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