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.
494 //#define BONDING_DEBUG 1
496 #include <linux/config.h>
497 #include <linux/kernel.h>
498 #include <linux/module.h>
499 #include <linux/sched.h>
500 #include <linux/types.h>
501 #include <linux/fcntl.h>
502 #include <linux/interrupt.h>
503 #include <linux/ptrace.h>
504 #include <linux/ioport.h>
505 #include <linux/in.h>
507 #include <linux/ip.h>
508 #include <linux/tcp.h>
509 #include <linux/udp.h>
510 #include <linux/slab.h>
511 #include <linux/string.h>
512 #include <linux/init.h>
513 #include <linux/timer.h>
514 #include <linux/socket.h>
515 #include <linux/ctype.h>
516 #include <linux/inet.h>
517 #include <linux/bitops.h>
518 #include <asm/system.h>
521 #include <asm/uaccess.h>
522 #include <linux/errno.h>
523 #include <linux/netdevice.h>
524 #include <linux/inetdevice.h>
525 #include <linux/etherdevice.h>
526 #include <linux/skbuff.h>
527 #include <net/sock.h>
528 #include <linux/rtnetlink.h>
529 #include <linux/proc_fs.h>
530 #include <linux/seq_file.h>
531 #include <linux/smp.h>
532 #include <linux/if_ether.h>
534 #include <linux/mii.h>
535 #include <linux/ethtool.h>
536 #include <linux/if_vlan.h>
537 #include <linux/if_bonding.h>
538 #include <net/route.h>
540 #include "bond_3ad.h"
541 #include "bond_alb.h"
543 /*---------------------------- Module parameters ----------------------------*/
545 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
546 #define BOND_LINK_MON_INTERV 0
547 #define BOND_LINK_ARP_INTERV 0
549 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
550 static int miimon = BOND_LINK_MON_INTERV;
551 static int updelay = 0;
552 static int downdelay = 0;
553 static int use_carrier = 1;
554 static char *mode = NULL;
555 static char *primary = NULL;
556 static char *lacp_rate = NULL;
557 static char *xmit_hash_policy = NULL;
558 static int arp_interval = BOND_LINK_ARP_INTERV;
559 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
560 struct bond_params bonding_defaults;
562 module_param(max_bonds, int, 0);
563 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
564 module_param(miimon, int, 0);
565 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
566 module_param(updelay, int, 0);
567 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
568 module_param(downdelay, int, 0);
569 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
571 module_param(use_carrier, int, 0);
572 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
573 "0 for off, 1 for on (default)");
574 module_param(mode, charp, 0);
575 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
576 "1 for active-backup, 2 for balance-xor, "
577 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
578 "6 for balance-alb");
579 module_param(primary, charp, 0);
580 MODULE_PARM_DESC(primary, "Primary network device to use");
581 module_param(lacp_rate, charp, 0);
582 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
584 module_param(xmit_hash_policy, charp, 0);
585 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
586 ", 1 for layer 3+4");
587 module_param(arp_interval, int, 0);
588 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
589 module_param_array(arp_ip_target, charp, NULL, 0);
590 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
592 /*----------------------------- Global variables ----------------------------*/
594 static const char *version =
595 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
597 LIST_HEAD(bond_dev_list);
599 #ifdef CONFIG_PROC_FS
600 static struct proc_dir_entry *bond_proc_dir = NULL;
603 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
604 static int arp_ip_count = 0;
605 static int bond_mode = BOND_MODE_ROUNDROBIN;
606 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
607 static int lacp_fast = 0;
610 struct bond_parm_tbl bond_lacp_tbl[] = {
611 { "slow", AD_LACP_SLOW},
612 { "fast", AD_LACP_FAST},
616 struct bond_parm_tbl bond_mode_tbl[] = {
617 { "balance-rr", BOND_MODE_ROUNDROBIN},
618 { "active-backup", BOND_MODE_ACTIVEBACKUP},
619 { "balance-xor", BOND_MODE_XOR},
620 { "broadcast", BOND_MODE_BROADCAST},
621 { "802.3ad", BOND_MODE_8023AD},
622 { "balance-tlb", BOND_MODE_TLB},
623 { "balance-alb", BOND_MODE_ALB},
627 struct bond_parm_tbl xmit_hashtype_tbl[] = {
628 { "layer2", BOND_XMIT_POLICY_LAYER2},
629 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
633 /*-------------------------- Forward declarations ---------------------------*/
635 static void bond_send_gratuitous_arp(struct bonding *bond);
637 /*---------------------------- General routines -----------------------------*/
639 const char *bond_mode_name(int mode)
642 case BOND_MODE_ROUNDROBIN :
643 return "load balancing (round-robin)";
644 case BOND_MODE_ACTIVEBACKUP :
645 return "fault-tolerance (active-backup)";
647 return "load balancing (xor)";
648 case BOND_MODE_BROADCAST :
649 return "fault-tolerance (broadcast)";
650 case BOND_MODE_8023AD:
651 return "IEEE 802.3ad Dynamic link aggregation";
653 return "transmit load balancing";
655 return "adaptive load balancing";
661 /*---------------------------------- VLAN -----------------------------------*/
664 * bond_add_vlan - add a new vlan id on bond
665 * @bond: bond that got the notification
666 * @vlan_id: the vlan id to add
668 * Returns -ENOMEM if allocation failed.
670 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
672 struct vlan_entry *vlan;
674 dprintk("bond: %s, vlan id %d\n",
675 (bond ? bond->dev->name: "None"), vlan_id);
677 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
682 INIT_LIST_HEAD(&vlan->vlan_list);
683 vlan->vlan_id = vlan_id;
686 write_lock_bh(&bond->lock);
688 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
690 write_unlock_bh(&bond->lock);
692 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
698 * bond_del_vlan - delete a vlan id from bond
699 * @bond: bond that got the notification
700 * @vlan_id: the vlan id to delete
702 * returns -ENODEV if @vlan_id was not found in @bond.
704 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
706 struct vlan_entry *vlan, *next;
709 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
711 write_lock_bh(&bond->lock);
713 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
714 if (vlan->vlan_id == vlan_id) {
715 list_del(&vlan->vlan_list);
717 if ((bond->params.mode == BOND_MODE_TLB) ||
718 (bond->params.mode == BOND_MODE_ALB)) {
719 bond_alb_clear_vlan(bond, vlan_id);
722 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
727 if (list_empty(&bond->vlan_list) &&
728 (bond->slave_cnt == 0)) {
729 /* Last VLAN removed and no slaves, so
730 * restore block on adding VLANs. This will
731 * be removed once new slaves that are not
732 * VLAN challenged will be added.
734 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
742 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
746 write_unlock_bh(&bond->lock);
751 * bond_has_challenged_slaves
752 * @bond: the bond we're working on
754 * Searches the slave list. Returns 1 if a vlan challenged slave
755 * was found, 0 otherwise.
757 * Assumes bond->lock is held.
759 static int bond_has_challenged_slaves(struct bonding *bond)
764 bond_for_each_slave(bond, slave, i) {
765 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
766 dprintk("found VLAN challenged slave - %s\n",
772 dprintk("no VLAN challenged slaves found\n");
777 * bond_next_vlan - safely skip to the next item in the vlans list.
778 * @bond: the bond we're working on
779 * @curr: item we're advancing from
781 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
782 * or @curr->next otherwise (even if it is @curr itself again).
784 * Caller must hold bond->lock
786 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
788 struct vlan_entry *next, *last;
790 if (list_empty(&bond->vlan_list)) {
795 next = list_entry(bond->vlan_list.next,
796 struct vlan_entry, vlan_list);
798 last = list_entry(bond->vlan_list.prev,
799 struct vlan_entry, vlan_list);
801 next = list_entry(bond->vlan_list.next,
802 struct vlan_entry, vlan_list);
804 next = list_entry(curr->vlan_list.next,
805 struct vlan_entry, vlan_list);
813 * bond_dev_queue_xmit - Prepare skb for xmit.
815 * @bond: bond device that got this skb for tx.
816 * @skb: hw accel VLAN tagged skb to transmit
817 * @slave_dev: slave that is supposed to xmit this skbuff
819 * When the bond gets an skb to transmit that is
820 * already hardware accelerated VLAN tagged, and it
821 * needs to relay this skb to a slave that is not
822 * hw accel capable, the skb needs to be "unaccelerated",
823 * i.e. strip the hwaccel tag and re-insert it as part
826 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
828 unsigned short vlan_id;
830 if (!list_empty(&bond->vlan_list) &&
831 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
832 vlan_get_tag(skb, &vlan_id) == 0) {
833 skb->dev = slave_dev;
834 skb = vlan_put_tag(skb, vlan_id);
836 /* vlan_put_tag() frees the skb in case of error,
837 * so return success here so the calling functions
838 * won't attempt to free is again.
843 skb->dev = slave_dev;
853 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
854 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
856 * a. This operation is performed in IOCTL context,
857 * b. The operation is protected by the RTNL semaphore in the 8021q code,
858 * c. Holding a lock with BH disabled while directly calling a base driver
859 * entry point is generally a BAD idea.
861 * The design of synchronization/protection for this operation in the 8021q
862 * module is good for one or more VLAN devices over a single physical device
863 * and cannot be extended for a teaming solution like bonding, so there is a
864 * potential race condition here where a net device from the vlan group might
865 * be referenced (either by a base driver or the 8021q code) while it is being
866 * removed from the system. However, it turns out we're not making matters
867 * worse, and if it works for regular VLAN usage it will work here too.
871 * bond_vlan_rx_register - Propagates registration to slaves
872 * @bond_dev: bonding net device that got called
873 * @grp: vlan group being registered
875 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
877 struct bonding *bond = bond_dev->priv;
883 bond_for_each_slave(bond, slave, i) {
884 struct net_device *slave_dev = slave->dev;
886 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
887 slave_dev->vlan_rx_register) {
888 slave_dev->vlan_rx_register(slave_dev, grp);
894 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
895 * @bond_dev: bonding net device that got called
896 * @vid: vlan id being added
898 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
900 struct bonding *bond = bond_dev->priv;
904 bond_for_each_slave(bond, slave, i) {
905 struct net_device *slave_dev = slave->dev;
907 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
908 slave_dev->vlan_rx_add_vid) {
909 slave_dev->vlan_rx_add_vid(slave_dev, vid);
913 res = bond_add_vlan(bond, vid);
915 printk(KERN_ERR DRV_NAME
916 ": %s: Error: Failed to add vlan id %d\n",
917 bond_dev->name, vid);
922 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
923 * @bond_dev: bonding net device that got called
924 * @vid: vlan id being removed
926 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
928 struct bonding *bond = bond_dev->priv;
930 struct net_device *vlan_dev;
933 bond_for_each_slave(bond, slave, i) {
934 struct net_device *slave_dev = slave->dev;
936 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
937 slave_dev->vlan_rx_kill_vid) {
938 /* Save and then restore vlan_dev in the grp array,
939 * since the slave's driver might clear it.
941 vlan_dev = bond->vlgrp->vlan_devices[vid];
942 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
943 bond->vlgrp->vlan_devices[vid] = vlan_dev;
947 res = bond_del_vlan(bond, vid);
949 printk(KERN_ERR DRV_NAME
950 ": %s: Error: Failed to remove vlan id %d\n",
951 bond_dev->name, vid);
955 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
957 struct vlan_entry *vlan;
959 write_lock_bh(&bond->lock);
961 if (list_empty(&bond->vlan_list)) {
965 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
966 slave_dev->vlan_rx_register) {
967 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
970 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
971 !(slave_dev->vlan_rx_add_vid)) {
975 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
976 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
980 write_unlock_bh(&bond->lock);
983 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
985 struct vlan_entry *vlan;
986 struct net_device *vlan_dev;
988 write_lock_bh(&bond->lock);
990 if (list_empty(&bond->vlan_list)) {
994 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
995 !(slave_dev->vlan_rx_kill_vid)) {
999 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
1000 /* Save and then restore vlan_dev in the grp array,
1001 * since the slave's driver might clear it.
1003 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
1004 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
1005 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
1009 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
1010 slave_dev->vlan_rx_register) {
1011 slave_dev->vlan_rx_register(slave_dev, NULL);
1015 write_unlock_bh(&bond->lock);
1018 /*------------------------------- Link status -------------------------------*/
1021 * Get link speed and duplex from the slave's base driver
1022 * using ethtool. If for some reason the call fails or the
1023 * values are invalid, fake speed and duplex to 100/Full
1026 static int bond_update_speed_duplex(struct slave *slave)
1028 struct net_device *slave_dev = slave->dev;
1029 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1031 struct ethtool_cmd etool;
1033 /* Fake speed and duplex */
1034 slave->speed = SPEED_100;
1035 slave->duplex = DUPLEX_FULL;
1037 if (slave_dev->ethtool_ops) {
1040 if (!slave_dev->ethtool_ops->get_settings) {
1044 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
1052 ioctl = slave_dev->do_ioctl;
1053 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1054 etool.cmd = ETHTOOL_GSET;
1055 ifr.ifr_data = (char*)&etool;
1056 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
1061 switch (etool.speed) {
1070 switch (etool.duplex) {
1078 slave->speed = etool.speed;
1079 slave->duplex = etool.duplex;
1085 * if <dev> supports MII link status reporting, check its link status.
1087 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
1088 * depening upon the setting of the use_carrier parameter.
1090 * Return either BMSR_LSTATUS, meaning that the link is up (or we
1091 * can't tell and just pretend it is), or 0, meaning that the link is
1094 * If reporting is non-zero, instead of faking link up, return -1 if
1095 * both ETHTOOL and MII ioctls fail (meaning the device does not
1096 * support them). If use_carrier is set, return whatever it says.
1097 * It'd be nice if there was a good way to tell if a driver supports
1098 * netif_carrier, but there really isn't.
1100 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
1102 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1104 struct mii_ioctl_data *mii;
1105 struct ethtool_value etool;
1107 if (bond->params.use_carrier) {
1108 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1111 ioctl = slave_dev->do_ioctl;
1113 /* TODO: set pointer to correct ioctl on a per team member */
1114 /* bases to make this more efficient. that is, once */
1115 /* we determine the correct ioctl, we will always */
1116 /* call it and not the others for that team */
1120 * We cannot assume that SIOCGMIIPHY will also read a
1121 * register; not all network drivers (e.g., e100)
1125 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
1126 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1128 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
1129 mii->reg_num = MII_BMSR;
1130 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
1131 return (mii->val_out & BMSR_LSTATUS);
1136 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
1137 /* for a period of time so we attempt to get link status */
1138 /* from it last if the above MII ioctls fail... */
1139 if (slave_dev->ethtool_ops) {
1140 if (slave_dev->ethtool_ops->get_link) {
1143 link = slave_dev->ethtool_ops->get_link(slave_dev);
1145 return link ? BMSR_LSTATUS : 0;
1150 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1151 etool.cmd = ETHTOOL_GLINK;
1152 ifr.ifr_data = (char*)&etool;
1153 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
1154 if (etool.data == 1) {
1155 return BMSR_LSTATUS;
1157 dprintk("SIOCETHTOOL shows link down\n");
1164 * If reporting, report that either there's no dev->do_ioctl,
1165 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
1166 * cannot report link status). If not reporting, pretend
1169 return (reporting ? -1 : BMSR_LSTATUS);
1172 /*----------------------------- Multicast list ------------------------------*/
1175 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
1177 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
1179 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
1180 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
1184 * returns dmi entry if found, NULL otherwise
1186 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
1188 struct dev_mc_list *idmi;
1190 for (idmi = mc_list; idmi; idmi = idmi->next) {
1191 if (bond_is_dmi_same(dmi, idmi)) {
1200 * Push the promiscuity flag down to appropriate slaves
1202 static void bond_set_promiscuity(struct bonding *bond, int inc)
1204 if (USES_PRIMARY(bond->params.mode)) {
1205 /* write lock already acquired */
1206 if (bond->curr_active_slave) {
1207 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
1210 struct slave *slave;
1212 bond_for_each_slave(bond, slave, i) {
1213 dev_set_promiscuity(slave->dev, inc);
1219 * Push the allmulti flag down to all slaves
1221 static void bond_set_allmulti(struct bonding *bond, int inc)
1223 if (USES_PRIMARY(bond->params.mode)) {
1224 /* write lock already acquired */
1225 if (bond->curr_active_slave) {
1226 dev_set_allmulti(bond->curr_active_slave->dev, inc);
1229 struct slave *slave;
1231 bond_for_each_slave(bond, slave, i) {
1232 dev_set_allmulti(slave->dev, inc);
1238 * Add a Multicast address to slaves
1241 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
1243 if (USES_PRIMARY(bond->params.mode)) {
1244 /* write lock already acquired */
1245 if (bond->curr_active_slave) {
1246 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
1249 struct slave *slave;
1251 bond_for_each_slave(bond, slave, i) {
1252 dev_mc_add(slave->dev, addr, alen, 0);
1258 * Remove a multicast address from slave
1261 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
1263 if (USES_PRIMARY(bond->params.mode)) {
1264 /* write lock already acquired */
1265 if (bond->curr_active_slave) {
1266 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
1269 struct slave *slave;
1271 bond_for_each_slave(bond, slave, i) {
1272 dev_mc_delete(slave->dev, addr, alen, 0);
1278 * Totally destroys the mc_list in bond
1280 static void bond_mc_list_destroy(struct bonding *bond)
1282 struct dev_mc_list *dmi;
1284 dmi = bond->mc_list;
1286 bond->mc_list = dmi->next;
1288 dmi = bond->mc_list;
1293 * Copy all the Multicast addresses from src to the bonding device dst
1295 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
1298 struct dev_mc_list *dmi, *new_dmi;
1300 for (dmi = mc_list; dmi; dmi = dmi->next) {
1301 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
1304 /* FIXME: Potential memory leak !!! */
1308 new_dmi->next = bond->mc_list;
1309 bond->mc_list = new_dmi;
1310 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
1311 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
1312 new_dmi->dmi_users = dmi->dmi_users;
1313 new_dmi->dmi_gusers = dmi->dmi_gusers;
1320 * flush all members of flush->mc_list from device dev->mc_list
1322 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
1324 struct bonding *bond = bond_dev->priv;
1325 struct dev_mc_list *dmi;
1327 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1328 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1331 if (bond->params.mode == BOND_MODE_8023AD) {
1332 /* del lacpdu mc addr from mc list */
1333 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1335 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1339 /*--------------------------- Active slave change ---------------------------*/
1342 * Update the mc list and multicast-related flags for the new and
1343 * old active slaves (if any) according to the multicast mode, and
1344 * promiscuous flags unconditionally.
1346 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
1348 struct dev_mc_list *dmi;
1350 if (!USES_PRIMARY(bond->params.mode)) {
1351 /* nothing to do - mc list is already up-to-date on
1358 if (bond->dev->flags & IFF_PROMISC) {
1359 dev_set_promiscuity(old_active->dev, -1);
1362 if (bond->dev->flags & IFF_ALLMULTI) {
1363 dev_set_allmulti(old_active->dev, -1);
1366 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1367 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1372 if (bond->dev->flags & IFF_PROMISC) {
1373 dev_set_promiscuity(new_active->dev, 1);
1376 if (bond->dev->flags & IFF_ALLMULTI) {
1377 dev_set_allmulti(new_active->dev, 1);
1380 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1381 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1387 * find_best_interface - select the best available slave to be the active one
1388 * @bond: our bonding struct
1390 * Warning: Caller must hold curr_slave_lock for writing.
1392 static struct slave *bond_find_best_slave(struct bonding *bond)
1394 struct slave *new_active, *old_active;
1395 struct slave *bestslave = NULL;
1396 int mintime = bond->params.updelay;
1399 new_active = old_active = bond->curr_active_slave;
1401 if (!new_active) { /* there were no active slaves left */
1402 if (bond->slave_cnt > 0) { /* found one slave */
1403 new_active = bond->first_slave;
1405 return NULL; /* still no slave, return NULL */
1409 /* first try the primary link; if arping, a link must tx/rx traffic
1410 * before it can be considered the curr_active_slave - also, we would skip
1411 * slaves between the curr_active_slave and primary_slave that may be up
1414 if ((bond->primary_slave) &&
1415 (!bond->params.arp_interval) &&
1416 (IS_UP(bond->primary_slave->dev))) {
1417 new_active = bond->primary_slave;
1420 /* remember where to stop iterating over the slaves */
1421 old_active = new_active;
1423 bond_for_each_slave_from(bond, new_active, i, old_active) {
1424 if (IS_UP(new_active->dev)) {
1425 if (new_active->link == BOND_LINK_UP) {
1427 } else if (new_active->link == BOND_LINK_BACK) {
1428 /* link up, but waiting for stabilization */
1429 if (new_active->delay < mintime) {
1430 mintime = new_active->delay;
1431 bestslave = new_active;
1441 * change_active_interface - change the active slave into the specified one
1442 * @bond: our bonding struct
1443 * @new: the new slave to make the active one
1445 * Set the new slave to the bond's settings and unset them on the old
1446 * curr_active_slave.
1447 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1449 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1450 * because it is apparently the best available slave we have, even though its
1451 * updelay hasn't timed out yet.
1453 * Warning: Caller must hold curr_slave_lock for writing.
1455 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1457 struct slave *old_active = bond->curr_active_slave;
1459 if (old_active == new_active) {
1464 if (new_active->link == BOND_LINK_BACK) {
1465 if (USES_PRIMARY(bond->params.mode)) {
1466 printk(KERN_INFO DRV_NAME
1467 ": %s: making interface %s the new "
1468 "active one %d ms earlier.\n",
1469 bond->dev->name, new_active->dev->name,
1470 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1473 new_active->delay = 0;
1474 new_active->link = BOND_LINK_UP;
1475 new_active->jiffies = jiffies;
1477 if (bond->params.mode == BOND_MODE_8023AD) {
1478 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1481 if ((bond->params.mode == BOND_MODE_TLB) ||
1482 (bond->params.mode == BOND_MODE_ALB)) {
1483 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1486 if (USES_PRIMARY(bond->params.mode)) {
1487 printk(KERN_INFO DRV_NAME
1488 ": %s: making interface %s the new "
1490 bond->dev->name, new_active->dev->name);
1495 if (USES_PRIMARY(bond->params.mode)) {
1496 bond_mc_swap(bond, new_active, old_active);
1499 if ((bond->params.mode == BOND_MODE_TLB) ||
1500 (bond->params.mode == BOND_MODE_ALB)) {
1501 bond_alb_handle_active_change(bond, new_active);
1503 bond->curr_active_slave = new_active;
1506 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1508 bond_set_slave_inactive_flags(old_active);
1512 bond_set_slave_active_flags(new_active);
1514 bond_send_gratuitous_arp(bond);
1519 * bond_select_active_slave - select a new active slave, if needed
1520 * @bond: our bonding struct
1522 * This functions shoud be called when one of the following occurs:
1523 * - The old curr_active_slave has been released or lost its link.
1524 * - The primary_slave has got its link back.
1525 * - A slave has got its link back and there's no old curr_active_slave.
1527 * Warning: Caller must hold curr_slave_lock for writing.
1529 void bond_select_active_slave(struct bonding *bond)
1531 struct slave *best_slave;
1533 best_slave = bond_find_best_slave(bond);
1534 if (best_slave != bond->curr_active_slave) {
1535 bond_change_active_slave(bond, best_slave);
1539 /*--------------------------- slave list handling ---------------------------*/
1542 * This function attaches the slave to the end of list.
1544 * bond->lock held for writing by caller.
1546 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1548 if (bond->first_slave == NULL) { /* attaching the first slave */
1549 new_slave->next = new_slave;
1550 new_slave->prev = new_slave;
1551 bond->first_slave = new_slave;
1553 new_slave->next = bond->first_slave;
1554 new_slave->prev = bond->first_slave->prev;
1555 new_slave->next->prev = new_slave;
1556 new_slave->prev->next = new_slave;
1563 * This function detaches the slave from the list.
1564 * WARNING: no check is made to verify if the slave effectively
1565 * belongs to <bond>.
1566 * Nothing is freed on return, structures are just unchained.
1567 * If any slave pointer in bond was pointing to <slave>,
1568 * it should be changed by the calling function.
1570 * bond->lock held for writing by caller.
1572 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1575 slave->next->prev = slave->prev;
1579 slave->prev->next = slave->next;
1582 if (bond->first_slave == slave) { /* slave is the first slave */
1583 if (bond->slave_cnt > 1) { /* there are more slave */
1584 bond->first_slave = slave->next;
1586 bond->first_slave = NULL; /* slave was the last one */
1595 /*---------------------------------- IOCTL ----------------------------------*/
1597 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1599 dprintk("bond_dev=%p\n", bond_dev);
1600 dprintk("slave_dev=%p\n", slave_dev);
1601 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1602 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1606 #define BOND_INTERSECT_FEATURES \
1607 (NETIF_F_SG|NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM)
1610 * Compute the common dev->feature set available to all slaves. Some
1611 * feature bits are managed elsewhere, so preserve feature bits set on
1612 * master device that are not part of the examined set.
1614 static int bond_compute_features(struct bonding *bond)
1616 unsigned long features = BOND_INTERSECT_FEATURES;
1617 struct slave *slave;
1618 struct net_device *bond_dev = bond->dev;
1621 bond_for_each_slave(bond, slave, i)
1622 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1624 if ((features & NETIF_F_SG) &&
1625 !(features & (NETIF_F_IP_CSUM |
1628 features &= ~NETIF_F_SG;
1630 features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1631 bond_dev->features = features;
1636 /* enslave device <slave> to bond device <master> */
1637 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1639 struct bonding *bond = bond_dev->priv;
1640 struct slave *new_slave = NULL;
1641 struct dev_mc_list *dmi;
1642 struct sockaddr addr;
1644 int old_features = bond_dev->features;
1647 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1648 slave_dev->do_ioctl == NULL) {
1649 printk(KERN_WARNING DRV_NAME
1650 ": %s: Warning: no link monitoring support for %s\n",
1651 bond_dev->name, slave_dev->name);
1654 /* bond must be initialized by bond_open() before enslaving */
1655 if (!(bond_dev->flags & IFF_UP)) {
1656 dprintk("Error, master_dev is not up\n");
1660 /* already enslaved */
1661 if (slave_dev->flags & IFF_SLAVE) {
1662 dprintk("Error, Device was already enslaved\n");
1666 /* vlan challenged mutual exclusion */
1667 /* no need to lock since we're protected by rtnl_lock */
1668 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1669 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1670 if (!list_empty(&bond->vlan_list)) {
1671 printk(KERN_ERR DRV_NAME
1672 ": %s: Error: cannot enslave VLAN "
1673 "challenged slave %s on VLAN enabled "
1674 "bond %s\n", bond_dev->name, slave_dev->name,
1678 printk(KERN_WARNING DRV_NAME
1679 ": %s: Warning: enslaved VLAN challenged "
1680 "slave %s. Adding VLANs will be blocked as "
1681 "long as %s is part of bond %s\n",
1682 bond_dev->name, slave_dev->name, slave_dev->name,
1684 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1687 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1688 if (bond->slave_cnt == 0) {
1689 /* First slave, and it is not VLAN challenged,
1690 * so remove the block of adding VLANs over the bond.
1692 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1697 * Old ifenslave binaries are no longer supported. These can
1698 * be identified with moderate accurary by the state of the slave:
1699 * the current ifenslave will set the interface down prior to
1700 * enslaving it; the old ifenslave will not.
1702 if ((slave_dev->flags & IFF_UP)) {
1703 printk(KERN_ERR DRV_NAME ": %s is up. "
1704 "This may be due to an out of date ifenslave.\n",
1707 goto err_undo_flags;
1710 if (slave_dev->set_mac_address == NULL) {
1711 printk(KERN_ERR DRV_NAME
1712 ": %s: Error: The slave device you specified does "
1713 "not support setting the MAC address. "
1714 "Your kernel likely does not support slave "
1715 "devices.\n", bond_dev->name);
1717 goto err_undo_flags;
1720 new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1723 goto err_undo_flags;
1726 memset(new_slave, 0, sizeof(struct slave));
1728 /* save slave's original flags before calling
1729 * netdev_set_master and dev_open
1731 new_slave->original_flags = slave_dev->flags;
1734 * Save slave's original ("permanent") mac address for modes
1735 * that need it, and for restoring it upon release, and then
1736 * set it to the master's address
1738 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1741 * Set slave to master's mac address. The application already
1742 * set the master's mac address to that of the first slave
1744 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1745 addr.sa_family = slave_dev->type;
1746 res = dev_set_mac_address(slave_dev, &addr);
1748 dprintk("Error %d calling set_mac_address\n", res);
1752 /* open the slave since the application closed it */
1753 res = dev_open(slave_dev);
1755 dprintk("Openning slave %s failed\n", slave_dev->name);
1756 goto err_restore_mac;
1759 res = netdev_set_master(slave_dev, bond_dev);
1761 dprintk("Error %d calling netdev_set_master\n", res);
1765 new_slave->dev = slave_dev;
1767 if ((bond->params.mode == BOND_MODE_TLB) ||
1768 (bond->params.mode == BOND_MODE_ALB)) {
1769 /* bond_alb_init_slave() must be called before all other stages since
1770 * it might fail and we do not want to have to undo everything
1772 res = bond_alb_init_slave(bond, new_slave);
1774 goto err_unset_master;
1778 /* If the mode USES_PRIMARY, then the new slave gets the
1779 * master's promisc (and mc) settings only if it becomes the
1780 * curr_active_slave, and that is taken care of later when calling
1781 * bond_change_active()
1783 if (!USES_PRIMARY(bond->params.mode)) {
1784 /* set promiscuity level to new slave */
1785 if (bond_dev->flags & IFF_PROMISC) {
1786 dev_set_promiscuity(slave_dev, 1);
1789 /* set allmulti level to new slave */
1790 if (bond_dev->flags & IFF_ALLMULTI) {
1791 dev_set_allmulti(slave_dev, 1);
1794 /* upload master's mc_list to new slave */
1795 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1796 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1800 if (bond->params.mode == BOND_MODE_8023AD) {
1801 /* add lacpdu mc addr to mc list */
1802 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1804 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1807 bond_add_vlans_on_slave(bond, slave_dev);
1809 write_lock_bh(&bond->lock);
1811 bond_attach_slave(bond, new_slave);
1813 new_slave->delay = 0;
1814 new_slave->link_failure_count = 0;
1816 bond_compute_features(bond);
1818 if (bond->params.miimon && !bond->params.use_carrier) {
1819 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1821 if ((link_reporting == -1) && !bond->params.arp_interval) {
1823 * miimon is set but a bonded network driver
1824 * does not support ETHTOOL/MII and
1825 * arp_interval is not set. Note: if
1826 * use_carrier is enabled, we will never go
1827 * here (because netif_carrier is always
1828 * supported); thus, we don't need to change
1829 * the messages for netif_carrier.
1831 printk(KERN_WARNING DRV_NAME
1832 ": %s: Warning: MII and ETHTOOL support not "
1833 "available for interface %s, and "
1834 "arp_interval/arp_ip_target module parameters "
1835 "not specified, thus bonding will not detect "
1836 "link failures! see bonding.txt for details.\n",
1837 bond_dev->name, slave_dev->name);
1838 } else if (link_reporting == -1) {
1839 /* unable get link status using mii/ethtool */
1840 printk(KERN_WARNING DRV_NAME
1841 ": %s: Warning: can't get link status from "
1842 "interface %s; the network driver associated "
1843 "with this interface does not support MII or "
1844 "ETHTOOL link status reporting, thus miimon "
1845 "has no effect on this interface.\n",
1846 bond_dev->name, slave_dev->name);
1850 /* check for initial state */
1851 if (!bond->params.miimon ||
1852 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1853 if (bond->params.updelay) {
1854 dprintk("Initial state of slave_dev is "
1855 "BOND_LINK_BACK\n");
1856 new_slave->link = BOND_LINK_BACK;
1857 new_slave->delay = bond->params.updelay;
1859 dprintk("Initial state of slave_dev is "
1861 new_slave->link = BOND_LINK_UP;
1863 new_slave->jiffies = jiffies;
1865 dprintk("Initial state of slave_dev is "
1866 "BOND_LINK_DOWN\n");
1867 new_slave->link = BOND_LINK_DOWN;
1870 if (bond_update_speed_duplex(new_slave) &&
1871 (new_slave->link != BOND_LINK_DOWN)) {
1872 printk(KERN_WARNING DRV_NAME
1873 ": %s: Warning: failed to get speed and duplex from %s, "
1874 "assumed to be 100Mb/sec and Full.\n",
1875 bond_dev->name, new_slave->dev->name);
1877 if (bond->params.mode == BOND_MODE_8023AD) {
1878 printk(KERN_WARNING DRV_NAME
1879 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1880 "support in base driver for proper aggregator "
1881 "selection.\n", bond_dev->name);
1885 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1886 /* if there is a primary slave, remember it */
1887 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1888 bond->primary_slave = new_slave;
1892 switch (bond->params.mode) {
1893 case BOND_MODE_ACTIVEBACKUP:
1894 /* if we're in active-backup mode, we need one and only one active
1895 * interface. The backup interfaces will have their NOARP flag set
1896 * because we need them to be completely deaf and not to respond to
1897 * any ARP request on the network to avoid fooling a switch. Thus,
1898 * since we guarantee that curr_active_slave always point to the last
1899 * usable interface, we just have to verify this interface's flag.
1901 if (((!bond->curr_active_slave) ||
1902 (bond->curr_active_slave->dev->flags & IFF_NOARP)) &&
1903 (new_slave->link != BOND_LINK_DOWN)) {
1904 dprintk("This is the first active slave\n");
1905 /* first slave or no active slave yet, and this link
1906 is OK, so make this interface the active one */
1907 bond_change_active_slave(bond, new_slave);
1909 dprintk("This is just a backup slave\n");
1910 bond_set_slave_inactive_flags(new_slave);
1913 case BOND_MODE_8023AD:
1914 /* in 802.3ad mode, the internal mechanism
1915 * will activate the slaves in the selected
1918 bond_set_slave_inactive_flags(new_slave);
1919 /* if this is the first slave */
1920 if (bond->slave_cnt == 1) {
1921 SLAVE_AD_INFO(new_slave).id = 1;
1922 /* Initialize AD with the number of times that the AD timer is called in 1 second
1923 * can be called only after the mac address of the bond is set
1925 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1926 bond->params.lacp_fast);
1928 SLAVE_AD_INFO(new_slave).id =
1929 SLAVE_AD_INFO(new_slave->prev).id + 1;
1932 bond_3ad_bind_slave(new_slave);
1936 new_slave->state = BOND_STATE_ACTIVE;
1937 if ((!bond->curr_active_slave) &&
1938 (new_slave->link != BOND_LINK_DOWN)) {
1939 /* first slave or no active slave yet, and this link
1940 * is OK, so make this interface the active one
1942 bond_change_active_slave(bond, new_slave);
1946 dprintk("This slave is always active in trunk mode\n");
1948 /* always active in trunk mode */
1949 new_slave->state = BOND_STATE_ACTIVE;
1951 /* In trunking mode there is little meaning to curr_active_slave
1952 * anyway (it holds no special properties of the bond device),
1953 * so we can change it without calling change_active_interface()
1955 if (!bond->curr_active_slave) {
1956 bond->curr_active_slave = new_slave;
1959 } /* switch(bond_mode) */
1961 write_unlock_bh(&bond->lock);
1963 printk(KERN_INFO DRV_NAME
1964 ": %s: enslaving %s as a%s interface with a%s link.\n",
1965 bond_dev->name, slave_dev->name,
1966 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1967 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1969 /* enslave is successful */
1972 /* Undo stages on error */
1974 netdev_set_master(slave_dev, NULL);
1977 dev_close(slave_dev);
1980 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1981 addr.sa_family = slave_dev->type;
1982 dev_set_mac_address(slave_dev, &addr);
1988 bond_dev->features = old_features;
1994 * Try to release the slave device <slave> from the bond device <master>
1995 * It is legal to access curr_active_slave without a lock because all the function
1998 * The rules for slave state should be:
1999 * for Active/Backup:
2000 * Active stays on all backups go down
2001 * for Bonded connections:
2002 * The first up interface should be left on and all others downed.
2004 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2006 struct bonding *bond = bond_dev->priv;
2007 struct slave *slave, *oldcurrent;
2008 struct sockaddr addr;
2009 int mac_addr_differ;
2011 /* slave is not a slave or master is not master of this slave */
2012 if (!(slave_dev->flags & IFF_SLAVE) ||
2013 (slave_dev->master != bond_dev)) {
2014 printk(KERN_ERR DRV_NAME
2015 ": %s: Error: cannot release %s.\n",
2016 bond_dev->name, slave_dev->name);
2020 write_lock_bh(&bond->lock);
2022 slave = bond_get_slave_by_dev(bond, slave_dev);
2024 /* not a slave of this bond */
2025 printk(KERN_INFO DRV_NAME
2026 ": %s: %s not enslaved\n",
2027 bond_dev->name, slave_dev->name);
2031 mac_addr_differ = memcmp(bond_dev->dev_addr,
2034 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
2035 printk(KERN_WARNING DRV_NAME
2036 ": %s: Warning: the permanent HWaddr of %s "
2037 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
2038 "still in use by %s. Set the HWaddr of "
2039 "%s to a different address to avoid "
2043 slave->perm_hwaddr[0],
2044 slave->perm_hwaddr[1],
2045 slave->perm_hwaddr[2],
2046 slave->perm_hwaddr[3],
2047 slave->perm_hwaddr[4],
2048 slave->perm_hwaddr[5],
2053 /* Inform AD package of unbinding of slave. */
2054 if (bond->params.mode == BOND_MODE_8023AD) {
2055 /* must be called before the slave is
2056 * detached from the list
2058 bond_3ad_unbind_slave(slave);
2061 printk(KERN_INFO DRV_NAME
2062 ": %s: releasing %s interface %s\n",
2064 (slave->state == BOND_STATE_ACTIVE)
2065 ? "active" : "backup",
2068 oldcurrent = bond->curr_active_slave;
2070 bond->current_arp_slave = NULL;
2072 /* release the slave from its bond */
2073 bond_detach_slave(bond, slave);
2075 bond_compute_features(bond);
2077 if (bond->primary_slave == slave) {
2078 bond->primary_slave = NULL;
2081 if (oldcurrent == slave) {
2082 bond_change_active_slave(bond, NULL);
2085 if ((bond->params.mode == BOND_MODE_TLB) ||
2086 (bond->params.mode == BOND_MODE_ALB)) {
2087 /* Must be called only after the slave has been
2088 * detached from the list and the curr_active_slave
2089 * has been cleared (if our_slave == old_current),
2090 * but before a new active slave is selected.
2092 bond_alb_deinit_slave(bond, slave);
2095 if (oldcurrent == slave) {
2096 bond_select_active_slave(bond);
2098 if (!bond->curr_active_slave) {
2099 printk(KERN_INFO DRV_NAME
2100 ": %s: now running without any active "
2106 if (bond->slave_cnt == 0) {
2107 /* if the last slave was removed, zero the mac address
2108 * of the master so it will be set by the application
2109 * to the mac address of the first slave
2111 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2113 if (list_empty(&bond->vlan_list)) {
2114 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2116 printk(KERN_WARNING DRV_NAME
2117 ": %s: Warning: clearing HW address of %s while it "
2118 "still has VLANs.\n",
2119 bond_dev->name, bond_dev->name);
2120 printk(KERN_WARNING DRV_NAME
2121 ": %s: When re-adding slaves, make sure the bond's "
2122 "HW address matches its VLANs'.\n",
2125 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2126 !bond_has_challenged_slaves(bond)) {
2127 printk(KERN_INFO DRV_NAME
2128 ": %s: last VLAN challenged slave %s "
2129 "left bond %s. VLAN blocking is removed\n",
2130 bond_dev->name, slave_dev->name, bond_dev->name);
2131 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2134 write_unlock_bh(&bond->lock);
2136 bond_del_vlans_from_slave(bond, slave_dev);
2138 /* If the mode USES_PRIMARY, then we should only remove its
2139 * promisc and mc settings if it was the curr_active_slave, but that was
2140 * already taken care of above when we detached the slave
2142 if (!USES_PRIMARY(bond->params.mode)) {
2143 /* unset promiscuity level from slave */
2144 if (bond_dev->flags & IFF_PROMISC) {
2145 dev_set_promiscuity(slave_dev, -1);
2148 /* unset allmulti level from slave */
2149 if (bond_dev->flags & IFF_ALLMULTI) {
2150 dev_set_allmulti(slave_dev, -1);
2153 /* flush master's mc_list from slave */
2154 bond_mc_list_flush(bond_dev, slave_dev);
2157 netdev_set_master(slave_dev, NULL);
2159 /* close slave before restoring its mac address */
2160 dev_close(slave_dev);
2162 /* restore original ("permanent") mac address */
2163 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2164 addr.sa_family = slave_dev->type;
2165 dev_set_mac_address(slave_dev, &addr);
2167 /* restore the original state of the
2168 * IFF_NOARP flag that might have been
2169 * set by bond_set_slave_inactive_flags()
2171 if ((slave->original_flags & IFF_NOARP) == 0) {
2172 slave_dev->flags &= ~IFF_NOARP;
2177 return 0; /* deletion OK */
2181 * This function releases all slaves.
2183 static int bond_release_all(struct net_device *bond_dev)
2185 struct bonding *bond = bond_dev->priv;
2186 struct slave *slave;
2187 struct net_device *slave_dev;
2188 struct sockaddr addr;
2190 write_lock_bh(&bond->lock);
2192 if (bond->slave_cnt == 0) {
2196 bond->current_arp_slave = NULL;
2197 bond->primary_slave = NULL;
2198 bond_change_active_slave(bond, NULL);
2200 while ((slave = bond->first_slave) != NULL) {
2201 /* Inform AD package of unbinding of slave
2202 * before slave is detached from the list.
2204 if (bond->params.mode == BOND_MODE_8023AD) {
2205 bond_3ad_unbind_slave(slave);
2208 slave_dev = slave->dev;
2209 bond_detach_slave(bond, slave);
2211 if ((bond->params.mode == BOND_MODE_TLB) ||
2212 (bond->params.mode == BOND_MODE_ALB)) {
2213 /* must be called only after the slave
2214 * has been detached from the list
2216 bond_alb_deinit_slave(bond, slave);
2219 bond_compute_features(bond);
2221 /* now that the slave is detached, unlock and perform
2222 * all the undo steps that should not be called from
2225 write_unlock_bh(&bond->lock);
2227 bond_del_vlans_from_slave(bond, slave_dev);
2229 /* If the mode USES_PRIMARY, then we should only remove its
2230 * promisc and mc settings if it was the curr_active_slave, but that was
2231 * already taken care of above when we detached the slave
2233 if (!USES_PRIMARY(bond->params.mode)) {
2234 /* unset promiscuity level from slave */
2235 if (bond_dev->flags & IFF_PROMISC) {
2236 dev_set_promiscuity(slave_dev, -1);
2239 /* unset allmulti level from slave */
2240 if (bond_dev->flags & IFF_ALLMULTI) {
2241 dev_set_allmulti(slave_dev, -1);
2244 /* flush master's mc_list from slave */
2245 bond_mc_list_flush(bond_dev, slave_dev);
2248 netdev_set_master(slave_dev, NULL);
2250 /* close slave before restoring its mac address */
2251 dev_close(slave_dev);
2253 /* restore original ("permanent") mac address*/
2254 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2255 addr.sa_family = slave_dev->type;
2256 dev_set_mac_address(slave_dev, &addr);
2258 /* restore the original state of the IFF_NOARP flag that might have
2259 * been set by bond_set_slave_inactive_flags()
2261 if ((slave->original_flags & IFF_NOARP) == 0) {
2262 slave_dev->flags &= ~IFF_NOARP;
2267 /* re-acquire the lock before getting the next slave */
2268 write_lock_bh(&bond->lock);
2271 /* zero the mac address of the master so it will be
2272 * set by the application to the mac address of the
2275 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2277 if (list_empty(&bond->vlan_list)) {
2278 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2280 printk(KERN_WARNING DRV_NAME
2281 ": %s: Warning: clearing HW address of %s while it "
2282 "still has VLANs.\n",
2283 bond_dev->name, bond_dev->name);
2284 printk(KERN_WARNING DRV_NAME
2285 ": %s: When re-adding slaves, make sure the bond's "
2286 "HW address matches its VLANs'.\n",
2290 printk(KERN_INFO DRV_NAME
2291 ": %s: released all slaves\n",
2295 write_unlock_bh(&bond->lock);
2301 * This function changes the active slave to slave <slave_dev>.
2302 * It returns -EINVAL in the following cases.
2303 * - <slave_dev> is not found in the list.
2304 * - There is not active slave now.
2305 * - <slave_dev> is already active.
2306 * - The link state of <slave_dev> is not BOND_LINK_UP.
2307 * - <slave_dev> is not running.
2308 * In these cases, this fuction does nothing.
2309 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2311 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2313 struct bonding *bond = bond_dev->priv;
2314 struct slave *old_active = NULL;
2315 struct slave *new_active = NULL;
2318 if (!USES_PRIMARY(bond->params.mode)) {
2322 /* Verify that master_dev is indeed the master of slave_dev */
2323 if (!(slave_dev->flags & IFF_SLAVE) ||
2324 (slave_dev->master != bond_dev)) {
2328 write_lock_bh(&bond->lock);
2330 old_active = bond->curr_active_slave;
2331 new_active = bond_get_slave_by_dev(bond, slave_dev);
2334 * Changing to the current active: do nothing; return success.
2336 if (new_active && (new_active == old_active)) {
2337 write_unlock_bh(&bond->lock);
2343 (new_active->link == BOND_LINK_UP) &&
2344 IS_UP(new_active->dev)) {
2345 bond_change_active_slave(bond, new_active);
2350 write_unlock_bh(&bond->lock);
2355 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2357 struct bonding *bond = bond_dev->priv;
2359 info->bond_mode = bond->params.mode;
2360 info->miimon = bond->params.miimon;
2362 read_lock_bh(&bond->lock);
2363 info->num_slaves = bond->slave_cnt;
2364 read_unlock_bh(&bond->lock);
2369 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2371 struct bonding *bond = bond_dev->priv;
2372 struct slave *slave;
2375 if (info->slave_id < 0) {
2379 read_lock_bh(&bond->lock);
2381 bond_for_each_slave(bond, slave, i) {
2382 if (i == (int)info->slave_id) {
2388 read_unlock_bh(&bond->lock);
2391 strcpy(info->slave_name, slave->dev->name);
2392 info->link = slave->link;
2393 info->state = slave->state;
2394 info->link_failure_count = slave->link_failure_count;
2402 /*-------------------------------- Monitoring -------------------------------*/
2404 /* this function is called regularly to monitor each slave's link. */
2405 void bond_mii_monitor(struct net_device *bond_dev)
2407 struct bonding *bond = bond_dev->priv;
2408 struct slave *slave, *oldcurrent;
2409 int do_failover = 0;
2413 read_lock(&bond->lock);
2415 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2417 if (bond->kill_timers) {
2421 if (bond->slave_cnt == 0) {
2425 /* we will try to read the link status of each of our slaves, and
2426 * set their IFF_RUNNING flag appropriately. For each slave not
2427 * supporting MII status, we won't do anything so that a user-space
2428 * program could monitor the link itself if needed.
2431 read_lock(&bond->curr_slave_lock);
2432 oldcurrent = bond->curr_active_slave;
2433 read_unlock(&bond->curr_slave_lock);
2435 bond_for_each_slave(bond, slave, i) {
2436 struct net_device *slave_dev = slave->dev;
2438 u16 old_speed = slave->speed;
2439 u8 old_duplex = slave->duplex;
2441 link_state = bond_check_dev_link(bond, slave_dev, 0);
2443 switch (slave->link) {
2444 case BOND_LINK_UP: /* the link was up */
2445 if (link_state == BMSR_LSTATUS) {
2446 /* link stays up, nothing more to do */
2448 } else { /* link going down */
2449 slave->link = BOND_LINK_FAIL;
2450 slave->delay = bond->params.downdelay;
2452 if (slave->link_failure_count < UINT_MAX) {
2453 slave->link_failure_count++;
2456 if (bond->params.downdelay) {
2457 printk(KERN_INFO DRV_NAME
2458 ": %s: link status down for %s "
2459 "interface %s, disabling it in "
2463 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2464 ? ((slave == oldcurrent)
2465 ? "active " : "backup ")
2469 bond->params.downdelay * bond->params.miimon);
2472 /* no break ! fall through the BOND_LINK_FAIL test to
2473 ensure proper action to be taken
2475 case BOND_LINK_FAIL: /* the link has just gone down */
2476 if (link_state != BMSR_LSTATUS) {
2477 /* link stays down */
2478 if (slave->delay <= 0) {
2479 /* link down for too long time */
2480 slave->link = BOND_LINK_DOWN;
2482 /* in active/backup mode, we must
2483 * completely disable this interface
2485 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2486 (bond->params.mode == BOND_MODE_8023AD)) {
2487 bond_set_slave_inactive_flags(slave);
2490 printk(KERN_INFO DRV_NAME
2491 ": %s: link status definitely "
2492 "down for interface %s, "
2497 /* notify ad that the link status has changed */
2498 if (bond->params.mode == BOND_MODE_8023AD) {
2499 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2502 if ((bond->params.mode == BOND_MODE_TLB) ||
2503 (bond->params.mode == BOND_MODE_ALB)) {
2504 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2507 if (slave == oldcurrent) {
2515 slave->link = BOND_LINK_UP;
2516 slave->jiffies = jiffies;
2517 printk(KERN_INFO DRV_NAME
2518 ": %s: link status up again after %d "
2519 "ms for interface %s.\n",
2521 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2525 case BOND_LINK_DOWN: /* the link was down */
2526 if (link_state != BMSR_LSTATUS) {
2527 /* the link stays down, nothing more to do */
2529 } else { /* link going up */
2530 slave->link = BOND_LINK_BACK;
2531 slave->delay = bond->params.updelay;
2533 if (bond->params.updelay) {
2534 /* if updelay == 0, no need to
2535 advertise about a 0 ms delay */
2536 printk(KERN_INFO DRV_NAME
2537 ": %s: link status up for "
2538 "interface %s, enabling it "
2542 bond->params.updelay * bond->params.miimon);
2545 /* no break ! fall through the BOND_LINK_BACK state in
2546 case there's something to do.
2548 case BOND_LINK_BACK: /* the link has just come back */
2549 if (link_state != BMSR_LSTATUS) {
2550 /* link down again */
2551 slave->link = BOND_LINK_DOWN;
2553 printk(KERN_INFO DRV_NAME
2554 ": %s: link status down again after %d "
2555 "ms for interface %s.\n",
2557 (bond->params.updelay - slave->delay) * bond->params.miimon,
2561 if (slave->delay == 0) {
2562 /* now the link has been up for long time enough */
2563 slave->link = BOND_LINK_UP;
2564 slave->jiffies = jiffies;
2566 if (bond->params.mode == BOND_MODE_8023AD) {
2567 /* prevent it from being the active one */
2568 slave->state = BOND_STATE_BACKUP;
2569 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2570 /* make it immediately active */
2571 slave->state = BOND_STATE_ACTIVE;
2572 } else if (slave != bond->primary_slave) {
2573 /* prevent it from being the active one */
2574 slave->state = BOND_STATE_BACKUP;
2577 printk(KERN_INFO DRV_NAME
2578 ": %s: link status definitely "
2579 "up for interface %s.\n",
2583 /* notify ad that the link status has changed */
2584 if (bond->params.mode == BOND_MODE_8023AD) {
2585 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2588 if ((bond->params.mode == BOND_MODE_TLB) ||
2589 (bond->params.mode == BOND_MODE_ALB)) {
2590 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2593 if ((!oldcurrent) ||
2594 (slave == bond->primary_slave)) {
2603 /* Should not happen */
2604 printk(KERN_ERR DRV_NAME
2605 ": %s: Error: %s Illegal value (link=%d)\n",
2610 } /* end of switch (slave->link) */
2612 bond_update_speed_duplex(slave);
2614 if (bond->params.mode == BOND_MODE_8023AD) {
2615 if (old_speed != slave->speed) {
2616 bond_3ad_adapter_speed_changed(slave);
2619 if (old_duplex != slave->duplex) {
2620 bond_3ad_adapter_duplex_changed(slave);
2627 write_lock(&bond->curr_slave_lock);
2629 bond_select_active_slave(bond);
2631 if (oldcurrent && !bond->curr_active_slave) {
2632 printk(KERN_INFO DRV_NAME
2633 ": %s: now running without any active "
2638 write_unlock(&bond->curr_slave_lock);
2642 if (bond->params.miimon) {
2643 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2646 read_unlock(&bond->lock);
2650 static u32 bond_glean_dev_ip(struct net_device *dev)
2652 struct in_device *idev;
2653 struct in_ifaddr *ifa;
2660 idev = __in_dev_get_rcu(dev);
2664 ifa = idev->ifa_list;
2668 addr = ifa->ifa_local;
2674 static int bond_has_ip(struct bonding *bond)
2676 struct vlan_entry *vlan, *vlan_next;
2678 if (bond->master_ip)
2681 if (list_empty(&bond->vlan_list))
2684 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2694 * We go to the (large) trouble of VLAN tagging ARP frames because
2695 * switches in VLAN mode (especially if ports are configured as
2696 * "native" to a VLAN) might not pass non-tagged frames.
2698 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2700 struct sk_buff *skb;
2702 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2703 slave_dev->name, dest_ip, src_ip, vlan_id);
2705 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2706 NULL, slave_dev->dev_addr, NULL);
2709 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2713 skb = vlan_put_tag(skb, vlan_id);
2715 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2723 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2726 u32 *targets = bond->params.arp_targets;
2727 struct vlan_entry *vlan, *vlan_next;
2728 struct net_device *vlan_dev;
2732 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2733 dprintk("basa: target %x\n", targets[i]);
2734 if (list_empty(&bond->vlan_list)) {
2735 dprintk("basa: empty vlan: arp_send\n");
2736 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2737 bond->master_ip, 0);
2742 * If VLANs are configured, we do a route lookup to
2743 * determine which VLAN interface would be used, so we
2744 * can tag the ARP with the proper VLAN tag.
2746 memset(&fl, 0, sizeof(fl));
2747 fl.fl4_dst = targets[i];
2748 fl.fl4_tos = RTO_ONLINK;
2750 rv = ip_route_output_key(&rt, &fl);
2752 if (net_ratelimit()) {
2753 printk(KERN_WARNING DRV_NAME
2754 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2755 bond->dev->name, NIPQUAD(fl.fl4_dst));
2761 * This target is not on a VLAN
2763 if (rt->u.dst.dev == bond->dev) {
2765 dprintk("basa: rtdev == bond->dev: arp_send\n");
2766 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2767 bond->master_ip, 0);
2772 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2774 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2775 if (vlan_dev == rt->u.dst.dev) {
2776 vlan_id = vlan->vlan_id;
2777 dprintk("basa: vlan match on %s %d\n",
2778 vlan_dev->name, vlan_id);
2785 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2786 vlan->vlan_ip, vlan_id);
2790 if (net_ratelimit()) {
2791 printk(KERN_WARNING DRV_NAME
2792 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2793 bond->dev->name, NIPQUAD(fl.fl4_dst),
2794 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2801 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2802 * for each VLAN above us.
2804 static void bond_send_gratuitous_arp(struct bonding *bond)
2806 struct slave *slave = bond->curr_active_slave;
2807 struct vlan_entry *vlan;
2808 struct net_device *vlan_dev;
2810 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2811 slave ? slave->dev->name : "NULL");
2815 if (bond->master_ip) {
2816 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2817 bond->master_ip, 0);
2820 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2821 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2822 if (vlan->vlan_ip) {
2823 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2824 vlan->vlan_ip, vlan->vlan_id);
2830 * this function is called regularly to monitor each slave's link
2831 * ensuring that traffic is being sent and received when arp monitoring
2832 * is used in load-balancing mode. if the adapter has been dormant, then an
2833 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2834 * arp monitoring in active backup mode.
2836 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2838 struct bonding *bond = bond_dev->priv;
2839 struct slave *slave, *oldcurrent;
2840 int do_failover = 0;
2844 read_lock(&bond->lock);
2846 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2848 if (bond->kill_timers) {
2852 if (bond->slave_cnt == 0) {
2856 read_lock(&bond->curr_slave_lock);
2857 oldcurrent = bond->curr_active_slave;
2858 read_unlock(&bond->curr_slave_lock);
2860 /* see if any of the previous devices are up now (i.e. they have
2861 * xmt and rcv traffic). the curr_active_slave does not come into
2862 * the picture unless it is null. also, slave->jiffies is not needed
2863 * here because we send an arp on each slave and give a slave as
2864 * long as it needs to get the tx/rx within the delta.
2865 * TODO: what about up/down delay in arp mode? it wasn't here before
2868 bond_for_each_slave(bond, slave, i) {
2869 if (slave->link != BOND_LINK_UP) {
2870 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2871 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2873 slave->link = BOND_LINK_UP;
2874 slave->state = BOND_STATE_ACTIVE;
2876 /* primary_slave has no meaning in round-robin
2877 * mode. the window of a slave being up and
2878 * curr_active_slave being null after enslaving
2882 printk(KERN_INFO DRV_NAME
2883 ": %s: link status definitely "
2884 "up for interface %s, ",
2889 printk(KERN_INFO DRV_NAME
2890 ": %s: interface %s is now up\n",
2896 /* slave->link == BOND_LINK_UP */
2898 /* not all switches will respond to an arp request
2899 * when the source ip is 0, so don't take the link down
2900 * if we don't know our ip yet
2902 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2903 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2904 bond_has_ip(bond))) {
2906 slave->link = BOND_LINK_DOWN;
2907 slave->state = BOND_STATE_BACKUP;
2909 if (slave->link_failure_count < UINT_MAX) {
2910 slave->link_failure_count++;
2913 printk(KERN_INFO DRV_NAME
2914 ": %s: interface %s is now down.\n",
2918 if (slave == oldcurrent) {
2924 /* note: if switch is in round-robin mode, all links
2925 * must tx arp to ensure all links rx an arp - otherwise
2926 * links may oscillate or not come up at all; if switch is
2927 * in something like xor mode, there is nothing we can
2928 * do - all replies will be rx'ed on same link causing slaves
2929 * to be unstable during low/no traffic periods
2931 if (IS_UP(slave->dev)) {
2932 bond_arp_send_all(bond, slave);
2937 write_lock(&bond->curr_slave_lock);
2939 bond_select_active_slave(bond);
2941 if (oldcurrent && !bond->curr_active_slave) {
2942 printk(KERN_INFO DRV_NAME
2943 ": %s: now running without any active "
2948 write_unlock(&bond->curr_slave_lock);
2952 if (bond->params.arp_interval) {
2953 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2956 read_unlock(&bond->lock);
2960 * When using arp monitoring in active-backup mode, this function is
2961 * called to determine if any backup slaves have went down or a new
2962 * current slave needs to be found.
2963 * The backup slaves never generate traffic, they are considered up by merely
2964 * receiving traffic. If the current slave goes down, each backup slave will
2965 * be given the opportunity to tx/rx an arp before being taken down - this
2966 * prevents all slaves from being taken down due to the current slave not
2967 * sending any traffic for the backups to receive. The arps are not necessarily
2968 * necessary, any tx and rx traffic will keep the current slave up. While any
2969 * rx traffic will keep the backup slaves up, the current slave is responsible
2970 * for generating traffic to keep them up regardless of any other traffic they
2971 * may have received.
2972 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2974 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2976 struct bonding *bond = bond_dev->priv;
2977 struct slave *slave;
2981 read_lock(&bond->lock);
2983 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2985 if (bond->kill_timers) {
2989 if (bond->slave_cnt == 0) {
2993 /* determine if any slave has come up or any backup slave has
2995 * TODO: what about up/down delay in arp mode? it wasn't here before
2998 bond_for_each_slave(bond, slave, i) {
2999 if (slave->link != BOND_LINK_UP) {
3000 if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
3002 slave->link = BOND_LINK_UP;
3004 write_lock(&bond->curr_slave_lock);
3006 if ((!bond->curr_active_slave) &&
3007 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
3008 bond_change_active_slave(bond, slave);
3009 bond->current_arp_slave = NULL;
3010 } else if (bond->curr_active_slave != slave) {
3011 /* this slave has just come up but we
3012 * already have a current slave; this
3013 * can also happen if bond_enslave adds
3014 * a new slave that is up while we are
3015 * searching for a new slave
3017 bond_set_slave_inactive_flags(slave);
3018 bond->current_arp_slave = NULL;
3021 if (slave == bond->curr_active_slave) {
3022 printk(KERN_INFO DRV_NAME
3023 ": %s: %s is up and now the "
3024 "active interface\n",
3028 printk(KERN_INFO DRV_NAME
3029 ": %s: backup interface %s is "
3035 write_unlock(&bond->curr_slave_lock);
3038 read_lock(&bond->curr_slave_lock);
3040 if ((slave != bond->curr_active_slave) &&
3041 (!bond->current_arp_slave) &&
3042 (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
3043 bond_has_ip(bond))) {
3044 /* a backup slave has gone down; three times
3045 * the delta allows the current slave to be
3046 * taken out before the backup slave.
3047 * note: a non-null current_arp_slave indicates
3048 * the curr_active_slave went down and we are
3049 * searching for a new one; under this
3050 * condition we only take the curr_active_slave
3051 * down - this gives each slave a chance to
3052 * tx/rx traffic before being taken out
3055 read_unlock(&bond->curr_slave_lock);
3057 slave->link = BOND_LINK_DOWN;
3059 if (slave->link_failure_count < UINT_MAX) {
3060 slave->link_failure_count++;
3063 bond_set_slave_inactive_flags(slave);
3065 printk(KERN_INFO DRV_NAME
3066 ": %s: backup interface %s is now down\n",
3070 read_unlock(&bond->curr_slave_lock);
3075 read_lock(&bond->curr_slave_lock);
3076 slave = bond->curr_active_slave;
3077 read_unlock(&bond->curr_slave_lock);
3080 /* if we have sent traffic in the past 2*arp_intervals but
3081 * haven't xmit and rx traffic in that time interval, select
3082 * a different slave. slave->jiffies is only updated when
3083 * a slave first becomes the curr_active_slave - not necessarily
3084 * after every arp; this ensures the slave has a full 2*delta
3085 * before being taken out. if a primary is being used, check
3086 * if it is up and needs to take over as the curr_active_slave
3088 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
3089 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
3090 bond_has_ip(bond))) &&
3091 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
3093 slave->link = BOND_LINK_DOWN;
3095 if (slave->link_failure_count < UINT_MAX) {
3096 slave->link_failure_count++;
3099 printk(KERN_INFO DRV_NAME
3100 ": %s: link status down for active interface "
3101 "%s, disabling it\n",
3105 write_lock(&bond->curr_slave_lock);
3107 bond_select_active_slave(bond);
3108 slave = bond->curr_active_slave;
3110 write_unlock(&bond->curr_slave_lock);
3112 bond->current_arp_slave = slave;
3115 slave->jiffies = jiffies;
3117 } else if ((bond->primary_slave) &&
3118 (bond->primary_slave != slave) &&
3119 (bond->primary_slave->link == BOND_LINK_UP)) {
3120 /* at this point, slave is the curr_active_slave */
3121 printk(KERN_INFO DRV_NAME
3122 ": %s: changing from interface %s to primary "
3126 bond->primary_slave->dev->name);
3128 /* primary is up so switch to it */
3129 write_lock(&bond->curr_slave_lock);
3130 bond_change_active_slave(bond, bond->primary_slave);
3131 write_unlock(&bond->curr_slave_lock);
3133 slave = bond->primary_slave;
3134 slave->jiffies = jiffies;
3136 bond->current_arp_slave = NULL;
3139 /* the current slave must tx an arp to ensure backup slaves
3142 if (slave && bond_has_ip(bond)) {
3143 bond_arp_send_all(bond, slave);
3147 /* if we don't have a curr_active_slave, search for the next available
3148 * backup slave from the current_arp_slave and make it the candidate
3149 * for becoming the curr_active_slave
3152 if (!bond->current_arp_slave) {
3153 bond->current_arp_slave = bond->first_slave;
3156 if (bond->current_arp_slave) {
3157 bond_set_slave_inactive_flags(bond->current_arp_slave);
3159 /* search for next candidate */
3160 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3161 if (IS_UP(slave->dev)) {
3162 slave->link = BOND_LINK_BACK;
3163 bond_set_slave_active_flags(slave);
3164 bond_arp_send_all(bond, slave);
3165 slave->jiffies = jiffies;
3166 bond->current_arp_slave = slave;
3170 /* if the link state is up at this point, we
3171 * mark it down - this can happen if we have
3172 * simultaneous link failures and
3173 * reselect_active_interface doesn't make this
3174 * one the current slave so it is still marked
3175 * up when it is actually down
3177 if (slave->link == BOND_LINK_UP) {
3178 slave->link = BOND_LINK_DOWN;
3179 if (slave->link_failure_count < UINT_MAX) {
3180 slave->link_failure_count++;
3183 bond_set_slave_inactive_flags(slave);
3185 printk(KERN_INFO DRV_NAME
3186 ": %s: backup interface %s is "
3196 if (bond->params.arp_interval) {
3197 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
3200 read_unlock(&bond->lock);
3203 /*------------------------------ proc/seq_file-------------------------------*/
3205 #ifdef CONFIG_PROC_FS
3207 #define SEQ_START_TOKEN ((void *)1)
3209 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3211 struct bonding *bond = seq->private;
3213 struct slave *slave;
3216 /* make sure the bond won't be taken away */
3217 read_lock(&dev_base_lock);
3218 read_lock_bh(&bond->lock);
3221 return SEQ_START_TOKEN;
3224 bond_for_each_slave(bond, slave, i) {
3225 if (++off == *pos) {
3233 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3235 struct bonding *bond = seq->private;
3236 struct slave *slave = v;
3239 if (v == SEQ_START_TOKEN) {
3240 return bond->first_slave;
3243 slave = slave->next;
3245 return (slave == bond->first_slave) ? NULL : slave;
3248 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3250 struct bonding *bond = seq->private;
3252 read_unlock_bh(&bond->lock);
3253 read_unlock(&dev_base_lock);
3256 static void bond_info_show_master(struct seq_file *seq)
3258 struct bonding *bond = seq->private;
3261 read_lock(&bond->curr_slave_lock);
3262 curr = bond->curr_active_slave;
3263 read_unlock(&bond->curr_slave_lock);
3265 seq_printf(seq, "Bonding Mode: %s\n",
3266 bond_mode_name(bond->params.mode));
3268 if (bond->params.mode == BOND_MODE_XOR ||
3269 bond->params.mode == BOND_MODE_8023AD) {
3270 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3271 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3272 bond->params.xmit_policy);
3275 if (USES_PRIMARY(bond->params.mode)) {
3276 seq_printf(seq, "Primary Slave: %s\n",
3277 (bond->primary_slave) ?
3278 bond->primary_slave->dev->name : "None");
3280 seq_printf(seq, "Currently Active Slave: %s\n",
3281 (curr) ? curr->dev->name : "None");
3284 seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down");
3285 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3286 seq_printf(seq, "Up Delay (ms): %d\n",
3287 bond->params.updelay * bond->params.miimon);
3288 seq_printf(seq, "Down Delay (ms): %d\n",
3289 bond->params.downdelay * bond->params.miimon);
3291 if (bond->params.mode == BOND_MODE_8023AD) {
3292 struct ad_info ad_info;
3294 seq_puts(seq, "\n802.3ad info\n");
3295 seq_printf(seq, "LACP rate: %s\n",
3296 (bond->params.lacp_fast) ? "fast" : "slow");
3298 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3299 seq_printf(seq, "bond %s has no active aggregator\n",
3302 seq_printf(seq, "Active Aggregator Info:\n");
3304 seq_printf(seq, "\tAggregator ID: %d\n",
3305 ad_info.aggregator_id);
3306 seq_printf(seq, "\tNumber of ports: %d\n",
3308 seq_printf(seq, "\tActor Key: %d\n",
3310 seq_printf(seq, "\tPartner Key: %d\n",
3311 ad_info.partner_key);
3312 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3313 ad_info.partner_system[0],
3314 ad_info.partner_system[1],
3315 ad_info.partner_system[2],
3316 ad_info.partner_system[3],
3317 ad_info.partner_system[4],
3318 ad_info.partner_system[5]);
3323 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3325 struct bonding *bond = seq->private;
3327 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3328 seq_printf(seq, "MII Status: %s\n",
3329 (slave->link == BOND_LINK_UP) ? "up" : "down");
3330 seq_printf(seq, "Link Failure Count: %d\n",
3331 slave->link_failure_count);
3334 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3335 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3336 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3337 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
3339 if (bond->params.mode == BOND_MODE_8023AD) {
3340 const struct aggregator *agg
3341 = SLAVE_AD_INFO(slave).port.aggregator;
3344 seq_printf(seq, "Aggregator ID: %d\n",
3345 agg->aggregator_identifier);
3347 seq_puts(seq, "Aggregator ID: N/A\n");
3352 static int bond_info_seq_show(struct seq_file *seq, void *v)
3354 if (v == SEQ_START_TOKEN) {
3355 seq_printf(seq, "%s\n", version);
3356 bond_info_show_master(seq);
3358 bond_info_show_slave(seq, v);
3364 static struct seq_operations bond_info_seq_ops = {
3365 .start = bond_info_seq_start,
3366 .next = bond_info_seq_next,
3367 .stop = bond_info_seq_stop,
3368 .show = bond_info_seq_show,
3371 static int bond_info_open(struct inode *inode, struct file *file)
3373 struct seq_file *seq;
3374 struct proc_dir_entry *proc;
3377 res = seq_open(file, &bond_info_seq_ops);
3379 /* recover the pointer buried in proc_dir_entry data */
3380 seq = file->private_data;
3382 seq->private = proc->data;
3388 static struct file_operations bond_info_fops = {
3389 .owner = THIS_MODULE,
3390 .open = bond_info_open,
3392 .llseek = seq_lseek,
3393 .release = seq_release,
3396 static int bond_create_proc_entry(struct bonding *bond)
3398 struct net_device *bond_dev = bond->dev;
3400 if (bond_proc_dir) {
3401 bond->proc_entry = create_proc_entry(bond_dev->name,
3404 if (bond->proc_entry == NULL) {
3405 printk(KERN_WARNING DRV_NAME
3406 ": Warning: Cannot create /proc/net/%s/%s\n",
3407 DRV_NAME, bond_dev->name);
3409 bond->proc_entry->data = bond;
3410 bond->proc_entry->proc_fops = &bond_info_fops;
3411 bond->proc_entry->owner = THIS_MODULE;
3412 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3419 static void bond_remove_proc_entry(struct bonding *bond)
3421 if (bond_proc_dir && bond->proc_entry) {
3422 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3423 memset(bond->proc_file_name, 0, IFNAMSIZ);
3424 bond->proc_entry = NULL;
3428 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3429 * Caller must hold rtnl_lock.
3431 static void bond_create_proc_dir(void)
3433 int len = strlen(DRV_NAME);
3435 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3436 bond_proc_dir = bond_proc_dir->next) {
3437 if ((bond_proc_dir->namelen == len) &&
3438 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3443 if (!bond_proc_dir) {
3444 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3445 if (bond_proc_dir) {
3446 bond_proc_dir->owner = THIS_MODULE;
3448 printk(KERN_WARNING DRV_NAME
3449 ": Warning: cannot create /proc/net/%s\n",
3455 /* Destroy the bonding directory under /proc/net, if empty.
3456 * Caller must hold rtnl_lock.
3458 static void bond_destroy_proc_dir(void)
3460 struct proc_dir_entry *de;
3462 if (!bond_proc_dir) {
3466 /* verify that the /proc dir is empty */
3467 for (de = bond_proc_dir->subdir; de; de = de->next) {
3468 /* ignore . and .. */
3469 if (*(de->name) != '.') {
3475 if (bond_proc_dir->owner == THIS_MODULE) {
3476 bond_proc_dir->owner = NULL;
3479 remove_proc_entry(DRV_NAME, proc_net);
3480 bond_proc_dir = NULL;
3483 #endif /* CONFIG_PROC_FS */
3485 /*-------------------------- netdev event handling --------------------------*/
3488 * Change device name
3490 static int bond_event_changename(struct bonding *bond)
3492 #ifdef CONFIG_PROC_FS
3493 bond_remove_proc_entry(bond);
3494 bond_create_proc_entry(bond);
3500 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3502 struct bonding *event_bond = bond_dev->priv;
3505 case NETDEV_CHANGENAME:
3506 return bond_event_changename(event_bond);
3507 case NETDEV_UNREGISTER:
3509 * TODO: remove a bond from the list?
3519 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3521 struct net_device *bond_dev = slave_dev->master;
3522 struct bonding *bond = bond_dev->priv;
3525 case NETDEV_UNREGISTER:
3527 bond_release(bond_dev, slave_dev);
3532 * TODO: is this what we get if somebody
3533 * sets up a hierarchical bond, then rmmod's
3534 * one of the slave bonding devices?
3539 * ... Or is it this?
3542 case NETDEV_CHANGEMTU:
3544 * TODO: Should slaves be allowed to
3545 * independently alter their MTU? For
3546 * an active-backup bond, slaves need
3547 * not be the same type of device, so
3548 * MTUs may vary. For other modes,
3549 * slaves arguably should have the
3550 * same MTUs. To do this, we'd need to
3551 * take over the slave's change_mtu
3552 * function for the duration of their
3556 case NETDEV_CHANGENAME:
3558 * TODO: handle changing the primary's name
3561 case NETDEV_FEAT_CHANGE:
3562 bond_compute_features(bond);
3572 * bond_netdev_event: handle netdev notifier chain events.
3574 * This function receives events for the netdev chain. The caller (an
3575 * ioctl handler calling notifier_call_chain) holds the necessary
3576 * locks for us to safely manipulate the slave devices (RTNL lock,
3579 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3581 struct net_device *event_dev = (struct net_device *)ptr;
3583 dprintk("event_dev: %s, event: %lx\n",
3584 (event_dev ? event_dev->name : "None"),
3587 if (event_dev->flags & IFF_MASTER) {
3588 dprintk("IFF_MASTER\n");
3589 return bond_master_netdev_event(event, event_dev);
3592 if (event_dev->flags & IFF_SLAVE) {
3593 dprintk("IFF_SLAVE\n");
3594 return bond_slave_netdev_event(event, event_dev);
3601 * bond_inetaddr_event: handle inetaddr notifier chain events.
3603 * We keep track of device IPs primarily to use as source addresses in
3604 * ARP monitor probes (rather than spewing out broadcasts all the time).
3606 * We track one IP for the main device (if it has one), plus one per VLAN.
3608 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3610 struct in_ifaddr *ifa = ptr;
3611 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3612 struct bonding *bond, *bond_next;
3613 struct vlan_entry *vlan, *vlan_next;
3615 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3616 if (bond->dev == event_dev) {
3619 bond->master_ip = ifa->ifa_local;
3622 bond->master_ip = bond_glean_dev_ip(bond->dev);
3629 if (list_empty(&bond->vlan_list))
3632 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3634 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3635 if (vlan_dev == event_dev) {
3638 vlan->vlan_ip = ifa->ifa_local;
3642 bond_glean_dev_ip(vlan_dev);
3653 static struct notifier_block bond_netdev_notifier = {
3654 .notifier_call = bond_netdev_event,
3657 static struct notifier_block bond_inetaddr_notifier = {
3658 .notifier_call = bond_inetaddr_event,
3661 /*-------------------------- Packet type handling ---------------------------*/
3663 /* register to receive lacpdus on a bond */
3664 static void bond_register_lacpdu(struct bonding *bond)
3666 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3668 /* initialize packet type */
3669 pk_type->type = PKT_TYPE_LACPDU;
3670 pk_type->dev = bond->dev;
3671 pk_type->func = bond_3ad_lacpdu_recv;
3673 dev_add_pack(pk_type);
3676 /* unregister to receive lacpdus on a bond */
3677 static void bond_unregister_lacpdu(struct bonding *bond)
3679 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3682 /*---------------------------- Hashing Policies -----------------------------*/
3685 * Hash for the the output device based upon layer 3 and layer 4 data. If
3686 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3687 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3689 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3690 struct net_device *bond_dev, int count)
3692 struct ethhdr *data = (struct ethhdr *)skb->data;
3693 struct iphdr *iph = skb->nh.iph;
3694 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3697 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3698 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3699 (iph->protocol == IPPROTO_TCP ||
3700 iph->protocol == IPPROTO_UDP)) {
3701 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3703 return (layer4_xor ^
3704 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3708 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3712 * Hash for the output device based upon layer 2 data
3714 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3715 struct net_device *bond_dev, int count)
3717 struct ethhdr *data = (struct ethhdr *)skb->data;
3719 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3722 /*-------------------------- Device entry points ----------------------------*/
3724 static int bond_open(struct net_device *bond_dev)
3726 struct bonding *bond = bond_dev->priv;
3727 struct timer_list *mii_timer = &bond->mii_timer;
3728 struct timer_list *arp_timer = &bond->arp_timer;
3730 bond->kill_timers = 0;
3732 if ((bond->params.mode == BOND_MODE_TLB) ||
3733 (bond->params.mode == BOND_MODE_ALB)) {
3734 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3736 /* bond_alb_initialize must be called before the timer
3739 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3740 /* something went wrong - fail the open operation */
3744 init_timer(alb_timer);
3745 alb_timer->expires = jiffies + 1;
3746 alb_timer->data = (unsigned long)bond;
3747 alb_timer->function = (void *)&bond_alb_monitor;
3748 add_timer(alb_timer);
3751 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3752 init_timer(mii_timer);
3753 mii_timer->expires = jiffies + 1;
3754 mii_timer->data = (unsigned long)bond_dev;
3755 mii_timer->function = (void *)&bond_mii_monitor;
3756 add_timer(mii_timer);
3759 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3760 init_timer(arp_timer);
3761 arp_timer->expires = jiffies + 1;
3762 arp_timer->data = (unsigned long)bond_dev;
3763 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3764 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3766 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3768 add_timer(arp_timer);
3771 if (bond->params.mode == BOND_MODE_8023AD) {
3772 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3773 init_timer(ad_timer);
3774 ad_timer->expires = jiffies + 1;
3775 ad_timer->data = (unsigned long)bond;
3776 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3777 add_timer(ad_timer);
3779 /* register to receive LACPDUs */
3780 bond_register_lacpdu(bond);
3786 static int bond_close(struct net_device *bond_dev)
3788 struct bonding *bond = bond_dev->priv;
3790 if (bond->params.mode == BOND_MODE_8023AD) {
3791 /* Unregister the receive of LACPDUs */
3792 bond_unregister_lacpdu(bond);
3795 write_lock_bh(&bond->lock);
3797 bond_mc_list_destroy(bond);
3799 /* signal timers not to re-arm */
3800 bond->kill_timers = 1;
3802 write_unlock_bh(&bond->lock);
3804 /* del_timer_sync must run without holding the bond->lock
3805 * because a running timer might be trying to hold it too
3808 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3809 del_timer_sync(&bond->mii_timer);
3812 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3813 del_timer_sync(&bond->arp_timer);
3816 switch (bond->params.mode) {
3817 case BOND_MODE_8023AD:
3818 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3822 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3828 /* Release the bonded slaves */
3829 bond_release_all(bond_dev);
3831 if ((bond->params.mode == BOND_MODE_TLB) ||
3832 (bond->params.mode == BOND_MODE_ALB)) {
3833 /* Must be called only after all
3834 * slaves have been released
3836 bond_alb_deinitialize(bond);
3842 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3844 struct bonding *bond = bond_dev->priv;
3845 struct net_device_stats *stats = &(bond->stats), *sstats;
3846 struct slave *slave;
3849 memset(stats, 0, sizeof(struct net_device_stats));
3851 read_lock_bh(&bond->lock);
3853 bond_for_each_slave(bond, slave, i) {
3854 sstats = slave->dev->get_stats(slave->dev);
3856 stats->rx_packets += sstats->rx_packets;
3857 stats->rx_bytes += sstats->rx_bytes;
3858 stats->rx_errors += sstats->rx_errors;
3859 stats->rx_dropped += sstats->rx_dropped;
3861 stats->tx_packets += sstats->tx_packets;
3862 stats->tx_bytes += sstats->tx_bytes;
3863 stats->tx_errors += sstats->tx_errors;
3864 stats->tx_dropped += sstats->tx_dropped;
3866 stats->multicast += sstats->multicast;
3867 stats->collisions += sstats->collisions;
3869 stats->rx_length_errors += sstats->rx_length_errors;
3870 stats->rx_over_errors += sstats->rx_over_errors;
3871 stats->rx_crc_errors += sstats->rx_crc_errors;
3872 stats->rx_frame_errors += sstats->rx_frame_errors;
3873 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3874 stats->rx_missed_errors += sstats->rx_missed_errors;
3876 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3877 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3878 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3879 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3880 stats->tx_window_errors += sstats->tx_window_errors;
3883 read_unlock_bh(&bond->lock);
3888 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3890 struct net_device *slave_dev = NULL;
3891 struct ifbond k_binfo;
3892 struct ifbond __user *u_binfo = NULL;
3893 struct ifslave k_sinfo;
3894 struct ifslave __user *u_sinfo = NULL;
3895 struct mii_ioctl_data *mii = NULL;
3898 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3899 bond_dev->name, cmd);
3911 * We do this again just in case we were called by SIOCGMIIREG
3912 * instead of SIOCGMIIPHY.
3919 if (mii->reg_num == 1) {
3920 struct bonding *bond = bond_dev->priv;
3922 read_lock_bh(&bond->lock);
3923 read_lock(&bond->curr_slave_lock);
3924 if (bond->curr_active_slave) {
3925 mii->val_out = BMSR_LSTATUS;
3927 read_unlock(&bond->curr_slave_lock);
3928 read_unlock_bh(&bond->lock);
3932 case BOND_INFO_QUERY_OLD:
3933 case SIOCBONDINFOQUERY:
3934 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3936 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3940 res = bond_info_query(bond_dev, &k_binfo);
3942 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3948 case BOND_SLAVE_INFO_QUERY_OLD:
3949 case SIOCBONDSLAVEINFOQUERY:
3950 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3952 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3956 res = bond_slave_info_query(bond_dev, &k_sinfo);
3958 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3969 if (!capable(CAP_NET_ADMIN)) {
3973 slave_dev = dev_get_by_name(ifr->ifr_slave);
3975 dprintk("slave_dev=%p: \n", slave_dev);
3980 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3982 case BOND_ENSLAVE_OLD:
3983 case SIOCBONDENSLAVE:
3984 res = bond_enslave(bond_dev, slave_dev);
3986 case BOND_RELEASE_OLD:
3987 case SIOCBONDRELEASE:
3988 res = bond_release(bond_dev, slave_dev);
3990 case BOND_SETHWADDR_OLD:
3991 case SIOCBONDSETHWADDR:
3992 res = bond_sethwaddr(bond_dev, slave_dev);
3994 case BOND_CHANGE_ACTIVE_OLD:
3995 case SIOCBONDCHANGEACTIVE:
3996 res = bond_ioctl_change_active(bond_dev, slave_dev);
4008 static void bond_set_multicast_list(struct net_device *bond_dev)
4010 struct bonding *bond = bond_dev->priv;
4011 struct dev_mc_list *dmi;
4013 write_lock_bh(&bond->lock);
4016 * Do promisc before checking multicast_mode
4018 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4019 bond_set_promiscuity(bond, 1);
4022 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4023 bond_set_promiscuity(bond, -1);
4026 /* set allmulti flag to slaves */
4027 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4028 bond_set_allmulti(bond, 1);
4031 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4032 bond_set_allmulti(bond, -1);
4035 bond->flags = bond_dev->flags;
4037 /* looking for addresses to add to slaves' mc list */
4038 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4039 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4040 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4044 /* looking for addresses to delete from slaves' list */
4045 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4046 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4047 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4051 /* save master's multicast list */
4052 bond_mc_list_destroy(bond);
4053 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4055 write_unlock_bh(&bond->lock);
4059 * Change the MTU of all of a master's slaves to match the master
4061 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4063 struct bonding *bond = bond_dev->priv;
4064 struct slave *slave, *stop_at;
4068 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4069 (bond_dev ? bond_dev->name : "None"), new_mtu);
4071 /* Can't hold bond->lock with bh disabled here since
4072 * some base drivers panic. On the other hand we can't
4073 * hold bond->lock without bh disabled because we'll
4074 * deadlock. The only solution is to rely on the fact
4075 * that we're under rtnl_lock here, and the slaves
4076 * list won't change. This doesn't solve the problem
4077 * of setting the slave's MTU while it is
4078 * transmitting, but the assumption is that the base
4079 * driver can handle that.
4081 * TODO: figure out a way to safely iterate the slaves
4082 * list, but without holding a lock around the actual
4083 * call to the base driver.
4086 bond_for_each_slave(bond, slave, i) {
4087 dprintk("s %p s->p %p c_m %p\n", slave,
4088 slave->prev, slave->dev->change_mtu);
4089 res = dev_set_mtu(slave->dev, new_mtu);
4092 /* If we failed to set the slave's mtu to the new value
4093 * we must abort the operation even in ACTIVE_BACKUP
4094 * mode, because if we allow the backup slaves to have
4095 * different mtu values than the active slave we'll
4096 * need to change their mtu when doing a failover. That
4097 * means changing their mtu from timer context, which
4098 * is probably not a good idea.
4100 dprintk("err %d %s\n", res, slave->dev->name);
4105 bond_dev->mtu = new_mtu;
4110 /* unwind from head to the slave that failed */
4112 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4115 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4117 dprintk("unwind err %d dev %s\n", tmp_res,
4128 * Note that many devices must be down to change the HW address, and
4129 * downing the master releases all slaves. We can make bonds full of
4130 * bonding devices to test this, however.
4132 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4134 struct bonding *bond = bond_dev->priv;
4135 struct sockaddr *sa = addr, tmp_sa;
4136 struct slave *slave, *stop_at;
4140 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4142 if (!is_valid_ether_addr(sa->sa_data)) {
4143 return -EADDRNOTAVAIL;
4146 /* Can't hold bond->lock with bh disabled here since
4147 * some base drivers panic. On the other hand we can't
4148 * hold bond->lock without bh disabled because we'll
4149 * deadlock. The only solution is to rely on the fact
4150 * that we're under rtnl_lock here, and the slaves
4151 * list won't change. This doesn't solve the problem
4152 * of setting the slave's hw address while it is
4153 * transmitting, but the assumption is that the base
4154 * driver can handle that.
4156 * TODO: figure out a way to safely iterate the slaves
4157 * list, but without holding a lock around the actual
4158 * call to the base driver.
4161 bond_for_each_slave(bond, slave, i) {
4162 dprintk("slave %p %s\n", slave, slave->dev->name);
4164 if (slave->dev->set_mac_address == NULL) {
4166 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4170 res = dev_set_mac_address(slave->dev, addr);
4172 /* TODO: consider downing the slave
4174 * User should expect communications
4175 * breakage anyway until ARP finish
4178 dprintk("err %d %s\n", res, slave->dev->name);
4184 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4188 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4189 tmp_sa.sa_family = bond_dev->type;
4191 /* unwind from head to the slave that failed */
4193 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4196 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4198 dprintk("unwind err %d dev %s\n", tmp_res,
4206 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4208 struct bonding *bond = bond_dev->priv;
4209 struct slave *slave, *start_at;
4213 read_lock(&bond->lock);
4215 if (!BOND_IS_OK(bond)) {
4219 read_lock(&bond->curr_slave_lock);
4220 slave = start_at = bond->curr_active_slave;
4221 read_unlock(&bond->curr_slave_lock);
4227 bond_for_each_slave_from(bond, slave, i, start_at) {
4228 if (IS_UP(slave->dev) &&
4229 (slave->link == BOND_LINK_UP) &&
4230 (slave->state == BOND_STATE_ACTIVE)) {
4231 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4233 write_lock(&bond->curr_slave_lock);
4234 bond->curr_active_slave = slave->next;
4235 write_unlock(&bond->curr_slave_lock);
4244 /* no suitable interface, frame not sent */
4247 read_unlock(&bond->lock);
4251 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
4252 struct bonding *bond,
4253 struct slave *slave)
4255 struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
4256 struct ethhdr *eth_data;
4261 printk(KERN_ERR DRV_NAME ": Error: "
4262 "bond_activebackup_xmit_copy(): skb_copy() failed\n");
4266 skb2->mac.raw = (unsigned char *)skb2->data;
4267 eth_data = eth_hdr(skb2);
4269 /* Pick an appropriate source MAC address
4270 * -- use slave's perm MAC addr, unless used by bond
4271 * -- otherwise, borrow active slave's perm MAC addr
4272 * since that will not be used
4274 hwaddr = slave->perm_hwaddr;
4275 if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
4276 hwaddr = bond->curr_active_slave->perm_hwaddr;
4278 /* Set source MAC address appropriately */
4279 memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
4281 res = bond_dev_queue_xmit(bond, skb2, slave->dev);
4283 dev_kfree_skb(skb2);
4289 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4290 * the bond has a usable interface.
4292 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4294 struct bonding *bond = bond_dev->priv;
4297 read_lock(&bond->lock);
4298 read_lock(&bond->curr_slave_lock);
4300 if (!BOND_IS_OK(bond)) {
4304 if (!bond->curr_active_slave)
4307 /* Xmit IGMP frames on all slaves to ensure rapid fail-over
4308 for multicast traffic on snooping switches */
4309 if (skb->protocol == __constant_htons(ETH_P_IP) &&
4310 skb->nh.iph->protocol == IPPROTO_IGMP) {
4311 struct slave *slave, *active_slave;
4314 active_slave = bond->curr_active_slave;
4315 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
4317 if (IS_UP(slave->dev) &&
4318 (slave->link == BOND_LINK_UP))
4319 bond_activebackup_xmit_copy(skb, bond, slave);
4322 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4326 /* no suitable interface, frame not sent */
4329 read_unlock(&bond->curr_slave_lock);
4330 read_unlock(&bond->lock);
4335 * In bond_xmit_xor() , we determine the output device by using a pre-
4336 * determined xmit_hash_policy(), If the selected device is not enabled,
4337 * find the next active slave.
4339 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4341 struct bonding *bond = bond_dev->priv;
4342 struct slave *slave, *start_at;
4347 read_lock(&bond->lock);
4349 if (!BOND_IS_OK(bond)) {
4353 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4355 bond_for_each_slave(bond, slave, i) {
4364 bond_for_each_slave_from(bond, slave, i, start_at) {
4365 if (IS_UP(slave->dev) &&
4366 (slave->link == BOND_LINK_UP) &&
4367 (slave->state == BOND_STATE_ACTIVE)) {
4368 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4375 /* no suitable interface, frame not sent */
4378 read_unlock(&bond->lock);
4383 * in broadcast mode, we send everything to all usable interfaces.
4385 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4387 struct bonding *bond = bond_dev->priv;
4388 struct slave *slave, *start_at;
4389 struct net_device *tx_dev = NULL;
4393 read_lock(&bond->lock);
4395 if (!BOND_IS_OK(bond)) {
4399 read_lock(&bond->curr_slave_lock);
4400 start_at = bond->curr_active_slave;
4401 read_unlock(&bond->curr_slave_lock);
4407 bond_for_each_slave_from(bond, slave, i, start_at) {
4408 if (IS_UP(slave->dev) &&
4409 (slave->link == BOND_LINK_UP) &&
4410 (slave->state == BOND_STATE_ACTIVE)) {
4412 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4414 printk(KERN_ERR DRV_NAME
4415 ": %s: Error: bond_xmit_broadcast(): "
4416 "skb_clone() failed\n",
4421 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4423 dev_kfree_skb(skb2);
4427 tx_dev = slave->dev;
4432 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4437 /* no suitable interface, frame not sent */
4440 /* frame sent to all suitable interfaces */
4441 read_unlock(&bond->lock);
4445 /*------------------------- Device initialization ---------------------------*/
4448 * set bond mode specific net device operations
4450 void bond_set_mode_ops(struct bonding *bond, int mode)
4452 struct net_device *bond_dev = bond->dev;
4455 case BOND_MODE_ROUNDROBIN:
4456 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4458 case BOND_MODE_ACTIVEBACKUP:
4459 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4462 bond_dev->hard_start_xmit = bond_xmit_xor;
4463 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4464 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4466 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4468 case BOND_MODE_BROADCAST:
4469 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4471 case BOND_MODE_8023AD:
4472 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4473 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4474 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4476 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4480 bond_dev->hard_start_xmit = bond_alb_xmit;
4481 bond_dev->set_mac_address = bond_alb_set_mac_address;
4484 /* Should never happen, mode already checked */
4485 printk(KERN_ERR DRV_NAME
4486 ": %s: Error: Unknown bonding mode %d\n",
4493 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4494 struct ethtool_drvinfo *drvinfo)
4496 strncpy(drvinfo->driver, DRV_NAME, 32);
4497 strncpy(drvinfo->version, DRV_VERSION, 32);
4498 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4501 static struct ethtool_ops bond_ethtool_ops = {
4502 .get_tx_csum = ethtool_op_get_tx_csum,
4503 .get_sg = ethtool_op_get_sg,
4504 .get_drvinfo = bond_ethtool_get_drvinfo,
4508 * Does not allocate but creates a /proc entry.
4511 static int __init bond_init(struct net_device *bond_dev, struct bond_params *params)
4513 struct bonding *bond = bond_dev->priv;
4515 dprintk("Begin bond_init for %s\n", bond_dev->name);
4517 /* initialize rwlocks */
4518 rwlock_init(&bond->lock);
4519 rwlock_init(&bond->curr_slave_lock);
4521 bond->params = *params; /* copy params struct */
4523 /* Initialize pointers */
4524 bond->first_slave = NULL;
4525 bond->curr_active_slave = NULL;
4526 bond->current_arp_slave = NULL;
4527 bond->primary_slave = NULL;
4528 bond->dev = bond_dev;
4529 INIT_LIST_HEAD(&bond->vlan_list);
4531 /* Initialize the device entry points */
4532 bond_dev->open = bond_open;
4533 bond_dev->stop = bond_close;
4534 bond_dev->get_stats = bond_get_stats;
4535 bond_dev->do_ioctl = bond_do_ioctl;
4536 bond_dev->ethtool_ops = &bond_ethtool_ops;
4537 bond_dev->set_multicast_list = bond_set_multicast_list;
4538 bond_dev->change_mtu = bond_change_mtu;
4539 bond_dev->set_mac_address = bond_set_mac_address;
4541 bond_set_mode_ops(bond, bond->params.mode);
4543 bond_dev->destructor = free_netdev;
4545 /* Initialize the device options */
4546 bond_dev->tx_queue_len = 0;
4547 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4549 /* At first, we block adding VLANs. That's the only way to
4550 * prevent problems that occur when adding VLANs over an
4551 * empty bond. The block will be removed once non-challenged
4552 * slaves are enslaved.
4554 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4556 /* don't acquire bond device's xmit_lock when
4558 bond_dev->features |= NETIF_F_LLTX;
4560 /* By default, we declare the bond to be fully
4561 * VLAN hardware accelerated capable. Special
4562 * care is taken in the various xmit functions
4563 * when there are slaves that are not hw accel
4566 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4567 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4568 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4569 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4570 NETIF_F_HW_VLAN_RX |
4571 NETIF_F_HW_VLAN_FILTER);
4573 #ifdef CONFIG_PROC_FS
4574 bond_create_proc_entry(bond);
4577 list_add_tail(&bond->bond_list, &bond_dev_list);
4582 /* De-initialize device specific data.
4583 * Caller must hold rtnl_lock.
4585 void bond_deinit(struct net_device *bond_dev)
4587 struct bonding *bond = bond_dev->priv;
4589 list_del(&bond->bond_list);
4591 #ifdef CONFIG_PROC_FS
4592 bond_remove_proc_entry(bond);
4596 /* Unregister and free all bond devices.
4597 * Caller must hold rtnl_lock.
4599 static void bond_free_all(void)
4601 struct bonding *bond, *nxt;
4603 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4604 struct net_device *bond_dev = bond->dev;
4606 unregister_netdevice(bond_dev);
4607 bond_deinit(bond_dev);
4610 #ifdef CONFIG_PROC_FS
4611 bond_destroy_proc_dir();
4615 /*------------------------- Module initialization ---------------------------*/
4618 * Convert string input module parms. Accept either the
4619 * number of the mode or its string name.
4621 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4625 for (i = 0; tbl[i].modename; i++) {
4626 if ((isdigit(*mode_arg) &&
4627 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4628 (strncmp(mode_arg, tbl[i].modename,
4629 strlen(tbl[i].modename)) == 0)) {
4637 static int bond_check_params(struct bond_params *params)
4640 * Convert string parameters.
4643 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4644 if (bond_mode == -1) {
4645 printk(KERN_ERR DRV_NAME
4646 ": Error: Invalid bonding mode \"%s\"\n",
4647 mode == NULL ? "NULL" : mode);
4652 if (xmit_hash_policy) {
4653 if ((bond_mode != BOND_MODE_XOR) &&
4654 (bond_mode != BOND_MODE_8023AD)) {
4655 printk(KERN_INFO DRV_NAME
4656 ": xor_mode param is irrelevant in mode %s\n",
4657 bond_mode_name(bond_mode));
4659 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4661 if (xmit_hashtype == -1) {
4662 printk(KERN_ERR DRV_NAME
4663 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4664 xmit_hash_policy == NULL ? "NULL" :
4672 if (bond_mode != BOND_MODE_8023AD) {
4673 printk(KERN_INFO DRV_NAME
4674 ": lacp_rate param is irrelevant in mode %s\n",
4675 bond_mode_name(bond_mode));
4677 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4678 if (lacp_fast == -1) {
4679 printk(KERN_ERR DRV_NAME
4680 ": Error: Invalid lacp rate \"%s\"\n",
4681 lacp_rate == NULL ? "NULL" : lacp_rate);
4687 if (max_bonds < 1 || max_bonds > INT_MAX) {
4688 printk(KERN_WARNING DRV_NAME
4689 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4690 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4691 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4692 max_bonds = BOND_DEFAULT_MAX_BONDS;
4696 printk(KERN_WARNING DRV_NAME
4697 ": Warning: miimon module parameter (%d), "
4698 "not in range 0-%d, so it was reset to %d\n",
4699 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4700 miimon = BOND_LINK_MON_INTERV;
4704 printk(KERN_WARNING DRV_NAME
4705 ": Warning: updelay module parameter (%d), "
4706 "not in range 0-%d, so it was reset to 0\n",
4711 if (downdelay < 0) {
4712 printk(KERN_WARNING DRV_NAME
4713 ": Warning: downdelay module parameter (%d), "
4714 "not in range 0-%d, so it was reset to 0\n",
4715 downdelay, INT_MAX);
4719 if ((use_carrier != 0) && (use_carrier != 1)) {
4720 printk(KERN_WARNING DRV_NAME
4721 ": Warning: use_carrier module parameter (%d), "
4722 "not of valid value (0/1), so it was set to 1\n",
4727 /* reset values for 802.3ad */
4728 if (bond_mode == BOND_MODE_8023AD) {
4730 printk(KERN_WARNING DRV_NAME
4731 ": Warning: miimon must be specified, "
4732 "otherwise bonding will not detect link "
4733 "failure, speed and duplex which are "
4734 "essential for 802.3ad operation\n");
4735 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4740 /* reset values for TLB/ALB */
4741 if ((bond_mode == BOND_MODE_TLB) ||
4742 (bond_mode == BOND_MODE_ALB)) {
4744 printk(KERN_WARNING DRV_NAME
4745 ": Warning: miimon must be specified, "
4746 "otherwise bonding will not detect link "
4747 "failure and link speed which are essential "
4748 "for TLB/ALB load balancing\n");
4749 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4754 if (bond_mode == BOND_MODE_ALB) {
4755 printk(KERN_NOTICE DRV_NAME
4756 ": In ALB mode you might experience client "
4757 "disconnections upon reconnection of a link if the "
4758 "bonding module updelay parameter (%d msec) is "
4759 "incompatible with the forwarding delay time of the "
4765 if (updelay || downdelay) {
4766 /* just warn the user the up/down delay will have
4767 * no effect since miimon is zero...
4769 printk(KERN_WARNING DRV_NAME
4770 ": Warning: miimon module parameter not set "
4771 "and updelay (%d) or downdelay (%d) module "
4772 "parameter is set; updelay and downdelay have "
4773 "no effect unless miimon is set\n",
4774 updelay, downdelay);
4777 /* don't allow arp monitoring */
4779 printk(KERN_WARNING DRV_NAME
4780 ": Warning: miimon (%d) and arp_interval (%d) "
4781 "can't be used simultaneously, disabling ARP "
4783 miimon, arp_interval);
4787 if ((updelay % miimon) != 0) {
4788 printk(KERN_WARNING DRV_NAME
4789 ": Warning: updelay (%d) is not a multiple "
4790 "of miimon (%d), updelay rounded to %d ms\n",
4791 updelay, miimon, (updelay / miimon) * miimon);
4796 if ((downdelay % miimon) != 0) {
4797 printk(KERN_WARNING DRV_NAME
4798 ": Warning: downdelay (%d) is not a multiple "
4799 "of miimon (%d), downdelay rounded to %d ms\n",
4801 (downdelay / miimon) * miimon);
4804 downdelay /= miimon;
4807 if (arp_interval < 0) {
4808 printk(KERN_WARNING DRV_NAME
4809 ": Warning: arp_interval module parameter (%d) "
4810 ", not in range 0-%d, so it was reset to %d\n",
4811 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4812 arp_interval = BOND_LINK_ARP_INTERV;
4815 for (arp_ip_count = 0;
4816 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4818 /* not complete check, but should be good enough to
4820 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4821 printk(KERN_WARNING DRV_NAME
4822 ": Warning: bad arp_ip_target module parameter "
4823 "(%s), ARP monitoring will not be performed\n",
4824 arp_ip_target[arp_ip_count]);
4827 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4828 arp_target[arp_ip_count] = ip;
4832 if (arp_interval && !arp_ip_count) {
4833 /* don't allow arping if no arp_ip_target given... */
4834 printk(KERN_WARNING DRV_NAME
4835 ": Warning: arp_interval module parameter (%d) "
4836 "specified without providing an arp_ip_target "
4837 "parameter, arp_interval was reset to 0\n",
4843 printk(KERN_INFO DRV_NAME
4844 ": MII link monitoring set to %d ms\n",
4846 } else if (arp_interval) {
4849 printk(KERN_INFO DRV_NAME
4850 ": ARP monitoring set to %d ms with %d target(s):",
4851 arp_interval, arp_ip_count);
4853 for (i = 0; i < arp_ip_count; i++)
4854 printk (" %s", arp_ip_target[i]);
4859 /* miimon and arp_interval not set, we need one so things
4860 * work as expected, see bonding.txt for details
4862 printk(KERN_WARNING DRV_NAME
4863 ": Warning: either miimon or arp_interval and "
4864 "arp_ip_target module parameters must be specified, "
4865 "otherwise bonding will not detect link failures! see "
4866 "bonding.txt for details.\n");
4869 if (primary && !USES_PRIMARY(bond_mode)) {
4870 /* currently, using a primary only makes sense
4871 * in active backup, TLB or ALB modes
4873 printk(KERN_WARNING DRV_NAME
4874 ": Warning: %s primary device specified but has no "
4875 "effect in %s mode\n",
4876 primary, bond_mode_name(bond_mode));
4880 /* fill params struct with the proper values */
4881 params->mode = bond_mode;
4882 params->xmit_policy = xmit_hashtype;
4883 params->miimon = miimon;
4884 params->arp_interval = arp_interval;
4885 params->updelay = updelay;
4886 params->downdelay = downdelay;
4887 params->use_carrier = use_carrier;
4888 params->lacp_fast = lacp_fast;
4889 params->primary[0] = 0;
4892 strncpy(params->primary, primary, IFNAMSIZ);
4893 params->primary[IFNAMSIZ - 1] = 0;
4896 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4901 /* Create a new bond based on the specified name and bonding parameters.
4902 * Caller must NOT hold rtnl_lock; we need to release it here before we
4903 * set up our sysfs entries.
4905 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4907 struct net_device *bond_dev;
4911 bond_dev = alloc_netdev(sizeof(struct bonding), name, ether_setup);
4913 printk(KERN_ERR DRV_NAME
4914 ": %s: eek! can't alloc netdev!\n",
4920 /* bond_init() must be called after dev_alloc_name() (for the
4921 * /proc files), but before register_netdevice(), because we
4922 * need to set function pointers.
4925 res = bond_init(bond_dev, params);
4930 SET_MODULE_OWNER(bond_dev);
4932 res = register_netdevice(bond_dev);
4937 *newbond = bond_dev->priv;
4939 rtnl_unlock(); /* allows sysfs registration of net device */
4942 bond_deinit(bond_dev);
4944 free_netdev(bond_dev);
4951 static int __init bonding_init(void)
4955 char new_bond_name[8]; /* Enough room for 999 bonds at init. */
4957 printk(KERN_INFO "%s", version);
4959 res = bond_check_params(&bonding_defaults);
4964 #ifdef CONFIG_PROC_FS
4965 bond_create_proc_dir();
4967 for (i = 0; i < max_bonds; i++) {
4968 sprintf(new_bond_name, "bond%d",i);
4969 res = bond_create(new_bond_name,&bonding_defaults, NULL);
4974 register_netdevice_notifier(&bond_netdev_notifier);
4975 register_inetaddr_notifier(&bond_inetaddr_notifier);
4987 static void __exit bonding_exit(void)
4989 unregister_netdevice_notifier(&bond_netdev_notifier);
4990 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4997 module_init(bonding_init);
4998 module_exit(bonding_exit);
4999 MODULE_LICENSE("GPL");
5000 MODULE_VERSION(DRV_VERSION);
5001 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5002 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5003 MODULE_SUPPORTED_DEVICE("most ethernet devices");