond_alb_init_slave(bond, new_slave);
1664                 if (res)
1665                         goto err_close;
1666         }
1667
1668         /* If the mode USES_PRIMARY, then the new slave gets the
1669          * master's promisc (and mc) settings only if it becomes the
1670          * curr_active_slave, and that is taken care of later when calling
1671          * bond_change_active()
1672          */
1673         if (!USES_PRIMARY(bond->params.mode)) {
1674                 /* set promiscuity level to new slave */
1675                 if (bond_dev->flags & IFF_PROMISC) {
1676                         res = dev_set_promiscuity(slave_dev, 1);
1677                         if (res)
1678                                 goto err_close;
1679                 }
1680
1681                 /* set allmulti level to new slave */
1682                 if (bond_dev->flags & IFF_ALLMULTI) {
1683                         res = dev_set_allmulti(slave_dev, 1);
1684                         if (res)
1685                                 goto err_close;
1686                 }
1687
1688                 netif_addr_lock_bh(bond_dev);
1689                 /* upload master's mc_list to new slave */
1690                 netdev_for_each_mc_addr(ha, bond_dev)
1691                         dev_mc_add(slave_dev, ha->addr);
1692                 netif_addr_unlock_bh(bond_dev);
1693         }
1694
1695         if (bond->params.mode == BOND_MODE_8023AD) {
1696                 /* add lacpdu mc addr to mc list */
1697                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1698
1699                 dev_mc_add(slave_dev, lacpdu_multicast);
1700         }
1701
1702         bond_add_vlans_on_slave(bond, slave_dev);
1703
1704         write_lock_bh(&bond->lock);
1705
1706         bond_attach_slave(bond, new_slave);
1707
1708         new_slave->delay = 0;
1709         new_slave->link_failure_count = 0;
1710
1711         write_unlock_bh(&bond->lock);
1712
1713         bond_compute_features(bond);
1714
1715         read_lock(&bond->lock);
1716
1717         new_slave->last_arp_rx = jiffies;
1718
1719         if (bond->params.miimon && !bond->params.use_carrier) {
1720                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1721
1722                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1723                         /*
1724                          * miimon is set but a bonded network driver
1725                          * does not support ETHTOOL/MII and
1726                          * arp_interval is not set.  Note: if
1727                          * use_carrier is enabled, we will never go
1728                          * here (because netif_carrier is always
1729                          * supported); thus, we don't need to change
1730                          * the messages for netif_carrier.
1731                          */
1732                         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",
1733                                bond_dev->name, slave_dev->name);
1734                 } else if (link_reporting == -1) {
1735                         /* unable get link status using mii/ethtool */
1736                         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",
1737                                    bond_dev->name, slave_dev->name);
1738                 }
1739         }
1740
1741         /* check for initial state */
1742         if (!bond->params.miimon ||
1743             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1744                 if (bond->params.updelay) {
1745                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1746                         new_slave->link  = BOND_LINK_BACK;
1747                         new_slave->delay = bond->params.updelay;
1748                 } else {
1749                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1750                         new_slave->link  = BOND_LINK_UP;
1751                 }
1752                 new_slave->jiffies = jiffies;
1753         } else {
1754                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1755                 new_slave->link  = BOND_LINK_DOWN;
1756         }
1757
1758         if (bond_update_speed_duplex(new_slave) &&
1759             (new_slave->link != BOND_LINK_DOWN)) {
1760                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1761                            bond_dev->name, new_slave->dev->name);
1762
1763                 if (bond->params.mode == BOND_MODE_8023AD) {
1764                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1765                                    bond_dev->name);
1766                 }
1767         }
1768
1769         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1770                 /* if there is a primary slave, remember it */
1771                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1772                         bond->primary_slave = new_slave;
1773                         bond->force_primary = true;
1774                 }
1775         }
1776
1777         write_lock_bh(&bond->curr_slave_lock);
1778
1779         switch (bond->params.mode) {
1780         case BOND_MODE_ACTIVEBACKUP:
1781                 bond_set_slave_inactive_flags(new_slave);
1782                 bond_select_active_slave(bond);
1783                 break;
1784         case BOND_MODE_8023AD:
1785                 /* in 802.3ad mode, the internal mechanism
1786                  * will activate the slaves in the selected
1787                  * aggregator
1788                  */
1789                 bond_set_slave_inactive_flags(new_slave);
1790                 /* if this is the first slave */
1791                 if (bond->slave_cnt == 1) {
1792                         SLAVE_AD_INFO(new_slave).id = 1;
1793                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1794                          * can be called only after the mac address of the bond is set
1795                          */
1796                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1797                 } else {
1798                         SLAVE_AD_INFO(new_slave).id =
1799                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1800                 }
1801
1802                 bond_3ad_bind_slave(new_slave);
1803                 break;
1804         case BOND_MODE_TLB:
1805         case BOND_MODE_ALB:
1806                 bond_set_active_slave(new_slave);
1807                 bond_set_slave_inactive_flags(new_slave);
1808                 bond_select_active_slave(bond);
1809                 break;
1810         default:
1811                 pr_debug("This slave is always active in trunk mode\n");
1812
1813                 /* always active in trunk mode */
1814                 bond_set_active_slave(new_slave);
1815
1816                 /* In trunking mode there is little meaning to curr_active_slave
1817                  * anyway (it holds no special properties of the bond device),
1818                  * so we can change it without calling change_active_interface()
1819                  */
1820                 if (!bond->curr_active_slave)
1821                         bond->curr_active_slave = new_slave;
1822
1823                 break;
1824         } /* switch(bond_mode) */
1825
1826         write_unlock_bh(&bond->curr_slave_lock);
1827
1828         bond_set_carrier(bond);
1829
1830 #ifdef CONFIG_NET_POLL_CONTROLLER
1831         slave_dev->npinfo = bond_netpoll_info(bond);
1832         if (slave_dev->npinfo) {
1833                 if (slave_enable_netpoll(new_slave)) {
1834                         read_unlock(&bond->lock);
1835                         pr_info("Error, %s: master_dev is using netpoll, "
1836                                  "but new slave device does not support netpoll.\n",
1837                                  bond_dev->name);
1838                         res = -EBUSY;
1839                         goto err_close;
1840                 }
1841         }
1842 #endif
1843
1844         read_unlock(&bond->lock);
1845
1846         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1847         if (res)
1848                 goto err_close;
1849
1850         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1851                                          new_slave);
1852         if (res) {
1853                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1854                 goto err_dest_symlinks;
1855         }
1856
1857         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1858                 bond_dev->name, slave_dev->name,
1859                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1860                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1861
1862         /* enslave is successful */
1863         return 0;
1864
1865 /* Undo stages on error */
1866 err_dest_symlinks:
1867         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1868
1869 err_close:
1870         dev_close(slave_dev);
1871
1872 err_unset_master:
1873         netdev_set_bond_master(slave_dev, NULL);
1874
1875 err_restore_mac:
1876         if (!bond->params.fail_over_mac) {
1877                 /* XXX TODO - fom follow mode needs to change master's
1878                  * MAC if this slave's MAC is in use by the bond, or at
1879                  * least print a warning.
1880                  */
1881                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1882                 addr.sa_family = slave_dev->type;
1883                 dev_set_mac_address(slave_dev, &addr);
1884         }
1885
1886 err_restore_mtu:
1887         dev_set_mtu(slave_dev, new_slave->original_mtu);
1888
1889 err_free:
1890         kfree(new_slave);
1891
1892 err_undo_flags:
1893         bond_compute_features(bond);
1894
1895         return res;
1896 }
1897
1898 /*
1899  * Try to release the slave device <slave> from the bond device <master>
1900  * It is legal to access curr_active_slave without a lock because all the function
1901  * is write-locked.
1902  *
1903  * The rules for slave state should be:
1904  *   for Active/Backup:
1905  *     Active stays on all backups go down
1906  *   for Bonded connections:
1907  *     The first up interface should be left on and all others downed.
1908  */
1909 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1910 {
1911         struct bonding *bond = netdev_priv(bond_dev);
1912         struct slave *slave, *oldcurrent;
1913         struct sockaddr addr;
1914         u32 old_features = bond_dev->features;
1915
1916         /* slave is not a slave or master is not master of this slave */
1917         if (!(slave_dev->flags & IFF_SLAVE) ||
1918             (slave_dev->master != bond_dev)) {
1919                 pr_err("%s: Error: cannot release %s.\n",
1920                        bond_dev->name, slave_dev->name);
1921                 return -EINVAL;
1922         }
1923
1924         block_netpoll_tx();
1925         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1926         write_lock_bh(&bond->lock);
1927
1928         slave = bond_get_slave_by_dev(bond, slave_dev);
1929         if (!slave) {
1930                 /* not a slave of this bond */
1931                 pr_info("%s: %s not enslaved\n",
1932                         bond_dev->name, slave_dev->name);
1933                 write_unlock_bh(&bond->lock);
1934                 unblock_netpoll_tx();
1935                 return -EINVAL;
1936         }
1937
1938         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1939          * for this slave anymore.
1940          */
1941         netdev_rx_handler_unregister(slave_dev);
1942         write_unlock_bh(&bond->lock);
1943         synchronize_net();
1944         write_lock_bh(&bond->lock);
1945
1946         if (!bond->params.fail_over_mac) {
1947                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1948                     bond->slave_cnt > 1)
1949                         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",
1950                                    bond_dev->name, slave_dev->name,
1951                                    slave->perm_hwaddr,
1952                                    bond_dev->name, slave_dev->name);
1953         }
1954
1955         /* Inform AD package of unbinding of slave. */
1956         if (bond->params.mode == BOND_MODE_8023AD) {
1957                 /* must be called before the slave is
1958                  * detached from the list
1959                  */
1960                 bond_3ad_unbind_slave(slave);
1961         }
1962
1963         pr_info("%s: releasing %s interface %s\n",
1964                 bond_dev->name,
1965                 bond_is_active_slave(slave) ? "active" : "backup",
1966                 slave_dev->name);
1967
1968         oldcurrent = bond->curr_active_slave;
1969
1970         bond->current_arp_slave = NULL;
1971
1972         /* release the slave from its bond */
1973         bond_detach_slave(bond, slave);
1974
1975         if (bond->primary_slave == slave)
1976                 bond->primary_slave = NULL;
1977
1978         if (oldcurrent == slave)
1979                 bond_change_active_slave(bond, NULL);
1980
1981         if (bond_is_lb(bond)) {
1982                 /* Must be called only after the slave has been
1983                  * detached from the list and the curr_active_slave
1984                  * has been cleared (if our_slave == old_current),
1985                  * but before a new active slave is selected.
1986                  */
1987                 write_unlock_bh(&bond->lock);
1988                 bond_alb_deinit_slave(bond, slave);
1989                 write_lock_bh(&bond->lock);
1990         }
1991
1992         if (oldcurrent == slave) {
1993                 /*
1994                  * Note that we hold RTNL over this sequence, so there
1995                  * is no concern that another slave add/remove event
1996                  * will interfere.
1997                  */
1998                 write_unlock_bh(&bond->lock);
1999                 read_lock(&bond->lock);
2000                 write_lock_bh(&bond->curr_slave_lock);
2001
2002                 bond_select_active_slave(bond);
2003
2004                 write_unlock_bh(&bond->curr_slave_lock);
2005                 read_unlock(&bond->lock);
2006                 write_lock_bh(&bond->lock);
2007         }
2008
2009         if (bond->slave_cnt == 0) {
2010                 bond_set_carrier(bond);
2011
2012                 /* if the last slave was removed, zero the mac address
2013                  * of the master so it will be set by the application
2014                  * to the mac address of the first slave
2015                  */
2016                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2017
2018                 if (bond_vlan_used(bond)) {
2019                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2020                                    bond_dev->name, bond_dev->name);
2021                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2022                                    bond_dev->name);
2023                 }
2024         }
2025
2026         write_unlock_bh(&bond->lock);
2027         unblock_netpoll_tx();
2028
2029         bond_compute_features(bond);
2030         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2031             (old_features & NETIF_F_VLAN_CHALLENGED))
2032                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2033                         bond_dev->name, slave_dev->name, bond_dev->name);
2034
2035         /* must do this from outside any spinlocks */
2036         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2037
2038         bond_del_vlans_from_slave(bond, slave_dev);
2039
2040         /* If the mode USES_PRIMARY, then we should only remove its
2041          * promisc and mc settings if it was the curr_active_slave, but that was
2042          * already taken care of above when we detached the slave
2043          */
2044         if (!USES_PRIMARY(bond->params.mode)) {
2045                 /* unset promiscuity level from slave */
2046                 if (bond_dev->flags & IFF_PROMISC)
2047                         dev_set_promiscuity(slave_dev, -1);
2048
2049                 /* unset allmulti level from slave */
2050                 if (bond_dev->flags & IFF_ALLMULTI)
2051                         dev_set_allmulti(slave_dev, -1);
2052
2053                 /* flush master's mc_list from slave */
2054                 netif_addr_lock_bh(bond_dev);
2055                 bond_mc_list_flush(bond_dev, slave_dev);
2056                 netif_addr_unlock_bh(bond_dev);
2057         }
2058
2059         netdev_set_bond_master(slave_dev, NULL);
2060
2061         slave_disable_netpoll(slave);
2062
2063         /* close slave before restoring its mac address */
2064         dev_close(slave_dev);
2065
2066         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2067                 /* restore original ("permanent") mac address */
2068                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2069                 addr.sa_family = slave_dev->type;
2070                 dev_set_mac_address(slave_dev, &addr);
2071         }
2072
2073         dev_set_mtu(slave_dev, slave->original_mtu);
2074
2075         slave_dev->priv_flags &= ~IFF_BONDING;
2076
2077         kfree(slave);
2078
2079         return 0;  /* deletion OK */
2080 }
2081
2082 /*
2083 * First release a slave and then destroy the bond if no more slaves are left.
2084 * Must be under rtnl_lock when this function is called.
2085 */
2086 static int  bond_release_and_destroy(struct net_device *bond_dev,
2087                                      struct net_device *slave_dev)
2088 {
2089         struct bonding *bond = netdev_priv(bond_dev);
2090         int ret;
2091
2092         ret = bond_release(bond_dev, slave_dev);
2093         if ((ret == 0) && (bond->slave_cnt == 0)) {
2094                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2095                 pr_info("%s: destroying bond %s.\n",
2096                         bond_dev->name, bond_dev->name);
2097                 unregister_netdevice(bond_dev);
2098         }
2099         return ret;
2100 }
2101
2102 /*
2103  * This function releases all slaves.
2104  */
2105 static int bond_release_all(struct net_device *bond_dev)
2106 {
2107         struct bonding *bond = netdev_priv(bond_dev);
2108         struct slave *slave;
2109         struct net_device *slave_dev;
2110         struct sockaddr addr;
2111
2112         write_lock_bh(&bond->lock);
2113
2114         netif_carrier_off(bond_dev);
2115
2116         if (bond->slave_cnt == 0)
2117                 goto out;
2118
2119         bond->current_arp_slave = NULL;
2120         bond->primary_slave = NULL;
2121         bond_change_active_slave(bond, NULL);
2122
2123         while ((slave = bond->first_slave) != NULL) {
2124                 /* Inform AD package of unbinding of slave
2125                  * before slave is detached from the list.
2126                  */
2127                 if (bond->params.mode == BOND_MODE_8023AD)
2128                         bond_3ad_unbind_slave(slave);
2129
2130                 slave_dev = slave->dev;
2131                 bond_detach_slave(bond, slave);
2132
2133                 /* now that the slave is detached, unlock and perform
2134                  * all the undo steps that should not be called from
2135                  * within a lock.
2136                  */
2137                 write_unlock_bh(&bond->lock);
2138
2139                 /* unregister rx_handler early so bond_handle_frame wouldn't
2140                  * be called for this slave anymore.
2141                  */
2142                 netdev_rx_handler_unregister(slave_dev);
2143                 synchronize_net();
2144
2145                 if (bond_is_lb(bond)) {
2146                         /* must be called only after the slave
2147                          * has been detached from the list
2148                          */
2149                         bond_alb_deinit_slave(bond, slave);
2150                 }
2151
2152                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2153                 bond_del_vlans_from_slave(bond, slave_dev);
2154
2155                 /* If the mode USES_PRIMARY, then we should only remove its
2156                  * promisc and mc settings if it was the curr_active_slave, but that was
2157                  * already taken care of above when we detached the slave
2158                  */
2159                 if (!USES_PRIMARY(bond->params.mode)) {
2160                         /* unset promiscuity level from slave */
2161                         if (bond_dev->flags & IFF_PROMISC)
2162                                 dev_set_promiscuity(slave_dev, -1);
2163
2164                         /* unset allmulti level from slave */
2165                         if (bond_dev->flags & IFF_ALLMULTI)
2166                                 dev_set_allmulti(slave_dev, -1);
2167
2168                         /* flush master's mc_list from slave */
2169                         netif_addr_lock_bh(bond_dev);
2170                         bond_mc_list_flush(bond_dev, slave_dev);
2171                         netif_addr_unlock_bh(bond_dev);
2172                 }
2173
2174                 netdev_set_bond_master(slave_dev, NULL);
2175
2176                 slave_disable_netpoll(slave);
2177
2178                 /* close slave before restoring its mac address */
2179                 dev_close(slave_dev);
2180
2181                 if (!bond->params.fail_over_mac) {
2182                         /* restore original ("permanent") mac address*/
2183                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2184                         addr.sa_family = slave_dev->type;
2185                         dev_set_mac_address(slave_dev, &addr);
2186                 }
2187
2188                 kfree(slave);
2189
2190                 /* re-acquire the lock before getting the next slave */
2191                 write_lock_bh(&bond->lock);
2192         }
2193
2194         /* zero the mac address of the master so it will be
2195          * set by the application to the mac address of the
2196          * first slave
2197          */
2198         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2199
2200         if (bond_vlan_used(bond)) {
2201                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2202                            bond_dev->name, bond_dev->name);
2203                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2204                            bond_dev->name);
2205         }
2206
2207         pr_info("%s: released all slaves\n", bond_dev->name);
2208
2209 out:
2210         write_unlock_bh(&bond->lock);
2211
2212         bond_compute_features(bond);
2213
2214         return 0;
2215 }
2216
2217 /*
2218  * This function changes the active slave to slave <slave_dev>.
2219  * It returns -EINVAL in the following cases.
2220  *  - <slave_dev> is not found in the list.
2221  *  - There is not active slave now.
2222  *  - <slave_dev> is already active.
2223  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2224  *  - <slave_dev> is not running.
2225  * In these cases, this function does nothing.
2226  * In the other cases, current_slave pointer is changed and 0 is returned.
2227  */
2228 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2229 {
2230         struct bonding *bond = netdev_priv(bond_dev);
2231         struct slave *old_active = NULL;
2232         struct slave *new_active = NULL;
2233         int res = 0;
2234
2235         if (!USES_PRIMARY(bond->params.mode))
2236                 return -EINVAL;
2237
2238         /* Verify that master_dev is indeed the master of slave_dev */
2239         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2240                 return -EINVAL;
2241
2242         read_lock(&bond->lock);
2243
2244         read_lock(&bond->curr_slave_lock);
2245         old_active = bond->curr_active_slave;
2246         read_unlock(&bond->curr_slave_lock);
2247
2248         new_active = bond_get_slave_by_dev(bond, slave_dev);
2249
2250         /*
2251          * Changing to the current active: do nothing; return success.
2252          */
2253         if (new_active && (new_active == old_active)) {
2254                 read_unlock(&bond->lock);
2255                 return 0;
2256         }
2257
2258         if ((new_active) &&
2259             (old_active) &&
2260             (new_active->link == BOND_LINK_UP) &&
2261             IS_UP(new_active->dev)) {
2262                 block_netpoll_tx();
2263                 write_lock_bh(&bond->curr_slave_lock);
2264                 bond_change_active_slave(bond, new_active);
2265                 write_unlock_bh(&bond->curr_slave_lock);
2266                 unblock_netpoll_tx();
2267         } else
2268                 res = -EINVAL;
2269
2270         read_unlock(&bond->lock);
2271
2272         return res;
2273 }
2274
2275 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2276 {
2277         struct bonding *bond = netdev_priv(bond_dev);
2278
2279         info->bond_mode = bond->params.mode;
2280         info->miimon = bond->params.miimon;
2281
2282         read_lock(&bond->lock);
2283         info->num_slaves = bond->slave_cnt;
2284         read_unlock(&bond->lock);
2285
2286         return 0;
2287 }
2288
2289 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2290 {
2291         struct bonding *bond = netdev_priv(bond_dev);
2292         struct slave *slave;
2293         int i, res = -ENODEV;
2294
2295         read_lock(&bond->lock);
2296
2297         bond_for_each_slave(bond, slave, i) {
2298                 if (i == (int)info->slave_id) {
2299                         res = 0;
2300                         strcpy(info->slave_name, slave->dev->name);
2301                         info->link = slave->link;
2302                         info->state = bond_slave_state(slave);
2303                         info->link_failure_count = slave->link_failure_count;
2304                         break;
2305                 }
2306         }
2307
2308         read_unlock(&bond->lock);
2309
2310         return res;
2311 }
2312
2313 /*-------------------------------- Monitoring -------------------------------*/
2314
2315
2316 static int bond_miimon_inspect(struct bonding *bond)
2317 {
2318         struct slave *slave;
2319         int i, link_state, commit = 0;
2320         bool ignore_updelay;
2321
2322         ignore_updelay = !bond->curr_active_slave ? true : false;
2323
2324         bond_for_each_slave(bond, slave, i) {
2325                 slave->new_link = BOND_LINK_NOCHANGE;
2326
2327                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2328
2329                 switch (slave->link) {
2330                 case BOND_LINK_UP:
2331                         if (link_state)
2332                                 continue;
2333
2334                         slave->link = BOND_LINK_FAIL;
2335                         slave->delay = bond->params.downdelay;
2336                         if (slave->delay) {
2337                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2338                                         bond->dev->name,
2339                                         (bond->params.mode ==
2340                                          BOND_MODE_ACTIVEBACKUP) ?
2341                                         (bond_is_active_slave(slave) ?
2342                                          "active " : "backup ") : "",
2343                                         slave->dev->name,
2344                                         bond->params.downdelay * bond->params.miimon);
2345                         }
2346                         /*FALLTHRU*/
2347                 case BOND_LINK_FAIL:
2348                         if (link_state) {
2349                                 /*
2350                                  * recovered before downdelay expired
2351                                  */
2352                                 slave->link = BOND_LINK_UP;
2353                                 slave->jiffies = jiffies;
2354                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2355                                         bond->dev->name,
2356                                         (bond->params.downdelay - slave->delay) *
2357                                         bond->params.miimon,
2358                                         slave->dev->name);
2359                                 continue;
2360                         }
2361
2362                         if (slave->delay <= 0) {
2363                                 slave->new_link = BOND_LINK_DOWN;
2364                                 commit++;
2365                                 continue;
2366                         }
2367
2368                         slave->delay--;
2369                         break;
2370
2371                 case BOND_LINK_DOWN:
2372                         if (!link_state)
2373                                 continue;
2374
2375                         slave->link = BOND_LINK_BACK;
2376                         slave->delay = bond->params.updelay;
2377
2378                         if (slave->delay) {
2379                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2380                                         bond->dev->name, slave->dev->name,
2381                                         ignore_updelay ? 0 :
2382                                         bond->params.updelay *
2383                                         bond->params.miimon);
2384                         }
2385                         /*FALLTHRU*/
2386                 case BOND_LINK_BACK:
2387                         if (!link_state) {
2388                                 slave->link = BOND_LINK_DOWN;
2389                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2390                                         bond->dev->name,
2391                                         (bond->params.updelay - slave->delay) *
2392                                         bond->params.miimon,
2393                                         slave->dev->name);
2394
2395                                 continue;
2396                         }
2397
2398                         if (ignore_updelay)
2399                                 slave->delay = 0;
2400
2401                         if (slave->delay <= 0) {
2402                                 slave->new_link = BOND_LINK_UP;
2403                                 commit++;
2404                                 ignore_updelay = false;
2405                                 continue;
2406                         }
2407
2408                         slave->delay--;
2409                         break;
2410                 }
2411         }
2412
2413         return commit;
2414 }
2415
2416 static void bond_miimon_commit(struct bonding *bond)
2417 {
2418         struct slave *slave;
2419         int i;
2420
2421         bond_for_each_slave(bond, slave, i) {
2422                 switch (slave->new_link) {
2423                 case BOND_LINK_NOCHANGE:
2424                         continue;
2425
2426                 case BOND_LINK_UP:
2427                         slave->link = BOND_LINK_UP;
2428                         slave->jiffies = jiffies;
2429
2430                         if (bond->params.mode == BOND_MODE_8023AD) {
2431                                 /* prevent it from being the active one */
2432                                 bond_set_backup_slave(slave);
2433                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2434                                 /* make it immediately active */
2435                                 bond_set_active_slave(slave);
2436                         } else if (slave != bond->primary_slave) {
2437                                 /* prevent it from being the active one */
2438                                 bond_set_backup_slave(slave);
2439                         }
2440
2441                         bond_update_speed_duplex(slave);
2442
2443                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2444                                 bond->dev->name, slave->dev->name,
2445                                 slave->speed, slave->duplex ? "full" : "half");
2446
2447                         /* notify ad that the link status has changed */
2448                         if (bond->params.mode == BOND_MODE_8023AD)
2449                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2450
2451                         if (bond_is_lb(bond))
2452                                 bond_alb_handle_link_change(bond, slave,
2453                                                             BOND_LINK_UP);
2454
2455                         if (!bond->curr_active_slave ||
2456                             (slave == bond->primary_slave))
2457                                 goto do_failover;
2458
2459                         continue;
2460
2461                 case BOND_LINK_DOWN:
2462                         if (slave->link_failure_count < UINT_MAX)
2463                                 slave->link_failure_count++;
2464
2465                         slave->link = BOND_LINK_DOWN;
2466
2467                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2468                             bond->params.mode == BOND_MODE_8023AD)
2469                                 bond_set_slave_inactive_flags(slave);
2470
2471                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2472                                 bond->dev->name, slave->dev->name);
2473
2474                         if (bond->params.mode == BOND_MODE_8023AD)
2475                                 bond_3ad_handle_link_change(slave,
2476                                                             BOND_LINK_DOWN);
2477
2478                         if (bond_is_lb(bond))
2479                                 bond_alb_handle_link_change(bond, slave,
2480                                                             BOND_LINK_DOWN);
2481
2482                         if (slave == bond->curr_active_slave)
2483                                 goto do_failover;
2484
2485                         continue;
2486
2487                 default:
2488                         pr_err("%s: invalid new link %d on slave %s\n",
2489                                bond->dev->name, slave->new_link,
2490                                slave->dev->name);
2491                         slave->new_link = BOND_LINK_NOCHANGE;
2492
2493                         continue;
2494                 }
2495
2496 do_failover:
2497                 ASSERT_RTNL();
2498                 block_netpoll_tx();
2499                 write_lock_bh(&bond->curr_slave_lock);
2500                 bond_select_active_slave(bond);
2501                 write_unlock_bh(&bond->curr_slave_lock);
2502                 unblock_netpoll_tx();
2503         }
2504
2505         bond_set_carrier(bond);
2506 }
2507
2508 /*
2509  * bond_mii_monitor
2510  *
2511  * Really a wrapper that splits the mii monitor into two phases: an
2512  * inspection, then (if inspection indicates something needs to be done)
2513  * an acquisition of appropriate locks followed by a commit phase to
2514  * implement whatever link state changes are indicated.
2515  */
2516 void bond_mii_monitor(struct work_struct *work)
2517 {
2518         struct bonding *bond = container_of(work, struct bonding,
2519                                             mii_work.work);
2520         bool should_notify_peers = false;
2521
2522         read_lock(&bond->lock);
2523         if (bond->kill_timers)
2524                 goto out;
2525
2526         if (bond->slave_cnt == 0)
2527                 goto re_arm;
2528
2529         should_notify_peers = bond_should_notify_peers(bond);
2530
2531         if (bond_miimon_inspect(bond)) {
2532                 read_unlock(&bond->lock);
2533                 rtnl_lock();
2534                 read_lock(&bond->lock);
2535
2536                 bond_miimon_commit(bond);
2537
2538                 read_unlock(&bond->lock);
2539                 rtnl_unlock();  /* might sleep, hold no other locks */
2540                 read_lock(&bond->lock);
2541         }
2542
2543 re_arm:
2544         if (bond->params.miimon && !bond->kill_timers)
2545                 queue_delayed_work(bond->wq, &bond->mii_work,
2546                                    msecs_to_jiffies(bond->params.miimon));
2547 out:
2548         read_unlock(&bond->lock);
2549
2550         if (should_notify_peers) {
2551                 rtnl_lock();
2552                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2553                 rtnl_unlock();
2554         }
2555 }
2556
2557 static __be32 bond_glean_dev_ip(struct net_device *dev)
2558 {
2559         struct in_device *idev;
2560         struct in_ifaddr *ifa;
2561         __be32 addr = 0;
2562
2563         if (!dev)
2564                 return 0;
2565
2566         rcu_read_lock();
2567         idev = __in_dev_get_rcu(dev);
2568         if (!idev)
2569                 goto out;
2570
2571         ifa = idev->ifa_list;
2572         if (!ifa)
2573                 goto out;
2574
2575         addr = ifa->ifa_local;
2576 out:
2577         rcu_read_unlock();
2578         return addr;
2579 }
2580
2581 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2582 {
2583         struct vlan_entry *vlan;
2584
2585         if (ip == bond->master_ip)
2586                 return 1;
2587
2588         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2589                 if (ip == vlan->vlan_ip)
2590                         return 1;
2591         }
2592
2593         return 0;
2594 }
2595
2596 /*
2597  * We go to the (large) trouble of VLAN tagging ARP frames because
2598  * switches in VLAN mode (especially if ports are configured as
2599  * "native" to a VLAN) might not pass non-tagged frames.
2600  */
2601 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2602 {
2603         struct sk_buff *skb;
2604
2605         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2606                  slave_dev->name, dest_ip, src_ip, vlan_id);
2607
2608         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2609                          NULL, slave_dev->dev_addr, NULL);
2610
2611         if (!skb) {
2612                 pr_err("ARP packet allocation failed\n");
2613                 return;
2614         }
2615         if (vlan_id) {
2616                 skb = vlan_put_tag(skb, vlan_id);
2617                 if (!skb) {
2618                         pr_err("failed to insert VLAN tag\n");
2619                         return;
2620                 }
2621         }
2622         arp_xmit(skb);
2623 }
2624
2625
2626 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2627 {
2628         int i, vlan_id;
2629         __be32 *targets = bond->params.arp_targets;
2630         struct vlan_entry *vlan;
2631         struct net_device *vlan_dev;
2632         struct rtable *rt;
2633
2634         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2635                 if (!targets[i])
2636                         break;
2637                 pr_debug("basa: target %x\n", targets[i]);
2638                 if (!bond_vlan_used(bond)) {
2639                         pr_debug("basa: empty vlan: arp_send\n");
2640                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2641                                       bond->master_ip, 0);
2642                         continue;
2643                 }
2644
2645                 /*
2646                  * If VLANs are configured, we do a route lookup to
2647                  * determine which VLAN interface would be used, so we
2648                  * can tag the ARP with the proper VLAN tag.
2649                  */
2650                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2651                                      RTO_ONLINK, 0);
2652                 if (IS_ERR(rt)) {
2653                         if (net_ratelimit()) {
2654                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2655                                            bond->dev->name, &targets[i]);
2656                         }
2657                         continue;
2658                 }
2659
2660                 /*
2661                  * This target is not on a VLAN
2662                  */
2663                 if (rt->dst.dev == bond->dev) {
2664                         ip_rt_put(rt);
2665                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2666                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2667                                       bond->master_ip, 0);
2668                         continue;
2669                 }
2670
2671                 vlan_id = 0;
2672                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2673                         rcu_read_lock();
2674                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2675                                                         vlan->vlan_id);
2676                         rcu_read_unlock();
2677                         if (vlan_dev == rt->dst.dev) {
2678                                 vlan_id = vlan->vlan_id;
2679                                 pr_debug("basa: vlan match on %s %d\n",
2680                                        vlan_dev->name, vlan_id);
2681                                 break;
2682                         }
2683                 }
2684
2685                 if (vlan_id) {
2686                         ip_rt_put(rt);
2687                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2688                                       vlan->vlan_ip, vlan_id);
2689                         continue;
2690                 }
2691
2692                 if (net_ratelimit()) {
2693                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2694                                    bond->dev->name, &targets[i],
2695                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2696                 }
2697                 ip_rt_put(rt);
2698         }
2699 }
2700
2701 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2702 {
2703         int i;
2704         __be32 *targets = bond->params.arp_targets;
2705
2706         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2707                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2708                          &sip, &tip, i, &targets[i],
2709                          bond_has_this_ip(bond, tip));
2710                 if (sip == targets[i]) {
2711                         if (bond_has_this_ip(bond, tip))
2712                                 slave->last_arp_rx = jiffies;
2713                         return;
2714                 }
2715         }
2716 }
2717
2718 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2719                          struct slave *slave)
2720 {
2721         struct arphdr *arp;
2722         unsigned char *arp_ptr;
2723         __be32 sip, tip;
2724
2725         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2726                 return;
2727
2728         read_lock(&bond->lock);
2729
2730         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2731                  bond->dev->name, skb->dev->name);
2732
2733         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2734                 goto out_unlock;
2735
2736         arp = arp_hdr(skb);
2737         if (arp->ar_hln != bond->dev->addr_len ||
2738             skb->pkt_type == PACKET_OTHERHOST ||
2739             skb->pkt_type == PACKET_LOOPBACK ||
2740             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2741             arp->ar_pro != htons(ETH_P_IP) ||
2742             arp->ar_pln != 4)
2743                 goto out_unlock;
2744
2745         arp_ptr = (unsigned char *)(arp + 1);
2746         arp_ptr += bond->dev->addr_len;
2747         memcpy(&sip, arp_ptr, 4);
2748         arp_ptr += 4 + bond->dev->addr_len;
2749         memcpy(&tip, arp_ptr, 4);
2750
2751         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2752                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2753                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2754                  &sip, &tip);
2755
2756         /*
2757          * Backup slaves won't see the ARP reply, but do come through
2758          * here for each ARP probe (so we swap the sip/tip to validate
2759          * the probe).  In a "redundant switch, common router" type of
2760          * configuration, the ARP probe will (hopefully) travel from
2761          * the active, through one switch, the router, then the other
2762          * switch before reaching the backup.
2763          */
2764         if (bond_is_active_slave(slave))
2765                 bond_validate_arp(bond, slave, sip, tip);
2766         else
2767                 bond_validate_arp(bond, slave, tip, sip);
2768
2769 out_unlock:
2770         read_unlock(&bond->lock);
2771 }
2772
2773 /*
2774  * this function is called regularly to monitor each slave's link
2775  * ensuring that traffic is being sent and received when arp monitoring
2776  * is used in load-balancing mode. if the adapter has been dormant, then an
2777  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2778  * arp monitoring in active backup mode.
2779  */
2780 void bond_loadbalance_arp_mon(struct work_struct *work)
2781 {
2782         struct bonding *bond = container_of(work, struct bonding,
2783                                             arp_work.work);
2784         struct slave *slave, *oldcurrent;
2785         int do_failover = 0;
2786         int delta_in_ticks;
2787         int i;
2788
2789         read_lock(&bond->lock);
2790
2791         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2792
2793         if (bond->kill_timers)
2794                 goto out;
2795
2796         if (bond->slave_cnt == 0)
2797                 goto re_arm;
2798
2799         read_lock(&bond->curr_slave_lock);
2800         oldcurrent = bond->curr_active_slave;
2801         read_unlock(&bond->curr_slave_lock);
2802
2803         /* see if any of the previous devices are up now (i.e. they have
2804          * xmt and rcv traffic). the curr_active_slave does not come into
2805          * the picture unless it is null. also, slave->jiffies is not needed
2806          * here because we send an arp on each slave and give a slave as
2807          * long as it needs to get the tx/rx within the delta.
2808          * TODO: what about up/down delay in arp mode? it wasn't here before
2809          *       so it can wait
2810          */
2811         bond_for_each_slave(bond, slave, i) {
2812                 unsigned long trans_start = dev_trans_start(slave->dev);
2813
2814                 if (slave->link != BOND_LINK_UP) {
2815                         if (time_in_range(jiffies,
2816                                 trans_start - delta_in_ticks,
2817                                 trans_start + delta_in_ticks) &&
2818                             time_in_range(jiffies,
2819                                 slave->dev->last_rx - delta_in_ticks,
2820                                 slave->dev->last_rx + delta_in_ticks)) {
2821
2822                                 slave->link  = BOND_LINK_UP;
2823                                 bond_set_active_slave(slave);
2824
2825                                 /* primary_slave has no meaning in round-robin
2826                                  * mode. the window of a slave being up and
2827                                  * curr_active_slave being null after enslaving
2828                                  * is closed.
2829                                  */
2830                                 if (!oldcurrent) {
2831                                         pr_info("%s: link status definitely up for interface %s, ",
2832                                                 bond->dev->name,
2833                                                 slave->dev->name);
2834                                         do_failover = 1;
2835                                 } else {
2836                                         pr_info("%s: interface %s is now up\n",
2837                                                 bond->dev->name,
2838                                                 slave->dev->name);
2839                                 }
2840                         }
2841                 } else {
2842                         /* slave->link == BOND_LINK_UP */
2843
2844                         /* not all switches will respond to an arp request
2845                          * when the source ip is 0, so don't take the link down
2846                          * if we don't know our ip yet
2847                          */
2848                         if (!time_in_range(jiffies,
2849                                 trans_start - delta_in_ticks,
2850                                 trans_start + 2 * delta_in_ticks) ||
2851                             !time_in_range(jiffies,
2852                                 slave->dev->last_rx - delta_in_ticks,
2853                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2854
2855                                 slave->link  = BOND_LINK_DOWN;
2856                                 bond_set_backup_slave(slave);
2857
2858                                 if (slave->link_failure_count < UINT_MAX)
2859                                         slave->link_failure_count++;
2860
2861                                 pr_info("%s: interface %s is now down.\n",
2862                                         bond->dev->name,
2863                                         slave->dev->name);
2864
2865                                 if (slave == oldcurrent)
2866                                         do_failover = 1;
2867                         }
2868                 }
2869
2870                 /* note: if switch is in round-robin mode, all links
2871                  * must tx arp to ensure all links rx an arp - otherwise
2872                  * links may oscillate or not come up at all; if switch is
2873                  * in something like xor mode, there is nothing we can
2874                  * do - all replies will be rx'ed on same link causing slaves
2875                  * to be unstable during low/no traffic periods
2876                  */
2877                 if (IS_UP(slave->dev))
2878                         bond_arp_send_all(bond, slave);
2879         }
2880
2881         if (do_failover) {
2882                 block_netpoll_tx();
2883                 write_lock_bh(&bond->curr_slave_lock);
2884
2885                 bond_select_active_slave(bond);
2886
2887                 write_unlock_bh(&bond->curr_slave_lock);
2888                 unblock_netpoll_tx();
2889         }
2890
2891 re_arm:
2892         if (bond->params.arp_interval && !bond->kill_timers)
2893                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2894 out:
2895         read_unlock(&bond->lock);
2896 }
2897
2898 /*
2899  * Called to inspect slaves for active-backup mode ARP monitor link state
2900  * changes.  Sets new_link in slaves to specify what action should take
2901  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2902  * to link states must be committed.
2903  *
2904  * Called with bond->lock held for read.
2905  */
2906 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2907 {
2908         struct slave *slave;
2909         int i, commit = 0;
2910         unsigned long trans_start;
2911
2912         bond_for_each_slave(bond, slave, i) {
2913                 slave->new_link = BOND_LINK_NOCHANGE;
2914
2915                 if (slave->link != BOND_LINK_UP) {
2916                         if (time_in_range(jiffies,
2917                                 slave_last_rx(bond, slave) - delta_in_ticks,
2918                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2919
2920                                 slave->new_link = BOND_LINK_UP;
2921                                 commit++;
2922                         }
2923
2924                         continue;
2925                 }
2926
2927                 /*
2928                  * Give slaves 2*delta after being enslaved or made
2929                  * active.  This avoids bouncing, as the last receive
2930                  * times need a full ARP monitor cycle to be updated.
2931                  */
2932                 if (time_in_range(jiffies,
2933                                   slave->jiffies - delta_in_ticks,
2934                                   slave->jiffies + 2 * delta_in_ticks))
2935                         continue;
2936
2937                 /*
2938                  * Backup slave is down if:
2939                  * - No current_arp_slave AND
2940                  * - more than 3*delta since last receive AND
2941                  * - the bond has an IP address
2942                  *
2943                  * Note: a non-null current_arp_slave indicates
2944                  * the curr_active_slave went down and we are
2945                  * searching for a new one; under this condition
2946                  * we only take the curr_active_slave down - this
2947                  * gives each slave a chance to tx/rx traffic
2948                  * before being taken out
2949                  */
2950                 if (!bond_is_active_slave(slave) &&
2951                     !bond->current_arp_slave &&
2952                     !time_in_range(jiffies,
2953                         slave_last_rx(bond, slave) - delta_in_ticks,
2954                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2955
2956                         slave->new_link = BOND_LINK_DOWN;
2957                         commit++;
2958                 }
2959
2960                 /*
2961                  * Active slave is down if:
2962                  * - more than 2*delta since transmitting OR
2963                  * - (more than 2*delta since receive AND
2964                  *    the bond has an IP address)
2965                  */
2966                 trans_start = dev_trans_start(slave->dev);
2967                 if (bond_is_active_slave(slave) &&
2968                     (!time_in_range(jiffies,
2969                         trans_start - delta_in_ticks,
2970                         trans_start + 2 * delta_in_ticks) ||
2971                      !time_in_range(jiffies,
2972                         slave_last_rx(bond, slave) - delta_in_ticks,
2973                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2974
2975                         slave->new_link = BOND_LINK_DOWN;
2976                         commit++;
2977                 }
2978         }
2979
2980         return commit;
2981 }
2982
2983 /*
2984  * Called to commit link state changes noted by inspection step of
2985  * active-backup mode ARP monitor.
2986  *
2987  * Called with RTNL and bond->lock for read.
2988  */
2989 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2990 {
2991         struct slave *slave;
2992         int i;
2993         unsigned long trans_start;
2994
2995         bond_for_each_slave(bond, slave, i) {
2996                 switch (slave->new_link) {
2997                 case BOND_LINK_NOCHANGE:
2998                         continue;
2999
3000                 case BOND_LINK_UP:
3001                         trans_start = dev_trans_start(slave->dev);
3002                         if ((!bond->curr_active_slave &&
3003                              time_in_range(jiffies,
3004                                            trans_start - delta_in_ticks,
3005                                            trans_start + delta_in_ticks)) ||
3006                             bond->curr_active_slave != slave) {
3007                                 slave->link = BOND_LINK_UP;
3008                                 bond->current_arp_slave = NULL;
3009
3010                                 pr_info("%s: link status definitely up for interface %s.\n",
3011                                         bond->dev->name, slave->dev->name);
3012
3013                                 if (!bond->curr_active_slave ||
3014                                     (slave == bond->primary_slave))
3015                                         goto do_failover;
3016
3017                         }
3018
3019                         continue;
3020
3021                 case BOND_LINK_DOWN:
3022                         if (slave->link_failure_count < UINT_MAX)
3023                                 slave->link_failure_count++;
3024
3025                         slave->link = BOND_LINK_DOWN;
3026                         bond_set_slave_inactive_flags(slave);
3027
3028                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3029                                 bond->dev->name, slave->dev->name);
3030
3031                         if (slave == bond->curr_active_slave) {
3032                                 bond->current_arp_slave = NULL;
3033                                 goto do_failover;
3034                         }
3035
3036                         continue;
3037
3038                 default:
3039                         pr_err("%s: impossible: new_link %d on slave %s\n",
3040                                bond->dev->name, slave->new_link,
3041                                slave->dev->name);
3042                         continue;
3043                 }
3044
3045 do_failover:
3046                 ASSERT_RTNL();
3047                 block_netpoll_tx();
3048                 write_lock_bh(&bond->curr_slave_lock);
3049                 bond_select_active_slave(bond);
3050                 write_unlock_bh(&bond->curr_slave_lock);
3051                 unblock_netpoll_tx();
3052         }
3053
3054         bond_set_carrier(bond);
3055 }
3056
3057 /*
3058  * Send ARP probes for active-backup mode ARP monitor.
3059  *
3060  * Called with bond->lock held for read.
3061  */
3062 static void bond_ab_arp_probe(struct bonding *bond)
3063 {
3064         struct slave *slave;
3065         int i;
3066
3067         read_lock(&bond->curr_slave_lock);
3068
3069         if (bond->current_arp_slave && bond->curr_active_slave)
3070                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3071                         bond->current_arp_slave->dev->name,
3072                         bond->curr_active_slave->dev->name);
3073
3074         if (bond->curr_active_slave) {
3075                 bond_arp_send_all(bond, bond->curr_active_slave);
3076                 read_unlock(&bond->curr_slave_lock);
3077                 return;
3078         }
3079
3080         read_unlock(&bond->curr_slave_lock);
3081
3082         /* if we don't have a curr_active_slave, search for the next available
3083          * backup slave from the current_arp_slave and make it the candidate
3084          * for becoming the curr_active_slave
3085          */
3086
3087         if (!bond->current_arp_slave) {
3088                 bond->current_arp_slave = bond->first_slave;
3089                 if (!bond->current_arp_slave)
3090                         return;
3091         }
3092
3093         bond_set_slave_inactive_flags(bond->current_arp_slave);
3094
3095         /* search for next candidate */
3096         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3097                 if (IS_UP(slave->dev)) {
3098                         slave->link = BOND_LINK_BACK;
3099                         bond_set_slave_active_flags(slave);
3100                         bond_arp_send_all(bond, slave);
3101                         slave->jiffies = jiffies;
3102                         bond->current_arp_slave = slave;
3103                         break;
3104                 }
3105
3106                 /* if the link state is up at this point, we
3107                  * mark it down - this can happen if we have
3108                  * simultaneous link failures and
3109                  * reselect_active_interface doesn't make this
3110                  * one the current slave so it is still marked
3111                  * up when it is actually down
3112                  */
3113                 if (slave->link == BOND_LINK_UP) {
3114                         slave->link = BOND_LINK_DOWN;
3115                         if (slave->link_failure_count < UINT_MAX)
3116                                 slave->link_failure_count++;
3117
3118                         bond_set_slave_inactive_flags(slave);
3119
3120                         pr_info("%s: backup interface %s is now down.\n",
3121                                 bond->dev->name, slave->dev->name);
3122                 }
3123         }
3124 }
3125
3126 void bond_activebackup_arp_mon(struct work_struct *work)
3127 {
3128         struct bonding *bond = container_of(work, struct bonding,
3129                                             arp_work.work);
3130         bool should_notify_peers = false;
3131         int delta_in_ticks;
3132
3133         read_lock(&bond->lock);
3134
3135         if (bond->kill_timers)
3136                 goto out;
3137
3138         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3139
3140         if (bond->slave_cnt == 0)
3141                 goto re_arm;
3142
3143         should_notify_peers = bond_should_notify_peers(bond);
3144
3145         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3146                 read_unlock(&bond->lock);
3147                 rtnl_lock();
3148                 read_lock(&bond->lock);
3149
3150                 bond_ab_arp_commit(bond, delta_in_ticks);
3151
3152                 read_unlock(&bond->lock);
3153                 rtnl_unlock();
3154                 read_lock(&bond->lock);
3155         }
3156
3157         bond_ab_arp_probe(bond);
3158
3159 re_arm:
3160         if (bond->params.arp_interval && !bond->kill_timers)
3161                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3162 out:
3163         read_unlock(&bond->lock);
3164
3165         if (should_notify_peers) {
3166                 rtnl_lock();
3167                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3168                 rtnl_unlock();
3169         }
3170 }
3171
3172 /*-------------------------- netdev event handling --------------------------*/
3173
3174 /*
3175  * Change device name
3176  */
3177 static int bond_event_changename(struct bonding *bond)
3178 {
3179         bond_remove_proc_entry(bond);
3180         bond_create_proc_entry(bond);
3181
3182         bond_debug_reregister(bond);
3183
3184         return NOTIFY_DONE;
3185 }
3186
3187 static int bond_master_netdev_event(unsigned long event,
3188                                     struct net_device *bond_dev)
3189 {
3190         struct bonding *event_bond = netdev_priv(bond_dev);
3191
3192         switch (event) {
3193         case NETDEV_CHANGENAME:
3194                 return bond_event_changename(event_bond);
3195         default:
3196                 break;
3197         }
3198
3199         return NOTIFY_DONE;
3200 }
3201
3202 static int bond_slave_netdev_event(unsigned long event,
3203                                    struct net_device *slave_dev)
3204 {
3205         struct net_device *bond_dev = slave_dev->master;
3206         struct bonding *bond = netdev_priv(bond_dev);
3207
3208         switch (event) {
3209         case NETDEV_UNREGISTER:
3210                 if (bond_dev) {
3211                         if (bond->setup_by_slave)
3212                                 bond_release_and_destroy(bond_dev, slave_dev);
3213                         else
3214                                 bond_release(bond_dev, slave_dev);
3215                 }
3216                 break;
3217         case NETDEV_CHANGE:
3218                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3219                         struct slave *slave;
3220
3221                         slave = bond_get_slave_by_dev(bond, slave_dev);
3222                         if (slave) {
3223                                 u32 old_speed = slave->speed;
3224                                 u8  old_duplex = slave->duplex;
3225
3226                                 bond_update_speed_duplex(slave);
3227
3228                                 if (bond_is_lb(bond))
3229                                         break;
3230
3231                                 if (old_speed != slave->speed)
3232                                         bond_3ad_adapter_speed_changed(slave);
3233                                 if (old_duplex != slave->duplex)
3234                                         bond_3ad_adapter_duplex_changed(slave);
3235                         }
3236                 }
3237
3238                 break;
3239         case NETDEV_DOWN:
3240                 /*
3241                  * ... Or is it this?
3242                  */
3243                 break;
3244         case NETDEV_CHANGEMTU:
3245                 /*
3246                  * TODO: Should slaves be allowed to
3247                  * independently alter their MTU?  For
3248                  * an active-backup bond, slaves need
3249                  * not be the same type of device, so
3250                  * MTUs may vary.  For other modes,
3251                  * slaves arguably should have the
3252                  * same MTUs. To do this, we'd need to
3253                  * take over the slave's change_mtu
3254                  * function for the duration of their
3255                  * servitude.
3256                  */
3257                 break;
3258         case NETDEV_CHANGENAME:
3259                 /*
3260                  * TODO: handle changing the primary's name
3261                  */
3262                 break;
3263         case NETDEV_FEAT_CHANGE:
3264                 bond_compute_features(bond);
3265                 break;
3266         default:
3267                 break;
3268         }
3269
3270         return NOTIFY_DONE;
3271 }
3272
3273 /*
3274  * bond_netdev_event: handle netdev notifier chain events.
3275  *
3276  * This function receives events for the netdev chain.  The caller (an
3277  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3278  * locks for us to safely manipulate the slave devices (RTNL lock,
3279  * dev_probe_lock).
3280  */
3281 static int bond_netdev_event(struct notifier_block *this,
3282                              unsigned long event, void *ptr)
3283 {
3284         struct net_device *event_dev = (struct net_device *)ptr;
3285
3286         pr_debug("event_dev: %s, event: %lx\n",
3287                  event_dev ? event_dev->name : "None",
3288                  event);
3289
3290         if (!(event_dev->priv_flags & IFF_BONDING))
3291                 return NOTIFY_DONE;
3292
3293         if (event_dev->flags & IFF_MASTER) {
3294                 pr_debug("IFF_MASTER\n");
3295                 return bond_master_netdev_event(event, event_dev);
3296         }
3297
3298         if (event_dev->flags & IFF_SLAVE) {
3299                 pr_debug("IFF_SLAVE\n");
3300                 return bond_slave_netdev_event(event, event_dev);
3301         }
3302
3303         return NOTIFY_DONE;
3304 }
3305
3306 /*
3307  * bond_inetaddr_event: handle inetaddr notifier chain events.
3308  *
3309  * We keep track of device IPs primarily to use as source addresses in
3310  * ARP monitor probes (rather than spewing out broadcasts all the time).
3311  *
3312  * We track one IP for the main device (if it has one), plus one per VLAN.
3313  */
3314 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3315 {
3316         struct in_ifaddr *ifa = ptr;
3317         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3318         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3319         struct bonding *bond;
3320         struct vlan_entry *vlan;
3321
3322         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3323                 if (bond->dev == event_dev) {
3324                         switch (event) {
3325                         case NETDEV_UP:
3326                                 bond->master_ip = ifa->ifa_local;
3327                                 return NOTIFY_OK;
3328                         case NETDEV_DOWN:
3329                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3330                                 return NOTIFY_OK;
3331                         default:
3332                                 return NOTIFY_DONE;
3333                         }
3334                 }
3335
3336                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3337                         vlan_dev = __vlan_find_dev_deep(bond->dev,
3338                                                         vlan->vlan_id);
3339                         if (vlan_dev == event_dev) {
3340                                 switch (event) {
3341                                 case NETDEV_UP:
3342                                         vlan->vlan_ip = ifa->ifa_local;
3343                                         return NOTIFY_OK;
3344                                 case NETDEV_DOWN:
3345                                         vlan->vlan_ip =
3346                                                 bond_glean_dev_ip(vlan_dev);
3347                                         return NOTIFY_OK;
3348                                 default:
3349                                         return NOTIFY_DONE;
3350                                 }
3351                         }
3352                 }
3353         }
3354         return NOTIFY_DONE;
3355 }
3356
3357 static struct notifier_block bond_netdev_notifier = {
3358         .notifier_call = bond_netdev_event,
3359 };
3360
3361 static struct notifier_block bond_inetaddr_notifier = {
3362         .notifier_call = bond_inetaddr_event,
3363 };
3364
3365 /*---------------------------- Hashing Policies -----------------------------*/
3366
3367 /*
3368  * Hash for the output device based upon layer 2 and layer 3 data. If
3369  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3370  */
3371 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3372 {
3373         struct ethhdr *data = (struct ethhdr *)skb->data;
3374         struct iphdr *iph = ip_hdr(skb);
3375
3376         if (skb->protocol == htons(ETH_P_IP)) {
3377                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3378                         (data->h_dest[5] ^ data->h_source[5])) % count;
3379         }
3380
3381         return (data->h_dest[5] ^ data->h_source[5]) % count;
3382 }
3383
3384 /*
3385  * Hash for the output device based upon layer 3 and layer 4 data. If
3386  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3387  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3388  */
3389 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3390 {
3391         struct ethhdr *data = (struct ethhdr *)skb->data;
3392         struct iphdr *iph = ip_hdr(skb);
3393         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3394         int layer4_xor = 0;
3395
3396         if (skb->protocol == htons(ETH_P_IP)) {
3397                 if (!ip_is_fragment(iph) &&
3398                     (iph->protocol == IPPROTO_TCP ||
3399                      iph->protocol == IPPROTO_UDP)) {
3400                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3401                 }
3402                 return (layer4_xor ^
3403                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3404
3405         }
3406
3407         return (data->h_dest[5] ^ data->h_source[5]) % count;
3408 }
3409
3410 /*
3411  * Hash for the output device based upon layer 2 data
3412  */
3413 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3414 {
3415         struct ethhdr *data = (struct ethhdr *)skb->data;
3416
3417         return (data->h_dest[5] ^ data->h_source[5]) % count;
3418 }
3419
3420 /*-------------------------- Device entry points ----------------------------*/
3421
3422 static int bond_open(struct net_device *bond_dev)
3423 {
3424         struct bonding *bond = netdev_priv(bond_dev);
3425         struct slave *slave;
3426         int i;
3427
3428         bond->kill_timers = 0;
3429
3430         /* reset slave->backup and slave->inactive */
3431         read_lock(&bond->lock);
3432         if (bond->slave_cnt > 0) {
3433                 read_lock(&bond->curr_slave_lock);
3434                 bond_for_each_slave(bond, slave, i) {
3435                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3436                                 && (slave != bond->curr_active_slave)) {
3437                                 bond_set_slave_inactive_flags(slave);
3438                         } else {
3439                                 bond_set_slave_active_flags(slave);
3440                         }
3441                 }
3442                 read_unlock(&bond->curr_slave_lock);
3443         }
3444         read_unlock(&bond->lock);
3445
3446         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3447
3448         if (bond_is_lb(bond)) {
3449                 /* bond_alb_initialize must be called before the timer
3450                  * is started.
3451                  */
3452                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3453                         /* something went wrong - fail the open operation */
3454                         return -ENOMEM;
3455                 }
3456
3457                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3458                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3459         }
3460
3461         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3462                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3463                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3464         }
3465
3466         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3467                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3468                         INIT_DELAYED_WORK(&bond->arp_work,
3469                                           bond_activebackup_arp_mon);
3470                 else
3471                         INIT_DELAYED_WORK(&bond->arp_work,
3472                                           bond_loadbalance_arp_mon);
3473
3474                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3475                 if (bond->params.arp_validate)
3476                         bond->recv_probe = bond_arp_rcv;
3477         }
3478
3479         if (bond->params.mode == BOND_MODE_8023AD) {
3480                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3481                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3482                 /* register to receive LACPDUs */
3483                 bond->recv_probe = bond_3ad_lacpdu_recv;
3484                 bond_3ad_initiate_agg_selection(bond, 1);
3485         }
3486
3487         return 0;
3488 }
3489
3490 static int bond_close(struct net_device *bond_dev)
3491 {
3492         struct bonding *bond = netdev_priv(bond_dev);
3493
3494         write_lock_bh(&bond->lock);
3495
3496         bond->send_peer_notif = 0;
3497
3498         /* signal timers not to re-arm */
3499         bond->kill_timers = 1;
3500
3501         write_unlock_bh(&bond->lock);
3502
3503         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3504                 cancel_delayed_work(&bond->mii_work);
3505         }
3506
3507         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3508                 cancel_delayed_work(&bond->arp_work);
3509         }
3510
3511         switch (bond->params.mode) {
3512         case BOND_MODE_8023AD:
3513                 cancel_delayed_work(&bond->ad_work);
3514                 break;
3515         case BOND_MODE_TLB:
3516         case BOND_MODE_ALB:
3517                 cancel_delayed_work(&bond->alb_work);
3518                 break;
3519         default:
3520                 break;
3521         }
3522
3523         if (delayed_work_pending(&bond->mcast_work))
3524                 cancel_delayed_work(&bond->mcast_work);
3525
3526         if (bond_is_lb(bond)) {
3527                 /* Must be called only after all
3528                  * slaves have been released
3529                  */
3530                 bond_alb_deinitialize(bond);
3531         }
3532         bond->recv_probe = NULL;
3533
3534         return 0;
3535 }
3536
3537 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3538                                                 struct rtnl_link_stats64 *stats)
3539 {
3540         struct bonding *bond = netdev_priv(bond_dev);
3541         struct rtnl_link_stats64 temp;
3542         struct slave *slave;
3543         int i;
3544
3545         memset(stats, 0, sizeof(*stats));
3546
3547         read_lock_bh(&bond->lock);
3548
3549         bond_for_each_slave(bond, slave, i) {
3550                 const struct rtnl_link_stats64 *sstats =
3551                         dev_get_stats(slave->dev, &temp);
3552
3553                 stats->rx_packets += sstats->rx_packets;
3554                 stats->rx_bytes += sstats->rx_bytes;
3555                 stats->rx_errors += sstats->rx_errors;
3556                 stats->rx_dropped += sstats->rx_dropped;
3557
3558                 stats->tx_packets += sstats->tx_packets;
3559                 stats->tx_bytes += sstats->tx_bytes;
3560                 stats->tx_errors += sstats->tx_errors;
3561                 stats->tx_dropped += sstats->tx_dropped;
3562
3563                 stats->multicast += sstats->multicast;
3564                 stats->collisions += sstats->collisions;
3565
3566                 stats->rx_length_errors += sstats->rx_length_errors;
3567                 stats->rx_over_errors += sstats->rx_over_errors;
3568                 stats->rx_crc_errors += sstats->rx_crc_errors;
3569                 stats->rx_frame_errors += sstats->rx_frame_errors;
3570                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3571                 stats->rx_missed_errors += sstats->rx_missed_errors;
3572
3573                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3574                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3575                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3576                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3577                 stats->tx_window_errors += sstats->tx_window_errors;
3578         }
3579
3580         read_unlock_bh(&bond->lock);
3581
3582         return stats;
3583 }
3584
3585 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3586 {
3587         struct net_device *slave_dev = NULL;
3588         struct ifbond k_binfo;
3589         struct ifbond __user *u_binfo = NULL;
3590         struct ifslave k_sinfo;
3591         struct ifslave __user *u_sinfo = NULL;
3592         struct mii_ioctl_data *mii = NULL;
3593         int res = 0;
3594
3595         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3596
3597         switch (cmd) {
3598         case SIOCGMIIPHY:
3599                 mii = if_mii(ifr);
3600                 if (!mii)
3601                         return -EINVAL;
3602
3603                 mii->phy_id = 0;
3604                 /* Fall Through */
3605         case SIOCGMIIREG:
3606                 /*
3607                  * We do this again just in case we were called by SIOCGMIIREG
3608                  * instead of SIOCGMIIPHY.
3609                  */
3610                 mii = if_mii(ifr);
3611                 if (!mii)
3612                         return -EINVAL;
3613
3614
3615                 if (mii->reg_num == 1) {
3616                         struct bonding *bond = netdev_priv(bond_dev);
3617                         mii->val_out = 0;
3618                         read_lock(&bond->lock);
3619                         read_lock(&bond->curr_slave_lock);
3620                         if (netif_carrier_ok(bond->dev))
3621                                 mii->val_out = BMSR_LSTATUS;
3622
3623                         read_unlock(&bond->curr_slave_lock);
3624                         read_unlock(&bond->lock);
3625                 }
3626
3627                 return 0;
3628         case BOND_INFO_QUERY_OLD:
3629         case SIOCBONDINFOQUERY:
3630                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3631
3632                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3633                         return -EFAULT;
3634
3635                 res = bond_info_query(bond_dev, &k_binfo);
3636                 if (res == 0 &&
3637                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3638                         return -EFAULT;
3639
3640                 return res;
3641         case BOND_SLAVE_INFO_QUERY_OLD:
3642         case SIOCBONDSLAVEINFOQUERY:
3643                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3644
3645                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3646                         return -EFAULT;
3647
3648                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3649                 if (res == 0 &&
3650                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3651                         return -EFAULT;
3652
3653                 return res;
3654         default:
3655                 /* Go on */
3656                 break;
3657         }
3658
3659         if (!capable(CAP_NET_ADMIN))
3660                 return -EPERM;
3661
3662         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3663
3664         pr_debug("slave_dev=%p:\n", slave_dev);
3665
3666         if (!slave_dev)
3667                 res = -ENODEV;
3668         else {
3669                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3670                 switch (cmd) {
3671                 case BOND_ENSLAVE_OLD:
3672                 case SIOCBONDENSLAVE:
3673                         res = bond_enslave(bond_dev, slave_dev);
3674                         break;
3675                 case BOND_RELEASE_OLD:
3676                 case SIOCBONDRELEASE:
3677                         res = bond_release(bond_dev, slave_dev);
3678                         break;
3679                 case BOND_SETHWADDR_OLD:
3680                 case SIOCBONDSETHWADDR:
3681                         res = bond_sethwaddr(bond_dev, slave_dev);
3682                         break;
3683                 case BOND_CHANGE_ACTIVE_OLD:
3684                 case SIOCBONDCHANGEACTIVE:
3685                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3686                         break;
3687                 default:
3688                         res = -EOPNOTSUPP;
3689                 }
3690
3691                 dev_put(slave_dev);
3692         }
3693
3694         return res;
3695 }
3696
3697 static bool bond_addr_in_mc_list(unsigned char *addr,
3698                                  struct netdev_hw_addr_list *list,
3699                                  int addrlen)
3700 {
3701         struct netdev_hw_addr *ha;
3702
3703         netdev_hw_addr_list_for_each(ha, list)
3704                 if (!memcmp(ha->addr, addr, addrlen))
3705                         return true;
3706
3707         return false;
3708 }
3709
3710 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3711 {
3712         struct bonding *bond = netdev_priv(bond_dev);
3713
3714         if (change & IFF_PROMISC)
3715                 bond_set_promiscuity(bond,
3716                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3717
3718         if (change & IFF_ALLMULTI)
3719                 bond_set_allmulti(bond,
3720                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3721 }
3722
3723 static void bond_set_multicast_list(struct net_device *bond_dev)
3724 {
3725         struct bonding *bond = netdev_priv(bond_dev);
3726         struct netdev_hw_addr *ha;
3727         bool found;
3728
3729         read_lock(&bond->lock);
3730
3731         /* looking for addresses to add to slaves' mc list */
3732         netdev_for_each_mc_addr(ha, bond_dev) {
3733                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3734                                              bond_dev->addr_len);
3735                 if (!found)
3736                         bond_mc_add(bond, ha->addr);
3737         }
3738
3739         /* looking for addresses to delete from slaves' list */
3740         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3741                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3742                                              bond_dev->addr_len);
3743                 if (!found)
3744                         bond_mc_del(bond, ha->addr);
3745         }
3746
3747         /* save master's multicast list */
3748         __hw_addr_flush(&bond->mc_list);
3749         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3750                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3751
3752         read_unlock(&bond->lock);
3753 }
3754
3755 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3756 {
3757         struct bonding *bond = netdev_priv(dev);
3758         struct slave *slave = bond->first_slave;
3759
3760         if (slave) {
3761                 const struct net_device_ops *slave_ops
3762                         = slave->dev->netdev_ops;
3763                 if (slave_ops->ndo_neigh_setup)
3764                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3765         }
3766         return 0;
3767 }
3768
3769 /*
3770  * Change the MTU of all of a master's slaves to match the master
3771  */
3772 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3773 {
3774         struct bonding *bond = netdev_priv(bond_dev);
3775         struct slave *slave, *stop_at;
3776         int res = 0;
3777         int i;
3778
3779         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3780                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3781
3782         /* Can't hold bond->lock with bh disabled here since
3783          * some base drivers panic. On the other hand we can't
3784          * hold bond->lock without bh disabled because we'll
3785          * deadlock. The only solution is to rely on the fact
3786          * that we're under rtnl_lock here, and the slaves
3787          * list won't change. This doesn't solve the problem
3788          * of setting the slave's MTU while it is
3789          * transmitting, but the assumption is that the base
3790          * driver can handle that.
3791          *
3792          * TODO: figure out a way to safely iterate the slaves
3793          * list, but without holding a lock around the actual
3794          * call to the base driver.
3795          */
3796
3797         bond_for_each_slave(bond, slave, i) {
3798                 pr_debug("s %p s->p %p c_m %p\n",
3799                          slave,
3800                          slave->prev,
3801                          slave->dev->netdev_ops->ndo_change_mtu);
3802
3803                 res = dev_set_mtu(slave->dev, new_mtu);
3804
3805                 if (res) {
3806                         /* If we failed to set the slave's mtu to the new value
3807                          * we must abort the operation even in ACTIVE_BACKUP
3808                          * mode, because if we allow the backup slaves to have
3809                          * different mtu values than the active slave we'll
3810                          * need to change their mtu when doing a failover. That
3811                          * means changing their mtu from timer context, which
3812                          * is probably not a good idea.
3813                          */
3814                         pr_debug("err %d %s\n", res, slave->dev->name);
3815                         goto unwind;
3816                 }
3817         }
3818
3819         bond_dev->mtu = new_mtu;
3820
3821         return 0;
3822
3823 unwind:
3824         /* unwind from head to the slave that failed */
3825         stop_at = slave;
3826         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3827                 int tmp_res;
3828
3829                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3830                 if (tmp_res) {
3831                         pr_debug("unwind err %d dev %s\n",
3832                                  tmp_res, slave->dev->name);
3833                 }
3834         }
3835
3836         return res;
3837 }
3838
3839 /*
3840  * Change HW address
3841  *
3842  * Note that many devices must be down to change the HW address, and
3843  * downing the master releases all slaves.  We can make bonds full of
3844  * bonding devices to test this, however.
3845  */
3846 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3847 {
3848         struct bonding *bond = netdev_priv(bond_dev);
3849         struct sockaddr *sa = addr, tmp_sa;
3850         struct slave *slave, *stop_at;
3851         int res = 0;
3852         int i;
3853
3854         if (bond->params.mode == BOND_MODE_ALB)
3855                 return bond_alb_set_mac_address(bond_dev, addr);
3856
3857
3858         pr_debug("bond=%p, name=%s\n",
3859                  bond, bond_dev ? bond_dev->name : "None");
3860
3861         /*
3862          * If fail_over_mac is set to active, do nothing and return
3863          * success.  Returning an error causes ifenslave to fail.
3864          */
3865         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3866                 return 0;
3867
3868         if (!is_valid_ether_addr(sa->sa_data))
3869                 return -EADDRNOTAVAIL;
3870
3871         /* Can't hold bond->lock with bh disabled here since
3872          * some base drivers panic. On the other hand we can't
3873          * hold bond->lock without bh disabled because we'll
3874          * deadlock. The only solution is to rely on the fact
3875          * that we're under rtnl_lock here, and the slaves
3876          * list won't change. This doesn't solve the problem
3877          * of setting the slave's hw address while it is
3878          * transmitting, but the assumption is that the base
3879          * driver can handle that.
3880          *
3881          * TODO: figure out a way to safely iterate the slaves
3882          * list, but without holding a lock around the actual
3883          * call to the base driver.
3884          */
3885
3886         bond_for_each_slave(bond, slave, i) {
3887                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3888                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3889
3890                 if (slave_ops->ndo_set_mac_address == NULL) {
3891                         res = -EOPNOTSUPP;
3892                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3893                         goto unwind;
3894                 }
3895
3896                 res = dev_set_mac_address(slave->dev, addr);
3897                 if (res) {
3898                         /* TODO: consider downing the slave
3899                          * and retry ?
3900                          * User should expect communications
3901                          * breakage anyway until ARP finish
3902                          * updating, so...
3903                          */
3904                         pr_debug("err %d %s\n", res, slave->dev->name);
3905                         goto unwind;
3906                 }
3907         }
3908
3909         /* success */
3910         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3911         return 0;
3912
3913 unwind:
3914         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3915         tmp_sa.sa_family = bond_dev->type;
3916
3917         /* unwind from head to the slave that failed */
3918         stop_at = slave;
3919         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3920                 int tmp_res;
3921
3922                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3923                 if (tmp_res) {
3924                         pr_debug("unwind err %d dev %s\n",
3925                                  tmp_res, slave->dev->name);
3926                 }
3927         }
3928
3929         return res;
3930 }
3931
3932 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3933 {
3934         struct bonding *bond = netdev_priv(bond_dev);
3935         struct slave *slave, *start_at;
3936         int i, slave_no, res = 1;
3937         struct iphdr *iph = ip_hdr(skb);
3938
3939         /*
3940          * Start with the curr_active_slave that joined the bond as the
3941          * default for sending IGMP traffic.  For failover purposes one
3942          * needs to maintain some consistency for the interface that will
3943          * send the join/membership reports.  The curr_active_slave found
3944          * will send all of this type of traffic.
3945          */
3946         if ((iph->protocol == IPPROTO_IGMP) &&
3947             (skb->protocol == htons(ETH_P_IP))) {
3948
3949                 read_lock(&bond->curr_slave_lock);
3950                 slave = bond->curr_active_slave;
3951                 read_unlock(&bond->curr_slave_lock);
3952
3953                 if (!slave)
3954                         goto out;
3955         } else {
3956                 /*
3957                  * Concurrent TX may collide on rr_tx_counter; we accept
3958                  * that as being rare enough not to justify using an
3959                  * atomic op here.
3960                  */
3961                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3962
3963                 bond_for_each_slave(bond, slave, i) {
3964                         slave_no--;
3965                         if (slave_no < 0)
3966                                 break;
3967                 }
3968         }
3969
3970         start_at = slave;
3971         bond_for_each_slave_from(bond, slave, i, start_at) {
3972                 if (IS_UP(slave->dev) &&
3973                     (slave->link == BOND_LINK_UP) &&
3974                     bond_is_active_slave(slave)) {
3975                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3976                         break;
3977                 }
3978         }
3979
3980 out:
3981         if (res) {
3982                 /* no suitable interface, frame not sent */
3983                 dev_kfree_skb(skb);
3984         }
3985
3986         return NETDEV_TX_OK;
3987 }
3988
3989
3990 /*
3991  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3992  * the bond has a usable interface.
3993  */
3994 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3995 {
3996         struct bonding *bond = netdev_priv(bond_dev);
3997         int res = 1;
3998
3999         read_lock(&bond->curr_slave_lock);
4000
4001         if (bond->curr_active_slave)
4002                 res = bond_dev_queue_xmit(bond, skb,
4003                         bond->curr_active_slave->dev);
4004
4005         if (res)
4006                 /* no suitable interface, frame not sent */
4007                 dev_kfree_skb(skb);
4008
4009         read_unlock(&bond->curr_slave_lock);
4010
4011         return NETDEV_TX_OK;
4012 }
4013
4014 /*
4015  * In bond_xmit_xor() , we determine the output device by using a pre-
4016  * determined xmit_hash_policy(), If the selected device is not enabled,
4017  * find the next active slave.
4018  */
4019 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4020 {
4021         struct bonding *bond = netdev_priv(bond_dev);
4022         struct slave *slave, *start_at;
4023         int slave_no;
4024         int i;
4025         int res = 1;
4026
4027         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4028
4029         bond_for_each_slave(bond, slave, i) {
4030                 slave_no--;
4031                 if (slave_no < 0)
4032                         break;
4033         }
4034
4035         start_at = slave;
4036
4037         bond_for_each_slave_from(bond, slave, i, start_at) {
4038                 if (IS_UP(slave->dev) &&
4039                     (slave->link == BOND_LINK_UP) &&
4040                     bond_is_active_slave(slave)) {
4041                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4042                         break;
4043                 }
4044         }
4045
4046         if (res) {
4047                 /* no suitable interface, frame not sent */
4048                 dev_kfree_skb(skb);
4049         }
4050
4051         return NETDEV_TX_OK;
4052 }
4053
4054 /*
4055  * in broadcast mode, we send everything to all usable interfaces.
4056  */
4057 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4058 {
4059         struct bonding *bond = netdev_priv(bond_dev);
4060         struct slave *slave, *start_at;
4061         struct net_device *tx_dev = NULL;
4062         int i;
4063         int res = 1;
4064
4065         read_lock(&bond->curr_slave_lock);
4066         start_at = bond->curr_active_slave;
4067         read_unlock(&bond->curr_slave_lock);
4068
4069         if (!start_at)
4070                 goto out;
4071
4072         bond_for_each_slave_from(bond, slave, i, start_at) {
4073                 if (IS_UP(slave->dev) &&
4074                     (slave->link == BOND_LINK_UP) &&
4075                     bond_is_active_slave(slave)) {
4076                         if (tx_dev) {
4077                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4078                                 if (!skb2) {
4079                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4080                                                bond_dev->name);
4081                                         continue;
4082                                 }
4083
4084                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4085                                 if (res) {
4086                                         dev_kfree_skb(skb2);
4087                                         continue;
4088                                 }
4089                         }
4090                         tx_dev = slave->dev;
4091                 }
4092         }
4093
4094         if (tx_dev)
4095                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4096
4097 out:
4098         if (res)
4099                 /* no suitable interface, frame not sent */
4100                 dev_kfree_skb(skb);
4101
4102         /* frame sent to all suitable interfaces */
4103         return NETDEV_TX_OK;
4104 }
4105
4106 /*------------------------- Device initialization ---------------------------*/
4107
4108 static void bond_set_xmit_hash_policy(struct bonding *bond)
4109 {
4110         switch (bond->params.xmit_policy) {
4111         case BOND_XMIT_POLICY_LAYER23:
4112                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4113                 break;
4114         case BOND_XMIT_POLICY_LAYER34:
4115                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4116                 break;
4117         case BOND_XMIT_POLICY_LAYER2:
4118         default:
4119                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4120                 break;
4121         }
4122 }
4123
4124 /*
4125  * Lookup the slave that corresponds to a qid
4126  */
4127 static inline int bond_slave_override(struct bonding *bond,
4128                                       struct sk_buff *skb)
4129 {
4130         int i, res = 1;
4131         struct slave *slave = NULL;
4132         struct slave *check_slave;
4133
4134         if (!skb->queue_mapping)
4135                 return 1;
4136
4137         /* Find out if any slaves have the same mapping as this skb. */
4138         bond_for_each_slave(bond, check_slave, i) {
4139                 if (check_slave->queue_id == skb->queue_mapping) {
4140                         slave = check_slave;
4141                         break;
4142                 }
4143         }
4144
4145         /* If the slave isn't UP, use default transmit policy. */
4146         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4147             (slave->link == BOND_LINK_UP)) {
4148                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4149         }
4150
4151         return res;
4152 }
4153
4154
4155 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4156 {
4157         /*
4158          * This helper function exists to help dev_pick_tx get the correct
4159          * destination queue.  Using a helper function skips a call to
4160          * skb_tx_hash and will put the skbs in the queue we expect on their
4161          * way down to the bonding driver.
4162          */
4163         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4164
4165         /*
4166          * Save the original txq to restore before passing to the driver
4167          */
4168         bond_queue_mapping(skb) = skb->queue_mapping;
4169
4170         if (unlikely(txq >= dev->real_num_tx_queues)) {
4171                 do {
4172                         txq -= dev->real_num_tx_queues;
4173                 } while (txq >= dev->real_num_tx_queues);
4174         }
4175         return txq;
4176 }
4177
4178 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4179 {
4180         struct bonding *bond = netdev_priv(dev);
4181
4182         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4183                 if (!bond_slave_override(bond, skb))
4184                         return NETDEV_TX_OK;
4185         }
4186
4187         switch (bond->params.mode) {
4188         case BOND_MODE_ROUNDROBIN:
4189                 return bond_xmit_roundrobin(skb, dev);
4190         case BOND_MODE_ACTIVEBACKUP:
4191                 return bond_xmit_activebackup(skb, dev);
4192         case BOND_MODE_XOR:
4193                 return bond_xmit_xor(skb, dev);
4194         case BOND_MODE_BROADCAST:
4195                 return bond_xmit_broadcast(skb, dev);
4196         case BOND_MODE_8023AD:
4197                 return bond_3ad_xmit_xor(skb, dev);
4198         case BOND_MODE_ALB:
4199         case BOND_MODE_TLB:
4200                 return bond_alb_xmit(skb, dev);
4201         default:
4202                 /* Should never happen, mode already checked */
4203                 pr_err("%s: Error: Unknown bonding mode %d\n",
4204                        dev->name, bond->params.mode);
4205                 WARN_ON_ONCE(1);
4206                 dev_kfree_skb(skb);
4207                 return NETDEV_TX_OK;
4208         }
4209 }
4210
4211 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4212 {
4213         struct bonding *bond = netdev_priv(dev);
4214         netdev_tx_t ret = NETDEV_TX_OK;
4215
4216         /*
4217          * If we risk deadlock from transmitting this in the
4218          * netpoll path, tell netpoll to queue the frame for later tx
4219          */
4220         if (is_netpoll_tx_blocked(dev))
4221                 return NETDEV_TX_BUSY;
4222
4223         read_lock(&bond->lock);
4224
4225         if (bond->slave_cnt)
4226                 ret = __bond_start_xmit(skb, dev);
4227         else
4228                 dev_kfree_skb(skb);
4229
4230         read_unlock(&bond->lock);
4231
4232         return ret;
4233 }
4234
4235 /*
4236  * set bond mode specific net device operations
4237  */
4238 void bond_set_mode_ops(struct bonding *bond, int mode)
4239 {
4240         struct net_device *bond_dev = bond->dev;
4241
4242         switch (mode) {
4243         case BOND_MODE_ROUNDROBIN:
4244                 break;
4245         case BOND_MODE_ACTIVEBACKUP:
4246                 break;
4247         case BOND_MODE_XOR:
4248                 bond_set_xmit_hash_policy(bond);
4249                 break;
4250         case BOND_MODE_BROADCAST:
4251                 break;
4252         case BOND_MODE_8023AD:
4253                 bond_set_xmit_hash_policy(bond);
4254                 break;
4255         case BOND_MODE_ALB:
4256                 /* FALLTHRU */
4257         case BOND_MODE_TLB:
4258                 break;
4259         default:
4260                 /* Should never happen, mode already checked */
4261                 pr_err("%s: Error: Unknown bonding mode %d\n",
4262                        bond_dev->name, mode);
4263                 break;
4264         }
4265 }
4266
4267 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4268                                     struct ethtool_drvinfo *drvinfo)
4269 {
4270         strncpy(drvinfo->driver, DRV_NAME, 32);
4271         strncpy(drvinfo->version, DRV_VERSION, 32);
4272         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4273 }
4274
4275 static const struct ethtool_ops bond_ethtool_ops = {
4276         .get_drvinfo            = bond_ethtool_get_drvinfo,
4277         .get_link               = ethtool_op_get_link,
4278 };
4279
4280 static const struct net_device_ops bond_netdev_ops = {
4281         .ndo_init               = bond_init,
4282         .ndo_uninit             = bond_uninit,
4283         .ndo_open               = bond_open,
4284         .ndo_stop               = bond_close,
4285         .ndo_start_xmit         = bond_start_xmit,
4286         .ndo_select_queue       = bond_select_queue,
4287         .ndo_get_stats64        = bond_get_stats,
4288         .ndo_do_ioctl           = bond_do_ioctl,
4289         .ndo_change_rx_flags    = bond_change_rx_flags,
4290         .ndo_set_rx_mode        = bond_set_multicast_list,
4291         .ndo_change_mtu         = bond_change_mtu,
4292         .ndo_set_mac_address    = bond_set_mac_address,
4293         .ndo_neigh_setup        = bond_neigh_setup,
4294         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4295         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4296 #ifdef CONFIG_NET_POLL_CONTROLLER
4297         .ndo_netpoll_setup      = bond_netpoll_setup,
4298         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4299         .ndo_poll_controller    = bond_poll_controller,
4300 #endif
4301         .ndo_add_slave          = bond_enslave,
4302         .ndo_del_slave          = bond_release,
4303         .ndo_fix_features       = bond_fix_features,
4304 };
4305
4306 static void bond_destructor(struct net_device *bond_dev)
4307 {
4308         struct bonding *bond = netdev_priv(bond_dev);
4309         if (bond->wq)
4310                 destroy_workqueue(bond->wq);
4311         free_netdev(bond_dev);
4312 }
4313
4314 static void bond_setup(struct net_device *bond_dev)
4315 {
4316         struct bonding *bond = netdev_priv(bond_dev);
4317
4318         /* initialize rwlocks */
4319         rwlock_init(&bond->lock);
4320         rwlock_init(&bond->curr_slave_lock);
4321
4322         bond->params = bonding_defaults;
4323
4324         /* Initialize pointers */
4325         bond->dev = bond_dev;
4326         INIT_LIST_HEAD(&bond->vlan_list);
4327
4328         /* Initialize the device entry points */
4329         ether_setup(bond_dev);
4330         bond_dev->netdev_ops = &bond_netdev_ops;
4331         bond_dev->ethtool_ops = &bond_ethtool_ops;
4332         bond_set_mode_ops(bond, bond->params.mode);
4333
4334         bond_dev->destructor = bond_destructor;
4335
4336         /* Initialize the device options */
4337         bond_dev->tx_queue_len = 0;
4338         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4339         bond_dev->priv_flags |= IFF_BONDING;
4340         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4341
4342         /* At first, we block adding VLANs. That's the only way to
4343          * prevent problems that occur when adding VLANs over an
4344          * empty bond. The block will be removed once non-challenged
4345          * slaves are enslaved.
4346          */
4347         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4348
4349         /* don't acquire bond device's netif_tx_lock when
4350          * transmitting */
4351         bond_dev->features |= NETIF_F_LLTX;
4352
4353         /* By default, we declare the bond to be fully
4354          * VLAN hardware accelerated capable. Special
4355          * care is taken in the various xmit functions
4356          * when there are slaves that are not hw accel
4357          * capable
4358          */
4359
4360         bond_dev->hw_features = BOND_VLAN_FEATURES |
4361                                 NETIF_F_HW_VLAN_TX |
4362                                 NETIF_F_HW_VLAN_RX |
4363                                 NETIF_F_HW_VLAN_FILTER;
4364
4365         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4366         bond_dev->features |= bond_dev->hw_features;
4367 }
4368
4369 static void bond_work_cancel_all(struct bonding *bond)
4370 {
4371         write_lock_bh(&bond->lock);
4372         bond->kill_timers = 1;
4373         write_unlock_bh(&bond->lock);
4374
4375         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4376                 cancel_delayed_work(&bond->mii_work);
4377
4378         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4379                 cancel_delayed_work(&bond->arp_work);
4380
4381         if (bond->params.mode == BOND_MODE_ALB &&
4382             delayed_work_pending(&bond->alb_work))
4383                 cancel_delayed_work(&bond->alb_work);
4384
4385         if (bond->params.mode == BOND_MODE_8023AD &&
4386             delayed_work_pending(&bond->ad_work))
4387                 cancel_delayed_work(&bond->ad_work);
4388
4389         if (delayed_work_pending(&bond->mcast_work))
4390                 cancel_delayed_work(&bond->mcast_work);
4391 }
4392
4393 /*
4394 * Destroy a bonding device.
4395 * Must be under rtnl_lock when this function is called.
4396 */
4397 static void bond_uninit(struct net_device *bond_dev)
4398 {
4399         struct bonding *bond = netdev_priv(bond_dev);
4400         struct vlan_entry *vlan, *tmp;
4401
4402         bond_netpoll_cleanup(bond_dev);
4403
4404         /* Release the bonded slaves */
4405         bond_release_all(bond_dev);
4406
4407         list_del(&bond->bond_list);
4408
4409         bond_work_cancel_all(bond);
4410
4411         bond_remove_proc_entry(bond);
4412
4413         bond_debug_unregister(bond);
4414
4415         __hw_addr_flush(&bond->mc_list);
4416
4417         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4418                 list_del(&vlan->vlan_list);
4419                 kfree(vlan);
4420         }
4421 }
4422
4423 /*------------------------- Module initialization ---------------------------*/
4424
4425 /*
4426  * Convert string input module parms.  Accept either the
4427  * number of the mode or its string name.  A bit complicated because
4428  * some mode names are substrings of other names, and calls from sysfs
4429  * may have whitespace in the name (trailing newlines, for example).
4430  */
4431 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4432 {
4433         int modeint = -1, i, rv;
4434         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4435
4436         for (p = (char *)buf; *p; p++)
4437                 if (!(isdigit(*p) || isspace(*p)))
4438                         break;
4439
4440         if (*p)
4441                 rv = sscanf(buf, "%20s", modestr);
4442         else
4443                 rv = sscanf(buf, "%d", &modeint);
4444
4445         if (!rv)
4446                 return -1;
4447
4448         for (i = 0; tbl[i].modename; i++) {
4449                 if (modeint == tbl[i].mode)
4450                         return tbl[i].mode;
4451                 if (strcmp(modestr, tbl[i].modename) == 0)
4452                         return tbl[i].mode;
4453         }
4454
4455         return -1;
4456 }
4457
4458 static int bond_check_params(struct bond_params *params)
4459 {
4460         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4461
4462         /*
4463          * Convert string parameters.
4464          */
4465         if (mode) {
4466                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4467                 if (bond_mode == -1) {
4468                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4469                                mode == NULL ? "NULL" : mode);
4470                         return -EINVAL;
4471                 }
4472         }
4473
4474         if (xmit_hash_policy) {
4475                 if ((bond_mode != BOND_MODE_XOR) &&
4476                     (bond_mode != BOND_MODE_8023AD)) {
4477                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4478                                bond_mode_name(bond_mode));
4479                 } else {
4480                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4481                                                         xmit_hashtype_tbl);
4482                         if (xmit_hashtype == -1) {
4483                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4484                                        xmit_hash_policy == NULL ? "NULL" :
4485                                        xmit_hash_policy);
4486                                 return -EINVAL;
4487                         }
4488                 }
4489         }
4490
4491         if (lacp_rate) {
4492                 if (bond_mode != BOND_MODE_8023AD) {
4493                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4494                                 bond_mode_name(bond_mode));
4495                 } else {
4496                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4497                         if (lacp_fast == -1) {
4498                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4499                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4500                                 return -EINVAL;
4501                         }
4502                 }
4503         }
4504
4505         if (ad_select) {
4506                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4507                 if (params->ad_select == -1) {
4508                         pr_err("Error: Invalid ad_select \"%s\"\n",
4509                                ad_select == NULL ? "NULL" : ad_select);
4510                         return -EINVAL;
4511                 }
4512
4513                 if (bond_mode != BOND_MODE_8023AD) {
4514                         pr_warning("ad_select param only affects 802.3ad mode\n");
4515                 }
4516         } else {
4517                 params->ad_select = BOND_AD_STABLE;
4518         }
4519
4520         if (max_bonds < 0) {
4521                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4522                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4523                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4524         }
4525
4526         if (miimon < 0) {
4527                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4528                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4529                 miimon = BOND_LINK_MON_INTERV;
4530         }
4531
4532         if (updelay < 0) {
4533                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4534                            updelay, INT_MAX);
4535                 updelay = 0;
4536         }
4537
4538         if (downdelay < 0) {
4539                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4540                            downdelay, INT_MAX);
4541                 downdelay = 0;
4542         }
4543
4544         if ((use_carrier != 0) && (use_carrier != 1)) {
4545                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4546                            use_carrier);
4547                 use_carrier = 1;
4548         }
4549
4550         if (num_peer_notif < 0 || num_peer_notif > 255) {
4551                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4552                            num_peer_notif);
4553                 num_peer_notif = 1;
4554         }
4555
4556         /* reset values for 802.3ad */
4557         if (bond_mode == BOND_MODE_8023AD) {
4558                 if (!miimon) {
4559                         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");
4560                         pr_warning("Forcing miimon to 100msec\n");
4561                         miimon = 100;
4562                 }
4563         }
4564
4565         if (tx_queues < 1 || tx_queues > 255) {
4566                 pr_warning("Warning: tx_queues (%d) should be between "
4567                            "1 and 255, resetting to %d\n",
4568                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4569                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4570         }
4571
4572         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4573                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4574                            "not of valid value (0/1), so it was set to "
4575                            "0\n", all_slaves_active);
4576                 all_slaves_active = 0;
4577         }
4578
4579         if (resend_igmp < 0 || resend_igmp > 255) {
4580                 pr_warning("Warning: resend_igmp (%d) should be between "
4581                            "0 and 255, resetting to %d\n",
4582                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4583                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4584         }
4585
4586         /* reset values for TLB/ALB */
4587         if ((bond_mode == BOND_MODE_TLB) ||
4588             (bond_mode == BOND_MODE_ALB)) {
4589                 if (!miimon) {
4590                         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");
4591                         pr_warning("Forcing miimon to 100msec\n");
4592                         miimon = 100;
4593                 }
4594         }
4595
4596         if (bond_mode == BOND_MODE_ALB) {
4597                 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",
4598                           updelay);
4599         }
4600
4601         if (!miimon) {
4602                 if (updelay || downdelay) {
4603                         /* just warn the user the up/down delay will have
4604                          * no effect since miimon is zero...
4605                          */
4606                         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",
4607                                    updelay, downdelay);
4608                 }
4609         } else {
4610                 /* don't allow arp monitoring */
4611                 if (arp_interval) {
4612                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4613                                    miimon, arp_interval);
4614                         arp_interval = 0;
4615                 }
4616
4617                 if ((updelay % miimon) != 0) {
4618                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4619                                    updelay, miimon,
4620                                    (updelay / miimon) * miimon);
4621                 }
4622
4623                 updelay /= miimon;
4624
4625                 if ((downdelay % miimon) != 0) {
4626                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4627                                    downdelay, miimon,
4628                                    (downdelay / miimon) * miimon);
4629                 }
4630
4631                 downdelay /= miimon;
4632         }
4633
4634         if (arp_interval < 0) {
4635                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4636                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4637                 arp_interval = BOND_LINK_ARP_INTERV;
4638         }
4639
4640         for (arp_ip_count = 0;
4641              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4642              arp_ip_count++) {
4643                 /* not complete check, but should be good enough to
4644                    catch mistakes */
4645                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4646                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4647                                    arp_ip_target[arp_ip_count]);
4648                         arp_interval = 0;
4649                 } else {
4650                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4651                         arp_target[arp_ip_count] = ip;
4652                 }
4653         }
4654
4655         if (arp_interval && !arp_ip_count) {
4656                 /* don't allow arping if no arp_ip_target given... */
4657                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4658                            arp_interval);
4659                 arp_interval = 0;
4660         }
4661
4662         if (arp_validate) {
4663                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4664                         pr_err("arp_validate only supported in active-backup mode\n");
4665                         return -EINVAL;
4666                 }
4667                 if (!arp_interval) {
4668                         pr_err("arp_validate requires arp_interval\n");
4669                         return -EINVAL;
4670                 }
4671
4672                 arp_validate_value = bond_parse_parm(arp_validate,
4673                                                      arp_validate_tbl);
4674                 if (arp_validate_value == -1) {
4675                         pr_err("Error: invalid arp_validate \"%s\"\n",
4676                                arp_validate == NULL ? "NULL" : arp_validate);
4677                         return -EINVAL;
4678                 }
4679         } else
4680                 arp_validate_value = 0;
4681
4682         if (miimon) {
4683                 pr_info("MII link monitoring set to %d ms\n", miimon);
4684         } else if (arp_interval) {
4685                 int i;
4686
4687                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4688                         arp_interval,
4689                         arp_validate_tbl[arp_validate_value].modename,
4690                         arp_ip_count);
4691
4692                 for (i = 0; i < arp_ip_count; i++)
4693                         pr_info(" %s", arp_ip_target[i]);
4694
4695                 pr_info("\n");
4696
4697         } else if (max_bonds) {
4698                 /* miimon and arp_interval not set, we need one so things
4699                  * work as expected, see bonding.txt for details
4700                  */
4701                 pr_debug("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");
4702         }
4703
4704         if (primary && !USES_PRIMARY(bond_mode)) {
4705                 /* currently, using a primary only makes sense
4706                  * in active backup, TLB or ALB modes
4707                  */
4708                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4709                            primary, bond_mode_name(bond_mode));
4710                 primary = NULL;
4711         }
4712
4713         if (primary && primary_reselect) {
4714                 primary_reselect_value = bond_parse_parm(primary_reselect,
4715                                                          pri_reselect_tbl);
4716                 if (primary_reselect_value == -1) {
4717                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4718                                primary_reselect ==
4719                                         NULL ? "NULL" : primary_reselect);
4720                         return -EINVAL;
4721                 }
4722         } else {
4723                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4724         }
4725
4726         if (fail_over_mac) {
4727                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4728                                                       fail_over_mac_tbl);
4729                 if (fail_over_mac_value == -1) {
4730                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4731                                arp_validate == NULL ? "NULL" : arp_validate);
4732                         return -EINVAL;
4733                 }
4734
4735                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4736                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4737         } else {
4738                 fail_over_mac_value = BOND_FOM_NONE;
4739         }
4740
4741         /* fill params struct with the proper values */
4742         params->mode = bond_mode;
4743         params->xmit_policy = xmit_hashtype;
4744         params->miimon = miimon;
4745         params->num_peer_notif = num_peer_notif;
4746         params->arp_interval = arp_interval;
4747         params->arp_validate = arp_validate_value;
4748         params->updelay = updelay;
4749         params->downdelay = downdelay;
4750         params->use_carrier = use_carrier;
4751         params->lacp_fast = lacp_fast;
4752         params->primary[0] = 0;
4753         params->primary_reselect = primary_reselect_value;
4754         params->fail_over_mac = fail_over_mac_value;
4755         params->tx_queues = tx_queues;
4756         params->all_slaves_active = all_slaves_active;
4757         params->resend_igmp = resend_igmp;
4758         params->min_links = min_links;
4759
4760         if (primary) {
4761                 strncpy(params->primary, primary, IFNAMSIZ);
4762                 params->primary[IFNAMSIZ - 1] = 0;
4763         }
4764
4765         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4766
4767         return 0;
4768 }
4769
4770 static struct lock_class_key bonding_netdev_xmit_lock_key;
4771 static struct lock_class_key bonding_netdev_addr_lock_key;
4772
4773 static void bond_set_lockdep_class_one(struct net_device *dev,
4774                                        struct netdev_queue *txq,
4775                                        void *_unused)
4776 {
4777         lockdep_set_class(&txq->_xmit_lock,
4778                           &bonding_netdev_xmit_lock_key);
4779 }
4780
4781 static void bond_set_lockdep_class(struct net_device *dev)
4782 {
4783         lockdep_set_class(&dev->addr_list_lock,
4784                           &bonding_netdev_addr_lock_key);
4785         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4786 }
4787
4788 /*
4789  * Called from registration process
4790  */
4791 static int bond_init(struct net_device *bond_dev)
4792 {
4793         struct bonding *bond = netdev_priv(bond_dev);
4794         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4795         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4796
4797         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4798
4799         /*
4800          * Initialize locks that may be required during
4801          * en/deslave operations.  All of the bond_open work
4802          * (of which this is part) should really be moved to
4803          * a phase prior to dev_open
4804          */
4805         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4806         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4807
4808         bond->wq = create_singlethread_workqueue(bond_dev->name);
4809         if (!bond->wq)
4810                 return -ENOMEM;
4811
4812         bond_set_lockdep_class(bond_dev);
4813
4814         bond_create_proc_entry(bond);
4815         list_add_tail(&bond->bond_list, &bn->dev_list);
4816
4817         bond_prepare_sysfs_group(bond);
4818
4819         bond_debug_register(bond);
4820
4821         __hw_addr_init(&bond->mc_list);
4822         return 0;
4823 }
4824
4825 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4826 {
4827         if (tb[IFLA_ADDRESS]) {
4828                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4829                         return -EINVAL;
4830                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4831                         return -EADDRNOTAVAIL;
4832         }
4833         return 0;
4834 }
4835
4836 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[],
4837                               unsigned int *num_queues,
4838                               unsigned int *real_num_queues)
4839 {
4840         *num_queues = tx_queues;
4841         return 0;
4842 }
4843
4844 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4845         .kind           = "bond",
4846         .priv_size      = sizeof(struct bonding),
4847         .setup          = bond_setup,
4848         .validate       = bond_validate,
4849         .get_tx_queues  = bond_get_tx_queues,
4850 };
4851
4852 /* Create a new bond based on the specified name and bonding parameters.
4853  * If name is NULL, obtain a suitable "bond%d" name for us.
4854  * Caller must NOT hold rtnl_lock; we need to release it here before we
4855  * set up our sysfs entries.
4856  */
4857 int bond_create(struct net *net, const char *name)
4858 {
4859         struct net_device *bond_dev;
4860         int res;
4861
4862         rtnl_lock();
4863
4864         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4865                                    name ? name : "bond%d",
4866                                    bond_setup, tx_queues);
4867         if (!bond_dev) {
4868                 pr_err("%s: eek! can't alloc netdev!\n", name);
4869                 rtnl_unlock();
4870                 return -ENOMEM;
4871         }
4872
4873         dev_net_set(bond_dev, net);
4874         bond_dev->rtnl_link_ops = &bond_link_ops;
4875
4876         res = register_netdevice(bond_dev);
4877
4878         netif_carrier_off(bond_dev);
4879
4880         rtnl_unlock();
4881         if (res < 0)
4882                 bond_destructor(bond_dev);
4883         return res;
4884 }
4885
4886 static int __net_init bond_net_init(struct net *net)
4887 {
4888         struct bond_net *bn = net_generic(net, bond_net_id);
4889
4890         bn->net = net;
4891         INIT_LIST_HEAD(&bn->dev_list);
4892
4893         bond_create_proc_dir(bn);
4894         bond_create_sysfs(bn);
4895         
4896         return 0;
4897 }
4898
4899 static void __net_exit bond_net_exit(struct net *net)
4900 {
4901         struct bond_net *bn = net_generic(net, bond_net_id);
4902
4903         bond_destroy_sysfs(bn);
4904         bond_destroy_proc_dir(bn);
4905 }
4906
4907 static struct pernet_operations bond_net_ops = {
4908         .init = bond_net_init,
4909         .exit = bond_net_exit,
4910         .id   = &bond_net_id,
4911         .size = sizeof(struct bond_net),
4912 };
4913
4914 static int __init bonding_init(void)
4915 {
4916         int i;
4917         int res;
4918
4919         pr_info("%s", bond_version);
4920
4921         res = bond_check_params(&bonding_defaults);
4922         if (res)
4923                 goto out;
4924
4925         res = register_pernet_subsys(&bond_net_ops);
4926         if (res)
4927                 goto out;
4928
4929         res = rtnl_link_register(&bond_link_ops);
4930         if (res)
4931                 goto err_link;
4932
4933         bond_create_debugfs();
4934
4935         for (i = 0; i < max_bonds; i++) {
4936                 res = bond_create(&init_net, NULL);
4937                 if (res)
4938                         goto err;
4939         }
4940
4941         register_netdevice_notifier(&bond_netdev_notifier);
4942         register_inetaddr_notifier(&bond_inetaddr_notifier);
4943 out:
4944         return res;
4945 err:
4946         rtnl_link_unregister(&bond_link_ops);
4947 err_link:
4948         unregister_pernet_subsys(&bond_net_ops);
4949         goto out;
4950
4951 }
4952
4953 static void __exit bonding_exit(void)
4954 {
4955         unregister_netdevice_notifier(&bond_netdev_notifier);
4956         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4957
4958         bond_destroy_debugfs();
4959
4960         rtnl_link_unregister(&bond_link_ops);
4961         unregister_pernet_subsys(&bond_net_ops);
4962
4963 #ifdef CONFIG_NET_POLL_CONTROLLER
4964         /*
4965          * Make sure we don't have an imbalance on our netpoll blocking
4966          */
4967         WARN_ON(atomic_read(&netpoll_block_tx));
4968 #endif
4969 }
4970
4971 module_init(bonding_init);
4972 module_exit(bonding_exit);
4973 MODULE_LICENSE("GPL");
4974 MODULE_VERSION(DRV_VERSION);
4975 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4976 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4977 MODULE_ALIAS_RTNL_LINK("bond");