2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
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.
24 * will release all slaves, marking them as down.
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.
32 * v0.1 - first working version.
33 * v0.2 - changed stats to be calculated by summing slaves stats.
36 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
37 * - fix leaks on failure at bond_init
39 * 2000/09/30 - Willy Tarreau <willy at meta-x.org>
40 * - added trivial code to release a slave device.
41 * - fixed security bug (CAP_NET_ADMIN not checked)
42 * - implemented MII link monitoring to disable dead links :
43 * All MII capable slaves are checked every <miimon> milliseconds
44 * (100 ms seems good). This value can be changed by passing it to
45 * insmod. A value of zero disables the monitoring (default).
46 * - fixed an infinite loop in bond_xmit_roundrobin() when there's no
48 * - made the code hopefully SMP safe
50 * 2000/10/03 - Willy Tarreau <willy at meta-x.org>
51 * - optimized slave lists based on relevant suggestions from Thomas Davis
52 * - implemented active-backup method to obtain HA with two switches:
53 * stay as long as possible on the same active interface, while we
54 * also monitor the backup one (MII link status) because we want to know
55 * if we are able to switch at any time. ( pass "mode=1" to insmod )
56 * - lots of stress testings because we need it to be more robust than the
59 * 2000/10/09 - Willy Tarreau <willy at meta-x.org>
60 * - added up and down delays after link state change.
61 * - optimized the slaves chaining so that when we run forward, we never
62 * repass through the bond itself, but we can find it by searching
63 * backwards. Renders the deletion more difficult, but accelerates the
65 * - smarter enslaving and releasing.
66 * - finer and more robust SMP locking
68 * 2000/10/17 - Willy Tarreau <willy at meta-x.org>
69 * - fixed two potential SMP race conditions
71 * 2000/10/18 - Willy Tarreau <willy at meta-x.org>
72 * - small fixes to the monitoring FSM in case of zero delays
73 * 2000/11/01 - Willy Tarreau <willy at meta-x.org>
74 * - fixed first slave not automatically used in trunk mode.
75 * 2000/11/10 : spelling of "EtherChannel" corrected.
76 * 2000/11/13 : fixed a race condition in case of concurrent accesses to ioctl().
77 * 2000/12/16 : fixed improper usage of rtnl_exlock_nowait().
79 * 2001/1/3 - Chad N. Tindel <ctindel at ieee dot org>
80 * - The bonding driver now simulates MII status monitoring, just like
81 * a normal network device. It will show that the link is down iff
82 * every slave in the bond shows that their links are down. If at least
83 * one slave is up, the bond's MII status will appear as up.
85 * 2001/2/7 - Chad N. Tindel <ctindel at ieee dot org>
86 * - Applications can now query the bond from user space to get
87 * information which may be useful. They do this by calling
88 * the BOND_INFO_QUERY ioctl. Once the app knows how many slaves
89 * are in the bond, it can call the BOND_SLAVE_INFO_QUERY ioctl to
90 * get slave specific information (# link failures, etc). See
91 * <linux/if_bonding.h> for more details. The structs of interest
92 * are ifbond and ifslave.
94 * 2001/4/5 - Chad N. Tindel <ctindel at ieee dot org>
95 * - Ported to 2.4 Kernel
97 * 2001/5/2 - Jeffrey E. Mast <jeff at mastfamily dot com>
98 * - When a device is detached from a bond, the slave device is no longer
99 * left thinking that is has a master.
101 * 2001/5/16 - Jeffrey E. Mast <jeff at mastfamily dot com>
102 * - memset did not appropriately initialized the bond rw_locks. Used
103 * rwlock_init to initialize to unlocked state to prevent deadlock when
104 * first attempting a lock
105 * - Called SET_MODULE_OWNER for bond device
107 * 2001/5/17 - Tim Anderson <tsa at mvista.com>
108 * - 2 paths for releasing for slave release; 1 through ioctl
109 * and 2) through close. Both paths need to release the same way.
110 * - the free slave in bond release is changing slave status before
111 * the free. The netdev_set_master() is intended to change slave state
112 * so it should not be done as part of the release process.
113 * - Simple rule for slave state at release: only the active in A/B and
114 * only one in the trunked case.
116 * 2001/6/01 - Tim Anderson <tsa at mvista.com>
117 * - Now call dev_close when releasing a slave so it doesn't screw up
120 * 2001/6/01 - Chad N. Tindel <ctindel at ieee dot org>
121 * - Added /proc support for getting bond and slave information.
122 * Information is in /proc/net/<bond device>/info.
123 * - Changed the locking when calling bond_close to prevent deadlock.
125 * 2001/8/05 - Janice Girouard <girouard at us.ibm.com>
126 * - correct problem where refcnt of slave is not incremented in bond_ioctl
127 * so the system hangs when halting.
128 * - correct locking problem when unable to malloc in bond_enslave.
129 * - adding bond_xmit_xor logic.
130 * - adding multiple bond device support.
132 * 2001/8/13 - Erik Habbinga <erik_habbinga at hp dot com>
133 * - correct locking problem with rtnl_exlock_nowait
135 * 2001/8/23 - Janice Girouard <girouard at us.ibm.com>
136 * - bzero initial dev_bonds, to correct oops
137 * - convert SIOCDEVPRIVATE to new MII ioctl calls
139 * 2001/9/13 - Takao Indoh <indou dot takao at jp dot fujitsu dot com>
140 * - Add the BOND_CHANGE_ACTIVE ioctl implementation
142 * 2001/9/14 - Mark Huth <mhuth at mvista dot com>
143 * - Change MII_LINK_READY to not check for end of auto-negotiation,
144 * but only for an up link.
146 * 2001/9/20 - Chad N. Tindel <ctindel at ieee dot org>
147 * - Add the device field to bonding_t. Previously the net_device
148 * corresponding to a bond wasn't available from the bonding_t
151 * 2001/9/25 - Janice Girouard <girouard at us.ibm.com>
152 * - add arp_monitor for active backup mode
154 * 2001/10/23 - Takao Indoh <indou dot takao at jp dot fujitsu dot com>
155 * - Various memory leak fixes
157 * 2001/11/5 - Mark Huth <mark dot huth at mvista dot com>
158 * - Don't take rtnl lock in bond_mii_monitor as it deadlocks under
159 * certain hotswap conditions.
160 * Note: this same change may be required in bond_arp_monitor ???
161 * - Remove possibility of calling bond_sethwaddr with NULL slave_dev ptr
162 * - Handle hot swap ethernet interface deregistration events to remove
163 * kernel oops following hot swap of enslaved interface
165 * 2002/1/2 - Chad N. Tindel <ctindel at ieee dot org>
166 * - Restore original slave flags at release time.
168 * 2002/02/18 - Erik Habbinga <erik_habbinga at hp dot com>
169 * - bond_release(): calling kfree on our_slave after call to
170 * bond_restore_slave_flags, not before
171 * - bond_enslave(): saving slave flags into original_flags before
172 * call to netdev_set_master, so the IFF_SLAVE flag doesn't end
173 * up in original_flags
175 * 2002/04/05 - Mark Smith <mark.smith at comdev dot cc> and
176 * Steve Mead <steve.mead at comdev dot cc>
177 * - Port Gleb Natapov's multicast support patchs from 2.4.12
178 * to 2.4.18 adding support for multicast.
180 * 2002/06/10 - Tony Cureington <tony.cureington * hp_com>
181 * - corrected uninitialized pointer (ifr.ifr_data) in bond_check_dev_link;
182 * actually changed function to use MIIPHY, then MIIREG, and finally
183 * ETHTOOL to determine the link status
184 * - fixed bad ifr_data pointer assignments in bond_ioctl
185 * - corrected mode 1 being reported as active-backup in bond_get_info;
186 * also added text to distinguish type of load balancing (rr or xor)
187 * - change arp_ip_target module param from "1-12s" (array of 12 ptrs)
188 * to "s" (a single ptr)
190 * 2002/08/30 - Jay Vosburgh <fubar at us dot ibm dot com>
191 * - Removed acquisition of xmit_lock in set_multicast_list; caused
192 * deadlock on SMP (lock is held by caller).
193 * - Revamped SIOCGMIIPHY, SIOCGMIIREG portion of bond_check_dev_link().
195 * 2002/09/18 - Jay Vosburgh <fubar at us dot ibm dot com>
196 * - Fixed up bond_check_dev_link() (and callers): removed some magic
197 * numbers, banished local MII_ defines, wrapped ioctl calls to
198 * prevent EFAULT errors
200 * 2002/9/30 - Jay Vosburgh <fubar at us dot ibm dot com>
201 * - make sure the ip target matches the arp_target before saving the
204 * 2002/9/30 - Dan Eisner <eisner at 2robots dot com>
205 * - make sure my_ip is set before taking down the link, since
206 * not all switches respond if the source ip is not set.
208 * 2002/10/8 - Janice Girouard <girouard at us dot ibm dot com>
209 * - read in the local ip address when enslaving a device
210 * - add primary support
211 * - make sure 2*arp_interval has passed when a new device
212 * is brought on-line before taking it down.
214 * 2002/09/11 - Philippe De Muyter <phdm at macqel dot be>
215 * - Added bond_xmit_broadcast logic.
216 * - Added bond_mode() support function.
218 * 2002/10/26 - Laurent Deniel <laurent.deniel at free.fr>
219 * - allow to register multicast addresses only on active slave
220 * (useful in active-backup mode)
221 * - add multicast module parameter
222 * - fix deletion of multicast groups after unloading module
224 * 2002/11/06 - Kameshwara Rayaprolu <kameshwara.rao * wipro_com>
225 * - Changes to prevent panic from closing the device twice; if we close
226 * the device in bond_release, we must set the original_flags to down
227 * so it won't be closed again by the network layer.
229 * 2002/11/07 - Tony Cureington <tony.cureington * hp_com>
230 * - Fix arp_target_hw_addr memory leak
231 * - Created activebackup_arp_monitor function to handle arp monitoring
232 * in active backup mode - the bond_arp_monitor had several problems...
233 * such as allowing slaves to tx arps sequentially without any delay
235 * - Renamed bond_arp_monitor to loadbalance_arp_monitor and re-wrote
236 * this function to just handle arp monitoring in load-balancing mode;
237 * it is a lot more compact now
238 * - Changes to ensure one and only one slave transmits in active-backup
240 * - Robustesize parameters; warn users about bad combinations of
241 * parameters; also if miimon is specified and a network driver does
242 * not support MII or ETHTOOL, inform the user of this
243 * - Changes to support link_failure_count when in arp monitoring mode
244 * - Fix up/down delay reported in /proc
245 * - Added version; log version; make version available from "modinfo -d"
246 * - Fixed problem in bond_check_dev_link - if the first IOCTL (SIOCGMIIPH)
247 * failed, the ETHTOOL ioctl never got a chance
249 * 2002/11/16 - Laurent Deniel <laurent.deniel at free.fr>
250 * - fix multicast handling in activebackup_arp_monitor
251 * - remove one unnecessary and confusing curr_active_slave == slave test
252 * in activebackup_arp_monitor
254 * 2002/11/17 - Laurent Deniel <laurent.deniel at free.fr>
255 * - fix bond_slave_info_query when slave_id = num_slaves
257 * 2002/11/19 - Janice Girouard <girouard at us dot ibm dot com>
258 * - correct ifr_data reference. Update ifr_data reference
259 * to mii_ioctl_data struct values to avoid confusion.
261 * 2002/11/22 - Bert Barbe <bert.barbe at oracle dot com>
262 * - Add support for multiple arp_ip_target
264 * 2002/12/13 - Jay Vosburgh <fubar at us dot ibm dot com>
265 * - Changed to allow text strings for mode and multicast, e.g.,
266 * insmod bonding mode=active-backup. The numbers still work.
267 * One change: an invalid choice will cause module load failure,
268 * rather than the previous behavior of just picking one.
269 * - Minor cleanups; got rid of dup ctype stuff, atoi function
271 * 2003/02/07 - Jay Vosburgh <fubar at us dot ibm dot com>
272 * - Added use_carrier module parameter that causes miimon to
273 * use netif_carrier_ok() test instead of MII/ETHTOOL ioctls.
274 * - Minor cleanups; consolidated ioctl calls to one function.
276 * 2003/02/07 - Tony Cureington <tony.cureington * hp_com>
277 * - Fix bond_mii_monitor() logic error that could result in
278 * bonding round-robin mode ignoring links after failover/recovery
280 * 2003/03/17 - Jay Vosburgh <fubar at us dot ibm dot com>
281 * - kmalloc fix (GFP_KERNEL to GFP_ATOMIC) reported by
282 * Shmulik dot Hen at intel.com.
283 * - Based on discussion on mailing list, changed use of
284 * update_slave_cnt(), created wrapper functions for adding/removing
285 * slaves, changed bond_xmit_xor() to check slave_cnt instead of
286 * checking slave and slave->dev (which only worked by accident).
287 * - Misc code cleanup: get arp_send() prototype from header file,
288 * add max_bonds to bonding.txt.
290 * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
291 * Shmulik Hen <shmulik.hen at intel dot com>
292 * - Make sure only bond_attach_slave() and bond_detach_slave() can
293 * manipulate the slave list, including slave_cnt, even when in
294 * bond_release_all().
295 * - Fixed hang in bond_release() with traffic running:
296 * netdev_set_master() must not be called from within the bond lock.
298 * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
299 * Shmulik Hen <shmulik.hen at intel dot com>
300 * - Fixed hang in bond_enslave() with traffic running:
301 * netdev_set_master() must not be called from within the bond lock.
303 * 2003/03/18 - Amir Noam <amir.noam at intel dot com>
304 * - Added support for getting slave's speed and duplex via ethtool.
305 * Needed for 802.3ad and other future modes.
307 * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
308 * Shmulik Hen <shmulik.hen at intel dot com>
309 * - Enable support of modes that need to use the unique mac address of
311 * * bond_enslave(): Moved setting the slave's mac address, and
312 * openning it, from the application to the driver. This breaks
313 * backward comaptibility with old versions of ifenslave that open
314 * the slave before enalsving it !!!.
315 * * bond_release(): The driver also takes care of closing the slave
316 * and restoring its original mac address.
317 * - Removed the code that restores all base driver's flags.
318 * Flags are automatically restored once all undo stages are done
320 * - Block possibility of enslaving before the master is up. This
321 * prevents putting the system in an unstable state.
323 * 2003/03/18 - Amir Noam <amir.noam at intel dot com>,
324 * Tsippy Mendelson <tsippy.mendelson at intel dot com> and
325 * Shmulik Hen <shmulik.hen at intel dot com>
326 * - Added support for IEEE 802.3ad Dynamic link aggregation mode.
328 * 2003/05/01 - Amir Noam <amir.noam at intel dot com>
329 * - Added ABI version control to restore compatibility between
330 * new/old ifenslave and new/old bonding.
332 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
333 * - Fixed bug in bond_release_all(): save old value of curr_active_slave
334 * before setting it to NULL.
335 * - Changed driver versioning scheme to include version number instead
336 * of release date (that is already in another field). There are 3
337 * fields X.Y.Z where:
338 * X - Major version - big behavior changes
339 * Y - Minor version - addition of features
340 * Z - Extra version - minor changes and bug fixes
341 * The current version is 1.0.0 as a base line.
343 * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
344 * Amir Noam <amir.noam at intel dot com>
345 * - Added support for lacp_rate module param.
346 * - Code beautification and style changes (mainly in comments).
347 * new version - 1.0.1
349 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
350 * - Based on discussion on mailing list, changed locking scheme
351 * to use lock/unlock or lock_bh/unlock_bh appropriately instead
352 * of lock_irqsave/unlock_irqrestore. The new scheme helps exposing
353 * hidden bugs and solves system hangs that occurred due to the fact
354 * that holding lock_irqsave doesn't prevent softirqs from running.
355 * This also increases total throughput since interrupts are not
356 * blocked on each transmitted packets or monitor timeout.
357 * new version - 2.0.0
359 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
360 * - Added support for Transmit load balancing mode.
361 * - Concentrate all assignments of curr_active_slave to a single point
362 * so specific modes can take actions when the primary adapter is
364 * - Take the updelay parameter into consideration during bond_enslave
365 * since some adapters loose their link during setting the device.
366 * - Renamed bond_3ad_link_status_changed() to
367 * bond_3ad_handle_link_change() for compatibility with TLB.
368 * new version - 2.1.0
370 * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com>
371 * - Added support for Adaptive load balancing mode which is
372 * equivalent to Transmit load balancing + Receive load balancing.
373 * new version - 2.2.0
375 * 2003/05/15 - Jay Vosburgh <fubar at us dot ibm dot com>
376 * - Applied fix to activebackup_arp_monitor posted to bonding-devel
377 * by Tony Cureington <tony.cureington * hp_com>. Fixes ARP
378 * monitor endless failover bug. Version to 2.2.10
380 * 2003/05/20 - Amir Noam <amir.noam at intel dot com>
381 * - Fixed bug in ABI version control - Don't commit to a specific
382 * ABI version if receiving unsupported ioctl commands.
384 * 2003/05/22 - Jay Vosburgh <fubar at us dot ibm dot com>
385 * - Fix ifenslave -c causing bond to loose existing routes;
386 * added bond_set_mac_address() that doesn't require the
388 * - In conjunction with fix for ifenslave -c, in
389 * bond_change_active(), changing to the already active slave
390 * is no longer an error (it successfully does nothing).
392 * 2003/06/30 - Amir Noam <amir.noam at intel dot com>
393 * - Fixed bond_change_active() for ALB/TLB modes.
396 * 2003/07/29 - Amir Noam <amir.noam at intel dot com>
397 * - Fixed ARP monitoring bug.
400 * 2003/07/31 - Willy Tarreau <willy at ods dot org>
401 * - Fixed kernel panic when using ARP monitoring without
402 * setting bond's IP address.
405 * 2003/08/06 - Amir Noam <amir.noam at intel dot com>
406 * - Back port from 2.6: use alloc_netdev(); fix /proc handling;
407 * made stats a part of bond struct so no need to allocate
408 * and free it separately; use standard list operations instead
409 * of pre-allocated array of bonds.
412 * 2003/08/07 - Jay Vosburgh <fubar at us dot ibm dot com>,
413 * Amir Noam <amir.noam at intel dot com> and
414 * Shmulik Hen <shmulik.hen at intel dot com>
415 * - Propagating master's settings: Distinguish between modes that
416 * use a primary slave from those that don't, and propagate settings
417 * accordingly; Consolidate change_active opeartions and add
418 * reselect_active and find_best opeartions; Decouple promiscuous
419 * handling from the multicast mode setting; Add support for changing
420 * HW address and MTU with proper unwind; Consolidate procfs code,
421 * add CHANGENAME handler; Enhance netdev notification handling.
424 * 2003/09/15 - Stephen Hemminger <shemminger at osdl dot org>,
425 * Amir Noam <amir.noam at intel dot com>
426 * - Convert /proc to seq_file interface.
427 * Change /proc/net/bondX/info to /proc/net/bonding/bondX.
428 * Set version to 2.4.1.
430 * 2003/11/20 - Amir Noam <amir.noam at intel dot com>
431 * - Fix /proc creation/destruction.
433 * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
434 * - Massive cleanup - Set version to 2.5.0
436 * o Consolidate format of prints and debug prints.
437 * o Remove bonding_t/slave_t typedefs and consolidate all casts.
438 * o Remove dead code and unnecessary checks.
439 * o Consolidate starting/stopping timers.
440 * o Consolidate handling of primary module param throughout the code.
441 * o Removed multicast module param support - all settings are done
443 * o Slave list iteration - bond is no longer part of the list,
444 * added cyclic list iteration macros.
445 * o Consolidate error handling in all xmit functions.
447 * o Consolidate function naming and declarations.
448 * o Consolidate function params and local variables names.
449 * o Consolidate return values.
450 * o Consolidate curly braces.
451 * o Consolidate conditionals format.
452 * o Change struct member names and types.
453 * o Chomp trailing spaces, remove empty lines, fix indentations.
454 * o Re-organize code according to context.
456 * 2003/12/30 - Amir Noam <amir.noam at intel dot com>
457 * - Fixed: Cannot remove and re-enslave the original active slave.
458 * - Fixed: Releasing the original active slave causes mac address
460 * - Add support for slaves that use ethtool_ops.
461 * Set version to 2.5.3.
463 * 2004/01/05 - Amir Noam <amir.noam at intel dot com>
464 * - Save bonding parameters per bond instead of using the global values.
465 * Set version to 2.5.4.
467 * 2004/01/14 - Shmulik Hen <shmulik.hen at intel dot com>
468 * - Enhance VLAN support:
469 * * Add support for VLAN hardware acceleration capable slaves.
470 * * Add capability to tag self generated packets in ALB/TLB modes.
471 * Set version to 2.6.0.
472 * 2004/10/29 - Mitch Williams <mitch.a.williams at intel dot com>
473 * - Fixed bug when unloading module while using 802.3ad. If
474 * spinlock debugging is turned on, this causes a stack dump.
475 * Solution is to move call to dev_remove_pack outside of the
477 * Set version to 2.6.1.
481 //#define BONDING_DEBUG 1
483 #include <linux/config.h>
484 #include <linux/kernel.h>
485 #include <linux/module.h>
486 #include <linux/sched.h>
487 #include <linux/types.h>
488 #include <linux/fcntl.h>
489 #include <linux/interrupt.h>
490 #include <linux/ptrace.h>
491 #include <linux/ioport.h>
492 #include <linux/in.h>
493 #include <linux/ip.h>
494 #include <linux/slab.h>
495 #include <linux/string.h>
496 #include <linux/init.h>
497 #include <linux/timer.h>
498 #include <linux/socket.h>
499 #include <linux/ctype.h>
500 #include <linux/inet.h>
501 #include <linux/bitops.h>
502 #include <asm/system.h>
505 #include <asm/uaccess.h>
506 #include <linux/errno.h>
507 #include <linux/netdevice.h>
508 #include <linux/inetdevice.h>
509 #include <linux/etherdevice.h>
510 #include <linux/skbuff.h>
511 #include <net/sock.h>
512 #include <linux/rtnetlink.h>
513 #include <linux/proc_fs.h>
514 #include <linux/seq_file.h>
515 #include <linux/smp.h>
516 #include <linux/if_ether.h>
518 #include <linux/mii.h>
519 #include <linux/ethtool.h>
520 #include <linux/if_vlan.h>
521 #include <linux/if_bonding.h>
523 #include "bond_3ad.h"
524 #include "bond_alb.h"
526 /*---------------------------- Module parameters ----------------------------*/
528 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
529 #define BOND_LINK_MON_INTERV 0
530 #define BOND_LINK_ARP_INTERV 0
532 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
533 static int miimon = BOND_LINK_MON_INTERV;
534 static int updelay = 0;
535 static int downdelay = 0;
536 static int use_carrier = 1;
537 static char *mode = NULL;
538 static char *primary = NULL;
539 static char *lacp_rate = NULL;
540 static int arp_interval = BOND_LINK_ARP_INTERV;
541 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
543 module_param(max_bonds, int, 0);
544 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
545 module_param(miimon, int, 0);
546 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
547 module_param(updelay, int, 0);
548 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
549 module_param(downdelay, int, 0);
550 MODULE_PARM_DESC(downdelay, "Delay before considering link down, in milliseconds");
551 module_param(use_carrier, int, 0);
552 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; 0 for off, 1 for on (default)");
553 module_param(mode, charp, 0);
554 MODULE_PARM_DESC(mode, "Mode of operation : 0 for round robin, 1 for active-backup, 2 for xor");
555 module_param(primary, charp, 0);
556 MODULE_PARM_DESC(primary, "Primary network device to use");
557 module_param(lacp_rate, charp, 0);
558 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner (slow/fast)");
559 module_param(arp_interval, int, 0);
560 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
561 module_param_array(arp_ip_target, charp, NULL, 0);
562 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
564 /*----------------------------- Global variables ----------------------------*/
566 static const char *version =
567 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
569 static LIST_HEAD(bond_dev_list);
571 #ifdef CONFIG_PROC_FS
572 static struct proc_dir_entry *bond_proc_dir = NULL;
575 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
576 static int arp_ip_count = 0;
577 static u32 my_ip = 0;
578 static int bond_mode = BOND_MODE_ROUNDROBIN;
579 static int lacp_fast = 0;
580 static int app_abi_ver = 0;
581 static int orig_app_abi_ver = -1; /* This is used to save the first ABI version
582 * we receive from the application. Once set,
583 * it won't be changed, and the module will
584 * refuse to enslave/release interfaces if the
585 * command comes from an application using
586 * another ABI version.
589 struct bond_parm_tbl {
594 static struct bond_parm_tbl bond_lacp_tbl[] = {
595 { "slow", AD_LACP_SLOW},
596 { "fast", AD_LACP_FAST},
600 static struct bond_parm_tbl bond_mode_tbl[] = {
601 { "balance-rr", BOND_MODE_ROUNDROBIN},
602 { "active-backup", BOND_MODE_ACTIVEBACKUP},
603 { "balance-xor", BOND_MODE_XOR},
604 { "broadcast", BOND_MODE_BROADCAST},
605 { "802.3ad", BOND_MODE_8023AD},
606 { "balance-tlb", BOND_MODE_TLB},
607 { "balance-alb", BOND_MODE_ALB},
611 /*-------------------------- Forward declarations ---------------------------*/
613 static inline void bond_set_mode_ops(struct net_device *bond_dev, int mode);
615 /*---------------------------- General routines -----------------------------*/
617 static const char *bond_mode_name(int mode)
620 case BOND_MODE_ROUNDROBIN :
621 return "load balancing (round-robin)";
622 case BOND_MODE_ACTIVEBACKUP :
623 return "fault-tolerance (active-backup)";
625 return "load balancing (xor)";
626 case BOND_MODE_BROADCAST :
627 return "fault-tolerance (broadcast)";
628 case BOND_MODE_8023AD:
629 return "IEEE 802.3ad Dynamic link aggregation";
631 return "transmit load balancing";
633 return "adaptive load balancing";
639 /*---------------------------------- VLAN -----------------------------------*/
642 * bond_add_vlan - add a new vlan id on bond
643 * @bond: bond that got the notification
644 * @vlan_id: the vlan id to add
646 * Returns -ENOMEM if allocation failed.
648 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
650 struct vlan_entry *vlan;
652 dprintk("bond: %s, vlan id %d\n",
653 (bond ? bond->dev->name: "None"), vlan_id);
655 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
660 INIT_LIST_HEAD(&vlan->vlan_list);
661 vlan->vlan_id = vlan_id;
663 write_lock_bh(&bond->lock);
665 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
667 write_unlock_bh(&bond->lock);
669 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
675 * bond_del_vlan - delete a vlan id from bond
676 * @bond: bond that got the notification
677 * @vlan_id: the vlan id to delete
679 * returns -ENODEV if @vlan_id was not found in @bond.
681 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
683 struct vlan_entry *vlan, *next;
686 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
688 write_lock_bh(&bond->lock);
690 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
691 if (vlan->vlan_id == vlan_id) {
692 list_del(&vlan->vlan_list);
694 if ((bond->params.mode == BOND_MODE_TLB) ||
695 (bond->params.mode == BOND_MODE_ALB)) {
696 bond_alb_clear_vlan(bond, vlan_id);
699 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
704 if (list_empty(&bond->vlan_list) &&
705 (bond->slave_cnt == 0)) {
706 /* Last VLAN removed and no slaves, so
707 * restore block on adding VLANs. This will
708 * be removed once new slaves that are not
709 * VLAN challenged will be added.
711 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
719 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
723 write_unlock_bh(&bond->lock);
728 * bond_has_challenged_slaves
729 * @bond: the bond we're working on
731 * Searches the slave list. Returns 1 if a vlan challenged slave
732 * was found, 0 otherwise.
734 * Assumes bond->lock is held.
736 static int bond_has_challenged_slaves(struct bonding *bond)
741 bond_for_each_slave(bond, slave, i) {
742 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
743 dprintk("found VLAN challenged slave - %s\n",
749 dprintk("no VLAN challenged slaves found\n");
754 * bond_next_vlan - safely skip to the next item in the vlans list.
755 * @bond: the bond we're working on
756 * @curr: item we're advancing from
758 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
759 * or @curr->next otherwise (even if it is @curr itself again).
761 * Caller must hold bond->lock
763 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
765 struct vlan_entry *next, *last;
767 if (list_empty(&bond->vlan_list)) {
772 next = list_entry(bond->vlan_list.next,
773 struct vlan_entry, vlan_list);
775 last = list_entry(bond->vlan_list.prev,
776 struct vlan_entry, vlan_list);
778 next = list_entry(bond->vlan_list.next,
779 struct vlan_entry, vlan_list);
781 next = list_entry(curr->vlan_list.next,
782 struct vlan_entry, vlan_list);
790 * bond_dev_queue_xmit - Prepare skb for xmit.
792 * @bond: bond device that got this skb for tx.
793 * @skb: hw accel VLAN tagged skb to transmit
794 * @slave_dev: slave that is supposed to xmit this skbuff
796 * When the bond gets an skb to tarnsmit that is
797 * already hardware accelerated VLAN tagged, and it
798 * needs to relay this skb to a slave that is not
799 * hw accel capable, the skb needs to be "unaccelerated",
800 * i.e. strip the hwaccel tag and re-insert it as part
803 * Assumption - once a VLAN device is created over the bond device, all
804 * packets are going to be hardware accelerated VLAN tagged since the IP
805 * binding is done over the VLAN device
807 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
809 unsigned short vlan_id;
812 if (!list_empty(&bond->vlan_list) &&
813 !(slave_dev->features & NETIF_F_HW_VLAN_TX)) {
814 res = vlan_get_tag(skb, &vlan_id);
819 skb->dev = slave_dev;
820 skb = vlan_put_tag(skb, vlan_id);
822 /* vlan_put_tag() frees the skb in case of error,
823 * so return success here so the calling functions
824 * won't attempt to free is again.
829 skb->dev = slave_dev;
839 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
840 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
842 * a. This operation is performed in IOCTL context,
843 * b. The operation is protected by the RTNL semaphore in the 8021q code,
844 * c. Holding a lock with BH disabled while directly calling a base driver
845 * entry point is generally a BAD idea.
847 * The design of synchronization/protection for this operation in the 8021q
848 * module is good for one or more VLAN devices over a single physical device
849 * and cannot be extended for a teaming solution like bonding, so there is a
850 * potential race condition here where a net device from the vlan group might
851 * be referenced (either by a base driver or the 8021q code) while it is being
852 * removed from the system. However, it turns out we're not making matters
853 * worse, and if it works for regular VLAN usage it will work here too.
857 * bond_vlan_rx_register - Propagates registration to slaves
858 * @bond_dev: bonding net device that got called
859 * @grp: vlan group being registered
861 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
863 struct bonding *bond = bond_dev->priv;
869 bond_for_each_slave(bond, slave, i) {
870 struct net_device *slave_dev = slave->dev;
872 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
873 slave_dev->vlan_rx_register) {
874 slave_dev->vlan_rx_register(slave_dev, grp);
880 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
881 * @bond_dev: bonding net device that got called
882 * @vid: vlan id being added
884 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
886 struct bonding *bond = bond_dev->priv;
890 bond_for_each_slave(bond, slave, i) {
891 struct net_device *slave_dev = slave->dev;
893 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
894 slave_dev->vlan_rx_add_vid) {
895 slave_dev->vlan_rx_add_vid(slave_dev, vid);
899 res = bond_add_vlan(bond, vid);
901 printk(KERN_ERR DRV_NAME
902 ": %s: Failed to add vlan id %d\n",
903 bond_dev->name, vid);
908 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
909 * @bond_dev: bonding net device that got called
910 * @vid: vlan id being removed
912 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
914 struct bonding *bond = bond_dev->priv;
916 struct net_device *vlan_dev;
919 bond_for_each_slave(bond, slave, i) {
920 struct net_device *slave_dev = slave->dev;
922 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
923 slave_dev->vlan_rx_kill_vid) {
924 /* Save and then restore vlan_dev in the grp array,
925 * since the slave's driver might clear it.
927 vlan_dev = bond->vlgrp->vlan_devices[vid];
928 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
929 bond->vlgrp->vlan_devices[vid] = vlan_dev;
933 res = bond_del_vlan(bond, vid);
935 printk(KERN_ERR DRV_NAME
936 ": %s: Failed to remove vlan id %d\n",
937 bond_dev->name, vid);
941 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
943 struct vlan_entry *vlan;
945 write_lock_bh(&bond->lock);
947 if (list_empty(&bond->vlan_list)) {
951 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
952 slave_dev->vlan_rx_register) {
953 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
956 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
957 !(slave_dev->vlan_rx_add_vid)) {
961 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
962 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
966 write_unlock_bh(&bond->lock);
969 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
971 struct vlan_entry *vlan;
972 struct net_device *vlan_dev;
974 write_lock_bh(&bond->lock);
976 if (list_empty(&bond->vlan_list)) {
980 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
981 !(slave_dev->vlan_rx_kill_vid)) {
985 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
986 /* Save and then restore vlan_dev in the grp array,
987 * since the slave's driver might clear it.
989 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
990 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
991 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
995 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
996 slave_dev->vlan_rx_register) {
997 slave_dev->vlan_rx_register(slave_dev, NULL);
1001 write_unlock_bh(&bond->lock);
1004 /*------------------------------- Link status -------------------------------*/
1007 * Get link speed and duplex from the slave's base driver
1008 * using ethtool. If for some reason the call fails or the
1009 * values are invalid, fake speed and duplex to 100/Full
1012 static int bond_update_speed_duplex(struct slave *slave)
1014 struct net_device *slave_dev = slave->dev;
1015 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1017 struct ethtool_cmd etool;
1019 /* Fake speed and duplex */
1020 slave->speed = SPEED_100;
1021 slave->duplex = DUPLEX_FULL;
1023 if (slave_dev->ethtool_ops) {
1026 if (!slave_dev->ethtool_ops->get_settings) {
1030 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
1038 ioctl = slave_dev->do_ioctl;
1039 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1040 etool.cmd = ETHTOOL_GSET;
1041 ifr.ifr_data = (char*)&etool;
1042 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
1047 switch (etool.speed) {
1056 switch (etool.duplex) {
1064 slave->speed = etool.speed;
1065 slave->duplex = etool.duplex;
1071 * if <dev> supports MII link status reporting, check its link status.
1073 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
1074 * depening upon the setting of the use_carrier parameter.
1076 * Return either BMSR_LSTATUS, meaning that the link is up (or we
1077 * can't tell and just pretend it is), or 0, meaning that the link is
1080 * If reporting is non-zero, instead of faking link up, return -1 if
1081 * both ETHTOOL and MII ioctls fail (meaning the device does not
1082 * support them). If use_carrier is set, return whatever it says.
1083 * It'd be nice if there was a good way to tell if a driver supports
1084 * netif_carrier, but there really isn't.
1086 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
1088 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1090 struct mii_ioctl_data *mii;
1091 struct ethtool_value etool;
1093 if (bond->params.use_carrier) {
1094 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1097 ioctl = slave_dev->do_ioctl;
1099 /* TODO: set pointer to correct ioctl on a per team member */
1100 /* bases to make this more efficient. that is, once */
1101 /* we determine the correct ioctl, we will always */
1102 /* call it and not the others for that team */
1106 * We cannot assume that SIOCGMIIPHY will also read a
1107 * register; not all network drivers (e.g., e100)
1111 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
1112 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1114 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
1115 mii->reg_num = MII_BMSR;
1116 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
1117 return (mii->val_out & BMSR_LSTATUS);
1122 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
1123 /* for a period of time so we attempt to get link status */
1124 /* from it last if the above MII ioctls fail... */
1125 if (slave_dev->ethtool_ops) {
1126 if (slave_dev->ethtool_ops->get_link) {
1129 link = slave_dev->ethtool_ops->get_link(slave_dev);
1131 return link ? BMSR_LSTATUS : 0;
1136 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1137 etool.cmd = ETHTOOL_GLINK;
1138 ifr.ifr_data = (char*)&etool;
1139 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
1140 if (etool.data == 1) {
1141 return BMSR_LSTATUS;
1143 dprintk("SIOCETHTOOL shows link down\n");
1150 * If reporting, report that either there's no dev->do_ioctl,
1151 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
1152 * cannot report link status). If not reporting, pretend
1155 return (reporting ? -1 : BMSR_LSTATUS);
1158 /*----------------------------- Multicast list ------------------------------*/
1161 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
1163 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
1165 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
1166 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
1170 * returns dmi entry if found, NULL otherwise
1172 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
1174 struct dev_mc_list *idmi;
1176 for (idmi = mc_list; idmi; idmi = idmi->next) {
1177 if (bond_is_dmi_same(dmi, idmi)) {
1186 * Push the promiscuity flag down to appropriate slaves
1188 static void bond_set_promiscuity(struct bonding *bond, int inc)
1190 if (USES_PRIMARY(bond->params.mode)) {
1191 /* write lock already acquired */
1192 if (bond->curr_active_slave) {
1193 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
1196 struct slave *slave;
1198 bond_for_each_slave(bond, slave, i) {
1199 dev_set_promiscuity(slave->dev, inc);
1205 * Push the allmulti flag down to all slaves
1207 static void bond_set_allmulti(struct bonding *bond, int inc)
1209 if (USES_PRIMARY(bond->params.mode)) {
1210 /* write lock already acquired */
1211 if (bond->curr_active_slave) {
1212 dev_set_allmulti(bond->curr_active_slave->dev, inc);
1215 struct slave *slave;
1217 bond_for_each_slave(bond, slave, i) {
1218 dev_set_allmulti(slave->dev, inc);
1224 * Add a Multicast address to slaves
1227 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
1229 if (USES_PRIMARY(bond->params.mode)) {
1230 /* write lock already acquired */
1231 if (bond->curr_active_slave) {
1232 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
1235 struct slave *slave;
1237 bond_for_each_slave(bond, slave, i) {
1238 dev_mc_add(slave->dev, addr, alen, 0);
1244 * Remove a multicast address from slave
1247 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
1249 if (USES_PRIMARY(bond->params.mode)) {
1250 /* write lock already acquired */
1251 if (bond->curr_active_slave) {
1252 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
1255 struct slave *slave;
1257 bond_for_each_slave(bond, slave, i) {
1258 dev_mc_delete(slave->dev, addr, alen, 0);
1264 * Totally destroys the mc_list in bond
1266 static void bond_mc_list_destroy(struct bonding *bond)
1268 struct dev_mc_list *dmi;
1270 dmi = bond->mc_list;
1272 bond->mc_list = dmi->next;
1274 dmi = bond->mc_list;
1279 * Copy all the Multicast addresses from src to the bonding device dst
1281 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond, int gpf_flag)
1283 struct dev_mc_list *dmi, *new_dmi;
1285 for (dmi = mc_list; dmi; dmi = dmi->next) {
1286 new_dmi = kmalloc(sizeof(struct dev_mc_list), gpf_flag);
1289 /* FIXME: Potential memory leak !!! */
1293 new_dmi->next = bond->mc_list;
1294 bond->mc_list = new_dmi;
1295 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
1296 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
1297 new_dmi->dmi_users = dmi->dmi_users;
1298 new_dmi->dmi_gusers = dmi->dmi_gusers;
1305 * flush all members of flush->mc_list from device dev->mc_list
1307 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
1309 struct bonding *bond = bond_dev->priv;
1310 struct dev_mc_list *dmi;
1312 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1313 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1316 if (bond->params.mode == BOND_MODE_8023AD) {
1317 /* del lacpdu mc addr from mc list */
1318 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1320 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1324 /*--------------------------- Active slave change ---------------------------*/
1327 * Update the mc list and multicast-related flags for the new and
1328 * old active slaves (if any) according to the multicast mode, and
1329 * promiscuous flags unconditionally.
1331 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
1333 struct dev_mc_list *dmi;
1335 if (!USES_PRIMARY(bond->params.mode)) {
1336 /* nothing to do - mc list is already up-to-date on
1343 if (bond->dev->flags & IFF_PROMISC) {
1344 dev_set_promiscuity(old_active->dev, -1);
1347 if (bond->dev->flags & IFF_ALLMULTI) {
1348 dev_set_allmulti(old_active->dev, -1);
1351 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1352 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1357 if (bond->dev->flags & IFF_PROMISC) {
1358 dev_set_promiscuity(new_active->dev, 1);
1361 if (bond->dev->flags & IFF_ALLMULTI) {
1362 dev_set_allmulti(new_active->dev, 1);
1365 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1366 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1372 * find_best_interface - select the best available slave to be the active one
1373 * @bond: our bonding struct
1375 * Warning: Caller must hold curr_slave_lock for writing.
1377 static struct slave *bond_find_best_slave(struct bonding *bond)
1379 struct slave *new_active, *old_active;
1380 struct slave *bestslave = NULL;
1381 int mintime = bond->params.updelay;
1384 new_active = old_active = bond->curr_active_slave;
1386 if (!new_active) { /* there were no active slaves left */
1387 if (bond->slave_cnt > 0) { /* found one slave */
1388 new_active = bond->first_slave;
1390 return NULL; /* still no slave, return NULL */
1394 /* first try the primary link; if arping, a link must tx/rx traffic
1395 * before it can be considered the curr_active_slave - also, we would skip
1396 * slaves between the curr_active_slave and primary_slave that may be up
1399 if ((bond->primary_slave) &&
1400 (!bond->params.arp_interval) &&
1401 (IS_UP(bond->primary_slave->dev))) {
1402 new_active = bond->primary_slave;
1405 /* remember where to stop iterating over the slaves */
1406 old_active = new_active;
1408 bond_for_each_slave_from(bond, new_active, i, old_active) {
1409 if (IS_UP(new_active->dev)) {
1410 if (new_active->link == BOND_LINK_UP) {
1412 } else if (new_active->link == BOND_LINK_BACK) {
1413 /* link up, but waiting for stabilization */
1414 if (new_active->delay < mintime) {
1415 mintime = new_active->delay;
1416 bestslave = new_active;
1426 * change_active_interface - change the active slave into the specified one
1427 * @bond: our bonding struct
1428 * @new: the new slave to make the active one
1430 * Set the new slave to the bond's settings and unset them on the old
1431 * curr_active_slave.
1432 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1434 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1435 * because it is apparently the best available slave we have, even though its
1436 * updelay hasn't timed out yet.
1438 * Warning: Caller must hold curr_slave_lock for writing.
1440 static void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1442 struct slave *old_active = bond->curr_active_slave;
1444 if (old_active == new_active) {
1449 if (new_active->link == BOND_LINK_BACK) {
1450 if (USES_PRIMARY(bond->params.mode)) {
1451 printk(KERN_INFO DRV_NAME
1452 ": %s: making interface %s the new "
1453 "active one %d ms earlier.\n",
1454 bond->dev->name, new_active->dev->name,
1455 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1458 new_active->delay = 0;
1459 new_active->link = BOND_LINK_UP;
1460 new_active->jiffies = jiffies;
1462 if (bond->params.mode == BOND_MODE_8023AD) {
1463 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1466 if ((bond->params.mode == BOND_MODE_TLB) ||
1467 (bond->params.mode == BOND_MODE_ALB)) {
1468 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1471 if (USES_PRIMARY(bond->params.mode)) {
1472 printk(KERN_INFO DRV_NAME
1473 ": %s: making interface %s the new "
1475 bond->dev->name, new_active->dev->name);
1480 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1482 bond_set_slave_inactive_flags(old_active);
1486 bond_set_slave_active_flags(new_active);
1490 if (USES_PRIMARY(bond->params.mode)) {
1491 bond_mc_swap(bond, new_active, old_active);
1494 if ((bond->params.mode == BOND_MODE_TLB) ||
1495 (bond->params.mode == BOND_MODE_ALB)) {
1496 bond_alb_handle_active_change(bond, new_active);
1498 bond->curr_active_slave = new_active;
1503 * bond_select_active_slave - select a new active slave, if needed
1504 * @bond: our bonding struct
1506 * This functions shoud be called when one of the following occurs:
1507 * - The old curr_active_slave has been released or lost its link.
1508 * - The primary_slave has got its link back.
1509 * - A slave has got its link back and there's no old curr_active_slave.
1511 * Warning: Caller must hold curr_slave_lock for writing.
1513 static void bond_select_active_slave(struct bonding *bond)
1515 struct slave *best_slave;
1517 best_slave = bond_find_best_slave(bond);
1518 if (best_slave != bond->curr_active_slave) {
1519 bond_change_active_slave(bond, best_slave);
1523 /*--------------------------- slave list handling ---------------------------*/
1526 * This function attaches the slave to the end of list.
1528 * bond->lock held for writing by caller.
1530 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1532 if (bond->first_slave == NULL) { /* attaching the first slave */
1533 new_slave->next = new_slave;
1534 new_slave->prev = new_slave;
1535 bond->first_slave = new_slave;
1537 new_slave->next = bond->first_slave;
1538 new_slave->prev = bond->first_slave->prev;
1539 new_slave->next->prev = new_slave;
1540 new_slave->prev->next = new_slave;
1547 * This function detaches the slave from the list.
1548 * WARNING: no check is made to verify if the slave effectively
1549 * belongs to <bond>.
1550 * Nothing is freed on return, structures are just unchained.
1551 * If any slave pointer in bond was pointing to <slave>,
1552 * it should be changed by the calling function.
1554 * bond->lock held for writing by caller.
1556 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1559 slave->next->prev = slave->prev;
1563 slave->prev->next = slave->next;
1566 if (bond->first_slave == slave) { /* slave is the first slave */
1567 if (bond->slave_cnt > 1) { /* there are more slave */
1568 bond->first_slave = slave->next;
1570 bond->first_slave = NULL; /* slave was the last one */
1579 /*---------------------------------- IOCTL ----------------------------------*/
1581 static int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1583 dprintk("bond_dev=%p\n", bond_dev);
1584 dprintk("slave_dev=%p\n", slave_dev);
1585 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1586 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1590 /* enslave device <slave> to bond device <master> */
1591 static int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1593 struct bonding *bond = bond_dev->priv;
1594 struct slave *new_slave = NULL;
1595 struct dev_mc_list *dmi;
1596 struct sockaddr addr;
1598 int old_features = bond_dev->features;
1601 if (slave_dev->do_ioctl == NULL) {
1602 printk(KERN_WARNING DRV_NAME
1603 ": Warning : no link monitoring support for %s\n",
1607 /* bond must be initialized by bond_open() before enslaving */
1608 if (!(bond_dev->flags & IFF_UP)) {
1609 dprintk("Error, master_dev is not up\n");
1613 /* already enslaved */
1614 if (slave_dev->flags & IFF_SLAVE) {
1615 dprintk("Error, Device was already enslaved\n");
1619 /* vlan challenged mutual exclusion */
1620 /* no need to lock since we're protected by rtnl_lock */
1621 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1622 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1623 if (!list_empty(&bond->vlan_list)) {
1624 printk(KERN_ERR DRV_NAME
1625 ": Error: cannot enslave VLAN "
1626 "challenged slave %s on VLAN enabled "
1627 "bond %s\n", slave_dev->name,
1631 printk(KERN_WARNING DRV_NAME
1632 ": Warning: enslaved VLAN challenged "
1633 "slave %s. Adding VLANs will be blocked as "
1634 "long as %s is part of bond %s\n",
1635 slave_dev->name, slave_dev->name,
1637 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1640 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1641 if (bond->slave_cnt == 0) {
1642 /* First slave, and it is not VLAN challenged,
1643 * so remove the block of adding VLANs over the bond.
1645 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1649 if (app_abi_ver >= 1) {
1650 /* The application is using an ABI, which requires the
1651 * slave interface to be closed.
1653 if ((slave_dev->flags & IFF_UP)) {
1654 printk(KERN_ERR DRV_NAME
1655 ": Error: %s is up\n",
1658 goto err_undo_flags;
1661 if (slave_dev->set_mac_address == NULL) {
1662 printk(KERN_ERR DRV_NAME
1663 ": Error: The slave device you specified does "
1664 "not support setting the MAC address.\n");
1666 "Your kernel likely does not support slave "
1670 goto err_undo_flags;
1673 /* The application is not using an ABI, which requires the
1674 * slave interface to be open.
1676 if (!(slave_dev->flags & IFF_UP)) {
1677 printk(KERN_ERR DRV_NAME
1678 ": Error: %s is not running\n",
1681 goto err_undo_flags;
1684 if ((bond->params.mode == BOND_MODE_8023AD) ||
1685 (bond->params.mode == BOND_MODE_TLB) ||
1686 (bond->params.mode == BOND_MODE_ALB)) {
1687 printk(KERN_ERR DRV_NAME
1688 ": Error: to use %s mode, you must upgrade "
1690 bond_mode_name(bond->params.mode));
1692 goto err_undo_flags;
1696 new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1699 goto err_undo_flags;
1702 memset(new_slave, 0, sizeof(struct slave));
1704 /* save slave's original flags before calling
1705 * netdev_set_master and dev_open
1707 new_slave->original_flags = slave_dev->flags;
1709 if (app_abi_ver >= 1) {
1710 /* save slave's original ("permanent") mac address for
1711 * modes that needs it, and for restoring it upon release,
1712 * and then set it to the master's address
1714 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1716 /* set slave to master's mac address
1717 * The application already set the master's
1718 * mac address to that of the first slave
1720 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1721 addr.sa_family = slave_dev->type;
1722 res = slave_dev->set_mac_address(slave_dev, &addr);
1724 dprintk("Error %d calling set_mac_address\n", res);
1728 /* open the slave since the application closed it */
1729 res = dev_open(slave_dev);
1731 dprintk("Openning slave %s failed\n", slave_dev->name);
1732 goto err_restore_mac;
1736 res = netdev_set_master(slave_dev, bond_dev);
1738 dprintk("Error %d calling netdev_set_master\n", res);
1739 if (app_abi_ver < 1) {
1746 new_slave->dev = slave_dev;
1748 if ((bond->params.mode == BOND_MODE_TLB) ||
1749 (bond->params.mode == BOND_MODE_ALB)) {
1750 /* bond_alb_init_slave() must be called before all other stages since
1751 * it might fail and we do not want to have to undo everything
1753 res = bond_alb_init_slave(bond, new_slave);
1755 goto err_unset_master;
1759 /* If the mode USES_PRIMARY, then the new slave gets the
1760 * master's promisc (and mc) settings only if it becomes the
1761 * curr_active_slave, and that is taken care of later when calling
1762 * bond_change_active()
1764 if (!USES_PRIMARY(bond->params.mode)) {
1765 /* set promiscuity level to new slave */
1766 if (bond_dev->flags & IFF_PROMISC) {
1767 dev_set_promiscuity(slave_dev, 1);
1770 /* set allmulti level to new slave */
1771 if (bond_dev->flags & IFF_ALLMULTI) {
1772 dev_set_allmulti(slave_dev, 1);
1775 /* upload master's mc_list to new slave */
1776 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1777 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1781 if (bond->params.mode == BOND_MODE_8023AD) {
1782 /* add lacpdu mc addr to mc list */
1783 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1785 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1788 bond_add_vlans_on_slave(bond, slave_dev);
1790 write_lock_bh(&bond->lock);
1792 bond_attach_slave(bond, new_slave);
1794 new_slave->delay = 0;
1795 new_slave->link_failure_count = 0;
1797 if (bond->params.miimon && !bond->params.use_carrier) {
1798 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1800 if ((link_reporting == -1) && !bond->params.arp_interval) {
1802 * miimon is set but a bonded network driver
1803 * does not support ETHTOOL/MII and
1804 * arp_interval is not set. Note: if
1805 * use_carrier is enabled, we will never go
1806 * here (because netif_carrier is always
1807 * supported); thus, we don't need to change
1808 * the messages for netif_carrier.
1810 printk(KERN_WARNING DRV_NAME
1811 ": Warning: MII and ETHTOOL support not "
1812 "available for interface %s, and "
1813 "arp_interval/arp_ip_target module parameters "
1814 "not specified, thus bonding will not detect "
1815 "link failures! see bonding.txt for details.\n",
1817 } else if (link_reporting == -1) {
1818 /* unable get link status using mii/ethtool */
1819 printk(KERN_WARNING DRV_NAME
1820 ": Warning: can't get link status from "
1821 "interface %s; the network driver associated "
1822 "with this interface does not support MII or "
1823 "ETHTOOL link status reporting, thus miimon "
1824 "has no effect on this interface.\n",
1829 /* check for initial state */
1830 if (!bond->params.miimon ||
1831 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1832 if (bond->params.updelay) {
1833 dprintk("Initial state of slave_dev is "
1834 "BOND_LINK_BACK\n");
1835 new_slave->link = BOND_LINK_BACK;
1836 new_slave->delay = bond->params.updelay;
1838 dprintk("Initial state of slave_dev is "
1840 new_slave->link = BOND_LINK_UP;
1842 new_slave->jiffies = jiffies;
1844 dprintk("Initial state of slave_dev is "
1845 "BOND_LINK_DOWN\n");
1846 new_slave->link = BOND_LINK_DOWN;
1849 if (bond_update_speed_duplex(new_slave) &&
1850 (new_slave->link != BOND_LINK_DOWN)) {
1851 printk(KERN_WARNING DRV_NAME
1852 ": Warning: failed to get speed/duplex from %s, speed "
1853 "forced to 100Mbps, duplex forced to Full.\n",
1854 new_slave->dev->name);
1856 if (bond->params.mode == BOND_MODE_8023AD) {
1858 "Operation of 802.3ad mode requires ETHTOOL "
1859 "support in base driver for proper aggregator "
1864 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1865 /* if there is a primary slave, remember it */
1866 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1867 bond->primary_slave = new_slave;
1871 switch (bond->params.mode) {
1872 case BOND_MODE_ACTIVEBACKUP:
1873 /* if we're in active-backup mode, we need one and only one active
1874 * interface. The backup interfaces will have their NOARP flag set
1875 * because we need them to be completely deaf and not to respond to
1876 * any ARP request on the network to avoid fooling a switch. Thus,
1877 * since we guarantee that curr_active_slave always point to the last
1878 * usable interface, we just have to verify this interface's flag.
1880 if (((!bond->curr_active_slave) ||
1881 (bond->curr_active_slave->dev->flags & IFF_NOARP)) &&
1882 (new_slave->link != BOND_LINK_DOWN)) {
1883 dprintk("This is the first active slave\n");
1884 /* first slave or no active slave yet, and this link
1885 is OK, so make this interface the active one */
1886 bond_change_active_slave(bond, new_slave);
1888 dprintk("This is just a backup slave\n");
1889 bond_set_slave_inactive_flags(new_slave);
1892 case BOND_MODE_8023AD:
1893 /* in 802.3ad mode, the internal mechanism
1894 * will activate the slaves in the selected
1897 bond_set_slave_inactive_flags(new_slave);
1898 /* if this is the first slave */
1899 if (bond->slave_cnt == 1) {
1900 SLAVE_AD_INFO(new_slave).id = 1;
1901 /* Initialize AD with the number of times that the AD timer is called in 1 second
1902 * can be called only after the mac address of the bond is set
1904 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1905 bond->params.lacp_fast);
1907 SLAVE_AD_INFO(new_slave).id =
1908 SLAVE_AD_INFO(new_slave->prev).id + 1;
1911 bond_3ad_bind_slave(new_slave);
1915 new_slave->state = BOND_STATE_ACTIVE;
1916 if ((!bond->curr_active_slave) &&
1917 (new_slave->link != BOND_LINK_DOWN)) {
1918 /* first slave or no active slave yet, and this link
1919 * is OK, so make this interface the active one
1921 bond_change_active_slave(bond, new_slave);
1925 dprintk("This slave is always active in trunk mode\n");
1927 /* always active in trunk mode */
1928 new_slave->state = BOND_STATE_ACTIVE;
1930 /* In trunking mode there is little meaning to curr_active_slave
1931 * anyway (it holds no special properties of the bond device),
1932 * so we can change it without calling change_active_interface()
1934 if (!bond->curr_active_slave) {
1935 bond->curr_active_slave = new_slave;
1938 } /* switch(bond_mode) */
1940 write_unlock_bh(&bond->lock);
1942 if (app_abi_ver < 1) {
1944 * !!! This is to support old versions of ifenslave.
1945 * We can remove this in 2.5 because our ifenslave takes
1946 * care of this for us.
1947 * We check to see if the master has a mac address yet.
1948 * If not, we'll give it the mac address of our slave device.
1952 for (ndx = 0; ndx < bond_dev->addr_len; ndx++) {
1953 dprintk("Checking ndx=%d of bond_dev->dev_addr\n",
1955 if (bond_dev->dev_addr[ndx] != 0) {
1956 dprintk("Found non-zero byte at ndx=%d\n",
1962 if (ndx == bond_dev->addr_len) {
1964 * We got all the way through the address and it was
1967 dprintk("%s doesn't have a MAC address yet. \n",
1969 dprintk("Going to give assign it from %s.\n",
1971 bond_sethwaddr(bond_dev, slave_dev);
1975 printk(KERN_INFO DRV_NAME
1976 ": %s: enslaving %s as a%s interface with a%s link.\n",
1977 bond_dev->name, slave_dev->name,
1978 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1979 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1981 /* enslave is successful */
1984 /* Undo stages on error */
1986 netdev_set_master(slave_dev, NULL);
1989 dev_close(slave_dev);
1992 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1993 addr.sa_family = slave_dev->type;
1994 slave_dev->set_mac_address(slave_dev, &addr);
2000 bond_dev->features = old_features;
2006 * Try to release the slave device <slave> from the bond device <master>
2007 * It is legal to access curr_active_slave without a lock because all the function
2010 * The rules for slave state should be:
2011 * for Active/Backup:
2012 * Active stays on all backups go down
2013 * for Bonded connections:
2014 * The first up interface should be left on and all others downed.
2016 static int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2018 struct bonding *bond = bond_dev->priv;
2019 struct slave *slave, *oldcurrent;
2020 struct sockaddr addr;
2021 int mac_addr_differ;
2023 /* slave is not a slave or master is not master of this slave */
2024 if (!(slave_dev->flags & IFF_SLAVE) ||
2025 (slave_dev->master != bond_dev)) {
2026 printk(KERN_ERR DRV_NAME
2027 ": Error: %s: cannot release %s.\n",
2028 bond_dev->name, slave_dev->name);
2032 write_lock_bh(&bond->lock);
2034 slave = bond_get_slave_by_dev(bond, slave_dev);
2036 /* not a slave of this bond */
2037 printk(KERN_INFO DRV_NAME
2038 ": %s: %s not enslaved\n",
2039 bond_dev->name, slave_dev->name);
2043 mac_addr_differ = memcmp(bond_dev->dev_addr,
2046 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
2047 printk(KERN_WARNING DRV_NAME
2048 ": Warning: the permanent HWaddr of %s "
2049 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
2050 "still in use by %s. Set the HWaddr of "
2051 "%s to a different address to avoid "
2054 slave->perm_hwaddr[0],
2055 slave->perm_hwaddr[1],
2056 slave->perm_hwaddr[2],
2057 slave->perm_hwaddr[3],
2058 slave->perm_hwaddr[4],
2059 slave->perm_hwaddr[5],
2064 /* Inform AD package of unbinding of slave. */
2065 if (bond->params.mode == BOND_MODE_8023AD) {
2066 /* must be called before the slave is
2067 * detached from the list
2069 bond_3ad_unbind_slave(slave);
2072 printk(KERN_INFO DRV_NAME
2073 ": %s: releasing %s interface %s\n",
2075 (slave->state == BOND_STATE_ACTIVE)
2076 ? "active" : "backup",
2079 oldcurrent = bond->curr_active_slave;
2081 bond->current_arp_slave = NULL;
2083 /* release the slave from its bond */
2084 bond_detach_slave(bond, slave);
2086 if (bond->primary_slave == slave) {
2087 bond->primary_slave = NULL;
2090 if (oldcurrent == slave) {
2091 bond_change_active_slave(bond, NULL);
2094 if ((bond->params.mode == BOND_MODE_TLB) ||
2095 (bond->params.mode == BOND_MODE_ALB)) {
2096 /* Must be called only after the slave has been
2097 * detached from the list and the curr_active_slave
2098 * has been cleared (if our_slave == old_current),
2099 * but before a new active slave is selected.
2101 bond_alb_deinit_slave(bond, slave);
2104 if (oldcurrent == slave) {
2105 bond_select_active_slave(bond);
2107 if (!bond->curr_active_slave) {
2108 printk(KERN_INFO DRV_NAME
2109 ": %s: now running without any active "
2115 if (bond->slave_cnt == 0) {
2116 /* if the last slave was removed, zero the mac address
2117 * of the master so it will be set by the application
2118 * to the mac address of the first slave
2120 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2122 if (list_empty(&bond->vlan_list)) {
2123 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2125 printk(KERN_WARNING DRV_NAME
2126 ": Warning: clearing HW address of %s while it "
2127 "still has VLANs.\n",
2129 printk(KERN_WARNING DRV_NAME
2130 ": When re-adding slaves, make sure the bond's "
2131 "HW address matches its VLANs'.\n");
2133 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2134 !bond_has_challenged_slaves(bond)) {
2135 printk(KERN_INFO DRV_NAME
2136 ": last VLAN challenged slave %s "
2137 "left bond %s. VLAN blocking is removed\n",
2138 slave_dev->name, bond_dev->name);
2139 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2142 write_unlock_bh(&bond->lock);
2144 bond_del_vlans_from_slave(bond, slave_dev);
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
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);
2156 /* unset allmulti level from slave */
2157 if (bond_dev->flags & IFF_ALLMULTI) {
2158 dev_set_allmulti(slave_dev, -1);
2161 /* flush master's mc_list from slave */
2162 bond_mc_list_flush(bond_dev, slave_dev);
2165 netdev_set_master(slave_dev, NULL);
2167 /* close slave before restoring its mac address */
2168 dev_close(slave_dev);
2170 if (app_abi_ver >= 1) {
2171 /* restore original ("permanent") mac address */
2172 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2173 addr.sa_family = slave_dev->type;
2174 slave_dev->set_mac_address(slave_dev, &addr);
2177 /* restore the original state of the
2178 * IFF_NOARP flag that might have been
2179 * set by bond_set_slave_inactive_flags()
2181 if ((slave->original_flags & IFF_NOARP) == 0) {
2182 slave_dev->flags &= ~IFF_NOARP;
2187 return 0; /* deletion OK */
2191 * This function releases all slaves.
2193 static int bond_release_all(struct net_device *bond_dev)
2195 struct bonding *bond = bond_dev->priv;
2196 struct slave *slave;
2197 struct net_device *slave_dev;
2198 struct sockaddr addr;
2200 write_lock_bh(&bond->lock);
2202 if (bond->slave_cnt == 0) {
2206 bond->current_arp_slave = NULL;
2207 bond->primary_slave = NULL;
2208 bond_change_active_slave(bond, NULL);
2210 while ((slave = bond->first_slave) != NULL) {
2211 /* Inform AD package of unbinding of slave
2212 * before slave is detached from the list.
2214 if (bond->params.mode == BOND_MODE_8023AD) {
2215 bond_3ad_unbind_slave(slave);
2218 slave_dev = slave->dev;
2219 bond_detach_slave(bond, slave);
2221 if ((bond->params.mode == BOND_MODE_TLB) ||
2222 (bond->params.mode == BOND_MODE_ALB)) {
2223 /* must be called only after the slave
2224 * has been detached from the list
2226 bond_alb_deinit_slave(bond, slave);
2229 /* now that the slave is detached, unlock and perform
2230 * all the undo steps that should not be called from
2233 write_unlock_bh(&bond->lock);
2235 bond_del_vlans_from_slave(bond, slave_dev);
2237 /* If the mode USES_PRIMARY, then we should only remove its
2238 * promisc and mc settings if it was the curr_active_slave, but that was
2239 * already taken care of above when we detached the slave
2241 if (!USES_PRIMARY(bond->params.mode)) {
2242 /* unset promiscuity level from slave */
2243 if (bond_dev->flags & IFF_PROMISC) {
2244 dev_set_promiscuity(slave_dev, -1);
2247 /* unset allmulti level from slave */
2248 if (bond_dev->flags & IFF_ALLMULTI) {
2249 dev_set_allmulti(slave_dev, -1);
2252 /* flush master's mc_list from slave */
2253 bond_mc_list_flush(bond_dev, slave_dev);
2256 netdev_set_master(slave_dev, NULL);
2258 /* close slave before restoring its mac address */
2259 dev_close(slave_dev);
2261 if (app_abi_ver >= 1) {
2262 /* restore original ("permanent") mac address*/
2263 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2264 addr.sa_family = slave_dev->type;
2265 slave_dev->set_mac_address(slave_dev, &addr);
2268 /* restore the original state of the IFF_NOARP flag that might have
2269 * been set by bond_set_slave_inactive_flags()
2271 if ((slave->original_flags & IFF_NOARP) == 0) {
2272 slave_dev->flags &= ~IFF_NOARP;
2277 /* re-acquire the lock before getting the next slave */
2278 write_lock_bh(&bond->lock);
2281 /* zero the mac address of the master so it will be
2282 * set by the application to the mac address of the
2285 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2287 if (list_empty(&bond->vlan_list)) {
2288 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2290 printk(KERN_WARNING DRV_NAME
2291 ": Warning: clearing HW address of %s while it "
2292 "still has VLANs.\n",
2294 printk(KERN_WARNING DRV_NAME
2295 ": When re-adding slaves, make sure the bond's "
2296 "HW address matches its VLANs'.\n");
2299 printk(KERN_INFO DRV_NAME
2300 ": %s: released all slaves\n",
2304 write_unlock_bh(&bond->lock);
2310 * This function changes the active slave to slave <slave_dev>.
2311 * It returns -EINVAL in the following cases.
2312 * - <slave_dev> is not found in the list.
2313 * - There is not active slave now.
2314 * - <slave_dev> is already active.
2315 * - The link state of <slave_dev> is not BOND_LINK_UP.
2316 * - <slave_dev> is not running.
2317 * In these cases, this fuction does nothing.
2318 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2320 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2322 struct bonding *bond = bond_dev->priv;
2323 struct slave *old_active = NULL;
2324 struct slave *new_active = NULL;
2327 if (!USES_PRIMARY(bond->params.mode)) {
2331 /* Verify that master_dev is indeed the master of slave_dev */
2332 if (!(slave_dev->flags & IFF_SLAVE) ||
2333 (slave_dev->master != bond_dev)) {
2337 write_lock_bh(&bond->lock);
2339 old_active = bond->curr_active_slave;
2340 new_active = bond_get_slave_by_dev(bond, slave_dev);
2343 * Changing to the current active: do nothing; return success.
2345 if (new_active && (new_active == old_active)) {
2346 write_unlock_bh(&bond->lock);
2352 (new_active->link == BOND_LINK_UP) &&
2353 IS_UP(new_active->dev)) {
2354 bond_change_active_slave(bond, new_active);
2359 write_unlock_bh(&bond->lock);
2364 static int bond_ethtool_ioctl(struct net_device *bond_dev, struct ifreq *ifr)
2366 struct ethtool_drvinfo info;
2367 void __user *addr = ifr->ifr_data;
2370 if (get_user(cmd, (uint32_t __user *)addr)) {
2375 case ETHTOOL_GDRVINFO:
2376 if (copy_from_user(&info, addr, sizeof(info))) {
2380 if (strcmp(info.driver, "ifenslave") == 0) {
2384 new_abi_ver = simple_strtoul(info.fw_version,
2387 printk(KERN_ERR DRV_NAME
2388 ": Error: got invalid ABI "
2389 "version from application\n");
2394 if (orig_app_abi_ver == -1) {
2395 orig_app_abi_ver = new_abi_ver;
2398 app_abi_ver = new_abi_ver;
2401 strncpy(info.driver, DRV_NAME, 32);
2402 strncpy(info.version, DRV_VERSION, 32);
2403 snprintf(info.fw_version, 32, "%d", BOND_ABI_VERSION);
2405 if (copy_to_user(addr, &info, sizeof(info))) {
2415 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2417 struct bonding *bond = bond_dev->priv;
2419 info->bond_mode = bond->params.mode;
2420 info->miimon = bond->params.miimon;
2422 read_lock_bh(&bond->lock);
2423 info->num_slaves = bond->slave_cnt;
2424 read_unlock_bh(&bond->lock);
2429 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2431 struct bonding *bond = bond_dev->priv;
2432 struct slave *slave;
2435 if (info->slave_id < 0) {
2439 read_lock_bh(&bond->lock);
2441 bond_for_each_slave(bond, slave, i) {
2442 if (i == (int)info->slave_id) {
2448 read_unlock_bh(&bond->lock);
2451 strcpy(info->slave_name, slave->dev->name);
2452 info->link = slave->link;
2453 info->state = slave->state;
2454 info->link_failure_count = slave->link_failure_count;
2462 /*-------------------------------- Monitoring -------------------------------*/
2464 /* this function is called regularly to monitor each slave's link. */
2465 static void bond_mii_monitor(struct net_device *bond_dev)
2467 struct bonding *bond = bond_dev->priv;
2468 struct slave *slave, *oldcurrent;
2469 int do_failover = 0;
2473 read_lock(&bond->lock);
2475 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2477 if (bond->kill_timers) {
2481 if (bond->slave_cnt == 0) {
2485 /* we will try to read the link status of each of our slaves, and
2486 * set their IFF_RUNNING flag appropriately. For each slave not
2487 * supporting MII status, we won't do anything so that a user-space
2488 * program could monitor the link itself if needed.
2491 read_lock(&bond->curr_slave_lock);
2492 oldcurrent = bond->curr_active_slave;
2493 read_unlock(&bond->curr_slave_lock);
2495 bond_for_each_slave(bond, slave, i) {
2496 struct net_device *slave_dev = slave->dev;
2498 u16 old_speed = slave->speed;
2499 u8 old_duplex = slave->duplex;
2501 link_state = bond_check_dev_link(bond, slave_dev, 0);
2503 switch (slave->link) {
2504 case BOND_LINK_UP: /* the link was up */
2505 if (link_state == BMSR_LSTATUS) {
2506 /* link stays up, nothing more to do */
2508 } else { /* link going down */
2509 slave->link = BOND_LINK_FAIL;
2510 slave->delay = bond->params.downdelay;
2512 if (slave->link_failure_count < UINT_MAX) {
2513 slave->link_failure_count++;
2516 if (bond->params.downdelay) {
2517 printk(KERN_INFO DRV_NAME
2518 ": %s: link status down for %s "
2519 "interface %s, disabling it in "
2523 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2524 ? ((slave == oldcurrent)
2525 ? "active " : "backup ")
2529 bond->params.downdelay * bond->params.miimon);
2532 /* no break ! fall through the BOND_LINK_FAIL test to
2533 ensure proper action to be taken
2535 case BOND_LINK_FAIL: /* the link has just gone down */
2536 if (link_state != BMSR_LSTATUS) {
2537 /* link stays down */
2538 if (slave->delay <= 0) {
2539 /* link down for too long time */
2540 slave->link = BOND_LINK_DOWN;
2542 /* in active/backup mode, we must
2543 * completely disable this interface
2545 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2546 (bond->params.mode == BOND_MODE_8023AD)) {
2547 bond_set_slave_inactive_flags(slave);
2550 printk(KERN_INFO DRV_NAME
2551 ": %s: link status definitely "
2552 "down for interface %s, "
2557 /* notify ad that the link status has changed */
2558 if (bond->params.mode == BOND_MODE_8023AD) {
2559 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2562 if ((bond->params.mode == BOND_MODE_TLB) ||
2563 (bond->params.mode == BOND_MODE_ALB)) {
2564 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2567 if (slave == oldcurrent) {
2575 slave->link = BOND_LINK_UP;
2576 slave->jiffies = jiffies;
2577 printk(KERN_INFO DRV_NAME
2578 ": %s: link status up again after %d "
2579 "ms for interface %s.\n",
2581 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2585 case BOND_LINK_DOWN: /* the link was down */
2586 if (link_state != BMSR_LSTATUS) {
2587 /* the link stays down, nothing more to do */
2589 } else { /* link going up */
2590 slave->link = BOND_LINK_BACK;
2591 slave->delay = bond->params.updelay;
2593 if (bond->params.updelay) {
2594 /* if updelay == 0, no need to
2595 advertise about a 0 ms delay */
2596 printk(KERN_INFO DRV_NAME
2597 ": %s: link status up for "
2598 "interface %s, enabling it "
2602 bond->params.updelay * bond->params.miimon);
2605 /* no break ! fall through the BOND_LINK_BACK state in
2606 case there's something to do.
2608 case BOND_LINK_BACK: /* the link has just come back */
2609 if (link_state != BMSR_LSTATUS) {
2610 /* link down again */
2611 slave->link = BOND_LINK_DOWN;
2613 printk(KERN_INFO DRV_NAME
2614 ": %s: link status down again after %d "
2615 "ms for interface %s.\n",
2617 (bond->params.updelay - slave->delay) * bond->params.miimon,
2621 if (slave->delay == 0) {
2622 /* now the link has been up for long time enough */
2623 slave->link = BOND_LINK_UP;
2624 slave->jiffies = jiffies;
2626 if (bond->params.mode == BOND_MODE_8023AD) {
2627 /* prevent it from being the active one */
2628 slave->state = BOND_STATE_BACKUP;
2629 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2630 /* make it immediately active */
2631 slave->state = BOND_STATE_ACTIVE;
2632 } else if (slave != bond->primary_slave) {
2633 /* prevent it from being the active one */
2634 slave->state = BOND_STATE_BACKUP;
2637 printk(KERN_INFO DRV_NAME
2638 ": %s: link status definitely "
2639 "up for interface %s.\n",
2643 /* notify ad that the link status has changed */
2644 if (bond->params.mode == BOND_MODE_8023AD) {
2645 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2648 if ((bond->params.mode == BOND_MODE_TLB) ||
2649 (bond->params.mode == BOND_MODE_ALB)) {
2650 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2653 if ((!oldcurrent) ||
2654 (slave == bond->primary_slave)) {
2663 /* Should not happen */
2664 printk(KERN_ERR "bonding: Error: %s Illegal value (link=%d)\n",
2665 slave->dev->name, slave->link);
2667 } /* end of switch (slave->link) */
2669 bond_update_speed_duplex(slave);
2671 if (bond->params.mode == BOND_MODE_8023AD) {
2672 if (old_speed != slave->speed) {
2673 bond_3ad_adapter_speed_changed(slave);
2676 if (old_duplex != slave->duplex) {
2677 bond_3ad_adapter_duplex_changed(slave);
2684 write_lock(&bond->curr_slave_lock);
2686 bond_select_active_slave(bond);
2688 if (oldcurrent && !bond->curr_active_slave) {
2689 printk(KERN_INFO DRV_NAME
2690 ": %s: now running without any active "
2695 write_unlock(&bond->curr_slave_lock);
2699 if (bond->params.miimon) {
2700 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2703 read_unlock(&bond->lock);
2706 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2709 u32 *targets = bond->params.arp_targets;
2711 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2712 arp_send(ARPOP_REQUEST, ETH_P_ARP, targets[i], slave->dev,
2713 my_ip, NULL, slave->dev->dev_addr,
2719 * this function is called regularly to monitor each slave's link
2720 * ensuring that traffic is being sent and received when arp monitoring
2721 * is used in load-balancing mode. if the adapter has been dormant, then an
2722 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2723 * arp monitoring in active backup mode.
2725 static void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2727 struct bonding *bond = bond_dev->priv;
2728 struct slave *slave, *oldcurrent;
2729 int do_failover = 0;
2733 read_lock(&bond->lock);
2735 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2737 if (bond->kill_timers) {
2741 if (bond->slave_cnt == 0) {
2745 read_lock(&bond->curr_slave_lock);
2746 oldcurrent = bond->curr_active_slave;
2747 read_unlock(&bond->curr_slave_lock);
2749 /* see if any of the previous devices are up now (i.e. they have
2750 * xmt and rcv traffic). the curr_active_slave does not come into
2751 * the picture unless it is null. also, slave->jiffies is not needed
2752 * here because we send an arp on each slave and give a slave as
2753 * long as it needs to get the tx/rx within the delta.
2754 * TODO: what about up/down delay in arp mode? it wasn't here before
2757 bond_for_each_slave(bond, slave, i) {
2758 if (slave->link != BOND_LINK_UP) {
2759 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2760 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2762 slave->link = BOND_LINK_UP;
2763 slave->state = BOND_STATE_ACTIVE;
2765 /* primary_slave has no meaning in round-robin
2766 * mode. the window of a slave being up and
2767 * curr_active_slave being null after enslaving
2771 printk(KERN_INFO DRV_NAME
2772 ": %s: link status definitely "
2773 "up for interface %s, ",
2778 printk(KERN_INFO DRV_NAME
2779 ": %s: interface %s is now up\n",
2785 /* slave->link == BOND_LINK_UP */
2787 /* not all switches will respond to an arp request
2788 * when the source ip is 0, so don't take the link down
2789 * if we don't know our ip yet
2791 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2792 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2795 slave->link = BOND_LINK_DOWN;
2796 slave->state = BOND_STATE_BACKUP;
2798 if (slave->link_failure_count < UINT_MAX) {
2799 slave->link_failure_count++;
2802 printk(KERN_INFO DRV_NAME
2803 ": %s: interface %s is now down.\n",
2807 if (slave == oldcurrent) {
2813 /* note: if switch is in round-robin mode, all links
2814 * must tx arp to ensure all links rx an arp - otherwise
2815 * links may oscillate or not come up at all; if switch is
2816 * in something like xor mode, there is nothing we can
2817 * do - all replies will be rx'ed on same link causing slaves
2818 * to be unstable during low/no traffic periods
2820 if (IS_UP(slave->dev)) {
2821 bond_arp_send_all(bond, slave);
2826 write_lock(&bond->curr_slave_lock);
2828 bond_select_active_slave(bond);
2830 if (oldcurrent && !bond->curr_active_slave) {
2831 printk(KERN_INFO DRV_NAME
2832 ": %s: now running without any active "
2837 write_unlock(&bond->curr_slave_lock);
2841 if (bond->params.arp_interval) {
2842 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2845 read_unlock(&bond->lock);
2849 * When using arp monitoring in active-backup mode, this function is
2850 * called to determine if any backup slaves have went down or a new
2851 * current slave needs to be found.
2852 * The backup slaves never generate traffic, they are considered up by merely
2853 * receiving traffic. If the current slave goes down, each backup slave will
2854 * be given the opportunity to tx/rx an arp before being taken down - this
2855 * prevents all slaves from being taken down due to the current slave not
2856 * sending any traffic for the backups to receive. The arps are not necessarily
2857 * necessary, any tx and rx traffic will keep the current slave up. While any
2858 * rx traffic will keep the backup slaves up, the current slave is responsible
2859 * for generating traffic to keep them up regardless of any other traffic they
2860 * may have received.
2861 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2863 static void bond_activebackup_arp_mon(struct net_device *bond_dev)
2865 struct bonding *bond = bond_dev->priv;
2866 struct slave *slave;
2870 read_lock(&bond->lock);
2872 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2874 if (bond->kill_timers) {
2878 if (bond->slave_cnt == 0) {
2882 /* determine if any slave has come up or any backup slave has
2884 * TODO: what about up/down delay in arp mode? it wasn't here before
2887 bond_for_each_slave(bond, slave, i) {
2888 if (slave->link != BOND_LINK_UP) {
2889 if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
2891 slave->link = BOND_LINK_UP;
2893 write_lock(&bond->curr_slave_lock);
2895 if ((!bond->curr_active_slave) &&
2896 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2897 bond_change_active_slave(bond, slave);
2898 bond->current_arp_slave = NULL;
2899 } else if (bond->curr_active_slave != slave) {
2900 /* this slave has just come up but we
2901 * already have a current slave; this
2902 * can also happen if bond_enslave adds
2903 * a new slave that is up while we are
2904 * searching for a new slave
2906 bond_set_slave_inactive_flags(slave);
2907 bond->current_arp_slave = NULL;
2910 if (slave == bond->curr_active_slave) {
2911 printk(KERN_INFO DRV_NAME
2912 ": %s: %s is up and now the "
2913 "active interface\n",
2917 printk(KERN_INFO DRV_NAME
2918 ": %s: backup interface %s is "
2924 write_unlock(&bond->curr_slave_lock);
2927 read_lock(&bond->curr_slave_lock);
2929 if ((slave != bond->curr_active_slave) &&
2930 (!bond->current_arp_slave) &&
2931 (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
2933 /* a backup slave has gone down; three times
2934 * the delta allows the current slave to be
2935 * taken out before the backup slave.
2936 * note: a non-null current_arp_slave indicates
2937 * the curr_active_slave went down and we are
2938 * searching for a new one; under this
2939 * condition we only take the curr_active_slave
2940 * down - this gives each slave a chance to
2941 * tx/rx traffic before being taken out
2944 read_unlock(&bond->curr_slave_lock);
2946 slave->link = BOND_LINK_DOWN;
2948 if (slave->link_failure_count < UINT_MAX) {
2949 slave->link_failure_count++;
2952 bond_set_slave_inactive_flags(slave);
2954 printk(KERN_INFO DRV_NAME
2955 ": %s: backup interface %s is now down\n",
2959 read_unlock(&bond->curr_slave_lock);
2964 read_lock(&bond->curr_slave_lock);
2965 slave = bond->curr_active_slave;
2966 read_unlock(&bond->curr_slave_lock);
2969 /* if we have sent traffic in the past 2*arp_intervals but
2970 * haven't xmit and rx traffic in that time interval, select
2971 * a different slave. slave->jiffies is only updated when
2972 * a slave first becomes the curr_active_slave - not necessarily
2973 * after every arp; this ensures the slave has a full 2*delta
2974 * before being taken out. if a primary is being used, check
2975 * if it is up and needs to take over as the curr_active_slave
2977 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2978 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2980 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2982 slave->link = BOND_LINK_DOWN;
2984 if (slave->link_failure_count < UINT_MAX) {
2985 slave->link_failure_count++;
2988 printk(KERN_INFO DRV_NAME
2989 ": %s: link status down for active interface "
2990 "%s, disabling it\n",
2994 write_lock(&bond->curr_slave_lock);
2996 bond_select_active_slave(bond);
2997 slave = bond->curr_active_slave;
2999 write_unlock(&bond->curr_slave_lock);
3001 bond->current_arp_slave = slave;
3004 slave->jiffies = jiffies;
3006 } else if ((bond->primary_slave) &&
3007 (bond->primary_slave != slave) &&
3008 (bond->primary_slave->link == BOND_LINK_UP)) {
3009 /* at this point, slave is the curr_active_slave */
3010 printk(KERN_INFO DRV_NAME
3011 ": %s: changing from interface %s to primary "
3015 bond->primary_slave->dev->name);
3017 /* primary is up so switch to it */
3018 write_lock(&bond->curr_slave_lock);
3019 bond_change_active_slave(bond, bond->primary_slave);
3020 write_unlock(&bond->curr_slave_lock);
3022 slave = bond->primary_slave;
3023 slave->jiffies = jiffies;
3025 bond->current_arp_slave = NULL;
3028 /* the current slave must tx an arp to ensure backup slaves
3031 if (slave && my_ip) {
3032 bond_arp_send_all(bond, slave);
3036 /* if we don't have a curr_active_slave, search for the next available
3037 * backup slave from the current_arp_slave and make it the candidate
3038 * for becoming the curr_active_slave
3041 if (!bond->current_arp_slave) {
3042 bond->current_arp_slave = bond->first_slave;
3045 if (bond->current_arp_slave) {
3046 bond_set_slave_inactive_flags(bond->current_arp_slave);
3048 /* search for next candidate */
3049 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave) {
3050 if (IS_UP(slave->dev)) {
3051 slave->link = BOND_LINK_BACK;
3052 bond_set_slave_active_flags(slave);
3053 bond_arp_send_all(bond, slave);
3054 slave->jiffies = jiffies;
3055 bond->current_arp_slave = slave;
3059 /* if the link state is up at this point, we
3060 * mark it down - this can happen if we have
3061 * simultaneous link failures and
3062 * reselect_active_interface doesn't make this
3063 * one the current slave so it is still marked
3064 * up when it is actually down
3066 if (slave->link == BOND_LINK_UP) {
3067 slave->link = BOND_LINK_DOWN;
3068 if (slave->link_failure_count < UINT_MAX) {
3069 slave->link_failure_count++;
3072 bond_set_slave_inactive_flags(slave);
3074 printk(KERN_INFO DRV_NAME
3075 ": %s: backup interface %s is "
3085 if (bond->params.arp_interval) {
3086 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
3089 read_unlock(&bond->lock);
3092 /*------------------------------ proc/seq_file-------------------------------*/
3094 #ifdef CONFIG_PROC_FS
3096 #define SEQ_START_TOKEN ((void *)1)
3098 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3100 struct bonding *bond = seq->private;
3102 struct slave *slave;
3105 /* make sure the bond won't be taken away */
3106 read_lock(&dev_base_lock);
3107 read_lock_bh(&bond->lock);
3110 return SEQ_START_TOKEN;
3113 bond_for_each_slave(bond, slave, i) {
3114 if (++off == *pos) {
3122 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3124 struct bonding *bond = seq->private;
3125 struct slave *slave = v;
3128 if (v == SEQ_START_TOKEN) {
3129 return bond->first_slave;
3132 slave = slave->next;
3134 return (slave == bond->first_slave) ? NULL : slave;
3137 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3139 struct bonding *bond = seq->private;
3141 read_unlock_bh(&bond->lock);
3142 read_unlock(&dev_base_lock);
3145 static void bond_info_show_master(struct seq_file *seq)
3147 struct bonding *bond = seq->private;
3150 read_lock(&bond->curr_slave_lock);
3151 curr = bond->curr_active_slave;
3152 read_unlock(&bond->curr_slave_lock);
3154 seq_printf(seq, "Bonding Mode: %s\n",
3155 bond_mode_name(bond->params.mode));
3157 if (USES_PRIMARY(bond->params.mode)) {
3158 seq_printf(seq, "Primary Slave: %s\n",
3159 (bond->params.primary[0]) ?
3160 bond->params.primary : "None");
3162 seq_printf(seq, "Currently Active Slave: %s\n",
3163 (curr) ? curr->dev->name : "None");
3166 seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down");
3167 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3168 seq_printf(seq, "Up Delay (ms): %d\n",
3169 bond->params.updelay * bond->params.miimon);
3170 seq_printf(seq, "Down Delay (ms): %d\n",
3171 bond->params.downdelay * bond->params.miimon);
3173 if (bond->params.mode == BOND_MODE_8023AD) {
3174 struct ad_info ad_info;
3176 seq_puts(seq, "\n802.3ad info\n");
3177 seq_printf(seq, "LACP rate: %s\n",
3178 (bond->params.lacp_fast) ? "fast" : "slow");
3180 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3181 seq_printf(seq, "bond %s has no active aggregator\n",
3184 seq_printf(seq, "Active Aggregator Info:\n");
3186 seq_printf(seq, "\tAggregator ID: %d\n",
3187 ad_info.aggregator_id);
3188 seq_printf(seq, "\tNumber of ports: %d\n",
3190 seq_printf(seq, "\tActor Key: %d\n",
3192 seq_printf(seq, "\tPartner Key: %d\n",
3193 ad_info.partner_key);
3194 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3195 ad_info.partner_system[0],
3196 ad_info.partner_system[1],
3197 ad_info.partner_system[2],
3198 ad_info.partner_system[3],
3199 ad_info.partner_system[4],
3200 ad_info.partner_system[5]);
3205 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3207 struct bonding *bond = seq->private;
3209 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3210 seq_printf(seq, "MII Status: %s\n",
3211 (slave->link == BOND_LINK_UP) ? "up" : "down");
3212 seq_printf(seq, "Link Failure Count: %d\n",
3213 slave->link_failure_count);
3215 if (app_abi_ver >= 1) {
3217 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3218 slave->perm_hwaddr[0],
3219 slave->perm_hwaddr[1],
3220 slave->perm_hwaddr[2],
3221 slave->perm_hwaddr[3],
3222 slave->perm_hwaddr[4],
3223 slave->perm_hwaddr[5]);
3226 if (bond->params.mode == BOND_MODE_8023AD) {
3227 const struct aggregator *agg
3228 = SLAVE_AD_INFO(slave).port.aggregator;
3231 seq_printf(seq, "Aggregator ID: %d\n",
3232 agg->aggregator_identifier);
3234 seq_puts(seq, "Aggregator ID: N/A\n");
3239 static int bond_info_seq_show(struct seq_file *seq, void *v)
3241 if (v == SEQ_START_TOKEN) {
3242 seq_printf(seq, "%s\n", version);
3243 bond_info_show_master(seq);
3245 bond_info_show_slave(seq, v);
3251 static struct seq_operations bond_info_seq_ops = {
3252 .start = bond_info_seq_start,
3253 .next = bond_info_seq_next,
3254 .stop = bond_info_seq_stop,
3255 .show = bond_info_seq_show,
3258 static int bond_info_open(struct inode *inode, struct file *file)
3260 struct seq_file *seq;
3261 struct proc_dir_entry *proc;
3264 res = seq_open(file, &bond_info_seq_ops);
3266 /* recover the pointer buried in proc_dir_entry data */
3267 seq = file->private_data;
3269 seq->private = proc->data;
3275 static struct file_operations bond_info_fops = {
3276 .owner = THIS_MODULE,
3277 .open = bond_info_open,
3279 .llseek = seq_lseek,
3280 .release = seq_release,
3283 static int bond_create_proc_entry(struct bonding *bond)
3285 struct net_device *bond_dev = bond->dev;
3287 if (bond_proc_dir) {
3288 bond->proc_entry = create_proc_entry(bond_dev->name,
3291 if (bond->proc_entry == NULL) {
3292 printk(KERN_WARNING DRV_NAME
3293 ": Warning: Cannot create /proc/net/%s/%s\n",
3294 DRV_NAME, bond_dev->name);
3296 bond->proc_entry->data = bond;
3297 bond->proc_entry->proc_fops = &bond_info_fops;
3298 bond->proc_entry->owner = THIS_MODULE;
3299 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3306 static void bond_remove_proc_entry(struct bonding *bond)
3308 if (bond_proc_dir && bond->proc_entry) {
3309 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3310 memset(bond->proc_file_name, 0, IFNAMSIZ);
3311 bond->proc_entry = NULL;
3315 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3316 * Caller must hold rtnl_lock.
3318 static void bond_create_proc_dir(void)
3320 int len = strlen(DRV_NAME);
3322 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3323 bond_proc_dir = bond_proc_dir->next) {
3324 if ((bond_proc_dir->namelen == len) &&
3325 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3330 if (!bond_proc_dir) {
3331 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3332 if (bond_proc_dir) {
3333 bond_proc_dir->owner = THIS_MODULE;
3335 printk(KERN_WARNING DRV_NAME
3336 ": Warning: cannot create /proc/net/%s\n",
3342 /* Destroy the bonding directory under /proc/net, if empty.
3343 * Caller must hold rtnl_lock.
3345 static void bond_destroy_proc_dir(void)
3347 struct proc_dir_entry *de;
3349 if (!bond_proc_dir) {
3353 /* verify that the /proc dir is empty */
3354 for (de = bond_proc_dir->subdir; de; de = de->next) {
3355 /* ignore . and .. */
3356 if (*(de->name) != '.') {
3362 if (bond_proc_dir->owner == THIS_MODULE) {
3363 bond_proc_dir->owner = NULL;
3366 remove_proc_entry(DRV_NAME, proc_net);
3367 bond_proc_dir = NULL;
3370 #endif /* CONFIG_PROC_FS */
3372 /*-------------------------- netdev event handling --------------------------*/
3375 * Change device name
3377 static int bond_event_changename(struct bonding *bond)
3379 #ifdef CONFIG_PROC_FS
3380 bond_remove_proc_entry(bond);
3381 bond_create_proc_entry(bond);
3387 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3389 struct bonding *event_bond = bond_dev->priv;
3392 case NETDEV_CHANGENAME:
3393 return bond_event_changename(event_bond);
3394 case NETDEV_UNREGISTER:
3396 * TODO: remove a bond from the list?
3406 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3408 struct net_device *bond_dev = slave_dev->master;
3411 case NETDEV_UNREGISTER:
3413 bond_release(bond_dev, slave_dev);
3418 * TODO: is this what we get if somebody
3419 * sets up a hierarchical bond, then rmmod's
3420 * one of the slave bonding devices?
3425 * ... Or is it this?
3428 case NETDEV_CHANGEMTU:
3430 * TODO: Should slaves be allowed to
3431 * independently alter their MTU? For
3432 * an active-backup bond, slaves need
3433 * not be the same type of device, so
3434 * MTUs may vary. For other modes,
3435 * slaves arguably should have the
3436 * same MTUs. To do this, we'd need to
3437 * take over the slave's change_mtu
3438 * function for the duration of their
3442 case NETDEV_CHANGENAME:
3444 * TODO: handle changing the primary's name
3455 * bond_netdev_event: handle netdev notifier chain events.
3457 * This function receives events for the netdev chain. The caller (an
3458 * ioctl handler calling notifier_call_chain) holds the necessary
3459 * locks for us to safely manipulate the slave devices (RTNL lock,
3462 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3464 struct net_device *event_dev = (struct net_device *)ptr;
3466 dprintk("event_dev: %s, event: %lx\n",
3467 (event_dev ? event_dev->name : "None"),
3470 if (event_dev->flags & IFF_MASTER) {
3471 dprintk("IFF_MASTER\n");
3472 return bond_master_netdev_event(event, event_dev);
3475 if (event_dev->flags & IFF_SLAVE) {
3476 dprintk("IFF_SLAVE\n");
3477 return bond_slave_netdev_event(event, event_dev);
3483 static struct notifier_block bond_netdev_notifier = {
3484 .notifier_call = bond_netdev_event,
3487 /*-------------------------- Packet type handling ---------------------------*/
3489 /* register to receive lacpdus on a bond */
3490 static void bond_register_lacpdu(struct bonding *bond)
3492 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3494 /* initialize packet type */
3495 pk_type->type = PKT_TYPE_LACPDU;
3496 pk_type->dev = bond->dev;
3497 pk_type->func = bond_3ad_lacpdu_recv;
3499 dev_add_pack(pk_type);
3502 /* unregister to receive lacpdus on a bond */
3503 static void bond_unregister_lacpdu(struct bonding *bond)
3505 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3508 /*-------------------------- Device entry points ----------------------------*/
3510 static int bond_open(struct net_device *bond_dev)
3512 struct bonding *bond = bond_dev->priv;
3513 struct timer_list *mii_timer = &bond->mii_timer;
3514 struct timer_list *arp_timer = &bond->arp_timer;
3516 bond->kill_timers = 0;
3518 if ((bond->params.mode == BOND_MODE_TLB) ||
3519 (bond->params.mode == BOND_MODE_ALB)) {
3520 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3522 /* bond_alb_initialize must be called before the timer
3525 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3526 /* something went wrong - fail the open operation */
3530 init_timer(alb_timer);
3531 alb_timer->expires = jiffies + 1;
3532 alb_timer->data = (unsigned long)bond;
3533 alb_timer->function = (void *)&bond_alb_monitor;
3534 add_timer(alb_timer);
3537 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3538 init_timer(mii_timer);
3539 mii_timer->expires = jiffies + 1;
3540 mii_timer->data = (unsigned long)bond_dev;
3541 mii_timer->function = (void *)&bond_mii_monitor;
3542 add_timer(mii_timer);
3545 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3546 init_timer(arp_timer);
3547 arp_timer->expires = jiffies + 1;
3548 arp_timer->data = (unsigned long)bond_dev;
3549 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3550 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3552 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3554 add_timer(arp_timer);
3557 if (bond->params.mode == BOND_MODE_8023AD) {
3558 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3559 init_timer(ad_timer);
3560 ad_timer->expires = jiffies + 1;
3561 ad_timer->data = (unsigned long)bond;
3562 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3563 add_timer(ad_timer);
3565 /* register to receive LACPDUs */
3566 bond_register_lacpdu(bond);
3572 static int bond_close(struct net_device *bond_dev)
3574 struct bonding *bond = bond_dev->priv;
3576 if (bond->params.mode == BOND_MODE_8023AD) {
3577 /* Unregister the receive of LACPDUs */
3578 bond_unregister_lacpdu(bond);
3581 write_lock_bh(&bond->lock);
3583 bond_mc_list_destroy(bond);
3585 /* signal timers not to re-arm */
3586 bond->kill_timers = 1;
3588 write_unlock_bh(&bond->lock);
3590 /* del_timer_sync must run without holding the bond->lock
3591 * because a running timer might be trying to hold it too
3594 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3595 del_timer_sync(&bond->mii_timer);
3598 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3599 del_timer_sync(&bond->arp_timer);
3602 switch (bond->params.mode) {
3603 case BOND_MODE_8023AD:
3604 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3608 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3614 /* Release the bonded slaves */
3615 bond_release_all(bond_dev);
3617 if ((bond->params.mode == BOND_MODE_TLB) ||
3618 (bond->params.mode == BOND_MODE_ALB)) {
3619 /* Must be called only after all
3620 * slaves have been released
3622 bond_alb_deinitialize(bond);
3628 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3630 struct bonding *bond = bond_dev->priv;
3631 struct net_device_stats *stats = &(bond->stats), *sstats;
3632 struct slave *slave;
3635 memset(stats, 0, sizeof(struct net_device_stats));
3637 read_lock_bh(&bond->lock);
3639 bond_for_each_slave(bond, slave, i) {
3640 sstats = slave->dev->get_stats(slave->dev);
3642 stats->rx_packets += sstats->rx_packets;
3643 stats->rx_bytes += sstats->rx_bytes;
3644 stats->rx_errors += sstats->rx_errors;
3645 stats->rx_dropped += sstats->rx_dropped;
3647 stats->tx_packets += sstats->tx_packets;
3648 stats->tx_bytes += sstats->tx_bytes;
3649 stats->tx_errors += sstats->tx_errors;
3650 stats->tx_dropped += sstats->tx_dropped;
3652 stats->multicast += sstats->multicast;
3653 stats->collisions += sstats->collisions;
3655 stats->rx_length_errors += sstats->rx_length_errors;
3656 stats->rx_over_errors += sstats->rx_over_errors;
3657 stats->rx_crc_errors += sstats->rx_crc_errors;
3658 stats->rx_frame_errors += sstats->rx_frame_errors;
3659 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3660 stats->rx_missed_errors += sstats->rx_missed_errors;
3662 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3663 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3664 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3665 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3666 stats->tx_window_errors += sstats->tx_window_errors;
3669 read_unlock_bh(&bond->lock);
3674 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3676 struct net_device *slave_dev = NULL;
3677 struct ifbond k_binfo;
3678 struct ifbond __user *u_binfo = NULL;
3679 struct ifslave k_sinfo;
3680 struct ifslave __user *u_sinfo = NULL;
3681 struct mii_ioctl_data *mii = NULL;
3682 int prev_abi_ver = orig_app_abi_ver;
3685 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3686 bond_dev->name, cmd);
3690 return bond_ethtool_ioctl(bond_dev, ifr);
3700 * We do this again just in case we were called by SIOCGMIIREG
3701 * instead of SIOCGMIIPHY.
3708 if (mii->reg_num == 1) {
3709 struct bonding *bond = bond_dev->priv;
3711 read_lock_bh(&bond->lock);
3712 read_lock(&bond->curr_slave_lock);
3713 if (bond->curr_active_slave) {
3714 mii->val_out = BMSR_LSTATUS;
3716 read_unlock(&bond->curr_slave_lock);
3717 read_unlock_bh(&bond->lock);
3721 case BOND_INFO_QUERY_OLD:
3722 case SIOCBONDINFOQUERY:
3723 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3725 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3729 res = bond_info_query(bond_dev, &k_binfo);
3731 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3737 case BOND_SLAVE_INFO_QUERY_OLD:
3738 case SIOCBONDSLAVEINFOQUERY:
3739 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3741 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3745 res = bond_slave_info_query(bond_dev, &k_sinfo);
3747 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3758 if (!capable(CAP_NET_ADMIN)) {
3762 if (orig_app_abi_ver == -1) {
3763 /* no orig_app_abi_ver was provided yet, so we'll use the
3764 * current one from now on, even if it's 0
3766 orig_app_abi_ver = app_abi_ver;
3768 } else if (orig_app_abi_ver != app_abi_ver) {
3769 printk(KERN_ERR DRV_NAME
3770 ": Error: already using ifenslave ABI version %d; to "
3771 "upgrade ifenslave to version %d, you must first "
3772 "reload bonding.\n",
3773 orig_app_abi_ver, app_abi_ver);
3777 slave_dev = dev_get_by_name(ifr->ifr_slave);
3779 dprintk("slave_dev=%p: \n", slave_dev);
3784 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3786 case BOND_ENSLAVE_OLD:
3787 case SIOCBONDENSLAVE:
3788 res = bond_enslave(bond_dev, slave_dev);
3790 case BOND_RELEASE_OLD:
3791 case SIOCBONDRELEASE:
3792 res = bond_release(bond_dev, slave_dev);
3794 case BOND_SETHWADDR_OLD:
3795 case SIOCBONDSETHWADDR:
3796 res = bond_sethwaddr(bond_dev, slave_dev);
3798 case BOND_CHANGE_ACTIVE_OLD:
3799 case SIOCBONDCHANGEACTIVE:
3800 res = bond_ioctl_change_active(bond_dev, slave_dev);
3810 /* The ioctl failed, so there's no point in changing the
3811 * orig_app_abi_ver. We'll restore it's value just in case
3812 * we've changed it earlier in this function.
3814 orig_app_abi_ver = prev_abi_ver;
3820 static void bond_set_multicast_list(struct net_device *bond_dev)
3822 struct bonding *bond = bond_dev->priv;
3823 struct dev_mc_list *dmi;
3825 write_lock_bh(&bond->lock);
3828 * Do promisc before checking multicast_mode
3830 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3831 bond_set_promiscuity(bond, 1);
3834 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3835 bond_set_promiscuity(bond, -1);
3838 /* set allmulti flag to slaves */
3839 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3840 bond_set_allmulti(bond, 1);
3843 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3844 bond_set_allmulti(bond, -1);
3847 bond->flags = bond_dev->flags;
3849 /* looking for addresses to add to slaves' mc list */
3850 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3851 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3852 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3856 /* looking for addresses to delete from slaves' list */
3857 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3858 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3859 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3863 /* save master's multicast list */
3864 bond_mc_list_destroy(bond);
3865 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3867 write_unlock_bh(&bond->lock);
3871 * Change the MTU of all of a master's slaves to match the master
3873 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3875 struct bonding *bond = bond_dev->priv;
3876 struct slave *slave, *stop_at;
3880 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3881 (bond_dev ? bond_dev->name : "None"), new_mtu);
3883 /* Can't hold bond->lock with bh disabled here since
3884 * some base drivers panic. On the other hand we can't
3885 * hold bond->lock without bh disabled because we'll
3886 * deadlock. The only solution is to rely on the fact
3887 * that we're under rtnl_lock here, and the slaves
3888 * list won't change. This doesn't solve the problem
3889 * of setting the slave's MTU while it is
3890 * transmitting, but the assumption is that the base
3891 * driver can handle that.
3893 * TODO: figure out a way to safely iterate the slaves
3894 * list, but without holding a lock around the actual
3895 * call to the base driver.
3898 bond_for_each_slave(bond, slave, i) {
3899 dprintk("s %p s->p %p c_m %p\n", slave,
3900 slave->prev, slave->dev->change_mtu);
3901 if (slave->dev->change_mtu) {
3902 res = slave->dev->change_mtu(slave->dev, new_mtu);
3904 slave->dev->mtu = new_mtu;
3909 /* If we failed to set the slave's mtu to the new value
3910 * we must abort the operation even in ACTIVE_BACKUP
3911 * mode, because if we allow the backup slaves to have
3912 * different mtu values than the active slave we'll
3913 * need to change their mtu when doing a failover. That
3914 * means changing their mtu from timer context, which
3915 * is probably not a good idea.
3917 dprintk("err %d %s\n", res, slave->dev->name);
3922 bond_dev->mtu = new_mtu;
3927 /* unwind from head to the slave that failed */
3929 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3932 if (slave->dev->change_mtu) {
3933 tmp_res = slave->dev->change_mtu(slave->dev, bond_dev->mtu);
3935 dprintk("unwind err %d dev %s\n", tmp_res,
3939 slave->dev->mtu = bond_dev->mtu;
3949 * Note that many devices must be down to change the HW address, and
3950 * downing the master releases all slaves. We can make bonds full of
3951 * bonding devices to test this, however.
3953 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3955 struct bonding *bond = bond_dev->priv;
3956 struct sockaddr *sa = addr, tmp_sa;
3957 struct slave *slave, *stop_at;
3961 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3963 if (!is_valid_ether_addr(sa->sa_data)) {
3964 return -EADDRNOTAVAIL;
3967 /* Can't hold bond->lock with bh disabled here since
3968 * some base drivers panic. On the other hand we can't
3969 * hold bond->lock without bh disabled because we'll
3970 * deadlock. The only solution is to rely on the fact
3971 * that we're under rtnl_lock here, and the slaves
3972 * list won't change. This doesn't solve the problem
3973 * of setting the slave's hw address while it is
3974 * transmitting, but the assumption is that the base
3975 * driver can handle that.
3977 * TODO: figure out a way to safely iterate the slaves
3978 * list, but without holding a lock around the actual
3979 * call to the base driver.
3982 bond_for_each_slave(bond, slave, i) {
3983 dprintk("slave %p %s\n", slave, slave->dev->name);
3985 if (slave->dev->set_mac_address == NULL) {
3987 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3991 res = slave->dev->set_mac_address(slave->dev, addr);
3993 /* TODO: consider downing the slave
3995 * User should expect communications
3996 * breakage anyway until ARP finish
3999 dprintk("err %d %s\n", res, slave->dev->name);
4005 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4009 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4010 tmp_sa.sa_family = bond_dev->type;
4012 /* unwind from head to the slave that failed */
4014 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4017 tmp_res = slave->dev->set_mac_address(slave->dev, &tmp_sa);
4019 dprintk("unwind err %d dev %s\n", tmp_res,
4027 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4029 struct bonding *bond = bond_dev->priv;
4030 struct slave *slave, *start_at;
4034 read_lock(&bond->lock);
4036 if (!BOND_IS_OK(bond)) {
4040 read_lock(&bond->curr_slave_lock);
4041 slave = start_at = bond->curr_active_slave;
4042 read_unlock(&bond->curr_slave_lock);
4048 bond_for_each_slave_from(bond, slave, i, start_at) {
4049 if (IS_UP(slave->dev) &&
4050 (slave->link == BOND_LINK_UP) &&
4051 (slave->state == BOND_STATE_ACTIVE)) {
4052 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4054 write_lock(&bond->curr_slave_lock);
4055 bond->curr_active_slave = slave->next;
4056 write_unlock(&bond->curr_slave_lock);
4065 /* no suitable interface, frame not sent */
4068 read_unlock(&bond->lock);
4073 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4074 * the bond has a usable interface.
4076 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4078 struct bonding *bond = bond_dev->priv;
4081 /* if we are sending arp packets, try to at least
4082 identify our own ip address */
4083 if (bond->params.arp_interval && !my_ip &&
4084 (skb->protocol == __constant_htons(ETH_P_ARP))) {
4085 char *the_ip = (char *)skb->data +
4086 sizeof(struct ethhdr) +
4087 sizeof(struct arphdr) +
4089 memcpy(&my_ip, the_ip, 4);
4092 read_lock(&bond->lock);
4093 read_lock(&bond->curr_slave_lock);
4095 if (!BOND_IS_OK(bond)) {
4099 if (bond->curr_active_slave) { /* one usable interface */
4100 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4105 /* no suitable interface, frame not sent */
4108 read_unlock(&bond->curr_slave_lock);
4109 read_unlock(&bond->lock);
4114 * in XOR mode, we determine the output device by performing xor on
4115 * the source and destination hw adresses. If this device is not
4116 * enabled, find the next slave following this xor slave.
4118 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4120 struct bonding *bond = bond_dev->priv;
4121 struct ethhdr *data = (struct ethhdr *)skb->data;
4122 struct slave *slave, *start_at;
4127 read_lock(&bond->lock);
4129 if (!BOND_IS_OK(bond)) {
4133 slave_no = (data->h_dest[5]^bond_dev->dev_addr[5]) % bond->slave_cnt;
4135 bond_for_each_slave(bond, slave, i) {
4144 bond_for_each_slave_from(bond, slave, i, start_at) {
4145 if (IS_UP(slave->dev) &&
4146 (slave->link == BOND_LINK_UP) &&
4147 (slave->state == BOND_STATE_ACTIVE)) {
4148 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4155 /* no suitable interface, frame not sent */
4158 read_unlock(&bond->lock);
4163 * in broadcast mode, we send everything to all usable interfaces.
4165 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4167 struct bonding *bond = bond_dev->priv;
4168 struct slave *slave, *start_at;
4169 struct net_device *tx_dev = NULL;
4173 read_lock(&bond->lock);
4175 if (!BOND_IS_OK(bond)) {
4179 read_lock(&bond->curr_slave_lock);
4180 start_at = bond->curr_active_slave;
4181 read_unlock(&bond->curr_slave_lock);
4187 bond_for_each_slave_from(bond, slave, i, start_at) {
4188 if (IS_UP(slave->dev) &&
4189 (slave->link == BOND_LINK_UP) &&
4190 (slave->state == BOND_STATE_ACTIVE)) {
4192 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4194 printk(KERN_ERR DRV_NAME
4195 ": Error: bond_xmit_broadcast(): "
4196 "skb_clone() failed\n");
4200 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4202 dev_kfree_skb(skb2);
4206 tx_dev = slave->dev;
4211 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4216 /* no suitable interface, frame not sent */
4219 /* frame sent to all suitable interfaces */
4220 read_unlock(&bond->lock);
4224 /*------------------------- Device initialization ---------------------------*/
4227 * set bond mode specific net device operations
4229 static inline void bond_set_mode_ops(struct net_device *bond_dev, int mode)
4232 case BOND_MODE_ROUNDROBIN:
4233 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4235 case BOND_MODE_ACTIVEBACKUP:
4236 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4239 bond_dev->hard_start_xmit = bond_xmit_xor;
4241 case BOND_MODE_BROADCAST:
4242 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4244 case BOND_MODE_8023AD:
4245 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4249 bond_dev->hard_start_xmit = bond_alb_xmit;
4250 bond_dev->set_mac_address = bond_alb_set_mac_address;
4253 /* Should never happen, mode already checked */
4254 printk(KERN_ERR DRV_NAME
4255 ": Error: Unknown bonding mode %d\n",
4262 * Does not allocate but creates a /proc entry.
4265 static int __init bond_init(struct net_device *bond_dev, struct bond_params *params)
4267 struct bonding *bond = bond_dev->priv;
4269 dprintk("Begin bond_init for %s\n", bond_dev->name);
4271 /* initialize rwlocks */
4272 rwlock_init(&bond->lock);
4273 rwlock_init(&bond->curr_slave_lock);
4275 bond->params = *params; /* copy params struct */
4277 /* Initialize pointers */
4278 bond->first_slave = NULL;
4279 bond->curr_active_slave = NULL;
4280 bond->current_arp_slave = NULL;
4281 bond->primary_slave = NULL;
4282 bond->dev = bond_dev;
4283 INIT_LIST_HEAD(&bond->vlan_list);
4285 /* Initialize the device entry points */
4286 bond_dev->open = bond_open;
4287 bond_dev->stop = bond_close;
4288 bond_dev->get_stats = bond_get_stats;
4289 bond_dev->do_ioctl = bond_do_ioctl;
4290 bond_dev->set_multicast_list = bond_set_multicast_list;
4291 bond_dev->change_mtu = bond_change_mtu;
4292 bond_dev->set_mac_address = bond_set_mac_address;
4294 bond_set_mode_ops(bond_dev, bond->params.mode);
4296 bond_dev->destructor = free_netdev;
4298 /* Initialize the device options */
4299 bond_dev->tx_queue_len = 0;
4300 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4302 /* At first, we block adding VLANs. That's the only way to
4303 * prevent problems that occur when adding VLANs over an
4304 * empty bond. The block will be removed once non-challenged
4305 * slaves are enslaved.
4307 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4309 /* By default, we declare the bond to be fully
4310 * VLAN hardware accelerated capable. Special
4311 * care is taken in the various xmit functions
4312 * when there are slaves that are not hw accel
4315 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4316 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4317 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4318 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4319 NETIF_F_HW_VLAN_RX |
4320 NETIF_F_HW_VLAN_FILTER);
4322 #ifdef CONFIG_PROC_FS
4323 bond_create_proc_entry(bond);
4326 list_add_tail(&bond->bond_list, &bond_dev_list);
4331 /* De-initialize device specific data.
4332 * Caller must hold rtnl_lock.
4334 static inline void bond_deinit(struct net_device *bond_dev)
4336 struct bonding *bond = bond_dev->priv;
4338 list_del(&bond->bond_list);
4340 #ifdef CONFIG_PROC_FS
4341 bond_remove_proc_entry(bond);
4345 /* Unregister and free all bond devices.
4346 * Caller must hold rtnl_lock.
4348 static void bond_free_all(void)
4350 struct bonding *bond, *nxt;
4352 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4353 struct net_device *bond_dev = bond->dev;
4355 unregister_netdevice(bond_dev);
4356 bond_deinit(bond_dev);
4359 #ifdef CONFIG_PROC_FS
4360 bond_destroy_proc_dir();
4364 /*------------------------- Module initialization ---------------------------*/
4367 * Convert string input module parms. Accept either the
4368 * number of the mode or its string name.
4370 static inline int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4374 for (i = 0; tbl[i].modename; i++) {
4375 if ((isdigit(*mode_arg) &&
4376 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4377 (strncmp(mode_arg, tbl[i].modename,
4378 strlen(tbl[i].modename)) == 0)) {
4386 static int bond_check_params(struct bond_params *params)
4389 * Convert string parameters.
4392 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4393 if (bond_mode == -1) {
4394 printk(KERN_ERR DRV_NAME
4395 ": Error: Invalid bonding mode \"%s\"\n",
4396 mode == NULL ? "NULL" : mode);
4402 if (bond_mode != BOND_MODE_8023AD) {
4403 printk(KERN_INFO DRV_NAME
4404 ": lacp_rate param is irrelevant in mode %s\n",
4405 bond_mode_name(bond_mode));
4407 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4408 if (lacp_fast == -1) {
4409 printk(KERN_ERR DRV_NAME
4410 ": Error: Invalid lacp rate \"%s\"\n",
4411 lacp_rate == NULL ? "NULL" : lacp_rate);
4417 if (max_bonds < 1 || max_bonds > INT_MAX) {
4418 printk(KERN_WARNING DRV_NAME
4419 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4420 "was reset to BOND_DEFAULT_MAX_BONDS (%d)",
4421 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4422 max_bonds = BOND_DEFAULT_MAX_BONDS;
4426 printk(KERN_WARNING DRV_NAME
4427 ": Warning: miimon module parameter (%d), "
4428 "not in range 0-%d, so it was reset to %d\n",
4429 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4430 miimon = BOND_LINK_MON_INTERV;
4434 printk(KERN_WARNING DRV_NAME
4435 ": Warning: updelay module parameter (%d), "
4436 "not in range 0-%d, so it was reset to 0\n",
4441 if (downdelay < 0) {
4442 printk(KERN_WARNING DRV_NAME
4443 ": Warning: downdelay module parameter (%d), "
4444 "not in range 0-%d, so it was reset to 0\n",
4445 downdelay, INT_MAX);
4449 if ((use_carrier != 0) && (use_carrier != 1)) {
4450 printk(KERN_WARNING DRV_NAME
4451 ": Warning: use_carrier module parameter (%d), "
4452 "not of valid value (0/1), so it was set to 1\n",
4457 /* reset values for 802.3ad */
4458 if (bond_mode == BOND_MODE_8023AD) {
4460 printk(KERN_WARNING DRV_NAME
4461 ": Warning: miimon must be specified, "
4462 "otherwise bonding will not detect link "
4463 "failure, speed and duplex which are "
4464 "essential for 802.3ad operation\n");
4465 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4470 /* reset values for TLB/ALB */
4471 if ((bond_mode == BOND_MODE_TLB) ||
4472 (bond_mode == BOND_MODE_ALB)) {
4474 printk(KERN_WARNING DRV_NAME
4475 ": Warning: miimon must be specified, "
4476 "otherwise bonding will not detect link "
4477 "failure and link speed which are essential "
4478 "for TLB/ALB load balancing\n");
4479 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4484 if (bond_mode == BOND_MODE_ALB) {
4485 printk(KERN_NOTICE DRV_NAME
4486 ": In ALB mode you might experience client "
4487 "disconnections upon reconnection of a link if the "
4488 "bonding module updelay parameter (%d msec) is "
4489 "incompatible with the forwarding delay time of the "
4495 if (updelay || downdelay) {
4496 /* just warn the user the up/down delay will have
4497 * no effect since miimon is zero...
4499 printk(KERN_WARNING DRV_NAME
4500 ": Warning: miimon module parameter not set "
4501 "and updelay (%d) or downdelay (%d) module "
4502 "parameter is set; updelay and downdelay have "
4503 "no effect unless miimon is set\n",
4504 updelay, downdelay);
4507 /* don't allow arp monitoring */
4509 printk(KERN_WARNING DRV_NAME
4510 ": Warning: miimon (%d) and arp_interval (%d) "
4511 "can't be used simultaneously, disabling ARP "
4513 miimon, arp_interval);
4517 if ((updelay % miimon) != 0) {
4518 printk(KERN_WARNING DRV_NAME
4519 ": Warning: updelay (%d) is not a multiple "
4520 "of miimon (%d), updelay rounded to %d ms\n",
4521 updelay, miimon, (updelay / miimon) * miimon);
4526 if ((downdelay % miimon) != 0) {
4527 printk(KERN_WARNING DRV_NAME
4528 ": Warning: downdelay (%d) is not a multiple "
4529 "of miimon (%d), downdelay rounded to %d ms\n",
4531 (downdelay / miimon) * miimon);
4534 downdelay /= miimon;
4537 if (arp_interval < 0) {
4538 printk(KERN_WARNING DRV_NAME
4539 ": Warning: arp_interval module parameter (%d) "
4540 ", not in range 0-%d, so it was reset to %d\n",
4541 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4542 arp_interval = BOND_LINK_ARP_INTERV;
4545 for (arp_ip_count = 0;
4546 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4548 /* not complete check, but should be good enough to
4550 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4551 printk(KERN_WARNING DRV_NAME
4552 ": Warning: bad arp_ip_target module parameter "
4553 "(%s), ARP monitoring will not be performed\n",
4554 arp_ip_target[arp_ip_count]);
4557 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4558 arp_target[arp_ip_count] = ip;
4562 if (arp_interval && !arp_ip_count) {
4563 /* don't allow arping if no arp_ip_target given... */
4564 printk(KERN_WARNING DRV_NAME
4565 ": Warning: arp_interval module parameter (%d) "
4566 "specified without providing an arp_ip_target "
4567 "parameter, arp_interval was reset to 0\n",
4573 printk(KERN_INFO DRV_NAME
4574 ": MII link monitoring set to %d ms\n",
4576 } else if (arp_interval) {
4579 printk(KERN_INFO DRV_NAME
4580 ": ARP monitoring set to %d ms with %d target(s):",
4581 arp_interval, arp_ip_count);
4583 for (i = 0; i < arp_ip_count; i++)
4584 printk (" %s", arp_ip_target[i]);
4589 /* miimon and arp_interval not set, we need one so things
4590 * work as expected, see bonding.txt for details
4592 printk(KERN_WARNING DRV_NAME
4593 ": Warning: either miimon or arp_interval and "
4594 "arp_ip_target module parameters must be specified, "
4595 "otherwise bonding will not detect link failures! see "
4596 "bonding.txt for details.\n");
4599 if (primary && !USES_PRIMARY(bond_mode)) {
4600 /* currently, using a primary only makes sense
4601 * in active backup, TLB or ALB modes
4603 printk(KERN_WARNING DRV_NAME
4604 ": Warning: %s primary device specified but has no "
4605 "effect in %s mode\n",
4606 primary, bond_mode_name(bond_mode));
4610 /* fill params struct with the proper values */
4611 params->mode = bond_mode;
4612 params->miimon = miimon;
4613 params->arp_interval = arp_interval;
4614 params->updelay = updelay;
4615 params->downdelay = downdelay;
4616 params->use_carrier = use_carrier;
4617 params->lacp_fast = lacp_fast;
4618 params->primary[0] = 0;
4621 strncpy(params->primary, primary, IFNAMSIZ);
4622 params->primary[IFNAMSIZ - 1] = 0;
4625 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4630 static int __init bonding_init(void)
4632 struct bond_params params;
4636 printk(KERN_INFO "%s", version);
4638 res = bond_check_params(¶ms);
4645 #ifdef CONFIG_PROC_FS
4646 bond_create_proc_dir();
4649 for (i = 0; i < max_bonds; i++) {
4650 struct net_device *bond_dev;
4652 bond_dev = alloc_netdev(sizeof(struct bonding), "", ether_setup);
4658 res = dev_alloc_name(bond_dev, "bond%d");
4660 free_netdev(bond_dev);
4664 /* bond_init() must be called after dev_alloc_name() (for the
4665 * /proc files), but before register_netdevice(), because we
4666 * need to set function pointers.
4668 res = bond_init(bond_dev, ¶ms);
4670 free_netdev(bond_dev);
4674 SET_MODULE_OWNER(bond_dev);
4676 res = register_netdevice(bond_dev);
4678 bond_deinit(bond_dev);
4679 free_netdev(bond_dev);
4685 register_netdevice_notifier(&bond_netdev_notifier);
4690 /* free and unregister all bonds that were successfully added */
4698 static void __exit bonding_exit(void)
4700 unregister_netdevice_notifier(&bond_netdev_notifier);
4707 module_init(bonding_init);
4708 module_exit(bonding_exit);
4709 MODULE_LICENSE("GPL");
4710 MODULE_VERSION(DRV_VERSION);
4711 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4712 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4713 MODULE_SUPPORTED_DEVICE("most ethernet devices");