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.
478 * 2005/06/05 - Jay Vosburgh <fubar@us.ibm.com>
479 * - Support for generating gratuitous ARPs in active-backup mode.
480 * Includes support for VLAN tagging all bonding-generated ARPs
481 * as needed. Set version to 2.6.2.
482 * 2005/06/08 - Jason Gabler <jygabler at lbl dot gov>
483 * - alternate hashing policy support for mode 2
484 * * Added kernel parameter "xmit_hash_policy" to allow the selection
485 * of different hashing policies for mode 2. The original mode 2
486 * policy is the default, now found in xmit_hash_policy_layer2().
487 * * Added xmit_hash_policy_layer34()
488 * - Modified by Jay Vosburgh <fubar@us.ibm.com> to also support mode 4.
489 * Set version to 2.6.3.
490 * 2005/09/26 - Jay Vosburgh <fubar@us.ibm.com>
491 * - Removed backwards compatibility for old ifenslaves. Version 2.6.4.
492 * 2005/09/27 - Mitch Williams <mitch.a.williams at intel dot com>
493 * - Radheka Godse <radheka.godse at intel dot com>
494 * - Split out bond creation code to allow for sysfs interface.
495 * - Removed static declaration on some functions and data items.
496 * - Added sysfs support, including capability to add/remove/change
497 * any bond at runtime.
500 * - Added bonding: <bondname>: prefix to sysfs log messages
501 * - Added arp_ip_targets to /proc entry
502 * - Allow ARP target table to have empty entries
503 * - trivial fix: added missing modes description to modinfo
504 * - Corrected bug in ALB init where kmalloc is called inside
506 * - Corrected behavior to maintain bond link when changing
507 * from arp monitor to miimon and vice versa
508 * - Added missing bonding: <bondname>: prefix to alb, ad log messages
509 * - Fixed stack dump warnings seen if changing between miimon
510 * and arp monitoring when the bond interface is down.
511 * - Fixed stack dump warnings seen when enslaving an e100
513 * - Set version to 3.0.0
516 //#define BONDING_DEBUG 1
518 #include <linux/config.h>
519 #include <linux/kernel.h>
520 #include <linux/module.h>
521 #include <linux/sched.h>
522 #include <linux/types.h>
523 #include <linux/fcntl.h>
524 #include <linux/interrupt.h>
525 #include <linux/ptrace.h>
526 #include <linux/ioport.h>
527 #include <linux/in.h>
529 #include <linux/ip.h>
530 #include <linux/tcp.h>
531 #include <linux/udp.h>
532 #include <linux/slab.h>
533 #include <linux/string.h>
534 #include <linux/init.h>
535 #include <linux/timer.h>
536 #include <linux/socket.h>
537 #include <linux/ctype.h>
538 #include <linux/inet.h>
539 #include <linux/bitops.h>
540 #include <asm/system.h>
543 #include <asm/uaccess.h>
544 #include <linux/errno.h>
545 #include <linux/netdevice.h>
546 #include <linux/inetdevice.h>
547 #include <linux/etherdevice.h>
548 #include <linux/skbuff.h>
549 #include <net/sock.h>
550 #include <linux/rtnetlink.h>
551 #include <linux/proc_fs.h>
552 #include <linux/seq_file.h>
553 #include <linux/smp.h>
554 #include <linux/if_ether.h>
556 #include <linux/mii.h>
557 #include <linux/ethtool.h>
558 #include <linux/if_vlan.h>
559 #include <linux/if_bonding.h>
560 #include <net/route.h>
562 #include "bond_3ad.h"
563 #include "bond_alb.h"
565 /*---------------------------- Module parameters ----------------------------*/
567 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
568 #define BOND_LINK_MON_INTERV 0
569 #define BOND_LINK_ARP_INTERV 0
571 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
572 static int miimon = BOND_LINK_MON_INTERV;
573 static int updelay = 0;
574 static int downdelay = 0;
575 static int use_carrier = 1;
576 static char *mode = NULL;
577 static char *primary = NULL;
578 static char *lacp_rate = NULL;
579 static char *xmit_hash_policy = NULL;
580 static int arp_interval = BOND_LINK_ARP_INTERV;
581 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
582 struct bond_params bonding_defaults;
584 module_param(max_bonds, int, 0);
585 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
586 module_param(miimon, int, 0);
587 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
588 module_param(updelay, int, 0);
589 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
590 module_param(downdelay, int, 0);
591 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
593 module_param(use_carrier, int, 0);
594 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
595 "0 for off, 1 for on (default)");
596 module_param(mode, charp, 0);
597 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
598 "1 for active-backup, 2 for balance-xor, "
599 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
600 "6 for balance-alb");
601 module_param(primary, charp, 0);
602 MODULE_PARM_DESC(primary, "Primary network device to use");
603 module_param(lacp_rate, charp, 0);
604 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
606 module_param(xmit_hash_policy, charp, 0);
607 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
608 ", 1 for layer 3+4");
609 module_param(arp_interval, int, 0);
610 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
611 module_param_array(arp_ip_target, charp, NULL, 0);
612 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
614 /*----------------------------- Global variables ----------------------------*/
616 static const char *version =
617 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
619 LIST_HEAD(bond_dev_list);
621 #ifdef CONFIG_PROC_FS
622 static struct proc_dir_entry *bond_proc_dir = NULL;
625 extern struct rw_semaphore bonding_rwsem;
626 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
627 static int arp_ip_count = 0;
628 static int bond_mode = BOND_MODE_ROUNDROBIN;
629 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
630 static int lacp_fast = 0;
633 struct bond_parm_tbl bond_lacp_tbl[] = {
634 { "slow", AD_LACP_SLOW},
635 { "fast", AD_LACP_FAST},
639 struct bond_parm_tbl bond_mode_tbl[] = {
640 { "balance-rr", BOND_MODE_ROUNDROBIN},
641 { "active-backup", BOND_MODE_ACTIVEBACKUP},
642 { "balance-xor", BOND_MODE_XOR},
643 { "broadcast", BOND_MODE_BROADCAST},
644 { "802.3ad", BOND_MODE_8023AD},
645 { "balance-tlb", BOND_MODE_TLB},
646 { "balance-alb", BOND_MODE_ALB},
650 struct bond_parm_tbl xmit_hashtype_tbl[] = {
651 { "layer2", BOND_XMIT_POLICY_LAYER2},
652 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
656 /*-------------------------- Forward declarations ---------------------------*/
658 static void bond_send_gratuitous_arp(struct bonding *bond);
660 /*---------------------------- General routines -----------------------------*/
662 const char *bond_mode_name(int mode)
665 case BOND_MODE_ROUNDROBIN :
666 return "load balancing (round-robin)";
667 case BOND_MODE_ACTIVEBACKUP :
668 return "fault-tolerance (active-backup)";
670 return "load balancing (xor)";
671 case BOND_MODE_BROADCAST :
672 return "fault-tolerance (broadcast)";
673 case BOND_MODE_8023AD:
674 return "IEEE 802.3ad Dynamic link aggregation";
676 return "transmit load balancing";
678 return "adaptive load balancing";
684 /*---------------------------------- VLAN -----------------------------------*/
687 * bond_add_vlan - add a new vlan id on bond
688 * @bond: bond that got the notification
689 * @vlan_id: the vlan id to add
691 * Returns -ENOMEM if allocation failed.
693 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
695 struct vlan_entry *vlan;
697 dprintk("bond: %s, vlan id %d\n",
698 (bond ? bond->dev->name: "None"), vlan_id);
700 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
705 INIT_LIST_HEAD(&vlan->vlan_list);
706 vlan->vlan_id = vlan_id;
709 write_lock_bh(&bond->lock);
711 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
713 write_unlock_bh(&bond->lock);
715 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
721 * bond_del_vlan - delete a vlan id from bond
722 * @bond: bond that got the notification
723 * @vlan_id: the vlan id to delete
725 * returns -ENODEV if @vlan_id was not found in @bond.
727 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
729 struct vlan_entry *vlan, *next;
732 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
734 write_lock_bh(&bond->lock);
736 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
737 if (vlan->vlan_id == vlan_id) {
738 list_del(&vlan->vlan_list);
740 if ((bond->params.mode == BOND_MODE_TLB) ||
741 (bond->params.mode == BOND_MODE_ALB)) {
742 bond_alb_clear_vlan(bond, vlan_id);
745 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
750 if (list_empty(&bond->vlan_list) &&
751 (bond->slave_cnt == 0)) {
752 /* Last VLAN removed and no slaves, so
753 * restore block on adding VLANs. This will
754 * be removed once new slaves that are not
755 * VLAN challenged will be added.
757 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
765 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
769 write_unlock_bh(&bond->lock);
774 * bond_has_challenged_slaves
775 * @bond: the bond we're working on
777 * Searches the slave list. Returns 1 if a vlan challenged slave
778 * was found, 0 otherwise.
780 * Assumes bond->lock is held.
782 static int bond_has_challenged_slaves(struct bonding *bond)
787 bond_for_each_slave(bond, slave, i) {
788 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
789 dprintk("found VLAN challenged slave - %s\n",
795 dprintk("no VLAN challenged slaves found\n");
800 * bond_next_vlan - safely skip to the next item in the vlans list.
801 * @bond: the bond we're working on
802 * @curr: item we're advancing from
804 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
805 * or @curr->next otherwise (even if it is @curr itself again).
807 * Caller must hold bond->lock
809 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
811 struct vlan_entry *next, *last;
813 if (list_empty(&bond->vlan_list)) {
818 next = list_entry(bond->vlan_list.next,
819 struct vlan_entry, vlan_list);
821 last = list_entry(bond->vlan_list.prev,
822 struct vlan_entry, vlan_list);
824 next = list_entry(bond->vlan_list.next,
825 struct vlan_entry, vlan_list);
827 next = list_entry(curr->vlan_list.next,
828 struct vlan_entry, vlan_list);
836 * bond_dev_queue_xmit - Prepare skb for xmit.
838 * @bond: bond device that got this skb for tx.
839 * @skb: hw accel VLAN tagged skb to transmit
840 * @slave_dev: slave that is supposed to xmit this skbuff
842 * When the bond gets an skb to transmit that is
843 * already hardware accelerated VLAN tagged, and it
844 * needs to relay this skb to a slave that is not
845 * hw accel capable, the skb needs to be "unaccelerated",
846 * i.e. strip the hwaccel tag and re-insert it as part
849 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
851 unsigned short vlan_id;
853 if (!list_empty(&bond->vlan_list) &&
854 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
855 vlan_get_tag(skb, &vlan_id) == 0) {
856 skb->dev = slave_dev;
857 skb = vlan_put_tag(skb, vlan_id);
859 /* vlan_put_tag() frees the skb in case of error,
860 * so return success here so the calling functions
861 * won't attempt to free is again.
866 skb->dev = slave_dev;
876 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
877 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
879 * a. This operation is performed in IOCTL context,
880 * b. The operation is protected by the RTNL semaphore in the 8021q code,
881 * c. Holding a lock with BH disabled while directly calling a base driver
882 * entry point is generally a BAD idea.
884 * The design of synchronization/protection for this operation in the 8021q
885 * module is good for one or more VLAN devices over a single physical device
886 * and cannot be extended for a teaming solution like bonding, so there is a
887 * potential race condition here where a net device from the vlan group might
888 * be referenced (either by a base driver or the 8021q code) while it is being
889 * removed from the system. However, it turns out we're not making matters
890 * worse, and if it works for regular VLAN usage it will work here too.
894 * bond_vlan_rx_register - Propagates registration to slaves
895 * @bond_dev: bonding net device that got called
896 * @grp: vlan group being registered
898 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
900 struct bonding *bond = bond_dev->priv;
906 bond_for_each_slave(bond, slave, i) {
907 struct net_device *slave_dev = slave->dev;
909 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
910 slave_dev->vlan_rx_register) {
911 slave_dev->vlan_rx_register(slave_dev, grp);
917 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
918 * @bond_dev: bonding net device that got called
919 * @vid: vlan id being added
921 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
923 struct bonding *bond = bond_dev->priv;
927 bond_for_each_slave(bond, slave, i) {
928 struct net_device *slave_dev = slave->dev;
930 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
931 slave_dev->vlan_rx_add_vid) {
932 slave_dev->vlan_rx_add_vid(slave_dev, vid);
936 res = bond_add_vlan(bond, vid);
938 printk(KERN_ERR DRV_NAME
939 ": %s: Error: Failed to add vlan id %d\n",
940 bond_dev->name, vid);
945 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
946 * @bond_dev: bonding net device that got called
947 * @vid: vlan id being removed
949 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
951 struct bonding *bond = bond_dev->priv;
953 struct net_device *vlan_dev;
956 bond_for_each_slave(bond, slave, i) {
957 struct net_device *slave_dev = slave->dev;
959 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
960 slave_dev->vlan_rx_kill_vid) {
961 /* Save and then restore vlan_dev in the grp array,
962 * since the slave's driver might clear it.
964 vlan_dev = bond->vlgrp->vlan_devices[vid];
965 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
966 bond->vlgrp->vlan_devices[vid] = vlan_dev;
970 res = bond_del_vlan(bond, vid);
972 printk(KERN_ERR DRV_NAME
973 ": %s: Error: Failed to remove vlan id %d\n",
974 bond_dev->name, vid);
978 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
980 struct vlan_entry *vlan;
982 write_lock_bh(&bond->lock);
984 if (list_empty(&bond->vlan_list)) {
988 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
989 slave_dev->vlan_rx_register) {
990 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
993 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
994 !(slave_dev->vlan_rx_add_vid)) {
998 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
999 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
1003 write_unlock_bh(&bond->lock);
1006 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
1008 struct vlan_entry *vlan;
1009 struct net_device *vlan_dev;
1011 write_lock_bh(&bond->lock);
1013 if (list_empty(&bond->vlan_list)) {
1017 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
1018 !(slave_dev->vlan_rx_kill_vid)) {
1022 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
1023 /* Save and then restore vlan_dev in the grp array,
1024 * since the slave's driver might clear it.
1026 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
1027 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
1028 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
1032 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
1033 slave_dev->vlan_rx_register) {
1034 slave_dev->vlan_rx_register(slave_dev, NULL);
1038 write_unlock_bh(&bond->lock);
1041 /*------------------------------- Link status -------------------------------*/
1044 * Get link speed and duplex from the slave's base driver
1045 * using ethtool. If for some reason the call fails or the
1046 * values are invalid, fake speed and duplex to 100/Full
1049 static int bond_update_speed_duplex(struct slave *slave)
1051 struct net_device *slave_dev = slave->dev;
1052 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1054 struct ethtool_cmd etool;
1056 /* Fake speed and duplex */
1057 slave->speed = SPEED_100;
1058 slave->duplex = DUPLEX_FULL;
1060 if (slave_dev->ethtool_ops) {
1063 if (!slave_dev->ethtool_ops->get_settings) {
1067 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
1075 ioctl = slave_dev->do_ioctl;
1076 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1077 etool.cmd = ETHTOOL_GSET;
1078 ifr.ifr_data = (char*)&etool;
1079 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
1084 switch (etool.speed) {
1093 switch (etool.duplex) {
1101 slave->speed = etool.speed;
1102 slave->duplex = etool.duplex;
1108 * if <dev> supports MII link status reporting, check its link status.
1110 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
1111 * depening upon the setting of the use_carrier parameter.
1113 * Return either BMSR_LSTATUS, meaning that the link is up (or we
1114 * can't tell and just pretend it is), or 0, meaning that the link is
1117 * If reporting is non-zero, instead of faking link up, return -1 if
1118 * both ETHTOOL and MII ioctls fail (meaning the device does not
1119 * support them). If use_carrier is set, return whatever it says.
1120 * It'd be nice if there was a good way to tell if a driver supports
1121 * netif_carrier, but there really isn't.
1123 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
1125 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1127 struct mii_ioctl_data *mii;
1128 struct ethtool_value etool;
1130 if (bond->params.use_carrier) {
1131 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1134 ioctl = slave_dev->do_ioctl;
1136 /* TODO: set pointer to correct ioctl on a per team member */
1137 /* bases to make this more efficient. that is, once */
1138 /* we determine the correct ioctl, we will always */
1139 /* call it and not the others for that team */
1143 * We cannot assume that SIOCGMIIPHY will also read a
1144 * register; not all network drivers (e.g., e100)
1148 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
1149 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1151 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
1152 mii->reg_num = MII_BMSR;
1153 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
1154 return (mii->val_out & BMSR_LSTATUS);
1159 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
1160 /* for a period of time so we attempt to get link status */
1161 /* from it last if the above MII ioctls fail... */
1162 if (slave_dev->ethtool_ops) {
1163 if (slave_dev->ethtool_ops->get_link) {
1166 link = slave_dev->ethtool_ops->get_link(slave_dev);
1168 return link ? BMSR_LSTATUS : 0;
1173 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1174 etool.cmd = ETHTOOL_GLINK;
1175 ifr.ifr_data = (char*)&etool;
1176 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
1177 if (etool.data == 1) {
1178 return BMSR_LSTATUS;
1180 dprintk("SIOCETHTOOL shows link down\n");
1187 * If reporting, report that either there's no dev->do_ioctl,
1188 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
1189 * cannot report link status). If not reporting, pretend
1192 return (reporting ? -1 : BMSR_LSTATUS);
1195 /*----------------------------- Multicast list ------------------------------*/
1198 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
1200 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
1202 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
1203 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
1207 * returns dmi entry if found, NULL otherwise
1209 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
1211 struct dev_mc_list *idmi;
1213 for (idmi = mc_list; idmi; idmi = idmi->next) {
1214 if (bond_is_dmi_same(dmi, idmi)) {
1223 * Push the promiscuity flag down to appropriate slaves
1225 static void bond_set_promiscuity(struct bonding *bond, int inc)
1227 if (USES_PRIMARY(bond->params.mode)) {
1228 /* write lock already acquired */
1229 if (bond->curr_active_slave) {
1230 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
1233 struct slave *slave;
1235 bond_for_each_slave(bond, slave, i) {
1236 dev_set_promiscuity(slave->dev, inc);
1242 * Push the allmulti flag down to all slaves
1244 static void bond_set_allmulti(struct bonding *bond, int inc)
1246 if (USES_PRIMARY(bond->params.mode)) {
1247 /* write lock already acquired */
1248 if (bond->curr_active_slave) {
1249 dev_set_allmulti(bond->curr_active_slave->dev, inc);
1252 struct slave *slave;
1254 bond_for_each_slave(bond, slave, i) {
1255 dev_set_allmulti(slave->dev, inc);
1261 * Add a Multicast address to slaves
1264 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
1266 if (USES_PRIMARY(bond->params.mode)) {
1267 /* write lock already acquired */
1268 if (bond->curr_active_slave) {
1269 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
1272 struct slave *slave;
1274 bond_for_each_slave(bond, slave, i) {
1275 dev_mc_add(slave->dev, addr, alen, 0);
1281 * Remove a multicast address from slave
1284 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
1286 if (USES_PRIMARY(bond->params.mode)) {
1287 /* write lock already acquired */
1288 if (bond->curr_active_slave) {
1289 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
1292 struct slave *slave;
1294 bond_for_each_slave(bond, slave, i) {
1295 dev_mc_delete(slave->dev, addr, alen, 0);
1301 * Totally destroys the mc_list in bond
1303 static void bond_mc_list_destroy(struct bonding *bond)
1305 struct dev_mc_list *dmi;
1307 dmi = bond->mc_list;
1309 bond->mc_list = dmi->next;
1311 dmi = bond->mc_list;
1316 * Copy all the Multicast addresses from src to the bonding device dst
1318 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
1321 struct dev_mc_list *dmi, *new_dmi;
1323 for (dmi = mc_list; dmi; dmi = dmi->next) {
1324 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1327 /* FIXME: Potential memory leak !!! */
1331 new_dmi->next = bond->mc_list;
1332 bond->mc_list = new_dmi;
1333 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
1334 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
1335 new_dmi->dmi_users = dmi->dmi_users;
1336 new_dmi->dmi_gusers = dmi->dmi_gusers;
1343 * flush all members of flush->mc_list from device dev->mc_list
1345 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
1347 struct bonding *bond = bond_dev->priv;
1348 struct dev_mc_list *dmi;
1350 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1351 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1354 if (bond->params.mode == BOND_MODE_8023AD) {
1355 /* del lacpdu mc addr from mc list */
1356 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1358 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1362 /*--------------------------- Active slave change ---------------------------*/
1365 * Update the mc list and multicast-related flags for the new and
1366 * old active slaves (if any) according to the multicast mode, and
1367 * promiscuous flags unconditionally.
1369 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
1371 struct dev_mc_list *dmi;
1373 if (!USES_PRIMARY(bond->params.mode)) {
1374 /* nothing to do - mc list is already up-to-date on
1381 if (bond->dev->flags & IFF_PROMISC) {
1382 dev_set_promiscuity(old_active->dev, -1);
1385 if (bond->dev->flags & IFF_ALLMULTI) {
1386 dev_set_allmulti(old_active->dev, -1);
1389 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1390 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1395 if (bond->dev->flags & IFF_PROMISC) {
1396 dev_set_promiscuity(new_active->dev, 1);
1399 if (bond->dev->flags & IFF_ALLMULTI) {
1400 dev_set_allmulti(new_active->dev, 1);
1403 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1404 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1410 * find_best_interface - select the best available slave to be the active one
1411 * @bond: our bonding struct
1413 * Warning: Caller must hold curr_slave_lock for writing.
1415 static struct slave *bond_find_best_slave(struct bonding *bond)
1417 struct slave *new_active, *old_active;
1418 struct slave *bestslave = NULL;
1419 int mintime = bond->params.updelay;
1422 new_active = old_active = bond->curr_active_slave;
1424 if (!new_active) { /* there were no active slaves left */
1425 if (bond->slave_cnt > 0) { /* found one slave */
1426 new_active = bond->first_slave;
1428 return NULL; /* still no slave, return NULL */
1432 /* first try the primary link; if arping, a link must tx/rx traffic
1433 * before it can be considered the curr_active_slave - also, we would skip
1434 * slaves between the curr_active_slave and primary_slave that may be up
1437 if ((bond->primary_slave) &&
1438 (!bond->params.arp_interval) &&
1439 (IS_UP(bond->primary_slave->dev))) {
1440 new_active = bond->primary_slave;
1443 /* remember where to stop iterating over the slaves */
1444 old_active = new_active;
1446 bond_for_each_slave_from(bond, new_active, i, old_active) {
1447 if (IS_UP(new_active->dev)) {
1448 if (new_active->link == BOND_LINK_UP) {
1450 } else if (new_active->link == BOND_LINK_BACK) {
1451 /* link up, but waiting for stabilization */
1452 if (new_active->delay < mintime) {
1453 mintime = new_active->delay;
1454 bestslave = new_active;
1464 * change_active_interface - change the active slave into the specified one
1465 * @bond: our bonding struct
1466 * @new: the new slave to make the active one
1468 * Set the new slave to the bond's settings and unset them on the old
1469 * curr_active_slave.
1470 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1472 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1473 * because it is apparently the best available slave we have, even though its
1474 * updelay hasn't timed out yet.
1476 * Warning: Caller must hold curr_slave_lock for writing.
1478 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1480 struct slave *old_active = bond->curr_active_slave;
1482 if (old_active == new_active) {
1487 if (new_active->link == BOND_LINK_BACK) {
1488 if (USES_PRIMARY(bond->params.mode)) {
1489 printk(KERN_INFO DRV_NAME
1490 ": %s: making interface %s the new "
1491 "active one %d ms earlier.\n",
1492 bond->dev->name, new_active->dev->name,
1493 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1496 new_active->delay = 0;
1497 new_active->link = BOND_LINK_UP;
1498 new_active->jiffies = jiffies;
1500 if (bond->params.mode == BOND_MODE_8023AD) {
1501 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1504 if ((bond->params.mode == BOND_MODE_TLB) ||
1505 (bond->params.mode == BOND_MODE_ALB)) {
1506 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1509 if (USES_PRIMARY(bond->params.mode)) {
1510 printk(KERN_INFO DRV_NAME
1511 ": %s: making interface %s the new "
1513 bond->dev->name, new_active->dev->name);
1518 if (USES_PRIMARY(bond->params.mode)) {
1519 bond_mc_swap(bond, new_active, old_active);
1522 if ((bond->params.mode == BOND_MODE_TLB) ||
1523 (bond->params.mode == BOND_MODE_ALB)) {
1524 bond_alb_handle_active_change(bond, new_active);
1526 bond->curr_active_slave = new_active;
1529 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1531 bond_set_slave_inactive_flags(old_active);
1535 bond_set_slave_active_flags(new_active);
1537 bond_send_gratuitous_arp(bond);
1542 * bond_select_active_slave - select a new active slave, if needed
1543 * @bond: our bonding struct
1545 * This functions shoud be called when one of the following occurs:
1546 * - The old curr_active_slave has been released or lost its link.
1547 * - The primary_slave has got its link back.
1548 * - A slave has got its link back and there's no old curr_active_slave.
1550 * Warning: Caller must hold curr_slave_lock for writing.
1552 void bond_select_active_slave(struct bonding *bond)
1554 struct slave *best_slave;
1556 best_slave = bond_find_best_slave(bond);
1557 if (best_slave != bond->curr_active_slave) {
1558 bond_change_active_slave(bond, best_slave);
1562 /*--------------------------- slave list handling ---------------------------*/
1565 * This function attaches the slave to the end of list.
1567 * bond->lock held for writing by caller.
1569 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1571 if (bond->first_slave == NULL) { /* attaching the first slave */
1572 new_slave->next = new_slave;
1573 new_slave->prev = new_slave;
1574 bond->first_slave = new_slave;
1576 new_slave->next = bond->first_slave;
1577 new_slave->prev = bond->first_slave->prev;
1578 new_slave->next->prev = new_slave;
1579 new_slave->prev->next = new_slave;
1586 * This function detaches the slave from the list.
1587 * WARNING: no check is made to verify if the slave effectively
1588 * belongs to <bond>.
1589 * Nothing is freed on return, structures are just unchained.
1590 * If any slave pointer in bond was pointing to <slave>,
1591 * it should be changed by the calling function.
1593 * bond->lock held for writing by caller.
1595 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1598 slave->next->prev = slave->prev;
1602 slave->prev->next = slave->next;
1605 if (bond->first_slave == slave) { /* slave is the first slave */
1606 if (bond->slave_cnt > 1) { /* there are more slave */
1607 bond->first_slave = slave->next;
1609 bond->first_slave = NULL; /* slave was the last one */
1618 /*---------------------------------- IOCTL ----------------------------------*/
1620 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1622 dprintk("bond_dev=%p\n", bond_dev);
1623 dprintk("slave_dev=%p\n", slave_dev);
1624 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1625 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1629 #define BOND_INTERSECT_FEATURES \
1630 (NETIF_F_SG|NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM)
1633 * Compute the common dev->feature set available to all slaves. Some
1634 * feature bits are managed elsewhere, so preserve feature bits set on
1635 * master device that are not part of the examined set.
1637 static int bond_compute_features(struct bonding *bond)
1639 unsigned long features = BOND_INTERSECT_FEATURES;
1640 struct slave *slave;
1641 struct net_device *bond_dev = bond->dev;
1644 bond_for_each_slave(bond, slave, i)
1645 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1647 if ((features & NETIF_F_SG) &&
1648 !(features & (NETIF_F_IP_CSUM |
1651 features &= ~NETIF_F_SG;
1653 features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1654 bond_dev->features = features;
1659 /* enslave device <slave> to bond device <master> */
1660 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1662 struct bonding *bond = bond_dev->priv;
1663 struct slave *new_slave = NULL;
1664 struct dev_mc_list *dmi;
1665 struct sockaddr addr;
1667 int old_features = bond_dev->features;
1670 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1671 slave_dev->do_ioctl == NULL) {
1672 printk(KERN_WARNING DRV_NAME
1673 ": %s: Warning: no link monitoring support for %s\n",
1674 bond_dev->name, slave_dev->name);
1677 /* bond must be initialized by bond_open() before enslaving */
1678 if (!(bond_dev->flags & IFF_UP)) {
1679 dprintk("Error, master_dev is not up\n");
1683 /* already enslaved */
1684 if (slave_dev->flags & IFF_SLAVE) {
1685 dprintk("Error, Device was already enslaved\n");
1689 /* vlan challenged mutual exclusion */
1690 /* no need to lock since we're protected by rtnl_lock */
1691 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1692 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1693 if (!list_empty(&bond->vlan_list)) {
1694 printk(KERN_ERR DRV_NAME
1695 ": %s: Error: cannot enslave VLAN "
1696 "challenged slave %s on VLAN enabled "
1697 "bond %s\n", bond_dev->name, slave_dev->name,
1701 printk(KERN_WARNING DRV_NAME
1702 ": %s: Warning: enslaved VLAN challenged "
1703 "slave %s. Adding VLANs will be blocked as "
1704 "long as %s is part of bond %s\n",
1705 bond_dev->name, slave_dev->name, slave_dev->name,
1707 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1710 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1711 if (bond->slave_cnt == 0) {
1712 /* First slave, and it is not VLAN challenged,
1713 * so remove the block of adding VLANs over the bond.
1715 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1720 * Old ifenslave binaries are no longer supported. These can
1721 * be identified with moderate accurary by the state of the slave:
1722 * the current ifenslave will set the interface down prior to
1723 * enslaving it; the old ifenslave will not.
1725 if ((slave_dev->flags & IFF_UP)) {
1726 printk(KERN_ERR DRV_NAME ": %s is up. "
1727 "This may be due to an out of date ifenslave.\n",
1730 goto err_undo_flags;
1733 if (slave_dev->set_mac_address == NULL) {
1734 printk(KERN_ERR DRV_NAME
1735 ": %s: Error: The slave device you specified does "
1736 "not support setting the MAC address. "
1737 "Your kernel likely does not support slave "
1738 "devices.\n", bond_dev->name);
1740 goto err_undo_flags;
1743 new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1746 goto err_undo_flags;
1749 memset(new_slave, 0, sizeof(struct slave));
1751 /* save slave's original flags before calling
1752 * netdev_set_master and dev_open
1754 new_slave->original_flags = slave_dev->flags;
1757 * Save slave's original ("permanent") mac address for modes
1758 * that need it, and for restoring it upon release, and then
1759 * set it to the master's address
1761 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1764 * Set slave to master's mac address. The application already
1765 * set the master's mac address to that of the first slave
1767 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1768 addr.sa_family = slave_dev->type;
1769 res = dev_set_mac_address(slave_dev, &addr);
1771 dprintk("Error %d calling set_mac_address\n", res);
1775 /* open the slave since the application closed it */
1776 res = dev_open(slave_dev);
1778 dprintk("Openning slave %s failed\n", slave_dev->name);
1779 goto err_restore_mac;
1782 res = netdev_set_master(slave_dev, bond_dev);
1784 dprintk("Error %d calling netdev_set_master\n", res);
1788 new_slave->dev = slave_dev;
1790 if ((bond->params.mode == BOND_MODE_TLB) ||
1791 (bond->params.mode == BOND_MODE_ALB)) {
1792 /* bond_alb_init_slave() must be called before all other stages since
1793 * it might fail and we do not want to have to undo everything
1795 res = bond_alb_init_slave(bond, new_slave);
1797 goto err_unset_master;
1801 /* If the mode USES_PRIMARY, then the new slave gets the
1802 * master's promisc (and mc) settings only if it becomes the
1803 * curr_active_slave, and that is taken care of later when calling
1804 * bond_change_active()
1806 if (!USES_PRIMARY(bond->params.mode)) {
1807 /* set promiscuity level to new slave */
1808 if (bond_dev->flags & IFF_PROMISC) {
1809 dev_set_promiscuity(slave_dev, 1);
1812 /* set allmulti level to new slave */
1813 if (bond_dev->flags & IFF_ALLMULTI) {
1814 dev_set_allmulti(slave_dev, 1);
1817 /* upload master's mc_list to new slave */
1818 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1819 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1823 if (bond->params.mode == BOND_MODE_8023AD) {
1824 /* add lacpdu mc addr to mc list */
1825 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1827 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1830 bond_add_vlans_on_slave(bond, slave_dev);
1832 write_lock_bh(&bond->lock);
1834 bond_attach_slave(bond, new_slave);
1836 new_slave->delay = 0;
1837 new_slave->link_failure_count = 0;
1839 bond_compute_features(bond);
1841 if (bond->params.miimon && !bond->params.use_carrier) {
1842 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1844 if ((link_reporting == -1) && !bond->params.arp_interval) {
1846 * miimon is set but a bonded network driver
1847 * does not support ETHTOOL/MII and
1848 * arp_interval is not set. Note: if
1849 * use_carrier is enabled, we will never go
1850 * here (because netif_carrier is always
1851 * supported); thus, we don't need to change
1852 * the messages for netif_carrier.
1854 printk(KERN_WARNING DRV_NAME
1855 ": %s: Warning: MII and ETHTOOL support not "
1856 "available for interface %s, and "
1857 "arp_interval/arp_ip_target module parameters "
1858 "not specified, thus bonding will not detect "
1859 "link failures! see bonding.txt for details.\n",
1860 bond_dev->name, slave_dev->name);
1861 } else if (link_reporting == -1) {
1862 /* unable get link status using mii/ethtool */
1863 printk(KERN_WARNING DRV_NAME
1864 ": %s: Warning: can't get link status from "
1865 "interface %s; the network driver associated "
1866 "with this interface does not support MII or "
1867 "ETHTOOL link status reporting, thus miimon "
1868 "has no effect on this interface.\n",
1869 bond_dev->name, slave_dev->name);
1873 /* check for initial state */
1874 if (!bond->params.miimon ||
1875 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1876 if (bond->params.updelay) {
1877 dprintk("Initial state of slave_dev is "
1878 "BOND_LINK_BACK\n");
1879 new_slave->link = BOND_LINK_BACK;
1880 new_slave->delay = bond->params.updelay;
1882 dprintk("Initial state of slave_dev is "
1884 new_slave->link = BOND_LINK_UP;
1886 new_slave->jiffies = jiffies;
1888 dprintk("Initial state of slave_dev is "
1889 "BOND_LINK_DOWN\n");
1890 new_slave->link = BOND_LINK_DOWN;
1893 if (bond_update_speed_duplex(new_slave) &&
1894 (new_slave->link != BOND_LINK_DOWN)) {
1895 printk(KERN_WARNING DRV_NAME
1896 ": %s: Warning: failed to get speed and duplex from %s, "
1897 "assumed to be 100Mb/sec and Full.\n",
1898 bond_dev->name, new_slave->dev->name);
1900 if (bond->params.mode == BOND_MODE_8023AD) {
1901 printk(KERN_WARNING DRV_NAME
1902 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1903 "support in base driver for proper aggregator "
1904 "selection.\n", bond_dev->name);
1908 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1909 /* if there is a primary slave, remember it */
1910 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1911 bond->primary_slave = new_slave;
1915 switch (bond->params.mode) {
1916 case BOND_MODE_ACTIVEBACKUP:
1917 /* if we're in active-backup mode, we need one and only one active
1918 * interface. The backup interfaces will have their NOARP flag set
1919 * because we need them to be completely deaf and not to respond to
1920 * any ARP request on the network to avoid fooling a switch. Thus,
1921 * since we guarantee that curr_active_slave always point to the last
1922 * usable interface, we just have to verify this interface's flag.
1924 if (((!bond->curr_active_slave) ||
1925 (bond->curr_active_slave->dev->flags & IFF_NOARP)) &&
1926 (new_slave->link != BOND_LINK_DOWN)) {
1927 dprintk("This is the first active slave\n");
1928 /* first slave or no active slave yet, and this link
1929 is OK, so make this interface the active one */
1930 bond_change_active_slave(bond, new_slave);
1932 dprintk("This is just a backup slave\n");
1933 bond_set_slave_inactive_flags(new_slave);
1936 case BOND_MODE_8023AD:
1937 /* in 802.3ad mode, the internal mechanism
1938 * will activate the slaves in the selected
1941 bond_set_slave_inactive_flags(new_slave);
1942 /* if this is the first slave */
1943 if (bond->slave_cnt == 1) {
1944 SLAVE_AD_INFO(new_slave).id = 1;
1945 /* Initialize AD with the number of times that the AD timer is called in 1 second
1946 * can be called only after the mac address of the bond is set
1948 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1949 bond->params.lacp_fast);
1951 SLAVE_AD_INFO(new_slave).id =
1952 SLAVE_AD_INFO(new_slave->prev).id + 1;
1955 bond_3ad_bind_slave(new_slave);
1959 new_slave->state = BOND_STATE_ACTIVE;
1960 if ((!bond->curr_active_slave) &&
1961 (new_slave->link != BOND_LINK_DOWN)) {
1962 /* first slave or no active slave yet, and this link
1963 * is OK, so make this interface the active one
1965 bond_change_active_slave(bond, new_slave);
1969 dprintk("This slave is always active in trunk mode\n");
1971 /* always active in trunk mode */
1972 new_slave->state = BOND_STATE_ACTIVE;
1974 /* In trunking mode there is little meaning to curr_active_slave
1975 * anyway (it holds no special properties of the bond device),
1976 * so we can change it without calling change_active_interface()
1978 if (!bond->curr_active_slave) {
1979 bond->curr_active_slave = new_slave;
1982 } /* switch(bond_mode) */
1984 write_unlock_bh(&bond->lock);
1986 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1988 goto err_unset_master;
1990 printk(KERN_INFO DRV_NAME
1991 ": %s: enslaving %s as a%s interface with a%s link.\n",
1992 bond_dev->name, slave_dev->name,
1993 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1994 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1996 /* enslave is successful */
1999 /* Undo stages on error */
2001 netdev_set_master(slave_dev, NULL);
2004 dev_close(slave_dev);
2007 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
2008 addr.sa_family = slave_dev->type;
2009 dev_set_mac_address(slave_dev, &addr);
2015 bond_dev->features = old_features;
2021 * Try to release the slave device <slave> from the bond device <master>
2022 * It is legal to access curr_active_slave without a lock because all the function
2025 * The rules for slave state should be:
2026 * for Active/Backup:
2027 * Active stays on all backups go down
2028 * for Bonded connections:
2029 * The first up interface should be left on and all others downed.
2031 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2033 struct bonding *bond = bond_dev->priv;
2034 struct slave *slave, *oldcurrent;
2035 struct sockaddr addr;
2036 int mac_addr_differ;
2038 /* slave is not a slave or master is not master of this slave */
2039 if (!(slave_dev->flags & IFF_SLAVE) ||
2040 (slave_dev->master != bond_dev)) {
2041 printk(KERN_ERR DRV_NAME
2042 ": %s: Error: cannot release %s.\n",
2043 bond_dev->name, slave_dev->name);
2047 write_lock_bh(&bond->lock);
2049 slave = bond_get_slave_by_dev(bond, slave_dev);
2051 /* not a slave of this bond */
2052 printk(KERN_INFO DRV_NAME
2053 ": %s: %s not enslaved\n",
2054 bond_dev->name, slave_dev->name);
2058 mac_addr_differ = memcmp(bond_dev->dev_addr,
2061 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
2062 printk(KERN_WARNING DRV_NAME
2063 ": %s: Warning: the permanent HWaddr of %s "
2064 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
2065 "still in use by %s. Set the HWaddr of "
2066 "%s to a different address to avoid "
2070 slave->perm_hwaddr[0],
2071 slave->perm_hwaddr[1],
2072 slave->perm_hwaddr[2],
2073 slave->perm_hwaddr[3],
2074 slave->perm_hwaddr[4],
2075 slave->perm_hwaddr[5],
2080 /* Inform AD package of unbinding of slave. */
2081 if (bond->params.mode == BOND_MODE_8023AD) {
2082 /* must be called before the slave is
2083 * detached from the list
2085 bond_3ad_unbind_slave(slave);
2088 printk(KERN_INFO DRV_NAME
2089 ": %s: releasing %s interface %s\n",
2091 (slave->state == BOND_STATE_ACTIVE)
2092 ? "active" : "backup",
2095 oldcurrent = bond->curr_active_slave;
2097 bond->current_arp_slave = NULL;
2099 /* release the slave from its bond */
2100 bond_detach_slave(bond, slave);
2102 bond_compute_features(bond);
2104 if (bond->primary_slave == slave) {
2105 bond->primary_slave = NULL;
2108 if (oldcurrent == slave) {
2109 bond_change_active_slave(bond, NULL);
2112 if ((bond->params.mode == BOND_MODE_TLB) ||
2113 (bond->params.mode == BOND_MODE_ALB)) {
2114 /* Must be called only after the slave has been
2115 * detached from the list and the curr_active_slave
2116 * has been cleared (if our_slave == old_current),
2117 * but before a new active slave is selected.
2119 bond_alb_deinit_slave(bond, slave);
2122 if (oldcurrent == slave) {
2123 bond_select_active_slave(bond);
2125 if (!bond->curr_active_slave) {
2126 printk(KERN_INFO DRV_NAME
2127 ": %s: now running without any active "
2133 if (bond->slave_cnt == 0) {
2134 /* if the last slave was removed, zero the mac address
2135 * of the master so it will be set by the application
2136 * to the mac address of the first slave
2138 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2140 if (list_empty(&bond->vlan_list)) {
2141 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2143 printk(KERN_WARNING DRV_NAME
2144 ": %s: Warning: clearing HW address of %s while it "
2145 "still has VLANs.\n",
2146 bond_dev->name, bond_dev->name);
2147 printk(KERN_WARNING DRV_NAME
2148 ": %s: When re-adding slaves, make sure the bond's "
2149 "HW address matches its VLANs'.\n",
2152 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2153 !bond_has_challenged_slaves(bond)) {
2154 printk(KERN_INFO DRV_NAME
2155 ": %s: last VLAN challenged slave %s "
2156 "left bond %s. VLAN blocking is removed\n",
2157 bond_dev->name, slave_dev->name, bond_dev->name);
2158 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2161 write_unlock_bh(&bond->lock);
2163 /* must do this from outside any spinlocks */
2164 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2166 bond_del_vlans_from_slave(bond, slave_dev);
2168 /* If the mode USES_PRIMARY, then we should only remove its
2169 * promisc and mc settings if it was the curr_active_slave, but that was
2170 * already taken care of above when we detached the slave
2172 if (!USES_PRIMARY(bond->params.mode)) {
2173 /* unset promiscuity level from slave */
2174 if (bond_dev->flags & IFF_PROMISC) {
2175 dev_set_promiscuity(slave_dev, -1);
2178 /* unset allmulti level from slave */
2179 if (bond_dev->flags & IFF_ALLMULTI) {
2180 dev_set_allmulti(slave_dev, -1);
2183 /* flush master's mc_list from slave */
2184 bond_mc_list_flush(bond_dev, slave_dev);
2187 netdev_set_master(slave_dev, NULL);
2189 /* close slave before restoring its mac address */
2190 dev_close(slave_dev);
2192 /* restore original ("permanent") mac address */
2193 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2194 addr.sa_family = slave_dev->type;
2195 dev_set_mac_address(slave_dev, &addr);
2197 /* restore the original state of the
2198 * IFF_NOARP flag that might have been
2199 * set by bond_set_slave_inactive_flags()
2201 if ((slave->original_flags & IFF_NOARP) == 0) {
2202 slave_dev->flags &= ~IFF_NOARP;
2207 return 0; /* deletion OK */
2211 * This function releases all slaves.
2213 static int bond_release_all(struct net_device *bond_dev)
2215 struct bonding *bond = bond_dev->priv;
2216 struct slave *slave;
2217 struct net_device *slave_dev;
2218 struct sockaddr addr;
2220 write_lock_bh(&bond->lock);
2222 if (bond->slave_cnt == 0) {
2226 bond->current_arp_slave = NULL;
2227 bond->primary_slave = NULL;
2228 bond_change_active_slave(bond, NULL);
2230 while ((slave = bond->first_slave) != NULL) {
2231 /* Inform AD package of unbinding of slave
2232 * before slave is detached from the list.
2234 if (bond->params.mode == BOND_MODE_8023AD) {
2235 bond_3ad_unbind_slave(slave);
2238 slave_dev = slave->dev;
2239 bond_detach_slave(bond, slave);
2241 if ((bond->params.mode == BOND_MODE_TLB) ||
2242 (bond->params.mode == BOND_MODE_ALB)) {
2243 /* must be called only after the slave
2244 * has been detached from the list
2246 bond_alb_deinit_slave(bond, slave);
2249 bond_compute_features(bond);
2251 /* now that the slave is detached, unlock and perform
2252 * all the undo steps that should not be called from
2255 write_unlock_bh(&bond->lock);
2257 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2258 bond_del_vlans_from_slave(bond, slave_dev);
2260 /* If the mode USES_PRIMARY, then we should only remove its
2261 * promisc and mc settings if it was the curr_active_slave, but that was
2262 * already taken care of above when we detached the slave
2264 if (!USES_PRIMARY(bond->params.mode)) {
2265 /* unset promiscuity level from slave */
2266 if (bond_dev->flags & IFF_PROMISC) {
2267 dev_set_promiscuity(slave_dev, -1);
2270 /* unset allmulti level from slave */
2271 if (bond_dev->flags & IFF_ALLMULTI) {
2272 dev_set_allmulti(slave_dev, -1);
2275 /* flush master's mc_list from slave */
2276 bond_mc_list_flush(bond_dev, slave_dev);
2279 netdev_set_master(slave_dev, NULL);
2281 /* close slave before restoring its mac address */
2282 dev_close(slave_dev);
2284 /* restore original ("permanent") mac address*/
2285 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2286 addr.sa_family = slave_dev->type;
2287 dev_set_mac_address(slave_dev, &addr);
2289 /* restore the original state of the IFF_NOARP flag that might have
2290 * been set by bond_set_slave_inactive_flags()
2292 if ((slave->original_flags & IFF_NOARP) == 0) {
2293 slave_dev->flags &= ~IFF_NOARP;
2298 /* re-acquire the lock before getting the next slave */
2299 write_lock_bh(&bond->lock);
2302 /* zero the mac address of the master so it will be
2303 * set by the application to the mac address of the
2306 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2308 if (list_empty(&bond->vlan_list)) {
2309 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2311 printk(KERN_WARNING DRV_NAME
2312 ": %s: Warning: clearing HW address of %s while it "
2313 "still has VLANs.\n",
2314 bond_dev->name, bond_dev->name);
2315 printk(KERN_WARNING DRV_NAME
2316 ": %s: When re-adding slaves, make sure the bond's "
2317 "HW address matches its VLANs'.\n",
2321 printk(KERN_INFO DRV_NAME
2322 ": %s: released all slaves\n",
2326 write_unlock_bh(&bond->lock);
2332 * This function changes the active slave to slave <slave_dev>.
2333 * It returns -EINVAL in the following cases.
2334 * - <slave_dev> is not found in the list.
2335 * - There is not active slave now.
2336 * - <slave_dev> is already active.
2337 * - The link state of <slave_dev> is not BOND_LINK_UP.
2338 * - <slave_dev> is not running.
2339 * In these cases, this fuction does nothing.
2340 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2342 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2344 struct bonding *bond = bond_dev->priv;
2345 struct slave *old_active = NULL;
2346 struct slave *new_active = NULL;
2349 if (!USES_PRIMARY(bond->params.mode)) {
2353 /* Verify that master_dev is indeed the master of slave_dev */
2354 if (!(slave_dev->flags & IFF_SLAVE) ||
2355 (slave_dev->master != bond_dev)) {
2359 write_lock_bh(&bond->lock);
2361 old_active = bond->curr_active_slave;
2362 new_active = bond_get_slave_by_dev(bond, slave_dev);
2365 * Changing to the current active: do nothing; return success.
2367 if (new_active && (new_active == old_active)) {
2368 write_unlock_bh(&bond->lock);
2374 (new_active->link == BOND_LINK_UP) &&
2375 IS_UP(new_active->dev)) {
2376 bond_change_active_slave(bond, new_active);
2381 write_unlock_bh(&bond->lock);
2386 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2388 struct bonding *bond = bond_dev->priv;
2390 info->bond_mode = bond->params.mode;
2391 info->miimon = bond->params.miimon;
2393 read_lock_bh(&bond->lock);
2394 info->num_slaves = bond->slave_cnt;
2395 read_unlock_bh(&bond->lock);
2400 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2402 struct bonding *bond = bond_dev->priv;
2403 struct slave *slave;
2406 if (info->slave_id < 0) {
2410 read_lock_bh(&bond->lock);
2412 bond_for_each_slave(bond, slave, i) {
2413 if (i == (int)info->slave_id) {
2419 read_unlock_bh(&bond->lock);
2422 strcpy(info->slave_name, slave->dev->name);
2423 info->link = slave->link;
2424 info->state = slave->state;
2425 info->link_failure_count = slave->link_failure_count;
2433 /*-------------------------------- Monitoring -------------------------------*/
2435 /* this function is called regularly to monitor each slave's link. */
2436 void bond_mii_monitor(struct net_device *bond_dev)
2438 struct bonding *bond = bond_dev->priv;
2439 struct slave *slave, *oldcurrent;
2440 int do_failover = 0;
2444 read_lock(&bond->lock);
2446 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2448 if (bond->kill_timers) {
2452 if (bond->slave_cnt == 0) {
2456 /* we will try to read the link status of each of our slaves, and
2457 * set their IFF_RUNNING flag appropriately. For each slave not
2458 * supporting MII status, we won't do anything so that a user-space
2459 * program could monitor the link itself if needed.
2462 read_lock(&bond->curr_slave_lock);
2463 oldcurrent = bond->curr_active_slave;
2464 read_unlock(&bond->curr_slave_lock);
2466 bond_for_each_slave(bond, slave, i) {
2467 struct net_device *slave_dev = slave->dev;
2469 u16 old_speed = slave->speed;
2470 u8 old_duplex = slave->duplex;
2472 link_state = bond_check_dev_link(bond, slave_dev, 0);
2474 switch (slave->link) {
2475 case BOND_LINK_UP: /* the link was up */
2476 if (link_state == BMSR_LSTATUS) {
2477 /* link stays up, nothing more to do */
2479 } else { /* link going down */
2480 slave->link = BOND_LINK_FAIL;
2481 slave->delay = bond->params.downdelay;
2483 if (slave->link_failure_count < UINT_MAX) {
2484 slave->link_failure_count++;
2487 if (bond->params.downdelay) {
2488 printk(KERN_INFO DRV_NAME
2489 ": %s: link status down for %s "
2490 "interface %s, disabling it in "
2494 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2495 ? ((slave == oldcurrent)
2496 ? "active " : "backup ")
2500 bond->params.downdelay * bond->params.miimon);
2503 /* no break ! fall through the BOND_LINK_FAIL test to
2504 ensure proper action to be taken
2506 case BOND_LINK_FAIL: /* the link has just gone down */
2507 if (link_state != BMSR_LSTATUS) {
2508 /* link stays down */
2509 if (slave->delay <= 0) {
2510 /* link down for too long time */
2511 slave->link = BOND_LINK_DOWN;
2513 /* in active/backup mode, we must
2514 * completely disable this interface
2516 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2517 (bond->params.mode == BOND_MODE_8023AD)) {
2518 bond_set_slave_inactive_flags(slave);
2521 printk(KERN_INFO DRV_NAME
2522 ": %s: link status definitely "
2523 "down for interface %s, "
2528 /* notify ad that the link status has changed */
2529 if (bond->params.mode == BOND_MODE_8023AD) {
2530 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2533 if ((bond->params.mode == BOND_MODE_TLB) ||
2534 (bond->params.mode == BOND_MODE_ALB)) {
2535 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2538 if (slave == oldcurrent) {
2546 slave->link = BOND_LINK_UP;
2547 slave->jiffies = jiffies;
2548 printk(KERN_INFO DRV_NAME
2549 ": %s: link status up again after %d "
2550 "ms for interface %s.\n",
2552 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2556 case BOND_LINK_DOWN: /* the link was down */
2557 if (link_state != BMSR_LSTATUS) {
2558 /* the link stays down, nothing more to do */
2560 } else { /* link going up */
2561 slave->link = BOND_LINK_BACK;
2562 slave->delay = bond->params.updelay;
2564 if (bond->params.updelay) {
2565 /* if updelay == 0, no need to
2566 advertise about a 0 ms delay */
2567 printk(KERN_INFO DRV_NAME
2568 ": %s: link status up for "
2569 "interface %s, enabling it "
2573 bond->params.updelay * bond->params.miimon);
2576 /* no break ! fall through the BOND_LINK_BACK state in
2577 case there's something to do.
2579 case BOND_LINK_BACK: /* the link has just come back */
2580 if (link_state != BMSR_LSTATUS) {
2581 /* link down again */
2582 slave->link = BOND_LINK_DOWN;
2584 printk(KERN_INFO DRV_NAME
2585 ": %s: link status down again after %d "
2586 "ms for interface %s.\n",
2588 (bond->params.updelay - slave->delay) * bond->params.miimon,
2592 if (slave->delay == 0) {
2593 /* now the link has been up for long time enough */
2594 slave->link = BOND_LINK_UP;
2595 slave->jiffies = jiffies;
2597 if (bond->params.mode == BOND_MODE_8023AD) {
2598 /* prevent it from being the active one */
2599 slave->state = BOND_STATE_BACKUP;
2600 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2601 /* make it immediately active */
2602 slave->state = BOND_STATE_ACTIVE;
2603 } else if (slave != bond->primary_slave) {
2604 /* prevent it from being the active one */
2605 slave->state = BOND_STATE_BACKUP;
2608 printk(KERN_INFO DRV_NAME
2609 ": %s: link status definitely "
2610 "up for interface %s.\n",
2614 /* notify ad that the link status has changed */
2615 if (bond->params.mode == BOND_MODE_8023AD) {
2616 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2619 if ((bond->params.mode == BOND_MODE_TLB) ||
2620 (bond->params.mode == BOND_MODE_ALB)) {
2621 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2624 if ((!oldcurrent) ||
2625 (slave == bond->primary_slave)) {
2634 /* Should not happen */
2635 printk(KERN_ERR DRV_NAME
2636 ": %s: Error: %s Illegal value (link=%d)\n",
2641 } /* end of switch (slave->link) */
2643 bond_update_speed_duplex(slave);
2645 if (bond->params.mode == BOND_MODE_8023AD) {
2646 if (old_speed != slave->speed) {
2647 bond_3ad_adapter_speed_changed(slave);
2650 if (old_duplex != slave->duplex) {
2651 bond_3ad_adapter_duplex_changed(slave);
2658 write_lock(&bond->curr_slave_lock);
2660 bond_select_active_slave(bond);
2662 if (oldcurrent && !bond->curr_active_slave) {
2663 printk(KERN_INFO DRV_NAME
2664 ": %s: now running without any active "
2669 write_unlock(&bond->curr_slave_lock);
2673 if (bond->params.miimon) {
2674 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2677 read_unlock(&bond->lock);
2681 static u32 bond_glean_dev_ip(struct net_device *dev)
2683 struct in_device *idev;
2684 struct in_ifaddr *ifa;
2691 idev = __in_dev_get_rcu(dev);
2695 ifa = idev->ifa_list;
2699 addr = ifa->ifa_local;
2705 static int bond_has_ip(struct bonding *bond)
2707 struct vlan_entry *vlan, *vlan_next;
2709 if (bond->master_ip)
2712 if (list_empty(&bond->vlan_list))
2715 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2725 * We go to the (large) trouble of VLAN tagging ARP frames because
2726 * switches in VLAN mode (especially if ports are configured as
2727 * "native" to a VLAN) might not pass non-tagged frames.
2729 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2731 struct sk_buff *skb;
2733 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2734 slave_dev->name, dest_ip, src_ip, vlan_id);
2736 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2737 NULL, slave_dev->dev_addr, NULL);
2740 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2744 skb = vlan_put_tag(skb, vlan_id);
2746 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2754 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2757 u32 *targets = bond->params.arp_targets;
2758 struct vlan_entry *vlan, *vlan_next;
2759 struct net_device *vlan_dev;
2763 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2766 dprintk("basa: target %x\n", targets[i]);
2767 if (list_empty(&bond->vlan_list)) {
2768 dprintk("basa: empty vlan: arp_send\n");
2769 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2770 bond->master_ip, 0);
2775 * If VLANs are configured, we do a route lookup to
2776 * determine which VLAN interface would be used, so we
2777 * can tag the ARP with the proper VLAN tag.
2779 memset(&fl, 0, sizeof(fl));
2780 fl.fl4_dst = targets[i];
2781 fl.fl4_tos = RTO_ONLINK;
2783 rv = ip_route_output_key(&rt, &fl);
2785 if (net_ratelimit()) {
2786 printk(KERN_WARNING DRV_NAME
2787 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2788 bond->dev->name, NIPQUAD(fl.fl4_dst));
2794 * This target is not on a VLAN
2796 if (rt->u.dst.dev == bond->dev) {
2798 dprintk("basa: rtdev == bond->dev: arp_send\n");
2799 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2800 bond->master_ip, 0);
2805 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2807 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2808 if (vlan_dev == rt->u.dst.dev) {
2809 vlan_id = vlan->vlan_id;
2810 dprintk("basa: vlan match on %s %d\n",
2811 vlan_dev->name, vlan_id);
2818 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2819 vlan->vlan_ip, vlan_id);
2823 if (net_ratelimit()) {
2824 printk(KERN_WARNING DRV_NAME
2825 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2826 bond->dev->name, NIPQUAD(fl.fl4_dst),
2827 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2834 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2835 * for each VLAN above us.
2837 static void bond_send_gratuitous_arp(struct bonding *bond)
2839 struct slave *slave = bond->curr_active_slave;
2840 struct vlan_entry *vlan;
2841 struct net_device *vlan_dev;
2843 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2844 slave ? slave->dev->name : "NULL");
2848 if (bond->master_ip) {
2849 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2850 bond->master_ip, 0);
2853 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2854 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2855 if (vlan->vlan_ip) {
2856 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2857 vlan->vlan_ip, vlan->vlan_id);
2863 * this function is called regularly to monitor each slave's link
2864 * ensuring that traffic is being sent and received when arp monitoring
2865 * is used in load-balancing mode. if the adapter has been dormant, then an
2866 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2867 * arp monitoring in active backup mode.
2869 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2871 struct bonding *bond = bond_dev->priv;
2872 struct slave *slave, *oldcurrent;
2873 int do_failover = 0;
2877 read_lock(&bond->lock);
2879 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2881 if (bond->kill_timers) {
2885 if (bond->slave_cnt == 0) {
2889 read_lock(&bond->curr_slave_lock);
2890 oldcurrent = bond->curr_active_slave;
2891 read_unlock(&bond->curr_slave_lock);
2893 /* see if any of the previous devices are up now (i.e. they have
2894 * xmt and rcv traffic). the curr_active_slave does not come into
2895 * the picture unless it is null. also, slave->jiffies is not needed
2896 * here because we send an arp on each slave and give a slave as
2897 * long as it needs to get the tx/rx within the delta.
2898 * TODO: what about up/down delay in arp mode? it wasn't here before
2901 bond_for_each_slave(bond, slave, i) {
2902 if (slave->link != BOND_LINK_UP) {
2903 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2904 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2906 slave->link = BOND_LINK_UP;
2907 slave->state = BOND_STATE_ACTIVE;
2909 /* primary_slave has no meaning in round-robin
2910 * mode. the window of a slave being up and
2911 * curr_active_slave being null after enslaving
2915 printk(KERN_INFO DRV_NAME
2916 ": %s: link status definitely "
2917 "up for interface %s, ",
2922 printk(KERN_INFO DRV_NAME
2923 ": %s: interface %s is now up\n",
2929 /* slave->link == BOND_LINK_UP */
2931 /* not all switches will respond to an arp request
2932 * when the source ip is 0, so don't take the link down
2933 * if we don't know our ip yet
2935 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2936 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2937 bond_has_ip(bond))) {
2939 slave->link = BOND_LINK_DOWN;
2940 slave->state = BOND_STATE_BACKUP;
2942 if (slave->link_failure_count < UINT_MAX) {
2943 slave->link_failure_count++;
2946 printk(KERN_INFO DRV_NAME
2947 ": %s: interface %s is now down.\n",
2951 if (slave == oldcurrent) {
2957 /* note: if switch is in round-robin mode, all links
2958 * must tx arp to ensure all links rx an arp - otherwise
2959 * links may oscillate or not come up at all; if switch is
2960 * in something like xor mode, there is nothing we can
2961 * do - all replies will be rx'ed on same link causing slaves
2962 * to be unstable during low/no traffic periods
2964 if (IS_UP(slave->dev)) {
2965 bond_arp_send_all(bond, slave);
2970 write_lock(&bond->curr_slave_lock);
2972 bond_select_active_slave(bond);
2974 if (oldcurrent && !bond->curr_active_slave) {
2975 printk(KERN_INFO DRV_NAME
2976 ": %s: now running without any active "
2981 write_unlock(&bond->curr_slave_lock);
2985 if (bond->params.arp_interval) {
2986 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2989 read_unlock(&bond->lock);
2993 * When using arp monitoring in active-backup mode, this function is
2994 * called to determine if any backup slaves have went down or a new
2995 * current slave needs to be found.
2996 * The backup slaves never generate traffic, they are considered up by merely
2997 * receiving traffic. If the current slave goes down, each backup slave will
2998 * be given the opportunity to tx/rx an arp before being taken down - this
2999 * prevents all slaves from being taken down due to the current slave not
3000 * sending any traffic for the backups to receive. The arps are not necessarily
3001 * necessary, any tx and rx traffic will keep the current slave up. While any
3002 * rx traffic will keep the backup slaves up, the current slave is responsible
3003 * for generating traffic to keep them up regardless of any other traffic they
3004 * may have received.
3005 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
3007 void bond_activebackup_arp_mon(struct net_device *bond_dev)
3009 struct bonding *bond = bond_dev->priv;
3010 struct slave *slave;
3014 read_lock(&bond->lock);
3016 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
3018 if (bond->kill_timers) {
3022 if (bond->slave_cnt == 0) {
3026 /* determine if any slave has come up or any backup slave has
3028 * TODO: what about up/down delay in arp mode? it wasn't here before
3031 bond_for_each_slave(bond, slave, i) {
3032 if (slave->link != BOND_LINK_UP) {
3033 if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
3035 slave->link = BOND_LINK_UP;
3037 write_lock(&bond->curr_slave_lock);
3039 if ((!bond->curr_active_slave) &&
3040 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
3041 bond_change_active_slave(bond, slave);
3042 bond->current_arp_slave = NULL;
3043 } else if (bond->curr_active_slave != slave) {
3044 /* this slave has just come up but we
3045 * already have a current slave; this
3046 * can also happen if bond_enslave adds
3047 * a new slave that is up while we are
3048 * searching for a new slave
3050 bond_set_slave_inactive_flags(slave);
3051 bond->current_arp_slave = NULL;
3054 if (slave == bond->curr_active_slave) {
3055 printk(KERN_INFO DRV_NAME
3056 ": %s: %s is up and now the "
3057 "active interface\n",
3061 printk(KERN_INFO DRV_NAME
3062 ": %s: backup interface %s is "
3068 write_unlock(&bond->curr_slave_lock);
3071 read_lock(&bond->curr_slave_lock);
3073 if ((slave != bond->curr_active_slave) &&
3074 (!bond->current_arp_slave) &&
3075 (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
3076 bond_has_ip(bond))) {
3077 /* a backup slave has gone down; three times
3078 * the delta allows the current slave to be
3079 * taken out before the backup slave.
3080 * note: a non-null current_arp_slave indicates
3081 * the curr_active_slave went down and we are
3082 * searching for a new one; under this
3083 * condition we only take the curr_active_slave
3084 * down - this gives each slave a chance to
3085 * tx/rx traffic before being taken out
3088 read_unlock(&bond->curr_slave_lock);
3090 slave->link = BOND_LINK_DOWN;
3092 if (slave->link_failure_count < UINT_MAX) {
3093 slave->link_failure_count++;
3096 bond_set_slave_inactive_flags(slave);
3098 printk(KERN_INFO DRV_NAME
3099 ": %s: backup interface %s is now down\n",
3103 read_unlock(&bond->curr_slave_lock);
3108 read_lock(&bond->curr_slave_lock);
3109 slave = bond->curr_active_slave;
3110 read_unlock(&bond->curr_slave_lock);
3113 /* if we have sent traffic in the past 2*arp_intervals but
3114 * haven't xmit and rx traffic in that time interval, select
3115 * a different slave. slave->jiffies is only updated when
3116 * a slave first becomes the curr_active_slave - not necessarily
3117 * after every arp; this ensures the slave has a full 2*delta
3118 * before being taken out. if a primary is being used, check
3119 * if it is up and needs to take over as the curr_active_slave
3121 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
3122 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
3123 bond_has_ip(bond))) &&
3124 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
3126 slave->link = BOND_LINK_DOWN;
3128 if (slave->link_failure_count < UINT_MAX) {
3129 slave->link_failure_count++;
3132 printk(KERN_INFO DRV_NAME
3133 ": %s: link status down for active interface "
3134 "%s, disabling it\n",
3138 write_lock(&bond->curr_slave_lock);
3140 bond_select_active_slave(bond);
3141 slave = bond->curr_active_slave;
3143 write_unlock(&bond->curr_slave_lock);
3145 bond->current_arp_slave = slave;
3148 slave->jiffies = jiffies;
3150 } else if ((bond->primary_slave) &&
3151 (bond->primary_slave != slave) &&
3152 (bond->primary_slave->link == BOND_LINK_UP)) {
3153 /* at this point, slave is the curr_active_slave */
3154 printk(KERN_INFO DRV_NAME
3155 ": %s: changing from interface %s to primary "
3159 bond->primary_slave->dev->name);
3161 /* primary is up so switch to it */
3162 write_lock(&bond->curr_slave_lock);
3163 bond_change_active_slave(bond, bond->primary_slave);
3164 write_unlock(&bond->curr_slave_lock);
3166 slave = bond->primary_slave;
3167 slave->jiffies = jiffies;
3169 bond->current_arp_slave = NULL;
3172 /* the current slave must tx an arp to ensure backup slaves
3175 if (slave && bond_has_ip(bond)) {
3176 bond_arp_send_all(bond, slave);
3180 /* if we don't have a curr_active_slave, search for the next available
3181 * backup slave from the current_arp_slave and make it the candidate
3182 * for becoming the curr_active_slave
3185 if (!bond->current_arp_slave) {
3186 bond->current_arp_slave = bond->first_slave;
3189 if (bond->current_arp_slave) {
3190 bond_set_slave_inactive_flags(bond->current_arp_slave);
3192 /* search for next candidate */
3193 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3194 if (IS_UP(slave->dev)) {
3195 slave->link = BOND_LINK_BACK;
3196 bond_set_slave_active_flags(slave);
3197 bond_arp_send_all(bond, slave);
3198 slave->jiffies = jiffies;
3199 bond->current_arp_slave = slave;
3203 /* if the link state is up at this point, we
3204 * mark it down - this can happen if we have
3205 * simultaneous link failures and
3206 * reselect_active_interface doesn't make this
3207 * one the current slave so it is still marked
3208 * up when it is actually down
3210 if (slave->link == BOND_LINK_UP) {
3211 slave->link = BOND_LINK_DOWN;
3212 if (slave->link_failure_count < UINT_MAX) {
3213 slave->link_failure_count++;
3216 bond_set_slave_inactive_flags(slave);
3218 printk(KERN_INFO DRV_NAME
3219 ": %s: backup interface %s is "
3229 if (bond->params.arp_interval) {
3230 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
3233 read_unlock(&bond->lock);
3236 /*------------------------------ proc/seq_file-------------------------------*/
3238 #ifdef CONFIG_PROC_FS
3240 #define SEQ_START_TOKEN ((void *)1)
3242 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3244 struct bonding *bond = seq->private;
3246 struct slave *slave;
3249 /* make sure the bond won't be taken away */
3250 read_lock(&dev_base_lock);
3251 read_lock_bh(&bond->lock);
3254 return SEQ_START_TOKEN;
3257 bond_for_each_slave(bond, slave, i) {
3258 if (++off == *pos) {
3266 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3268 struct bonding *bond = seq->private;
3269 struct slave *slave = v;
3272 if (v == SEQ_START_TOKEN) {
3273 return bond->first_slave;
3276 slave = slave->next;
3278 return (slave == bond->first_slave) ? NULL : slave;
3281 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3283 struct bonding *bond = seq->private;
3285 read_unlock_bh(&bond->lock);
3286 read_unlock(&dev_base_lock);
3289 static void bond_info_show_master(struct seq_file *seq)
3291 struct bonding *bond = seq->private;
3296 read_lock(&bond->curr_slave_lock);
3297 curr = bond->curr_active_slave;
3298 read_unlock(&bond->curr_slave_lock);
3300 seq_printf(seq, "Bonding Mode: %s\n",
3301 bond_mode_name(bond->params.mode));
3303 if (bond->params.mode == BOND_MODE_XOR ||
3304 bond->params.mode == BOND_MODE_8023AD) {
3305 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3306 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3307 bond->params.xmit_policy);
3310 if (USES_PRIMARY(bond->params.mode)) {
3311 seq_printf(seq, "Primary Slave: %s\n",
3312 (bond->primary_slave) ?
3313 bond->primary_slave->dev->name : "None");
3315 seq_printf(seq, "Currently Active Slave: %s\n",
3316 (curr) ? curr->dev->name : "None");
3319 seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down");
3320 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3321 seq_printf(seq, "Up Delay (ms): %d\n",
3322 bond->params.updelay * bond->params.miimon);
3323 seq_printf(seq, "Down Delay (ms): %d\n",
3324 bond->params.downdelay * bond->params.miimon);
3327 /* ARP information */
3328 if(bond->params.arp_interval > 0) {
3330 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3331 bond->params.arp_interval);
3333 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3335 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3336 if (!bond->params.arp_targets[i])
3339 seq_printf(seq, ",");
3340 target = ntohl(bond->params.arp_targets[i]);
3341 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3344 seq_printf(seq, "\n");
3347 if (bond->params.mode == BOND_MODE_8023AD) {
3348 struct ad_info ad_info;
3350 seq_puts(seq, "\n802.3ad info\n");
3351 seq_printf(seq, "LACP rate: %s\n",
3352 (bond->params.lacp_fast) ? "fast" : "slow");
3354 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3355 seq_printf(seq, "bond %s has no active aggregator\n",
3358 seq_printf(seq, "Active Aggregator Info:\n");
3360 seq_printf(seq, "\tAggregator ID: %d\n",
3361 ad_info.aggregator_id);
3362 seq_printf(seq, "\tNumber of ports: %d\n",
3364 seq_printf(seq, "\tActor Key: %d\n",
3366 seq_printf(seq, "\tPartner Key: %d\n",
3367 ad_info.partner_key);
3368 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3369 ad_info.partner_system[0],
3370 ad_info.partner_system[1],
3371 ad_info.partner_system[2],
3372 ad_info.partner_system[3],
3373 ad_info.partner_system[4],
3374 ad_info.partner_system[5]);
3379 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3381 struct bonding *bond = seq->private;
3383 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3384 seq_printf(seq, "MII Status: %s\n",
3385 (slave->link == BOND_LINK_UP) ? "up" : "down");
3386 seq_printf(seq, "Link Failure Count: %d\n",
3387 slave->link_failure_count);
3390 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3391 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3392 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3393 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
3395 if (bond->params.mode == BOND_MODE_8023AD) {
3396 const struct aggregator *agg
3397 = SLAVE_AD_INFO(slave).port.aggregator;
3400 seq_printf(seq, "Aggregator ID: %d\n",
3401 agg->aggregator_identifier);
3403 seq_puts(seq, "Aggregator ID: N/A\n");
3408 static int bond_info_seq_show(struct seq_file *seq, void *v)
3410 if (v == SEQ_START_TOKEN) {
3411 seq_printf(seq, "%s\n", version);
3412 bond_info_show_master(seq);
3414 bond_info_show_slave(seq, v);
3420 static struct seq_operations bond_info_seq_ops = {
3421 .start = bond_info_seq_start,
3422 .next = bond_info_seq_next,
3423 .stop = bond_info_seq_stop,
3424 .show = bond_info_seq_show,
3427 static int bond_info_open(struct inode *inode, struct file *file)
3429 struct seq_file *seq;
3430 struct proc_dir_entry *proc;
3433 res = seq_open(file, &bond_info_seq_ops);
3435 /* recover the pointer buried in proc_dir_entry data */
3436 seq = file->private_data;
3438 seq->private = proc->data;
3444 static struct file_operations bond_info_fops = {
3445 .owner = THIS_MODULE,
3446 .open = bond_info_open,
3448 .llseek = seq_lseek,
3449 .release = seq_release,
3452 static int bond_create_proc_entry(struct bonding *bond)
3454 struct net_device *bond_dev = bond->dev;
3456 if (bond_proc_dir) {
3457 bond->proc_entry = create_proc_entry(bond_dev->name,
3460 if (bond->proc_entry == NULL) {
3461 printk(KERN_WARNING DRV_NAME
3462 ": Warning: Cannot create /proc/net/%s/%s\n",
3463 DRV_NAME, bond_dev->name);
3465 bond->proc_entry->data = bond;
3466 bond->proc_entry->proc_fops = &bond_info_fops;
3467 bond->proc_entry->owner = THIS_MODULE;
3468 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3475 static void bond_remove_proc_entry(struct bonding *bond)
3477 if (bond_proc_dir && bond->proc_entry) {
3478 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3479 memset(bond->proc_file_name, 0, IFNAMSIZ);
3480 bond->proc_entry = NULL;
3484 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3485 * Caller must hold rtnl_lock.
3487 static void bond_create_proc_dir(void)
3489 int len = strlen(DRV_NAME);
3491 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3492 bond_proc_dir = bond_proc_dir->next) {
3493 if ((bond_proc_dir->namelen == len) &&
3494 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3499 if (!bond_proc_dir) {
3500 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3501 if (bond_proc_dir) {
3502 bond_proc_dir->owner = THIS_MODULE;
3504 printk(KERN_WARNING DRV_NAME
3505 ": Warning: cannot create /proc/net/%s\n",
3511 /* Destroy the bonding directory under /proc/net, if empty.
3512 * Caller must hold rtnl_lock.
3514 static void bond_destroy_proc_dir(void)
3516 struct proc_dir_entry *de;
3518 if (!bond_proc_dir) {
3522 /* verify that the /proc dir is empty */
3523 for (de = bond_proc_dir->subdir; de; de = de->next) {
3524 /* ignore . and .. */
3525 if (*(de->name) != '.') {
3531 if (bond_proc_dir->owner == THIS_MODULE) {
3532 bond_proc_dir->owner = NULL;
3535 remove_proc_entry(DRV_NAME, proc_net);
3536 bond_proc_dir = NULL;
3539 #endif /* CONFIG_PROC_FS */
3541 /*-------------------------- netdev event handling --------------------------*/
3544 * Change device name
3546 static int bond_event_changename(struct bonding *bond)
3548 #ifdef CONFIG_PROC_FS
3549 bond_remove_proc_entry(bond);
3550 bond_create_proc_entry(bond);
3552 down_write(&(bonding_rwsem));
3553 bond_destroy_sysfs_entry(bond);
3554 bond_create_sysfs_entry(bond);
3555 up_write(&(bonding_rwsem));
3559 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3561 struct bonding *event_bond = bond_dev->priv;
3564 case NETDEV_CHANGENAME:
3565 return bond_event_changename(event_bond);
3566 case NETDEV_UNREGISTER:
3568 * TODO: remove a bond from the list?
3578 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3580 struct net_device *bond_dev = slave_dev->master;
3581 struct bonding *bond = bond_dev->priv;
3584 case NETDEV_UNREGISTER:
3586 bond_release(bond_dev, slave_dev);
3591 * TODO: is this what we get if somebody
3592 * sets up a hierarchical bond, then rmmod's
3593 * one of the slave bonding devices?
3598 * ... Or is it this?
3601 case NETDEV_CHANGEMTU:
3603 * TODO: Should slaves be allowed to
3604 * independently alter their MTU? For
3605 * an active-backup bond, slaves need
3606 * not be the same type of device, so
3607 * MTUs may vary. For other modes,
3608 * slaves arguably should have the
3609 * same MTUs. To do this, we'd need to
3610 * take over the slave's change_mtu
3611 * function for the duration of their
3615 case NETDEV_CHANGENAME:
3617 * TODO: handle changing the primary's name
3620 case NETDEV_FEAT_CHANGE:
3621 bond_compute_features(bond);
3631 * bond_netdev_event: handle netdev notifier chain events.
3633 * This function receives events for the netdev chain. The caller (an
3634 * ioctl handler calling notifier_call_chain) holds the necessary
3635 * locks for us to safely manipulate the slave devices (RTNL lock,
3638 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3640 struct net_device *event_dev = (struct net_device *)ptr;
3642 dprintk("event_dev: %s, event: %lx\n",
3643 (event_dev ? event_dev->name : "None"),
3646 if (event_dev->flags & IFF_MASTER) {
3647 dprintk("IFF_MASTER\n");
3648 return bond_master_netdev_event(event, event_dev);
3651 if (event_dev->flags & IFF_SLAVE) {
3652 dprintk("IFF_SLAVE\n");
3653 return bond_slave_netdev_event(event, event_dev);
3660 * bond_inetaddr_event: handle inetaddr notifier chain events.
3662 * We keep track of device IPs primarily to use as source addresses in
3663 * ARP monitor probes (rather than spewing out broadcasts all the time).
3665 * We track one IP for the main device (if it has one), plus one per VLAN.
3667 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3669 struct in_ifaddr *ifa = ptr;
3670 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3671 struct bonding *bond, *bond_next;
3672 struct vlan_entry *vlan, *vlan_next;
3674 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3675 if (bond->dev == event_dev) {
3678 bond->master_ip = ifa->ifa_local;
3681 bond->master_ip = bond_glean_dev_ip(bond->dev);
3688 if (list_empty(&bond->vlan_list))
3691 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3693 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3694 if (vlan_dev == event_dev) {
3697 vlan->vlan_ip = ifa->ifa_local;
3701 bond_glean_dev_ip(vlan_dev);
3712 static struct notifier_block bond_netdev_notifier = {
3713 .notifier_call = bond_netdev_event,
3716 static struct notifier_block bond_inetaddr_notifier = {
3717 .notifier_call = bond_inetaddr_event,
3720 /*-------------------------- Packet type handling ---------------------------*/
3722 /* register to receive lacpdus on a bond */
3723 static void bond_register_lacpdu(struct bonding *bond)
3725 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3727 /* initialize packet type */
3728 pk_type->type = PKT_TYPE_LACPDU;
3729 pk_type->dev = bond->dev;
3730 pk_type->func = bond_3ad_lacpdu_recv;
3732 dev_add_pack(pk_type);
3735 /* unregister to receive lacpdus on a bond */
3736 static void bond_unregister_lacpdu(struct bonding *bond)
3738 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3741 /*---------------------------- Hashing Policies -----------------------------*/
3744 * Hash for the the output device based upon layer 3 and layer 4 data. If
3745 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3746 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3748 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3749 struct net_device *bond_dev, int count)
3751 struct ethhdr *data = (struct ethhdr *)skb->data;
3752 struct iphdr *iph = skb->nh.iph;
3753 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3756 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3757 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3758 (iph->protocol == IPPROTO_TCP ||
3759 iph->protocol == IPPROTO_UDP)) {
3760 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3762 return (layer4_xor ^
3763 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3767 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3771 * Hash for the output device based upon layer 2 data
3773 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3774 struct net_device *bond_dev, int count)
3776 struct ethhdr *data = (struct ethhdr *)skb->data;
3778 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3781 /*-------------------------- Device entry points ----------------------------*/
3783 static int bond_open(struct net_device *bond_dev)
3785 struct bonding *bond = bond_dev->priv;
3786 struct timer_list *mii_timer = &bond->mii_timer;
3787 struct timer_list *arp_timer = &bond->arp_timer;
3789 bond->kill_timers = 0;
3791 if ((bond->params.mode == BOND_MODE_TLB) ||
3792 (bond->params.mode == BOND_MODE_ALB)) {
3793 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3795 /* bond_alb_initialize must be called before the timer
3798 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3799 /* something went wrong - fail the open operation */
3803 init_timer(alb_timer);
3804 alb_timer->expires = jiffies + 1;
3805 alb_timer->data = (unsigned long)bond;
3806 alb_timer->function = (void *)&bond_alb_monitor;
3807 add_timer(alb_timer);
3810 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3811 init_timer(mii_timer);
3812 mii_timer->expires = jiffies + 1;
3813 mii_timer->data = (unsigned long)bond_dev;
3814 mii_timer->function = (void *)&bond_mii_monitor;
3815 add_timer(mii_timer);
3818 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3819 init_timer(arp_timer);
3820 arp_timer->expires = jiffies + 1;
3821 arp_timer->data = (unsigned long)bond_dev;
3822 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3823 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3825 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3827 add_timer(arp_timer);
3830 if (bond->params.mode == BOND_MODE_8023AD) {
3831 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3832 init_timer(ad_timer);
3833 ad_timer->expires = jiffies + 1;
3834 ad_timer->data = (unsigned long)bond;
3835 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3836 add_timer(ad_timer);
3838 /* register to receive LACPDUs */
3839 bond_register_lacpdu(bond);
3845 static int bond_close(struct net_device *bond_dev)
3847 struct bonding *bond = bond_dev->priv;
3849 if (bond->params.mode == BOND_MODE_8023AD) {
3850 /* Unregister the receive of LACPDUs */
3851 bond_unregister_lacpdu(bond);
3854 write_lock_bh(&bond->lock);
3856 bond_mc_list_destroy(bond);
3858 /* signal timers not to re-arm */
3859 bond->kill_timers = 1;
3861 write_unlock_bh(&bond->lock);
3863 /* del_timer_sync must run without holding the bond->lock
3864 * because a running timer might be trying to hold it too
3867 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3868 del_timer_sync(&bond->mii_timer);
3871 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3872 del_timer_sync(&bond->arp_timer);
3875 switch (bond->params.mode) {
3876 case BOND_MODE_8023AD:
3877 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3881 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3887 /* Release the bonded slaves */
3888 bond_release_all(bond_dev);
3890 if ((bond->params.mode == BOND_MODE_TLB) ||
3891 (bond->params.mode == BOND_MODE_ALB)) {
3892 /* Must be called only after all
3893 * slaves have been released
3895 bond_alb_deinitialize(bond);
3901 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3903 struct bonding *bond = bond_dev->priv;
3904 struct net_device_stats *stats = &(bond->stats), *sstats;
3905 struct slave *slave;
3908 memset(stats, 0, sizeof(struct net_device_stats));
3910 read_lock_bh(&bond->lock);
3912 bond_for_each_slave(bond, slave, i) {
3913 sstats = slave->dev->get_stats(slave->dev);
3915 stats->rx_packets += sstats->rx_packets;
3916 stats->rx_bytes += sstats->rx_bytes;
3917 stats->rx_errors += sstats->rx_errors;
3918 stats->rx_dropped += sstats->rx_dropped;
3920 stats->tx_packets += sstats->tx_packets;
3921 stats->tx_bytes += sstats->tx_bytes;
3922 stats->tx_errors += sstats->tx_errors;
3923 stats->tx_dropped += sstats->tx_dropped;
3925 stats->multicast += sstats->multicast;
3926 stats->collisions += sstats->collisions;
3928 stats->rx_length_errors += sstats->rx_length_errors;
3929 stats->rx_over_errors += sstats->rx_over_errors;
3930 stats->rx_crc_errors += sstats->rx_crc_errors;
3931 stats->rx_frame_errors += sstats->rx_frame_errors;
3932 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3933 stats->rx_missed_errors += sstats->rx_missed_errors;
3935 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3936 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3937 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3938 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3939 stats->tx_window_errors += sstats->tx_window_errors;
3942 read_unlock_bh(&bond->lock);
3947 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3949 struct net_device *slave_dev = NULL;
3950 struct ifbond k_binfo;
3951 struct ifbond __user *u_binfo = NULL;
3952 struct ifslave k_sinfo;
3953 struct ifslave __user *u_sinfo = NULL;
3954 struct mii_ioctl_data *mii = NULL;
3957 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3958 bond_dev->name, cmd);
3970 * We do this again just in case we were called by SIOCGMIIREG
3971 * instead of SIOCGMIIPHY.
3978 if (mii->reg_num == 1) {
3979 struct bonding *bond = bond_dev->priv;
3981 read_lock_bh(&bond->lock);
3982 read_lock(&bond->curr_slave_lock);
3983 if (bond->curr_active_slave) {
3984 mii->val_out = BMSR_LSTATUS;
3986 read_unlock(&bond->curr_slave_lock);
3987 read_unlock_bh(&bond->lock);
3991 case BOND_INFO_QUERY_OLD:
3992 case SIOCBONDINFOQUERY:
3993 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3995 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3999 res = bond_info_query(bond_dev, &k_binfo);
4001 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
4007 case BOND_SLAVE_INFO_QUERY_OLD:
4008 case SIOCBONDSLAVEINFOQUERY:
4009 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4011 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4015 res = bond_slave_info_query(bond_dev, &k_sinfo);
4017 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4028 if (!capable(CAP_NET_ADMIN)) {
4032 down_write(&(bonding_rwsem));
4033 slave_dev = dev_get_by_name(ifr->ifr_slave);
4035 dprintk("slave_dev=%p: \n", slave_dev);
4040 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4042 case BOND_ENSLAVE_OLD:
4043 case SIOCBONDENSLAVE:
4044 res = bond_enslave(bond_dev, slave_dev);
4046 case BOND_RELEASE_OLD:
4047 case SIOCBONDRELEASE:
4048 res = bond_release(bond_dev, slave_dev);
4050 case BOND_SETHWADDR_OLD:
4051 case SIOCBONDSETHWADDR:
4052 res = bond_sethwaddr(bond_dev, slave_dev);
4054 case BOND_CHANGE_ACTIVE_OLD:
4055 case SIOCBONDCHANGEACTIVE:
4056 res = bond_ioctl_change_active(bond_dev, slave_dev);
4065 up_write(&(bonding_rwsem));
4069 static void bond_set_multicast_list(struct net_device *bond_dev)
4071 struct bonding *bond = bond_dev->priv;
4072 struct dev_mc_list *dmi;
4074 write_lock_bh(&bond->lock);
4077 * Do promisc before checking multicast_mode
4079 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4080 bond_set_promiscuity(bond, 1);
4083 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4084 bond_set_promiscuity(bond, -1);
4087 /* set allmulti flag to slaves */
4088 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4089 bond_set_allmulti(bond, 1);
4092 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4093 bond_set_allmulti(bond, -1);
4096 bond->flags = bond_dev->flags;
4098 /* looking for addresses to add to slaves' mc list */
4099 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4100 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4101 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4105 /* looking for addresses to delete from slaves' list */
4106 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4107 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4108 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4112 /* save master's multicast list */
4113 bond_mc_list_destroy(bond);
4114 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4116 write_unlock_bh(&bond->lock);
4120 * Change the MTU of all of a master's slaves to match the master
4122 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4124 struct bonding *bond = bond_dev->priv;
4125 struct slave *slave, *stop_at;
4129 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4130 (bond_dev ? bond_dev->name : "None"), new_mtu);
4132 /* Can't hold bond->lock with bh disabled here since
4133 * some base drivers panic. On the other hand we can't
4134 * hold bond->lock without bh disabled because we'll
4135 * deadlock. The only solution is to rely on the fact
4136 * that we're under rtnl_lock here, and the slaves
4137 * list won't change. This doesn't solve the problem
4138 * of setting the slave's MTU while it is
4139 * transmitting, but the assumption is that the base
4140 * driver can handle that.
4142 * TODO: figure out a way to safely iterate the slaves
4143 * list, but without holding a lock around the actual
4144 * call to the base driver.
4147 bond_for_each_slave(bond, slave, i) {
4148 dprintk("s %p s->p %p c_m %p\n", slave,
4149 slave->prev, slave->dev->change_mtu);
4151 res = dev_set_mtu(slave->dev, new_mtu);
4154 /* If we failed to set the slave's mtu to the new value
4155 * we must abort the operation even in ACTIVE_BACKUP
4156 * mode, because if we allow the backup slaves to have
4157 * different mtu values than the active slave we'll
4158 * need to change their mtu when doing a failover. That
4159 * means changing their mtu from timer context, which
4160 * is probably not a good idea.
4162 dprintk("err %d %s\n", res, slave->dev->name);
4167 bond_dev->mtu = new_mtu;
4172 /* unwind from head to the slave that failed */
4174 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4177 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4179 dprintk("unwind err %d dev %s\n", tmp_res,
4190 * Note that many devices must be down to change the HW address, and
4191 * downing the master releases all slaves. We can make bonds full of
4192 * bonding devices to test this, however.
4194 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4196 struct bonding *bond = bond_dev->priv;
4197 struct sockaddr *sa = addr, tmp_sa;
4198 struct slave *slave, *stop_at;
4202 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4204 if (!is_valid_ether_addr(sa->sa_data)) {
4205 return -EADDRNOTAVAIL;
4208 /* Can't hold bond->lock with bh disabled here since
4209 * some base drivers panic. On the other hand we can't
4210 * hold bond->lock without bh disabled because we'll
4211 * deadlock. The only solution is to rely on the fact
4212 * that we're under rtnl_lock here, and the slaves
4213 * list won't change. This doesn't solve the problem
4214 * of setting the slave's hw address while it is
4215 * transmitting, but the assumption is that the base
4216 * driver can handle that.
4218 * TODO: figure out a way to safely iterate the slaves
4219 * list, but without holding a lock around the actual
4220 * call to the base driver.
4223 bond_for_each_slave(bond, slave, i) {
4224 dprintk("slave %p %s\n", slave, slave->dev->name);
4226 if (slave->dev->set_mac_address == NULL) {
4228 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4232 res = dev_set_mac_address(slave->dev, addr);
4234 /* TODO: consider downing the slave
4236 * User should expect communications
4237 * breakage anyway until ARP finish
4240 dprintk("err %d %s\n", res, slave->dev->name);
4246 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4250 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4251 tmp_sa.sa_family = bond_dev->type;
4253 /* unwind from head to the slave that failed */
4255 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4258 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4260 dprintk("unwind err %d dev %s\n", tmp_res,
4268 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4270 struct bonding *bond = bond_dev->priv;
4271 struct slave *slave, *start_at;
4275 read_lock(&bond->lock);
4277 if (!BOND_IS_OK(bond)) {
4281 read_lock(&bond->curr_slave_lock);
4282 slave = start_at = bond->curr_active_slave;
4283 read_unlock(&bond->curr_slave_lock);
4289 bond_for_each_slave_from(bond, slave, i, start_at) {
4290 if (IS_UP(slave->dev) &&
4291 (slave->link == BOND_LINK_UP) &&
4292 (slave->state == BOND_STATE_ACTIVE)) {
4293 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4295 write_lock(&bond->curr_slave_lock);
4296 bond->curr_active_slave = slave->next;
4297 write_unlock(&bond->curr_slave_lock);
4306 /* no suitable interface, frame not sent */
4309 read_unlock(&bond->lock);
4313 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
4314 struct bonding *bond,
4315 struct slave *slave)
4317 struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
4318 struct ethhdr *eth_data;
4323 printk(KERN_ERR DRV_NAME ": Error: "
4324 "bond_activebackup_xmit_copy(): skb_copy() failed\n");
4328 skb2->mac.raw = (unsigned char *)skb2->data;
4329 eth_data = eth_hdr(skb2);
4331 /* Pick an appropriate source MAC address
4332 * -- use slave's perm MAC addr, unless used by bond
4333 * -- otherwise, borrow active slave's perm MAC addr
4334 * since that will not be used
4336 hwaddr = slave->perm_hwaddr;
4337 if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
4338 hwaddr = bond->curr_active_slave->perm_hwaddr;
4340 /* Set source MAC address appropriately */
4341 memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
4343 res = bond_dev_queue_xmit(bond, skb2, slave->dev);
4345 dev_kfree_skb(skb2);
4351 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4352 * the bond has a usable interface.
4354 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4356 struct bonding *bond = bond_dev->priv;
4359 read_lock(&bond->lock);
4360 read_lock(&bond->curr_slave_lock);
4362 if (!BOND_IS_OK(bond)) {
4366 if (!bond->curr_active_slave)
4369 /* Xmit IGMP frames on all slaves to ensure rapid fail-over
4370 for multicast traffic on snooping switches */
4371 if (skb->protocol == __constant_htons(ETH_P_IP) &&
4372 skb->nh.iph->protocol == IPPROTO_IGMP) {
4373 struct slave *slave, *active_slave;
4376 active_slave = bond->curr_active_slave;
4377 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
4379 if (IS_UP(slave->dev) &&
4380 (slave->link == BOND_LINK_UP))
4381 bond_activebackup_xmit_copy(skb, bond, slave);
4384 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4388 /* no suitable interface, frame not sent */
4391 read_unlock(&bond->curr_slave_lock);
4392 read_unlock(&bond->lock);
4397 * In bond_xmit_xor() , we determine the output device by using a pre-
4398 * determined xmit_hash_policy(), If the selected device is not enabled,
4399 * find the next active slave.
4401 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4403 struct bonding *bond = bond_dev->priv;
4404 struct slave *slave, *start_at;
4409 read_lock(&bond->lock);
4411 if (!BOND_IS_OK(bond)) {
4415 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4417 bond_for_each_slave(bond, slave, i) {
4426 bond_for_each_slave_from(bond, slave, i, start_at) {
4427 if (IS_UP(slave->dev) &&
4428 (slave->link == BOND_LINK_UP) &&
4429 (slave->state == BOND_STATE_ACTIVE)) {
4430 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4437 /* no suitable interface, frame not sent */
4440 read_unlock(&bond->lock);
4445 * in broadcast mode, we send everything to all usable interfaces.
4447 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4449 struct bonding *bond = bond_dev->priv;
4450 struct slave *slave, *start_at;
4451 struct net_device *tx_dev = NULL;
4455 read_lock(&bond->lock);
4457 if (!BOND_IS_OK(bond)) {
4461 read_lock(&bond->curr_slave_lock);
4462 start_at = bond->curr_active_slave;
4463 read_unlock(&bond->curr_slave_lock);
4469 bond_for_each_slave_from(bond, slave, i, start_at) {
4470 if (IS_UP(slave->dev) &&
4471 (slave->link == BOND_LINK_UP) &&
4472 (slave->state == BOND_STATE_ACTIVE)) {
4474 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4476 printk(KERN_ERR DRV_NAME
4477 ": %s: Error: bond_xmit_broadcast(): "
4478 "skb_clone() failed\n",
4483 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4485 dev_kfree_skb(skb2);
4489 tx_dev = slave->dev;
4494 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4499 /* no suitable interface, frame not sent */
4502 /* frame sent to all suitable interfaces */
4503 read_unlock(&bond->lock);
4507 /*------------------------- Device initialization ---------------------------*/
4510 * set bond mode specific net device operations
4512 void bond_set_mode_ops(struct bonding *bond, int mode)
4514 struct net_device *bond_dev = bond->dev;
4517 case BOND_MODE_ROUNDROBIN:
4518 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4520 case BOND_MODE_ACTIVEBACKUP:
4521 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4524 bond_dev->hard_start_xmit = bond_xmit_xor;
4525 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4526 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4528 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4530 case BOND_MODE_BROADCAST:
4531 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4533 case BOND_MODE_8023AD:
4534 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4535 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4536 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4538 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4542 bond_dev->hard_start_xmit = bond_alb_xmit;
4543 bond_dev->set_mac_address = bond_alb_set_mac_address;
4546 /* Should never happen, mode already checked */
4547 printk(KERN_ERR DRV_NAME
4548 ": %s: Error: Unknown bonding mode %d\n",
4555 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4556 struct ethtool_drvinfo *drvinfo)
4558 strncpy(drvinfo->driver, DRV_NAME, 32);
4559 strncpy(drvinfo->version, DRV_VERSION, 32);
4560 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4563 static struct ethtool_ops bond_ethtool_ops = {
4564 .get_tx_csum = ethtool_op_get_tx_csum,
4565 .get_sg = ethtool_op_get_sg,
4566 .get_drvinfo = bond_ethtool_get_drvinfo,
4570 * Does not allocate but creates a /proc entry.
4573 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4575 struct bonding *bond = bond_dev->priv;
4577 dprintk("Begin bond_init for %s\n", bond_dev->name);
4579 /* initialize rwlocks */
4580 rwlock_init(&bond->lock);
4581 rwlock_init(&bond->curr_slave_lock);
4583 bond->params = *params; /* copy params struct */
4585 /* Initialize pointers */
4586 bond->first_slave = NULL;
4587 bond->curr_active_slave = NULL;
4588 bond->current_arp_slave = NULL;
4589 bond->primary_slave = NULL;
4590 bond->dev = bond_dev;
4591 INIT_LIST_HEAD(&bond->vlan_list);
4593 /* Initialize the device entry points */
4594 bond_dev->open = bond_open;
4595 bond_dev->stop = bond_close;
4596 bond_dev->get_stats = bond_get_stats;
4597 bond_dev->do_ioctl = bond_do_ioctl;
4598 bond_dev->ethtool_ops = &bond_ethtool_ops;
4599 bond_dev->set_multicast_list = bond_set_multicast_list;
4600 bond_dev->change_mtu = bond_change_mtu;
4601 bond_dev->set_mac_address = bond_set_mac_address;
4603 bond_set_mode_ops(bond, bond->params.mode);
4605 bond_dev->destructor = free_netdev;
4607 /* Initialize the device options */
4608 bond_dev->tx_queue_len = 0;
4609 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4611 /* At first, we block adding VLANs. That's the only way to
4612 * prevent problems that occur when adding VLANs over an
4613 * empty bond. The block will be removed once non-challenged
4614 * slaves are enslaved.
4616 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4618 /* don't acquire bond device's xmit_lock when
4620 bond_dev->features |= NETIF_F_LLTX;
4622 /* By default, we declare the bond to be fully
4623 * VLAN hardware accelerated capable. Special
4624 * care is taken in the various xmit functions
4625 * when there are slaves that are not hw accel
4628 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4629 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4630 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4631 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4632 NETIF_F_HW_VLAN_RX |
4633 NETIF_F_HW_VLAN_FILTER);
4635 #ifdef CONFIG_PROC_FS
4636 bond_create_proc_entry(bond);
4639 list_add_tail(&bond->bond_list, &bond_dev_list);
4644 /* De-initialize device specific data.
4645 * Caller must hold rtnl_lock.
4647 void bond_deinit(struct net_device *bond_dev)
4649 struct bonding *bond = bond_dev->priv;
4651 list_del(&bond->bond_list);
4653 #ifdef CONFIG_PROC_FS
4654 bond_remove_proc_entry(bond);
4658 /* Unregister and free all bond devices.
4659 * Caller must hold rtnl_lock.
4661 static void bond_free_all(void)
4663 struct bonding *bond, *nxt;
4665 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4666 struct net_device *bond_dev = bond->dev;
4668 unregister_netdevice(bond_dev);
4669 bond_deinit(bond_dev);
4672 #ifdef CONFIG_PROC_FS
4673 bond_destroy_proc_dir();
4677 /*------------------------- Module initialization ---------------------------*/
4680 * Convert string input module parms. Accept either the
4681 * number of the mode or its string name.
4683 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4687 for (i = 0; tbl[i].modename; i++) {
4688 if ((isdigit(*mode_arg) &&
4689 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4690 (strncmp(mode_arg, tbl[i].modename,
4691 strlen(tbl[i].modename)) == 0)) {
4699 static int bond_check_params(struct bond_params *params)
4702 * Convert string parameters.
4705 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4706 if (bond_mode == -1) {
4707 printk(KERN_ERR DRV_NAME
4708 ": Error: Invalid bonding mode \"%s\"\n",
4709 mode == NULL ? "NULL" : mode);
4714 if (xmit_hash_policy) {
4715 if ((bond_mode != BOND_MODE_XOR) &&
4716 (bond_mode != BOND_MODE_8023AD)) {
4717 printk(KERN_INFO DRV_NAME
4718 ": xor_mode param is irrelevant in mode %s\n",
4719 bond_mode_name(bond_mode));
4721 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4723 if (xmit_hashtype == -1) {
4724 printk(KERN_ERR DRV_NAME
4725 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4726 xmit_hash_policy == NULL ? "NULL" :
4734 if (bond_mode != BOND_MODE_8023AD) {
4735 printk(KERN_INFO DRV_NAME
4736 ": lacp_rate param is irrelevant in mode %s\n",
4737 bond_mode_name(bond_mode));
4739 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4740 if (lacp_fast == -1) {
4741 printk(KERN_ERR DRV_NAME
4742 ": Error: Invalid lacp rate \"%s\"\n",
4743 lacp_rate == NULL ? "NULL" : lacp_rate);
4749 if (max_bonds < 1 || max_bonds > INT_MAX) {
4750 printk(KERN_WARNING DRV_NAME
4751 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4752 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4753 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4754 max_bonds = BOND_DEFAULT_MAX_BONDS;
4758 printk(KERN_WARNING DRV_NAME
4759 ": Warning: miimon module parameter (%d), "
4760 "not in range 0-%d, so it was reset to %d\n",
4761 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4762 miimon = BOND_LINK_MON_INTERV;
4766 printk(KERN_WARNING DRV_NAME
4767 ": Warning: updelay module parameter (%d), "
4768 "not in range 0-%d, so it was reset to 0\n",
4773 if (downdelay < 0) {
4774 printk(KERN_WARNING DRV_NAME
4775 ": Warning: downdelay module parameter (%d), "
4776 "not in range 0-%d, so it was reset to 0\n",
4777 downdelay, INT_MAX);
4781 if ((use_carrier != 0) && (use_carrier != 1)) {
4782 printk(KERN_WARNING DRV_NAME
4783 ": Warning: use_carrier module parameter (%d), "
4784 "not of valid value (0/1), so it was set to 1\n",
4789 /* reset values for 802.3ad */
4790 if (bond_mode == BOND_MODE_8023AD) {
4792 printk(KERN_WARNING DRV_NAME
4793 ": Warning: miimon must be specified, "
4794 "otherwise bonding will not detect link "
4795 "failure, speed and duplex which are "
4796 "essential for 802.3ad operation\n");
4797 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4802 /* reset values for TLB/ALB */
4803 if ((bond_mode == BOND_MODE_TLB) ||
4804 (bond_mode == BOND_MODE_ALB)) {
4806 printk(KERN_WARNING DRV_NAME
4807 ": Warning: miimon must be specified, "
4808 "otherwise bonding will not detect link "
4809 "failure and link speed which are essential "
4810 "for TLB/ALB load balancing\n");
4811 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4816 if (bond_mode == BOND_MODE_ALB) {
4817 printk(KERN_NOTICE DRV_NAME
4818 ": In ALB mode you might experience client "
4819 "disconnections upon reconnection of a link if the "
4820 "bonding module updelay parameter (%d msec) is "
4821 "incompatible with the forwarding delay time of the "
4827 if (updelay || downdelay) {
4828 /* just warn the user the up/down delay will have
4829 * no effect since miimon is zero...
4831 printk(KERN_WARNING DRV_NAME
4832 ": Warning: miimon module parameter not set "
4833 "and updelay (%d) or downdelay (%d) module "
4834 "parameter is set; updelay and downdelay have "
4835 "no effect unless miimon is set\n",
4836 updelay, downdelay);
4839 /* don't allow arp monitoring */
4841 printk(KERN_WARNING DRV_NAME
4842 ": Warning: miimon (%d) and arp_interval (%d) "
4843 "can't be used simultaneously, disabling ARP "
4845 miimon, arp_interval);
4849 if ((updelay % miimon) != 0) {
4850 printk(KERN_WARNING DRV_NAME
4851 ": Warning: updelay (%d) is not a multiple "
4852 "of miimon (%d), updelay rounded to %d ms\n",
4853 updelay, miimon, (updelay / miimon) * miimon);
4858 if ((downdelay % miimon) != 0) {
4859 printk(KERN_WARNING DRV_NAME
4860 ": Warning: downdelay (%d) is not a multiple "
4861 "of miimon (%d), downdelay rounded to %d ms\n",
4863 (downdelay / miimon) * miimon);
4866 downdelay /= miimon;
4869 if (arp_interval < 0) {
4870 printk(KERN_WARNING DRV_NAME
4871 ": Warning: arp_interval module parameter (%d) "
4872 ", not in range 0-%d, so it was reset to %d\n",
4873 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4874 arp_interval = BOND_LINK_ARP_INTERV;
4877 for (arp_ip_count = 0;
4878 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4880 /* not complete check, but should be good enough to
4882 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4883 printk(KERN_WARNING DRV_NAME
4884 ": Warning: bad arp_ip_target module parameter "
4885 "(%s), ARP monitoring will not be performed\n",
4886 arp_ip_target[arp_ip_count]);
4889 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4890 arp_target[arp_ip_count] = ip;
4894 if (arp_interval && !arp_ip_count) {
4895 /* don't allow arping if no arp_ip_target given... */
4896 printk(KERN_WARNING DRV_NAME
4897 ": Warning: arp_interval module parameter (%d) "
4898 "specified without providing an arp_ip_target "
4899 "parameter, arp_interval was reset to 0\n",
4905 printk(KERN_INFO DRV_NAME
4906 ": MII link monitoring set to %d ms\n",
4908 } else if (arp_interval) {
4911 printk(KERN_INFO DRV_NAME
4912 ": ARP monitoring set to %d ms with %d target(s):",
4913 arp_interval, arp_ip_count);
4915 for (i = 0; i < arp_ip_count; i++)
4916 printk (" %s", arp_ip_target[i]);
4921 /* miimon and arp_interval not set, we need one so things
4922 * work as expected, see bonding.txt for details
4924 printk(KERN_WARNING DRV_NAME
4925 ": Warning: either miimon or arp_interval and "
4926 "arp_ip_target module parameters must be specified, "
4927 "otherwise bonding will not detect link failures! see "
4928 "bonding.txt for details.\n");
4931 if (primary && !USES_PRIMARY(bond_mode)) {
4932 /* currently, using a primary only makes sense
4933 * in active backup, TLB or ALB modes
4935 printk(KERN_WARNING DRV_NAME
4936 ": Warning: %s primary device specified but has no "
4937 "effect in %s mode\n",
4938 primary, bond_mode_name(bond_mode));
4942 /* fill params struct with the proper values */
4943 params->mode = bond_mode;
4944 params->xmit_policy = xmit_hashtype;
4945 params->miimon = miimon;
4946 params->arp_interval = arp_interval;
4947 params->updelay = updelay;
4948 params->downdelay = downdelay;
4949 params->use_carrier = use_carrier;
4950 params->lacp_fast = lacp_fast;
4951 params->primary[0] = 0;
4954 strncpy(params->primary, primary, IFNAMSIZ);
4955 params->primary[IFNAMSIZ - 1] = 0;
4958 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4963 /* Create a new bond based on the specified name and bonding parameters.
4964 * Caller must NOT hold rtnl_lock; we need to release it here before we
4965 * set up our sysfs entries.
4967 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4969 struct net_device *bond_dev;
4973 bond_dev = alloc_netdev(sizeof(struct bonding), name, ether_setup);
4975 printk(KERN_ERR DRV_NAME
4976 ": %s: eek! can't alloc netdev!\n",
4982 /* bond_init() must be called after dev_alloc_name() (for the
4983 * /proc files), but before register_netdevice(), because we
4984 * need to set function pointers.
4987 res = bond_init(bond_dev, params);
4992 SET_MODULE_OWNER(bond_dev);
4994 res = register_netdevice(bond_dev);
4999 *newbond = bond_dev->priv;
5001 rtnl_unlock(); /* allows sysfs registration of net device */
5002 res = bond_create_sysfs_entry(bond_dev->priv);
5005 bond_deinit(bond_dev);
5007 free_netdev(bond_dev);
5014 static int __init bonding_init(void)
5018 char new_bond_name[8]; /* Enough room for 999 bonds at init. */
5020 printk(KERN_INFO "%s", version);
5022 res = bond_check_params(&bonding_defaults);
5027 #ifdef CONFIG_PROC_FS
5028 bond_create_proc_dir();
5030 for (i = 0; i < max_bonds; i++) {
5031 sprintf(new_bond_name, "bond%d",i);
5032 res = bond_create(new_bond_name,&bonding_defaults, NULL);
5037 res = bond_create_sysfs();
5041 register_netdevice_notifier(&bond_netdev_notifier);
5042 register_inetaddr_notifier(&bond_inetaddr_notifier);
5048 bond_destroy_sysfs();
5055 static void __exit bonding_exit(void)
5057 unregister_netdevice_notifier(&bond_netdev_notifier);
5058 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5062 bond_destroy_sysfs();
5066 module_init(bonding_init);
5067 module_exit(bonding_exit);
5068 MODULE_LICENSE("GPL");
5069 MODULE_VERSION(DRV_VERSION);
5070 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5071 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5072 MODULE_SUPPORTED_DEVICE("most ethernet devices");