Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * originally based on the dummy device. | |
3 | * | |
4 | * Copyright 1999, Thomas Davis, tadavis@lbl.gov. | |
5 | * Licensed under the GPL. Based on dummy.c, and eql.c devices. | |
6 | * | |
7 | * bonding.c: an Ethernet Bonding driver | |
8 | * | |
9 | * This is useful to talk to a Cisco EtherChannel compatible equipment: | |
10 | * Cisco 5500 | |
11 | * Sun Trunking (Solaris) | |
12 | * Alteon AceDirector Trunks | |
13 | * Linux Bonding | |
14 | * and probably many L2 switches ... | |
15 | * | |
16 | * How it works: | |
17 | * ifconfig bond0 ipaddress netmask up | |
18 | * will setup a network device, with an ip address. No mac address | |
19 | * will be assigned at this time. The hw mac address will come from | |
20 | * the first slave bonded to the channel. All slaves will then use | |
21 | * this hw mac address. | |
22 | * | |
23 | * ifconfig bond0 down | |
24 | * will release all slaves, marking them as down. | |
25 | * | |
26 | * ifenslave bond0 eth0 | |
27 | * will attach eth0 to bond0 as a slave. eth0 hw mac address will either | |
28 | * a: be used as initial mac address | |
29 | * b: if a hw mac address already is there, eth0's hw mac address | |
30 | * will then be set from bond0. | |
31 | * | |
32 | * v0.1 - first working version. | |
33 | * v0.2 - changed stats to be calculated by summing slaves stats. | |
34 | * | |
35 | * Changes: | |
36 | * Arnaldo Carvalho de Melo <acme@conectiva.com.br> | |
37 | * - fix leaks on failure at bond_init | |
38 | * | |
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 | |
47 | * good slave. | |
48 | * - made the code hopefully SMP safe | |
49 | * | |
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 | |
57 | * wires ! :-> | |
58 | * | |
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 | |
64 | * scan. | |
65 | * - smarter enslaving and releasing. | |
66 | * - finer and more robust SMP locking | |
67 | * | |
68 | * 2000/10/17 - Willy Tarreau <willy at meta-x.org> | |
69 | * - fixed two potential SMP race conditions | |
70 | * | |
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(). | |
78 | * | |
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. | |
84 | * | |
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. | |
93 | * | |
94 | * 2001/4/5 - Chad N. Tindel <ctindel at ieee dot org> | |
95 | * - Ported to 2.4 Kernel | |
96 | * | |
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. | |
100 | * | |
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 | |
106 | * | |
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. | |
115 | * | |
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 | |
118 | * out routing table. | |
119 | * | |
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. | |
124 | * | |
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. | |
131 | * | |
132 | * 2001/8/13 - Erik Habbinga <erik_habbinga at hp dot com> | |
133 | * - correct locking problem with rtnl_exlock_nowait | |
134 | * | |
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 | |
138 | * | |
139 | * 2001/9/13 - Takao Indoh <indou dot takao at jp dot fujitsu dot com> | |
140 | * - Add the BOND_CHANGE_ACTIVE ioctl implementation | |
141 | * | |
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. | |
145 | * | |
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 | |
149 | * structure. | |
150 | * | |
151 | * 2001/9/25 - Janice Girouard <girouard at us.ibm.com> | |
152 | * - add arp_monitor for active backup mode | |
153 | * | |
154 | * 2001/10/23 - Takao Indoh <indou dot takao at jp dot fujitsu dot com> | |
155 | * - Various memory leak fixes | |
156 | * | |
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 | |
164 | * | |
165 | * 2002/1/2 - Chad N. Tindel <ctindel at ieee dot org> | |
166 | * - Restore original slave flags at release time. | |
167 | * | |
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 | |
174 | * | |
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. | |
179 | * | |
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) | |
189 | * | |
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(). | |
194 | * | |
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 | |
199 | * | |
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 | |
202 | * hw address. | |
203 | * | |
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. | |
207 | * | |
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. | |
213 | * | |
214 | * 2002/09/11 - Philippe De Muyter <phdm at macqel dot be> | |
215 | * - Added bond_xmit_broadcast logic. | |
216 | * - Added bond_mode() support function. | |
217 | * | |
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 | |
223 | * | |
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. | |
228 | * | |
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 | |
234 | * for a response | |
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 | |
239 | * mode | |
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 | |
248 | * | |
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 | |
253 | * | |
254 | * 2002/11/17 - Laurent Deniel <laurent.deniel at free.fr> | |
255 | * - fix bond_slave_info_query when slave_id = num_slaves | |
256 | * | |
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. | |
260 | * | |
261 | * 2002/11/22 - Bert Barbe <bert.barbe at oracle dot com> | |
262 | * - Add support for multiple arp_ip_target | |
263 | * | |
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 | |
270 | * | |
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. | |
275 | * | |
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 | |
279 | * | |
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. | |
289 | * | |
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. | |
297 | * | |
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. | |
302 | * | |
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. | |
306 | * | |
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 | |
310 | * each slave. | |
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 | |
319 | * properly. | |
320 | * - Block possibility of enslaving before the master is up. This | |
321 | * prevents putting the system in an unstable state. | |
322 | * | |
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. | |
327 | * | |
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. | |
331 | * | |
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. | |
342 | * | |
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 | |
348 | * | |
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 | |
358 | * | |
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 | |
363 | * changed. | |
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 | |
369 | * | |
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 | |
374 | * | |
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 | |
379 | * | |
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. | |
383 | * | |
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 | |
387 | * bond to be down. | |
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). | |
391 | * | |
392 | * 2003/06/30 - Amir Noam <amir.noam at intel dot com> | |
393 | * - Fixed bond_change_active() for ALB/TLB modes. | |
394 | * Version to 2.2.14. | |
395 | * | |
396 | * 2003/07/29 - Amir Noam <amir.noam at intel dot com> | |
397 | * - Fixed ARP monitoring bug. | |
398 | * Version to 2.2.15. | |
399 | * | |
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. | |
403 | * Version to 2.2.16. | |
404 | * | |
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. | |
410 | * Version to 2.3.0. | |
411 | * | |
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. | |
422 | * Version to 2.4.0. | |
423 | * | |
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. | |
429 | * | |
430 | * 2003/11/20 - Amir Noam <amir.noam at intel dot com> | |
431 | * - Fix /proc creation/destruction. | |
432 | * | |
433 | * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com> | |
434 | * - Massive cleanup - Set version to 2.5.0 | |
435 | * Code changes: | |
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 | |
442 | * according to mode. | |
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. | |
446 | * Style changes: | |
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. | |
455 | * | |
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 | |
459 | * duplication. | |
460 | * - Add support for slaves that use ethtool_ops. | |
461 | * Set version to 2.5.3. | |
462 | * | |
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. | |
466 | * | |
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 | |
476 | * spinlock. | |
477 | * Set version to 2.6.1. | |
c3ade5ca JV |
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. | |
169a3e66 JV |
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. | |
1da177e4 LT |
490 | */ |
491 | ||
492 | //#define BONDING_DEBUG 1 | |
493 | ||
494 | #include <linux/config.h> | |
495 | #include <linux/kernel.h> | |
496 | #include <linux/module.h> | |
497 | #include <linux/sched.h> | |
498 | #include <linux/types.h> | |
499 | #include <linux/fcntl.h> | |
500 | #include <linux/interrupt.h> | |
501 | #include <linux/ptrace.h> | |
502 | #include <linux/ioport.h> | |
503 | #include <linux/in.h> | |
169a3e66 | 504 | #include <net/ip.h> |
1da177e4 | 505 | #include <linux/ip.h> |
169a3e66 JV |
506 | #include <linux/tcp.h> |
507 | #include <linux/udp.h> | |
1da177e4 LT |
508 | #include <linux/slab.h> |
509 | #include <linux/string.h> | |
510 | #include <linux/init.h> | |
511 | #include <linux/timer.h> | |
512 | #include <linux/socket.h> | |
513 | #include <linux/ctype.h> | |
514 | #include <linux/inet.h> | |
515 | #include <linux/bitops.h> | |
516 | #include <asm/system.h> | |
517 | #include <asm/io.h> | |
518 | #include <asm/dma.h> | |
519 | #include <asm/uaccess.h> | |
520 | #include <linux/errno.h> | |
521 | #include <linux/netdevice.h> | |
522 | #include <linux/inetdevice.h> | |
523 | #include <linux/etherdevice.h> | |
524 | #include <linux/skbuff.h> | |
525 | #include <net/sock.h> | |
526 | #include <linux/rtnetlink.h> | |
527 | #include <linux/proc_fs.h> | |
528 | #include <linux/seq_file.h> | |
529 | #include <linux/smp.h> | |
530 | #include <linux/if_ether.h> | |
531 | #include <net/arp.h> | |
532 | #include <linux/mii.h> | |
533 | #include <linux/ethtool.h> | |
534 | #include <linux/if_vlan.h> | |
535 | #include <linux/if_bonding.h> | |
c3ade5ca | 536 | #include <net/route.h> |
1da177e4 LT |
537 | #include "bonding.h" |
538 | #include "bond_3ad.h" | |
539 | #include "bond_alb.h" | |
540 | ||
541 | /*---------------------------- Module parameters ----------------------------*/ | |
542 | ||
543 | /* monitor all links that often (in milliseconds). <=0 disables monitoring */ | |
544 | #define BOND_LINK_MON_INTERV 0 | |
545 | #define BOND_LINK_ARP_INTERV 0 | |
546 | ||
547 | static int max_bonds = BOND_DEFAULT_MAX_BONDS; | |
548 | static int miimon = BOND_LINK_MON_INTERV; | |
549 | static int updelay = 0; | |
550 | static int downdelay = 0; | |
551 | static int use_carrier = 1; | |
552 | static char *mode = NULL; | |
553 | static char *primary = NULL; | |
554 | static char *lacp_rate = NULL; | |
169a3e66 | 555 | static char *xmit_hash_policy = NULL; |
1da177e4 LT |
556 | static int arp_interval = BOND_LINK_ARP_INTERV; |
557 | static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, }; | |
558 | ||
559 | module_param(max_bonds, int, 0); | |
560 | MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); | |
561 | module_param(miimon, int, 0); | |
562 | MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); | |
563 | module_param(updelay, int, 0); | |
564 | MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); | |
565 | module_param(downdelay, int, 0); | |
566 | MODULE_PARM_DESC(downdelay, "Delay before considering link down, in milliseconds"); | |
567 | module_param(use_carrier, int, 0); | |
568 | MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; 0 for off, 1 for on (default)"); | |
569 | module_param(mode, charp, 0); | |
570 | MODULE_PARM_DESC(mode, "Mode of operation : 0 for round robin, 1 for active-backup, 2 for xor"); | |
571 | module_param(primary, charp, 0); | |
572 | MODULE_PARM_DESC(primary, "Primary network device to use"); | |
573 | module_param(lacp_rate, charp, 0); | |
574 | MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner (slow/fast)"); | |
169a3e66 JV |
575 | module_param(xmit_hash_policy, charp, 0); |
576 | MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method : 0 for layer 2 (default), 1 for layer 3+4"); | |
1da177e4 LT |
577 | module_param(arp_interval, int, 0); |
578 | MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); | |
579 | module_param_array(arp_ip_target, charp, NULL, 0); | |
580 | MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); | |
581 | ||
582 | /*----------------------------- Global variables ----------------------------*/ | |
583 | ||
584 | static const char *version = | |
585 | DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n"; | |
586 | ||
587 | static LIST_HEAD(bond_dev_list); | |
588 | ||
589 | #ifdef CONFIG_PROC_FS | |
590 | static struct proc_dir_entry *bond_proc_dir = NULL; | |
591 | #endif | |
592 | ||
593 | static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ; | |
594 | static int arp_ip_count = 0; | |
1da177e4 | 595 | static int bond_mode = BOND_MODE_ROUNDROBIN; |
169a3e66 | 596 | static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2; |
1da177e4 LT |
597 | static int lacp_fast = 0; |
598 | static int app_abi_ver = 0; | |
599 | static int orig_app_abi_ver = -1; /* This is used to save the first ABI version | |
600 | * we receive from the application. Once set, | |
601 | * it won't be changed, and the module will | |
602 | * refuse to enslave/release interfaces if the | |
603 | * command comes from an application using | |
604 | * another ABI version. | |
605 | */ | |
1da177e4 LT |
606 | struct bond_parm_tbl { |
607 | char *modename; | |
608 | int mode; | |
609 | }; | |
610 | ||
611 | static struct bond_parm_tbl bond_lacp_tbl[] = { | |
612 | { "slow", AD_LACP_SLOW}, | |
613 | { "fast", AD_LACP_FAST}, | |
614 | { NULL, -1}, | |
615 | }; | |
616 | ||
617 | static struct bond_parm_tbl bond_mode_tbl[] = { | |
618 | { "balance-rr", BOND_MODE_ROUNDROBIN}, | |
619 | { "active-backup", BOND_MODE_ACTIVEBACKUP}, | |
620 | { "balance-xor", BOND_MODE_XOR}, | |
621 | { "broadcast", BOND_MODE_BROADCAST}, | |
622 | { "802.3ad", BOND_MODE_8023AD}, | |
623 | { "balance-tlb", BOND_MODE_TLB}, | |
624 | { "balance-alb", BOND_MODE_ALB}, | |
625 | { NULL, -1}, | |
626 | }; | |
627 | ||
169a3e66 JV |
628 | static struct bond_parm_tbl xmit_hashtype_tbl[] = { |
629 | { "layer2", BOND_XMIT_POLICY_LAYER2}, | |
630 | { "layer3+4", BOND_XMIT_POLICY_LAYER34}, | |
631 | { NULL, -1}, | |
632 | }; | |
633 | ||
1da177e4 LT |
634 | /*-------------------------- Forward declarations ---------------------------*/ |
635 | ||
169a3e66 | 636 | static inline void bond_set_mode_ops(struct bonding *bond, int mode); |
c3ade5ca | 637 | static void bond_send_gratuitous_arp(struct bonding *bond); |
1da177e4 LT |
638 | |
639 | /*---------------------------- General routines -----------------------------*/ | |
640 | ||
641 | static const char *bond_mode_name(int mode) | |
642 | { | |
643 | switch (mode) { | |
644 | case BOND_MODE_ROUNDROBIN : | |
645 | return "load balancing (round-robin)"; | |
646 | case BOND_MODE_ACTIVEBACKUP : | |
647 | return "fault-tolerance (active-backup)"; | |
648 | case BOND_MODE_XOR : | |
649 | return "load balancing (xor)"; | |
650 | case BOND_MODE_BROADCAST : | |
651 | return "fault-tolerance (broadcast)"; | |
652 | case BOND_MODE_8023AD: | |
653 | return "IEEE 802.3ad Dynamic link aggregation"; | |
654 | case BOND_MODE_TLB: | |
655 | return "transmit load balancing"; | |
656 | case BOND_MODE_ALB: | |
657 | return "adaptive load balancing"; | |
658 | default: | |
659 | return "unknown"; | |
660 | } | |
661 | } | |
662 | ||
663 | /*---------------------------------- VLAN -----------------------------------*/ | |
664 | ||
665 | /** | |
666 | * bond_add_vlan - add a new vlan id on bond | |
667 | * @bond: bond that got the notification | |
668 | * @vlan_id: the vlan id to add | |
669 | * | |
670 | * Returns -ENOMEM if allocation failed. | |
671 | */ | |
672 | static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id) | |
673 | { | |
674 | struct vlan_entry *vlan; | |
675 | ||
676 | dprintk("bond: %s, vlan id %d\n", | |
677 | (bond ? bond->dev->name: "None"), vlan_id); | |
678 | ||
679 | vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL); | |
680 | if (!vlan) { | |
681 | return -ENOMEM; | |
682 | } | |
683 | ||
684 | INIT_LIST_HEAD(&vlan->vlan_list); | |
685 | vlan->vlan_id = vlan_id; | |
c3ade5ca | 686 | vlan->vlan_ip = 0; |
1da177e4 LT |
687 | |
688 | write_lock_bh(&bond->lock); | |
689 | ||
690 | list_add_tail(&vlan->vlan_list, &bond->vlan_list); | |
691 | ||
692 | write_unlock_bh(&bond->lock); | |
693 | ||
694 | dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name); | |
695 | ||
696 | return 0; | |
697 | } | |
698 | ||
699 | /** | |
700 | * bond_del_vlan - delete a vlan id from bond | |
701 | * @bond: bond that got the notification | |
702 | * @vlan_id: the vlan id to delete | |
703 | * | |
704 | * returns -ENODEV if @vlan_id was not found in @bond. | |
705 | */ | |
706 | static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id) | |
707 | { | |
708 | struct vlan_entry *vlan, *next; | |
709 | int res = -ENODEV; | |
710 | ||
711 | dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id); | |
712 | ||
713 | write_lock_bh(&bond->lock); | |
714 | ||
715 | list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) { | |
716 | if (vlan->vlan_id == vlan_id) { | |
717 | list_del(&vlan->vlan_list); | |
718 | ||
719 | if ((bond->params.mode == BOND_MODE_TLB) || | |
720 | (bond->params.mode == BOND_MODE_ALB)) { | |
721 | bond_alb_clear_vlan(bond, vlan_id); | |
722 | } | |
723 | ||
724 | dprintk("removed VLAN ID %d from bond %s\n", vlan_id, | |
725 | bond->dev->name); | |
726 | ||
727 | kfree(vlan); | |
728 | ||
729 | if (list_empty(&bond->vlan_list) && | |
730 | (bond->slave_cnt == 0)) { | |
731 | /* Last VLAN removed and no slaves, so | |
732 | * restore block on adding VLANs. This will | |
733 | * be removed once new slaves that are not | |
734 | * VLAN challenged will be added. | |
735 | */ | |
736 | bond->dev->features |= NETIF_F_VLAN_CHALLENGED; | |
737 | } | |
738 | ||
739 | res = 0; | |
740 | goto out; | |
741 | } | |
742 | } | |
743 | ||
744 | dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id, | |
745 | bond->dev->name); | |
746 | ||
747 | out: | |
748 | write_unlock_bh(&bond->lock); | |
749 | return res; | |
750 | } | |
751 | ||
752 | /** | |
753 | * bond_has_challenged_slaves | |
754 | * @bond: the bond we're working on | |
755 | * | |
756 | * Searches the slave list. Returns 1 if a vlan challenged slave | |
757 | * was found, 0 otherwise. | |
758 | * | |
759 | * Assumes bond->lock is held. | |
760 | */ | |
761 | static int bond_has_challenged_slaves(struct bonding *bond) | |
762 | { | |
763 | struct slave *slave; | |
764 | int i; | |
765 | ||
766 | bond_for_each_slave(bond, slave, i) { | |
767 | if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) { | |
768 | dprintk("found VLAN challenged slave - %s\n", | |
769 | slave->dev->name); | |
770 | return 1; | |
771 | } | |
772 | } | |
773 | ||
774 | dprintk("no VLAN challenged slaves found\n"); | |
775 | return 0; | |
776 | } | |
777 | ||
778 | /** | |
779 | * bond_next_vlan - safely skip to the next item in the vlans list. | |
780 | * @bond: the bond we're working on | |
781 | * @curr: item we're advancing from | |
782 | * | |
783 | * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL, | |
784 | * or @curr->next otherwise (even if it is @curr itself again). | |
785 | * | |
786 | * Caller must hold bond->lock | |
787 | */ | |
788 | struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr) | |
789 | { | |
790 | struct vlan_entry *next, *last; | |
791 | ||
792 | if (list_empty(&bond->vlan_list)) { | |
793 | return NULL; | |
794 | } | |
795 | ||
796 | if (!curr) { | |
797 | next = list_entry(bond->vlan_list.next, | |
798 | struct vlan_entry, vlan_list); | |
799 | } else { | |
800 | last = list_entry(bond->vlan_list.prev, | |
801 | struct vlan_entry, vlan_list); | |
802 | if (last == curr) { | |
803 | next = list_entry(bond->vlan_list.next, | |
804 | struct vlan_entry, vlan_list); | |
805 | } else { | |
806 | next = list_entry(curr->vlan_list.next, | |
807 | struct vlan_entry, vlan_list); | |
808 | } | |
809 | } | |
810 | ||
811 | return next; | |
812 | } | |
813 | ||
814 | /** | |
815 | * bond_dev_queue_xmit - Prepare skb for xmit. | |
816 | * | |
817 | * @bond: bond device that got this skb for tx. | |
818 | * @skb: hw accel VLAN tagged skb to transmit | |
819 | * @slave_dev: slave that is supposed to xmit this skbuff | |
820 | * | |
821 | * When the bond gets an skb to transmit that is | |
822 | * already hardware accelerated VLAN tagged, and it | |
823 | * needs to relay this skb to a slave that is not | |
824 | * hw accel capable, the skb needs to be "unaccelerated", | |
825 | * i.e. strip the hwaccel tag and re-insert it as part | |
826 | * of the payload. | |
827 | */ | |
828 | int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev) | |
829 | { | |
830 | unsigned short vlan_id; | |
831 | ||
832 | if (!list_empty(&bond->vlan_list) && | |
833 | !(slave_dev->features & NETIF_F_HW_VLAN_TX) && | |
834 | vlan_get_tag(skb, &vlan_id) == 0) { | |
835 | skb->dev = slave_dev; | |
836 | skb = vlan_put_tag(skb, vlan_id); | |
837 | if (!skb) { | |
838 | /* vlan_put_tag() frees the skb in case of error, | |
839 | * so return success here so the calling functions | |
840 | * won't attempt to free is again. | |
841 | */ | |
842 | return 0; | |
843 | } | |
844 | } else { | |
845 | skb->dev = slave_dev; | |
846 | } | |
847 | ||
848 | skb->priority = 1; | |
849 | dev_queue_xmit(skb); | |
850 | ||
851 | return 0; | |
852 | } | |
853 | ||
854 | /* | |
855 | * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid | |
856 | * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a | |
857 | * lock because: | |
858 | * a. This operation is performed in IOCTL context, | |
859 | * b. The operation is protected by the RTNL semaphore in the 8021q code, | |
860 | * c. Holding a lock with BH disabled while directly calling a base driver | |
861 | * entry point is generally a BAD idea. | |
862 | * | |
863 | * The design of synchronization/protection for this operation in the 8021q | |
864 | * module is good for one or more VLAN devices over a single physical device | |
865 | * and cannot be extended for a teaming solution like bonding, so there is a | |
866 | * potential race condition here where a net device from the vlan group might | |
867 | * be referenced (either by a base driver or the 8021q code) while it is being | |
868 | * removed from the system. However, it turns out we're not making matters | |
869 | * worse, and if it works for regular VLAN usage it will work here too. | |
870 | */ | |
871 | ||
872 | /** | |
873 | * bond_vlan_rx_register - Propagates registration to slaves | |
874 | * @bond_dev: bonding net device that got called | |
875 | * @grp: vlan group being registered | |
876 | */ | |
877 | static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp) | |
878 | { | |
879 | struct bonding *bond = bond_dev->priv; | |
880 | struct slave *slave; | |
881 | int i; | |
882 | ||
883 | bond->vlgrp = grp; | |
884 | ||
885 | bond_for_each_slave(bond, slave, i) { | |
886 | struct net_device *slave_dev = slave->dev; | |
887 | ||
888 | if ((slave_dev->features & NETIF_F_HW_VLAN_RX) && | |
889 | slave_dev->vlan_rx_register) { | |
890 | slave_dev->vlan_rx_register(slave_dev, grp); | |
891 | } | |
892 | } | |
893 | } | |
894 | ||
895 | /** | |
896 | * bond_vlan_rx_add_vid - Propagates adding an id to slaves | |
897 | * @bond_dev: bonding net device that got called | |
898 | * @vid: vlan id being added | |
899 | */ | |
900 | static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid) | |
901 | { | |
902 | struct bonding *bond = bond_dev->priv; | |
903 | struct slave *slave; | |
904 | int i, res; | |
905 | ||
906 | bond_for_each_slave(bond, slave, i) { | |
907 | struct net_device *slave_dev = slave->dev; | |
908 | ||
909 | if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) && | |
910 | slave_dev->vlan_rx_add_vid) { | |
911 | slave_dev->vlan_rx_add_vid(slave_dev, vid); | |
912 | } | |
913 | } | |
914 | ||
915 | res = bond_add_vlan(bond, vid); | |
916 | if (res) { | |
917 | printk(KERN_ERR DRV_NAME | |
918 | ": %s: Failed to add vlan id %d\n", | |
919 | bond_dev->name, vid); | |
920 | } | |
921 | } | |
922 | ||
923 | /** | |
924 | * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves | |
925 | * @bond_dev: bonding net device that got called | |
926 | * @vid: vlan id being removed | |
927 | */ | |
928 | static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid) | |
929 | { | |
930 | struct bonding *bond = bond_dev->priv; | |
931 | struct slave *slave; | |
932 | struct net_device *vlan_dev; | |
933 | int i, res; | |
934 | ||
935 | bond_for_each_slave(bond, slave, i) { | |
936 | struct net_device *slave_dev = slave->dev; | |
937 | ||
938 | if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) && | |
939 | slave_dev->vlan_rx_kill_vid) { | |
940 | /* Save and then restore vlan_dev in the grp array, | |
941 | * since the slave's driver might clear it. | |
942 | */ | |
943 | vlan_dev = bond->vlgrp->vlan_devices[vid]; | |
944 | slave_dev->vlan_rx_kill_vid(slave_dev, vid); | |
945 | bond->vlgrp->vlan_devices[vid] = vlan_dev; | |
946 | } | |
947 | } | |
948 | ||
949 | res = bond_del_vlan(bond, vid); | |
950 | if (res) { | |
951 | printk(KERN_ERR DRV_NAME | |
952 | ": %s: Failed to remove vlan id %d\n", | |
953 | bond_dev->name, vid); | |
954 | } | |
955 | } | |
956 | ||
957 | static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev) | |
958 | { | |
959 | struct vlan_entry *vlan; | |
960 | ||
961 | write_lock_bh(&bond->lock); | |
962 | ||
963 | if (list_empty(&bond->vlan_list)) { | |
964 | goto out; | |
965 | } | |
966 | ||
967 | if ((slave_dev->features & NETIF_F_HW_VLAN_RX) && | |
968 | slave_dev->vlan_rx_register) { | |
969 | slave_dev->vlan_rx_register(slave_dev, bond->vlgrp); | |
970 | } | |
971 | ||
972 | if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) || | |
973 | !(slave_dev->vlan_rx_add_vid)) { | |
974 | goto out; | |
975 | } | |
976 | ||
977 | list_for_each_entry(vlan, &bond->vlan_list, vlan_list) { | |
978 | slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id); | |
979 | } | |
980 | ||
981 | out: | |
982 | write_unlock_bh(&bond->lock); | |
983 | } | |
984 | ||
985 | static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev) | |
986 | { | |
987 | struct vlan_entry *vlan; | |
988 | struct net_device *vlan_dev; | |
989 | ||
990 | write_lock_bh(&bond->lock); | |
991 | ||
992 | if (list_empty(&bond->vlan_list)) { | |
993 | goto out; | |
994 | } | |
995 | ||
996 | if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) || | |
997 | !(slave_dev->vlan_rx_kill_vid)) { | |
998 | goto unreg; | |
999 | } | |
1000 | ||
1001 | list_for_each_entry(vlan, &bond->vlan_list, vlan_list) { | |
1002 | /* Save and then restore vlan_dev in the grp array, | |
1003 | * since the slave's driver might clear it. | |
1004 | */ | |
1005 | vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id]; | |
1006 | slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id); | |
1007 | bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev; | |
1008 | } | |
1009 | ||
1010 | unreg: | |
1011 | if ((slave_dev->features & NETIF_F_HW_VLAN_RX) && | |
1012 | slave_dev->vlan_rx_register) { | |
1013 | slave_dev->vlan_rx_register(slave_dev, NULL); | |
1014 | } | |
1015 | ||
1016 | out: | |
1017 | write_unlock_bh(&bond->lock); | |
1018 | } | |
1019 | ||
1020 | /*------------------------------- Link status -------------------------------*/ | |
1021 | ||
1022 | /* | |
1023 | * Get link speed and duplex from the slave's base driver | |
1024 | * using ethtool. If for some reason the call fails or the | |
1025 | * values are invalid, fake speed and duplex to 100/Full | |
1026 | * and return error. | |
1027 | */ | |
1028 | static int bond_update_speed_duplex(struct slave *slave) | |
1029 | { | |
1030 | struct net_device *slave_dev = slave->dev; | |
1031 | static int (* ioctl)(struct net_device *, struct ifreq *, int); | |
1032 | struct ifreq ifr; | |
1033 | struct ethtool_cmd etool; | |
1034 | ||
1035 | /* Fake speed and duplex */ | |
1036 | slave->speed = SPEED_100; | |
1037 | slave->duplex = DUPLEX_FULL; | |
1038 | ||
1039 | if (slave_dev->ethtool_ops) { | |
1040 | u32 res; | |
1041 | ||
1042 | if (!slave_dev->ethtool_ops->get_settings) { | |
1043 | return -1; | |
1044 | } | |
1045 | ||
1046 | res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool); | |
1047 | if (res < 0) { | |
1048 | return -1; | |
1049 | } | |
1050 | ||
1051 | goto verify; | |
1052 | } | |
1053 | ||
1054 | ioctl = slave_dev->do_ioctl; | |
1055 | strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); | |
1056 | etool.cmd = ETHTOOL_GSET; | |
1057 | ifr.ifr_data = (char*)&etool; | |
1058 | if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) { | |
1059 | return -1; | |
1060 | } | |
1061 | ||
1062 | verify: | |
1063 | switch (etool.speed) { | |
1064 | case SPEED_10: | |
1065 | case SPEED_100: | |
1066 | case SPEED_1000: | |
1067 | break; | |
1068 | default: | |
1069 | return -1; | |
1070 | } | |
1071 | ||
1072 | switch (etool.duplex) { | |
1073 | case DUPLEX_FULL: | |
1074 | case DUPLEX_HALF: | |
1075 | break; | |
1076 | default: | |
1077 | return -1; | |
1078 | } | |
1079 | ||
1080 | slave->speed = etool.speed; | |
1081 | slave->duplex = etool.duplex; | |
1082 | ||
1083 | return 0; | |
1084 | } | |
1085 | ||
1086 | /* | |
1087 | * if <dev> supports MII link status reporting, check its link status. | |
1088 | * | |
1089 | * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), | |
1090 | * depening upon the setting of the use_carrier parameter. | |
1091 | * | |
1092 | * Return either BMSR_LSTATUS, meaning that the link is up (or we | |
1093 | * can't tell and just pretend it is), or 0, meaning that the link is | |
1094 | * down. | |
1095 | * | |
1096 | * If reporting is non-zero, instead of faking link up, return -1 if | |
1097 | * both ETHTOOL and MII ioctls fail (meaning the device does not | |
1098 | * support them). If use_carrier is set, return whatever it says. | |
1099 | * It'd be nice if there was a good way to tell if a driver supports | |
1100 | * netif_carrier, but there really isn't. | |
1101 | */ | |
1102 | static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting) | |
1103 | { | |
1104 | static int (* ioctl)(struct net_device *, struct ifreq *, int); | |
1105 | struct ifreq ifr; | |
1106 | struct mii_ioctl_data *mii; | |
1107 | struct ethtool_value etool; | |
1108 | ||
1109 | if (bond->params.use_carrier) { | |
1110 | return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; | |
1111 | } | |
1112 | ||
1113 | ioctl = slave_dev->do_ioctl; | |
1114 | if (ioctl) { | |
1115 | /* TODO: set pointer to correct ioctl on a per team member */ | |
1116 | /* bases to make this more efficient. that is, once */ | |
1117 | /* we determine the correct ioctl, we will always */ | |
1118 | /* call it and not the others for that team */ | |
1119 | /* member. */ | |
1120 | ||
1121 | /* | |
1122 | * We cannot assume that SIOCGMIIPHY will also read a | |
1123 | * register; not all network drivers (e.g., e100) | |
1124 | * support that. | |
1125 | */ | |
1126 | ||
1127 | /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ | |
1128 | strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); | |
1129 | mii = if_mii(&ifr); | |
1130 | if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) { | |
1131 | mii->reg_num = MII_BMSR; | |
1132 | if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) { | |
1133 | return (mii->val_out & BMSR_LSTATUS); | |
1134 | } | |
1135 | } | |
1136 | } | |
1137 | ||
1138 | /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */ | |
1139 | /* for a period of time so we attempt to get link status */ | |
1140 | /* from it last if the above MII ioctls fail... */ | |
1141 | if (slave_dev->ethtool_ops) { | |
1142 | if (slave_dev->ethtool_ops->get_link) { | |
1143 | u32 link; | |
1144 | ||
1145 | link = slave_dev->ethtool_ops->get_link(slave_dev); | |
1146 | ||
1147 | return link ? BMSR_LSTATUS : 0; | |
1148 | } | |
1149 | } | |
1150 | ||
1151 | if (ioctl) { | |
1152 | strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); | |
1153 | etool.cmd = ETHTOOL_GLINK; | |
1154 | ifr.ifr_data = (char*)&etool; | |
1155 | if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) { | |
1156 | if (etool.data == 1) { | |
1157 | return BMSR_LSTATUS; | |
1158 | } else { | |
1159 | dprintk("SIOCETHTOOL shows link down\n"); | |
1160 | return 0; | |
1161 | } | |
1162 | } | |
1163 | } | |
1164 | ||
1165 | /* | |
1166 | * If reporting, report that either there's no dev->do_ioctl, | |
1167 | * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we | |
1168 | * cannot report link status). If not reporting, pretend | |
1169 | * we're ok. | |
1170 | */ | |
1171 | return (reporting ? -1 : BMSR_LSTATUS); | |
1172 | } | |
1173 | ||
1174 | /*----------------------------- Multicast list ------------------------------*/ | |
1175 | ||
1176 | /* | |
1177 | * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise | |
1178 | */ | |
1179 | static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2) | |
1180 | { | |
1181 | return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 && | |
1182 | dmi1->dmi_addrlen == dmi2->dmi_addrlen; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * returns dmi entry if found, NULL otherwise | |
1187 | */ | |
1188 | static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list) | |
1189 | { | |
1190 | struct dev_mc_list *idmi; | |
1191 | ||
1192 | for (idmi = mc_list; idmi; idmi = idmi->next) { | |
1193 | if (bond_is_dmi_same(dmi, idmi)) { | |
1194 | return idmi; | |
1195 | } | |
1196 | } | |
1197 | ||
1198 | return NULL; | |
1199 | } | |
1200 | ||
1201 | /* | |
1202 | * Push the promiscuity flag down to appropriate slaves | |
1203 | */ | |
1204 | static void bond_set_promiscuity(struct bonding *bond, int inc) | |
1205 | { | |
1206 | if (USES_PRIMARY(bond->params.mode)) { | |
1207 | /* write lock already acquired */ | |
1208 | if (bond->curr_active_slave) { | |
1209 | dev_set_promiscuity(bond->curr_active_slave->dev, inc); | |
1210 | } | |
1211 | } else { | |
1212 | struct slave *slave; | |
1213 | int i; | |
1214 | bond_for_each_slave(bond, slave, i) { | |
1215 | dev_set_promiscuity(slave->dev, inc); | |
1216 | } | |
1217 | } | |
1218 | } | |
1219 | ||
1220 | /* | |
1221 | * Push the allmulti flag down to all slaves | |
1222 | */ | |
1223 | static void bond_set_allmulti(struct bonding *bond, int inc) | |
1224 | { | |
1225 | if (USES_PRIMARY(bond->params.mode)) { | |
1226 | /* write lock already acquired */ | |
1227 | if (bond->curr_active_slave) { | |
1228 | dev_set_allmulti(bond->curr_active_slave->dev, inc); | |
1229 | } | |
1230 | } else { | |
1231 | struct slave *slave; | |
1232 | int i; | |
1233 | bond_for_each_slave(bond, slave, i) { | |
1234 | dev_set_allmulti(slave->dev, inc); | |
1235 | } | |
1236 | } | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * Add a Multicast address to slaves | |
1241 | * according to mode | |
1242 | */ | |
1243 | static void bond_mc_add(struct bonding *bond, void *addr, int alen) | |
1244 | { | |
1245 | if (USES_PRIMARY(bond->params.mode)) { | |
1246 | /* write lock already acquired */ | |
1247 | if (bond->curr_active_slave) { | |
1248 | dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0); | |
1249 | } | |
1250 | } else { | |
1251 | struct slave *slave; | |
1252 | int i; | |
1253 | bond_for_each_slave(bond, slave, i) { | |
1254 | dev_mc_add(slave->dev, addr, alen, 0); | |
1255 | } | |
1256 | } | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * Remove a multicast address from slave | |
1261 | * according to mode | |
1262 | */ | |
1263 | static void bond_mc_delete(struct bonding *bond, void *addr, int alen) | |
1264 | { | |
1265 | if (USES_PRIMARY(bond->params.mode)) { | |
1266 | /* write lock already acquired */ | |
1267 | if (bond->curr_active_slave) { | |
1268 | dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0); | |
1269 | } | |
1270 | } else { | |
1271 | struct slave *slave; | |
1272 | int i; | |
1273 | bond_for_each_slave(bond, slave, i) { | |
1274 | dev_mc_delete(slave->dev, addr, alen, 0); | |
1275 | } | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Totally destroys the mc_list in bond | |
1281 | */ | |
1282 | static void bond_mc_list_destroy(struct bonding *bond) | |
1283 | { | |
1284 | struct dev_mc_list *dmi; | |
1285 | ||
1286 | dmi = bond->mc_list; | |
1287 | while (dmi) { | |
1288 | bond->mc_list = dmi->next; | |
1289 | kfree(dmi); | |
1290 | dmi = bond->mc_list; | |
1291 | } | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Copy all the Multicast addresses from src to the bonding device dst | |
1296 | */ | |
1297 | static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond, int gpf_flag) | |
1298 | { | |
1299 | struct dev_mc_list *dmi, *new_dmi; | |
1300 | ||
1301 | for (dmi = mc_list; dmi; dmi = dmi->next) { | |
1302 | new_dmi = kmalloc(sizeof(struct dev_mc_list), gpf_flag); | |
1303 | ||
1304 | if (!new_dmi) { | |
1305 | /* FIXME: Potential memory leak !!! */ | |
1306 | return -ENOMEM; | |
1307 | } | |
1308 | ||
1309 | new_dmi->next = bond->mc_list; | |
1310 | bond->mc_list = new_dmi; | |
1311 | new_dmi->dmi_addrlen = dmi->dmi_addrlen; | |
1312 | memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen); | |
1313 | new_dmi->dmi_users = dmi->dmi_users; | |
1314 | new_dmi->dmi_gusers = dmi->dmi_gusers; | |
1315 | } | |
1316 | ||
1317 | return 0; | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * flush all members of flush->mc_list from device dev->mc_list | |
1322 | */ | |
1323 | static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev) | |
1324 | { | |
1325 | struct bonding *bond = bond_dev->priv; | |
1326 | struct dev_mc_list *dmi; | |
1327 | ||
1328 | for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) { | |
1329 | dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0); | |
1330 | } | |
1331 | ||
1332 | if (bond->params.mode == BOND_MODE_8023AD) { | |
1333 | /* del lacpdu mc addr from mc list */ | |
1334 | u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; | |
1335 | ||
1336 | dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0); | |
1337 | } | |
1338 | } | |
1339 | ||
1340 | /*--------------------------- Active slave change ---------------------------*/ | |
1341 | ||
1342 | /* | |
1343 | * Update the mc list and multicast-related flags for the new and | |
1344 | * old active slaves (if any) according to the multicast mode, and | |
1345 | * promiscuous flags unconditionally. | |
1346 | */ | |
1347 | static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active) | |
1348 | { | |
1349 | struct dev_mc_list *dmi; | |
1350 | ||
1351 | if (!USES_PRIMARY(bond->params.mode)) { | |
1352 | /* nothing to do - mc list is already up-to-date on | |
1353 | * all slaves | |
1354 | */ | |
1355 | return; | |
1356 | } | |
1357 | ||
1358 | if (old_active) { | |
1359 | if (bond->dev->flags & IFF_PROMISC) { | |
1360 | dev_set_promiscuity(old_active->dev, -1); | |
1361 | } | |
1362 | ||
1363 | if (bond->dev->flags & IFF_ALLMULTI) { | |
1364 | dev_set_allmulti(old_active->dev, -1); | |
1365 | } | |
1366 | ||
1367 | for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) { | |
1368 | dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0); | |
1369 | } | |
1370 | } | |
1371 | ||
1372 | if (new_active) { | |
1373 | if (bond->dev->flags & IFF_PROMISC) { | |
1374 | dev_set_promiscuity(new_active->dev, 1); | |
1375 | } | |
1376 | ||
1377 | if (bond->dev->flags & IFF_ALLMULTI) { | |
1378 | dev_set_allmulti(new_active->dev, 1); | |
1379 | } | |
1380 | ||
1381 | for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) { | |
1382 | dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0); | |
1383 | } | |
1384 | } | |
1385 | } | |
1386 | ||
1387 | /** | |
1388 | * find_best_interface - select the best available slave to be the active one | |
1389 | * @bond: our bonding struct | |
1390 | * | |
1391 | * Warning: Caller must hold curr_slave_lock for writing. | |
1392 | */ | |
1393 | static struct slave *bond_find_best_slave(struct bonding *bond) | |
1394 | { | |
1395 | struct slave *new_active, *old_active; | |
1396 | struct slave *bestslave = NULL; | |
1397 | int mintime = bond->params.updelay; | |
1398 | int i; | |
1399 | ||
1400 | new_active = old_active = bond->curr_active_slave; | |
1401 | ||
1402 | if (!new_active) { /* there were no active slaves left */ | |
1403 | if (bond->slave_cnt > 0) { /* found one slave */ | |
1404 | new_active = bond->first_slave; | |
1405 | } else { | |
1406 | return NULL; /* still no slave, return NULL */ | |
1407 | } | |
1408 | } | |
1409 | ||
1410 | /* first try the primary link; if arping, a link must tx/rx traffic | |
1411 | * before it can be considered the curr_active_slave - also, we would skip | |
1412 | * slaves between the curr_active_slave and primary_slave that may be up | |
1413 | * and able to arp | |
1414 | */ | |
1415 | if ((bond->primary_slave) && | |
1416 | (!bond->params.arp_interval) && | |
1417 | (IS_UP(bond->primary_slave->dev))) { | |
1418 | new_active = bond->primary_slave; | |
1419 | } | |
1420 | ||
1421 | /* remember where to stop iterating over the slaves */ | |
1422 | old_active = new_active; | |
1423 | ||
1424 | bond_for_each_slave_from(bond, new_active, i, old_active) { | |
1425 | if (IS_UP(new_active->dev)) { | |
1426 | if (new_active->link == BOND_LINK_UP) { | |
1427 | return new_active; | |
1428 | } else if (new_active->link == BOND_LINK_BACK) { | |
1429 | /* link up, but waiting for stabilization */ | |
1430 | if (new_active->delay < mintime) { | |
1431 | mintime = new_active->delay; | |
1432 | bestslave = new_active; | |
1433 | } | |
1434 | } | |
1435 | } | |
1436 | } | |
1437 | ||
1438 | return bestslave; | |
1439 | } | |
1440 | ||
1441 | /** | |
1442 | * change_active_interface - change the active slave into the specified one | |
1443 | * @bond: our bonding struct | |
1444 | * @new: the new slave to make the active one | |
1445 | * | |
1446 | * Set the new slave to the bond's settings and unset them on the old | |
1447 | * curr_active_slave. | |
1448 | * Setting include flags, mc-list, promiscuity, allmulti, etc. | |
1449 | * | |
1450 | * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, | |
1451 | * because it is apparently the best available slave we have, even though its | |
1452 | * updelay hasn't timed out yet. | |
1453 | * | |
1454 | * Warning: Caller must hold curr_slave_lock for writing. | |
1455 | */ | |
1456 | static void bond_change_active_slave(struct bonding *bond, struct slave *new_active) | |
1457 | { | |
1458 | struct slave *old_active = bond->curr_active_slave; | |
1459 | ||
1460 | if (old_active == new_active) { | |
1461 | return; | |
1462 | } | |
1463 | ||
1464 | if (new_active) { | |
1465 | if (new_active->link == BOND_LINK_BACK) { | |
1466 | if (USES_PRIMARY(bond->params.mode)) { | |
1467 | printk(KERN_INFO DRV_NAME | |
1468 | ": %s: making interface %s the new " | |
1469 | "active one %d ms earlier.\n", | |
1470 | bond->dev->name, new_active->dev->name, | |
1471 | (bond->params.updelay - new_active->delay) * bond->params.miimon); | |
1472 | } | |
1473 | ||
1474 | new_active->delay = 0; | |
1475 | new_active->link = BOND_LINK_UP; | |
1476 | new_active->jiffies = jiffies; | |
1477 | ||
1478 | if (bond->params.mode == BOND_MODE_8023AD) { | |
1479 | bond_3ad_handle_link_change(new_active, BOND_LINK_UP); | |
1480 | } | |
1481 | ||
1482 | if ((bond->params.mode == BOND_MODE_TLB) || | |
1483 | (bond->params.mode == BOND_MODE_ALB)) { | |
1484 | bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); | |
1485 | } | |
1486 | } else { | |
1487 | if (USES_PRIMARY(bond->params.mode)) { | |
1488 | printk(KERN_INFO DRV_NAME | |
1489 | ": %s: making interface %s the new " | |
1490 | "active one.\n", | |
1491 | bond->dev->name, new_active->dev->name); | |
1492 | } | |
1493 | } | |
1494 | } | |
1495 | ||
1da177e4 LT |
1496 | if (USES_PRIMARY(bond->params.mode)) { |
1497 | bond_mc_swap(bond, new_active, old_active); | |
1498 | } | |
1499 | ||
1500 | if ((bond->params.mode == BOND_MODE_TLB) || | |
1501 | (bond->params.mode == BOND_MODE_ALB)) { | |
1502 | bond_alb_handle_active_change(bond, new_active); | |
1503 | } else { | |
1504 | bond->curr_active_slave = new_active; | |
1505 | } | |
c3ade5ca JV |
1506 | |
1507 | if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) { | |
1508 | if (old_active) { | |
1509 | bond_set_slave_inactive_flags(old_active); | |
1510 | } | |
1511 | ||
1512 | if (new_active) { | |
1513 | bond_set_slave_active_flags(new_active); | |
1514 | } | |
1515 | bond_send_gratuitous_arp(bond); | |
1516 | } | |
1da177e4 LT |
1517 | } |
1518 | ||
1519 | /** | |
1520 | * bond_select_active_slave - select a new active slave, if needed | |
1521 | * @bond: our bonding struct | |
1522 | * | |
1523 | * This functions shoud be called when one of the following occurs: | |
1524 | * - The old curr_active_slave has been released or lost its link. | |
1525 | * - The primary_slave has got its link back. | |
1526 | * - A slave has got its link back and there's no old curr_active_slave. | |
1527 | * | |
1528 | * Warning: Caller must hold curr_slave_lock for writing. | |
1529 | */ | |
1530 | static void bond_select_active_slave(struct bonding *bond) | |
1531 | { | |
1532 | struct slave *best_slave; | |
1533 | ||
1534 | best_slave = bond_find_best_slave(bond); | |
1535 | if (best_slave != bond->curr_active_slave) { | |
1536 | bond_change_active_slave(bond, best_slave); | |
1537 | } | |
1538 | } | |
1539 | ||
1540 | /*--------------------------- slave list handling ---------------------------*/ | |
1541 | ||
1542 | /* | |
1543 | * This function attaches the slave to the end of list. | |
1544 | * | |
1545 | * bond->lock held for writing by caller. | |
1546 | */ | |
1547 | static void bond_attach_slave(struct bonding *bond, struct slave *new_slave) | |
1548 | { | |
1549 | if (bond->first_slave == NULL) { /* attaching the first slave */ | |
1550 | new_slave->next = new_slave; | |
1551 | new_slave->prev = new_slave; | |
1552 | bond->first_slave = new_slave; | |
1553 | } else { | |
1554 | new_slave->next = bond->first_slave; | |
1555 | new_slave->prev = bond->first_slave->prev; | |
1556 | new_slave->next->prev = new_slave; | |
1557 | new_slave->prev->next = new_slave; | |
1558 | } | |
1559 | ||
1560 | bond->slave_cnt++; | |
1561 | } | |
1562 | ||
1563 | /* | |
1564 | * This function detaches the slave from the list. | |
1565 | * WARNING: no check is made to verify if the slave effectively | |
1566 | * belongs to <bond>. | |
1567 | * Nothing is freed on return, structures are just unchained. | |
1568 | * If any slave pointer in bond was pointing to <slave>, | |
1569 | * it should be changed by the calling function. | |
1570 | * | |
1571 | * bond->lock held for writing by caller. | |
1572 | */ | |
1573 | static void bond_detach_slave(struct bonding *bond, struct slave *slave) | |
1574 | { | |
1575 | if (slave->next) { | |
1576 | slave->next->prev = slave->prev; | |
1577 | } | |
1578 | ||
1579 | if (slave->prev) { | |
1580 | slave->prev->next = slave->next; | |
1581 | } | |
1582 | ||
1583 | if (bond->first_slave == slave) { /* slave is the first slave */ | |
1584 | if (bond->slave_cnt > 1) { /* there are more slave */ | |
1585 | bond->first_slave = slave->next; | |
1586 | } else { | |
1587 | bond->first_slave = NULL; /* slave was the last one */ | |
1588 | } | |
1589 | } | |
1590 | ||
1591 | slave->next = NULL; | |
1592 | slave->prev = NULL; | |
1593 | bond->slave_cnt--; | |
1594 | } | |
1595 | ||
1596 | /*---------------------------------- IOCTL ----------------------------------*/ | |
1597 | ||
1598 | static int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev) | |
1599 | { | |
1600 | dprintk("bond_dev=%p\n", bond_dev); | |
1601 | dprintk("slave_dev=%p\n", slave_dev); | |
1602 | dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len); | |
1603 | memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len); | |
1604 | return 0; | |
1605 | } | |
1606 | ||
8531c5ff AK |
1607 | #define BOND_INTERSECT_FEATURES \ |
1608 | (NETIF_F_SG|NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM) | |
1609 | ||
1610 | /* | |
1611 | * Compute the features available to the bonding device by | |
1612 | * intersection of all of the slave devices' BOND_INTERSECT_FEATURES. | |
1613 | * Call this after attaching or detaching a slave to update the | |
1614 | * bond's features. | |
1615 | */ | |
1616 | static int bond_compute_features(struct bonding *bond) | |
1617 | { | |
1618 | int i; | |
1619 | struct slave *slave; | |
1620 | struct net_device *bond_dev = bond->dev; | |
1621 | int features = bond->bond_features; | |
1622 | ||
1623 | bond_for_each_slave(bond, slave, i) { | |
1624 | struct net_device * slave_dev = slave->dev; | |
1625 | if (i == 0) { | |
1626 | features |= BOND_INTERSECT_FEATURES; | |
1627 | } | |
1628 | features &= | |
1629 | ~(~slave_dev->features & BOND_INTERSECT_FEATURES); | |
1630 | } | |
1631 | ||
1632 | /* turn off NETIF_F_SG if we need a csum and h/w can't do it */ | |
1633 | if ((features & NETIF_F_SG) && | |
1634 | !(features & (NETIF_F_IP_CSUM | | |
1635 | NETIF_F_NO_CSUM | | |
1636 | NETIF_F_HW_CSUM))) { | |
1637 | features &= ~NETIF_F_SG; | |
1638 | } | |
1639 | ||
1640 | bond_dev->features = features; | |
1641 | ||
1642 | return 0; | |
1643 | } | |
1644 | ||
1da177e4 LT |
1645 | /* enslave device <slave> to bond device <master> */ |
1646 | static int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev) | |
1647 | { | |
1648 | struct bonding *bond = bond_dev->priv; | |
1649 | struct slave *new_slave = NULL; | |
1650 | struct dev_mc_list *dmi; | |
1651 | struct sockaddr addr; | |
1652 | int link_reporting; | |
1653 | int old_features = bond_dev->features; | |
1654 | int res = 0; | |
1655 | ||
552709d5 | 1656 | if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL && |
1657 | slave_dev->do_ioctl == NULL) { | |
1da177e4 LT |
1658 | printk(KERN_WARNING DRV_NAME |
1659 | ": Warning : no link monitoring support for %s\n", | |
1660 | slave_dev->name); | |
1661 | } | |
1662 | ||
1663 | /* bond must be initialized by bond_open() before enslaving */ | |
1664 | if (!(bond_dev->flags & IFF_UP)) { | |
1665 | dprintk("Error, master_dev is not up\n"); | |
1666 | return -EPERM; | |
1667 | } | |
1668 | ||
1669 | /* already enslaved */ | |
1670 | if (slave_dev->flags & IFF_SLAVE) { | |
1671 | dprintk("Error, Device was already enslaved\n"); | |
1672 | return -EBUSY; | |
1673 | } | |
1674 | ||
1675 | /* vlan challenged mutual exclusion */ | |
1676 | /* no need to lock since we're protected by rtnl_lock */ | |
1677 | if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { | |
1678 | dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name); | |
1679 | if (!list_empty(&bond->vlan_list)) { | |
1680 | printk(KERN_ERR DRV_NAME | |
1681 | ": Error: cannot enslave VLAN " | |
1682 | "challenged slave %s on VLAN enabled " | |
1683 | "bond %s\n", slave_dev->name, | |
1684 | bond_dev->name); | |
1685 | return -EPERM; | |
1686 | } else { | |
1687 | printk(KERN_WARNING DRV_NAME | |
1688 | ": Warning: enslaved VLAN challenged " | |
1689 | "slave %s. Adding VLANs will be blocked as " | |
1690 | "long as %s is part of bond %s\n", | |
1691 | slave_dev->name, slave_dev->name, | |
1692 | bond_dev->name); | |
1693 | bond_dev->features |= NETIF_F_VLAN_CHALLENGED; | |
1694 | } | |
1695 | } else { | |
1696 | dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name); | |
1697 | if (bond->slave_cnt == 0) { | |
1698 | /* First slave, and it is not VLAN challenged, | |
1699 | * so remove the block of adding VLANs over the bond. | |
1700 | */ | |
1701 | bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED; | |
1702 | } | |
1703 | } | |
1704 | ||
1705 | if (app_abi_ver >= 1) { | |
1706 | /* The application is using an ABI, which requires the | |
1707 | * slave interface to be closed. | |
1708 | */ | |
1709 | if ((slave_dev->flags & IFF_UP)) { | |
1710 | printk(KERN_ERR DRV_NAME | |
1711 | ": Error: %s is up\n", | |
1712 | slave_dev->name); | |
1713 | res = -EPERM; | |
1714 | goto err_undo_flags; | |
1715 | } | |
1716 | ||
1717 | if (slave_dev->set_mac_address == NULL) { | |
1718 | printk(KERN_ERR DRV_NAME | |
1719 | ": Error: The slave device you specified does " | |
1720 | "not support setting the MAC address.\n"); | |
1721 | printk(KERN_ERR | |
1722 | "Your kernel likely does not support slave " | |
1723 | "devices.\n"); | |
1724 | ||
1725 | res = -EOPNOTSUPP; | |
1726 | goto err_undo_flags; | |
1727 | } | |
1728 | } else { | |
1729 | /* The application is not using an ABI, which requires the | |
1730 | * slave interface to be open. | |
1731 | */ | |
1732 | if (!(slave_dev->flags & IFF_UP)) { | |
1733 | printk(KERN_ERR DRV_NAME | |
1734 | ": Error: %s is not running\n", | |
1735 | slave_dev->name); | |
1736 | res = -EINVAL; | |
1737 | goto err_undo_flags; | |
1738 | } | |
1739 | ||
1740 | if ((bond->params.mode == BOND_MODE_8023AD) || | |
1741 | (bond->params.mode == BOND_MODE_TLB) || | |
1742 | (bond->params.mode == BOND_MODE_ALB)) { | |
1743 | printk(KERN_ERR DRV_NAME | |
1744 | ": Error: to use %s mode, you must upgrade " | |
1745 | "ifenslave.\n", | |
1746 | bond_mode_name(bond->params.mode)); | |
1747 | res = -EOPNOTSUPP; | |
1748 | goto err_undo_flags; | |
1749 | } | |
1750 | } | |
1751 | ||
1752 | new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL); | |
1753 | if (!new_slave) { | |
1754 | res = -ENOMEM; | |
1755 | goto err_undo_flags; | |
1756 | } | |
1757 | ||
1758 | memset(new_slave, 0, sizeof(struct slave)); | |
1759 | ||
1760 | /* save slave's original flags before calling | |
1761 | * netdev_set_master and dev_open | |
1762 | */ | |
1763 | new_slave->original_flags = slave_dev->flags; | |
1764 | ||
1765 | if (app_abi_ver >= 1) { | |
1766 | /* save slave's original ("permanent") mac address for | |
1767 | * modes that needs it, and for restoring it upon release, | |
1768 | * and then set it to the master's address | |
1769 | */ | |
1770 | memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN); | |
1771 | ||
1772 | /* set slave to master's mac address | |
1773 | * The application already set the master's | |
1774 | * mac address to that of the first slave | |
1775 | */ | |
1776 | memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len); | |
1777 | addr.sa_family = slave_dev->type; | |
1778 | res = dev_set_mac_address(slave_dev, &addr); | |
1779 | if (res) { | |
1780 | dprintk("Error %d calling set_mac_address\n", res); | |
1781 | goto err_free; | |
1782 | } | |
1783 | ||
1784 | /* open the slave since the application closed it */ | |
1785 | res = dev_open(slave_dev); | |
1786 | if (res) { | |
1787 | dprintk("Openning slave %s failed\n", slave_dev->name); | |
1788 | goto err_restore_mac; | |
1789 | } | |
1790 | } | |
1791 | ||
1792 | res = netdev_set_master(slave_dev, bond_dev); | |
1793 | if (res) { | |
1794 | dprintk("Error %d calling netdev_set_master\n", res); | |
1795 | if (app_abi_ver < 1) { | |
1796 | goto err_free; | |
1797 | } else { | |
1798 | goto err_close; | |
1799 | } | |
1800 | } | |
1801 | ||
1802 | new_slave->dev = slave_dev; | |
1803 | ||
1804 | if ((bond->params.mode == BOND_MODE_TLB) || | |
1805 | (bond->params.mode == BOND_MODE_ALB)) { | |
1806 | /* bond_alb_init_slave() must be called before all other stages since | |
1807 | * it might fail and we do not want to have to undo everything | |
1808 | */ | |
1809 | res = bond_alb_init_slave(bond, new_slave); | |
1810 | if (res) { | |
1811 | goto err_unset_master; | |
1812 | } | |
1813 | } | |
1814 | ||
1815 | /* If the mode USES_PRIMARY, then the new slave gets the | |
1816 | * master's promisc (and mc) settings only if it becomes the | |
1817 | * curr_active_slave, and that is taken care of later when calling | |
1818 | * bond_change_active() | |
1819 | */ | |
1820 | if (!USES_PRIMARY(bond->params.mode)) { | |
1821 | /* set promiscuity level to new slave */ | |
1822 | if (bond_dev->flags & IFF_PROMISC) { | |
1823 | dev_set_promiscuity(slave_dev, 1); | |
1824 | } | |
1825 | ||
1826 | /* set allmulti level to new slave */ | |
1827 | if (bond_dev->flags & IFF_ALLMULTI) { | |
1828 | dev_set_allmulti(slave_dev, 1); | |
1829 | } | |
1830 | ||
1831 | /* upload master's mc_list to new slave */ | |
1832 | for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) { | |
1833 | dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0); | |
1834 | } | |
1835 | } | |
1836 | ||
1837 | if (bond->params.mode == BOND_MODE_8023AD) { | |
1838 | /* add lacpdu mc addr to mc list */ | |
1839 | u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; | |
1840 | ||
1841 | dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0); | |
1842 | } | |
1843 | ||
1844 | bond_add_vlans_on_slave(bond, slave_dev); | |
1845 | ||
1846 | write_lock_bh(&bond->lock); | |
1847 | ||
1848 | bond_attach_slave(bond, new_slave); | |
1849 | ||
1850 | new_slave->delay = 0; | |
1851 | new_slave->link_failure_count = 0; | |
1852 | ||
8531c5ff AK |
1853 | bond_compute_features(bond); |
1854 | ||
1da177e4 LT |
1855 | if (bond->params.miimon && !bond->params.use_carrier) { |
1856 | link_reporting = bond_check_dev_link(bond, slave_dev, 1); | |
1857 | ||
1858 | if ((link_reporting == -1) && !bond->params.arp_interval) { | |
1859 | /* | |
1860 | * miimon is set but a bonded network driver | |
1861 | * does not support ETHTOOL/MII and | |
1862 | * arp_interval is not set. Note: if | |
1863 | * use_carrier is enabled, we will never go | |
1864 | * here (because netif_carrier is always | |
1865 | * supported); thus, we don't need to change | |
1866 | * the messages for netif_carrier. | |
1867 | */ | |
1868 | printk(KERN_WARNING DRV_NAME | |
1869 | ": Warning: MII and ETHTOOL support not " | |
1870 | "available for interface %s, and " | |
1871 | "arp_interval/arp_ip_target module parameters " | |
1872 | "not specified, thus bonding will not detect " | |
1873 | "link failures! see bonding.txt for details.\n", | |
1874 | slave_dev->name); | |
1875 | } else if (link_reporting == -1) { | |
1876 | /* unable get link status using mii/ethtool */ | |
1877 | printk(KERN_WARNING DRV_NAME | |
1878 | ": Warning: can't get link status from " | |
1879 | "interface %s; the network driver associated " | |
1880 | "with this interface does not support MII or " | |
1881 | "ETHTOOL link status reporting, thus miimon " | |
1882 | "has no effect on this interface.\n", | |
1883 | slave_dev->name); | |
1884 | } | |
1885 | } | |
1886 | ||
1887 | /* check for initial state */ | |
1888 | if (!bond->params.miimon || | |
1889 | (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) { | |
1890 | if (bond->params.updelay) { | |
1891 | dprintk("Initial state of slave_dev is " | |
1892 | "BOND_LINK_BACK\n"); | |
1893 | new_slave->link = BOND_LINK_BACK; | |
1894 | new_slave->delay = bond->params.updelay; | |
1895 | } else { | |
1896 | dprintk("Initial state of slave_dev is " | |
1897 | "BOND_LINK_UP\n"); | |
1898 | new_slave->link = BOND_LINK_UP; | |
1899 | } | |
1900 | new_slave->jiffies = jiffies; | |
1901 | } else { | |
1902 | dprintk("Initial state of slave_dev is " | |
1903 | "BOND_LINK_DOWN\n"); | |
1904 | new_slave->link = BOND_LINK_DOWN; | |
1905 | } | |
1906 | ||
1907 | if (bond_update_speed_duplex(new_slave) && | |
1908 | (new_slave->link != BOND_LINK_DOWN)) { | |
1909 | printk(KERN_WARNING DRV_NAME | |
1910 | ": Warning: failed to get speed and duplex from %s, " | |
1911 | "assumed to be 100Mb/sec and Full.\n", | |
1912 | new_slave->dev->name); | |
1913 | ||
1914 | if (bond->params.mode == BOND_MODE_8023AD) { | |
1915 | printk(KERN_WARNING | |
1916 | "Operation of 802.3ad mode requires ETHTOOL " | |
1917 | "support in base driver for proper aggregator " | |
1918 | "selection.\n"); | |
1919 | } | |
1920 | } | |
1921 | ||
1922 | if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) { | |
1923 | /* if there is a primary slave, remember it */ | |
1924 | if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { | |
1925 | bond->primary_slave = new_slave; | |
1926 | } | |
1927 | } | |
1928 | ||
1929 | switch (bond->params.mode) { | |
1930 | case BOND_MODE_ACTIVEBACKUP: | |
1931 | /* if we're in active-backup mode, we need one and only one active | |
1932 | * interface. The backup interfaces will have their NOARP flag set | |
1933 | * because we need them to be completely deaf and not to respond to | |
1934 | * any ARP request on the network to avoid fooling a switch. Thus, | |
1935 | * since we guarantee that curr_active_slave always point to the last | |
1936 | * usable interface, we just have to verify this interface's flag. | |
1937 | */ | |
1938 | if (((!bond->curr_active_slave) || | |
1939 | (bond->curr_active_slave->dev->flags & IFF_NOARP)) && | |
1940 | (new_slave->link != BOND_LINK_DOWN)) { | |
1941 | dprintk("This is the first active slave\n"); | |
1942 | /* first slave or no active slave yet, and this link | |
1943 | is OK, so make this interface the active one */ | |
1944 | bond_change_active_slave(bond, new_slave); | |
1945 | } else { | |
1946 | dprintk("This is just a backup slave\n"); | |
1947 | bond_set_slave_inactive_flags(new_slave); | |
1948 | } | |
1949 | break; | |
1950 | case BOND_MODE_8023AD: | |
1951 | /* in 802.3ad mode, the internal mechanism | |
1952 | * will activate the slaves in the selected | |
1953 | * aggregator | |
1954 | */ | |
1955 | bond_set_slave_inactive_flags(new_slave); | |
1956 | /* if this is the first slave */ | |
1957 | if (bond->slave_cnt == 1) { | |
1958 | SLAVE_AD_INFO(new_slave).id = 1; | |
1959 | /* Initialize AD with the number of times that the AD timer is called in 1 second | |
1960 | * can be called only after the mac address of the bond is set | |
1961 | */ | |
1962 | bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL, | |
1963 | bond->params.lacp_fast); | |
1964 | } else { | |
1965 | SLAVE_AD_INFO(new_slave).id = | |
1966 | SLAVE_AD_INFO(new_slave->prev).id + 1; | |
1967 | } | |
1968 | ||
1969 | bond_3ad_bind_slave(new_slave); | |
1970 | break; | |
1971 | case BOND_MODE_TLB: | |
1972 | case BOND_MODE_ALB: | |
1973 | new_slave->state = BOND_STATE_ACTIVE; | |
1974 | if ((!bond->curr_active_slave) && | |
1975 | (new_slave->link != BOND_LINK_DOWN)) { | |
1976 | /* first slave or no active slave yet, and this link | |
1977 | * is OK, so make this interface the active one | |
1978 | */ | |
1979 | bond_change_active_slave(bond, new_slave); | |
1980 | } | |
1981 | break; | |
1982 | default: | |
1983 | dprintk("This slave is always active in trunk mode\n"); | |
1984 | ||
1985 | /* always active in trunk mode */ | |
1986 | new_slave->state = BOND_STATE_ACTIVE; | |
1987 | ||
1988 | /* In trunking mode there is little meaning to curr_active_slave | |
1989 | * anyway (it holds no special properties of the bond device), | |
1990 | * so we can change it without calling change_active_interface() | |
1991 | */ | |
1992 | if (!bond->curr_active_slave) { | |
1993 | bond->curr_active_slave = new_slave; | |
1994 | } | |
1995 | break; | |
1996 | } /* switch(bond_mode) */ | |
1997 | ||
1998 | write_unlock_bh(&bond->lock); | |
1999 | ||
2000 | if (app_abi_ver < 1) { | |
2001 | /* | |
2002 | * !!! This is to support old versions of ifenslave. | |
2003 | * We can remove this in 2.5 because our ifenslave takes | |
2004 | * care of this for us. | |
2005 | * We check to see if the master has a mac address yet. | |
2006 | * If not, we'll give it the mac address of our slave device. | |
2007 | */ | |
2008 | int ndx = 0; | |
2009 | ||
2010 | for (ndx = 0; ndx < bond_dev->addr_len; ndx++) { | |
2011 | dprintk("Checking ndx=%d of bond_dev->dev_addr\n", | |
2012 | ndx); | |
2013 | if (bond_dev->dev_addr[ndx] != 0) { | |
2014 | dprintk("Found non-zero byte at ndx=%d\n", | |
2015 | ndx); | |
2016 | break; | |
2017 | } | |
2018 | } | |
2019 | ||
2020 | if (ndx == bond_dev->addr_len) { | |
2021 | /* | |
2022 | * We got all the way through the address and it was | |
2023 | * all 0's. | |
2024 | */ | |
2025 | dprintk("%s doesn't have a MAC address yet. \n", | |
2026 | bond_dev->name); | |
2027 | dprintk("Going to give assign it from %s.\n", | |
2028 | slave_dev->name); | |
2029 | bond_sethwaddr(bond_dev, slave_dev); | |
2030 | } | |
2031 | } | |
2032 | ||
2033 | printk(KERN_INFO DRV_NAME | |
2034 | ": %s: enslaving %s as a%s interface with a%s link.\n", | |
2035 | bond_dev->name, slave_dev->name, | |
2036 | new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup", | |
2037 | new_slave->link != BOND_LINK_DOWN ? "n up" : " down"); | |
2038 | ||
2039 | /* enslave is successful */ | |
2040 | return 0; | |
2041 | ||
2042 | /* Undo stages on error */ | |
2043 | err_unset_master: | |
2044 | netdev_set_master(slave_dev, NULL); | |
2045 | ||
2046 | err_close: | |
2047 | dev_close(slave_dev); | |
2048 | ||
2049 | err_restore_mac: | |
2050 | memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN); | |
2051 | addr.sa_family = slave_dev->type; | |
2052 | dev_set_mac_address(slave_dev, &addr); | |
2053 | ||
2054 | err_free: | |
2055 | kfree(new_slave); | |
2056 | ||
2057 | err_undo_flags: | |
2058 | bond_dev->features = old_features; | |
8531c5ff | 2059 | |
1da177e4 LT |
2060 | return res; |
2061 | } | |
2062 | ||
2063 | /* | |
2064 | * Try to release the slave device <slave> from the bond device <master> | |
2065 | * It is legal to access curr_active_slave without a lock because all the function | |
2066 | * is write-locked. | |
2067 | * | |
2068 | * The rules for slave state should be: | |
2069 | * for Active/Backup: | |
2070 | * Active stays on all backups go down | |
2071 | * for Bonded connections: | |
2072 | * The first up interface should be left on and all others downed. | |
2073 | */ | |
2074 | static int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) | |
2075 | { | |
2076 | struct bonding *bond = bond_dev->priv; | |
2077 | struct slave *slave, *oldcurrent; | |
2078 | struct sockaddr addr; | |
2079 | int mac_addr_differ; | |
2080 | ||
2081 | /* slave is not a slave or master is not master of this slave */ | |
2082 | if (!(slave_dev->flags & IFF_SLAVE) || | |
2083 | (slave_dev->master != bond_dev)) { | |
2084 | printk(KERN_ERR DRV_NAME | |
2085 | ": Error: %s: cannot release %s.\n", | |
2086 | bond_dev->name, slave_dev->name); | |
2087 | return -EINVAL; | |
2088 | } | |
2089 | ||
2090 | write_lock_bh(&bond->lock); | |
2091 | ||
2092 | slave = bond_get_slave_by_dev(bond, slave_dev); | |
2093 | if (!slave) { | |
2094 | /* not a slave of this bond */ | |
2095 | printk(KERN_INFO DRV_NAME | |
2096 | ": %s: %s not enslaved\n", | |
2097 | bond_dev->name, slave_dev->name); | |
2098 | return -EINVAL; | |
2099 | } | |
2100 | ||
2101 | mac_addr_differ = memcmp(bond_dev->dev_addr, | |
2102 | slave->perm_hwaddr, | |
2103 | ETH_ALEN); | |
2104 | if (!mac_addr_differ && (bond->slave_cnt > 1)) { | |
2105 | printk(KERN_WARNING DRV_NAME | |
2106 | ": Warning: the permanent HWaddr of %s " | |
2107 | "- %02X:%02X:%02X:%02X:%02X:%02X - is " | |
2108 | "still in use by %s. Set the HWaddr of " | |
2109 | "%s to a different address to avoid " | |
2110 | "conflicts.\n", | |
2111 | slave_dev->name, | |
2112 | slave->perm_hwaddr[0], | |
2113 | slave->perm_hwaddr[1], | |
2114 | slave->perm_hwaddr[2], | |
2115 | slave->perm_hwaddr[3], | |
2116 | slave->perm_hwaddr[4], | |
2117 | slave->perm_hwaddr[5], | |
2118 | bond_dev->name, | |
2119 | slave_dev->name); | |
2120 | } | |
2121 | ||
2122 | /* Inform AD package of unbinding of slave. */ | |
2123 | if (bond->params.mode == BOND_MODE_8023AD) { | |
2124 | /* must be called before the slave is | |
2125 | * detached from the list | |
2126 | */ | |
2127 | bond_3ad_unbind_slave(slave); | |
2128 | } | |
2129 | ||
2130 | printk(KERN_INFO DRV_NAME | |
2131 | ": %s: releasing %s interface %s\n", | |
2132 | bond_dev->name, | |
2133 | (slave->state == BOND_STATE_ACTIVE) | |
2134 | ? "active" : "backup", | |
2135 | slave_dev->name); | |
2136 | ||
2137 | oldcurrent = bond->curr_active_slave; | |
2138 | ||
2139 | bond->current_arp_slave = NULL; | |
2140 | ||
2141 | /* release the slave from its bond */ | |
2142 | bond_detach_slave(bond, slave); | |
2143 | ||
8531c5ff AK |
2144 | bond_compute_features(bond); |
2145 | ||
1da177e4 LT |
2146 | if (bond->primary_slave == slave) { |
2147 | bond->primary_slave = NULL; | |
2148 | } | |
2149 | ||
2150 | if (oldcurrent == slave) { | |
2151 | bond_change_active_slave(bond, NULL); | |
2152 | } | |
2153 | ||
2154 | if ((bond->params.mode == BOND_MODE_TLB) || | |
2155 | (bond->params.mode == BOND_MODE_ALB)) { | |
2156 | /* Must be called only after the slave has been | |
2157 | * detached from the list and the curr_active_slave | |
2158 | * has been cleared (if our_slave == old_current), | |
2159 | * but before a new active slave is selected. | |
2160 | */ | |
2161 | bond_alb_deinit_slave(bond, slave); | |
2162 | } | |
2163 | ||
2164 | if (oldcurrent == slave) { | |
2165 | bond_select_active_slave(bond); | |
2166 | ||
2167 | if (!bond->curr_active_slave) { | |
2168 | printk(KERN_INFO DRV_NAME | |
2169 | ": %s: now running without any active " | |
2170 | "interface !\n", | |
2171 | bond_dev->name); | |
2172 | } | |
2173 | } | |
2174 | ||
2175 | if (bond->slave_cnt == 0) { | |
2176 | /* if the last slave was removed, zero the mac address | |
2177 | * of the master so it will be set by the application | |
2178 | * to the mac address of the first slave | |
2179 | */ | |
2180 | memset(bond_dev->dev_addr, 0, bond_dev->addr_len); | |
2181 | ||
2182 | if (list_empty(&bond->vlan_list)) { | |
2183 | bond_dev->features |= NETIF_F_VLAN_CHALLENGED; | |
2184 | } else { | |
2185 | printk(KERN_WARNING DRV_NAME | |
2186 | ": Warning: clearing HW address of %s while it " | |
2187 | "still has VLANs.\n", | |
2188 | bond_dev->name); | |
2189 | printk(KERN_WARNING DRV_NAME | |
2190 | ": When re-adding slaves, make sure the bond's " | |
2191 | "HW address matches its VLANs'.\n"); | |
2192 | } | |
2193 | } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) && | |
2194 | !bond_has_challenged_slaves(bond)) { | |
2195 | printk(KERN_INFO DRV_NAME | |
2196 | ": last VLAN challenged slave %s " | |
2197 | "left bond %s. VLAN blocking is removed\n", | |
2198 | slave_dev->name, bond_dev->name); | |
2199 | bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED; | |
2200 | } | |
2201 | ||
2202 | write_unlock_bh(&bond->lock); | |
2203 | ||
2204 | bond_del_vlans_from_slave(bond, slave_dev); | |
2205 | ||
2206 | /* If the mode USES_PRIMARY, then we should only remove its | |
2207 | * promisc and mc settings if it was the curr_active_slave, but that was | |
2208 | * already taken care of above when we detached the slave | |
2209 | */ | |
2210 | if (!USES_PRIMARY(bond->params.mode)) { | |
2211 | /* unset promiscuity level from slave */ | |
2212 | if (bond_dev->flags & IFF_PROMISC) { | |
2213 | dev_set_promiscuity(slave_dev, -1); | |
2214 | } | |
2215 | ||
2216 | /* unset allmulti level from slave */ | |
2217 | if (bond_dev->flags & IFF_ALLMULTI) { | |
2218 | dev_set_allmulti(slave_dev, -1); | |
2219 | } | |
2220 | ||
2221 | /* flush master's mc_list from slave */ | |
2222 | bond_mc_list_flush(bond_dev, slave_dev); | |
2223 | } | |
2224 | ||
2225 | netdev_set_master(slave_dev, NULL); | |
2226 | ||
2227 | /* close slave before restoring its mac address */ | |
2228 | dev_close(slave_dev); | |
2229 | ||
2230 | if (app_abi_ver >= 1) { | |
2231 | /* restore original ("permanent") mac address */ | |
2232 | memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN); | |
2233 | addr.sa_family = slave_dev->type; | |
2234 | dev_set_mac_address(slave_dev, &addr); | |
2235 | } | |
2236 | ||
2237 | /* restore the original state of the | |
2238 | * IFF_NOARP flag that might have been | |
2239 | * set by bond_set_slave_inactive_flags() | |
2240 | */ | |
2241 | if ((slave->original_flags & IFF_NOARP) == 0) { | |
2242 | slave_dev->flags &= ~IFF_NOARP; | |
2243 | } | |
2244 | ||
2245 | kfree(slave); | |
2246 | ||
2247 | return 0; /* deletion OK */ | |
2248 | } | |
2249 | ||
2250 | /* | |
2251 | * This function releases all slaves. | |
2252 | */ | |
2253 | static int bond_release_all(struct net_device *bond_dev) | |
2254 | { | |
2255 | struct bonding *bond = bond_dev->priv; | |
2256 | struct slave *slave; | |
2257 | struct net_device *slave_dev; | |
2258 | struct sockaddr addr; | |
2259 | ||
2260 | write_lock_bh(&bond->lock); | |
2261 | ||
2262 | if (bond->slave_cnt == 0) { | |
2263 | goto out; | |
2264 | } | |
2265 | ||
2266 | bond->current_arp_slave = NULL; | |
2267 | bond->primary_slave = NULL; | |
2268 | bond_change_active_slave(bond, NULL); | |
2269 | ||
2270 | while ((slave = bond->first_slave) != NULL) { | |
2271 | /* Inform AD package of unbinding of slave | |
2272 | * before slave is detached from the list. | |
2273 | */ | |
2274 | if (bond->params.mode == BOND_MODE_8023AD) { | |
2275 | bond_3ad_unbind_slave(slave); | |
2276 | } | |
2277 | ||
2278 | slave_dev = slave->dev; | |
2279 | bond_detach_slave(bond, slave); | |
2280 | ||
2281 | if ((bond->params.mode == BOND_MODE_TLB) || | |
2282 | (bond->params.mode == BOND_MODE_ALB)) { | |
2283 | /* must be called only after the slave | |
2284 | * has been detached from the list | |
2285 | */ | |
2286 | bond_alb_deinit_slave(bond, slave); | |
2287 | } | |
2288 | ||
8531c5ff AK |
2289 | bond_compute_features(bond); |
2290 | ||
1da177e4 LT |
2291 | /* now that the slave is detached, unlock and perform |
2292 | * all the undo steps that should not be called from | |
2293 | * within a lock. | |
2294 | */ | |
2295 | write_unlock_bh(&bond->lock); | |
2296 | ||
2297 | bond_del_vlans_from_slave(bond, slave_dev); | |
2298 | ||
2299 | /* If the mode USES_PRIMARY, then we should only remove its | |
2300 | * promisc and mc settings if it was the curr_active_slave, but that was | |
2301 | * already taken care of above when we detached the slave | |
2302 | */ | |
2303 | if (!USES_PRIMARY(bond->params.mode)) { | |
2304 | /* unset promiscuity level from slave */ | |
2305 | if (bond_dev->flags & IFF_PROMISC) { | |
2306 | dev_set_promiscuity(slave_dev, -1); | |
2307 | } | |
2308 | ||
2309 | /* unset allmulti level from slave */ | |
2310 | if (bond_dev->flags & IFF_ALLMULTI) { | |
2311 | dev_set_allmulti(slave_dev, -1); | |
2312 | } | |
2313 | ||
2314 | /* flush master's mc_list from slave */ | |
2315 | bond_mc_list_flush(bond_dev, slave_dev); | |
2316 | } | |
2317 | ||
2318 | netdev_set_master(slave_dev, NULL); | |
2319 | ||
2320 | /* close slave before restoring its mac address */ | |
2321 | dev_close(slave_dev); | |
2322 | ||
2323 | if (app_abi_ver >= 1) { | |
2324 | /* restore original ("permanent") mac address*/ | |
2325 | memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN); | |
2326 | addr.sa_family = slave_dev->type; | |
2327 | dev_set_mac_address(slave_dev, &addr); | |
2328 | } | |
2329 | ||
2330 | /* restore the original state of the IFF_NOARP flag that might have | |
2331 | * been set by bond_set_slave_inactive_flags() | |
2332 | */ | |
2333 | if ((slave->original_flags & IFF_NOARP) == 0) { | |
2334 | slave_dev->flags &= ~IFF_NOARP; | |
2335 | } | |
2336 | ||
2337 | kfree(slave); | |
2338 | ||
2339 | /* re-acquire the lock before getting the next slave */ | |
2340 | write_lock_bh(&bond->lock); | |
2341 | } | |
2342 | ||
2343 | /* zero the mac address of the master so it will be | |
2344 | * set by the application to the mac address of the | |
2345 | * first slave | |
2346 | */ | |
2347 | memset(bond_dev->dev_addr, 0, bond_dev->addr_len); | |
2348 | ||
2349 | if (list_empty(&bond->vlan_list)) { | |
2350 | bond_dev->features |= NETIF_F_VLAN_CHALLENGED; | |
2351 | } else { | |
2352 | printk(KERN_WARNING DRV_NAME | |
2353 | ": Warning: clearing HW address of %s while it " | |
2354 | "still has VLANs.\n", | |
2355 | bond_dev->name); | |
2356 | printk(KERN_WARNING DRV_NAME | |
2357 | ": When re-adding slaves, make sure the bond's " | |
2358 | "HW address matches its VLANs'.\n"); | |
2359 | } | |
2360 | ||
2361 | printk(KERN_INFO DRV_NAME | |
2362 | ": %s: released all slaves\n", | |
2363 | bond_dev->name); | |
2364 | ||
2365 | out: | |
2366 | write_unlock_bh(&bond->lock); | |
2367 | ||
2368 | return 0; | |
2369 | } | |
2370 | ||
2371 | /* | |
2372 | * This function changes the active slave to slave <slave_dev>. | |
2373 | * It returns -EINVAL in the following cases. | |
2374 | * - <slave_dev> is not found in the list. | |
2375 | * - There is not active slave now. | |
2376 | * - <slave_dev> is already active. | |
2377 | * - The link state of <slave_dev> is not BOND_LINK_UP. | |
2378 | * - <slave_dev> is not running. | |
2379 | * In these cases, this fuction does nothing. | |
2380 | * In the other cases, currnt_slave pointer is changed and 0 is returned. | |
2381 | */ | |
2382 | static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev) | |
2383 | { | |
2384 | struct bonding *bond = bond_dev->priv; | |
2385 | struct slave *old_active = NULL; | |
2386 | struct slave *new_active = NULL; | |
2387 | int res = 0; | |
2388 | ||
2389 | if (!USES_PRIMARY(bond->params.mode)) { | |
2390 | return -EINVAL; | |
2391 | } | |
2392 | ||
2393 | /* Verify that master_dev is indeed the master of slave_dev */ | |
2394 | if (!(slave_dev->flags & IFF_SLAVE) || | |
2395 | (slave_dev->master != bond_dev)) { | |
2396 | return -EINVAL; | |
2397 | } | |
2398 | ||
2399 | write_lock_bh(&bond->lock); | |
2400 | ||
2401 | old_active = bond->curr_active_slave; | |
2402 | new_active = bond_get_slave_by_dev(bond, slave_dev); | |
2403 | ||
2404 | /* | |
2405 | * Changing to the current active: do nothing; return success. | |
2406 | */ | |
2407 | if (new_active && (new_active == old_active)) { | |
2408 | write_unlock_bh(&bond->lock); | |
2409 | return 0; | |
2410 | } | |
2411 | ||
2412 | if ((new_active) && | |
2413 | (old_active) && | |
2414 | (new_active->link == BOND_LINK_UP) && | |
2415 | IS_UP(new_active->dev)) { | |
2416 | bond_change_active_slave(bond, new_active); | |
2417 | } else { | |
2418 | res = -EINVAL; | |
2419 | } | |
2420 | ||
2421 | write_unlock_bh(&bond->lock); | |
2422 | ||
2423 | return res; | |
2424 | } | |
2425 | ||
2426 | static int bond_ethtool_ioctl(struct net_device *bond_dev, struct ifreq *ifr) | |
2427 | { | |
2428 | struct ethtool_drvinfo info; | |
2429 | void __user *addr = ifr->ifr_data; | |
2430 | uint32_t cmd; | |
2431 | ||
2432 | if (get_user(cmd, (uint32_t __user *)addr)) { | |
2433 | return -EFAULT; | |
2434 | } | |
2435 | ||
2436 | switch (cmd) { | |
2437 | case ETHTOOL_GDRVINFO: | |
2438 | if (copy_from_user(&info, addr, sizeof(info))) { | |
2439 | return -EFAULT; | |
2440 | } | |
2441 | ||
2442 | if (strcmp(info.driver, "ifenslave") == 0) { | |
2443 | int new_abi_ver; | |
2444 | char *endptr; | |
2445 | ||
2446 | new_abi_ver = simple_strtoul(info.fw_version, | |
2447 | &endptr, 0); | |
2448 | if (*endptr) { | |
2449 | printk(KERN_ERR DRV_NAME | |
2450 | ": Error: got invalid ABI " | |
2451 | "version from application\n"); | |
2452 | ||
2453 | return -EINVAL; | |
2454 | } | |
2455 | ||
2456 | if (orig_app_abi_ver == -1) { | |
2457 | orig_app_abi_ver = new_abi_ver; | |
2458 | } | |
2459 | ||
2460 | app_abi_ver = new_abi_ver; | |
2461 | } | |
2462 | ||
2463 | strncpy(info.driver, DRV_NAME, 32); | |
2464 | strncpy(info.version, DRV_VERSION, 32); | |
2465 | snprintf(info.fw_version, 32, "%d", BOND_ABI_VERSION); | |
2466 | ||
2467 | if (copy_to_user(addr, &info, sizeof(info))) { | |
2468 | return -EFAULT; | |
2469 | } | |
2470 | ||
2471 | return 0; | |
2472 | default: | |
2473 | return -EOPNOTSUPP; | |
2474 | } | |
2475 | } | |
2476 | ||
2477 | static int bond_info_query(struct net_device *bond_dev, struct ifbond *info) | |
2478 | { | |
2479 | struct bonding *bond = bond_dev->priv; | |
2480 | ||
2481 | info->bond_mode = bond->params.mode; | |
2482 | info->miimon = bond->params.miimon; | |
2483 | ||
2484 | read_lock_bh(&bond->lock); | |
2485 | info->num_slaves = bond->slave_cnt; | |
2486 | read_unlock_bh(&bond->lock); | |
2487 | ||
2488 | return 0; | |
2489 | } | |
2490 | ||
2491 | static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) | |
2492 | { | |
2493 | struct bonding *bond = bond_dev->priv; | |
2494 | struct slave *slave; | |
2495 | int i, found = 0; | |
2496 | ||
2497 | if (info->slave_id < 0) { | |
2498 | return -ENODEV; | |
2499 | } | |
2500 | ||
2501 | read_lock_bh(&bond->lock); | |
2502 | ||
2503 | bond_for_each_slave(bond, slave, i) { | |
2504 | if (i == (int)info->slave_id) { | |
2505 | found = 1; | |
2506 | break; | |
2507 | } | |
2508 | } | |
2509 | ||
2510 | read_unlock_bh(&bond->lock); | |
2511 | ||
2512 | if (found) { | |
2513 | strcpy(info->slave_name, slave->dev->name); | |
2514 | info->link = slave->link; | |
2515 | info->state = slave->state; | |
2516 | info->link_failure_count = slave->link_failure_count; | |
2517 | } else { | |
2518 | return -ENODEV; | |
2519 | } | |
2520 | ||
2521 | return 0; | |
2522 | } | |
2523 | ||
2524 | /*-------------------------------- Monitoring -------------------------------*/ | |
2525 | ||
2526 | /* this function is called regularly to monitor each slave's link. */ | |
2527 | static void bond_mii_monitor(struct net_device *bond_dev) | |
2528 | { | |
2529 | struct bonding *bond = bond_dev->priv; | |
2530 | struct slave *slave, *oldcurrent; | |
2531 | int do_failover = 0; | |
2532 | int delta_in_ticks; | |
2533 | int i; | |
2534 | ||
2535 | read_lock(&bond->lock); | |
2536 | ||
2537 | delta_in_ticks = (bond->params.miimon * HZ) / 1000; | |
2538 | ||
2539 | if (bond->kill_timers) { | |
2540 | goto out; | |
2541 | } | |
2542 | ||
2543 | if (bond->slave_cnt == 0) { | |
2544 | goto re_arm; | |
2545 | } | |
2546 | ||
2547 | /* we will try to read the link status of each of our slaves, and | |
2548 | * set their IFF_RUNNING flag appropriately. For each slave not | |
2549 | * supporting MII status, we won't do anything so that a user-space | |
2550 | * program could monitor the link itself if needed. | |
2551 | */ | |
2552 | ||
2553 | read_lock(&bond->curr_slave_lock); | |
2554 | oldcurrent = bond->curr_active_slave; | |
2555 | read_unlock(&bond->curr_slave_lock); | |
2556 | ||
2557 | bond_for_each_slave(bond, slave, i) { | |
2558 | struct net_device *slave_dev = slave->dev; | |
2559 | int link_state; | |
2560 | u16 old_speed = slave->speed; | |
2561 | u8 old_duplex = slave->duplex; | |
2562 | ||
2563 | link_state = bond_check_dev_link(bond, slave_dev, 0); | |
2564 | ||
2565 | switch (slave->link) { | |
2566 | case BOND_LINK_UP: /* the link was up */ | |
2567 | if (link_state == BMSR_LSTATUS) { | |
2568 | /* link stays up, nothing more to do */ | |
2569 | break; | |
2570 | } else { /* link going down */ | |
2571 | slave->link = BOND_LINK_FAIL; | |
2572 | slave->delay = bond->params.downdelay; | |
2573 | ||
2574 | if (slave->link_failure_count < UINT_MAX) { | |
2575 | slave->link_failure_count++; | |
2576 | } | |
2577 | ||
2578 | if (bond->params.downdelay) { | |
2579 | printk(KERN_INFO DRV_NAME | |
2580 | ": %s: link status down for %s " | |
2581 | "interface %s, disabling it in " | |
2582 | "%d ms.\n", | |
2583 | bond_dev->name, | |
2584 | IS_UP(slave_dev) | |
2585 | ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) | |
2586 | ? ((slave == oldcurrent) | |
2587 | ? "active " : "backup ") | |
2588 | : "") | |
2589 | : "idle ", | |
2590 | slave_dev->name, | |
2591 | bond->params.downdelay * bond->params.miimon); | |
2592 | } | |
2593 | } | |
2594 | /* no break ! fall through the BOND_LINK_FAIL test to | |
2595 | ensure proper action to be taken | |
2596 | */ | |
2597 | case BOND_LINK_FAIL: /* the link has just gone down */ | |
2598 | if (link_state != BMSR_LSTATUS) { | |
2599 | /* link stays down */ | |
2600 | if (slave->delay <= 0) { | |
2601 | /* link down for too long time */ | |
2602 | slave->link = BOND_LINK_DOWN; | |
2603 | ||
2604 | /* in active/backup mode, we must | |
2605 | * completely disable this interface | |
2606 | */ | |
2607 | if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) || | |
2608 | (bond->params.mode == BOND_MODE_8023AD)) { | |
2609 | bond_set_slave_inactive_flags(slave); | |
2610 | } | |
2611 | ||
2612 | printk(KERN_INFO DRV_NAME | |
2613 | ": %s: link status definitely " | |
2614 | "down for interface %s, " | |
2615 | "disabling it\n", | |
2616 | bond_dev->name, | |
2617 | slave_dev->name); | |
2618 | ||
2619 | /* notify ad that the link status has changed */ | |
2620 | if (bond->params.mode == BOND_MODE_8023AD) { | |
2621 | bond_3ad_handle_link_change(slave, BOND_LINK_DOWN); | |
2622 | } | |
2623 | ||
2624 | if ((bond->params.mode == BOND_MODE_TLB) || | |
2625 | (bond->params.mode == BOND_MODE_ALB)) { | |
2626 | bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN); | |
2627 | } | |
2628 | ||
2629 | if (slave == oldcurrent) { | |
2630 | do_failover = 1; | |
2631 | } | |
2632 | } else { | |
2633 | slave->delay--; | |
2634 | } | |
2635 | } else { | |
2636 | /* link up again */ | |
2637 | slave->link = BOND_LINK_UP; | |
2638 | slave->jiffies = jiffies; | |
2639 | printk(KERN_INFO DRV_NAME | |
2640 | ": %s: link status up again after %d " | |
2641 | "ms for interface %s.\n", | |
2642 | bond_dev->name, | |
2643 | (bond->params.downdelay - slave->delay) * bond->params.miimon, | |
2644 | slave_dev->name); | |
2645 | } | |
2646 | break; | |
2647 | case BOND_LINK_DOWN: /* the link was down */ | |
2648 | if (link_state != BMSR_LSTATUS) { | |
2649 | /* the link stays down, nothing more to do */ | |
2650 | break; | |
2651 | } else { /* link going up */ | |
2652 | slave->link = BOND_LINK_BACK; | |
2653 | slave->delay = bond->params.updelay; | |
2654 | ||
2655 | if (bond->params.updelay) { | |
2656 | /* if updelay == 0, no need to | |
2657 | advertise about a 0 ms delay */ | |
2658 | printk(KERN_INFO DRV_NAME | |
2659 | ": %s: link status up for " | |
2660 | "interface %s, enabling it " | |
2661 | "in %d ms.\n", | |
2662 | bond_dev->name, | |
2663 | slave_dev->name, | |
2664 | bond->params.updelay * bond->params.miimon); | |
2665 | } | |
2666 | } | |
2667 | /* no break ! fall through the BOND_LINK_BACK state in | |
2668 | case there's something to do. | |
2669 | */ | |
2670 | case BOND_LINK_BACK: /* the link has just come back */ | |
2671 | if (link_state != BMSR_LSTATUS) { | |
2672 | /* link down again */ | |
2673 | slave->link = BOND_LINK_DOWN; | |
2674 | ||
2675 | printk(KERN_INFO DRV_NAME | |
2676 | ": %s: link status down again after %d " | |
2677 | "ms for interface %s.\n", | |
2678 | bond_dev->name, | |
2679 | (bond->params.updelay - slave->delay) * bond->params.miimon, | |
2680 | slave_dev->name); | |
2681 | } else { | |
2682 | /* link stays up */ | |
2683 | if (slave->delay == 0) { | |
2684 | /* now the link has been up for long time enough */ | |
2685 | slave->link = BOND_LINK_UP; | |
2686 | slave->jiffies = jiffies; | |
2687 | ||
2688 | if (bond->params.mode == BOND_MODE_8023AD) { | |
2689 | /* prevent it from being the active one */ | |
2690 | slave->state = BOND_STATE_BACKUP; | |
2691 | } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) { | |
2692 | /* make it immediately active */ | |
2693 | slave->state = BOND_STATE_ACTIVE; | |
2694 | } else if (slave != bond->primary_slave) { | |
2695 | /* prevent it from being the active one */ | |
2696 | slave->state = BOND_STATE_BACKUP; | |
2697 | } | |
2698 | ||
2699 | printk(KERN_INFO DRV_NAME | |
2700 | ": %s: link status definitely " | |
2701 | "up for interface %s.\n", | |
2702 | bond_dev->name, | |
2703 | slave_dev->name); | |
2704 | ||
2705 | /* notify ad that the link status has changed */ | |
2706 | if (bond->params.mode == BOND_MODE_8023AD) { | |
2707 | bond_3ad_handle_link_change(slave, BOND_LINK_UP); | |
2708 | } | |
2709 | ||
2710 | if ((bond->params.mode == BOND_MODE_TLB) || | |
2711 | (bond->params.mode == BOND_MODE_ALB)) { | |
2712 | bond_alb_handle_link_change(bond, slave, BOND_LINK_UP); | |
2713 | } | |
2714 | ||
2715 | if ((!oldcurrent) || | |
2716 | (slave == bond->primary_slave)) { | |
2717 | do_failover = 1; | |
2718 | } | |
2719 | } else { | |
2720 | slave->delay--; | |
2721 | } | |
2722 | } | |
2723 | break; | |
2724 | default: | |
2725 | /* Should not happen */ | |
2726 | printk(KERN_ERR "bonding: Error: %s Illegal value (link=%d)\n", | |
2727 | slave->dev->name, slave->link); | |
2728 | goto out; | |
2729 | } /* end of switch (slave->link) */ | |
2730 | ||
2731 | bond_update_speed_duplex(slave); | |
2732 | ||
2733 | if (bond->params.mode == BOND_MODE_8023AD) { | |
2734 | if (old_speed != slave->speed) { | |
2735 | bond_3ad_adapter_speed_changed(slave); | |
2736 | } | |
2737 | ||
2738 | if (old_duplex != slave->duplex) { | |
2739 | bond_3ad_adapter_duplex_changed(slave); | |
2740 | } | |
2741 | } | |
2742 | ||
2743 | } /* end of for */ | |
2744 | ||
2745 | if (do_failover) { | |
2746 | write_lock(&bond->curr_slave_lock); | |
2747 | ||
2748 | bond_select_active_slave(bond); | |
2749 | ||
2750 | if (oldcurrent && !bond->curr_active_slave) { | |
2751 | printk(KERN_INFO DRV_NAME | |
2752 | ": %s: now running without any active " | |
2753 | "interface !\n", | |
2754 | bond_dev->name); | |
2755 | } | |
2756 | ||
2757 | write_unlock(&bond->curr_slave_lock); | |
2758 | } | |
2759 | ||
2760 | re_arm: | |
2761 | if (bond->params.miimon) { | |
2762 | mod_timer(&bond->mii_timer, jiffies + delta_in_ticks); | |
2763 | } | |
2764 | out: | |
2765 | read_unlock(&bond->lock); | |
2766 | } | |
2767 | ||
c3ade5ca JV |
2768 | |
2769 | static u32 bond_glean_dev_ip(struct net_device *dev) | |
2770 | { | |
2771 | struct in_device *idev; | |
2772 | struct in_ifaddr *ifa; | |
2773 | u32 addr = 0; | |
2774 | ||
2775 | if (!dev) | |
2776 | return 0; | |
2777 | ||
2778 | rcu_read_lock(); | |
e5ed6399 | 2779 | idev = __in_dev_get_rcu(dev); |
c3ade5ca JV |
2780 | if (!idev) |
2781 | goto out; | |
2782 | ||
2783 | ifa = idev->ifa_list; | |
2784 | if (!ifa) | |
2785 | goto out; | |
2786 | ||
2787 | addr = ifa->ifa_local; | |
2788 | out: | |
2789 | rcu_read_unlock(); | |
2790 | return addr; | |
2791 | } | |
2792 | ||
2793 | static int bond_has_ip(struct bonding *bond) | |
2794 | { | |
2795 | struct vlan_entry *vlan, *vlan_next; | |
2796 | ||
2797 | if (bond->master_ip) | |
2798 | return 1; | |
2799 | ||
2800 | if (list_empty(&bond->vlan_list)) | |
2801 | return 0; | |
2802 | ||
2803 | list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list, | |
2804 | vlan_list) { | |
2805 | if (vlan->vlan_ip) | |
2806 | return 1; | |
2807 | } | |
2808 | ||
2809 | return 0; | |
2810 | } | |
2811 | ||
2812 | /* | |
2813 | * We go to the (large) trouble of VLAN tagging ARP frames because | |
2814 | * switches in VLAN mode (especially if ports are configured as | |
2815 | * "native" to a VLAN) might not pass non-tagged frames. | |
2816 | */ | |
2817 | static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id) | |
2818 | { | |
2819 | struct sk_buff *skb; | |
2820 | ||
2821 | dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op, | |
2822 | slave_dev->name, dest_ip, src_ip, vlan_id); | |
2823 | ||
2824 | skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, | |
2825 | NULL, slave_dev->dev_addr, NULL); | |
2826 | ||
2827 | if (!skb) { | |
2828 | printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n"); | |
2829 | return; | |
2830 | } | |
2831 | if (vlan_id) { | |
2832 | skb = vlan_put_tag(skb, vlan_id); | |
2833 | if (!skb) { | |
2834 | printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n"); | |
2835 | return; | |
2836 | } | |
2837 | } | |
2838 | arp_xmit(skb); | |
2839 | } | |
2840 | ||
2841 | ||
1da177e4 LT |
2842 | static void bond_arp_send_all(struct bonding *bond, struct slave *slave) |
2843 | { | |
c3ade5ca | 2844 | int i, vlan_id, rv; |
1da177e4 | 2845 | u32 *targets = bond->params.arp_targets; |
c3ade5ca JV |
2846 | struct vlan_entry *vlan, *vlan_next; |
2847 | struct net_device *vlan_dev; | |
2848 | struct flowi fl; | |
2849 | struct rtable *rt; | |
1da177e4 LT |
2850 | |
2851 | for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) { | |
c3ade5ca JV |
2852 | dprintk("basa: target %x\n", targets[i]); |
2853 | if (list_empty(&bond->vlan_list)) { | |
2854 | dprintk("basa: empty vlan: arp_send\n"); | |
2855 | bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], | |
2856 | bond->master_ip, 0); | |
2857 | continue; | |
2858 | } | |
2859 | ||
2860 | /* | |
2861 | * If VLANs are configured, we do a route lookup to | |
2862 | * determine which VLAN interface would be used, so we | |
2863 | * can tag the ARP with the proper VLAN tag. | |
2864 | */ | |
2865 | memset(&fl, 0, sizeof(fl)); | |
2866 | fl.fl4_dst = targets[i]; | |
2867 | fl.fl4_tos = RTO_ONLINK; | |
2868 | ||
2869 | rv = ip_route_output_key(&rt, &fl); | |
2870 | if (rv) { | |
2871 | if (net_ratelimit()) { | |
2872 | printk(KERN_WARNING DRV_NAME | |
2873 | ": %s: no route to arp_ip_target %u.%u.%u.%u\n", | |
2874 | bond->dev->name, NIPQUAD(fl.fl4_dst)); | |
2875 | } | |
2876 | continue; | |
2877 | } | |
2878 | ||
2879 | /* | |
2880 | * This target is not on a VLAN | |
2881 | */ | |
2882 | if (rt->u.dst.dev == bond->dev) { | |
ed4b9f80 | 2883 | ip_rt_put(rt); |
c3ade5ca JV |
2884 | dprintk("basa: rtdev == bond->dev: arp_send\n"); |
2885 | bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], | |
2886 | bond->master_ip, 0); | |
2887 | continue; | |
2888 | } | |
2889 | ||
2890 | vlan_id = 0; | |
2891 | list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list, | |
2892 | vlan_list) { | |
2893 | vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id]; | |
2894 | if (vlan_dev == rt->u.dst.dev) { | |
2895 | vlan_id = vlan->vlan_id; | |
2896 | dprintk("basa: vlan match on %s %d\n", | |
2897 | vlan_dev->name, vlan_id); | |
2898 | break; | |
2899 | } | |
2900 | } | |
2901 | ||
2902 | if (vlan_id) { | |
ed4b9f80 | 2903 | ip_rt_put(rt); |
c3ade5ca JV |
2904 | bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], |
2905 | vlan->vlan_ip, vlan_id); | |
2906 | continue; | |
2907 | } | |
2908 | ||
2909 | if (net_ratelimit()) { | |
2910 | printk(KERN_WARNING DRV_NAME | |
2911 | ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n", | |
2912 | bond->dev->name, NIPQUAD(fl.fl4_dst), | |
2913 | rt->u.dst.dev ? rt->u.dst.dev->name : "NULL"); | |
2914 | } | |
ed4b9f80 | 2915 | ip_rt_put(rt); |
c3ade5ca JV |
2916 | } |
2917 | } | |
2918 | ||
2919 | /* | |
2920 | * Kick out a gratuitous ARP for an IP on the bonding master plus one | |
2921 | * for each VLAN above us. | |
2922 | */ | |
2923 | static void bond_send_gratuitous_arp(struct bonding *bond) | |
2924 | { | |
2925 | struct slave *slave = bond->curr_active_slave; | |
2926 | struct vlan_entry *vlan; | |
2927 | struct net_device *vlan_dev; | |
2928 | ||
2929 | dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name, | |
2930 | slave ? slave->dev->name : "NULL"); | |
2931 | if (!slave) | |
2932 | return; | |
2933 | ||
2934 | if (bond->master_ip) { | |
2935 | bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip, | |
2936 | bond->master_ip, 0); | |
2937 | } | |
2938 | ||
2939 | list_for_each_entry(vlan, &bond->vlan_list, vlan_list) { | |
2940 | vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id]; | |
2941 | if (vlan->vlan_ip) { | |
2942 | bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip, | |
2943 | vlan->vlan_ip, vlan->vlan_id); | |
2944 | } | |
1da177e4 LT |
2945 | } |
2946 | } | |
2947 | ||
2948 | /* | |
2949 | * this function is called regularly to monitor each slave's link | |
2950 | * ensuring that traffic is being sent and received when arp monitoring | |
2951 | * is used in load-balancing mode. if the adapter has been dormant, then an | |
2952 | * arp is transmitted to generate traffic. see activebackup_arp_monitor for | |
2953 | * arp monitoring in active backup mode. | |
2954 | */ | |
2955 | static void bond_loadbalance_arp_mon(struct net_device *bond_dev) | |
2956 | { | |
2957 | struct bonding *bond = bond_dev->priv; | |
2958 | struct slave *slave, *oldcurrent; | |
2959 | int do_failover = 0; | |
2960 | int delta_in_ticks; | |
2961 | int i; | |
2962 | ||
2963 | read_lock(&bond->lock); | |
2964 | ||
2965 | delta_in_ticks = (bond->params.arp_interval * HZ) / 1000; | |
2966 | ||
2967 | if (bond->kill_timers) { | |
2968 | goto out; | |
2969 | } | |
2970 | ||
2971 | if (bond->slave_cnt == 0) { | |
2972 | goto re_arm; | |
2973 | } | |
2974 | ||
2975 | read_lock(&bond->curr_slave_lock); | |
2976 | oldcurrent = bond->curr_active_slave; | |
2977 | read_unlock(&bond->curr_slave_lock); | |
2978 | ||
2979 | /* see if any of the previous devices are up now (i.e. they have | |
2980 | * xmt and rcv traffic). the curr_active_slave does not come into | |
2981 | * the picture unless it is null. also, slave->jiffies is not needed | |
2982 | * here because we send an arp on each slave and give a slave as | |
2983 | * long as it needs to get the tx/rx within the delta. | |
2984 | * TODO: what about up/down delay in arp mode? it wasn't here before | |
2985 | * so it can wait | |
2986 | */ | |
2987 | bond_for_each_slave(bond, slave, i) { | |
2988 | if (slave->link != BOND_LINK_UP) { | |
2989 | if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) && | |
2990 | ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) { | |
2991 | ||
2992 | slave->link = BOND_LINK_UP; | |
2993 | slave->state = BOND_STATE_ACTIVE; | |
2994 | ||
2995 | /* primary_slave has no meaning in round-robin | |
2996 | * mode. the window of a slave being up and | |
2997 | * curr_active_slave being null after enslaving | |
2998 | * is closed. | |
2999 | */ | |
3000 | if (!oldcurrent) { | |
3001 | printk(KERN_INFO DRV_NAME | |
3002 | ": %s: link status definitely " | |
3003 | "up for interface %s, ", | |
3004 | bond_dev->name, | |
3005 | slave->dev->name); | |
3006 | do_failover = 1; | |
3007 | } else { | |
3008 | printk(KERN_INFO DRV_NAME | |
3009 | ": %s: interface %s is now up\n", | |
3010 | bond_dev->name, | |
3011 | slave->dev->name); | |
3012 | } | |
3013 | } | |
3014 | } else { | |
3015 | /* slave->link == BOND_LINK_UP */ | |
3016 | ||
3017 | /* not all switches will respond to an arp request | |
3018 | * when the source ip is 0, so don't take the link down | |
3019 | * if we don't know our ip yet | |
3020 | */ | |
3021 | if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) || | |
3022 | (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) && | |
c3ade5ca | 3023 | bond_has_ip(bond))) { |
1da177e4 LT |
3024 | |
3025 | slave->link = BOND_LINK_DOWN; | |
3026 | slave->state = BOND_STATE_BACKUP; | |
3027 | ||
3028 | if (slave->link_failure_count < UINT_MAX) { | |
3029 | slave->link_failure_count++; | |
3030 | } | |
3031 | ||
3032 | printk(KERN_INFO DRV_NAME | |
3033 | ": %s: interface %s is now down.\n", | |
3034 | bond_dev->name, | |
3035 | slave->dev->name); | |
3036 | ||
3037 | if (slave == oldcurrent) { | |
3038 | do_failover = 1; | |
3039 | } | |
3040 | } | |
3041 | } | |
3042 | ||
3043 | /* note: if switch is in round-robin mode, all links | |
3044 | * must tx arp to ensure all links rx an arp - otherwise | |
3045 | * links may oscillate or not come up at all; if switch is | |
3046 | * in something like xor mode, there is nothing we can | |
3047 | * do - all replies will be rx'ed on same link causing slaves | |
3048 | * to be unstable during low/no traffic periods | |
3049 | */ | |
3050 | if (IS_UP(slave->dev)) { | |
3051 | bond_arp_send_all(bond, slave); | |
3052 | } | |
3053 | } | |
3054 | ||
3055 | if (do_failover) { | |
3056 | write_lock(&bond->curr_slave_lock); | |
3057 | ||
3058 | bond_select_active_slave(bond); | |
3059 | ||
3060 | if (oldcurrent && !bond->curr_active_slave) { | |
3061 | printk(KERN_INFO DRV_NAME | |
3062 | ": %s: now running without any active " | |
3063 | "interface !\n", | |
3064 | bond_dev->name); | |
3065 | } | |
3066 | ||
3067 | write_unlock(&bond->curr_slave_lock); | |
3068 | } | |
3069 | ||
3070 | re_arm: | |
3071 | if (bond->params.arp_interval) { | |
3072 | mod_timer(&bond->arp_timer, jiffies + delta_in_ticks); | |
3073 | } | |
3074 | out: | |
3075 | read_unlock(&bond->lock); | |
3076 | } | |
3077 | ||
3078 | /* | |
3079 | * When using arp monitoring in active-backup mode, this function is | |
3080 | * called to determine if any backup slaves have went down or a new | |
3081 | * current slave needs to be found. | |
3082 | * The backup slaves never generate traffic, they are considered up by merely | |
3083 | * receiving traffic. If the current slave goes down, each backup slave will | |
3084 | * be given the opportunity to tx/rx an arp before being taken down - this | |
3085 | * prevents all slaves from being taken down due to the current slave not | |
3086 | * sending any traffic for the backups to receive. The arps are not necessarily | |
3087 | * necessary, any tx and rx traffic will keep the current slave up. While any | |
3088 | * rx traffic will keep the backup slaves up, the current slave is responsible | |
3089 | * for generating traffic to keep them up regardless of any other traffic they | |
3090 | * may have received. | |
3091 | * see loadbalance_arp_monitor for arp monitoring in load balancing mode | |
3092 | */ | |
3093 | static void bond_activebackup_arp_mon(struct net_device *bond_dev) | |
3094 | { | |
3095 | struct bonding *bond = bond_dev->priv; | |
3096 | struct slave *slave; | |
3097 | int delta_in_ticks; | |
3098 | int i; | |
3099 | ||
3100 | read_lock(&bond->lock); | |
3101 | ||
3102 | delta_in_ticks = (bond->params.arp_interval * HZ) / 1000; | |
3103 | ||
3104 | if (bond->kill_timers) { | |
3105 | goto out; | |
3106 | } | |
3107 | ||
3108 | if (bond->slave_cnt == 0) { | |
3109 | goto re_arm; | |
3110 | } | |
3111 | ||
3112 | /* determine if any slave has come up or any backup slave has | |
3113 | * gone down | |
3114 | * TODO: what about up/down delay in arp mode? it wasn't here before | |
3115 | * so it can wait | |
3116 | */ | |
3117 | bond_for_each_slave(bond, slave, i) { | |
3118 | if (slave->link != BOND_LINK_UP) { | |
3119 | if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) { | |
3120 | ||
3121 | slave->link = BOND_LINK_UP; | |
3122 | ||
3123 | write_lock(&bond->curr_slave_lock); | |
3124 | ||
3125 | if ((!bond->curr_active_slave) && | |
3126 | ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) { | |
3127 | bond_change_active_slave(bond, slave); | |
3128 | bond->current_arp_slave = NULL; | |
3129 | } else if (bond->curr_active_slave != slave) { | |
3130 | /* this slave has just come up but we | |
3131 | * already have a current slave; this | |
3132 | * can also happen if bond_enslave adds | |
3133 | * a new slave that is up while we are | |
3134 | * searching for a new slave | |
3135 | */ | |
3136 | bond_set_slave_inactive_flags(slave); | |
3137 | bond->current_arp_slave = NULL; | |
3138 | } | |
3139 | ||
3140 | if (slave == bond->curr_active_slave) { | |
3141 | printk(KERN_INFO DRV_NAME | |
3142 | ": %s: %s is up and now the " | |
3143 | "active interface\n", | |
3144 | bond_dev->name, | |
3145 | slave->dev->name); | |
3146 | } else { | |
3147 | printk(KERN_INFO DRV_NAME | |
3148 | ": %s: backup interface %s is " | |
3149 | "now up\n", | |
3150 | bond_dev->name, | |
3151 | slave->dev->name); | |
3152 | } | |
3153 | ||
3154 | write_unlock(&bond->curr_slave_lock); | |
3155 | } | |
3156 | } else { | |
3157 | read_lock(&bond->curr_slave_lock); | |
3158 | ||
3159 | if ((slave != bond->curr_active_slave) && | |
3160 | (!bond->current_arp_slave) && | |
3161 | (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) && | |
c3ade5ca | 3162 | bond_has_ip(bond))) { |
1da177e4 LT |
3163 | /* a backup slave has gone down; three times |
3164 | * the delta allows the current slave to be | |
3165 | * taken out before the backup slave. | |
3166 | * note: a non-null current_arp_slave indicates | |
3167 | * the curr_active_slave went down and we are | |
3168 | * searching for a new one; under this | |
3169 | * condition we only take the curr_active_slave | |
3170 | * down - this gives each slave a chance to | |
3171 | * tx/rx traffic before being taken out | |
3172 | */ | |
3173 | ||
3174 | read_unlock(&bond->curr_slave_lock); | |
3175 | ||
3176 | slave->link = BOND_LINK_DOWN; | |
3177 | ||
3178 | if (slave->link_failure_count < UINT_MAX) { | |
3179 | slave->link_failure_count++; | |
3180 | } | |
3181 | ||
3182 | bond_set_slave_inactive_flags(slave); | |
3183 | ||
3184 | printk(KERN_INFO DRV_NAME | |
3185 | ": %s: backup interface %s is now down\n", | |
3186 | bond_dev->name, | |
3187 | slave->dev->name); | |
3188 | } else { | |
3189 | read_unlock(&bond->curr_slave_lock); | |
3190 | } | |
3191 | } | |
3192 | } | |
3193 | ||
3194 | read_lock(&bond->curr_slave_lock); | |
3195 | slave = bond->curr_active_slave; | |
3196 | read_unlock(&bond->curr_slave_lock); | |
3197 | ||
3198 | if (slave) { | |
3199 | /* if we have sent traffic in the past 2*arp_intervals but | |
3200 | * haven't xmit and rx traffic in that time interval, select | |
3201 | * a different slave. slave->jiffies is only updated when | |
3202 | * a slave first becomes the curr_active_slave - not necessarily | |
3203 | * after every arp; this ensures the slave has a full 2*delta | |
3204 | * before being taken out. if a primary is being used, check | |
3205 | * if it is up and needs to take over as the curr_active_slave | |
3206 | */ | |
3207 | if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) || | |
c3ade5ca JV |
3208 | (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) && |
3209 | bond_has_ip(bond))) && | |
1da177e4 LT |
3210 | ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) { |
3211 | ||
3212 | slave->link = BOND_LINK_DOWN; | |
3213 | ||
3214 | if (slave->link_failure_count < UINT_MAX) { | |
3215 | slave->link_failure_count++; | |
3216 | } | |
3217 | ||
3218 | printk(KERN_INFO DRV_NAME | |
3219 | ": %s: link status down for active interface " | |
3220 | "%s, disabling it\n", | |
3221 | bond_dev->name, | |
3222 | slave->dev->name); | |
3223 | ||
3224 | write_lock(&bond->curr_slave_lock); | |
3225 | ||
3226 | bond_select_active_slave(bond); | |
3227 | slave = bond->curr_active_slave; | |
3228 | ||
3229 | write_unlock(&bond->curr_slave_lock); | |
3230 | ||
3231 | bond->current_arp_slave = slave; | |
3232 | ||
3233 | if (slave) { | |
3234 | slave->jiffies = jiffies; | |
3235 | } | |
3236 | } else if ((bond->primary_slave) && | |
3237 | (bond->primary_slave != slave) && | |
3238 | (bond->primary_slave->link == BOND_LINK_UP)) { | |
3239 | /* at this point, slave is the curr_active_slave */ | |
3240 | printk(KERN_INFO DRV_NAME | |
3241 | ": %s: changing from interface %s to primary " | |
3242 | "interface %s\n", | |
3243 | bond_dev->name, | |
3244 | slave->dev->name, | |
3245 | bond->primary_slave->dev->name); | |
3246 | ||
3247 | /* primary is up so switch to it */ | |
3248 | write_lock(&bond->curr_slave_lock); | |
3249 | bond_change_active_slave(bond, bond->primary_slave); | |
3250 | write_unlock(&bond->curr_slave_lock); | |
3251 | ||
3252 | slave = bond->primary_slave; | |
3253 | slave->jiffies = jiffies; | |
3254 | } else { | |
3255 | bond->current_arp_slave = NULL; | |
3256 | } | |
3257 | ||
3258 | /* the current slave must tx an arp to ensure backup slaves | |
3259 | * rx traffic | |
3260 | */ | |
c3ade5ca | 3261 | if (slave && bond_has_ip(bond)) { |
1da177e4 LT |
3262 | bond_arp_send_all(bond, slave); |
3263 | } | |
3264 | } | |
3265 | ||
3266 | /* if we don't have a curr_active_slave, search for the next available | |
3267 | * backup slave from the current_arp_slave and make it the candidate | |
3268 | * for becoming the curr_active_slave | |
3269 | */ | |
3270 | if (!slave) { | |
3271 | if (!bond->current_arp_slave) { | |
3272 | bond->current_arp_slave = bond->first_slave; | |
3273 | } | |
3274 | ||
3275 | if (bond->current_arp_slave) { | |
3276 | bond_set_slave_inactive_flags(bond->current_arp_slave); | |
3277 | ||
3278 | /* search for next candidate */ | |
2f872f04 | 3279 | bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) { |
1da177e4 LT |
3280 | if (IS_UP(slave->dev)) { |
3281 | slave->link = BOND_LINK_BACK; | |
3282 | bond_set_slave_active_flags(slave); | |
3283 | bond_arp_send_all(bond, slave); | |
3284 | slave->jiffies = jiffies; | |
3285 | bond->current_arp_slave = slave; | |
3286 | break; | |
3287 | } | |
3288 | ||
3289 | /* if the link state is up at this point, we | |
3290 | * mark it down - this can happen if we have | |
3291 | * simultaneous link failures and | |
3292 | * reselect_active_interface doesn't make this | |
3293 | * one the current slave so it is still marked | |
3294 | * up when it is actually down | |
3295 | */ | |
3296 | if (slave->link == BOND_LINK_UP) { | |
3297 | slave->link = BOND_LINK_DOWN; | |
3298 | if (slave->link_failure_count < UINT_MAX) { | |
3299 | slave->link_failure_count++; | |
3300 | } | |
3301 | ||
3302 | bond_set_slave_inactive_flags(slave); | |
3303 | ||
3304 | printk(KERN_INFO DRV_NAME | |
3305 | ": %s: backup interface %s is " | |
3306 | "now down.\n", | |
3307 | bond_dev->name, | |
3308 | slave->dev->name); | |
3309 | } | |
3310 | } | |
3311 | } | |
3312 | } | |
3313 | ||
3314 | re_arm: | |
3315 | if (bond->params.arp_interval) { | |
3316 | mod_timer(&bond->arp_timer, jiffies + delta_in_ticks); | |
3317 | } | |
3318 | out: | |
3319 | read_unlock(&bond->lock); | |
3320 | } | |
3321 | ||
3322 | /*------------------------------ proc/seq_file-------------------------------*/ | |
3323 | ||
3324 | #ifdef CONFIG_PROC_FS | |
3325 | ||
3326 | #define SEQ_START_TOKEN ((void *)1) | |
3327 | ||
3328 | static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos) | |
3329 | { | |
3330 | struct bonding *bond = seq->private; | |
3331 | loff_t off = 0; | |
3332 | struct slave *slave; | |
3333 | int i; | |
3334 | ||
3335 | /* make sure the bond won't be taken away */ | |
3336 | read_lock(&dev_base_lock); | |
3337 | read_lock_bh(&bond->lock); | |
3338 | ||
3339 | if (*pos == 0) { | |
3340 | return SEQ_START_TOKEN; | |
3341 | } | |
3342 | ||
3343 | bond_for_each_slave(bond, slave, i) { | |
3344 | if (++off == *pos) { | |
3345 | return slave; | |
3346 | } | |
3347 | } | |
3348 | ||
3349 | return NULL; | |
3350 | } | |
3351 | ||
3352 | static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3353 | { | |
3354 | struct bonding *bond = seq->private; | |
3355 | struct slave *slave = v; | |
3356 | ||
3357 | ++*pos; | |
3358 | if (v == SEQ_START_TOKEN) { | |
3359 | return bond->first_slave; | |
3360 | } | |
3361 | ||
3362 | slave = slave->next; | |
3363 | ||
3364 | return (slave == bond->first_slave) ? NULL : slave; | |
3365 | } | |
3366 | ||
3367 | static void bond_info_seq_stop(struct seq_file *seq, void *v) | |
3368 | { | |
3369 | struct bonding *bond = seq->private; | |
3370 | ||
3371 | read_unlock_bh(&bond->lock); | |
3372 | read_unlock(&dev_base_lock); | |
3373 | } | |
3374 | ||
3375 | static void bond_info_show_master(struct seq_file *seq) | |
3376 | { | |
3377 | struct bonding *bond = seq->private; | |
3378 | struct slave *curr; | |
3379 | ||
3380 | read_lock(&bond->curr_slave_lock); | |
3381 | curr = bond->curr_active_slave; | |
3382 | read_unlock(&bond->curr_slave_lock); | |
3383 | ||
3384 | seq_printf(seq, "Bonding Mode: %s\n", | |
3385 | bond_mode_name(bond->params.mode)); | |
3386 | ||
3387 | if (USES_PRIMARY(bond->params.mode)) { | |
3388 | seq_printf(seq, "Primary Slave: %s\n", | |
3389 | (bond->params.primary[0]) ? | |
3390 | bond->params.primary : "None"); | |
3391 | ||
3392 | seq_printf(seq, "Currently Active Slave: %s\n", | |
3393 | (curr) ? curr->dev->name : "None"); | |
3394 | } | |
3395 | ||
3396 | seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down"); | |
3397 | seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon); | |
3398 | seq_printf(seq, "Up Delay (ms): %d\n", | |
3399 | bond->params.updelay * bond->params.miimon); | |
3400 | seq_printf(seq, "Down Delay (ms): %d\n", | |
3401 | bond->params.downdelay * bond->params.miimon); | |
3402 | ||
3403 | if (bond->params.mode == BOND_MODE_8023AD) { | |
3404 | struct ad_info ad_info; | |
3405 | ||
3406 | seq_puts(seq, "\n802.3ad info\n"); | |
3407 | seq_printf(seq, "LACP rate: %s\n", | |
3408 | (bond->params.lacp_fast) ? "fast" : "slow"); | |
3409 | ||
3410 | if (bond_3ad_get_active_agg_info(bond, &ad_info)) { | |
3411 | seq_printf(seq, "bond %s has no active aggregator\n", | |
3412 | bond->dev->name); | |
3413 | } else { | |
3414 | seq_printf(seq, "Active Aggregator Info:\n"); | |
3415 | ||
3416 | seq_printf(seq, "\tAggregator ID: %d\n", | |
3417 | ad_info.aggregator_id); | |
3418 | seq_printf(seq, "\tNumber of ports: %d\n", | |
3419 | ad_info.ports); | |
3420 | seq_printf(seq, "\tActor Key: %d\n", | |
3421 | ad_info.actor_key); | |
3422 | seq_printf(seq, "\tPartner Key: %d\n", | |
3423 | ad_info.partner_key); | |
3424 | seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n", | |
3425 | ad_info.partner_system[0], | |
3426 | ad_info.partner_system[1], | |
3427 | ad_info.partner_system[2], | |
3428 | ad_info.partner_system[3], | |
3429 | ad_info.partner_system[4], | |
3430 | ad_info.partner_system[5]); | |
3431 | } | |
3432 | } | |
3433 | } | |
3434 | ||
3435 | static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave) | |
3436 | { | |
3437 | struct bonding *bond = seq->private; | |
3438 | ||
3439 | seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name); | |
3440 | seq_printf(seq, "MII Status: %s\n", | |
3441 | (slave->link == BOND_LINK_UP) ? "up" : "down"); | |
3442 | seq_printf(seq, "Link Failure Count: %d\n", | |
3443 | slave->link_failure_count); | |
3444 | ||
3445 | if (app_abi_ver >= 1) { | |
3446 | seq_printf(seq, | |
3447 | "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n", | |
3448 | slave->perm_hwaddr[0], | |
3449 | slave->perm_hwaddr[1], | |
3450 | slave->perm_hwaddr[2], | |
3451 | slave->perm_hwaddr[3], | |
3452 | slave->perm_hwaddr[4], | |
3453 | slave->perm_hwaddr[5]); | |
3454 | } | |
3455 | ||
3456 | if (bond->params.mode == BOND_MODE_8023AD) { | |
3457 | const struct aggregator *agg | |
3458 | = SLAVE_AD_INFO(slave).port.aggregator; | |
3459 | ||
3460 | if (agg) { | |
3461 | seq_printf(seq, "Aggregator ID: %d\n", | |
3462 | agg->aggregator_identifier); | |
3463 | } else { | |
3464 | seq_puts(seq, "Aggregator ID: N/A\n"); | |
3465 | } | |
3466 | } | |
3467 | } | |
3468 | ||
3469 | static int bond_info_seq_show(struct seq_file *seq, void *v) | |
3470 | { | |
3471 | if (v == SEQ_START_TOKEN) { | |
3472 | seq_printf(seq, "%s\n", version); | |
3473 | bond_info_show_master(seq); | |
3474 | } else { | |
3475 | bond_info_show_slave(seq, v); | |
3476 | } | |
3477 | ||
3478 | return 0; | |
3479 | } | |
3480 | ||
3481 | static struct seq_operations bond_info_seq_ops = { | |
3482 | .start = bond_info_seq_start, | |
3483 | .next = bond_info_seq_next, | |
3484 | .stop = bond_info_seq_stop, | |
3485 | .show = bond_info_seq_show, | |
3486 | }; | |
3487 | ||
3488 | static int bond_info_open(struct inode *inode, struct file *file) | |
3489 | { | |
3490 | struct seq_file *seq; | |
3491 | struct proc_dir_entry *proc; | |
3492 | int res; | |
3493 | ||
3494 | res = seq_open(file, &bond_info_seq_ops); | |
3495 | if (!res) { | |
3496 | /* recover the pointer buried in proc_dir_entry data */ | |
3497 | seq = file->private_data; | |
3498 | proc = PDE(inode); | |
3499 | seq->private = proc->data; | |
3500 | } | |
3501 | ||
3502 | return res; | |
3503 | } | |
3504 | ||
3505 | static struct file_operations bond_info_fops = { | |
3506 | .owner = THIS_MODULE, | |
3507 | .open = bond_info_open, | |
3508 | .read = seq_read, | |
3509 | .llseek = seq_lseek, | |
3510 | .release = seq_release, | |
3511 | }; | |
3512 | ||
3513 | static int bond_create_proc_entry(struct bonding *bond) | |
3514 | { | |
3515 | struct net_device *bond_dev = bond->dev; | |
3516 | ||
3517 | if (bond_proc_dir) { | |
3518 | bond->proc_entry = create_proc_entry(bond_dev->name, | |
3519 | S_IRUGO, | |
3520 | bond_proc_dir); | |
3521 | if (bond->proc_entry == NULL) { | |
3522 | printk(KERN_WARNING DRV_NAME | |
3523 | ": Warning: Cannot create /proc/net/%s/%s\n", | |
3524 | DRV_NAME, bond_dev->name); | |
3525 | } else { | |
3526 | bond->proc_entry->data = bond; | |
3527 | bond->proc_entry->proc_fops = &bond_info_fops; | |
3528 | bond->proc_entry->owner = THIS_MODULE; | |
3529 | memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ); | |
3530 | } | |
3531 | } | |
3532 | ||
3533 | return 0; | |
3534 | } | |
3535 | ||
3536 | static void bond_remove_proc_entry(struct bonding *bond) | |
3537 | { | |
3538 | if (bond_proc_dir && bond->proc_entry) { | |
3539 | remove_proc_entry(bond->proc_file_name, bond_proc_dir); | |
3540 | memset(bond->proc_file_name, 0, IFNAMSIZ); | |
3541 | bond->proc_entry = NULL; | |
3542 | } | |
3543 | } | |
3544 | ||
3545 | /* Create the bonding directory under /proc/net, if doesn't exist yet. | |
3546 | * Caller must hold rtnl_lock. | |
3547 | */ | |
3548 | static void bond_create_proc_dir(void) | |
3549 | { | |
3550 | int len = strlen(DRV_NAME); | |
3551 | ||
3552 | for (bond_proc_dir = proc_net->subdir; bond_proc_dir; | |
3553 | bond_proc_dir = bond_proc_dir->next) { | |
3554 | if ((bond_proc_dir->namelen == len) && | |
3555 | !memcmp(bond_proc_dir->name, DRV_NAME, len)) { | |
3556 | break; | |
3557 | } | |
3558 | } | |
3559 | ||
3560 | if (!bond_proc_dir) { | |
3561 | bond_proc_dir = proc_mkdir(DRV_NAME, proc_net); | |
3562 | if (bond_proc_dir) { | |
3563 | bond_proc_dir->owner = THIS_MODULE; | |
3564 | } else { | |
3565 | printk(KERN_WARNING DRV_NAME | |
3566 | ": Warning: cannot create /proc/net/%s\n", | |
3567 | DRV_NAME); | |
3568 | } | |
3569 | } | |
3570 | } | |
3571 | ||
3572 | /* Destroy the bonding directory under /proc/net, if empty. | |
3573 | * Caller must hold rtnl_lock. | |
3574 | */ | |
3575 | static void bond_destroy_proc_dir(void) | |
3576 | { | |
3577 | struct proc_dir_entry *de; | |
3578 | ||
3579 | if (!bond_proc_dir) { | |
3580 | return; | |
3581 | } | |
3582 | ||
3583 | /* verify that the /proc dir is empty */ | |
3584 | for (de = bond_proc_dir->subdir; de; de = de->next) { | |
3585 | /* ignore . and .. */ | |
3586 | if (*(de->name) != '.') { | |
3587 | break; | |
3588 | } | |
3589 | } | |
3590 | ||
3591 | if (de) { | |
3592 | if (bond_proc_dir->owner == THIS_MODULE) { | |
3593 | bond_proc_dir->owner = NULL; | |
3594 | } | |
3595 | } else { | |
3596 | remove_proc_entry(DRV_NAME, proc_net); | |
3597 | bond_proc_dir = NULL; | |
3598 | } | |
3599 | } | |
3600 | #endif /* CONFIG_PROC_FS */ | |
3601 | ||
3602 | /*-------------------------- netdev event handling --------------------------*/ | |
3603 | ||
3604 | /* | |
3605 | * Change device name | |
3606 | */ | |
3607 | static int bond_event_changename(struct bonding *bond) | |
3608 | { | |
3609 | #ifdef CONFIG_PROC_FS | |
3610 | bond_remove_proc_entry(bond); | |
3611 | bond_create_proc_entry(bond); | |
3612 | #endif | |
3613 | ||
3614 | return NOTIFY_DONE; | |
3615 | } | |
3616 | ||
3617 | static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev) | |
3618 | { | |
3619 | struct bonding *event_bond = bond_dev->priv; | |
3620 | ||
3621 | switch (event) { | |
3622 | case NETDEV_CHANGENAME: | |
3623 | return bond_event_changename(event_bond); | |
3624 | case NETDEV_UNREGISTER: | |
3625 | /* | |
3626 | * TODO: remove a bond from the list? | |
3627 | */ | |
3628 | break; | |
3629 | default: | |
3630 | break; | |
3631 | } | |
3632 | ||
3633 | return NOTIFY_DONE; | |
3634 | } | |
3635 | ||
3636 | static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev) | |
3637 | { | |
3638 | struct net_device *bond_dev = slave_dev->master; | |
8531c5ff | 3639 | struct bonding *bond = bond_dev->priv; |
1da177e4 LT |
3640 | |
3641 | switch (event) { | |
3642 | case NETDEV_UNREGISTER: | |
3643 | if (bond_dev) { | |
3644 | bond_release(bond_dev, slave_dev); | |
3645 | } | |
3646 | break; | |
3647 | case NETDEV_CHANGE: | |
3648 | /* | |
3649 | * TODO: is this what we get if somebody | |
3650 | * sets up a hierarchical bond, then rmmod's | |
3651 | * one of the slave bonding devices? | |
3652 | */ | |
3653 | break; | |
3654 | case NETDEV_DOWN: | |
3655 | /* | |
3656 | * ... Or is it this? | |
3657 | */ | |
3658 | break; | |
3659 | case NETDEV_CHANGEMTU: | |
3660 | /* | |
3661 | * TODO: Should slaves be allowed to | |
3662 | * independently alter their MTU? For | |
3663 | * an active-backup bond, slaves need | |
3664 | * not be the same type of device, so | |
3665 | * MTUs may vary. For other modes, | |
3666 | * slaves arguably should have the | |
3667 | * same MTUs. To do this, we'd need to | |
3668 | * take over the slave's change_mtu | |
3669 | * function for the duration of their | |
3670 | * servitude. | |
3671 | */ | |
3672 | break; | |
3673 | case NETDEV_CHANGENAME: | |
3674 | /* | |
3675 | * TODO: handle changing the primary's name | |
3676 | */ | |
3677 | break; | |
8531c5ff AK |
3678 | case NETDEV_FEAT_CHANGE: |
3679 | bond_compute_features(bond); | |
3680 | break; | |
1da177e4 LT |
3681 | default: |
3682 | break; | |
3683 | } | |
3684 | ||
3685 | return NOTIFY_DONE; | |
3686 | } | |
3687 | ||
3688 | /* | |
3689 | * bond_netdev_event: handle netdev notifier chain events. | |
3690 | * | |
3691 | * This function receives events for the netdev chain. The caller (an | |
3692 | * ioctl handler calling notifier_call_chain) holds the necessary | |
3693 | * locks for us to safely manipulate the slave devices (RTNL lock, | |
3694 | * dev_probe_lock). | |
3695 | */ | |
3696 | static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) | |
3697 | { | |
3698 | struct net_device *event_dev = (struct net_device *)ptr; | |
3699 | ||
3700 | dprintk("event_dev: %s, event: %lx\n", | |
3701 | (event_dev ? event_dev->name : "None"), | |
3702 | event); | |
3703 | ||
3704 | if (event_dev->flags & IFF_MASTER) { | |
3705 | dprintk("IFF_MASTER\n"); | |
3706 | return bond_master_netdev_event(event, event_dev); | |
3707 | } | |
3708 | ||
3709 | if (event_dev->flags & IFF_SLAVE) { | |
3710 | dprintk("IFF_SLAVE\n"); | |
3711 | return bond_slave_netdev_event(event, event_dev); | |
3712 | } | |
3713 | ||
3714 | return NOTIFY_DONE; | |
3715 | } | |
3716 | ||
c3ade5ca JV |
3717 | /* |
3718 | * bond_inetaddr_event: handle inetaddr notifier chain events. | |
3719 | * | |
3720 | * We keep track of device IPs primarily to use as source addresses in | |
3721 | * ARP monitor probes (rather than spewing out broadcasts all the time). | |
3722 | * | |
3723 | * We track one IP for the main device (if it has one), plus one per VLAN. | |
3724 | */ | |
3725 | static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) | |
3726 | { | |
3727 | struct in_ifaddr *ifa = ptr; | |
3728 | struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev; | |
3729 | struct bonding *bond, *bond_next; | |
3730 | struct vlan_entry *vlan, *vlan_next; | |
3731 | ||
3732 | list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) { | |
3733 | if (bond->dev == event_dev) { | |
3734 | switch (event) { | |
3735 | case NETDEV_UP: | |
3736 | bond->master_ip = ifa->ifa_local; | |
3737 | return NOTIFY_OK; | |
3738 | case NETDEV_DOWN: | |
3739 | bond->master_ip = bond_glean_dev_ip(bond->dev); | |
3740 | return NOTIFY_OK; | |
3741 | default: | |
3742 | return NOTIFY_DONE; | |
3743 | } | |
3744 | } | |
3745 | ||
3746 | if (list_empty(&bond->vlan_list)) | |
3747 | continue; | |
3748 | ||
3749 | list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list, | |
3750 | vlan_list) { | |
3751 | vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id]; | |
3752 | if (vlan_dev == event_dev) { | |
3753 | switch (event) { | |
3754 | case NETDEV_UP: | |
3755 | vlan->vlan_ip = ifa->ifa_local; | |
3756 | return NOTIFY_OK; | |
3757 | case NETDEV_DOWN: | |
3758 | vlan->vlan_ip = | |
3759 | bond_glean_dev_ip(vlan_dev); | |
3760 | return NOTIFY_OK; | |
3761 | default: | |
3762 | return NOTIFY_DONE; | |
3763 | } | |
3764 | } | |
3765 | } | |
3766 | } | |
3767 | return NOTIFY_DONE; | |
3768 | } | |
3769 | ||
1da177e4 LT |
3770 | static struct notifier_block bond_netdev_notifier = { |
3771 | .notifier_call = bond_netdev_event, | |
3772 | }; | |
3773 | ||
c3ade5ca JV |
3774 | static struct notifier_block bond_inetaddr_notifier = { |
3775 | .notifier_call = bond_inetaddr_event, | |
3776 | }; | |
3777 | ||
1da177e4 LT |
3778 | /*-------------------------- Packet type handling ---------------------------*/ |
3779 | ||
3780 | /* register to receive lacpdus on a bond */ | |
3781 | static void bond_register_lacpdu(struct bonding *bond) | |
3782 | { | |
3783 | struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type); | |
3784 | ||
3785 | /* initialize packet type */ | |
3786 | pk_type->type = PKT_TYPE_LACPDU; | |
3787 | pk_type->dev = bond->dev; | |
3788 | pk_type->func = bond_3ad_lacpdu_recv; | |
3789 | ||
3790 | dev_add_pack(pk_type); | |
3791 | } | |
3792 | ||
3793 | /* unregister to receive lacpdus on a bond */ | |
3794 | static void bond_unregister_lacpdu(struct bonding *bond) | |
3795 | { | |
3796 | dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type)); | |
3797 | } | |
3798 | ||
169a3e66 JV |
3799 | /*---------------------------- Hashing Policies -----------------------------*/ |
3800 | ||
3801 | /* | |
3802 | * Hash for the the output device based upon layer 3 and layer 4 data. If | |
3803 | * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is | |
3804 | * altogether not IP, mimic bond_xmit_hash_policy_l2() | |
3805 | */ | |
3806 | static int bond_xmit_hash_policy_l34(struct sk_buff *skb, | |
3807 | struct net_device *bond_dev, int count) | |
3808 | { | |
3809 | struct ethhdr *data = (struct ethhdr *)skb->data; | |
3810 | struct iphdr *iph = skb->nh.iph; | |
3811 | u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl); | |
3812 | int layer4_xor = 0; | |
3813 | ||
3814 | if (skb->protocol == __constant_htons(ETH_P_IP)) { | |
3815 | if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) && | |
3816 | (iph->protocol == IPPROTO_TCP || | |
3817 | iph->protocol == IPPROTO_UDP)) { | |
3818 | layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1))); | |
3819 | } | |
3820 | return (layer4_xor ^ | |
3821 | ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count; | |
3822 | ||
3823 | } | |
3824 | ||
3825 | return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count; | |
3826 | } | |
3827 | ||
3828 | /* | |
3829 | * Hash for the output device based upon layer 2 data | |
3830 | */ | |
3831 | static int bond_xmit_hash_policy_l2(struct sk_buff *skb, | |
3832 | struct net_device *bond_dev, int count) | |
3833 | { | |
3834 | struct ethhdr *data = (struct ethhdr *)skb->data; | |
3835 | ||
3836 | return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count; | |
3837 | } | |
3838 | ||
1da177e4 LT |
3839 | /*-------------------------- Device entry points ----------------------------*/ |
3840 | ||
3841 | static int bond_open(struct net_device *bond_dev) | |
3842 | { | |
3843 | struct bonding *bond = bond_dev->priv; | |
3844 | struct timer_list *mii_timer = &bond->mii_timer; | |
3845 | struct timer_list *arp_timer = &bond->arp_timer; | |
3846 | ||
3847 | bond->kill_timers = 0; | |
3848 | ||
3849 | if ((bond->params.mode == BOND_MODE_TLB) || | |
3850 | (bond->params.mode == BOND_MODE_ALB)) { | |
3851 | struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer); | |
3852 | ||
3853 | /* bond_alb_initialize must be called before the timer | |
3854 | * is started. | |
3855 | */ | |
3856 | if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) { | |
3857 | /* something went wrong - fail the open operation */ | |
3858 | return -1; | |
3859 | } | |
3860 | ||
3861 | init_timer(alb_timer); | |
3862 | alb_timer->expires = jiffies + 1; | |
3863 | alb_timer->data = (unsigned long)bond; | |
3864 | alb_timer->function = (void *)&bond_alb_monitor; | |
3865 | add_timer(alb_timer); | |
3866 | } | |
3867 | ||
3868 | if (bond->params.miimon) { /* link check interval, in milliseconds. */ | |
3869 | init_timer(mii_timer); | |
3870 | mii_timer->expires = jiffies + 1; | |
3871 | mii_timer->data = (unsigned long)bond_dev; | |
3872 | mii_timer->function = (void *)&bond_mii_monitor; | |
3873 | add_timer(mii_timer); | |
3874 | } | |
3875 | ||
3876 | if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ | |
3877 | init_timer(arp_timer); | |
3878 | arp_timer->expires = jiffies + 1; | |
3879 | arp_timer->data = (unsigned long)bond_dev; | |
3880 | if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) { | |
3881 | arp_timer->function = (void *)&bond_activebackup_arp_mon; | |
3882 | } else { | |
3883 | arp_timer->function = (void *)&bond_loadbalance_arp_mon; | |
3884 | } | |
3885 | add_timer(arp_timer); | |
3886 | } | |
3887 | ||
3888 | if (bond->params.mode == BOND_MODE_8023AD) { | |
3889 | struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer); | |
3890 | init_timer(ad_timer); | |
3891 | ad_timer->expires = jiffies + 1; | |
3892 | ad_timer->data = (unsigned long)bond; | |
3893 | ad_timer->function = (void *)&bond_3ad_state_machine_handler; | |
3894 | add_timer(ad_timer); | |
3895 | ||
3896 | /* register to receive LACPDUs */ | |
3897 | bond_register_lacpdu(bond); | |
3898 | } | |
3899 | ||
3900 | return 0; | |
3901 | } | |
3902 | ||
3903 | static int bond_close(struct net_device *bond_dev) | |
3904 | { | |
3905 | struct bonding *bond = bond_dev->priv; | |
3906 | ||
3907 | if (bond->params.mode == BOND_MODE_8023AD) { | |
3908 | /* Unregister the receive of LACPDUs */ | |
3909 | bond_unregister_lacpdu(bond); | |
3910 | } | |
3911 | ||
3912 | write_lock_bh(&bond->lock); | |
3913 | ||
3914 | bond_mc_list_destroy(bond); | |
3915 | ||
3916 | /* signal timers not to re-arm */ | |
3917 | bond->kill_timers = 1; | |
3918 | ||
3919 | write_unlock_bh(&bond->lock); | |
3920 | ||
3921 | /* del_timer_sync must run without holding the bond->lock | |
3922 | * because a running timer might be trying to hold it too | |
3923 | */ | |
3924 | ||
3925 | if (bond->params.miimon) { /* link check interval, in milliseconds. */ | |
3926 | del_timer_sync(&bond->mii_timer); | |
3927 | } | |
3928 | ||
3929 | if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ | |
3930 | del_timer_sync(&bond->arp_timer); | |
3931 | } | |
3932 | ||
3933 | switch (bond->params.mode) { | |
3934 | case BOND_MODE_8023AD: | |
3935 | del_timer_sync(&(BOND_AD_INFO(bond).ad_timer)); | |
3936 | break; | |
3937 | case BOND_MODE_TLB: | |
3938 | case BOND_MODE_ALB: | |
3939 | del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer)); | |
3940 | break; | |
3941 | default: | |
3942 | break; | |
3943 | } | |
3944 | ||
3945 | /* Release the bonded slaves */ | |
3946 | bond_release_all(bond_dev); | |
3947 | ||
3948 | if ((bond->params.mode == BOND_MODE_TLB) || | |
3949 | (bond->params.mode == BOND_MODE_ALB)) { | |
3950 | /* Must be called only after all | |
3951 | * slaves have been released | |
3952 | */ | |
3953 | bond_alb_deinitialize(bond); | |
3954 | } | |
3955 | ||
3956 | return 0; | |
3957 | } | |
3958 | ||
3959 | static struct net_device_stats *bond_get_stats(struct net_device *bond_dev) | |
3960 | { | |
3961 | struct bonding *bond = bond_dev->priv; | |
3962 | struct net_device_stats *stats = &(bond->stats), *sstats; | |
3963 | struct slave *slave; | |
3964 | int i; | |
3965 | ||
3966 | memset(stats, 0, sizeof(struct net_device_stats)); | |
3967 | ||
3968 | read_lock_bh(&bond->lock); | |
3969 | ||
3970 | bond_for_each_slave(bond, slave, i) { | |
3971 | sstats = slave->dev->get_stats(slave->dev); | |
3972 | ||
3973 | stats->rx_packets += sstats->rx_packets; | |
3974 | stats->rx_bytes += sstats->rx_bytes; | |
3975 | stats->rx_errors += sstats->rx_errors; | |
3976 | stats->rx_dropped += sstats->rx_dropped; | |
3977 | ||
3978 | stats->tx_packets += sstats->tx_packets; | |
3979 | stats->tx_bytes += sstats->tx_bytes; | |
3980 | stats->tx_errors += sstats->tx_errors; | |
3981 | stats->tx_dropped += sstats->tx_dropped; | |
3982 | ||
3983 | stats->multicast += sstats->multicast; | |
3984 | stats->collisions += sstats->collisions; | |
3985 | ||
3986 | stats->rx_length_errors += sstats->rx_length_errors; | |
3987 | stats->rx_over_errors += sstats->rx_over_errors; | |
3988 | stats->rx_crc_errors += sstats->rx_crc_errors; | |
3989 | stats->rx_frame_errors += sstats->rx_frame_errors; | |
3990 | stats->rx_fifo_errors += sstats->rx_fifo_errors; | |
3991 | stats->rx_missed_errors += sstats->rx_missed_errors; | |
3992 | ||
3993 | stats->tx_aborted_errors += sstats->tx_aborted_errors; | |
3994 | stats->tx_carrier_errors += sstats->tx_carrier_errors; | |
3995 | stats->tx_fifo_errors += sstats->tx_fifo_errors; | |
3996 | stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors; | |
3997 | stats->tx_window_errors += sstats->tx_window_errors; | |
3998 | } | |
3999 | ||
4000 | read_unlock_bh(&bond->lock); | |
4001 | ||
4002 | return stats; | |
4003 | } | |
4004 | ||
4005 | static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) | |
4006 | { | |
4007 | struct net_device *slave_dev = NULL; | |
4008 | struct ifbond k_binfo; | |
4009 | struct ifbond __user *u_binfo = NULL; | |
4010 | struct ifslave k_sinfo; | |
4011 | struct ifslave __user *u_sinfo = NULL; | |
4012 | struct mii_ioctl_data *mii = NULL; | |
4013 | int prev_abi_ver = orig_app_abi_ver; | |
4014 | int res = 0; | |
4015 | ||
4016 | dprintk("bond_ioctl: master=%s, cmd=%d\n", | |
4017 | bond_dev->name, cmd); | |
4018 | ||
4019 | switch (cmd) { | |
4020 | case SIOCETHTOOL: | |
4021 | return bond_ethtool_ioctl(bond_dev, ifr); | |
4022 | case SIOCGMIIPHY: | |
4023 | mii = if_mii(ifr); | |
4024 | if (!mii) { | |
4025 | return -EINVAL; | |
4026 | } | |
4027 | mii->phy_id = 0; | |
4028 | /* Fall Through */ | |
4029 | case SIOCGMIIREG: | |
4030 | /* | |
4031 | * We do this again just in case we were called by SIOCGMIIREG | |
4032 | * instead of SIOCGMIIPHY. | |
4033 | */ | |
4034 | mii = if_mii(ifr); | |
4035 | if (!mii) { | |
4036 | return -EINVAL; | |
4037 | } | |
4038 | ||
4039 | if (mii->reg_num == 1) { | |
4040 | struct bonding *bond = bond_dev->priv; | |
4041 | mii->val_out = 0; | |
4042 | read_lock_bh(&bond->lock); | |
4043 | read_lock(&bond->curr_slave_lock); | |
4044 | if (bond->curr_active_slave) { | |
4045 | mii->val_out = BMSR_LSTATUS; | |
4046 | } | |
4047 | read_unlock(&bond->curr_slave_lock); | |
4048 | read_unlock_bh(&bond->lock); | |
4049 | } | |
4050 | ||
4051 | return 0; | |
4052 | case BOND_INFO_QUERY_OLD: | |
4053 | case SIOCBONDINFOQUERY: | |
4054 | u_binfo = (struct ifbond __user *)ifr->ifr_data; | |
4055 | ||
4056 | if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) { | |
4057 | return -EFAULT; | |
4058 | } | |
4059 | ||
4060 | res = bond_info_query(bond_dev, &k_binfo); | |
4061 | if (res == 0) { | |
4062 | if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) { | |
4063 | return -EFAULT; | |
4064 | } | |
4065 | } | |
4066 | ||
4067 | return res; | |
4068 | case BOND_SLAVE_INFO_QUERY_OLD: | |
4069 | case SIOCBONDSLAVEINFOQUERY: | |
4070 | u_sinfo = (struct ifslave __user *)ifr->ifr_data; | |
4071 | ||
4072 | if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) { | |
4073 | return -EFAULT; | |
4074 | } | |
4075 | ||
4076 | res = bond_slave_info_query(bond_dev, &k_sinfo); | |
4077 | if (res == 0) { | |
4078 | if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) { | |
4079 | return -EFAULT; | |
4080 | } | |
4081 | } | |
4082 | ||
4083 | return res; | |
4084 | default: | |
4085 | /* Go on */ | |
4086 | break; | |
4087 | } | |
4088 | ||
4089 | if (!capable(CAP_NET_ADMIN)) { | |
4090 | return -EPERM; | |
4091 | } | |
4092 | ||
4093 | if (orig_app_abi_ver == -1) { | |
4094 | /* no orig_app_abi_ver was provided yet, so we'll use the | |
4095 | * current one from now on, even if it's 0 | |
4096 | */ | |
4097 | orig_app_abi_ver = app_abi_ver; | |
4098 | ||
4099 | } else if (orig_app_abi_ver != app_abi_ver) { | |
4100 | printk(KERN_ERR DRV_NAME | |
4101 | ": Error: already using ifenslave ABI version %d; to " | |
4102 | "upgrade ifenslave to version %d, you must first " | |
4103 | "reload bonding.\n", | |
4104 | orig_app_abi_ver, app_abi_ver); | |
4105 | return -EINVAL; | |
4106 | } | |
4107 | ||
4108 | slave_dev = dev_get_by_name(ifr->ifr_slave); | |
4109 | ||
4110 | dprintk("slave_dev=%p: \n", slave_dev); | |
4111 | ||
4112 | if (!slave_dev) { | |
4113 | res = -ENODEV; | |
4114 | } else { | |
4115 | dprintk("slave_dev->name=%s: \n", slave_dev->name); | |
4116 | switch (cmd) { | |
4117 | case BOND_ENSLAVE_OLD: | |
4118 | case SIOCBONDENSLAVE: | |
4119 | res = bond_enslave(bond_dev, slave_dev); | |
4120 | break; | |
4121 | case BOND_RELEASE_OLD: | |
4122 | case SIOCBONDRELEASE: | |
4123 | res = bond_release(bond_dev, slave_dev); | |
4124 | break; | |
4125 | case BOND_SETHWADDR_OLD: | |
4126 | case SIOCBONDSETHWADDR: | |
4127 | res = bond_sethwaddr(bond_dev, slave_dev); | |
4128 | break; | |
4129 | case BOND_CHANGE_ACTIVE_OLD: | |
4130 | case SIOCBONDCHANGEACTIVE: | |
4131 | res = bond_ioctl_change_active(bond_dev, slave_dev); | |
4132 | break; | |
4133 | default: | |
4134 | res = -EOPNOTSUPP; | |
4135 | } | |
4136 | ||
4137 | dev_put(slave_dev); | |
4138 | } | |
4139 | ||
4140 | if (res < 0) { | |
4141 | /* The ioctl failed, so there's no point in changing the | |
4142 | * orig_app_abi_ver. We'll restore it's value just in case | |
4143 | * we've changed it earlier in this function. | |
4144 | */ | |
4145 | orig_app_abi_ver = prev_abi_ver; | |
4146 | } | |
4147 | ||
4148 | return res; | |
4149 | } | |
4150 | ||
4151 | static void bond_set_multicast_list(struct net_device *bond_dev) | |
4152 | { | |
4153 | struct bonding *bond = bond_dev->priv; | |
4154 | struct dev_mc_list *dmi; | |
4155 | ||
4156 | write_lock_bh(&bond->lock); | |
4157 | ||
4158 | /* | |
4159 | * Do promisc before checking multicast_mode | |
4160 | */ | |
4161 | if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) { | |
4162 | bond_set_promiscuity(bond, 1); | |
4163 | } | |
4164 | ||
4165 | if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) { | |
4166 | bond_set_promiscuity(bond, -1); | |
4167 | } | |
4168 | ||
4169 | /* set allmulti flag to slaves */ | |
4170 | if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) { | |
4171 | bond_set_allmulti(bond, 1); | |
4172 | } | |
4173 | ||
4174 | if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) { | |
4175 | bond_set_allmulti(bond, -1); | |
4176 | } | |
4177 | ||
4178 | bond->flags = bond_dev->flags; | |
4179 | ||
4180 | /* looking for addresses to add to slaves' mc list */ | |
4181 | for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) { | |
4182 | if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) { | |
4183 | bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen); | |
4184 | } | |
4185 | } | |
4186 | ||
4187 | /* looking for addresses to delete from slaves' list */ | |
4188 | for (dmi = bond->mc_list; dmi; dmi = dmi->next) { | |
4189 | if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) { | |
4190 | bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen); | |
4191 | } | |
4192 | } | |
4193 | ||
4194 | /* save master's multicast list */ | |
4195 | bond_mc_list_destroy(bond); | |
4196 | bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC); | |
4197 | ||
4198 | write_unlock_bh(&bond->lock); | |
4199 | } | |
4200 | ||
4201 | /* | |
4202 | * Change the MTU of all of a master's slaves to match the master | |
4203 | */ | |
4204 | static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) | |
4205 | { | |
4206 | struct bonding *bond = bond_dev->priv; | |
4207 | struct slave *slave, *stop_at; | |
4208 | int res = 0; | |
4209 | int i; | |
4210 | ||
4211 | dprintk("bond=%p, name=%s, new_mtu=%d\n", bond, | |
4212 | (bond_dev ? bond_dev->name : "None"), new_mtu); | |
4213 | ||
4214 | /* Can't hold bond->lock with bh disabled here since | |
4215 | * some base drivers panic. On the other hand we can't | |
4216 | * hold bond->lock without bh disabled because we'll | |
4217 | * deadlock. The only solution is to rely on the fact | |
4218 | * that we're under rtnl_lock here, and the slaves | |
4219 | * list won't change. This doesn't solve the problem | |
4220 | * of setting the slave's MTU while it is | |
4221 | * transmitting, but the assumption is that the base | |
4222 | * driver can handle that. | |
4223 | * | |
4224 | * TODO: figure out a way to safely iterate the slaves | |
4225 | * list, but without holding a lock around the actual | |
4226 | * call to the base driver. | |
4227 | */ | |
4228 | ||
4229 | bond_for_each_slave(bond, slave, i) { | |
4230 | dprintk("s %p s->p %p c_m %p\n", slave, | |
4231 | slave->prev, slave->dev->change_mtu); | |
4232 | res = dev_set_mtu(slave->dev, new_mtu); | |
4233 | ||
4234 | if (res) { | |
4235 | /* If we failed to set the slave's mtu to the new value | |
4236 | * we must abort the operation even in ACTIVE_BACKUP | |
4237 | * mode, because if we allow the backup slaves to have | |
4238 | * different mtu values than the active slave we'll | |
4239 | * need to change their mtu when doing a failover. That | |
4240 | * means changing their mtu from timer context, which | |
4241 | * is probably not a good idea. | |
4242 | */ | |
4243 | dprintk("err %d %s\n", res, slave->dev->name); | |
4244 | goto unwind; | |
4245 | } | |
4246 | } | |
4247 | ||
4248 | bond_dev->mtu = new_mtu; | |
4249 | ||
4250 | return 0; | |
4251 | ||
4252 | unwind: | |
4253 | /* unwind from head to the slave that failed */ | |
4254 | stop_at = slave; | |
4255 | bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) { | |
4256 | int tmp_res; | |
4257 | ||
4258 | tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu); | |
4259 | if (tmp_res) { | |
4260 | dprintk("unwind err %d dev %s\n", tmp_res, | |
4261 | slave->dev->name); | |
4262 | } | |
4263 | } | |
4264 | ||
4265 | return res; | |
4266 | } | |
4267 | ||
4268 | /* | |
4269 | * Change HW address | |
4270 | * | |
4271 | * Note that many devices must be down to change the HW address, and | |
4272 | * downing the master releases all slaves. We can make bonds full of | |
4273 | * bonding devices to test this, however. | |
4274 | */ | |
4275 | static int bond_set_mac_address(struct net_device *bond_dev, void *addr) | |
4276 | { | |
4277 | struct bonding *bond = bond_dev->priv; | |
4278 | struct sockaddr *sa = addr, tmp_sa; | |
4279 | struct slave *slave, *stop_at; | |
4280 | int res = 0; | |
4281 | int i; | |
4282 | ||
4283 | dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None")); | |
4284 | ||
4285 | if (!is_valid_ether_addr(sa->sa_data)) { | |
4286 | return -EADDRNOTAVAIL; | |
4287 | } | |
4288 | ||
4289 | /* Can't hold bond->lock with bh disabled here since | |
4290 | * some base drivers panic. On the other hand we can't | |
4291 | * hold bond->lock without bh disabled because we'll | |
4292 | * deadlock. The only solution is to rely on the fact | |
4293 | * that we're under rtnl_lock here, and the slaves | |
4294 | * list won't change. This doesn't solve the problem | |
4295 | * of setting the slave's hw address while it is | |
4296 | * transmitting, but the assumption is that the base | |
4297 | * driver can handle that. | |
4298 | * | |
4299 | * TODO: figure out a way to safely iterate the slaves | |
4300 | * list, but without holding a lock around the actual | |
4301 | * call to the base driver. | |
4302 | */ | |
4303 | ||
4304 | bond_for_each_slave(bond, slave, i) { | |
4305 | dprintk("slave %p %s\n", slave, slave->dev->name); | |
4306 | ||
4307 | if (slave->dev->set_mac_address == NULL) { | |
4308 | res = -EOPNOTSUPP; | |
4309 | dprintk("EOPNOTSUPP %s\n", slave->dev->name); | |
4310 | goto unwind; | |
4311 | } | |
4312 | ||
4313 | res = dev_set_mac_address(slave->dev, addr); | |
4314 | if (res) { | |
4315 | /* TODO: consider downing the slave | |
4316 | * and retry ? | |
4317 | * User should expect communications | |
4318 | * breakage anyway until ARP finish | |
4319 | * updating, so... | |
4320 | */ | |
4321 | dprintk("err %d %s\n", res, slave->dev->name); | |
4322 | goto unwind; | |
4323 | } | |
4324 | } | |
4325 | ||
4326 | /* success */ | |
4327 | memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len); | |
4328 | return 0; | |
4329 | ||
4330 | unwind: | |
4331 | memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len); | |
4332 | tmp_sa.sa_family = bond_dev->type; | |
4333 | ||
4334 | /* unwind from head to the slave that failed */ | |
4335 | stop_at = slave; | |
4336 | bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) { | |
4337 | int tmp_res; | |
4338 | ||
4339 | tmp_res = dev_set_mac_address(slave->dev, &tmp_sa); | |
4340 | if (tmp_res) { | |
4341 | dprintk("unwind err %d dev %s\n", tmp_res, | |
4342 | slave->dev->name); | |
4343 | } | |
4344 | } | |
4345 | ||
4346 | return res; | |
4347 | } | |
4348 | ||
4349 | static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev) | |
4350 | { | |
4351 | struct bonding *bond = bond_dev->priv; | |
4352 | struct slave *slave, *start_at; | |
4353 | int i; | |
4354 | int res = 1; | |
4355 | ||
4356 | read_lock(&bond->lock); | |
4357 | ||
4358 | if (!BOND_IS_OK(bond)) { | |
4359 | goto out; | |
4360 | } | |
4361 | ||
4362 | read_lock(&bond->curr_slave_lock); | |
4363 | slave = start_at = bond->curr_active_slave; | |
4364 | read_unlock(&bond->curr_slave_lock); | |
4365 | ||
4366 | if (!slave) { | |
4367 | goto out; | |
4368 | } | |
4369 | ||
4370 | bond_for_each_slave_from(bond, slave, i, start_at) { | |
4371 | if (IS_UP(slave->dev) && | |
4372 | (slave->link == BOND_LINK_UP) && | |
4373 | (slave->state == BOND_STATE_ACTIVE)) { | |
4374 | res = bond_dev_queue_xmit(bond, skb, slave->dev); | |
4375 | ||
4376 | write_lock(&bond->curr_slave_lock); | |
4377 | bond->curr_active_slave = slave->next; | |
4378 | write_unlock(&bond->curr_slave_lock); | |
4379 | ||
4380 | break; | |
4381 | } | |
4382 | } | |
4383 | ||
4384 | ||
4385 | out: | |
4386 | if (res) { | |
4387 | /* no suitable interface, frame not sent */ | |
4388 | dev_kfree_skb(skb); | |
4389 | } | |
4390 | read_unlock(&bond->lock); | |
4391 | return 0; | |
4392 | } | |
4393 | ||
4394 | /* | |
4395 | * in active-backup mode, we know that bond->curr_active_slave is always valid if | |
4396 | * the bond has a usable interface. | |
4397 | */ | |
4398 | static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev) | |
4399 | { | |
4400 | struct bonding *bond = bond_dev->priv; | |
4401 | int res = 1; | |
4402 | ||
1da177e4 LT |
4403 | read_lock(&bond->lock); |
4404 | read_lock(&bond->curr_slave_lock); | |
4405 | ||
4406 | if (!BOND_IS_OK(bond)) { | |
4407 | goto out; | |
4408 | } | |
4409 | ||
4410 | if (bond->curr_active_slave) { /* one usable interface */ | |
4411 | res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev); | |
4412 | } | |
4413 | ||
4414 | out: | |
4415 | if (res) { | |
4416 | /* no suitable interface, frame not sent */ | |
4417 | dev_kfree_skb(skb); | |
4418 | } | |
4419 | read_unlock(&bond->curr_slave_lock); | |
4420 | read_unlock(&bond->lock); | |
4421 | return 0; | |
4422 | } | |
4423 | ||
4424 | /* | |
169a3e66 JV |
4425 | * In bond_xmit_xor() , we determine the output device by using a pre- |
4426 | * determined xmit_hash_policy(), If the selected device is not enabled, | |
4427 | * find the next active slave. | |
1da177e4 LT |
4428 | */ |
4429 | static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev) | |
4430 | { | |
4431 | struct bonding *bond = bond_dev->priv; | |
1da177e4 LT |
4432 | struct slave *slave, *start_at; |
4433 | int slave_no; | |
4434 | int i; | |
4435 | int res = 1; | |
4436 | ||
4437 | read_lock(&bond->lock); | |
4438 | ||
4439 | if (!BOND_IS_OK(bond)) { | |
4440 | goto out; | |
4441 | } | |
4442 | ||
169a3e66 | 4443 | slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt); |
1da177e4 LT |
4444 | |
4445 | bond_for_each_slave(bond, slave, i) { | |
4446 | slave_no--; | |
4447 | if (slave_no < 0) { | |
4448 | break; | |
4449 | } | |
4450 | } | |
4451 | ||
4452 | start_at = slave; | |
4453 | ||
4454 | bond_for_each_slave_from(bond, slave, i, start_at) { | |
4455 | if (IS_UP(slave->dev) && | |
4456 | (slave->link == BOND_LINK_UP) && | |
4457 | (slave->state == BOND_STATE_ACTIVE)) { | |
4458 | res = bond_dev_queue_xmit(bond, skb, slave->dev); | |
4459 | break; | |
4460 | } | |
4461 | } | |
4462 | ||
4463 | out: | |
4464 | if (res) { | |
4465 | /* no suitable interface, frame not sent */ | |
4466 | dev_kfree_skb(skb); | |
4467 | } | |
4468 | read_unlock(&bond->lock); | |
4469 | return 0; | |
4470 | } | |
4471 | ||
4472 | /* | |
4473 | * in broadcast mode, we send everything to all usable interfaces. | |
4474 | */ | |
4475 | static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev) | |
4476 | { | |
4477 | struct bonding *bond = bond_dev->priv; | |
4478 | struct slave *slave, *start_at; | |
4479 | struct net_device *tx_dev = NULL; | |
4480 | int i; | |
4481 | int res = 1; | |
4482 | ||
4483 | read_lock(&bond->lock); | |
4484 | ||
4485 | if (!BOND_IS_OK(bond)) { | |
4486 | goto out; | |
4487 | } | |
4488 | ||
4489 | read_lock(&bond->curr_slave_lock); | |
4490 | start_at = bond->curr_active_slave; | |
4491 | read_unlock(&bond->curr_slave_lock); | |
4492 | ||
4493 | if (!start_at) { | |
4494 | goto out; | |
4495 | } | |
4496 | ||
4497 | bond_for_each_slave_from(bond, slave, i, start_at) { | |
4498 | if (IS_UP(slave->dev) && | |
4499 | (slave->link == BOND_LINK_UP) && | |
4500 | (slave->state == BOND_STATE_ACTIVE)) { | |
4501 | if (tx_dev) { | |
4502 | struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); | |
4503 | if (!skb2) { | |
4504 | printk(KERN_ERR DRV_NAME | |
4505 | ": Error: bond_xmit_broadcast(): " | |
4506 | "skb_clone() failed\n"); | |
4507 | continue; | |
4508 | } | |
4509 | ||
4510 | res = bond_dev_queue_xmit(bond, skb2, tx_dev); | |
4511 | if (res) { | |
4512 | dev_kfree_skb(skb2); | |
4513 | continue; | |
4514 | } | |
4515 | } | |
4516 | tx_dev = slave->dev; | |
4517 | } | |
4518 | } | |
4519 | ||
4520 | if (tx_dev) { | |
4521 | res = bond_dev_queue_xmit(bond, skb, tx_dev); | |
4522 | } | |
4523 | ||
4524 | out: | |
4525 | if (res) { | |
4526 | /* no suitable interface, frame not sent */ | |
4527 | dev_kfree_skb(skb); | |
4528 | } | |
4529 | /* frame sent to all suitable interfaces */ | |
4530 | read_unlock(&bond->lock); | |
4531 | return 0; | |
4532 | } | |
4533 | ||
4534 | /*------------------------- Device initialization ---------------------------*/ | |
4535 | ||
4536 | /* | |
4537 | * set bond mode specific net device operations | |
4538 | */ | |
169a3e66 | 4539 | static inline void bond_set_mode_ops(struct bonding *bond, int mode) |
1da177e4 | 4540 | { |
169a3e66 JV |
4541 | struct net_device *bond_dev = bond->dev; |
4542 | ||
1da177e4 LT |
4543 | switch (mode) { |
4544 | case BOND_MODE_ROUNDROBIN: | |
4545 | bond_dev->hard_start_xmit = bond_xmit_roundrobin; | |
4546 | break; | |
4547 | case BOND_MODE_ACTIVEBACKUP: | |
4548 | bond_dev->hard_start_xmit = bond_xmit_activebackup; | |
4549 | break; | |
4550 | case BOND_MODE_XOR: | |
4551 | bond_dev->hard_start_xmit = bond_xmit_xor; | |
169a3e66 JV |
4552 | if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) |
4553 | bond->xmit_hash_policy = bond_xmit_hash_policy_l34; | |
4554 | else | |
4555 | bond->xmit_hash_policy = bond_xmit_hash_policy_l2; | |
1da177e4 LT |
4556 | break; |
4557 | case BOND_MODE_BROADCAST: | |
4558 | bond_dev->hard_start_xmit = bond_xmit_broadcast; | |
4559 | break; | |
4560 | case BOND_MODE_8023AD: | |
4561 | bond_dev->hard_start_xmit = bond_3ad_xmit_xor; | |
169a3e66 JV |
4562 | if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) |
4563 | bond->xmit_hash_policy = bond_xmit_hash_policy_l34; | |
4564 | else | |
4565 | bond->xmit_hash_policy = bond_xmit_hash_policy_l2; | |
1da177e4 LT |
4566 | break; |
4567 | case BOND_MODE_TLB: | |
4568 | case BOND_MODE_ALB: | |
4569 | bond_dev->hard_start_xmit = bond_alb_xmit; | |
4570 | bond_dev->set_mac_address = bond_alb_set_mac_address; | |
4571 | break; | |
4572 | default: | |
4573 | /* Should never happen, mode already checked */ | |
4574 | printk(KERN_ERR DRV_NAME | |
4575 | ": Error: Unknown bonding mode %d\n", | |
4576 | mode); | |
4577 | break; | |
4578 | } | |
4579 | } | |
4580 | ||
8531c5ff AK |
4581 | static struct ethtool_ops bond_ethtool_ops = { |
4582 | .get_tx_csum = ethtool_op_get_tx_csum, | |
4583 | .get_sg = ethtool_op_get_sg, | |
4584 | }; | |
4585 | ||
1da177e4 LT |
4586 | /* |
4587 | * Does not allocate but creates a /proc entry. | |
4588 | * Allowed to fail. | |
4589 | */ | |
4590 | static int __init bond_init(struct net_device *bond_dev, struct bond_params *params) | |
4591 | { | |
4592 | struct bonding *bond = bond_dev->priv; | |
4593 | ||
4594 | dprintk("Begin bond_init for %s\n", bond_dev->name); | |
4595 | ||
4596 | /* initialize rwlocks */ | |
4597 | rwlock_init(&bond->lock); | |
4598 | rwlock_init(&bond->curr_slave_lock); | |
4599 | ||
4600 | bond->params = *params; /* copy params struct */ | |
4601 | ||
4602 | /* Initialize pointers */ | |
4603 | bond->first_slave = NULL; | |
4604 | bond->curr_active_slave = NULL; | |
4605 | bond->current_arp_slave = NULL; | |
4606 | bond->primary_slave = NULL; | |
4607 | bond->dev = bond_dev; | |
4608 | INIT_LIST_HEAD(&bond->vlan_list); | |
4609 | ||
4610 | /* Initialize the device entry points */ | |
4611 | bond_dev->open = bond_open; | |
4612 | bond_dev->stop = bond_close; | |
4613 | bond_dev->get_stats = bond_get_stats; | |
4614 | bond_dev->do_ioctl = bond_do_ioctl; | |
8531c5ff | 4615 | bond_dev->ethtool_ops = &bond_ethtool_ops; |
1da177e4 LT |
4616 | bond_dev->set_multicast_list = bond_set_multicast_list; |
4617 | bond_dev->change_mtu = bond_change_mtu; | |
4618 | bond_dev->set_mac_address = bond_set_mac_address; | |
4619 | ||
169a3e66 | 4620 | bond_set_mode_ops(bond, bond->params.mode); |
1da177e4 LT |
4621 | |
4622 | bond_dev->destructor = free_netdev; | |
4623 | ||
4624 | /* Initialize the device options */ | |
4625 | bond_dev->tx_queue_len = 0; | |
4626 | bond_dev->flags |= IFF_MASTER|IFF_MULTICAST; | |
4627 | ||
4628 | /* At first, we block adding VLANs. That's the only way to | |
4629 | * prevent problems that occur when adding VLANs over an | |
4630 | * empty bond. The block will be removed once non-challenged | |
4631 | * slaves are enslaved. | |
4632 | */ | |
4633 | bond_dev->features |= NETIF_F_VLAN_CHALLENGED; | |
4634 | ||
4635 | /* don't acquire bond device's xmit_lock when | |
4636 | * transmitting */ | |
4637 | bond_dev->features |= NETIF_F_LLTX; | |
4638 | ||
4639 | /* By default, we declare the bond to be fully | |
4640 | * VLAN hardware accelerated capable. Special | |
4641 | * care is taken in the various xmit functions | |
4642 | * when there are slaves that are not hw accel | |
4643 | * capable | |
4644 | */ | |
4645 | bond_dev->vlan_rx_register = bond_vlan_rx_register; | |
4646 | bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid; | |
4647 | bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid; | |
4648 | bond_dev->features |= (NETIF_F_HW_VLAN_TX | | |
4649 | NETIF_F_HW_VLAN_RX | | |
4650 | NETIF_F_HW_VLAN_FILTER); | |
4651 | ||
8531c5ff AK |
4652 | bond->bond_features = bond_dev->features; |
4653 | ||
1da177e4 LT |
4654 | #ifdef CONFIG_PROC_FS |
4655 | bond_create_proc_entry(bond); | |
4656 | #endif | |
4657 | ||
4658 | list_add_tail(&bond->bond_list, &bond_dev_list); | |
4659 | ||
4660 | return 0; | |
4661 | } | |
4662 | ||
4663 | /* De-initialize device specific data. | |
4664 | * Caller must hold rtnl_lock. | |
4665 | */ | |
4666 | static inline void bond_deinit(struct net_device *bond_dev) | |
4667 | { | |
4668 | struct bonding *bond = bond_dev->priv; | |
4669 | ||
4670 | list_del(&bond->bond_list); | |
4671 | ||
4672 | #ifdef CONFIG_PROC_FS | |
4673 | bond_remove_proc_entry(bond); | |
4674 | #endif | |
4675 | } | |
4676 | ||
4677 | /* Unregister and free all bond devices. | |
4678 | * Caller must hold rtnl_lock. | |
4679 | */ | |
4680 | static void bond_free_all(void) | |
4681 | { | |
4682 | struct bonding *bond, *nxt; | |
4683 | ||
4684 | list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) { | |
4685 | struct net_device *bond_dev = bond->dev; | |
4686 | ||
4687 | unregister_netdevice(bond_dev); | |
4688 | bond_deinit(bond_dev); | |
4689 | } | |
4690 | ||
4691 | #ifdef CONFIG_PROC_FS | |
4692 | bond_destroy_proc_dir(); | |
4693 | #endif | |
4694 | } | |
4695 | ||
4696 | /*------------------------- Module initialization ---------------------------*/ | |
4697 | ||
4698 | /* | |
4699 | * Convert string input module parms. Accept either the | |
4700 | * number of the mode or its string name. | |
4701 | */ | |
4702 | static inline int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl) | |
4703 | { | |
4704 | int i; | |
4705 | ||
4706 | for (i = 0; tbl[i].modename; i++) { | |
4707 | if ((isdigit(*mode_arg) && | |
4708 | tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) || | |
4709 | (strncmp(mode_arg, tbl[i].modename, | |
4710 | strlen(tbl[i].modename)) == 0)) { | |
4711 | return tbl[i].mode; | |
4712 | } | |
4713 | } | |
4714 | ||
4715 | return -1; | |
4716 | } | |
4717 | ||
4718 | static int bond_check_params(struct bond_params *params) | |
4719 | { | |
4720 | /* | |
4721 | * Convert string parameters. | |
4722 | */ | |
4723 | if (mode) { | |
4724 | bond_mode = bond_parse_parm(mode, bond_mode_tbl); | |
4725 | if (bond_mode == -1) { | |
4726 | printk(KERN_ERR DRV_NAME | |
4727 | ": Error: Invalid bonding mode \"%s\"\n", | |
4728 | mode == NULL ? "NULL" : mode); | |
4729 | return -EINVAL; | |
4730 | } | |
4731 | } | |
4732 | ||
169a3e66 JV |
4733 | if (xmit_hash_policy) { |
4734 | if ((bond_mode != BOND_MODE_XOR) && | |
4735 | (bond_mode != BOND_MODE_8023AD)) { | |
4736 | printk(KERN_INFO DRV_NAME | |
4737 | ": xor_mode param is irrelevant in mode %s\n", | |
4738 | bond_mode_name(bond_mode)); | |
4739 | } else { | |
4740 | xmit_hashtype = bond_parse_parm(xmit_hash_policy, | |
4741 | xmit_hashtype_tbl); | |
4742 | if (xmit_hashtype == -1) { | |
4743 | printk(KERN_ERR DRV_NAME | |
4744 | ": Error: Invalid xmit_hash_policy \"%s\"\n", | |
4745 | xmit_hash_policy == NULL ? "NULL" : | |
4746 | xmit_hash_policy); | |
4747 | return -EINVAL; | |
4748 | } | |
4749 | } | |
4750 | } | |
4751 | ||
1da177e4 LT |
4752 | if (lacp_rate) { |
4753 | if (bond_mode != BOND_MODE_8023AD) { | |
4754 | printk(KERN_INFO DRV_NAME | |
4755 | ": lacp_rate param is irrelevant in mode %s\n", | |
4756 | bond_mode_name(bond_mode)); | |
4757 | } else { | |
4758 | lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl); | |
4759 | if (lacp_fast == -1) { | |
4760 | printk(KERN_ERR DRV_NAME | |
4761 | ": Error: Invalid lacp rate \"%s\"\n", | |
4762 | lacp_rate == NULL ? "NULL" : lacp_rate); | |
4763 | return -EINVAL; | |
4764 | } | |
4765 | } | |
4766 | } | |
4767 | ||
4768 | if (max_bonds < 1 || max_bonds > INT_MAX) { | |
4769 | printk(KERN_WARNING DRV_NAME | |
4770 | ": Warning: max_bonds (%d) not in range %d-%d, so it " | |
4771 | "was reset to BOND_DEFAULT_MAX_BONDS (%d)", | |
4772 | max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS); | |
4773 | max_bonds = BOND_DEFAULT_MAX_BONDS; | |
4774 | } | |
4775 | ||
4776 | if (miimon < 0) { | |
4777 | printk(KERN_WARNING DRV_NAME | |
4778 | ": Warning: miimon module parameter (%d), " | |
4779 | "not in range 0-%d, so it was reset to %d\n", | |
4780 | miimon, INT_MAX, BOND_LINK_MON_INTERV); | |
4781 | miimon = BOND_LINK_MON_INTERV; | |
4782 | } | |
4783 | ||
4784 | if (updelay < 0) { | |
4785 | printk(KERN_WARNING DRV_NAME | |
4786 | ": Warning: updelay module parameter (%d), " | |
4787 | "not in range 0-%d, so it was reset to 0\n", | |
4788 | updelay, INT_MAX); | |
4789 | updelay = 0; | |
4790 | } | |
4791 | ||
4792 | if (downdelay < 0) { | |
4793 | printk(KERN_WARNING DRV_NAME | |
4794 | ": Warning: downdelay module parameter (%d), " | |
4795 | "not in range 0-%d, so it was reset to 0\n", | |
4796 | downdelay, INT_MAX); | |
4797 | downdelay = 0; | |
4798 | } | |
4799 | ||
4800 | if ((use_carrier != 0) && (use_carrier != 1)) { | |
4801 | printk(KERN_WARNING DRV_NAME | |
4802 | ": Warning: use_carrier module parameter (%d), " | |
4803 | "not of valid value (0/1), so it was set to 1\n", | |
4804 | use_carrier); | |
4805 | use_carrier = 1; | |
4806 | } | |
4807 | ||
4808 | /* reset values for 802.3ad */ | |
4809 | if (bond_mode == BOND_MODE_8023AD) { | |
4810 | if (!miimon) { | |
4811 | printk(KERN_WARNING DRV_NAME | |
4812 | ": Warning: miimon must be specified, " | |
4813 | "otherwise bonding will not detect link " | |
4814 | "failure, speed and duplex which are " | |
4815 | "essential for 802.3ad operation\n"); | |
4816 | printk(KERN_WARNING "Forcing miimon to 100msec\n"); | |
4817 | miimon = 100; | |
4818 | } | |
4819 | } | |
4820 | ||
4821 | /* reset values for TLB/ALB */ | |
4822 | if ((bond_mode == BOND_MODE_TLB) || | |
4823 | (bond_mode == BOND_MODE_ALB)) { | |
4824 | if (!miimon) { | |
4825 | printk(KERN_WARNING DRV_NAME | |
4826 | ": Warning: miimon must be specified, " | |
4827 | "otherwise bonding will not detect link " | |
4828 | "failure and link speed which are essential " | |
4829 | "for TLB/ALB load balancing\n"); | |
4830 | printk(KERN_WARNING "Forcing miimon to 100msec\n"); | |
4831 | miimon = 100; | |
4832 | } | |
4833 | } | |
4834 | ||
4835 | if (bond_mode == BOND_MODE_ALB) { | |
4836 | printk(KERN_NOTICE DRV_NAME | |
4837 | ": In ALB mode you might experience client " | |
4838 | "disconnections upon reconnection of a link if the " | |
4839 | "bonding module updelay parameter (%d msec) is " | |
4840 | "incompatible with the forwarding delay time of the " | |
4841 | "switch\n", | |
4842 | updelay); | |
4843 | } | |
4844 | ||
4845 | if (!miimon) { | |
4846 | if (updelay || downdelay) { | |
4847 | /* just warn the user the up/down delay will have | |
4848 | * no effect since miimon is zero... | |
4849 | */ | |
4850 | printk(KERN_WARNING DRV_NAME | |
4851 | ": Warning: miimon module parameter not set " | |
4852 | "and updelay (%d) or downdelay (%d) module " | |
4853 | "parameter is set; updelay and downdelay have " | |
4854 | "no effect unless miimon is set\n", | |
4855 | updelay, downdelay); | |
4856 | } | |
4857 | } else { | |
4858 | /* don't allow arp monitoring */ | |
4859 | if (arp_interval) { | |
4860 | printk(KERN_WARNING DRV_NAME | |
4861 | ": Warning: miimon (%d) and arp_interval (%d) " | |
4862 | "can't be used simultaneously, disabling ARP " | |
4863 | "monitoring\n", | |
4864 | miimon, arp_interval); | |
4865 | arp_interval = 0; | |
4866 | } | |
4867 | ||
4868 | if ((updelay % miimon) != 0) { | |
4869 | printk(KERN_WARNING DRV_NAME | |
4870 | ": Warning: updelay (%d) is not a multiple " | |
4871 | "of miimon (%d), updelay rounded to %d ms\n", | |
4872 | updelay, miimon, (updelay / miimon) * miimon); | |
4873 | } | |
4874 | ||
4875 | updelay /= miimon; | |
4876 | ||
4877 | if ((downdelay % miimon) != 0) { | |
4878 | printk(KERN_WARNING DRV_NAME | |
4879 | ": Warning: downdelay (%d) is not a multiple " | |
4880 | "of miimon (%d), downdelay rounded to %d ms\n", | |
4881 | downdelay, miimon, | |
4882 | (downdelay / miimon) * miimon); | |
4883 | } | |
4884 | ||
4885 | downdelay /= miimon; | |
4886 | } | |
4887 | ||
4888 | if (arp_interval < 0) { | |
4889 | printk(KERN_WARNING DRV_NAME | |
4890 | ": Warning: arp_interval module parameter (%d) " | |
4891 | ", not in range 0-%d, so it was reset to %d\n", | |
4892 | arp_interval, INT_MAX, BOND_LINK_ARP_INTERV); | |
4893 | arp_interval = BOND_LINK_ARP_INTERV; | |
4894 | } | |
4895 | ||
4896 | for (arp_ip_count = 0; | |
4897 | (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count]; | |
4898 | arp_ip_count++) { | |
4899 | /* not complete check, but should be good enough to | |
4900 | catch mistakes */ | |
4901 | if (!isdigit(arp_ip_target[arp_ip_count][0])) { | |
4902 | printk(KERN_WARNING DRV_NAME | |
4903 | ": Warning: bad arp_ip_target module parameter " | |
4904 | "(%s), ARP monitoring will not be performed\n", | |
4905 | arp_ip_target[arp_ip_count]); | |
4906 | arp_interval = 0; | |
4907 | } else { | |
4908 | u32 ip = in_aton(arp_ip_target[arp_ip_count]); | |
4909 | arp_target[arp_ip_count] = ip; | |
4910 | } | |
4911 | } | |
4912 | ||
4913 | if (arp_interval && !arp_ip_count) { | |
4914 | /* don't allow arping if no arp_ip_target given... */ | |
4915 | printk(KERN_WARNING DRV_NAME | |
4916 | ": Warning: arp_interval module parameter (%d) " | |
4917 | "specified without providing an arp_ip_target " | |
4918 | "parameter, arp_interval was reset to 0\n", | |
4919 | arp_interval); | |
4920 | arp_interval = 0; | |
4921 | } | |
4922 | ||
4923 | if (miimon) { | |
4924 | printk(KERN_INFO DRV_NAME | |
4925 | ": MII link monitoring set to %d ms\n", | |
4926 | miimon); | |
4927 | } else if (arp_interval) { | |
4928 | int i; | |
4929 | ||
4930 | printk(KERN_INFO DRV_NAME | |
4931 | ": ARP monitoring set to %d ms with %d target(s):", | |
4932 | arp_interval, arp_ip_count); | |
4933 | ||
4934 | for (i = 0; i < arp_ip_count; i++) | |
4935 | printk (" %s", arp_ip_target[i]); | |
4936 | ||
4937 | printk("\n"); | |
4938 | ||
4939 | } else { | |
4940 | /* miimon and arp_interval not set, we need one so things | |
4941 | * work as expected, see bonding.txt for details | |
4942 | */ | |
4943 | printk(KERN_WARNING DRV_NAME | |
4944 | ": Warning: either miimon or arp_interval and " | |
4945 | "arp_ip_target module parameters must be specified, " | |
4946 | "otherwise bonding will not detect link failures! see " | |
4947 | "bonding.txt for details.\n"); | |
4948 | } | |
4949 | ||
4950 | if (primary && !USES_PRIMARY(bond_mode)) { | |
4951 | /* currently, using a primary only makes sense | |
4952 | * in active backup, TLB or ALB modes | |
4953 | */ | |
4954 | printk(KERN_WARNING DRV_NAME | |
4955 | ": Warning: %s primary device specified but has no " | |
4956 | "effect in %s mode\n", | |
4957 | primary, bond_mode_name(bond_mode)); | |
4958 | primary = NULL; | |
4959 | } | |
4960 | ||
4961 | /* fill params struct with the proper values */ | |
4962 | params->mode = bond_mode; | |
169a3e66 | 4963 | params->xmit_policy = xmit_hashtype; |
1da177e4 LT |
4964 | params->miimon = miimon; |
4965 | params->arp_interval = arp_interval; | |
4966 | params->updelay = updelay; | |
4967 | params->downdelay = downdelay; | |
4968 | params->use_carrier = use_carrier; | |
4969 | params->lacp_fast = lacp_fast; | |
4970 | params->primary[0] = 0; | |
4971 | ||
4972 | if (primary) { | |
4973 | strncpy(params->primary, primary, IFNAMSIZ); | |
4974 | params->primary[IFNAMSIZ - 1] = 0; | |
4975 | } | |
4976 | ||
4977 | memcpy(params->arp_targets, arp_target, sizeof(arp_target)); | |
4978 | ||
4979 | return 0; | |
4980 | } | |
4981 | ||
4982 | static int __init bonding_init(void) | |
4983 | { | |
4984 | struct bond_params params; | |
4985 | int i; | |
4986 | int res; | |
4987 | ||
4988 | printk(KERN_INFO "%s", version); | |
4989 | ||
4990 | res = bond_check_params(¶ms); | |
4991 | if (res) { | |
4992 | return res; | |
4993 | } | |
4994 | ||
4995 | rtnl_lock(); | |
4996 | ||
4997 | #ifdef CONFIG_PROC_FS | |
4998 | bond_create_proc_dir(); | |
4999 | #endif | |
5000 | ||
5001 | for (i = 0; i < max_bonds; i++) { | |
5002 | struct net_device *bond_dev; | |
5003 | ||
5004 | bond_dev = alloc_netdev(sizeof(struct bonding), "", ether_setup); | |
5005 | if (!bond_dev) { | |
5006 | res = -ENOMEM; | |
5007 | goto out_err; | |
5008 | } | |
5009 | ||
5010 | res = dev_alloc_name(bond_dev, "bond%d"); | |
5011 | if (res < 0) { | |
5012 | free_netdev(bond_dev); | |
5013 | goto out_err; | |
5014 | } | |
5015 | ||
5016 | /* bond_init() must be called after dev_alloc_name() (for the | |
5017 | * /proc files), but before register_netdevice(), because we | |
5018 | * need to set function pointers. | |
5019 | */ | |
5020 | res = bond_init(bond_dev, ¶ms); | |
5021 | if (res < 0) { | |
5022 | free_netdev(bond_dev); | |
5023 | goto out_err; | |
5024 | } | |
5025 | ||
5026 | SET_MODULE_OWNER(bond_dev); | |
5027 | ||
5028 | res = register_netdevice(bond_dev); | |
5029 | if (res < 0) { | |
5030 | bond_deinit(bond_dev); | |
5031 | free_netdev(bond_dev); | |
5032 | goto out_err; | |
5033 | } | |
5034 | } | |
5035 | ||
5036 | rtnl_unlock(); | |
5037 | register_netdevice_notifier(&bond_netdev_notifier); | |
c3ade5ca | 5038 | register_inetaddr_notifier(&bond_inetaddr_notifier); |
1da177e4 LT |
5039 | |
5040 | return 0; | |
5041 | ||
5042 | out_err: | |
40abc270 FM |
5043 | /* |
5044 | * rtnl_unlock() will run netdev_run_todo(), putting the | |
5045 | * thus-far-registered bonding devices into a state which | |
5046 | * unregigister_netdevice() will accept | |
5047 | */ | |
5048 | rtnl_unlock(); | |
5049 | rtnl_lock(); | |
5050 | ||
1da177e4 LT |
5051 | /* free and unregister all bonds that were successfully added */ |
5052 | bond_free_all(); | |
5053 | ||
5054 | rtnl_unlock(); | |
5055 | ||
5056 | return res; | |
5057 | } | |
5058 | ||
5059 | static void __exit bonding_exit(void) | |
5060 | { | |
5061 | unregister_netdevice_notifier(&bond_netdev_notifier); | |
c3ade5ca | 5062 | unregister_inetaddr_notifier(&bond_inetaddr_notifier); |
1da177e4 LT |
5063 | |
5064 | rtnl_lock(); | |
5065 | bond_free_all(); | |
5066 | rtnl_unlock(); | |
5067 | } | |
5068 | ||
5069 | module_init(bonding_init); | |
5070 | module_exit(bonding_exit); | |
5071 | MODULE_LICENSE("GPL"); | |
5072 | MODULE_VERSION(DRV_VERSION); | |
5073 | MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION); | |
5074 | MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); | |
5075 | MODULE_SUPPORTED_DEVICE("most ethernet devices"); | |
5076 | ||
5077 | /* | |
5078 | * Local variables: | |
5079 | * c-indent-level: 8 | |
5080 | * c-basic-offset: 8 | |
5081 | * tab-width: 8 | |
5082 | * End: | |
5083 | */ | |
5084 |